Extraction, Refining, and Fabrication of Light Metals. Proceedings of the International Symposium on Extraction, Refining and Fabrication of Light Metals, Ottawa, Ontario, August 18â€“21,

Titles of Related Interest— Ashby ENGINEERING MATERIALS 1

Ashby ENGINEERING MATERIALS 2

Brook IMPACT OF NON-DESTRUCTIVE TESTING

Koppel AUTOMATION IN MINING, MINERAL AND METAL PROCESSING 1989

Ruhle METAL-CERAMIC INTERFACES

Taya METAL MATRIX COMPOSITES

Other CIM Proceedings Published by Pergamon

Bergman FERROUS AND NON-FERROUS ALLOY PROCESSES

Bickert REDUCTION AND CASTING OF ALUMINUM

Chalkley TAILING AND EFFLUENT MANAGEMENT

Closset PRODUCTION AND ELECTROLYSIS OF LIGHT METALS

Dobby PROCESSING OF COMPLEX ORES

Embury HIGH TEMPERATURE OXIDATION AND SULPHIDATION PROCESSES

Jaeck PRIMARY AND SECONDARY LEAD PROCESSING

Jonas DIRECT ROLLING AND HOT CHARGING OF STRAND CAST BILLETS

Kachanlwsky IMPACT OF OXYGEN ON THE PRODUCTIVITY OF NON-FERROUS

METALLURGICAL PROCESSES

Lalt F. WEINBERG INTERNATIONAL SYMPOSIUM ON SOLIDIFICATION PROCESSING

Macmillan QUALITY AND PROCESS CONTROL IN REDUCTION AND CASTING OF ALUMINUM AND OTHER LIGHT METALS

Mostaghacl PROCESSING OF CERAMIC AND METAL MATRIX COMPOSITES

Plumpton PRODUCTION AND PROCESSING OF FINE PARTICLES

Purely FUNDAMENTALS AND APPLICATIONS OF TERNARY DIFFUSION

Rigaud ADVANCES IN REFRACTORIES FOR THE METALLURGICAL INDUSTRIES

Ruddle ACCELERATED COOLING OF ROLLED STEEL

Salter GOLD METALLURGY

Thompson COMPUTER SOFTWARE IN CHEMICAL AND EXTRACTIVE METALLURGY

TWIgge-Molecey MATERIALS HANDLING IN PYROMETALLURGY

IWigge-Molecey PROCESS GAS HANDLING AND CLEANING

TVson FRACTURE MECHANICS

Wilkinson ADVANCED STRUCTURAL MATERIALS

Related Journals (Free sample copies available upon request)

ACTA METALLURGICA CANADIAN METALLURGICAL QUARTERLY MATERIALS RESEARCH BULLETIN MINERALS ENGINEERING SCRIPTA METALLURGICA

Titles of Related Interest— Ashby ENGINEERING MATERIALS 1

Ashby ENGINEERING MATERIALS 2

Brook IMPACT OF NON-DESTRUCTIVE TESTING

Koppel AUTOMATION IN MINING, MINERAL AND METAL PROCESSING 1989

Ruhle METAL-CERAMIC INTERFACES

Taya METAL MATRIX COMPOSITES

Other CIM Proceedings Published by Pergamon

Bergman FERROUS AND NON-FERROUS ALLOY PROCESSES

Bickert REDUCTION AND CASTING OF ALUMINUM

Chalkley TAILING AND EFFLUENT MANAGEMENT

Closset PRODUCTION AND ELECTROLYSIS OF LIGHT METALS

Dobby PROCESSING OF COMPLEX ORES

Embury HIGH TEMPERATURE OXIDATION AND SULPHIDATION PROCESSES

Jaeck PRIMARY AND SECONDARY LEAD PROCESSING

Jonas DIRECT ROLLING AND HOT CHARGING OF STRAND CAST BILLETS

Kachanlwsky IMPACT OF OXYGEN ON THE PRODUCTIVITY OF NON-FERROUS

METALLURGICAL PROCESSES

Lalt F. WEINBERG INTERNATIONAL SYMPOSIUM ON SOLIDIFICATION PROCESSING

Macmillan QUALITY AND PROCESS CONTROL IN REDUCTION AND CASTING OF ALUMINUM AND OTHER LIGHT METALS

Mostaghacl PROCESSING OF CERAMIC AND METAL MATRIX COMPOSITES

Plumpton PRODUCTION AND PROCESSING OF FINE PARTICLES

Purely FUNDAMENTALS AND APPLICATIONS OF TERNARY DIFFUSION

Rigaud ADVANCES IN REFRACTORIES FOR THE METALLURGICAL INDUSTRIES

Ruddle ACCELERATED COOLING OF ROLLED STEEL

Salter GOLD METALLURGY

Thompson COMPUTER SOFTWARE IN CHEMICAL AND EXTRACTIVE METALLURGY

TWIgge-Molecey MATERIALS HANDLING IN PYROMETALLURGY

IWigge-Molecey PROCESS GAS HANDLING AND CLEANING

TVson FRACTURE MECHANICS

Wilkinson ADVANCED STRUCTURAL MATERIALS

Related Journals (Free sample copies available upon request)

ACTA METALLURGICA CANADIAN METALLURGICAL QUARTERLY MATERIALS RESEARCH BULLETIN MINERALS ENGINEERING SCRIPTA METALLURGICA

PROCEEDINGS OF THE INTERNATIONAL SYMPOSIUM ON EXTRACTION, REFINING AND FABRICATION OF LIGHT METALS, OTTAWA, ONTARIO, AUGUST 18-21, 1991

Extraction, Refining, and Fabrication of Light Metals

Editors

Mahi Sahoo CANMET Ottawa, Ontario

Peter Pinfold Norsk Hydro Canada Inc. Becancour, Quebec

Symposium organized by the Light Metals Section of The Metallurgical Society of CIM

30th ANNUAL CONFERENCE OF METALLURGISTS OF CIM 30e CONFERENCE ANNUELLE DES M E T A L L U R G I E S DE LTCM

Pergamon Press New York • Oxford • Beijing • Frankfurt

Pergamon Press Offices:

U.S.A.

U.K.

PEOPLE'S REPUBLIC OF CHINA

FEDERAL REPUBLIC OF GERMANY

BRAZIL

AUSTRALIA

JAPAN

CANADA

Pergamon Press, Inc., Maxwell House, Fairview Park, Elmsford, New York 10523, U.S.A.

Pergamon Press pic, Headington Hill Hall, Oxford 0X3 OBW, England

Pergamon Press, 0909 China World Tower, No. 1 Jian Guo Men Wai Avenue, Beijing 1000004, People's Republic of China

Pergamon Press GmbH, Hammerweg 6, D-6242 Kronberg, Federal Republic of Germany

Pergamon Editora Ltda, Rua Ega de Queiros, 346 CEP 04011, Paraiso, Sao Paulo, Brazil

Pergamon Press Australia Pty Ltd., P.O. Box 544, Potts Point, NSW 2011, Australia

Pergamon Press, 8th Floor, Matsuoka Central Building, 1-7-1 Nishishinjuku, Shinjuku-ku, Tokyo 160, Japan

Pergamon Press Canada Ltd., Suite 271, 253 College Street, Toronto, Ontario M5T 1R5 Canada

Copyright © 1991 Pergamon Press Inc.

All rights reserved. No part of this publication may be reproduced in a retrieval system or transmitted in any form or by any means: electronic, electrostatic, magnetic tape, mechanical, photocopying, recording or otherwise, without permission in writing from the publishers.

Library of Congress Cataloging in Publication Data

ISBN 0-08-041444-3

Printing: 1 2 3 4 5 6 7 8 9 Year: 0 1 2 3 4 5 6 7 8 9

Printed In the United States of America

0™ The paper used in this publication meets the minimum require-ments of American National Standard for Information Sciences-Permanence of Paper for Printed Library Materials, ANSI Z 39.48-1984

Session Chairmen

Aspects of Magnesium Technology M. Sahoo M. Avedesian

CANMET Institute of Magnesium Technology Ottawa, Ontario Ste-Foy, Quebec

Light Metal Matrix Composites B. Closset J. Masounave

Timminco Metals ETS, Universite du Quebec a Montreal Toronto, Ontario Montreal, Quebec

Aspects of Light Metals Reduction Technology L. Larouche C. Bickert

Canadian Reynolds Metals Co. Ltd. Pechiney Corp. Baie Comeau, Quebec Greenwich, Connecticut, U.S.A.

Casting and Solidification of Aluminum Base Alloys P. Pinfold P. Tremblay

Norsk Hydro Canada Inc. Arvida R&D Centre, Alcan International Ltd. Becancour, Quebec Jonquiere, Quebec

Mathematical Modelling and Computer Simulation R. Guthrie M. Bouchard

McGill University Universite du Quebec a Chicoutimi Montreal, Quebec Chicoutimi, Quebec

Melting, Alloying and Properties of Light Metals P. Aylen P. Pinfold

Alumax Inc. Norsk Hydro Canada Inc. Montreal, Quebec Becancour, Quebec

v

Foreword

The Light Metals Section of the Metallurgical Society of CIM has been organizing international symposia for the last three years on advances in different aspects of light metals technology during the Annual Conference of Metallurgists.

This year in conjunction with the 30th Annual Conference of Metallurgists of CIM, the Light Metals Section is proud to present its Fourth International Symposium on "Extraction, Refin-ing and Fabrication of Light Metals". World-class scientists and engineers from more than six countries are presenting over thirty papers on topics such as magnesium casting technology, metal matrix composites, mathematical modelling, solidification, and reduction of light metals.

In view of Canada's recent emergence as an important magnesium producer, a separate session will be held to focus on the recent advances in magnesium technology.

Metal matrix composites are an important class of the advanced industrial materials and signifi-cant advances have been achieved recently on the fabrication and characterization of their micro-structures and mechanical properties. The session on light metal matrix composites comprises papers on some of these recent developments.

As the Symposium chairpeople, we wish to express our sincere thanks to all the authors for preparing their manuscripts in time for publication, and for presenting their papers at the Symposium.

One of us (M.S.) would like to acknowledge the assistance provided by Messrs. Peter Newcombe and Bob Emmett in taking care of the preliminary planning and correspondence for the Symposium announcements. Mr. Newcombe, in particular spent considerable time compiling the addresses and preparing the technical program for the sessions. The moral support of the management of the Metals Technology Laboratories of CANMET is also gratefully acknowledged.

August 1991 Mahi Sahoo Peter Pinfold Editors

vii

3

Production and applications of ultra high purity magnesium

S.G. Hibbins, F.C. Dimayuga Timminco Metals, A Division of Timminco Limited, Haley, Ontario, Canada

INTRODUCTION

Timminco Metals

1 Haley, Ontario facility has produced

magnesium metal and alloys since 1942. Under the guidance of Dr. L.N. Pidgeon, this plant became the first commercialized application of the classical Pidgeon Process for producing magnesium metal (1,2,3,4). Today, this 800 acre facility is the recognized leader in the production of the world's highest purity magnesium metal and alloys.

Timminco has been responsible for significant advancements in the production and application of light and reactive metals. The original thermal magnesium process has been adapted to the production of calcium and strontium metals. Research and development has led to the introduction of such products as CAL-AL and MAGCAL for use by the lead industry, a variety of strontium products including a 90% Sr - 10% Al alloy for use by the aluminum foundry industry and various granular and particulate magnesium and calcium products. Timminco has made significant process improvements over the years to increase automation and efficiency of the silicothermic magnesium process and also to refine the process to consolidate its status as producer of the highest purity magnesium commercially available.

Today, Timminco Metals operates three production facilities which are shown in Figure 1. The Haley plant remains the primary source of magnesium metal and alloys, ingots, extrusions, granules and particulates. The Westmeath plant, which was commissioned in 1987, is the source of strontium and calcium metal and special alloys. The Beauharnois facility continues to supply 85% and 75% ferrosilicon to Haley and a limited number of customers in the steel industry. The unique nature of the metallothermic process facilities enables Timminco to maintain a high degree of operating flexibility with a variety of products being manufactured at either the Haley or Westmeath plants, as circumstances require.

As a major part of the company's efforts to continuously improve the quality and delivery of its products, Timminco has implemented a comprehensive Quality Assurance Program which is designed to meet and in some cases, exceed the requirements of the Canadian Standard Z299.3 (5). As a result, Timminco's Westmeath operations have been given a Ql Preferred Quality Supplier Award by the Ford Motor Company.

4 EXTRACTION, REFINING AND FABRICATION OF LIGHT METALS

In recent years, Timminco has reaffirmed its commitment to research and development, especially towards new process and product development. In 1988, with the transfer of specialty metals production to the Westmeath plant, investments have been made at the 7500 square foot Haley Research facility, with the installation of 3 new experimental melting furnaces and additional small-scale vacuum furnace capacity. Timminco also has an active out of house R & D programme, which has included research at several universities, the Canada Centre for Mineral and Energy Technology (CANMET) and the Institute of Magnesium Technology (IMT).

Integrated nesii Biter

Magnesium Smell

magnesium metal S alloys; ingots, extrusions, granules

TIMMINCO METALS

HALEY

R&D Pilot Plant

new products + processes

WESTMEATH

Sr + Ca products,

special alloys

BEAUHARNOIS

ferrosilicon

Figure 1 - Timminco Metals Facilities.

THE SILICOTHERMIC MAGNESIUM PROCESS

Process Description

In the production of magnesium by the Pidgeon silicothermic process (Figure 2), high purity dolomite, quarried from an on-site open pit, is crushed, classified and calcined. The burned dolomite is subsequently mixed with a controlled amount of ferrosilicon obtained from Timminco^ Beauharnois, Quebec ferroalloy plant. The mixture is briquetted and charged to high-alloy steel retorts contained in gas-fired furnaces.

Reduction is carried out under vacuum at about 1200°C according to the following chemical reaction:

2MgO CaO + Si = 2Mg + 2CaO Si02 (1)

During this reaction, metallic magnesium is vaporized and transported to a controlled temperature condenser where sublimation takes place. After a specified period of time, the vacuum is broken and the condensed "crown" of magnesium metal and the dicalcium silicate slag are discharged from the retort. The metallic crowns are subsequently processed to produce a variety of pure metal and alloy products including ingots, billets, sacrificial anodes, extrusions and granules.

EXTRACTION, REFINING AND FABRICATION OF LIGHT METALS 5

Reductant

Briquetting

Chemical Laboratory

Computer

Figure 2 - Timminco

fs Metallothermic Vacuum Reduction Process.

Over the years, substantial progress has been made in automating the Pidgeon process. Froats (4) outlined the advances which have been implemented to improve the tooling for charging and discharging the multiplicity of small units characteristic of the process. Automated equipment was developed to distribute a predetermined quantity of briquette charge in the horizontal retorts prior to the vacuum reduction cycle. The technology to remove hot spent residue by a vacuum pneumatic system was also developed. By these developments, the laborious manual tasks of charging briquettes and discharging residue were eliminated.

Since 1980, further improvements have been implemented in the process, the most significant of these being:

upgrading of plant milling, mixing and briquetting equipment, including improved, automatic control of raw material batch weights. conversion of electric retort furnace capacity to gas-fired furnaces incorporating energy-efficient regenerative burner technology. implementation of statistical process control (SPC) in all phases of plant operation, as part of a comprehensive quality assurance program (5). automation and streamlining of several downstream operations, e.g., metal crown handling, granulating and extrusion.

Metallurgical Efficiency

A significant body of knowledge has been developed at Timminco on the metallurgy of the Pidgeon process (1,2,3,4). Operating efficiency is dependent on such variables as temperature, vacuum, cycle time, type of reductant, raw material sizing, briquette density, retort alloy composition and dimensions.

Ingots & Extrusions

Melt & Alloy

Cut & Size

Granulate

Package & Ship

Slag

On-Line Data Base


The roost significant factor affecting operating efficiency is the relationship between magnesium reduction recovery and ferrosilicon use. According to reaction (1), the stoichiometric requirement is 0.577 kg Si per kg of Mg contained in the retort charge. In practice, an excess of the stoichiometric amount is necessary to obtain high Mg recovery. The excess amount must be determined empirically, as no adequate theoretical prediction based on the Fe-Si phase diagram can explain the optimum operating region.

Plant operations can be run equally well using either 75% or 85% ferrosilicon, although at present, 85% ferrosilicon is preferred due to slightly higher operating productivity. The optimum operating condition is based on a combination of ferrosilicon usage, magnesium recovery, ferrosilicon price, production rate and other operating factors.

HIGH PURITY MAGNESIUM

Overview of Applications

Magnesium has a diversity of applications, as Figure 3 shows. The main applications can be grouped as aluminum alloying, chemical and structural uses. As shown, 52% of world magnesium shipments in 1990 were used in aluminum alloying. Chemical applications, including nodular iron, steel desulphurization, metal reduction, sacrificial anodes and Grignard reagents, accounted for about 27% of magnesium consumption. Structural applications (die and gravity casting, wrought products) were about 18% of world magnesium consumption in 1990. Chemical applications account for most of the consumption of higher purity magnesium. The development of high purity magnesium alloys has alleviated many of the concerns related to corrosion of magnesium in structural applications. The most significant growth potential for magnesium consumption is expected to be in magnesium casting alloys.

Aluminium 52%

Chemical 10%

Total shipments 1990; 252.000 mt Figure 3 - Primary Magnesium Shipment by Market Segment, 1990.

(Source IMA).

Nodular Iron

Desulfurization 11%

Structural 18%

Other 3%


Pure Magnesium - Specifications and Products

In magnesium manufacture, Timminco can control individual impurity elements to meet specialized requirements. Timminco

1s

control of impurity elements has led to the development of three unique grades superior to the generic 99.8% magnesium grade. These are designated High Purity 99.90%, Super Purity 99.95% and Ultra Purity 99.98%. The high purity grades are normally produced in a variety of ingot forms. Recent R & D has also enabled Timminco to produce the complete range of magnesium purities as granular and particulate products. The typical size distribution of the high purity particulate products is shown in Table I.

Table I - Timminco High Purity Magnesium Particulates -Typical Size Distribution.

Product Screen Size, % Retained (Tyler Mesh)

4 6 10 14 20 35 65 100 200 -200

Granules 2 25 60 10 2 1

Particulate 0 0 5 35 30 20 5 2 2 1

Fine Particulate

0 0 0 2 25 50 15 3 4 1

Pure Magnesium - Process Capability

Table II shows the specifications and mean impurity content of Timminco's pure magnesium for both 99.95% and 99.98% magnesium grades. As shown, the mean impurity contents of both grades are well below the maximum specifications.

The capability of a process is measured statistically by its capability index, Cpk, which is defined as the ratio of the upper specification limit minus the mean to three times the standard deviation. Values of Cpk much greater than one indicate that the process produces according to specifications more than 99.7% of the time. Table III lists the process capability index for various magnesium grades. The Cpk is normally well above 1, thus ensuring that all casts meet the required specification.

With the objective of continuously increasing Timminco's process capability, a research project was undertaken to identify the effects of plant operating practice on final magnesium ingot aluminum content. The tests showed that the crown and un-refined melt must contain very little aluminum to begin with, if aluminum levels below .0030% are to be achieved in refined magnesium.

Table II - Pure Magnesium - Specifications and Mean Impurity Content

Commercial

Timminco High Purity Grades

Element

99.80

Grade

%max

High Purity

99.90 Grade

%max

Super Purity

99.95 Grade

Spec

Mean

%max

%

Ultra Purity

99.98 Grade

Spec

Mean

%max

%

Aluminum

(1)

(1)

(1)

0.0040

0.004

0.0030

Zinc

(1)

(1)

(1)

0.0045

0.007

0.0045

Manganese

0.10

0.01

0.01

0.0030

0.002

0.0015

Iron

(1)

0.0070

0.003

0.0015

0.002

0.0015

Nickel

0.001

0.001

0.001

<0.0005

0.0005

<0.0005

Copper

0.02

0.005

0.002

<0.0005

0.0005

<0.0005

Silicon

(1)

0.010

0.010

0.0045

0.003

0.0025

Lead

0.01

0.005

0.003

0.0010

0.001

<0.0010

Calcium

0.010

0.005

0.003

0.0012

0.003

0.0012

Tin

0.01

0.001

0.001

<0.001

0.001

<0.0010

Cadmium

—

—

—

—

—

<0.0001

Others,each

0. 05

0.012

0.01

0. 005

Others,total

0.20

0.100

0.03

0.02

(1) Controlled by limits for others, each.

oo

EXTRACTION, REFINING AND FABRICATION OF LIGHT METALS


Table III - Pure Magnesium Ingot - Standard Purity Process Capability Index - Cpk.

Element Mean (%)

Standard Deviation

99.80 Grade

cP |c 99.90 Grade

99.95 Grade

Aluminum 0.0040 0.0006 27.7 4.4 5.6

Zinc 0.0045 0.0010 16.5 2.5 3.2

Manganese 0.0030 0.0005 67.3 7.3 7.3

Iron 0.0015 0.0007 25.0 2.6 0.9

Silicon 0.0045 0.0008 20.6 2.3 2.3

Calcium 0.0012 0.0004 7.3 3.5 1.8

Note: 1. Nickel, Copper, Lead, Tin, Cadmium - Typical values are virtually constant at or below analytical detection limits

2. Results are typical of 4th quarter - 1991

Cpk = Upper specification - Mean 3 sigma

Operating practice E, shown in Figure 4, further lowers the aluminum content of 99.98% magnesium. With this operating practice, levels of other impurities are also further reduced. The results were achieved under plant operating conditions and are thus representative of the capability of the silicothermic process for commercial production of ultra-high purity magnesium.

Weight Percent

0.007

0.006

0.005

0.004

0.003

0.002

0.001

0.000

Al Zn Mn Fe Ni Cu Si Figure 4 - Mean Impurity Content of Magnesium Ingot.

9

• • 9 9 . 9 8 8 p e c .

EZ3 9 9 . 9 8 M e a n

EHB T r i a l E


HIGH PURITY MAGNESIUM CASTING ALLOYS

Magnesium Casting Alloys - Specifications and Products

Timminco Metals was the first to offer a high purity alloy for die and gravity casting, AZ91X. As shown in Table IV, the AZ91X alloy contains low amounts of iron and nickel and is particularly low in copper and silicon. Timminco also produces super and ultra-high purity alloys, AZ91SX and AZ91UX, respectively (6) . These alloys have extremely low levels of iron, nickel and copper and have been designed to provide exceptionally low and consistent corrosion rates for both die and gravity cast applications where corrosion is a prime concern.

Corrosion of Magnesium Casting Alloys

Although the detrimental effects of heavy metal impurities on the corrosion resistance of magnesium alloys have been known for many years (7), only recently have the benefits of higher purity alloys been fully appreciated, (8,9,10).

Recent studies using the ASTM B117 salt spray corrosion test on AZ91 and AM60 magnesium casting alloys have shown that the resistance to salt spray is dramatically improved by reducing trace amounts of nickel and copper as well as lowering the iron to manganese ratio in the casting. These findings formed the basis for the specifications the industry has adopted for most of the high purity alloys available today.

Ultra-High Purity Die Cast Alloys

Very little data on the corrosion of magnesium alloys with extremely low levels of impurities exist in the literature. Timminco^ pioneering efforts into the investigation of corrosion properties of ultra-high purity alloys have now resulted in data on die cast parts (6).

To broaden the range of compositions from the very low levels of impurities to higher ones, Zuliani combined Timminco's data with those reported by Hillis (9,10) and developed the following multiple regression equation to account for the combined effects of heavy metal impurities on the salt spray corrosion rate of AZ91 die castings:

Log (corrosion rate, mils/yr) = 1.5657 + 0.4931 log (%Cu) (2) + 168.8215 (% Ni) + 18.8154 (Fe/Mn)

r

2 = 0.83; standard error = 0.275

F-ratio = 124.85; degrees of freedom = 3,79

Using equation (2) the maximum corrosion rates of various grades of AZ91 were calculated and listed in Table IV. Due to its lower copper specification, Timminco

1s AZ91X is expected to have

lower corrosion rates up to around 12 mils per year as compared with AZ91D whose corrosion rates may be as high as 25 mils per year.


Table IV - Predicted Maximum Corrosion Rates of Various Grades of Die-Cast AZ91

Max. Salt Spray Alloy Cu, Ni, Fe, Mn, Corrosion Rate

%max %max %max %min mils per yr.**

AZ91C .10 .01 (.01)* .13 >3,000 AZ91D .015 .001 .004 .17 25 AZ91X .003 .001 .004 .17 12 AZ91SX .0024 .0010 .0024 .17 5 AZ91UX .0010 .0010 .0015 .17 3

* No max. Fe is specified; assume typical value is 0.01%

** Calculated using equation 2 and specified chemical analysis in Table IV taken to the nearest part per million (4 decimal points) . For example, 0.001% Ni spec can be satisfied by rounding down an actual analysis of 0.0014% Ni.

Further reduction in anticipated maximum corrosion rates are seen in the super and ultra-purity grades with corrosion as low as 5 mils per year for AZ91SX and 3 mils per year for AZ91UX.

In addition, by simultaneously reducing iron, nickel and copper impurity levels in the casting, the variability in component-to-component corrosion performance is significantly reduced. Thus, castings which meet AZ91D specifications may exhibit corrosion rates anywhere from about 1 up to 25 mils per year depending on the exact chemistry of each particular casting. However, castings of AZ91UX specification may exhibit corrosion rates from about 1 to 3 mils per year.

Figure 5 illustrates the visual appearance of AZ91 die cast test panels as a function of salt spray corrosion rates.

As shown;

at 31.9 mils per year, about the maximum expected with AZ91D, surface pitting is heavy and deep; however, no perforations exist as would have been the case prior to the development of the AZ91D high purity alloy. at 6.9 mils per, surface pitting is significantly reduced; however, localized deep pits are still evident, at 5.3 mils per year, about the maximum expected with AZ91SX, surface pitting is minimal and not deep; the sample is more blemished that pitted. Sample identification scribe marks are clearly visible. at 2.9 mils per year and less, which is typical of AZ91UX, no pits are evident. The panel surface has a very fine roughened appearance and is clean enough to reveal the flow pattern of the liquid metal as it was injected into the die cavity.

Figure 5 - Visual Appearance of AZ91 Magnesium Alloy Test Panels of Varying Purity After 10 Days Exposure to Salt Spray.


6.9 mpy 31.9 m p y

5.3 m p y 2.9 m p y

1.2 mpy 0.4 m p y


Importance of the Fe/Mn Ratio

In the regression analysis of die casting corrosion data, the Fe/Mn ratio in the casting was more highly correlated to the corrosion rate than was the iron analysis. Manganese appears to have a twofold effect, first precipitating iron to the solubility limit prior to casting the melt and second, coating the remaining iron particles during solidification, thereby inhibiting their cathodic corrosion effect in the final casting (12).

The solubility of manganese in AZ91 is strongly dependent on the iron content of the alloy and the melt temperature (9). The lower metal temperatures encountered in many die casting foundries compared to primary metal operations often leads to significant manganese precipitation during primary ingot remelting. Thus, the Fe/Mn ratio of the primary ingots is not a good indicator for predicting the corrosion resistance of the final casting.

Thus, even though the corrosion rate is dependent on the Fe/Mn ratio in the casting, the addition of large amounts of manganese to the primary metal will not negate the harmful effects of excessively high iron levels. In view of the propensity for manganese precipitation, reducing the iron content of the primary metal and following good foundry practice to minimize iron pickup during processing are the only effective ways of ensuring low corrosion rates.

HIGH PURITY MAGNESIUM ANODES

Cathodic Protection

Today, a principal means of preventing corrosion of metals like iron or copper is cathodic protection, where a counter current is imposed either from an external source or through the use of a sacrificial anode. Magnesium and its alloys, due to their highly anodic position in the electromotive series and favourable electrochemical equivalent weight, have been widely used as sacrificial anodes for underground structures, hot water tanks and chemical equipment.

Underground Anodes

Overall anode performance in the protection of underground structures is dependent on several factors namely:

anode composition and impurities anode current density time backfill composition, preparation and installation practice soil composition, temperatures, moisture and other characteristics.

Osborn and Robinson (13) reported the results of both laboratory and field tests which highlight the importance of purity in ground anode performance. Figure 6 shows the effect of various impurities on anode efficiency of AZ63 alloy.


The field test which was conducted for 6 months demonstrated that efficiencies of high purity primary alloys rise rapidly to a maximum of 1300 ampere-hours per kilogram, whereas the efficiency of lower purity secondary AZ63 rises less rapidly to a maximum of only 600 ampere-hours per kilogram. The levels of impurity contained in the tested anodes are shown in Table V.

In practice, therefore, using anodes of high purity AZ63 will result in more efficient cathodic protection and consequently, longer service life of the protected structure such as underground pipelines and storage tanks.

Table V - Composition of AZ63 Ground Anodes Referred to in Figure 6.

Alloy Purity Al Zn

Analysis, Fe

Weight % Ni Cu Mn

High 6.5 2.9 .001 <.001 <.01 .23

Low 6.1 2.9 .004 .006 .05 .14

Current Capacity, A-h/kg

1400

1200

1000

800

600

400

200

0

0.00 0.05 0.10 0.15 0.20 0.25 Cur ren t Dens i ty , m A / c m 2

Figure 6 - Effect of Impurities on Current Capacity of AZ63 Ground Anodes.

Hot Water Tank Anodes

Similar to ground anodes, hot water tank anodes exhibit different protection efficiencies depending on a number of factors including anode composition and impurity levels, water conditions, tank construction, etc..


Researchers at Curtin University of Technology (14) studied the effect of iron content on the corrosion of AZ31 anodes. They simulated hot water tank conditions by imposing a range of current densities, 0 to .114 mA cm"

2 through a potentiostat-galvanostat and

using flowing pre-heated water as the electrolyte. They monitored the galvanic current and correlated it to the total weight loss of the anodes. The anodes are of AZ31 composition with two iron levels: high (0.0061% Fe) and low (0.0028% Fe).

Their findings are plotted in Figure 7 where it can be seen that using AZ31 anodes of low iron contents results in reduced corrosion rates and longer life. As a result, Hughes et al have recommended that iron contents of AZ31 anodes be kept below 0.003%

The effects of alloy purity on anode performance was also the subject of a Timminco-sponsored study at the New Brunswick Research and Productivity Council. The 14-day test employed a current density of 0.5 mA cm"

2 and compared two impurity levels.

The results presented in Figure 8 point out that the high purity AZ31 anode exhibits a current capacity in excess of 1300 ampere-hours per kg while the lower purity anode exhibits only 1150 ampere-hours per kg.

(15).

Total Corrosion Rate, g/m2/day

20

10h

15

- B - L o w f - A - H i g h Wm

5 0 2 4 6 8 10

Galvanic Corrosion Rate, g/m2/day

12 14

Figure 7 Effect of Iron Content on Corrosion of AZ31.


Current Capacity, A-h/kg 1,400 r

High Purity Low Purity

Figure 8 - Effect of Alloy Purity on Current Capacity of AZ31 Anodes.

Therefore, at the same level of protection (ie: same galvanic current load), higher purity anodes will corrode at a slower rate providing a longer service life of the water tank. Other advantages follow including the less incidence of "white water" which is the result of evolution of small amounts of hydrogen gas from the excessive self-corrosion of the anode. In addition, the rapid consumption of anodes associated with lower purity alloys means that scale and residue materials build up rapidly inside the tank which usually impedes efficient heat transfer.

In view of the aforementioned data on the beneficial effects of alloy purity on anode performance, Timminco Metals had specified impurity levels lower than that specified by ASTM in its alloys used for anode manufacture. As listed in Table VI, the maximum iron, nickel and copper contents in Timminco's AZ31 and AZ63 alloys are significantly lower than those specified by ASTM.

Table VI - Maximum Impurity Contents of Magnesium Alloys for Anode Manufacture.

Alloy Spec'n Fe Ni Cu Si Total Others

AZ31 ASTM .005 .005 .05 .10 .30 Timminco .002 .001 .002 .02 .10

AZ63 ASTM — .010 .20 .20 .30 Timminco .008 .005 .005 .02 .10

High Low Purity Purity

Al 2.8 2.8 Zn 0.87 0.84 Mn 0.51 0.43 Fe 0.0012 0.0045 Ni 0 .0003 0.0020 Cu 0.0011 0.027


SUMMARY

Timminco Metals has accumulated substantial experience and expertise in the application of the silicothermic reduction process to the commercial production of standard and high purity magnesium and its alloys. Under the close scrutiny of its quality assurance program, Timminco has the ability to produce a wide range of purity specifications and product forms to satisfy a diversity of customer demands.

The advantages offered by high purity magnesium and alloys are clearly demonstrated. The improved corrosion resistance of Timminco's super and ultra high purity casting alloys is highlighted for die cast applications. Use of Timminco

fs high

purity magnesium alloys, whether for underground or hot water tank anodes, results in more efficient cathodic protection and thus, longer service life of the protected structure.

Timminco is continuing its leadership in the development and application of high purity magnesium metal and magnesium alloys. Timminco focuses on process optimization and product development in its R & D strategy in order to fulfill the Company's commitment to the highest standards of quality, technology and customer service.

ACKNOWLEDGEMENTS

Thanks are due to Mr. L.S. Ball for providing statistical data on process capability. The authors are grateful for helpful discussions with Dr. D.J. Zuliani. We would like to thank the senior management of Timminco Metals for permission to publish this paper.

REFERENCES

1. L.M. Pidgeon and W.A. Alexander "Thermal Production of Magnesium - Pilot Plant Studies on the Retort Ferrosilicon Process", Trans AIME. 159, (1944), 315-351.

2. L.M. Pidgeon "Thermal Production of Magnesium", Trans.Can.Inst. Mining Met.. XLIX, (1946), 621-635.

3. D.M. Peplinski, "Factors Affecting Magnesium Reduction by the Pidgeon Process at Dominion Magnesium", Proc. of 1966 AIME Annual Meeting. Feb 27 - Mar 3, 1966.

4. A. Froats, "Pidgeon Silicothermic Process in the 70's", Proc. TMS Light Metals. 109th AIME Annual Meeting. Feb 2 4 - 2 8 , 1980, 969-979.

5. D.J. Zuliani, "Quality Assurance in Timminco's Magnesium Production", Proc. 45th World Magnesium Conference. IMA, Washington, 1988, 72-87.

6. D.J. Zuliani, "The Improved Corrosion Resistance of Ultra High Purity Magnesium Alloy Castings", Proc. 67th AGARD Conference. Mierlo, Netherlands, October 2-7, 1988.


REFERENCES

7. E.F. Emley, Principles of Magnesium Technology. Pergamon Press, New York, N.Y., 1966, pp 687-693.

8. J.K. Aune, "Minimizing Base Metal Corrosion in Magnesium Products. The Effect of Element Distribution (Structure) on Corrosion Behaviour", Proc. 40th World Magnesium Conference. Toronto, 1983.

9. K.N. Reichek, K.J. Clark and J.E. Hillis, "Controlling the Salt Water Corrosion Performance of Magnesium AZ91 Alloy", SAE Technical Paper 850417. Society of Automotive Engineers, 1985.

10. J.E. Hillis, K.N. Reichek and K.J. Clark, " Controlling the Salt Water Corrosion Performance of Magnesium AZ91 Alloy in High and Low Pressure Cast Form", Recent Advances in Magnesium Technology. Proc. of the AFS/CMI Conference, June 25-26, 1985, City of Commerce, California, 87-106.

11. J.E. Hillis and K.N. Reichek, "High Purity AM60 Alloy. The Critical Contaminant Limits and the Salt Water Corrosion Performance", SAE Technical Paper 860288. Society of Automotive Engineers, 1986.

12. C. Sheldon Roberts, Magnesium and its Alloys. John Wiley and Sons, Inc., 1960, 196.

13. 0. Osborn and H.A. Robinson, "Performance of Magnesium Galvanic Anodes in Underground Service", Corrosion. (April 1952), V.8, n.4, 114-129.

14. Private communication with the Curtin University of Technology, Perth, Australia.

15. H.C. Hughes, A.M. Peek and T. Pyle, "Cathodic Protection in Domestic Hot Water Storage Tanks", Paper Presented at the 28th Australasian Corrosion Conference, Perth, Australia, Nov. 1988.

19

Influence of the microstructure on the corrosion resistance of Mg-Al based alloys

F. Lefebvre Centre d'Etudes Nucteaires de Grenoble, Grenoble Cedex, France

G. Nussbaum Pechiney Electrome'tallurgie Laboratoire, d'Electrothermie de Chedde, Passy, France

The detrimental effect of cathodic impurities (Fe, Cu, Ni) or the presence of chloride and fluoride particles on the corrosion resistance of magnesium and its alloys has long been known. This has resulted in the use of high purity alloys and fluxless casting procedures by industrials.

In this paper we propose a tentative explanation for the great corrosion resistance of high purity Al-containing alloys.The corrosion mechanisms of AZ91 (Mg-9%Al-l%Zn) alloy is discussed in terms of solid solution and B - M g ^ A l ^ precipitates effects.

New address : *Centre d'Etudes Nucleaires de Grenoble, 85X, 38041 Grenoble Cedex, France ** Pechiney Electrometallurgie, Laboratoire d'Electrothermie de Chedde, 74 190 Passy, France.


INTRODUCTION

With a density of 1.74, Magnesium is the lightest structural metal. Its mechanical properties, great machinability and castability could make it a serious competitor for Aluminium and Zinc alloys as well as for plastics in a great number of applications.

However, it is also the most anodic of all structural metals, with a standard electrochemical potential of-2.4 V to the Hydrogen electrode [1]. Only Li and Na are more anodic. Furthermore, a layer of Mg(OH>2, also containing Mg oxyde and carbonate, forms on a surface

of Mg freshly exposed to natural or aqueous environment which is not protective[l]. For these two reasons, Mg and Mg based alloys are sensitive to corrosion and because of their anodic nature, especially to galvanic corrosion.

Injudicious joining of Mg with Al, Zn or steel parts will therefore lead to rapid dissolution of Mg. Most designers know how to deal with this problem. A more tricky phenomenon is microgalvanic corrosion : when even small cathodic and active particles are present on the surface of Mg parts, the corrosion rate of the alloy is increased up to several orders of magnitude. The origin of these particles can be twofold : high content of elements such as Fe, Ni or Cu in the Mg melt during elaboration of the alloy (these elements present almost no solubility in Mg in the solid state) Fig. 1, or contamination of the surface of the part during the last steps of the production process (extrusion, sawing, shot peening or blasting ...). Such an acute sensibility to micro-galvanic corrosion is equally devastating for the various Mg based alloys since no great difference exists between their standard potential. This has well been documented in [1] and [2] and has driven the search for procedures that allow the elaboration of high purity alloys on an industrial scale. These alloys indeed present a high corrosion resistance.

Corrosion r a t e

(mcd)

300 + >• NI

100 + •o Cu

50 100 150 ppm Fe NI

0 1000 2000 3000 4000 5000 I ppm Cu

6000

Figure 1 : influence of the presence of cathodic impurities on the corrosion rate of Mg.


It is also well known that, when alloys that present very low levels of cathodic impurities are used, the corrosion behavior of Mg parts varies greatly from one alloy to the other, i.e. that the contents in major alloying elements can greatly influence the corrosion resistance of the products. From this point of view, Al and Y seem to be the most promising elements [ 1-4-5].

The objective of this paper is to determine the effect of up to 9 wt% Al on the corrosion resistance of Mg and relate it to the changes induced on the microstructure of the material. We will therefore first describe the effect of Al in solid solution on the corrosion kinetics of Mg. Then, we will show the role of 6 precipitates on the corrosion pathways. A corrosion mechanism will finally be proposed for the alloy AZ91 and discussed owing to the results obtained by T.J. Warneretal at the CRV, reported in [6,7].

MATERIALS AND EXPERIMENTAL PROCEDURES

Three alloys were elaborated at 7 0 0 ^ in a BN-coated mild steel crucible under a Ar + 2 % SF6 gas flow, starting from the same Mg ingot, with the following chemical composition (wt%):

- Mg-7.5% Al ; - AZ 91 : Mg-8.1% Al-0.6% Zn-0.25% Mn. - Mg-40% Al ; (corresponding to the 6 precipitates composition)

The Mg samples and both alloys contain less than 40 |Xg/g Fe, 10 }ig/g Ni, 50 |ig/g Cu and 300 |ig/g Si. The Mg ingot also contains 430 jig/g Mn and less than 100 jig/g Al. All the alloys can then be considered as high purity grades.

After elaboration and casting in an iron mold, Mg-7.5% Al and AZ 91 were extruded into 15 mm bars (extrusion ratio - 1 0 , temperature 300^) and heat treated to :

- T4 temper (solid solution): 8 hrs at 390<€, 8 hrs at 400<€ and 8 hrs at 4 1 0 ^ + water quenching. A typical micrograph for a T4 treated specimen is shown in fig. 2a. Al is present in solid solution in the matrix. Si and Mn have formed precipitated phases. - T6 temper (Artificial ageing): T4 +16 hrs at 200

c€.

The microstructure of Mg-7.5A1 and AZ91 alloys treated to the T6 temper has been described in [8] and [9] respectively. A typical micrograph is shown in fig. 2b. It is composed of a Mg matrix with 2-3 % Al in solid solution and M g ^ A l ^ precipitates, quite homogeneously dispersed in the matrix, under 2 types of morphology :

- fine platelets inside the grains : continuous precipitation, - coarser lamellaes that have pushed the grain boundaries while growing according to a conventional discontinuous precipitation process.

The corrosion test that has been used is derived from the MEL procedure : immersion for 24 hours in a 5 % NaCl solution buffered to pH - 11 with Mg(OH)2« The weight loss is measured

and reported in mg/cm^ day (mcd).

Polarization curves were drawn in a 5 % NaCl solution buffered to pH ~ 11 with Mg(OH)2- The

use of an aerated or de-aerated solution was found of minor importance.


• * '\ *

• • •

* *

*

•?

F i g u r e d : Typical microstructure of AZ91. In the T4 temper most B phases are dissolved while Mg2Si and (AI.Mn.Fe) precipitates are randomly scattered on the solid solution . In the T6

temper B phases are formed, cellular on grain bounderies, fine and continuous in the grains.

AZ91 in the T4 temper 2a , 50 urn

| A Z 9 1 in the T6 temper 2b , , 5 0


EFFECT of Al in SOLID SOLUTION

Weight loss measurement

3 solid solution treated samples of each material were tested for 24 hrs. The results are:

- M g : 50± lOmcd;

- Mg-7.5% A l : 2.2± 0.2 mcd ; - AZ 91 : 1.5±0.5 mcd.

The main point of these results is the very beneficial effect of Al in solid solution on the weight loss of Mg. The difference between the two Al-containing alloys reveals the influence of Zn, also present in solid solution, and Mn. Addition of Mn to Mg-Al alloys reduces the effects of microgalvanic corrosion [1,10,13] by eliminating pure Fe particles. Because of the very low content in Fe of the Mg used for these experiments, we believe that the presence of Zn in solid solution in the alloy has a beneficial effect on its corrosion resistance.

Corrosion potential and polarization curves

The open circuit corrosion potential Eo of Mg and Mg-7.5% A1-T4 samples was measured in the same solution (5 % NaCl + Mg(OH)2>. Both potential were found to evolve in the opposite

sense with immersion time (figure 3), leading to an equilibrium potential significantly lower for Mg-7.5% Al than for Mg (-1710 mV/SCE vs - 1630 mV/SCE respectively).

Ecor(mV/SCE) Corrosion potential evolution with time : Influence of aluminium in solid solution on the surface reactivity of Mg

- 1 6 0 0 T

- 1 6 5 0

- 1 7 0 0 +

- 1 7 5 0 t

- 1 8 0 0 4-3 0 60 90 120 150 180

M g - A I 9 % T 4

2 1 0

t(mn)

2 4 0

Figure 3 : Corrosion potential in 5 % NaCl + Mg(OH)2.Mg and Mg-7,5%A1 show an opposite

evolution with time, expressing a different interaction between the sample surface and the electrolyte.

The polarization curves for the same samples are presented in figure 4 : Both materials are very reactive and present large current densities for small polarizations. The anodic branches are not linear for either material suggesting that several reactions superposed control the dissolution, whereas on the cathodic branches a linear domain can be found and associated to a modification of the water reduction kinetics.


E (mV/s.c.e.) - 8 0 0 T

Polarization curves Mg and Mg-AI9% T4

Mg .

-1 000 t

- 1200 t

- 2000 t

- 2400

-2200 t

-1 800 t

-1 600 f

-1400 +

Mg-AI9% T4

Log I i I ( m A / c m 2

0,01 0,1 1 10 100

Figure 4 : polarization curves of Mg and Mg-7.4% Al T4 samples in a 5% NaCl + Mg(OH)2

solution. The anodic branches of both materials are similar. The presence of Al in solid solution decreases the kinetics of decomposition of water.

Discussion : Role of Al in solid solution

Adding 7.5 % Al in solid solution to Mg has resulted in a very large increase of its corrosion resistance and slight modifications of its electrochemical properties:

- the corrosion potential and the polarization curve are characteristic of reactive materials in both cases - the non-linearity of the anodic branches of the polarization curve can be interpreted as the superposition of several phenomena controlling the corrosion of both materials ; furthermore, there is no difference between the two anodic curves that could account for the increase of the corrosion resistance - on the other hand, the opposite evolution, with time, of the corrosion potential of both materials expresses a significant modification of the Mg surface reactivity with the addition of Al. - the cathodic branch of the polarization curve of the Mg-7.5%A1 samples is shifted toward smaller current densities compared to Mg, indicating that the kinetics of the cathodic reaction, i.e. decomposition of water, is slower on this alloy than on pure Mg. Such an effect can be understood as a modification of the charge transfert kinetics at the surface of the immersed sample.

Owing to these first results, the large increase of the corrosion resistance of Mg with the addition of Al in solid solution can't be attributed to the existence of different electrochemical reactions involved in the dissolution process but rather to modifications of the surface reactivity. Such an approach is consistent with the results reported by T.J. Warner etal. [7] suggesting that the increase in the corrosion resistance of the Mg-Al alloys could result from a change in their surface microstructures : on the basis of angle resolved XPS data, they calculated an increase in


the thickness of the unattacked surface layer from approximately 60 A to 80 A when 9% Al was present in the alloy. TEM plane view observations [7] also indicate that the surface evolves from a nearly amorphous layer of Mg to a layer containing fine (<10 nm) crystallites of Mg-Al alloys. Does this account for the slower penetration of CI" ions in Al-containing alloy surface product layers as observed in [6] ? And consequently for the increase in the corrosion resistance we have measured ? Observations reported in [6] and [7] were performed on samples at the corrosion initiation stage. They nevertheless appear to be in good agreement with the weight loss measurements and the electrochemical tests. Whatever the observation scale is, the presence of Al in solid solution in the matrix decreases the dissolution rate of the material in the electrolyte.

ROLE of M g 1 7A l 12 PRECIPITATES in the CORROSION PROCESS

Weight IQSS measurements

Samples treated to the T6 temper were tested for their corrosion resistance in the 5 % NaCl + Mg(OH)2 solution. The weight loss is :

- Mg-7.4% Al T6 : 0.5 ± 0.2 mcd - A Z 9 1 T 6 : 0 . 5 ± 0 . 2 m c d

Two observations can be made :

- the alloys treated to the T6 temper present the same corrosion resistance in this solution, considerably larger than that of Mg, - the T6 treated samples present a higher corrosion resistance than the T4 treated specimens.

Corrosion potential and polarization m m s

The open circuit potential Eo for Mg-7,5% A1-T6 samples was measured in a 5 % NaCl + Mg(OH>2 solution. It was found to be significantly more noble than that of the Mg-7,5% A1-T4

samples (-1620 mV/SCE vs - 1710 mV/SCE) and comparable to that of Mg (-1630 mV/SCE). It had the same evolution with time than pure Mg.

To investigate further the role of the B - M g ^ A l ^ phase in the corrosion process of Mg-Al alloys, the corrosion potential of Mg-40 % Al was also measured, in the same solution. It was foud to be - 1440 mV/SCE, whereas that of the matrix, which contains about 2% Al in solid solution was similar to that of pure Mg (- 1620 mV/SCE). These results are in good agreement with those reported in [10].

The polarization curves of Mg, Mg-7.5% A1-T4 and Mg-40% Al samples are presented in figure 5. Mg and Mg-7,5% A1-T4 appear to be much more reactive than the 6-phase since a small potential variation induces on these two materials a large current density increase in opposition to the precipitates behaviour on which the water reduction is very slow.


Log I i I

( i iiA/cm2)

E (mV/sce)

Figure 5 : polarization curves for the B-phase, Mg and Mg-7.4% Al T4 materials in a de-aerated, 5% NaCl + Mg(OH)2 solution. The 6- precipitates appear as very slow cathodes.

Initiation of corrosion

Initiation of corrosion was studied by optical microscopy on Mg-7.5% Al samples after several short immersions in the corrosive solution.

With respect to the observation scale, initiation seems to occur randomly on the T4 treated samples, while it always happens at the interface between cellular 6 phases and the matrix on the T6 ones (figure 6).

A previous study on as-cast AZ91 samples, showing large chemical heterogeneities in the grains due to solidification dendrites, had shown initiation to occur inside the grains [12].

It seems therefore possible to correlate corrosion initiation firstly to coarse chemical heterogeneities and secondly to 6 precipitation.

M g

M g 1 7 A I 1 2

M g - 7 , 5 % Al T4


After 1 mn Immersion After 5 mn immersion

After 15 mn Immersion After 25 mn Immersion

Figure 6 : The initiation of corrosion is followed using alternate period of immersion and observation on a polished sample.


Propagation of corrosion

Propagation of corrosion also appears to be very different on T4 and T6 samples. On T4 samples corrosion essentially occurs through large pits digging the surface from place to place while it spreads slowly on the surface of T6 samples.

On the other hand, even if it had no effect on polarization curves, aeration of the solution has a great impact on both tempers, and considerably increases the corrosion rate. In a previous work, on pressure die cast AZ91 a significant increase of the corrosion rate with the aeration of the solution had been pointed out (figure 7) [13]. Such different behaviors, of electochemical characteristics and effective corrosion rate in aerated solution gives further evidence of the existence of several reactions controlling Mg dissolution.

Corrosion kinetics of Pressure Die Cast AZ91

cm3/cm2

Figure 7 : kinetics of corrosion of AZ 91 samples in the T4 and T6 temper. Replacement of the electrolyte corresponds here to aeration of the solution. We then notice the influence of oxygen on the corrosion rate of Mg alloys.

Discussion : influence of the microstructure on corrosion

Thermal treatments, and therefore the microstructure, appear to have a great influence on the corrosion patterns and rates of Mg-Al alloys :

- on T4 samples, corrosion has the same pattern as pure Mg but is much less violent; this was attributed to the lower reactivity of the matrix when it contains Al towards saline solutions; - on T6 samples, corrosion is decreased again and has a completely different aspect: it spreads slowly on the surface.

electrolyte replacement


Optical observations show corrosion to start in the vicinity of the 6 precipitates. TEM cross-sectional observations of the corroded near surface of rapidly solidified Mg-A18% with (3 intergranular precipitation reported by T.J. Warner et al. [7] are consistent with these results. SIMS analysis of the oxidation layer of the Mg-A13% sample, after immersion, also shows preferential penetration of CI" ions in the B-rich areas [7].

However, because of their low cathodic activity, the current distribution near the 6 phases will impose a decrease in the oxidation reaction rate. Moreover, the propagation will develop preferentially on the surface of the sample where the aeration is maximal. Such a scheme is consistent with the drastic slow down of corrosion propagation near the intergranular B phases reported in [14-15].

We can then assume that the localisation of the attack near the B phases combined with their low cathodic activity is the explanation of the corrosion resistance improvement observed in the T6 temper compared to T4.

CONCLUSION

When magnesium alloys are completely homogeneised and free from cathodic impurities, their corrosion behaviour is controlled by their microstructure. In the case of Mg-Al alloys the influence of two major parameters has been designated and discussed : the presence of Al in solid solution and the presence of a fairly homogeneous dispersion of Mg|yAlj2 precipitates.

The presence of aluminium in solid solution decreases the reactivity of the alloy. Electrochemical measurements show significant surface evolutions due to the presence of aluminium, pointing out some changes in the kinetics of the mechanisms that govern the dissolution. Various reasons for these changes may be invoked such as an increase of the dissolution energy of magnesium atoms or the formation of a protective aluminium rich layer at the surface of the alloy counteracting the dissolution process. The use of surface characterization techniques such as ESCA or SIMS to study this surface reactivity appear to be of great interest.

Artificial ageing treatments, which induce the formation of M g j y A l ^ precipitates, have also a

significant effect on corrosion. Electrochemical measurements, on a Mg-A140% alloy, have shown that their corrosion potential is much more noble than that of pure magnesium but that they produce a very low current density when polarised. Thus, it is possible to explain owing to the electrochemical results that corrosion initiates at the interface between the precipitates and the matrix where there is a large thermodynamical driving force, and that it propagates slowly on the surface of T6 treated samples because of the low cathodic activity of B precipitates spread all over the grains.

ACKNOWLEDGEMENT

The authors wish to thank F. Saillard, E.Laclau and P. Bridot for their contribution to the experimental part of this work. Invaluable discussions with N. Thorne, T. Warner, D. Duly, G. Regazzoni, K. Nisancioglu, P. Tsakiropoulos, L. Robbiola and Pr. Fiaud are also acknowledged.


REFERENCES

[1] E.F.Emley, Principles of Magnesium technology, Pergamon Press (1966) [2] J.D.Hanawalt, C.E. Nelson, J.A.Peloubet, Trans.AIME 147 (1942) p.273 [3] O.Lunder, T.K.RAune, N.Ass.Corr.Eng.42£5), 1987, p.291 [4] O.Lunder, proc.Corrosion 88, St-Louis (Missouri, USA), 1988 [5] F.Hehmannn, Ph.D. thesis, Inst.fiir Metallkunde Stuttgart, 16 sept. 1988 [6] T.J.Warner, N.A.Thorne, G.Nussbaum, W.M.Stobbs, to be published [7] TJ.Warner, N.A.Thorne, H.Dunlop, to be published in Applied Surface Science [8] D.Duly, Memoire de DEA, Universite de Grenoble, Sept. 1990 [9] J.B.Clark Acta Met, voL16 (1968). p. 141 [10]. O.Lunder, J.E.Lein, T.K.Aune, K.Nisancioglu, Corrosion, sept 1989.45(9). pp 741-748 [11]. T.Beljoudi, Memoire de DEA, University Paris VI, septembre 1991 [12]. Unpublished Pechiney results [13]. F.Lefebvre, La corrosion des alliages de magnesium, Internal Pechiney report. [14]. T.K. Aune, O.Lunder, K.Nisancioglu, Proc. 21st Annual Technical Meeting of the International Metallographic Society, Toronto, Ontario, July 1988. [15]. K.Nisancioglu, O.lunder, T.K.Aune, Proc. I.M.A. Congress, Cannes, May 1990

31

Magnesium plaster cast prototypes versus diecastings — a comparative evalution of properties

N. Fantetti, M.O. Pekguleryuz, M.M. Avedesian Institute of Magnesium Technology, Inc., Ste-Foy, Quebec, Canada

ABSTRACT

The demand for magnesium prototypes is growing very rapidly with the increasing number of new applications for magnesium diecastings. Plaster mold casting is becoming more popular amongst all other prototyping methods because it can produce castings with thin walls, smooth surface finish and close dimensional tolerances. However, there is very little quantitative data published which compares the properties of the plaster castings with the eventual diecast parts.

In this study, tensile, impact, fatigue, and microstructural properties of plaster cast and diecast alloys AZ91 (E and D) and AM60B are reported as a first phase of research being conducted at the Institute of Magnesium Technology (ITM). Comparative data on corrosion resistance of plaster cast and diecast plates are also reported.

KEYWORDS

Plaster, casting, diecasting, magnesium, prototypes, tensile, corrosion, fatigue, impact.

INTRODUCTION

Developments in the 1980's paved the way to a plaster mold casting process for magnesium alloys where the risks for a reaction between the molten metal and the molding plaster were reduced (1). More recent in-house developments have made plaster mold casting very safe and it is now used routinely for casting prototypes for new diecast applications.

Thin walls and smooth surface finish are the two most important attributes of the plaster casting prototyping process (2-3). However, in order to make the prototype properties representative of the eventual magnesium diecastings, the plaster mold castings must not only meet the dimensional specifications but also reproduce as closely as possible the characteristics of the diecast part.

Because of the dissimilar process parameters and molding material used in plaster and diecasting, the associated cooling rates may differ by as much as 2-3 orders of magnitude (4). This results in major microstructural differences. The plaster mold castings produced by conventional methods generally have a coarser grain structure and can be prone to shrinkage porosity. This explains the general characteristic that plaster cast parts have lower mechanical properties than diecastings.


Another characteristic of plaster mold castings for which very little quantitative data are available is corrosion resistance. This property is not only affected by the chemical composition but also by the microstructure (5). The objective was to generate basic knowledge on the corrosion resistance of plaster mold castings versus diecastings.

Several prototypes have been produced at the Institute (ITM) using plaster and sand casting among which are automobile doorhandles, variable speed transmission pulleys for snowmobiles, turbo charger wheels, transmission cases, camera housings as can be seen in Figures 1 and 2.

Figure 1 - Fixed pulley (snowmobile Figure 2 - Transmission case, mobile transmission), steering column pulley (snowmobile trans-support bracket, vacuum cleaner mission), turbocharger wheel, fan, camera housing (welding bearing housing, hardware cell), doorhandles. appliances.

Better knowledge of the properties of magnesium cast prototypes will help avoid unfair rejection of magnesium for new diecast magnesium applications.

EXPERIMENTAL PROCEDURE

Plaster Molds

Fabrication of Plaster Cast Test Specimens and Corrosion Plates

The plaster molding process developed and used at the ITM is proprietary and can be described generally as follows:

• a slurry containing plaster, reinforcing particles, inhibitors and water is mixed; • the slurry is poured into the pattern; • after solidifying, the mold is stripped; • the molds are dried according to a cycle which is a function of their weight; • before metal pouring, the molds are purged with a gas mixture.

Special attention to materials, gases and process parameters is critical in order to avoid a reaction between the molten magnesium and the plaster molds. Such a reaction can take place if the residual moisture content is not reduced below a critical limit or if contaminants are present in the raw materials used during the mold fabrication process.


A typical plaster mold used for casting the test bars and the resulting test specimens are shown in Figures 3 and 4, respectively. The test bars consist of a tensile specimen (6.3 mm diameter), an impact specimen (10x10x54 mm) and a flat tensile ( 3 x 6 mm) in the reduced section. Figure 5 shows the plaster cast corrosion plates with gating system. The dimensions of each plate are 100x140x3 mm.

Figure 3 - Tensile and impact test bar Figure 4 - Plaster cast tensile and plaster mold. impact test bars.

Figure 5 - Plaster cast corrosion plates with gating system.

Melting and Casting Parameters

Melting was carried out using a 10 kg capacity nickel-free stainless steel crucible under SF6(1%) - C 02 atmosphere using an electric resistance furnace. Only primary ingots were charged.

Grain refining treatment consisted of submerging hexachloroethane tablets into the melt using a bell plunger. This method is documented by several authors as an established procedure for grain refining in sand casting foundries (6). The effect of this technique for plaster cast magnesium alloys was studied in earlier work (7).


Chilling consisted of placing steel inserts approximately 5 x 5 x 10 mm in the plaster mold in the middle of the test specimens. The objective was to increase heat transfer and promote directional solidification (4).

Each casting was poured at 700 °C directly from the crucible to avoid the temperature drop caused by metal transfer. Mold temperature was kept at 205 ± 6 °C.

Heat Treatment

Heat treatments were conducted in an electric resistance oven with forced air circulation. Solution treatment (T4) and solution and ageing treatment (T6) were carried out according to ASTM B661-87. Heat treatment parameters are listed in Table I. Again the effect ot these treatments is well documented for sand castings (8,9) but little has been published for plaster cast magnesium alloys.

Table I - Heat Treatment Parameters According to ASTM B661-87

Designation Temperature Time °C hours

T4 413 20

T6 169 16 (T4 above followed by)

Diecasting

Equipment and Specimens

The diecast specimens were produced at the ITM using a 250 tonne Freeh hot-chamber diecasting machine equiped with a Dataprocess monitoring system and a Thermocast heating and cooling unit for die temperature control. Two dies donated to the Institute by Magnesium Corporation of America (Magcorp) were used for producing the diecast test specimens and corrosion plates. The resulting parts are shown in Figures 6 and 7. The dimensions and geometry of the diecast specimens are similar to that of the plaster cast specimens.

Figure 6 - Diecast test specimens. Figure 7 - Diecast corrosion plates.


Diecasting Parameters

Table II summarizes the diecasting process operating parameters used during the production of the diecast specimens (both test specimens and corrosion plates).

Table II - Diecasting Process Operating Parameters

Operating parameters Value

Metal temperature

Filling time

Injection speed

Freezing time

Injection pressure

Die temperature

Cycle time

Shot weight

620- 650 °C

0.10-0.18 ms

650 - 800 mm / s

4 s

120 bars

260 - 270 °C

20 s

100 g (test specimens) 170 g (corrosion plates)

Chemical Composition

The chemical composition of the melts used throughout these experiments is given in Table III.

Table III - Average Chemical Composition of the Melts (weight %)*

Alloy Al Mn Zn Si Cu Fe Ni

AZ91E 8.4 0.26 0.70 0.002 0.001 0.002 < 0.001

AZ91D 9.45 0.15 0.75 0.020 0.0023 0.0019 0.0020

AM60B 6.25 0.24 0.0017 0.082 0.0019 0.0019 0.0010

* From spectrograph^, AA and ICP analyses.

Testing Procedures

Tensile Testing

Minimum machining was done on the plaster cast test bars to remove burrs and flash. A computer controlled Instron tensile tester was used for all tensile tests according to ASTM B557-84.

Impact Testing

Impact tests were done using a Charpy Pendulum on unnotched specimens at room temperature according to ASTM E23-87.


Rotating Bending Fatigue Testing

Plaster cast unnotched fatigue test specimens were machined from 1.25 mm diameter test bars. Diecast tensile specimens were used in the as cast condition for the fatigue tests (only the extremities were machined to fit the fixtures). The tests were done at 8,000 rpm with various loads using a conventional R.R. Moore machine.

Corrosion Testing

Plaster and diecast corrosion plates were trimmed, washed with acetone and weighed prior to corrosion testing (5). They were exposed for 200 hours to salt spray corrosion in accordance with ASTM Bl 17-85.

RESULTS AND DISCUSSION

Tensile Properties

The results of tensile testing are shown in Table IV. All testing carried out at 20 °C.

Table IV - Average Mechanical Properties at 20 °C

Alloy-condition

Number of samples

U.T.S. MPa

Y.S. MPa

E (%)

Diecast AZ91D 21 245 153 5.9

Plaster cast AZ91E (with chill, T6) 16 213 124 3.0

Plaster cast AZ91E (without chill, T6) 8 165 118 1.4

Diecast AM60B 19 238 117 13.2

Plaster cast AM60B (with chill, T6) 10 212 71 9.1

Plaster cast AM60B (without chill, T6) 12 163 70 5.9

These results show that for the ultimate tensile strength, the plaster cast AZ91E test bars (with chill) and T6 reach 85% of the value obtained for the diecast test bars. Plaster casting without chilling produces a lower U.T.S. value, in the range of 70% of the diecast value. For the same alloy, the yield strength obtained by plaster casting is approximately 80% of that of diecastings while the elongation is considerably lower at 3.3% as compared with the value achieved by diecasting (5.9%).

For alloy AM60B, the results are very similar. It should be pointed out that AM60B alloy is not usually gravity cast. It was included in our test program for completeness only. The ultimate tensile strength of the plaster castings (T6 - chill) exceeds 85% of the values obtained by the diecast specimens. However, for the same conditions, the yield strength is only 61 % of that obtained by diecasting AM60B while the elongation values are approximately 70% of the diecast values.


The increased localized cooling rate and the directional solidification caused by the chills have a significant effect as seen in Table IV. These differences in tensile properties of diecast and plaster cast (T6) samples are largely due to differences in grain size caused by different solidification and cooling rates. Figures 8,9 and 10 show typical microstructures of diecast and plaster cast AZ91 (T6) samples with and without chills, respectively. Grain sizes of these samples were determined using a LECO image analysis method. As seen in Table V, diecast samples are very fine grained (9 -11 um) while plaster cast samples with no chills possess coarse grains of 200 - 230 um. The use of chills has reduced the grain size of plaster cast samples to 150 - 180 um.

The mechanism that explains the increase of tensile properties and the loss of ductility with decreasing grain size is related to the effect of grain boundaries on metal deformation. In polycrys-talline metals, grain boundaries act as obstacles to slip dislocations. Hence, in fine-grained materials a much larger applied stress is needed to cause deformation than in coarse-grained materials.

Normally, loss of ductility is also associated with coarse grain size because dislocation pile-up is more severe in coarser grains. Earlier work (7) has shown that a solution annealing heat treatment will produce higher elongation values than solution annealing and ageing (T6) treatment. The former should be used in cases where elongation is important.

Table V - Grain Size Measurements

Casting condition Grain Size Range pm

Plaster (with chilling) AZ91E 120-180

Plaster (without chilling) AZ91E 200-230

Diecast AZ91D 9-11

Figure 8 - Diecast AZ91D (As cast).

Figure 9 - Plaster cast AZ91E (T6 with chills).


Figure 10 - Plaster cast AZ9IE (T6 without chills).

impact Properties

The results of Charpy impact tests (unnotched) are shown in Table VI.

Table VI - Charpy Impact Tests (Unnotched)

Alloy-condition

Number of samples

Impact Energy J

Diecast AZ91D 15 5.0

Plaster cast AZ91E T6-chill 10 5.7

Plaster cast AZ91E-T6 10 4.5

Plaster cast AZ91E As Cast 10 3.0

Diecast AM60B 15 8.3

Plaster cast AM60B T6-chill 12 16.6

Plaster cast AM60B-T6 12 11.8

Plaster cast AM60B As Cast 10 6.9

The general trend shown by these tests is that the impact energy of the plaster cast specimens is either equal or superior to that of the diecast specimens for both magnesium alloys. This can be explained by the porosity present in diecast samples.


As seen in the micrographs of Figures 9 and 10, the directional solidification has produced sound test specimens free of porosity while the diecast impact specimens possess porosity which reduces the resistance to the propagation of the flaw. Density measurements carried out on plaster cast and diecast AZ91 impact specimens revealed that plaster cast samples have a higher density (1.79 g/cm3) compared to the diecast specimens (1.71 g/cm3). This indicates a higher porosity for diecast specimens.

A SEM micrograph of a diecast fracture surface given in Figure 11 reveals the presence of a large central shrinkage pore while the plaster cast specimen (see figure 12) has negligible porosity.

Figure 11 - Impact surface of Figure 12 - Impact surface plaster of diecast AZ91D. cast AZ91E (T6-chill).

Rotating Bending Fatigue Properties

The fatigue test results are presented in Figures 13 and 14.

in

# no rupture

Number of cycles

Figure 13 - Rotating bending fatigue tests, diecast AZ91D.


no rupture

Number of cycles

Figure 14 - Rotating bending fatigue tests, plaster cast (T6-chill) AZ91E.

Fatigue strength for diecast specimens is in the range of 83 to 90 MPa while it is in the range of 54 to 61 Mpa for the plaster cast - T6 chilled specimens. In both cases, these values are approximately one third of the ultimate tensile strength and agrees with earlier findings (10) where fatigue strength was found to be between 0.20 - 0.43 of the U.T.S. for cast magnesium alloys.

Corrosion Testing

The appearance of typical plaster AZ91E and diecast AZ91D corrosion plates after 200 hours exposure to salt spray is shown in Figure 15. The corrosion rates are reported in Table VII.

Figure 15 - Appearance of typical diecast AZ91D (left) and plaster cast AZ91E corrosion plates.


Table VII - Average Corrosion Rate for Plaster Cast and Diecast AZ91 Alloy (200 hours)

Casting Process Number of Corrosion Rate plates mg/cm

2-day

Diecast (AZ9ID) 10 0.16

Plaster cast T6

(AZ91E) 6 0.40

These results show that plaster cast plates have a corrosion rate which is close to double that of the diecast plates. Coarse constituents in the plaster castings act as cathodic sites for corrosion while most of the elements are in solution or exist as very fine constituents in the diecast plates (11). The fact that the AZ91E ingot contained slightly less aluminum than AZ91D diecasting ingots may also account for a portion of the difference in rates (12). Several surface treatments are commonly used for cleaning the surface of sand castings and improving the corrosion resistance (11,13,14). However, their effect has not yet been verified on plaster castings.

CONCLUSIONS

Ultimate tensile strength of plaster cast (chill and T6) AZ91E tests bars is in the range of 70 to 80% of the diecast values. The use of T6 heat treatment alone on plaster castings will produce U.T.S. values in the range of 60 to 65% of the diecast. The yield strength values for plaster cast (chill and T6) are approximately 70% of the diecast. Elongation value for plaster cast (chill and T6) is 3% as compared with 5% for hot-chamber diecast AZ91D.

The U.T.S., yield strength and elongation values for plaster cast AM60B alloy have similar percentage values compared with its hot-chamber diecast counterpart.

Impact values for plaster cast AZ91E and AM60B (chill and T6 or T6 alone) are equal or superior to diecast impact properties. This was verified by metallography and SEM analysis that indicated less porosity in the plaster cast specimens.

The corrosion rate was measured at 0.40 mg/cm

2 - day for plaster cast AZ91E (T6) as compared

to 0.16 mg/cm

2 - day for hot-chamber diecast AZ91D.

Fatigue resistance of plaster cast AZ9 IE (T6-chill) is in the range of 54 to 61 MPa as compared to 83 to 90 MPa for the diecast alloy.

Control of metal quality and process variables must be followed for the production of high quality magnesium prototype castings which will meet today's stringent mechanical and physical requirements and eventually match those of their diecast counterparts. Recent advances in the plaster casting process will help enforce the prototyping route for new magnesium applications. The Institute of Magnesium Technology is continuing its investiga-tion into the comparative properties of plaster cast and diecast parts.


REFERENCES

1. "Magnesium Alloy Casting in Plaster Molds", U.S. Patent, number: 4579166, 1986.

2. P.S. Frederick, "Prototyping Magnesium Alloy Castings", Dow Chemical Co., Paper 880511, SAE International Congress & Exposition, 1988.

3. A. Torok and P. Wilson, "Investment Cast Premium Quality Magnesium Alloys", 1

st Society

for Advancement of Material and Process Engineering Conference, August 1987.

4. M.V.Chamberlin and J.G.Mezoff, "Effect of Mold Materials", Dow Chemicals Co., American Foundryman, October 1946.

5. D.L.Albright and C. Suman, "Understanding Corrosion in Magnesium Die Casting Alloys", Paper 880510, SAE International Congress & Exposition, 1988.

6. E.F. Emley, "Principles of Magnesium Technology", Pergamon Press, Oxford, 1966.

7. N. Fantetti, A. Couture and A. Thorvaldsen, "Properties of Magnesium Plaster Castings", Paper 910413, SAE International Congress & Exposition, 1991.

8. C. Brooks, "Heat Treatment, Structure and Properties of Nonferrous Alloys", American Society for Metals, 1982.

9. C. Suman, "Heat Treatment of Magnesium Diecasting Alloys AZ91D and AM60B", Paper 890207, SAE International Congress & Exposition, 1989.

10. V.V. Ogaveric and R.I. Stephens, "Fatigue of Magnesium Alloys". Annual Review of Material Science. Vol. 20, pp. 141-177,1990.

11. K. Nisancioglu, O. Lunder and T. Aune, "Corrosion Mechanism of AZ91 Magnesium Alloy", IMA 47

th Conference, Cannes 1990.

12. O. Lunder, J.E. Lein, T.Kr. Aune and K. Nisancioglu, "The Role of Mg1 7Al12 Phase in the Corrosion of Mg Alloy AZ91", Corrosion Science. Vol. 45, No. 9, NACE, 1989.

13. A. Nassar, "Design Concept of Magnesium Accessory Drive Brackets", Paper 910553, SAE International Congress & Exposition, Detroit, February 25 - March 1, 1991.

14. K.J. Clark, "AZ91E Magnesium Sand Casting Alloy, the Standard for Excellent Corrosion Performance", 43

rd Annual IMA Conference, Los Angeles, 1986.

AKNOWLEDGEMENT

The authors would like to thank Magnesium Corporation of America (Magcorp) for the donation of the dies which were used for the production of the diecast test specimens and corrosion plates.

Primary magnesium production at Becancour

P.M.D. Pinfold Norsk Hydro Canada Inc., Becancour, Quebec, Canada

ABSTRACT

The Becancour Magnesium Plant of Norsk Hydro Canada Inc., employs a unique five (5) stage Production Process, commencing with the acid dissolution of magnesite.

The author examines how the choice of this process influences the chemical quality of the final product, and presents Process Capability Studies for both Pure and Alloy products.

Plans to extend the product range and further enhance Process Capability are also discussed.


INTRODUCTION

Norsk Hydro a.s commenced magnesium production in 1951 and by the mid * 80's had established a 50% share of the Western European market. However, access to 50% of the western world market, in North America, was hampered by a customs duty barrier. (See Fig. 1)

The Becancour plant was conceived to serve the North and South American markets. This paper examines how and why the design of this 45,000 m.t.p.y. plant has evolved, and to what extent the chemistry of the products currently produced satisfies various market requirements.

The reasoning behind other process developments announced since plant start up, and plans to improve process capability, are also discussed.

Fig. I - PRIMARY Mg SHIPMENTS 1990 BY WORLD ZONES

North America 5 0 %

AREA % MARKET SHARE

North America 5 0

Western Europe 2 7

Asia/Oceania 15

Latin America 5

Others 3

Total Shipments 1 9 9 0 : 2 5 2 . 0 0 0 mt SOURCE I.M.A.

Western Europe 2 7 % Asia/Oceania 1 5 %

Latin America 5 %

Others 3 %


PROCESS DESCRIPTION

The production process employed at Becancour, comprises of 5 distinct stages, (See Fig. 2)

These are:

1. Raw material Dissolution and Brine Purifi-cation.

2. Dehydration of the Brine. 3. Electrolysis. 4. Hydrochloric Acid Synthesis. 5. Product casting.

and will be reviewed individually.

Fig. 2 - PRODUCTION PROCESS BECANCOUR Mg PLANT

RAW MATERIAL HANDLING & STORAGE

M g C 0 3

0)

MAGNESITE DISSOLVING {*"

MgCI brine

EVAPORATION PRILLING DEHYDRATION

I

M g C I 2 PRILLS _ J

PL ELECTROLYSIS

ALLOY

©

MOLTEN Mg

j j MATERIALS

FOUNDRY

PURE Mg ALLOY Mg INGOTS

MAGNESIUM METAL STORAGE

SHIPMENT

-HCI-

HCI

HCI

HCI

HCI

SYNTHESIS CI

45

0 ©


STAGE 1

The dissolution process

Much thought and research effort went into esta-blishing a flexible process capable of producing a purified MgCl2 brine, from low cost magnesium bearing raw materials.

The chosen route for Phase 1 of the Becancour Plant is the dissolution of magnesite rock in hot concentrated hydrochloric acid viz:

MgC03 + 2 HC1 = MgCl2 + H20 + Co2 (1)

Common impurities in magnesite include CaO, Si02, Al203, Fe203 together with minor amounts of other non metallics and some heavy metals,

Al, Fe, Mn, Si, P, Cu, Ni are removed by precipitation in basic solution at this stage, where final metal purity is substantially determined. Si02 is practically insoluble in HC1. However, Ca remains in solution along with MgCl2, and this eventually reports in the electrolyte at Stage 3.

The plant design was based on pilot plant studies. Experience to date confirms that capacity calculations were in fact conservative.

STAGE 2

Dehydration

The dehydration process converts a concentrated MgCl2 brine solution to an anhydrous granular feed stock for electrolysis.

Thus, in contrast to many conventional Primary Aluminium Smelters, Becancour has its' own raw material refinery on site. Also, approx 4 m.t. of anhydrous MgCl2 are required to produce 1 m.t. of magnesium, in contrast to approx 2 m.t. of refined alumina per m.t. of aluminium.

Becancour features a second generation design based on 12 years of dehydration production experience, and a continuous R & D program.

The substages of dehydration are:

a) Evaporat ion. b) Prilling. c) Prill Drying.

Evaporation needs no explanation. When the resulting solution is almost supersaturated, it is spun through a centrifuge and cooled, producing a granular

46


product or "prill". The prills are easily handled with conventional bulk handling equipment, but at this stage have the formula:

MgCl2 6H20 (2)

This chemically combined water is removed by drying with hot air and hydrochloric acid gas.

As a result of the dehydration process, the impurity profile of the MgCl2 remains unchanged.

Anhydrous MgCl2 is unfortunately hygroscopic, requiring careful handling techniques.

Performance of the unit has exceeded expectations in both volume and quality. The incorporation of R & D results achieved since the design was frozen for construction, will allow further capacity increases and a reduction in energy consumption to be realised.

STAGE 3

Electrolysis

Electrolysis, in principle employs a NaCl - CaCl2 electrolyte, into which the anhydrous MgCl2 prills are fed at a controlled rate.

The Becancour cell design results from a large and continuous R & D effort, and features a dense electrode pack of non consumable graphite anodes and steel cathodes in a simple cell

1. It combines high amperage and current

efficiency to give the worlds• most productive magnesium cell. Such a performance would not be possible without the anhydrous feed stock. The metal is collected on top of the electrolyte in a separate settling chamber. Power consumption is of the order of 12kwh/kg Mg.

The electrolytic decomposition of MgCl2 releases approx 3mt of chlorine gas for each tonne of Mg metal produced.

Except for pick up of Fe, Na and Ca (from the cathode and electrolyte respectively) impurity levels remain substantially as established in Stage 1 of the process.

Performance of the electrolysis cells has also exceeded expectations. R & D work has indicated that when cell rebuilds are required, a modified design can give a further productivity boost.

47


STAGE 4

Acid synthesis

The chlorine gas from electrolysis is burnt together with a hydrogen containing gas to produce, after absorption in water, concentrated Hydrochloric Acid which reverts to Stage 1 & 2 of the process.

Thus, a closed loop is established for chlorine which incidentally contains the largest HC1 plant in North America.

Acid synthesis is not an area of Norsk Hydro developed technology. The acid synthesis units have performed well since start up and were purchased as a Turn-Key project from a specialist German company.

STAGE 5

Product casting

Currently, the Becancour foundry has 3 production lines. (See Fig. 3)

Fig. 3 - FOUNDRY FLOW CHART BECANCOUR M g PLANT

Mg FROM

ELECTROLYSIS PLANT

PREHEAT

FURNACE

ALLOYING

FURNACE

Mg ALLOY

CASTING

FURNACE

PURE Mg

CASTING

FURNACE

DOUBLE ALLOYS SMALL PALLETIZING ^ TO

CASTING BELT INGOT COOLER & WEIGHING STORAGE

SINGLE

CASTING BELT

Mg INGOT

COOLER

PALLETIZING

& WEIGHING • TO

STORAGE

D.C.

PIT

CUTTING

WEIGHING

TO

STORAGE

ALLOYING

ELEMENTS

BILLETS

"T"BARS


High Purity alloy ingots (8 kg & 12 kg), for the Die Casting Industry are produced on one line, which has a potential to cast all of the current production.

Alloy is prepared in a batch furnace to which cell metal and preheated alloying elements are added. Iron is removed by precipitation with manganese in this furnace, in order to meet the high purity requirements.

The chemical analysis of each batch is verified by the laboratory before being transferred to the casting furnace. This unit is a multichamber gravity settling furnace which, supposing that the batch preparation is well coordinated, allows the casting of ingots for many hours in an uninterrupted run.

Thus, the "Continuous Refining Furnace", which is another product of our extensive R & D program, plays a key role in ensuring the cleanliness of our foundry products.

The other two production lines for Pure metal, share a common casting or "C.R.

M furnace. The ingot machine

has a potential to cast all the current production. Our standard products are 22 kg ingots for aluminium alloying and 17 kg ingots for steel desulfurisation.

A Vertical Direct Chill Casting pit produces T-Bars for the Aluminium industry and large diameter cylinders for nodular iron production. "Hot Top - Level Pour" is employed on these products for which potential capacity is 30,000 m.t.p.y.

PROCESS CONTINUITY

Continuity in electrolysis is vital for economic operation, as in other industries. Buffer storage capacity is in fact installed for purified brine, prills and hydrochloric acid. Provision is made to absorb Cl2 gas for short periods if acid synthesis capacity is not available. Additionally, the electricity supply can be routed from any of three major supply systems should the need arise.

PROCESS CONTROL

All production stages are continually monitored by an advanced computerized control system.

Process Control, Environment Control, Final Product Quality and Technical Support functions are provided by a well equipped laboratory.

A comprehensive Quality Assurance Program has led to a high level of customer satisfaction with Norsk Hydro Canada products and services.

49


METAL MARKETS

Primary Magnesium shipments in U.S.A. and Canada for 1990, by Market Segments, are shown in Fig. 4.

Fig. 4 PRIMARY Mg SHIPMENTS 1990 IN U.S.A. & CANADA,

BY MARKET SEGMENTS

Aluminium 47%

Desulfurization 16%

Total Shipments 1990: 127.300 mt SOURCE I.M.A.

Quickly we can see that about 80% of sales are in pure metal, and alloy applications, (structual) account for somewhat less than 20%


Pure metal

Generally speaking, aluminium alloying requirements are based on the ASTM 9980A specification, whilst the Nodular Iron and Desulfurization markets specify ASTM 9980B.

Becancour Process Capability for these products is shown in Table I.

Table I - Becancour process capability for ASTM 9980 pure magnesium

SPECIFICATION BECANCOUR 9980 A ASTM ASTM PROCESS CAPABILITY STUDY 9980A 9980B MAY 8th-JULY 1st 1990

(752 samples)

ELEMENT % % % MEAN. STANDARD DEVIATION

Al. (max.) 0.05 0.05 0.007 0.003 5.42 Cu (max.) 0.02 0.02 0.0006 0.0003 26.93 Fe (max.) 0.05 0.05 0.030 0.002 4.92 Mn (max.) 0.10 0.10 0.060 0.003 5.49 Ni (max.) 0.001 0.005 0.007 0.0002 1.43 Si (max.) 0.05 0.05 0.002 0.0009 19.04 Zn (max.) 0.05 0.05 0.0021 0.0002 101.76 Mg (by diffe-) 99.80 99.80 99.89 0.006 5.27 rence) (minimum)

The analytical results shown in Tables I & II, were all determined on a spark emission spectrograph and are quoted following a study of the precision and detection limits of this equipment.

Additionally, ASTM 9980A stipulates a maximum level of Na at 0.006% and both 9980A & 9980B limit Pb and Sn to 0.01% each. Background levels of these elements are monitored using an inductive coupled plasma. Precision studies on this technique are not yet completed. However concentration levels of these elements report consistantly as:

Na < 0.003%, Pb < 0.005%, Sn < 0.002%

The Cpk figures indicate that the Becancour process can consistantly satisfy both the ASTM 9980A and 9980B specifications.

In fact, iron levels are typical of electrolytic metal. The "thumb print" for Becancour metal is the rather high, but generally harmless, manganese content. Cu, Si, and Zn are particularly low.


Alloy

Currently, production is directed at high volume, high purity products, which translates into ASTM AZ91D.

Becancour Process Capability for these products is shown in Table II.

Table II - Becancour process capability for ASTM AZ91D alloy

BECANCOUR PROCESS CAPABILITY SPECIFICATION AZ91D NOV. 16th TO DEC. 31st 1990

(2105 samples)

ELEMENT % MIN. % MAX. % MEAN. STANDARD °pk DEVIATION

Al 8.5 9.5 9.0 0.2 0.97 Zn 0.45 0.9 0.65 0.02 2.62 Cu 0.015 0.0007 0.0002 28.3 Fe 0.004 0.0024 0.0004 1.65 Mn 0.17 0.25 0.01 1.70 Ni 0.001 0.0010 0.0001 1.37 Si 0.05 0.006 0.0006 24.9

Die castings account for about 70% of 1990 structural sales in North America, and have shown an average annual growth rate of 18% over the last 6 years.

The capability index exceeds 1.33 for all elements except Al. The low levels of Cu & Si are particularly noteworthy, and are directly attributable to the brine purification effected in Stage 1 of the process.

FUTURE DEVELOPMENTS

Having established a volume operation in Becancour, Norsk Hydro a.s now has a production unit in each of the two largest market areas.

Immediate plans in Canada are to extend D.C. casting applications to include alloy billet production. Alloy extrusions are a natural complement to die castings in the drive by the automobile and transportation industries to reduce vehicule weight.

A further reduction in the Manganese content of pure metal is also planned, in order to accomodate certain specialist applications. A stepwise reduction is possible, depending on the consumption of precipitation reagents. An interim target will be < 300 PPM Mn.


SUMMARY

The Becancour magnesium plant has demonstrated the soundness of the experience which went into its' design.

Metal chemistry has met with wide acceptance in the market place and represents an excellent feed stock for the preparation of high purity alloys.

Considerable gains in productivity and energy consumption are possible as the current Process Optimization Program is supplemented by the application of ongoing research and development results.

The full potential of the Becancour complex will not however be realized, until market growth allows the plant to assume the same economies of scale, as those enjoyed by modern aluminium smelters.

ACKNOWLEDGEMENTS

The author's wishes to acknowledge the unselfish efforts of all Norsk Hydro Canada employees and technical support groups in Norway. Their sustained team effort has ensured the successful startup of a greenfield magnesium complex.

REFERENCES

N. Hoy-Petersen, "From Past to Future", Proc. 47th World Magnesium Conference, I.M.A., Cannes, 1990, 18-23.

53

55

Potential and limitations of rapidly solidified magnesium alloys

G. Nussbaum Pechiney Electrometallurgie laboratoire, d'Electrothermie de Chedde, Passy, France

G. Regazzoni Pechiney, Centre de Recherches de Voreppe, Voreppe, France

H. Gjestland Norsk Hydro, Magnesium Materials Technology, Porsgrunn, Norway

This paper will critically review the main achievements of a 3 year joint research program between Pechiney and Norsk Hydro.

Principles for alloy design were deduced from experiments which led to RS bulk products presenting a density of 1.8, a submicronic grain size and tensile strength as high as 600 MPa, with 2-5 % elongation to fracture.

In NaCl solutions, the corrosion resistance of the alloys of the Mg-Al-Ca system is at least equal to that to AZ 91 E T6 conventional cast alloy. The corrosion rate of AZ 91 + 2 % Ca products is 3 times slower than that of the conventional alloy.

The damage tolerance properties (ductility, toughness and fatigue) seem to depend essentially upon the consolidation conditions.

Finally, the presence of fine and stable dispersoids as well as of rare earths in the alloys increases the elevated temperature properties of the products.

57

Low pressure sand casting of magnesium alloys

J.C. Hogg Rolls-Royce, U.K. (formerly with Norsk Hydro Magnesium, Porsgrunn, Norway)

H. Westengen Magnesium Matrials Technology, Norsk Hydro Magnesium, Porsgrunn, Norway

D.L. Albright Market Development Center, Norsk Hydro Magnesium, Southfield, Michigan, U.S.A.

ABSTRACT

The integrity of a casting depends u p o n the satisfactory complet ion of procedures to insure p r o p e r mel t chemis t ry and t rea tment , meta l transfer, m o l d filling and solidification. This p a p e r r ev iews va r ious aspects of the e q u i p m e n t wh ich is necessary for effective low pressure casting and further demonst ra tes the feasibility of p ressure assisted filling of molds . The descr ipt ion of a series of cast ing trials of s imple shapes is inc luded , as well as reference to more complex par t s . The results ind ica te tha t the i nhe ren t a d v a n t a g e s of cont ro l led l ow p r e s s u r e filling offer possibilities for the produc t ion of p r emium quality castings.


INTRODUCTION

The final quality of a casting depends u p o n the success of a number of important steps in the product ion process. These operations include the following:

Purification of the melt; Selection of the alloying elements and addit ives to refine grain size and

modify microstructure; Transfer of the metal to the feeding system; Filling of the mold; Control of the solidification process.

Considerable progress has been m a d e in recent years to opt imize each of the above aspects of the casting process.

For the m a g n e s i u m casting indus t ry , realization of the full advan tages of strict control of the impur i ty levels, part icularly the copper , iron and nickel contents, has led to rap id deve lopment of certain commercial alloys (1, 2), exemplified by AZ91D and AM60B.

For a luminum-base cast ing alloys, a n u m b e r of m e t h o d s to descr ibe the mel t quali ty has been established. Along wi th gas analyzers and methods to characterize the inclusion content , thermal analysis equ ipmen t is also being ut i l ized in casting facilities (3). From thermal analysis, the microstructure of a casting can be partially forecast For example, the grain size and the degree of silicon modification correlate well w i th s imple cooling curve analysis. In the case of m a g n e s i u m alloys, these methods are still generally confined to laboratory studies.

Significant emphas i s has been pa id to the u n d e r s t a n d i n g of the solidification pat tern of complex castings. Compute r s imulat ion of solidification is n o w becoming a powerful tool in the design of casting molds . With the adven t of more powerful computer systems, the models are gradually being extended to cover the mold filling phase , a long wi th convect ion in the mold cavity, in addi t ion to solving the heat transfer equat ion (4).

A prerequisi te for utilizing such tools to opt imize the casting quality is to develop the practical aspects of the casting process. Especially impor tan t par ts of the process are the transfer of mol ten metal from the crucible to the feeding sys tem and the filling of the molds . Magnes ium and its alloys are k n o w n to have a high affinity for oxygen, and it is manda to ry to avoid turbulence and excessive oxidation dur ing the m o l d filling pe r iod . O the rwise , an apprec iable n u m b e r of cas t ing defects can originate at this stage of the casting process.

In an earlier pape r (5), a complete appara tus for the l ow pressure die casting of magnes ium alloys was described. The basic concepts of this process are adaptable to low pressure sand casting (LPSC). To demonst ra te the feasibility of pressure assisted filling of sand molds , a series of trials involving simple shapes , as well as a complex au tomobi l e inlet manifo ld , w a s conduc ted . Resul ts of these inves t iga t ions are reported below.


LOW PRESSURE FEEDING SYSTEM

A special appa ra tu s for l ow pressure sand m o l d casting of m a g n e s i u m alloys is s h o w schematical ly in Figure 1. The concept po r t r ayed relies u p o n a m e t h o d to transfer mol ten m a g n e s i u m in heated, Cr-Mo alloy steel tubes (6). U n d e r the action of gravi ty , a cons tan t level of l iquid meta l is ma in t a ined t h r o u g h o u t the ent i re sys t em, f rom the m e l t i n g furnace to the e n t r a n c e to the m o l d . This z e ro overpressure level is main ta ined by controll ing the vertical posi t ion of the mel t ing furnace to w i th in p l u s or m i n u s one cent imeter . Since the comple te sys tem of hea ted tubes is a lways filled wi th l iquid meta l , p rob lems wi th oxide format ion within the tubes are el iminated. Also, since the level change in the casting furnace is minimized, erosion of the l inings is marked ly reduced.

VALVE PRESSURIZED AIR

Figure 1 - Schematic d iagram of low pressure sand casting system.

W h e n the casting furnace is pressur ized du r ing filling of the mold , a valve closes the transfer tube connecting the casting furnace a n d the mel t ing furnace. Thus there can be no backflow of l iquid metal . After releasing the pressure , the va lve opens , and meta l flows in to the cast ing furnace to compensa te for the level difference. S imul t aneous ly , the me l t i ng furnace is lifted to adjust the meta l level of the complete system.

All mel t t r ea tment is carr ied ou t in the mel t ing furnace. This inc ludes grain re f inement a n d the a d d i t i o n of m a n g a n e s e ch lo r ide for m a i n t a i n i n g i m p u r i t y control .

In the casting furnace, a vo lume percentage mixture of 0.2% sulfur hexafluoride, 20% carbon dioxide a n d 79.8% d r y air is used to protect the mel t from oxidat ion. Ambient air is used to pressur ize the casting furnace. To insure close control of the pressure, a flow control valve is used.

LIFTING

T A B L E

CASTING F U R N A C E

* MELTING F U R N A C E

CONNECTING

T U B E

C A S T I N G

M A C H I N E

C O N S T A N T

LEVEL

S A N D M O L D SIPHON TUBE SFfi/COo/AIH

M E L T } P R O T E C T I O N |

LEVEL CONTROL


Convent iona l d ry silica sand b o n d e d wi th a chemical b inder w a s used in the p roduc t ion of the sand mo lds a n d cores. The b inder is an alkali phenol ic resin, h a r d e n e d by an al iphat ic ester-base ha rdener . N o inhibi tors w e r e a d d e d to the mo ld ing sand. A sand mix ture containing 1.25 we igh t percent b inder (with a 5:1 rat io of resin to hardener ) has a bench life of approximate ly 10-12 minutes , wi th a m o l d / c u r e s t r ipping time of approximately 60 minutes .

The water content of the molds is significant, especially in h u m i d weather , and it is impor tan t to d ry all molds and cores thoroughly prior to assembly. Ideally, this o p e r a t i o n is d o n e a t H O C for 30-40 m i n u t e s in a d r y i n g o v e n . F lushing of the molds wi th the protect ive gas mix ture men t ioned above prior to casting el iminated problems otherwise associated wi th oxidation.

Molds of varying complexity were p repared to test the system. Initially, a simple open top, cylindrical billet mo ld was filled by manua l control of the pressure in the casting furnace. Somet ime later, a complex automobi le intake manifold, Figure 2, was selected for trials.

Figure 2 - Photographs of low pressure sand cast automobile intake manifold.

The basis of pressure assisted bot tom feeding is to control the flow of metal into the mold . Consider the si tuation shown in Figure 3, whe re a constant cross section casting is gravi ty fed wi th a metal head, h t. The posit ion of the metal front, h, as a function of time, t, is given by:

MOLDS

MOLD FILLING

h = N/ 2 g h ^ ( A i/ A f) . t (1)

Here , Ai is the area of the ingate, Af is the constant cross sectional area of the casting, and g is the gravity constant.


Figure 3 - Principle of gravity feeding of casting.

The vertical filling velocity, V, as a function of t ime will then be:

V2

j% ( A j / A f ) ^ g (2)

The filling velocity can also be expressed as a function of the metal front posit ion, that is:

v = ii-N/ii(vw (3)

Equations 1 a n d 2 are shown graphically in Figure 4, where h and V are plot ted as functions of t ime. It is apparen t that the filling ra te varies considerably wi th time. Initially, the vertical velocity is high, wi th the possibility of creating turbulence and associated oxidat ion p rob lems . As the level in the casting approaches that of the feeder, the velocity decreases toward zero. This is also i l lustrated in Figure 5, which shows V as a function of the position of the molten metal front (Eqn. 3).

5 0 -

4 5 -

4 0 -

3 5 -

3 0 -

2 5 -

2 0 -

1 5 -

1 0 -

5 -

0 -

fit**

^%*h(eq.1)

V (eq . 2)

i i i i i r 0 5 10 15 20 25 30

FILLING TIME (sec)

4.0

— 3.5

— 3.0

— 2.5

2.0

1.5

1.0

0.5

0

Figure 4 - Variation of filling velocity and metal height wi th t ime du r ing gravity casting.

(wo) M1H

0I3H O

NH

IId

VE

RT

ICA

L F

ILLIN

G V

EL

OC

ITY

V (cm

/sec)


8 | 3 . 5 -

> 3 . 0 -

o O 2 . 5 -UJ > O 2 . 0 -z - I —I EE 1.5 — _i < g i . o -DC Ul > 0 . 5 -

0.0 0.0 0.2

~"T~ 0.4

I 0.6

I 0.8

RELATIVE FILLING HEIGHT, h / ht

Figure 5 - Variation of filling velocity as a function of relative filling height.

By control l ing the p ressure in the crucible, the filling rate can in pr inciple be precisely chosen. As a first approximat ion , the equat ion which connects pressure and level is given by:

p = P g h (4)

Here, p is the pressure and p is the density of the liquid metal.

The vertical filling velocity will then be:

V = ( l / p g ) d p / d t (5)

A more accurate description, which includes the effect of l iquid flow, but neglects frictional effects, is based u p o n Bernoulli's Equation:

Based on this equat ion, the pressure versus t ime relat ionship can be calculated for a specific filling sequence. For example , one choice might be a constant vertical fill rate, while another could be a constant mass flow.

W h e n the cross-sectional area of a casting as a function of he ight is known , Equat ion 6 can be uti l ized to develop the pressure- t ime relation even for complex geometr ies .

CASTING TRIALS AZ91

Several au tomobi le in take manifolds we re cast from high pu r i ty m a g n e s i u m alloy AZ91. Figure 6 shows the calculated pressure versus t ime relat ionship which arises from the applicat ion of Equat ion 6 and a constant metal flow of 0.5 k g / s e c . Figure 6 also shows the variat ion of pressure which was measured . It is apparen t that there is only slight deviation between these curves.


+ - CALCULATED

~ i — i — i — I — i — i — i — r — i — r 0 10 20 30 40 50 60 70 80 90 100

CASTING TIME (sec)

Figure 6 - Calculated and measured pressure versus t ime curves for sand cast intake manifolds.

To follow the mold filling, a series of thermocouples was located in the mold wall at var ious levels. Figure 7 shows the t ime dependence of the measu red metal level compared to the calculated level. It can be concluded that it takes an appreciably longer t ime to fill the mo ld than is ant icipated from the calculations. Some of this d iscrepancy can be a t t r ibuted to frictional losses in the transfer tube system. It is, however , m p r e likely that the resistance to flow deduced from Figure 7 is connected to part ia l solidification in the thinner sections of the mold . For AZ91 alloy, wi th a l iquidus t empera tu re of about 600C, a dendri t ic ne twork which restricts metal flow significantly is formed at temperatures only slightly be low the l iquidus.

C A L C U L A T E D

M E A S U R E D

CASTING T IME (sec)

Figure 7 - Calculated and measured mold filling behavior for intake manifolds.

The mechan ica l p r o p e r t i e s of tensi le spec imens m a c h i n e d from the in take manifold castings are s h o w n in Table I. Since p rocedures were not unde r t aken to control the solidification pa t te rn of these par ts , casting defects such as interdendri t ic porosi ty , sh r inkage poros i ty a n d hot cracking we re occasionally observed . These defects are k n o w to have a significant influence on mechanical proper t ies , and thus on the va lues l is ted in Table I. Defects which migh t h a v e ar isen from improper mold filling we re not encountered .


Table I - Mechanical Properties of Samples from Intake Manifold Castings (AZ91)

Cond i t ion 0.2% Offset Tensile Elongat ion Yield Strength S t rength

Elongat ion

(MPa) (MPa) (%)

T4 90 190 5.0

T6 120 210 4.5

The pho tomic rograph of Figure 8 shows the micros t ructure which is typical of sand cast AZ91. The grain size is of the order of 100-150 microns. Along the grain boundar ies and cell boundar i e s , mass ive Mgi7Al i2 particles are found, as well as eutectic colonies composed of lamellae of Mg and Mgi7Ali2. The dark grey, generally polyhedral phase particles in the interior of the grains are Al-Mn intermetallics.

Figure 8 - Microstructure of sand cast AZ91 alloy.

Other Alloys

A few casting trials were also conducted with two other magnesium alloys. Castings were produced from zirconium-refined ZE41 alloy, as well as from ZCM730 alloy. The ZCM alloy was readily adaptable to the low pressure mold filling process, whereas special attention was required for ZE41, primarily to try to insure successful grain refining.


S U M M A R Y

It has been demons t r a t ed that controlled low pressure assisted mo ld filling is a practical a n d viable process for the sand casting of magnes ium alloys. The inherent advantages of this process offer possibilities for the p roduc t ion of p r e m i u m quali ty castings.

REFERENCES

1. D. L. Albright, "Relationship of Microstructure and Corrosion Behavior in Magnes ium Alloy Ingots and Castings", Advances in M a g n e s i u m Alloys a n d Composi tes , H. Paris and W. H. Hunt , Eds., The Metallurgical Society, 1988.

2. J. E. Hillis and K. N . Reichek, "High Puri ty Magnes ium AM60 Alloy: The Critical Contaminan t Limits and the Salt Water Corrosion Performance", SAE Technical Paper 860288, February, 1986.

3. P roceed ings , Conference on Thermal Analys is of Mol ten A l u m i n u m , American Foundrymen ' s Society/Cast Metals Institute, Chicago, 1985.

4. P. R. Sahm, K. Weiss, H. Walther and D. Rosenthal, "Das rechnerische Modell ieren der Ers ta r rung bei NE-Metall-Dauerformguss", Giesserei, Volume 75,1988.

5. H. Westengen and O. Holta, "Low Pressure Permanent Mold Casting of Magnes ium - Recent Developments", SAE Technical Paper 880509, February 1988.

6. O. M. Hustoft and E. E. Estergaard, "Gravity Metering for Magnes ium Cold Chamber Die Casting", SDCE Paper G-T87-002, Proceedings, 14th International Die Casting Congress and Exposition, Toronto, 1987.

69

Ultimate sediment concentration in directionally solidified AlSi-SiC particulate metal matrix composites

M. Gallerneault, M. Kaya, R.W. Smith Department of Materials and Metallurgical Engineering, Queen's University, Kingston, Ontario, Canada

G.W. Dellamore Wollongong University, Wollongong, New South Wales, Australia

ABSTRACT

Samples of hypo-eutectic based Al-Si A356 and 15 volume per cent SiC particulate ( nominal diameter 10-15 micrometres) composites were solidified at rates ranging from 1 to 1500 micrometres per second, using both a Bridgman and an accelerated growth apparatus. The resulting microstructures are discussed with respect to the movement of the particulate reinforcement and the growth of the primary and eutectic phases. Results indicate that the primary phase (a-aluminum) is more effective at pushing the SiC particulate than previously postulated. There is no evidence that even at the lowest growth rate, where particle entrapment is predicted, the primary phase entraps the reinforcement.

KEYWORDS

Metal matrix composite, directional solidification, particle pushing, entrapment, SiC particulate.

INTRODUCTION

Metal matrix composites (MMC's) continue to make inroads into more commonly used and high tonnage areas such as automotive components. While a good deal of work has been done regarding fixed reinforcement systems (where the reinforcement is held in place as either a handleable preform or fibre tow) and the resulting solidification microstructures [1-10], there is much less information in the literature regarding the freely dispersed reinforcement or castable MMC's (i.e. composites processed with a molten matrix) [11-15]. As the distribution of the reinforcement has important consequences upon many of the properties of the MMC (e.g. fracture toughness, elongation to fracture, etc.), the establishment of the relationship between solidification mechanism(s) and reinforcement distribution will be an important task.

When a small reinforcing phase (i.e. the characteristic dimension -an effective diameter-is less than say 20 microns the material is referred to as particulate reinforced) is present in the hypo-eutectic system one is normally concerned with two phenomena:

i) particle pushing ,and,

1 Wollongong University, Wollongong, NSW, Australia.


ii) entrapment. Pushing occurs when the advancing solidifying interface (a dendrite in all practical

instances) forces the particulate into the interdendritic channels, thus the particulate distribution will be controlled by the dendrite size. If, on the other hand, particulate entrapment occurs, then the particulate should be incorporated in the advancing primary phase. Consequently, the particulate reinforcement will be largely unaffected by the advancing dendrite and its distribution will be more directly a function of the larger-scale fluctuations such as sedimentation, turbulence and convection. At present the understanding of the interactions between a reinforcement and the solidifying matrix is unclear [16-20], especially in terms of the reinforcement distribution. Using the method by Uhlmann et al[16] particulate pushing should occur at growth rates above about 1 micrometres per second. There is, however, little or no information regarding the effects of alloy or particulate build-up, and no experimental validation of that work for commercial MMC's.

A typical hypo-eutectic microstructure (A356 + 15 vol. % SiCp) appears as aluminum primaries with an interdendritic region comprised of eutectic AISi and the reinforcement (FIGURE 1). In the literature the rates of growth are usually confined to those (large) found in commercial processes. While fast rates of growth may be necessary for a variety of reasons (e.g. the particulate is heavier than the molten matrix and so it may be expected that sedimentation of the SiCp may introduce other complexities into the problem) it is often useful to perform slow growth experiments to analyze various steady state aspects of the solidification process.

The purpose of this study was to examine the directional solidification of hypo-eutectic AISi + 15 vol.% SiCp with the aim to understanding the micro-mechanisms which establish the ultimate distribution of the reinforcement in the MMC. Of particular interest was the determination the reinforcement build-up in the interdendritic regions and the effect of growth rate upon it.

EXPERIMENTAL

Extruded 9.0 mm. diameter rod of A356 (see TABLE 1) and A356 + 15 vol.% SiCp -500 grit - were swaged to 5.6 mm. diameter and grown vertically as 6 cm. long samples in a sealed alumina crucible (O.D.= 8 mm., I.D. = 6 mm. , LENGTH 300 mm.) using a Bridgman crystal grower; temperature gradient = 40Kcm"

1 over the liquidus/solidus range of 828 - 883K. The

composite (DURALCAN™

2) and alloy were supplied by Alcan International, Kingston Labs,

Kingston, Ontario, Canada. The samples were grown at rates of 1, 5, 10, 20 and 50 micrometres/second. In order to grow samples at the higher rates (above 50 micrometres/second) an accelerated growth apparatus was used, which has been described elsewhere [21].

TABLE 1 COMPOSITION LIMITS FOR A356 (WEIGHT PER CENT)

Si Fe Cu Mn Mg Zn Ti Other 6.5-7.5 0.12 0.10 0.05 0.30-0.45 0.05 0.20 0.15

2 DURALCAN™ is the proprietary process of Dural Aluminum Composite Corporation, a

wholly-owned subsidiary of Alcan Aluminium Limited, Montreal, Canada.


Sections of the grown samples were taken both longitudinally (using electro-discharge machining) and transverse to the growth direction ( with a diamond saw at 10 mm. intervals) to the growth direction and prepared as metallographic specimens using standard procedures for these types of materials [22]. As required, both optical and scanning electron microscopy (and EDS) were performed on the samples as well as image analysis (Kontron VIDAS® set at 480 by 512 pixels - Vertical by Horizontal resolution) to establish the microstructure constituents, measure dendrite size, volume fraction reinforcement and segregation in the samples. Measurements were made some 40 mm. from the starting (bottom) end of the crystals, though for the case of the slowest growing samples the 50 mm. section had to be used to obtain the desired number of cells for an average cell size determination.

RESULTS

The solidification microstructure of the A356 alloy is fairly complex and at the established growth conditions a variety of intermetallic phases can be observed (see FIGURE 2). The predominant solidification reaction is the hypo-eutectic AlSi. The form of the microstructure was cellular/dendritic at all examined growth conditions, FIGURES 3,5,7,9. At growth rates above 50 micrometres/second, directionality of the microstructure was lost, and large amounts of lateral growth were observed. Composite samples, grown at identical rates, displayed similar degrees of microstructural alignment, FIGURES 4,6,8,10, with the particulate always being associated with the interdendritic eutectic (last to freeze) phase(s) at all growth rates.

Measurement of the dendrite arm spacing was performed from optical micrographs. Individual cell widths were measured as the minimum dimension from sections parallel to the principal growth direction; about 120 measurements were made for each sample. The results for both composite and non-reinforced materials, given in FIGURE 11.

Due to the variations in the total solidification time for the samples (ranging from a few hundred to tens of thousands of seconds) there was some initial concern about the possible effect(s) that settling of the SiCp may have introduced into the measurements. To evaluate this a series of measurements was made to confirm the degree of settling. The transverse sections (taken every 10 mm.'s) were analyzed quantitatively for the area fraction of the primary phase. The measurements were performed at a magnification of 75X (screen magnification) which allowed for relatively easy discrimination of the primary and eutectic/reinforcement. Due to sectioning limitations, it was only possible to measure approximately 50 % of the entire cross-section of the sample. (Note that the values in TABLES 2 and 3 are of the mean and first standard deviation.) TABLE 2 outlines the results of the transverse area fraction measurements of the primary phase over the length of some of the samples.

Another consideration was the degree to which the reinforcement level varied within the interdendritic regions. Measurements of the area per cent SiCp were made at the interdendritic regions, which were a mixture of eutectic and SiCp, for the various levels of reinforcement using optical microscopy at a magnification of 1100 times (screen magnification). This higher magnification was necessary to ensure that over the entire range of growth rates/interdendritic region sizes, a good value for reinforcement level was possible. The results, TABLE 3, indicated that little variation of reinforcement in the eutectic occurred for the directionally grown materials. Note that the case for the most rapidly solidified material was not included as the eutectic colony width (and primary aluminum phase) was less than the width of the particulate.


TABLE 2 PRIMARY PHASE (AREA %) THROUGH LENGTH OF BRIDGMAN GROWN SAMPLE - TRANSVERSE SECTIONS (MAG = 75X)

Distance from Growth Rate (micrometres/second) Bottom (cm.'s) 1 5 50

1 51 ± 2 52 ± 2 54 ± 2 2 48 ± 3 58 ± 1 59 ± 2 3 50 ± 2 59 ± 2 60 ± 2 4 51 ± 5 57 ± 1 56 ± 1 5 50 ± 2 58 ± 2 57 ± 2

TABLE 3 AMOUNT OF INTERDENDRITIC SiCp (AREA %) FOR A356 + 15 Volume % SiCp - TRANSVERSE SECTIONS (MAG = 1100X)

Growth Rate (urns'

1) Interdendritic Conc'n (vol.% SiCp)

1 36.1 ± 4.7 5 35.4 ± 4.3 10 34.1 ± 4.5 20 31.2 ± 4.4 50 32.1 ± 5.7

DISCUSSION

The effect of reinforcement loading upon the cell spacing appears to be minor, even at the lowest rates of growth. At growth rates below about 50 micrometres/second there is a steady decrease in the cell size versus the non-reinforced material. Previously published results indicate a sedimentation rate for 15 volume per cent SiCp in A356 of about 20 micrometres/second, based on hindered Stoke's approximation [12],

_ £ p g ( p p- P i ) , . _r)t.es Vh 1 8 ^

U L)

where, Vh is the hindered settling velocity, Dp is the particle diameter, g is the gravitational constant, pp is the particulate density, p, is the liquid density, rj is the liquid viscosity and C is the concentration of particles. If one accepts that the hindered Stoke's case is plausible - there are fairly broad definitions for both Dp and r\ - then one possibility would be that the available settling particulate will fill-in the interdendritic region. (Recall that in these experiments the solidification zone extended some 10 mm.'s and at the lowest growth rates a large amount of


time is available for settling.) At the rates of growth near and below that of the hindered settling condition -about 20 micrometres/second- it is likely that the pile-up of particulate reached a level where refinement of the primary phase must occur in order to have enough energy to push the reinforcement. Typical commercial processes would employ much greater rates of growth and hence the final distribution of the particulate will reflect the more macroscopic variations in reinforcement associated with turbulence, convection and associated mould-filling considerations.

As shown in TABLE 2 (values given are for the mean and first standard deviation), the amount of primary phase was relatively constant when measured transversely. Longitudinal sections of the same samples showed no banding. Even at the top of the samples no significant amount of enrichment or depletion of the SiCp was observed in any of the directionally grown samples. Some small amount of enrichment was observed in the slowest grown composite sample (about the top 1 mm.), however this could be due to some oxide film effect at the top of the specimen. Indeed, the near total lack of any clustering or settling in the directionally grown specimens was first interpreted as evidence of a highly viscous 'melt' which did not experience sedimentation. However, this possibility was discounted by the presence of appreciable amount of settling below a stable oxide (made by a fine cut into the extruded rod prior to melting and growing), see FIGURE 12. In this case significant settling of the particulate may be seen, though as might be expected, on slow directional growth, little lateral pushing occurs. It was noted, however, that the reinforced samples displayed greater amounts of cellular asymmetry during growth. For example, often a growth twin formed during the solidification of the alloys at the slower growth rates; this never occurred with the composite samples. Similarly, the longitudinal sections did not show the same degree of cellular alignment with the growth direction in the composite as the for the alloy. Both of. these phenomena indicate a degree of interaction between the particulate and solidifying matrix.

It is unclear why neither settling nor banding of the particulate, due to primary pushing, occurred in these experiments; perhaps our growth rates were still too high. While it may be argued that any settling tendencies were overcome by the strength of the primary pushing, the lack of any longitudinal changes in the particulate distribution is puzzling. Specimen geometry could lead to the present situation. As the length of the solidification (mushy) zone comprised about 10 mm. ( 20 %) of the total length of the sample it could have acted as a physical damper to accommodate off-axis growth around particles or particle clusters. This would suggest that a shorter mushy zone or slower growth rate may reduce the degree of off-axis growth.

In conventional alloy solidification the interdendritic composition is often a function of the solubility limits of the various phases. Also, the space between primaries decreases with increased cooling, provided the undercoolings are kept fairly constant (the usual case for many Al-based systems). We note in this work that the build-up of particulate at the interdendritic region appears to have a maximum value for this system of approximately 35 volume per cent. While it is known that the A356 + 15 SiCp system is sensitive to shear rate, and its rheology is similar to a Bingham solid [23], the micro-mechanics of these materials are largely unknown. In this MMC system, the amount of material between the cells is a function of the ability of the particulate to 'fall' between the advancing cells (thus shape and loading should be important), the void created by the decrease in volume on solidification (and the rate it occurs), and perhaps most importantly, the rheology of the particulate/eutectic couple as it progresses down the interdendritic channel. These points are beyond the scope of this study.


1. The phenomenon of particle pushing by the primary phase is noted at growth rates as low as 1 micrometre/second for the A356 + 15 volume per cent SiCp system.

2. The presence of 15 volume per cent SiCp in A356 results in minor primary refinement at growth rates below 50 micrometres/second.

3. Interdendritic reinforcement levels for the A356 + 15 SiCp system were measured to be in the range of 36.1 ± 4.7 volume per cent. This may be taken as a maximum loading for the eutectic system.

ACKNOWLEDGEMENTS

The authors are indebted to Alcan International and NSERC-CRD for the funding of this work and to Drs. D.J. Lloyd and I. Jin and Mr. M. Ryvola of the Alcan International, Kingston Labs for their useful discussions during the course of this work and (M.R.) metallographic assistance.

REFERENCES

1. T.W. Clyne, M.G. Bader, G.R. Cappelman and P.A. Hubert, J. of Mat. Sci, 20, (1985), p. 20.

2. H. Fukunaga and K. Goda, Bull, of the Jap. Soc. of Mech. Eng., 28, 235, (1985), p. 1. 3. J. A. Cornie, A. Mortensen, M.N. Gungor and M.C. Flemings, ICCM V, (1985), ed.

Harrigan et. al., p. 809. 4. A. Mortensen, PhD Thesis, Dep't of Materials Science and Engineering, MIT, (1986). 5. M. Gallerneault and D.J. Lloyd, Can. Met. Quart., 28, 3, (1989), p. 265. 6. T.W. Clyne, Met. Trans., 18A, (1987), p. 1519. 7. R. Warren and C.H. Andersson, Composites, 15, 1, (1984), p. 101. 8. W.H. Hunt, "Interfaces in Metal Matrix Composites", Eds. A.K. Dhingra and S.G.

Fishman, TMS-AIME, (1986), p. 3. 9. A. Mortensen, M.N. Gungor, J.A. Cornie and M.C. Flemings, J. of Metals, (1986),

p.30. 10. A. Mortensen, J.A. Cornie and M.C. Flemings, J. of Metals, (1988), p. 12. 11. I. Jin and D.J. Lloyd, "ASM Int'l. Conf. of Fabrication of Reinforced Metal

Composites", Sept. 17-19, 1990, Montreal. 12. D.J. Lloyd, Composites Science and Technology, 35, (1989), p. 159. 13. D.M. Stefanescu and B.K. Dhindaw, ASM Handbook, vol. 15, p. 142. 14. P.K. Rohatgi, R. Asthana and S. Das, Int'l. Met. Rev., 13, 3, (1983), p. 115. 15. LA. Ibrahim, F.A. Mohamed and F.J. Laveraia, J. of Mat. Sci., 26, 5, (1991), p. 1. 16. D.R. Uhlmann, B. Chalmers and K.A. Jackson, J. Appl. Phys.,35, 10, (1964), p. 2986. 17. S.N. Omenyi and A.W. Neumann, J. Appl. Phys., 47, 9, (1976), p. 3956.


17. S.N. Omenyi and A.W. Neumann, J. Appl. Phys., 47, 9, (1976), p. 3956. 18. C.E. Schevoz and F. Weinberg, Met. Trans., 16B, (1985), p. 367. 19. D.M. Stefanescu, A. Moitra, A S . Kacar and B.K. Dhindaw, Met. Trans, 21A, (1990),

p. 231.

20. R. Sasikumar, T.R. Ramamohan and B.C. Pai, Acta Metall., 37, 7, (1989), p. 2085. 21. M. Gallerneault and R.W. Smith, Proc. of 6th Int'l. Conf. on Composite Structures,

(1991). 22. D J . Lloyd, M. Ryvola and E. Saez, Microstructural Science, 17, (1988), p. 189. 23. F. Ayersch and M. Mada, C.D.T. Project P1274, Ecole Polytechnique, Montreal.


FIGURE 2: Backscatter electron image of A356 grown at 1 ums 1 displaying interdendritic phases of ArAISi eutectic, B:Al3Fe, C:Mg2Si and DrAlFeSi.

FIGURE 1: Optical micrograph of the typical as-cast microstructure of A356 + 15 SiCp.


FIGURE 4: Optical micrograph of A356 taken parallel to the growth direction. Growth is UP. Growth rate = 1 pirns"1

FIGURE 3: Optical micrograph of A356 taken transverse to the growth direction. Growth rate = 1 pirns"1


FIGURE 5: Optical micrograph of A356 + 15 SiCp taken transverse to the growth direction. Growth Rate = 1 ums"1

FIGURE 6: Optical micrograph of A356 taken parallel to the growth direction. Growth is UP. Growth Rate = 1 ums"1


\ y. Vfc*. V^.^ * ^ .M r ^ .

FIGURE 7: Optical micrograph of A356 taken transverse to the growth direction. Growth Rate = 10 pirns*1

FIGURE 8: Optical micrograph of A356 + 15 vol. % SiCp transverse to the growth direction. Growth Rate = 10 pirns"1


" * J A. >\ / > • - x vi / - < f j—y< v

FIGURE 9: Optical micrograph of A356 taken transverse to the growth direction. Growth Rate = 50 ums"1

FIGURE 10: Optical micrograph of A356 + 15 vol. % SiC transverse to the growth direction. Growth Rate = 50 ums"1


1000

E

£ 100 +

"q3 O

• ALLOY

OMMC

10-1 l 1 1 1—

0 .200 1.000 10 .000 100 .000 1000 .000 Growth Rate (/zms"')

FIGURE 11: Variation of Cell Size with Growth Rate for A356 and A356 + 15 Vol. % SiCp

FIGURE 12: Optical micrograph of oxide 'cut' on side of sample. Growth direction UP. Growth Rate - 50 urns"1

83

On the wettability of ceramic fibres by metals in various metal matrix composite systems

H. Liu Departement des science appliquees, Universite du Quebec a Chicoutimi, Chicoutimi, Quebec, Canada

T. Suzuki Department of Metallurgical Engineering, Tokyo Institute of Technology, Ookayama, Tokyo, Japan

F.H. Samuel Departement des sciences Appliquees, Universite du Quebec a Chicoutimi, Chicoutimi, Quebec, Canada

ABSTRACT

Interface wettability and reactions between reinforcing fibers and matrix metals in the C/Cu, A1203/A1 and SiC/Al systems were investigated. It was found that the wettability between the ceramic fibers and matrix metals was generally poor, and that the interface reactions between the two phases degraded the tensile strength of reinforcing fibers. Alloying additions in small quantities of Mo, Cr, Fe, Co and V were found effective in improving the interface wettability for the C/Cu system. At the interface between A1203 and Al, the formation of volatile A120 was detected and the wetting behaviour was found to be complicated. The addition of Si to Al was found to be effective in controlling A14C3 formation for the reactive system of SiC/Al.

INTRODUCTION

Metal matrix composites (MMCs) are fabricated at the temperature of the molten matrix metal in the case of liquid metal processes such as infiltration or casting. In such processes, a good wettability between reinforcements and matrices is always required to obtain a homogeneous distribution of the reinforcements in the matrix. In most composite systems, however, the contact between the molten matrix metal and reinforcing fibers unavoidably results in interfacial reactions between the two phases. While a moderate amount of such reactions is necessary for improving the interface wettability and the interface bounding between the fibers and matrix, an excess amount will degrade the fibers.

The mechanical properties of MMCs are estimated from the mechanical properties of the components according to the rule of mixture (ROM); however, measured values are always found to deviate from (being in most cases lower than) the estimated ROM values, due to the fact that the ROM does not take into account the possible influence of any physical/chemical process occurring at the interface.

Although interface wettability and reactions for the C/Cu [1-3], A1203/A1 [4-6] and SiC/Al [7-8] systems have been investigated extensively, these studies have been carried out on pure or single crystal substrates and the wettability evaluated mostly by the sessile drop method. Also, a direct correlation between the wettability/interface reactions and mechanical properties has not been obtained. In the present investigation, therefore, the C/Cu, A1203/A1 and SiC/Al composite systems were examined by directly employing carbon, A1203 and SiC fibers and the respective


matrix metals, Cu and Al and their alloys. As the C/Cu system is known to be interface inert, even at molten copper temperatures, the

investigation was mainly directed at improving the wettability of the system by alloying additions.

For the A1203/A1 system, as previous studies (employing various types of alumina materials) gave scattered results, the investigation was carried out using commercial A1203 fibers and aluminum matrix.

For the SiC/Al system, the influence of the interface reactions on the tensile strength of the SiC fibers and the effect of Si addition to the Al matrix in controlling the reaction were investigated. Reaction zone theory was employed to discuss and explain the fracture modes of the SiC fibers attached to the reaction zone.

EXPERIMENTAL METHOD

As the systems have been investigated, the experimental procedure common to all is described herein. Deviations special to each case are mentioned in the respective sections.

The fibers used in the present work were PAN based HM carbon fibers, silicon carbide fibers (Nicalon) and alumina fibers. Their physical properties are listed in Table I. All the alloys used were prepared by the arc melting technique. The purity of both matrix metal (Cu, Al) as well as alloying elements was over 99.99%.

Pure matrix metals or their binary alloys were coated onto the surface of corresponding reinforcing fibers, in an ULVAC DRP-40E high vacuum electron beam evaporation apparatus. Reinforcing fibers in tow were unravelled and wound on the surface of a drum frame as uniformly as possible. The drum frame was then installed on a spindle in the evaporation chamber and driven at a rate of 30 rpm whereby a uniform coating of 0.2 /xm thickness could be achieved after 9.8 K sec deposition time. The coated fibers were encapsulated in a quartz tube under vacuum (10~

5 torr) and annealed for 1.8 K sec at temperatures corresponding to the

purpose of investigation, followed by water quenching.

TABLE I - Nominal properties of the fibers used in the present research.

Fibers Carbon SiC A1203

Impurity(mass%) - 0*11.2 Si2O:15.0 Density(g cm

1) 1.8 2.55 3.2

Diameter (jxm) 6.5 15.5 10.5 Tensile strength (GPa) 2.43 3.2 1.8 Eastic Modulus (GPa) 243.0 206.2 210.0

After the heat treatment, the wetting behaviour and interface reactions between the fibers and the coating metal for each composite system were examined using a JEOL JSM-150 scanning electron microscope.

To investigate the influence of interface reactions on the tensile strength of the reinforcing fibers before and after heat treatment, tensile tests were carried out on single fibers, using an


Instron type testing machine. The gauge length was set to be 20 mm. The Weibull distribution theory [9] was used to analyse the results obtained.


C/Cu System

Since copper is known to be chemically inert with respect to carbon, alloying additions were deemed necessary to increase the interface bonding and improve the interface wettability. Binary copper alloys can be classified into three categories: alloys with miscibility gaps, obtained by the addition of Pb, Nb, Ta, Cr, V, Mo, Fe or Co to copper; alloys forming continuous solid solutions with Mn, Pd, and Ni; and alloys producing intermetallic compounds with Si, Sn, In, Ti and Mg. Heat treatment of the carbon fibers coated with these Cu alloys was carried out at 1403 K for 30 min.

Effect of Alloying Addition on Wetting

The surface morphologies of carbon fibers as-received and coated with copper are shown in Fig.l . The presence of Cu droplets seen in Figs, lb and lc (formed in the liquid state and remaining intact after water quenching), on the surface of the carbon fibers indicates the poor wettability of the molten copper on the fibers.

Fig.2 displays the effect of alloying additions of Cr and V in improving the wettability of copper on carbon fibers. It can be seen that the contact angle between pure copper and the carbon fibers (Fig. lc) is somewhat reduced by the additions of 0.4 at% Cr (Fig.2a) and 0.3 at% V (Fig.2c). When the concentrations were increased to 1 at%, the contact angle was reduced to < 90° as shown in Figs.2b and 2d.

The wetting behaviour of the binary alloys was likewise observed and evaluated on the basis of "good" and "poor" levels. These are given in Table II. It was found that besides Cr and V, the elements Mo, Fe and Co were also effective in improving the wettability.

In a physico-chemical sense, a liquid metal that wets a solid phase should spread over its surface as shown in Fig.3. This spreading ability is generally expressed in terms of a spreading coefficient, Wg, defined as

wa = y

w - ( Y

W + y

b )

( 1)

where 7

v l, 7 " and 7**, are the surface tensions of liquid and solid and interface tension

between the liquid and the solid, respectively. For a given composite system, 7 " is a constant. Therefore, the smaller the values of 7

vl and

7

b, the larger should be the value of Wt. For the binary copper alloys used in the present work,

7

vl and 7

18 were calculated on the basis of a thermodynamic model and the results are shown i

Fig.4. Comparing experimental and calculated results (Table II and Fig.4, respectively), it is found that suitable elements are those which effectively reduce the interface tension between the copper and carbon fibers. It is also found that the elements that improve the wettability in the C/Cu system are those which form miscibility gaps with Cu. In addition, these elements have a rather low solid solubility in Cu and exhibit a strong tendency for carbide formation (with the exception of Pb) than the others, due to the positive heats of mixing of their corresponding


alloys which motivates the segregation of the alloying elements to the C/Cu interface to form carbides.

Figure 1-Scanning electron micrographs of carbon fiber surface after heat treatment at 1403 K for 30 min: (a) as-received; (b) and (c) coated with copper.

Figure 2-Reduction in the contact angle between coating alloys and carbon fiber surface coated with: (a) Cu-0.4 at% Cr; (b) Cu-1 at% Cr; (c) Cu-0.3 at% V and (d) Cu-1 at% V.

SoW 7 ,s Solid 7 ,s

Figure 3-Schematic representation of a liquid spreading on the surface of a solid substrate.

Fig.5 exhibits the calculated values of the surface and interface tensions as a function of the bulk concentration of the alloying elements. The reduction in the interface tension can be

i d 0 Mm ^


attributed to the interface adsorption of the alloying elements due to carbide formation. A consequent improvement in the wettability is thus obtained.

TABLE II - Alloying elements and their addition effects.

Category Elements Addition effects

Miscibility

gap

Pb Nb Ta Cr V Mo Fe Co Good except for P b , N b a n d T i

Continuous solid solution

Mn Pd Ni Poor

Intermetallic compound

Si Sn In Ti Mg Poor

1.1c*-0.80-

"oi 0.2 OA ' 0!6 0.8 To Bulk a tom fraction of alloying e lements , X? (%)

Figure 4-Variation in surface and interface tensions, 7

vl and y**, vs. the bulk

concentration of alloying elements, X?.

AHi<KJ m o l e "

1)

-110 -70 -30

-7

o I n t e r f a c e

a S u r f a c e

_l_ _l_ I i i I i i I i i -90 -60 -30 0 30 60 90 120

Ah|(KJ m o l e

- 1)

Figure 5 - S u r f a c e and i n t e r f a c e concentrations of alloying elements, X?, X|, vs. the heat of surface and interface adsorption, AH*, AI-& Bulk concentrations of alloying elements: 1 at%.

I n t e r f a c e t e n s i o n

S u r f a c e t e n s i o n


Tensile strength

Figure 6-Influence of heat treatment at 1403 K on the Weibull distribution of the tensile strength of carbon fibers coated with pure Cu, Cu-lat% Co, Cu-lat% Fe and Cu-lat% Cr.

The tensile strength of the as-received carbon fibers and those coated with pure Cu, Cu-1 at% Co, Cu-1 at% Fe and Cu-1 at% Cr were measured before and after heat treatment and the results were analysed by the Weibull distribution theory [9]. Fig.6 shows the Weibull plot of the test results. It is observed that, for the fibers coated with binary alloys containing 1 at% Co, Fe and Cr, the failure strength is reduced by the heat treatment. This may be attributed to the degrading effect of the interface reactions between the alloying elements and the fibers.

AUO JM System

The wetting behaviour of aluminum on sapphire, ruby and recrystallized alumina has been reported by many workers. However, due to the complicated wetting behaviour of the system, the results are rather scattered in nature. It was decided, therefore, to study the wetting and interface reactions of the system by directly investigating Al-coated A1203 fibers after heat treatment at several temperature levels above the melting point of pure Al.

Interface Reactions

The fracture surface of an as-coated fiber is shown in Fig. 7. It is apparent that the bonding between the coating and the fiber is weak. Fig.8 shows a coated fiber after annealing at 1173 K. Thermal stresses due to the large difference in thermal expansion coefficients of the Al2 O 3 and the Al coating open cracks that result in the peeling off of the Al layer after water quenching.

Associated with the peeling off, an interface reaction was found to occur above 1123 K. Many researchers have investigated the reaction between molten aluminum and alumina. At temperatures below 2000 K, the reaction has been reported as [10]

44/(/) +Al203(s)-+3Al20(g) (2)

The swelling up and bursting of the coating observed in the present study, displayed in Fig.9,

[(U

id-D

ul-] 6°

l=x


suggest a gaseous phase as one of the reaction products. When coated fibers, sealed in an evacuated quartz tube were heated above 1123 K for over 900 sec, a thin Al foil was observed on the inner wall of the quartz tube. However, carbon fibers coated with Al and heated likewise showed no such behaviour.

It is known that when volatile A120 meets a cold substrate, Al is deposited according to the reaction [10]

2Al2(Xg) - 44/(s) + 02(g) (3)

Figure 7-As-coated alumina fibers showing Figure 8-Cracking and peeling off of Al interface bonding. coating on the fiber surface.

Figure 10-Surface morphology of the fiber after interfacial reaction.

Figure 9-(a) Swelling up and (b) bursting of Al coating on the alumina fiber surface caused by the formation of volatile A120.


We can conclude that this is what occurs in the present case. The fiber surface subsequent to the interface reaction at 1273 K is shown in Fig. 10, where the bright area marks the region that suffered an extensive reaction.

Wettability at the A1203/A1 Interface

In the present investigation, no wetting of the alumina fibers by molten aluminum was observed below the maximum observation temperature of 1273 K. As seen from Eqn.l and Fig.3, the wettability is improved if the interface tension, V

-, between the two phases is reduced.

While an interface reaction tends to reduce the value of V8, Livey and Murray [11] have shown

that Vs is not reduced very much, even at temperatures of 1243 K (see Table III).

Table III Surface, interface tensions and contact angle in the A1203/A1 system.

Temperature (K) 7

v l(N m

1) 7

V ,(N m"') V ( N iff

1) Contact angle (°)

1213 0.819 0.991 1.797 170 1243 0.808 0.991 1.673 148 1528 0.712 0.959 0.897 85

This can be explained as follows. In general, an interface reaction results in a solid interface phase termed the reaction zone. This zone reduces the interface tension and thereby improves the interface wettability. However, due to the formation of volatile A120 at the A1203/A1 interface, no stable interface phase/reaction zone is created. Thus, not much reduction in y

1' is

obtained. Fig. 11 gives a schematic representation of the above.

Figure 11-Schematic diagram of the decohesion of Al coating from the surface of alumina fiber due to the formation of volatile A120: (a) swelling up and (b) breakage of the Al coating.

Influence of Interface Reaction on Fiber Strength

To investigate the effect of interface reactions on the fiber strength, tensile tests were performed on annealed fibers. Prior testing, the Al coating was removed from the fiber using HC1 solution. The effect of the latter on the fiber strength was previously proven negligible by comparing the measured strength of uncoated fibers before and after acid treatment.

( a ) <b)


Fig. 12 depicts the variation in tensile strength with heat treatment temperature for both coated and uncoated fibers. It is evident that the tensile strength for coated fibers is greatly decreased when contact is made with molten aluminum.

Al melting point

3 00 4 0 0 6 0 0 6 0 0 9 00 1000 1100 1200

Figure 12-Effect of heat treatment on the tensile strength of the coated (A) and uncoated (O) alumina fibers. Each strength value is the average of 40 measurements made on single fibers.

SiC/Al System

On account of the importance of this system, many workers have attempted to study the problems associated with the interface reactions for the SiC/Al system.

In the present work, Al-Si alloys were prepared and coated onto the surface of SiC fibers. The effects of Si content/addition in controlling the interface reaction was investigated. Fracture surface of SiC fibers coated with pure Al and Al-Si alloys were observed after various heat treatment regimes. The fracture modes were discussed using the Reaction Zone theory.

Influence of Interface Reaction on Fiber Strength

Fig. 13 shows the effect of annealing temperature on the tensile strength of coated and uncoated fibers, heated for 1.8 K sec. Prior testing, the coating was removed from the fibers after annealing, employing 0.2 N NaOH solution. From this figure, it is seen that for the coated fibers, a sudden decrease in tensile strength around the melting point of Al is observed, while in the case of the uncoated fibers, the strength retains approximately its original value. The reduction in strength is obviously due to the reaction between molten Al and SiC.

Fig. 14 shows the changes in tensile strength of SiC fibers coated with pure Al and different Al-Si alloys versus annealing time for a fixed annealing temperature of 833 K. It is seen that the fibers coated with Al-Si alloys exhibit less strength loss with increasing annealing time than those coated with pure Al. Also, no substantial difference is observed for fibers with different Al-Si alloy coatings (with differing Si contents). It is also evident from this figure, that each of the curves can be divided into two parts, one where the tensile strength remains nearly constant, and the second, where it decreases with annealing time.


Melting point of aluminium

300 4 0 0 5 0 0 6 0 0 700 8 0 0 9 0 0 1000 1 1 0 0 '

Annealing temperature (K)

Figure 13-Annealing temperature dependence of tensile strength of SiC fibers: (A ) coated with Al and (O) uncoated with Al. Annealing time: 1.8 K sec.

Annealing t i m e , t (h)

4 9 16

1 2 3 4

Square root of anneal ing time , P'

2 (h

1 / 2)

Figure 14-Relation between annealing time and tensile strength of the SiC fibers coated with Al ( O ) , Al-1 at% Si (A ) and Al-5 at% Si (°). The arrows separate the growth stages. Annealing temperature is 823 K.

Obviously, the heat treatment results in the formation of a reaction zone. This is easily confirmed by observing the polished transverse sections of coated fibers before and after heat treatment at 823 K for 16 hr, as shown in Figs. 15a and 15b.

The fracture behaviour of the coated fibers is greatly influenced by the presence of the reaction zone. The fracture modes of variously treated fibers are shown in Figs. 16a through 16h, where Fig. 16a corresponds to a fiber associated with pure Al, Figs. 16b through 16d show the fibers coated with pure Al, Al-1 at% Si, and Al-5 at% Si, respectively, all after being annealed at 823 K for 1 hr. Fibers with the same coatings as those corresponding to Figs. 16b through 16d, but annealed at 823 K for 16 hr are shown in Figs. 16e through 16g. It can be seen from Figs. 16b through 16g, that the fibers coated with pure Al exhibit a shorter pull-out length than

BdO) MlBuens aijsuei


Figure 15-Polished transverse sections of coated SiC fibers: (a) prior annealing; (b) after annealing.

those coated with Al-Si alloys. Fig.l6h shows the fracture surface of a fiber coated with pure Al heat treated at 823 K for 25 hr, revealing the flat nature of the surface.

Reaction Zone Growth

The tensile strength, o^, of a coated fiber is estimated from the rule of mixture by the equation

where af is the strength of the fiber including the reaction zone, am is the strength of the matrix at the fracture strain of the fiber, and Vf and V m are the volume fractions of the fibers including the reaction zone and unreacted matrix, respectively. For SiC fibers with 0.2 /xm thick coatings, the difference between the ah and o> values is less than 4%. In Fig. 14, ah is employed, since it can be measured by the tensile test. For theoretical considerations, however, it is convenient to choose o> to represent the changing trends in ab, as is often done in discussions of interface zone theories concerning composites.

The value of a{ in Eqn.14, as a function of the reaction zone thickness, 5, has been theoretically estimated by Ochiai et al [12]. According to their theory, where the dependence of of on b has been deduced, there are two situations to consider:

(a) In the first stage of the reaction zone growth, the defects caused by the interfacial reaction are less significant than those inherent in the fibers. In this case, the fiber strength is controlled by the inherent defects and therefore retains its original strength, a9 i.e.

This is what is observed in the first part of the curves shown in Fig. 14. (b) With increasing annealing time, the reaction zone grows, and the effect of the notch

formed in the reaction zone (see Fig. 17) in limiting the fiber strength becomes more significant. Thus the fiber strength decreases as the thickness of the reaction zone increases. The reaction zone is generally brittle and fractures at a small strain. The situation here is much more

(5)


complicated than in the case (a), and therefore needs to be discussed in detail. Fig. 17 is a schematic representation of the situation when a notch forms under loading, at a stress level of aM and extends into the fiber at a stress level of 0^. When aM < 0^, the notch will form first and then extend into the fiber. In this case the value of a{ is given by 0^. If, however, aM > 0^, as soon as the notch is formed, it will propagate catastrophically through the fiber. In this case, the value of a{ will be given by (%. In the theoretical treatment by Ochiai et al [12], the expressions for 0^ and 0^ are given by

a = c ( - 71 r L )

1 /m ( Ef ET) T ( 1 + - ) 6"

/m (6)

and

* 1.122 ^

where C is a constant, ru, the radius of the fiber core remaining unreacted, L, the gauge length of the tensile test, m, the Weibull distribution shape parameter, T, the gamma function, Kq, the fracture toughness of the fiber and Ef and Ez, the modulus for the fiber and reaction zone, respectively. The value of m for the fibers tested in the present case was evaluated to be > 2 by the Weibull distribution analysis.

According to Eqns.5, 6 and 7, the variations of 0*, aM and 0^ with increasing reaction zone thickness, 5, can be represented as shown in Fig. 18a, remembering that a{ = 0* in the first stage of the reaction zone growth. In the second stage, if (% < 0^, then af = 0^, and if > 0^, then 0f = 0W, and the tensile strength of the fibers, crf, will vary with 5 as shown in Fig. 18b, following the path A -* B -* C -* D.

Comparing Fig. 18b with Fig. 14, one can see that a{ changes with 5 in the same way as does 0h with t

1 / 2. It is to be noted that the relation between 5 and the annealing time, t, during the

growth of the reaction zone is generally given by

6

2 = Kt (8)

where K represents the reaction zone growth rate constant.

Explanation of Fracture Modes by Reaction Zone Theory

Since the changes in af in the two representative stages of reaction zone growth are found to be different, it is appropriate to discuss the fracture behaviour of the coated/annealed fibers separately for each stage.

It is easy to understand that the interface shear strength, T{, will increase with increasing 5. The value of T{ with respect to a critical interface shear strength, rc, is what determines the fracture mode of the coated fibers: when r{ < rc, interfacial debonding will occur, prior to the the failure of the fiber; when T{ > rc, the fiber will fracture before the debonding occurs. For the first reaction zone growth stage, shown in Figs. 14 and 18, the reaction zone is thin and the interface shear strength is weak, so that T{ < rc. Thus, for a uniaxial tensile test carried out on a coated fiber, interfacial debounding occurs prior to notch extension and the fiber fractures due


to the inherent defects present, resulting in a long pull-out length of the fiber, as seen in Fig. 16a for the as-coated fiber and Figs. 16b through 16d for coated fibers annealed at 823 K for 1 hr.

In the second growth stage of the reaction zone, the interface shear strength becomes strong, so that ^ > rc. In this stage, the strength of the coated fiber is reduced, depending on the thickness of the reaction zone. Fracture of the coated fibers occurs due to the extension of the

Figure 16-Fracture modes of SiC fibers coated with:(a) pure Al;(b) pure Al,(c) Al-lat% Si,(d) Al-5at% Si,all annealed for lhr;(e) pure Al,(f) Al-lat% Si,(g) Al-5at% Si,all annealed for 16 hr;(h) pure Al annealed for 25 hr. Temperature: 823 K.

i 15 um t|

.20 um t\

15 um J


Reaction / zone

Notch Extension of the notch

Figure 17-Models for analysis of the effects of notch formation on the tensile strength of the fiber: (a) a notch forms but does not extend; (b) a notch forms and extends into the fiber.

Figure 18-Schematic illustration of the variations in: (a) the tensile strength level a*, aM and (V, and (b) the strength of the fiber, af, with increasing thickness of the reaction zone, 5.

PU

B * uo

'o

'au

oz u

oiio

ea.' i

\\m

9iqu am io ss

ajjs

9|is

uai

Thickness of reaction zone , 5


notch, resulting in a short pull-out length, as seen in Figs.l6e through 16g, for coated fibersln the second growth stage of the reaction zone, the interface shear strength becomes strong, annealed at 823 K for 16 hr, or in a flat fracture surface as in the case of the coated fiber annealed at the same temperature for 25 hr, Fig. 16h. It should be noted that the pull-out length for the Al-Si alloy coated fibers is longer than that for the Al coated fiber, for the same heat treatment (823 K, 1 hr). This observation indicate the effectiveness of Si additions in suppressing the interface reaction zone growth.

CONCLUSION

Interfacial properties such as wettability and interface reactions and their effects on the tensile strength of reinforcing fibers in the composite systems C/Cu, A120 3/Al and SiC/Al were investigated and the following results obtained:

1. For the C/Cu system, those transition metals whose binary Cu alloys form miscibility gaps in the liquid state and which exhibit very low solid solubilities in Cu are judged suitable for improving the interfacial wettability between copper and carbon fibers. The elements Mo, Cr, V, Fe and Co were found to be effective. The resulting improvement in the wettability is attributed to the interface adsorption of the alloying elements and the interface reactions between them and the carbon fibers, leading to a reduction in the interface tension.

2. For the A1203/A1 system, no wetting was observed below 1273 K, the maximum temperature in the present investigations. Volatile A120 was formed above 1123 K. The interface reaction between molten Al and A120 was found to degrade the fiber strength greatly.

3. For the SiC/Al system, the reaction zone growth between SiC and Al comprises two stages. In the first stage, the fiber strength is not affected by the interfacial reaction. In the second stage, the interfacial reaction increases and the fiber strength is reduced with increasing thickness of the reaction zone. Fiber fracture is caused by the propagation of a notch formed in the reaction zone into the fiber. The interfacial shear strength determines the fracture mode of the fiber. Alloying additions of Si could retard the growth rate of the reaction zone.

ACKNOWLEDGEMENTS

The authors acknowledge the cooperation and the contribution of Dr T. Shinoda, Hitachi Research Laboratory, Hitachi Ltd, Japan, and Dr Y. Mishima, Research Laboratory of Precision Machinery and Electronics, Tokyo Institute of Technology, Japan. The financial support received from the Natural Science and Engineering Research Council of Canada, the foundation de TUQAC, and the Societe d'eletrolyse et de chimie Alcan(SECAL) is acknowledged with gratitude.


REFERENCES

1 A. G. Metcalfe, Interface in Metal Matrix Composites. Academic Press, New York, USA 1974, 1.

2 D. M. Goddad, Journal of Material Science. VOL.51, 1978, 1840. 3 D. A. Mortimer and M. Nicholas, Journal of Material Science. Vol.5, 1970, 149. 4 C. G. Levi, G. J. Abbaschian and R. Mehrabian, Metallurgical Transactions. 9A, 1978,

697. 5 R. D. Carnahan, T. L. Johnston and C. H. Li, Journal of American Ceramic Society.

Vol.441, 1958, 343. 6 J. A. Champion, B. J. Keene and J. M. Sillwood, Journal of Material Science. Vol.4,

1969,39. 7 S. Tawata and S. Yamada, Journal of Japanese Institute of Metals. Vol.47, 1983, 159. 8 S. Kohara, "Compatibility of SiC fibers with Aluminium", Proceedings of the Japan-US

Conference on Composite Materials. K. Kawata and Y. Akasaka, Eds., Japanese society for composite Materials, Tokyo, 1982, 1451.

9 W. Weibull, Fatigue Testing and Analysis of Results. Pergamon Press Ltd., Oxford, 1961, 134.

10 L. F. Mondolfo, Aluminium Alloys:Structure and Properties. Bustterworths, London, 1976,344.

11 D. T. Livey and P. Murray, "The Wetting Properties of Solid Oxides and Carbides by Liquid Metals", Plansee Proceedings. F. Bebesovsky, Ed., Reutte/Tyrol and Pergamon Press Ltd., London, 1955, 387.

12 S. Ochiai, S. Urakawa, K. Ameyama and Y. Murakami, Metallurgical Transactions. Vol. 11 A, 1980, 525.

99

Microstructures and properties of Ti-6Al-4V/TiB composites produced by rapid solidification processing

S.M.L. Sastry, R.J. Lederich, W.O. Soboyejo McDonnell Douglas Research Laboratories, St. Louis, Missouri, U.S.A.

ABSTRACT

Rapid solidification processing (RSP) of Ti-B alloys, with accompanying large undercoolings and high cooling rates, has been proven to be very effective for producing in-situ titanium composites containing very large volume fractions of reinforcing second-phase particles reminiscent of filaments or whiskers used in metal matrix composites. The reinforcing dispersoids are formed in-situ in Ti-B alloys either upon solidification or subsequently by the controlled decomposition of the supersaturated solid solutions. The reinforcements are typically 100-1000 nm in diameter and 1-10 y.m long, and they produce significant strength and modulus increases.


INTRODUCTION

A major advantage of rapid solidification processing (RSP) is its ability to produce alloy compositions not obtainable by conventional processing methods. RSP alloys have excellent compositional homogeneity, small grain sizes, and homogeneously distributed fine precipitates and dispersoids. Boron additions to RSP titanium alloys result in in-situ rod-like titanium boride reinforcements and such reinforcements significantly increase the moduli and strengths of the alloys (1-5). This paper reports the mechanical property improvements resulting from boron additions to Ti-6A1-4V.


Rapidly solidified powders were produced at the MDRL plasma arc melting/centrifugal atomization (PAMCA) facility, which is described in detail elsewhere (6). Source materials consisted of high purity titanium pieces, aluminum pellets, vanadium turnings, and small boron chunks. Each PAMCA production run yielded 5-kg of -80 mesh powder (< 180 diameter). The compositions of the five alloys investigated are shown in Table 1. The aim chemistries were attained and the interstitial levels were within acceptable limits.

Table 1. Chemical composition (wt %).

Alloy Al V B o C N H (ppm)

Ti-6A1-0.5B 6.20 - 0.53 0.11 0.077 0.0065 11 Ti-6A1-4V-0.5B 6.10 4.03 0.58 0.12 0.082 0.0053 13 Ti-6A1-1B 6.00 - 1.12 0.12 0.067 0.0051 27 Ti-7.5A1-4V-0.5B 7.43 4.12 0.53 0.12 0.017 0.0037 25 Ti-7.5A1-1B 7.54 - 1.10 0.12 0.13 0.027 29

90-224-1157

The rapidly solidified alloy powders were consolidated by extruding at the Air Force Materials Laboratories (Wright Patterson AFB, OH). The powders were packed inside thick-walled, 70 mm diameter Ti-6A1-4V extrusion cans, vacuum outgassed, and sealed. Each can contained 1 kg of powder. The cans were then heated to 1065°C (1950°F) for 2 h and extruded through a circular die at a reduction ratio of 14:1. All five alloys extruded very well.

Smooth, cylindrical, button-headed, tensile specimens having a gauge diameter of 37 mm (0.145 in.) were tested. Fracture toughness values were determined from single edge notched (SEN) bend specimens under three-point bend loading. The notches were introduced by electro-discharge machining using a 0.1 mm (0.004 in.) diameter wire, and the SEN specimens were pre-cracked using far-field compressive loading techniques (7).


The densities of these alloys are between 0 and 3% less than that of Ti-6A1-4V. The

variations in aluminum and vanadium contents have a greater effect on density than the boron

additions.


The alloys were evaluated in stabilized conditions and in a beta annealed (1200°C/1 h/FC + 600°C/24 h/AC) condition. Stabilization anneals of 704°C/24 h/AC and 815°C/24 h/AC were performed, and both produced nearly identical microstructures and mechanical properties. After the stabilization anneal the microstructure consists of fine equiaxed grains with rodlike TiB reinforcements aligned along the extrusion direction (Figs, la and lb). The equiaxed grains of the stabilized microstructure are between 3 and 8 |xm in diameter with the vanadium containing alloys having a slightly finer microstructure. The boride reinforcements were about 5 to 10 |im long, with a length to diameter ratio of about 8:1.

( a )

Figure 1. Optical micrographs of (a) Ti-6A1-0.5B and (b) Ti-7.5A1-4V-0.5B

following the 704°C/24 h/AC heat treatment.

Annealing the alloys at 1200°C results in coarsened borides and large grains (Figs. 2a and 2b). The borides are typically between 10 and 40 |j,m in length, and their high aspect ratio is retained. However these alloys did not form a Widmanstatten microstructure upon cooling from this temperature despite the fact that the beta transus is ~1040°C. The microstructure of the boron containing alloys contains two distinct phases as shown in the SEM micrograph in Fig. 3. Energy dispersive x-ray analysis identified the continuous (or grain boundary) phase as Ti-4A1-12V (higher in V and lower in Al) than the discontinuous phase (Ti-8A1-2V).

( a )

4

10 um

Figure 2. Optical micrographs of (a) Ti-6A1-0.5B and (b) Ti-7.5A1-4V-0.5B following the 1200°C/1 h/FC + 600°C/24 h/AC heat treatment.

(b)


Figure 3. Scanning electron micrograph of Ti-7.5A1-4V-0.5B following the 1200°C/1 h/FC + 600°C/24 h/AC heat treatment.

Alloys containing 0.5% B have very attractive combinations of strength and ductility at 25 ° C following the stabilization anneal (Fig. 4). The alloys containing 1% B have limited ductility at the present impurity levels. As shown in Fig. 5, the 1200°C annealing treatment produces a significant reduction in ductility of all the alloys with the exception of Ti-6A1-4V-0.5B, which has 9% ductility. Figure 6 compares the elevated temperature yield stresses of Ti-6A1-4V-0.5B and Ti-7.5A1-4V-0.5B with those of two conventional titanium alloys in the mill-annealed condition. The Ti-7.5A1-4V-0.5B alloy has a temperature advantage of more than 160°C over Ti-6A1-4V for the entire temperature range.

l,500r

^l .oooh

V7~7\ Percent elongation EX3 Yield stress

[53 £

i n 5CM*- KYA

i l l Is

H20

mi

1

A

16

12

• I

A' y y

Figure 4. Tensile properties at 25 °C of the stabilization annealed Ti-B alloys compared with mill annealed Ti-6A1-4V.

(«dW)

SS

3*S

PPM


1,500

-2 1,000

3 500

Percent elongation S X 1 Yield stress

I

0 H VT)

20

a Hi6

1.

I S 3

12

/ / / / 90-224-1151

Figure 5. Tensile properties at 25 °C of the beta annealed Ti-B alloys compared with mill annealed Ti-6A1-4V.

100 200 300 400 500 600 700

Temperature (°C)

90-224-1152

Figure 6. Elevated temperature tensile properties of (o) Ti-7.5A1-4V-0.5B and (A) Ti-6A1-4V-0.5B in the stabilization annealed condition, compared with (--) mill annealed Ti-6A1-4V, and (••) duplex annealed Ti-6A1-2Sn-4Zr-2Mo.

103


The modulus values are significantly higher in boron containing alloys than in Ti-6A1-4V and Ti-6Al-2Sn-4Zr-2Mo (Fig. 7). The fracture toughnesses of these alloys are plotted in Fig. 8. The ductile alloys have toughness values of between 26 and 42 MPav'm. The Ti-6A1-4V-0.6B alloys, with appreciable ductility in the beta annealed condition have a fracture toughness values of «47 MPaVm. Figures 9 and 10 plot secondary creep rates of the Ti-6A1-4V-0.5B and Ti-7.5A1-4V-0.5B alloys at 540 and 650 °C respectively. Shown for comparison are creep rates of mill-annealed Ti-6A1-4V and duplex annealed Ti-6Al-2Sn-4Zr-2Mo (note that the comparison alloys in Fig. 10 are at 600 °C rather than 650 °C and thus the improvement is even greater than what it appears). The TiB reinforcements decrease the creep rates by approximately 3 orders of magnitude.

Modulus, E (Msi)

5 10 15

Ti-6A1-.5B

TI-6A1-4V-.5B

Ti-6A1-1B

Ti-7.5A1-4V-.5B

T1-6A1-4V

Ti-6Al-2Sn-4Zr-2 Mo

V///////////A

/////////TTTl

50 100

Modulus, E (GPa)

150

Figure 7. Elastic moduli of Ti-B alloys, Ti-6A1-4V, and Ti-6Al-2Sn-4Zr-2Mo.

90-224-1154

Figure 8. Fracture toughness of the Ti-B alloys following (•) 704°C/24 h/AC, (•) 815°C/24 h/AC, and (A) 1200°C/1 h/AC + 600°C/24 h/AC annealing treatments.

*Z 90-2

3 Si

http://Ti-7.5A1-4V-.5B


1 10 100 1000

Initial stress, a (MPa) 90-224-1155

Figure 9. Stress dependence of steady-state creep at 540 °C of (A) mill annealed Ti-6A1-4V, (V) duplex annealed Ti-6Al-2Sn-4Zr-2Mo, and stabilization annealed (o) Ti-7.5A1-4V-0.5B and (°) Ti-6A1-4V-0.5B.

1 10 100 1000

Initial stress, a (MPa) 90-224-1156

Figure 10. Stress dependence of steady-state creep at 600° C of (A) mill annealed Ti-6A1-4V and (V) duplex annealed Ti-6Al-2Sn-4Zr-2Mo and at 650 °C of (•) Ti-6A1-4V-0.5B and (o) Ti-7.5A1-4V-0.5B in the stabilized condition, and (•) Ti-6A1-4V-0.5B in the beta annealed condition.

CONCLUSIONS

Titanium alloys containing rodlike TiB reinforcements were produced by rapid solidification processing. These alloys have attractive combination of room temperature strength, modulus, ductility, and fracture toughness. These TiB reinforcements extend the elevated temperature mechanical properties of Ti-6A1-4V by 160°C and reduce its creep rates by three orders of magnitude.

j_s) 3jfj daau) aims-X

prois ( j_S) 3JBJ daOJO 51FJS-XpB31S


ACKNOWLEDGEMENT

This investigation was supported by the McDonnell Douglas Corporation Independent Research and Development program.

REFERENCES

1. S.M.L. Sastry, J.E. O'Neal and T.C. Peng, "Thermal Stability and Growth Kinetics of Oxide, Boride, and Carbide Phases in Rapidly Solidified Titanium Alloys", Microstructural Science: Proceedings of the International Metallographic Society. M.E. Blum, P.M. French, R. M. Middleton, and G. F. Vander Voort, Eds., Elsevier, New York, NY, USA, 1986, 275-284.

2. S.M.L. Sastry, J.E.O'Neal and T.C. Peng, "Advanced Titanium Composites", U.S. Patent. No. 4639281, 21 January 1987.

3. S.M.L. Sastry, T.C. Peng and R.J. Lederich, "Mechanical Behavior of Rapidly Solidified Titanium Alloys", Mechanical Behavior of Rapidly Solidified Materials. S.M.L. Sastry and B.A. MacDonald, Eds., The Metallurgical Society of AIME, Warrendale, PA, U.S.A., 1986, 207-230.

4. S.M.L. Sastry, "Improvements of Titanium Alloy Microstructure and Properties by Rapid Solidification Processing", Rapid Solidification Processing Principles and Technologies, IV, R. Mehrabian and P. A. Parrish, Eds. Claitor's Publishing Division, Baton Rouge, LA, U.S.A., 1988, 165-173.

5. M. Taya and R.J. Arsenault, "A Comparison Between a Shear Lag Type Model and an Eshelby Type Model in Predicting the Mechanical Properties of a Short Fiber Composite", Scr. Met. 21, 3, 1987, 349-354.

6. T.C. Peng, B.D. London and S.M.L. Sastry, "Characteristic of Rapidly Solidified Titanium Alloy Powders Produced by Plasma-Arc-Melting/Centrifugal Atomization", Adv. in Powder Met.-1989. T.G. Gassbarre and W.F. Jandeska, Eds., MPEF, Princeton, NJ, U.S.A., 1989, 387-398.

7. S. Suresh and J.R. Brockenbrough, "Theory and Experiments of Fracture in Cyclic Compression: Single Phase Ceramics, Transforming Ceramics and Ceramic Composites", Acta Met. 36,1988,1455-1470.

107

Microstructure/property relationships in SiCp-reinforced aluminum composites

S. Dionne MTL/CANMET, Ottawa, Ontario, Canada

M.R. Krishnadev Departement de mines et metallurgie, Universite Laval, Ste-Foy, Quebec, Canada

ABSTRACT

Changes in the matrix microstructure have significant effects upon the mechanical properties and the fracture behaviour of SiC particulate-reinforced aluminum (SiCp/Al) composites. To evaluate these effects, various heat-treatments were used to modify the matrix microstructure and properties of SiCp/Al composites. Fracture surfaces of tensile specimens were examined in the scanning electron microscope and quantitative fractography was performed. The study revealed that the fracture mechanisms operative in the aluminum matrix of the SiC/Al tensile specimens are similar to those encountered in monolithic heat-treatable aluminum alloys and are dependant upon the aging condition. The occurrence and extent of reinforcement cracking and/or decohesion was related to microstructural changes produced by age-hardening.

KEY WORDS

Metal Matrix Composites: Aluminum Alloys: Silicon Carbide: Heat Treatment: Tensile Properties.


INTRODUCTION

Aluminum alloys reinforced with particulate SiC, compared with their unreinforced, monolithic counterparts , offer potentially higher specific strength and stiffness as well as improved fatigue and wear properties (1) . However, a serious drawback of these materials is their low ductility and fracture toughness. Tensile elongations of the composites, even with small volume fractions of reinforcement, are typically an order of magnitude less than those of the unreinforced alloys. Recently, several studies have focussed on the role of the microstructure on the deformation and fracture of metal matrix composites (2-7). It was found that matrix properties exert a significant influence on fatigue and fracture toughness of the composites. In the present work, in order to extend our knowledge of the effects of matrix properties on the behaviour of composites, various heat-treatments were imposed on a commercially fabricated SiCp reinforced 7091 aluminum composite and ensuing changes in the matrix microstructure and properties were studied.

MATERIAL AND EXPERIMENTAL PROCEDURE

The SiC/Al composite material for this study was purchased from DWA Composite Specialties Inc. in the form of plates (1.3 cm thick by 13 cm wide). These plates were extruded using conical dies with an area ratio of 11.6:1 from cylindrical billets which had been consolidated using a proprietary practice.

For chemical analysis, the composite material was dissolved in HCI / H 2 O 2 and the SiC particulate was filtered out. The composition of the dissolved matrix is given in Table I.

Table I-Composition (wt%) of the aluminum matrix of the SiC/7091 composite.

Zn Mq Cu Fe Si 0 0 1 A n 00

The particulate volume fraction, average size and aspect ratio were determined by quantitative image analysis using a Quantimet 920 image analyser. The measured volume fraction of SiC was 30.9% and the average particle area size was 15 |lm

2f with a significant number

of particles being larger than 20 |lm

2 (see Fig.l). The aspect ratio

of the SiC particles was close to one and they were uniformly distributed (Fig.2) .

To establish the age-hardening behaviour of the material, heat-treatments were carried out on polished specimens. The specimens were solution annealed at 4 90±5°C for one hour in an argon atmosphere and quenched in water at 25°C. Aging was carried out at 120±2°C or 160±2°C for the required period of time and the specimens were air cooled to room temperature.


0 . 6

0 - 1 0 1 0 - 2 0 2 0 - 3 0 3 0 - 4 0 4 0 - 5 0 5 0 - 6 0 6 0 - 7 0 7 0 - 8 0 > 8 0

PART ICULATE AREA (Jim2 )

Figure 1-SiC particulate size distribution in a transverse section of the SiC/7091 composite

Figure 2-Optical micrograph of SiC/7091 (transverse section etched with Graff and Sargent's etch).


Microhardness measurements were carried out on polished specimens using a diamond pyramid indenter with a 50 gf load applied for 15 seconds. Indentations were made in the aluminum matrix between SiC particles within 30 minutes of completion of the heat-treatment. Ten to twenty measurements were made on each specimen. The age-hardening curves obtained in this manner were used to select heat-treatment procedures corresponding to under-aged (120°C/lh) and peak-aged (120°C/24h and 160°C/4h) matrix conditions.

For tensile testing, round specimens with a gauge diameter of 6.35 mm, gauge length of 25.4 mm and button ends were diamond machined with their long axis perpendicular to the extrusion direction. The specimens were heat-treated after machining and immediately tested at room temperature on an Instron Model 8562 frame at a strain rate of 1.68X10"

4 s"l. Quantitative fractography

was carried out on broken tensile specimens. The area fraction of SiC on the fracture surface was evaluated by a manual point count method.

MICROSTRUCTURE

The distribution of intermetallic constituents in the as-fabricated SiC/7091 composite was examined using the back scattered electron mode of the SEM, which provides atomic contrast images. Small intermetallics (0.5 to 2 Jim) aligned in the extrusion direction were found, by EDX analysis, to consist of Al, Mg, Cu and Zn (Fig.3) . A few large (5 to 20 |im) Ti- and V-containing inclusions were also found as well as smaller (about 1 to 2 (im in diameter) Fe- and Al-containing inclusions. After solution annealing, the A1-, Mg-, Cu- and Zn-containing intermetallics were completely dissolved and only the Fe+Al and Ti+V inclusions remained undissolved, thereby greatly reducing the volume fraction of large intermetallics in the material. A peak-aged specimen was etched with Graff and Sargent

1s etch to reveal the grain boundaries

(Fig.l). Most grains were quite small and the average grain diameter was smaller than the interparticle spacing. A few larger grains were observed in areas with small local volume fraction of SiC.

MECHANICAL PROPERTIES

Figure 4 shows a plot of the microhardness of the aluminum matrix of the composie as a function of aging time at 120°C and at 160°C. Peak hardness was reached after 24 to 28 hours at 120°C Matrix hardness increased from about 85 kgmm~2 (solution-annealed state) to 185 kgmm"2 (peak-aged state). Aging was very rapid : 70% of the total hardness increment was obtained after one hour of aging. When the aging treatment was carried out at 160°C, the time for peak hardness decreased to 4 hours but the matrix microhardness decreased to 148 kgmm~2. This effect of aging temperature on the peak hardness and aging kinetics is similar to that reported for unreinforced wrought Al-Zn-Mg-Cu alloys such as 7075 (8).


Figure 3-Back scattered electron micrograph of as-fabricated SiC/7091 composite showing the distribution of intermetallics.

200 I — i — • • ••••••I

TIME (HOUR)

Figure 4-Plot of microhardness of the aluminum matrix of the composite as a function of aging time at 120°C and 160°C.

CM I

.* CO CO

$

§ H 2


T e n s i l e P r o p e r t i e s

Figure 5 shows the effects of the different heat-treatments on the yield and ultimate tensile strength of the SiC/7091 composites. Heat-treatment which produced peak hardness of the aluminum matrix also resulted in significant increases in the yield strength and ultimate tensile strength of the composite compared with those of the as-fabricated and solution-annealed materials. The increase in yield strength was more pronounced than the increase in ultimate tensile strength for both aging temperatures, as expected for artificial aging of aluminum alloys. This indicates a loss in the work hardening capacity during aging. This loss is also implied by the fact that the ratios of yield strength to tensile strength in the aged materials are considerably higher than in the case of solution annealed material (Table II).

The strength increments achieved by a peak-aging treatment at 120°C were 40 to 45 MPa higher than those measured for material aged at 160°C. This may be related to the variation in the size and distribution of precipitated strengthening phases, a finer size and distribution corresponding to a higher yield strength. Higher aging temperatures usually result in coarser precipitate distributions (9). However, the tensile elongation of material aged at 160°C was slightly greater than that of material aged at 120°C (see Table II) . All three artificially-aged materials had significantly lower ductility than materials tested in the solution-annealed condition.

A F : a s - f a b r i c a t e d

S A : s o l u t i o n - a n n e a l e d

U A : u n d e r - a g e d 1 2 0 ° C / l h

PA 1 2 0 ° C : p e a k - a g e d 1 2 0 ° C / 2 4 h

PA 160°C: p e a k - a g e d 1 6 0 ° C / 4 h

H y . s .

• T . S .

Figure 5-Effect of heat-treatment on the strength of SiC/7091 Composites.

a

w o 53 w CO


Fracfcography

Fracture surfaces of as-fabricated, under-aged and peak-aged SiC/7091 tensile specimens were quite flat and perpendicular to the specimen axis. Common fractographic features included cracked SiC particles located at the bottom of large aluminum dimples, surrounded by numerous smaller dimples in the aluminum matrix (Fig. 6a) . The size of the larger dimples tended to be closer to that of the parent SiC for the aged specimens than for specimens tested in the as-fabricated condition. In many areas, aluminum could be observed to adhere to the SiC particles, indicating a strong SiC/Al bond (Fig.7a). On the fracture surfaces of specimens tested in the as-fabricated condition, void nucleation at the large intermetallic phases in the aluminum matrix and near the SiC/Al interfaces was observed (Fig.7b).

Table II-Tensile properties of heat-treated SiC/7091 composites.

Average Area Fraction Heat Treatment Elongation dtf/d£* Y.S.to of Cracked SiC on

Condition (%) (GPa) T.S. Ratio

Tensile Fracture Surface (%)

As-Fabricated 2 . .5±0. .4 45 0.75 3 6±8

Solution- 5. .6±0. .2 32 .8 0.54 20±4 Annealed

Under-Aged 2 . . 0±0 . .5 68 0.79 51±5 (120°C/lh) Peak-Aged 1. . 4±0. .5 97.8 0.88 4 6±3 (120°C/24h) Peak-Aged 1, .7±0. .4 — 0.87 32±4 (160°C/4h)

do/de is the rate of work hardening measured as the slope of

the true stress-true strain plot at a true strain of 0.003.

In the case of specimens tested in the solution-annealed condition, the fracture plane was slanted approximately 45° from the tensile axis. The fracture surfaces were composed of cracked SiC particles surrounded by flat matrix areas with some elongated dimples (Fig.6b). Many of the cracked SiC lay at the bottom of the elongated dimples. Fracture in the matrix region immediately adjacent to the SiC/Al interface was also observed in a few instances. This resulted in decohered SiC with a thin layer of finely dimpled aluminum adhering to the particle. The occurrence of this SiC/Al decohesion was greater in the solution-annealed specimens than in the as-fabricated and aged specimens. Nevertheless, the number of decohered SiC observed on the fracture surfaces of solution-annealed specimens was small and this fracture mode obviously did not significantly contribute to the overall failure of the material.

Measurements of area fraction of cracked SiC particles indicated that the solution-annealed material had the lowest number of cracked particles (see Table I I ) . The area fraction of cracked particles increased rapidly with aging but tended to decrease


Figure 6b-SEM fractograph of a solution-annealed SiC/7091 tensile specimen.

114

Figure 6a-SEM fractograph of an under-aged (120°C/lh) SiC/7091 tensile specimen.


Figure 7b-SEM fractograph of a SiC/7091 tensile specimen tested in the as-fabricated condition.

Arrows show dimples initiated at large intermetallics.

115

Figure 7b-SEM fractograph of a SiC/7091 tensile specFigure 7b-SEM fractograph


somewhat on approaching the peak in hardness. Thus the tendency for the composite to fail by particulate cracking was enhanced by aging. In the case of the solution-annealed material, the area fraction of SiC on the fracture surface was smaller than the fraction of SiC in the bulk material. This indicates that SiC particle fracture played a minor role in the fracture of this material.

DISCUSSION

This study has highlighted the complex relationship between the fracture mechanisms of MMC's and the microstructural features of the matrix and particulate. In examining the results of the present study, two conditions can be identified depending upon the level of stress transferred to the reinforcement particulate. In the first case, the load transferred to the particulate is such that particle cracking initiates the failure. In the second case, the matrix is so weak that the fracture stress of the particles is not exceeded and the failure is dependent upon fracture mechanisms operative in the matrix.

The reason for these different fracture mechanisms may lie in the effect of aging on the strength and work hardening behaviour of the matrix. In addition to reducing the capacity for work hardening, the aging causes a rapid increase in the rate of work hardening during the initial part of the tensile test (see Table II). The higher yield strength and work-hardening rates of the aged matrices result in higher values of stress in matrix areas around the SiC particles. In the case of particulate-reinforced alloys, the increase in local stress is expected to be accelerated by the presence of stress concentration sites at the sharp corners of the reinforcement particles and by the presence of highly-constrained matrix areas between closely-spaced particles. Transfer of these high stresses to the particles leads to an increased incidence of the particulate breaking strength being exceeded and results in enhanced void nucleation.

For specimens tested in the solution-annealed condition, the presence of magnesium in solid solution promoted a shear mode of matrix failure, similar to the failure mode observed in solution-annealed, unreinforced 7075 alloys (10) . In the solution-annealed composite, the general stress levels attained in the matrix were insufficient to produce particulate cracking before the occurrence of void initiation in the matrix. In the case of material tested in the as-fabricated condition, the age-hardening elements were present as large intermetallic particles. Void initiation in the matrix alloy occurred at these large intermetallics, resulting in low ductility of the composite, despite a relatively soft matrix.

Although failure of the composite by matrix shear led to lower ultimate tensile strength values, it was accompanied by greater ductility, with a tensile elongation closer to the range of values considered as useful for engineering applications. In the case of an Al-Zn-Mg-Cu matrix alloy, the solution-annealed condition is not stable at room temperature. However, the possibility of improving the ductility of metal matrix composites reinforced with relatively coarse SiC by optimizing the matrix composition and heat-treatment


should be explored. This approach could be valuable for applications where an improvement in properties such as stiffness and wear resistance, coupled with adequate ductility, are of greater importance than achieving higher strengths.

REFERENCES

(1) S. V. Nair, J. K. Tien and R. C. Bates, I n t . M e t . R e v . 1985, 275-290.

(2) C P . You, A. W. Thompson and I. M. Bernstein, Scr. Metall. 2 1 , 1987, 181-185.

(3) M. Manoharan and J. J. Lewandowski, Scr. Metall. Zl, 1989, 301-304 .

(4) T. Christman, A. Needleman, S. Nutt and S. Suresh, Mater. Sci. Eng. A1Q7. 1989, 49-61.

(5) S. Dionne and M. R. Krishnadev, Fabrication of Particulates Reinforced Metal Composites, J. Masounave and F. G. Hamel, Eds., ASM International, Materials Park, Ohio, 1990, 261-270.

(6) D. L. Davidson, Metallurgical Transactions 22&, No.l, 1991, 113-123.

(7) Z. Zhao, S. Zhijian and X. Yingkun, Mat. Sci. Eng. A132. 1991, 83-88.

(8) J. E. Hatch, Ed., A luminum:Properties and Physical Metallurgy. ASM, Metals Park, Ohio, 1984, 184.

(9) D. A. Porter and K. E. Easterling, Phase Transformations in Metals and Alloysr Van Nostrand Reinhold Co., Wokingham, England, 1981, 291-308.

(10) H. Chandra-Holm and J. D. Embury, Y i e l d , Flow and F r a c t u r e o f Polycrystals, T. N. Baker, Ed., Elsevier, 1983, 275-310.

ACKNOWLEDGEMENTS

The authors would like to thank Dr. M. Shehata and Mr. B. Casault (CANMET) for performing the quantitative metallographic evaluation, Mr. R. Bouchard and G. Weatherhall (CANMET) for performing the tensile tests and Mr. B. Durocher and Y. Lavoie (CANMET) for their help with photomicroscopy and heat-treatments.

119

The wetting of carbon/TiB2 composite materials by aluminum

K.D. Watson, J.M. Toguri Department of Metallurgy and Materials Science, University of Toronto, Toronto, Ontario, Canada

ABSTRACT

The specific energy requirement for the production of aluminium in Hall-Heroult reduction cells could be decreased if an inert, aluminium wetted cathode was used. A potential material for this application is carbon/TiB2 composite. A sessile drop technique combined with X-ray radiography was used to measure the contact angle formed between aluminium and pure hot-pressed TiB2, carbon/TiB2 composite, graphite and a carbonaceous cement in cryolite-alumina melts.

INTRODUCTION

The conventional electrolytic process for the production of aluminium using carbon cathode linings is energy inefficient [1]. Carbon is not wetted by aluminium which necessitates that the reduction cells must operate with a deep aluminium pad. The interface between the aluminium pad and the electrolyte is mobile and the interpolar gap in an aluminium reduction cell must be maintained at about 40-50 mm to prevent intermittent shorting [2]. The resultant ohmic voltage drop in the electrolyte of from 1.5 to 2.5 V, represents between 30 to 40 percent of the total electrical energy consumption of the process [3].

The efficiency of aluminium production could be improved using drained cathode reduction cells with sloping, aluminium wetted cathodes [1]. In this design the aluminium pad is replaced by an adherent wetting aluminium film. A drained cathode cell could be operated at a reduced interpolar gap with attendant energy savings. Since the 1950's extensive research has been carried out to find a material suitable for use in drained cathode cell design [4]. A candidate material is carbon/TiB2 composite [5,6]. An important requirement of this material is that it be wetted by aluminium.

It is reported extensively in the literature that TiB2 is wetted by aluminium with reference to aluminium electrolysis [1-7]. Most of the reports are qualitative in nature. The wetting of aluminium upon hot pressed TiB2 in vacuum was investigated by Rhee [8] and Samsonov [9]. Both reported that the contact angle was temperature dependent with wetting being observed as temperature increased above 980 K and 1425 K respectively. Both of these studies were conducted in the absence of cryolite-alumina melts.

Liao and Liu [10] measured the contact angle of aluminium on hot-pressed TiB2 and cathode carbon coated with a TiB2 paste, in cryolite-alumina melts. An X-ray radiographic sessile drop


technique was used. The TiB2 coating paste contained TiB2 powder, resin, pitch and additives. Two pastes were used, one containing 40 mass % TiB2 (#1) and the other 60 mass % TiB2 (#2). They reported contact angle values for paste #1 of between 63° and 65° and for paste #2 of between 51° and 57°. The hot-pressed TiB2 was found to be completely wetted by aluminium.

In the present study the contact angles formed by aluminium on pure hot-pressed TiB2, graphite, carbonaceous cement and carbon/TiB2 composites in the presence of cryolite-alumina melts were determined.

EXPERIMENTAL

The sessile drop technique incorporating X-ray imaging of the drop profile was used for contact angle measurements. The experimental apparatus and technique used was that described by Utigard [11].

The specifications of the chemicals used to prepare the cryolite-alumina melts are given in Table I. A standard melt composition of 8 mass % (excess) A1F3, 5 mass % CaF2, 3 mass % A1203, balance cryolite was used for all tests.

Table I - Specifications of Chemicals

CHEMICAL COMPANY SPECIFICATIONS

Na3AlF6 Alfa Products 97.6% (1.0% A1203 -0.2% CaF2)

Al2Os J.T. Baker Alfa Products

Reagent (99.1%) (99.99%)

A1F3 Alfa Products Anhyd. (99.5%)

CaF2 Fischer Certified

Al Alfa Products 99.999%

Contact angle measurements were conducted on pure hot-pressed TiB2, carbon/TiB2 composites of varying TiB2 levels, graphite and a carbonaceous cement. The composition of the carbon/TiB2 composites are shown in Table II. The samples were 19 mm or 14 mm in diameter. The hot pressed TiB2 samples were 6.8 mm in thickness and all the other samples were 5 mm in thickness. The surface of the samples was prepared using silicon carbide paper. The final finish was obtained with 600 grit paper.

Table II - Carbon/TiB2 Composite Samples (mass %)

TiB2 Powder 44.6 54.0 62.0 70.0

Carbon Matrix 55.4 46.0 38.0 30.0

The samples were glued to the bottom of a graphite crucible using the carbonaceous cement. 0.55 g of aluminium and 15 g of melt were used in each test. The crucible was placed into the reaction tube of the furnace which was at the test temperature. The interaction between the aluminium


and the sample was monitored using an X-ray/TV system. The time at which the cryolite-alumina mixture was observed to be fully molten was designated time zero for the test. Radiographs were taken of the sessile drop as a function of time. The radiograph exposure time was 10 seconds. The crucible was rotated 90° to assess drop symmetry. All tests were limited to less than 2 hours as the fused silica reaction tubes were prone to failure at longer periods. The contact angles were determined from the radiographs using the method of Utigard and Toguri [12]. At the completion of the test, the crucible was removed from the furnace and cooled in a stream of fanned air. The cooled crucibles were sectioned.


In all tests, during the period of crucible heat-up, the aluminium retained the original shape of the pellet, even when the temperature was above the melting point of aluminium. It appears that the thin adherent solid oxide film on the aluminum surface constrained any shape change. This phenomena is well documented in the literature [13]. When the cryolite-alumina mixture started to melt, the aluminium drop shape changed, presumably due to the removal of the oxide film from the surface of the aluminium by dissolution into the melt. The behaviour of the aluminium drop then became dependent upon the substrate material.

Graphite, Carbonaceous Cement, Hot-pressed TiB,

The contact angle values measured on the graphite, carbonaceous cement and hot pressed TiB2 are given in Table III.

Table III - Contact Angle Values.

SAMPLE TEMP.(°C) CONTACT ANGLE

(DEGREES) (AT TIME = 30 MINUTES)

(1) (2) AVERAGE

Graphite 980 155 150 153

Graphite 1000 161 155 158

Graphite 1010 152 156 154

Carbonaceous cement 980 141 147 144

Carbonaceous cement 1010 144 148 146

Hot pressed TiB2 (A) 980 24 28 26

Hot pressed TiB2 (A) 980 27 25 26

Hot pressed TiB2 (B)

1 980 0 0 0

Hot pressed TiB2 (B)

1 980 0 0 0

1 See Results and Discussion.

The values given are those measured 30 minutes after the melt was fully molten. Contact angle (1) is an average of the contact angles measured on each side of the drop. Contact angle (2) is the value after rotation of the sample by 90°.


Figure 1 - Radiograph showing non-wetting of aluminium on graphite in a cryolite-alumina melt, at 1000°C, magnification 2.25X.

The graphite and carbonaceous cement were not wetted by the aluminium. A typical radiograph showing non-wetting between the aluminium and the graphite is given in FIG. 1. The average contact angles were in the range 144°-158°. The contact angles were independent of temperature in the range (980-1010°C) and did not change significantly with time. Utigard [11], reported contact angles of between 150° and 170° for aluminium on graphite in cryolite-alumina melts at various temperatures.

The hot-pressed TiB2 samples were wetted by the aluminium immediately following the melting of the cryolite-alumina mixture. The aluminium spread rapidly to the perimeter of samples. In the initial tests (A), the measured contact angles were approximately 26°. The aluminium had spread to and was confined by the wall of the crucible, as illustrated by the radiograph in FIG. 2. Using samples of smaller diameter (B) placed in the centre of the crucible, with the outside of the sample well away from the crucible wall, the aluminium spread rapidly right across the top of the sample. Within a few seconds the aluminium could not be visually discerned and no contact angle could be measured. Subsequent examination of the sectioned crucibles revealed complete coverage of the TiB2 samples by a film of aluminium. The hot-pressed TiB2 samples were completely wetted by the aluminium, contact angle of zero, which is consistent with literature reports [2,4,5,7,10].

Carbon/TiB, Composite

The contact angle measurements for the carbon/TiB2 composites are plotted on FIG. 3. The values presented are an average of four contact angles measured at the drop periphery each 90° apart. The plots in FIG. 3 indicate that in all tests the contact angle was time dependent. Two possible explanations for this time dependency are (i) removal of contaminants from the composite surface and (ii) a reduction in the aluminium drop volume.

Surface Contaminant Removal

The results obtained in this and other studies [10,11], suggest that pure TiB2 is wetted and carbon non-wetted by aluminium in cryolite-alumina melts. The carbon/TiB2 composite therefore consists of a wettable component in a non-wettable matrix. The contact angle is determined, in part, by the concentration of TiB2 at the composite surface.

Examination of the carbon/TiB2 composite using an electron microprobe suggest that a carbon film is present on the surface the TiB2 particles. Such a film may play a role in the time dependency of the contact angle. The presence of a carbon film would lower the effective surface concentration of TiB2. If the carbon film is removed, the wettability of the composite would increase. If this occurs

Figure 2 - Radiograph showing complete spreading of aluminium on hot-pressed TiB2 in a cryolite-alumina melt, at 1000°C, magnification 2.25X.


120i

110'

100-

90j

80"

70ii

60^

50-

40-

30-

20-

10-

a a

X

D •

X

4 L •

45 x

54 •

62 •

70

~80~ 20 40 60 TIME (minutes)

100 120

Figure 3 - Contact angle versus time for carbon/TiB2 composites of various TiB2 levels (mass %), at 1000°C.

as a function of time during the contact angle test, the measured contact angle will decrease over time.

One possible mechanism for the removal of the carbon film from the TiB2 particles is the formation and subsequent dissolution of aluminium carbide into the melt. This will increase the concentration of TiB2 at the composite surface and consequently, the wettability of the composite. The contact angle would decrease at a rate dependent upon the removal of the carbon through the A14C3 formation and dissolution. Similarly, other contaminants at the composite surface which may affect wetting, such as oxides, may be removed by dissolution into the melt and so result in the time dependency of contact angle.

A reduction in contact angle due to this mechanism would result in the advance of the drop periphery. A decrease in contact angle and a corresponding advancement of the drop periphery, as revealed by inspection of the radiographs, was observed in tests carried out on these composites. Inspection of the sectioned crucibles revealed a yellow material on the carbon/TiB2 composite surface adjacent to the aluminium drop periphery. This material was assumed to be AI4C3 which has a characteristic yellow colour [2].

Aluminium Removal

The time dependence of the contact angle may also be due to the removal of aluminium from the drop. Real surfaces generally exhibit contact angle hysteresis. There is a range of contact angles which are stable on the surface [14]. The size of this range is dependent upon the degree of surface

i O )

LD _ J

8


heterogeneity and surface roughness. The smallest of this range is termed the receding angle, 0r, and the largest the advancing angle, 0 a. If the liquid is withdrawn from a drop initially at a contact angle greater than 0r, the contact angle will decrease, while the drop periphery remains stationary, until 0 r is reached. If further liquid is withdrawn, the drop will no longer be stable and the drop periphery will retreat to maintain the contact angle at 0r.

(a) (b) Figure 4 - Radiograph of aluminium on a carbon/TiB2 composite in a cryolite

-alumina melt, at 1000°C, magnification 2.25X. Test time (a) zero, (b) 100 minutes.

As the contact angle tests proceeded, the aluminium drop decreased in size as shown by comparison of the radiographs in FIG. 4 (a) and (b). In this case the drop volume decreased from about 204 to 65 mm3 over a period of 100 minutes, as calculated from the drop diameters. The loss of aluminium may be due to formation of sodium via the reaction,

3NaF + Al -> A1F3 + 3Na

or by A14C3 formation or by dissolution of aluminium into the melt. This loss of aluminium was also observed in cases where there was no wetting, suggesting loss by penetration into the substrate was not probable.

The contact angle decrease observed on the carbon/TiB2 samples was characterised by an advancement of the aluminium drop periphery. This suggests that although removal of aluminium from the drop may have contributed to contact angle change, the dominant mechanism for the contact angle decrease was an increase in the substrate wettability via time dependent removal of surface contamination.

No time dependency was observed for the contact angles measured on the graphite or carbonaceous cement samples although loss of aluminium was observed. In these tests the drop peripheries retreated as the drop volume decreased and the contact angle remained constant. This suggests that, due to the homogeneous nature of the graphite and carbonaceous cement, the extent of contact angle hysteresis on these samples was small i.e. the values of 0 a and 0 r were similar.


Equilibrium Contact Angle and Complete Wetting

A decrease in contact angle over time and spreading of the aluminium was observed for all of the carbon/TiB2 samples. The decrease in contact angle was more rapid the higher the TiB2 content of the composite. This implies that the wettability of the composite increased as the TiB2 content increased.

At the maximum test time the contact angles were either still decreasing or had levelled out to a value of about 20° due to the confinement of the crucible walls. The final or equilibrium contact angles were therefore unknown. To assess if further reduction in the contact angles would occur, tests were conducted using samples of composite containing 62 and 70 mass % TiB2, of diameter smaller than the diameter of the crucible, as done with the hot-pressed TiB2 samples.

30 40 50 TIME (minutes)

Figure 5 - Contact angle versus time for carbon/TiB2 composites containing 62 and 70 mass % TiB2, at 1000°C (small diameter samples).

The results of these tests are given in FIG. 5. The measured contact angles showed a decrease over time similar to that observed in tests with samples of larger diameter, until a contact angle of about 45°. This contact angle corresponded to the aluminium having spread to the edges of the smaller sample. The contact angle continued to decrease. At a contact angle of about 8°, the point of contact between the aluminium and the composite surface could no longer be resolved. Examination of the sectioned crucibles revealed that the aluminium had spread completely over the top and down the sides of the composite surface. This indicates that these carbon/TiB2 samples were completely wetted, contact angle of zero.

CO CD

9> 9 LU

I

CD

I O o


Similar tests were not conducted on the carbon/TiB2 composites containing 45 and 54 mass % TiB2 due to the constraint on the maximum test time. The equilibrium contact angle formed by aluminium on these composites was therefore not determined and it is uncertain whether they would achieve complete wetting given sufficient time.

Liao and Liu [10] measured the contact angles formed between aluminium and carbon/TiB2 composite materials in cryolite-alumina melts. The composites they studied contained 40 and 60 mass % TiB2. They did not report the contact angle as a function of time or refer to any time dependency for the contact angles. Their technique was similar to that used in the present study. However, in the present study the radiograph exposure time was 10 seconds whereas Liao and Liu used an exposure time of 8 minutes. The sensitivity of a technique that employs such a long exposure time may be questionable.

CONCLUSIONS

From the present study, the following conclusions are made: 1. Graphite and a carbonaceous cement were not wetted by aluminium in a cryolite-alumina

melt. The average contact angles were in the range 144°-158° and were independent of time and temperature in the range studied (980-1010°C).

2. Hot-pressed TiB2 was found to be completely wetted, contact angle of zero, by aluminium in a cryolite-alumina melt, at 980°C.

3. The contact angle formed by aluminium on carbon/TiB2 composites in a cryolite-alumina melt at 1000°C exhibited time dependency. It was proposed the time dependency was due to: (i) removal of contamination from the composite surface; (ii) removal of aluminium from the drop.

4. The wettability of the carbon/TiB2 composite by aluminium in a cryolite-alumina melt at 1000°C increased as the TiB2 content increased. Complete wetting, contact angle of zero, was observed within 90 minutes on composites containing 62 and 70 mass % TiB2.

ACKNOWLEDGEMENTS

One of the authors (K.D.W.) wishes to thank Comalco Aluminium Ltd., Australia for Financial support. Technical discussions with R. Shaw, D. Juric and G.J. Houston were also greatly appreciated.

REFERENCES

1. J.B. Todd: Journal of Metals, 1981, Sept., pp. 42-45.

2. K. Grjotheim and B.J. Welch: Aluminium Smelter Technology, Aluminium-Verlag, Dusseldorf, 1988.

3. R.C. Dorward: Journal of Applied Electrochemistry, 1983, vol.13, pp. 569-575.

4. K. Billehaug and H.A. Oye: Aluminium, 1980, vol. 56, No. 10, pp. 642-648, No. 11, pp. 713-718.

5. L.G. Boxall, A.V. Cooke and H.W. Hayden: Light Metals 1984, A.I.M.E., pp. 573-588.

6. J.T. Gee, K.W. Tucker, L.A. Joo, D.V. Stewart, T. Alcorn and A. Tabereaux: U.S. Dept. of Energy Report, No. DOE/ID/12689-1, 1989, Prepared by Great Lakes Research and Reynolds Metals Company.

7. C.E. Ransley: Journal of Metals, 1962, vol. 14, No. 2, pp. 129-135.


8. S.K. Rhee: Journal of the American Ceramic Society. 1970, vol. 53, No. 7, pp. 386-389.

9. G.V. Samsonov, A.D. Panasyuk and M.S. Borovikova: Poroshkovaya Metallurgiya, 1973, No. 6 (126), pp. 51-57.

10. X. Liao and Y. Liu: Light Metals 1990, A.I.M.E., pp. 409-412.

11. T.A. Utigard: PhJD. Thesis, 1985, University of Toronto, Toronto, Ontario, Canada.

12. T.A. Utigard and J.M. Toguri: Metall. Trans. B, 1985, vol. 16B, pp. 333-338.

13. H. John and H. Hausner: Journal of Material Science Letters, 1986, vol. 5, pp. 549-551.

14. R.E. Johnson and R.H. Dettre: Amer. Chem. Soc., 1963, No. 43, pp. 112-115.

131

Cathode heat flux measurements

W. Posey, N.E. Richards, A.T. Tabereaux Manufacturing Technology Laboratory, Reynolds Materials Company, Sheffield, Alabama, U.S.A.

Abstract

Heat flux from heated surfaces can be obtained by (1) calculating heat transfer equations using measurements of surface temperature and associated parameters, (2) determining the difference in temperatures across a known thermal conductivity material, and (3) measuring directly with a calibrated heat flux meter. In this work, heat flux obtained using a commercial heat flux meter is compared with heat flux values calculated from surface temperature measurements of vertical and horizontal steel surfaces as a function of the air velocities under controlled conditions.

Keywords

Energy balance, heat flux, heat flux meter, heat losses, heat transfer, heat transfer coefficient.

Introduction

The measurement of heat losses is of great importance in determining the overall cell energy balance for specific reduction cell designs and/or operations. For example, in center-worked cells the cathode side ledge freeze profile is determined from the cells' thermal balance. The operation of the cell without a side ledge freeze leads to the shortening of pot life. Consideration must be giving to the reduction of cell heat loss and heat distribution to keep the thermal energy balance corresponding to the lower cell voltage operation in new or retrofitted cells using energy-saving designs and materials.

A number of reduction cell heat balance studies have been reported to date using a commercial heat flux meter, specifically the Shoterm HFMR developed by Showa Denko K.K. (1 - 4), which has a reported accuracy of 2 - 3% of the indicated value. The heat flux meter simplifies the measurement of heat flux by measuring the small temperature difference across a heat flow transducer of a known thermal conductivity that has been acclimatized with the temperature of a heated surface. The transducer should not be an insulator or it will be bypassed by the heat flow, and the sensor thickness must be small relative to its surface area.

The heat flux meter is widely used in the aluminum industry because of its ease of installation and direct output of heat transfer results. It eliminates the need to analyze the complex contribution of thermal properties and geometry of the surface materials, outside thermal characteristics, and conditions which may restrain surface heat transfer mechanisms.

However, no data has been presented evaluating the accuracy of the heat flux meters under analogous conditions experienced with reduction cell cathode measurements. The objective of this work was to validate the accuracy of the results obtained with the heat flux meter on heated horizontal and vertical surfaces and at different air velocities.


Experimental Procedures

The evaluation of the Shoterm HFMR entailed the measurement and comparison of the direct output value heat flux (watts/m2) with heat flux determined from the measurement of surface temperatures and other heat transfer variables. Two interchangeable factory-calibrated probes for steel surfaces were used in the measurements.

E500B-R For surfaces with a radiation rate of about 0.9, such as those of steel articles, brick, and objects painted in colors other than silver.

Heated Surfaces

Heat flux measurements were made on the exterior surfaces of a mild steel box (12.5 in. tall x 20 in. wide x 48.25 in. long) lined with firebricks and heated on the inside with a gas burner at one end of the box. Individual surfaces of the box were maintained at test conditions of 100°, 200°, 300°, and 400°C, and at a range of wind velocity from 0 to 6 meters per second (m/s). The range of temperature and air velocities were chosen to represent conditions analogous to those generally found on reduction cell cathodes surfaces.

The outside steel surface temperatures were measured at locations on the box corresponding with the heat stabilizing firebricks inside the steel box before and after each measurement with the heat flux meter. A type k thermocouple, previously calibrated against a NBS certified thermocouple, was held magnetically in contact with the steel box surface. The exact position of the direct-contact thermocouple was noted on the steel box after the surface temperature reached a constant value. The heat flux probe was then attached at the same position and heat flux measurements were taken at one-minute intervals.

Test periods were initially set at 10 minutes for measurements with the heat flux meter at the beginning of the work, but heat fluxes were found to stabilized at a constant value in 2 to 3 minutes as the work progressed, and the duration of the tests was reduced to 5 minutes. The steel surface in the area of measurements was lightly sanded prior to each test to eliminate surface scale, buildup, etc., to ensure a good contact and heat transfer between the probe and the heated steel surface.

Air Velocities

Heat flux measurements were first taken with zero wind velocities on the vertical (side), horizontal (top and bottom) surfaces of the steel box at different surface temperatures to satisfy the natural convection requirement of the heat transfer equation: (a) 0.20 for a horizontal plate facing down, (b) 0.2/ for a vertical plate, and (c) 0.38 for a horizontal plate facing up. These different heat transfer coefficients represent the ability or effect of naturally-rising air replenishing or sweeping the surface and affecting the convective heat transfer. A radiation factor of 1.0 was used in all heat transfer calculations because there were no adjacent perpendicular or parallel conducting surfaces or enclosures such as the cradles found on reduction cell cathodes.

Next, heat flux measurements were taken on the vertical (side), horizontal-up (top), and horizontal-down (bottom) surfaces of the steel box at various wind velocities. Surface and ambient temperatures were taken at the beginning and at the end of each test. Ambient air temperatures were taken with a shielded ambient thermocouple approximately 10 to 20 feet away from the heated box surfaces.

An electric motor-driven, caged-wheel fan blower was used to changes the air velocity at the surfaces of the box. The air was directed from the blower across the surface of the heated steel box by means of a 6 ft. long duct system f 11 3/16-inch diameter) constructed of six round metal buckets brazed end to end. A Pitot tube and manometer was used to conduct velocity traverses in the duct to determine the average air velocity and provide a check for two


continuous reading air velocity meters. The blower inlet was incrementally covered (3/4, 1/2, and 1/4 of the inlet opening), resulting in reduced velocities. The air temperature was measured continuously througn one of the holes in the duct.

Air velocities of approximately 6.0 m/s (meters/second) have been measured around the outside edges of the shell bottom in plant measurements for a prebake reduction cell with an open basement and with the ambient changing little year around. Zero air velocity is typically found for some sidewall areas shielded behind the bus or for either end wall of the potshell with no floor openings which also are shielded by the bus. The floor openings around the top of the sidewall around the cradle and the rising heat from the sidewall induce a velocity of approximately 3.0 m/s for the top 12 inches of the sidewall. The ambient air temperatures vary considerably and are directly proportional to stagnation in the different areas.

The following table gives the matrix of the test parameters for the 32 individual heat flux tests.

Average Velocity

Position m/sec 100

Test Number (Temperature. ° Q

200 300 400

Vertical: 0.0 1 5

2 3 4

Horizontal-up: 0.0 6 10

7 8 9

Horizontal-down: 0.0 11 12 13 14 1.021 15 16 17 2.316 18 19 2.584 21 22 20 24 4.566 25 26 27 28 6.634 29 30 31 32

Results

Incremental increases in the heated surface temperature and air velocity result in a corresponding incremental increase in the calculated heat flux (shown in Figures 2,4, 6 and 8) for the different surface positions. Thus, the calculated heat flux information, determined under very carefully controlled conditions, was used to compare and evaluate heat flux data obtained by directly using the heat flux meter.

The average heat flux values obtained for the measurements for individual tests are summarized in Tables I and II with regard to surface position, temperature, and wind velocity. An 11 to 20% average difference was found to exist for Probe No. 1 and 4 to 17% for Probe No. 2 between the heat flux calculated from surface temperature and air velocity measurements and obtained directly from the heat flux meter for the different surfaces. Compared with calculated heat flux, the heat flux from the vertical surface had almost twice the difference of the heat flux from horizontal surfaces at zero velocity.

Plots of the calculated and measured heat flux at zero wind velocity are given in Figures 1 and 2 for Probe No. 1 and in Figures 5 and 6 for Probe No. 2. Both sets of heat flux data at zero velocity have an excellent fit with a power regression equation as indicated by the high correlation coefficients.


TABLE 1. HEAT FLUX MEASUREMENTS - NO. 1 Air HEAT FLUX

Velocity Temp. (HFM) Calc. DIFFERENCE m/sec ° C W/m

2 W/m

z W/m %

Vertical Surface: 1 0 102 1172 994 -178 -17.9 2 0 200 4715 3173 -1543 -48.6 3 0 300 7976 6812 -1164 -17.1 4 0 400 12161 12547 386 3JL

Average -625 -20.1

Horizontal - Up Surface: 6 0 103 1344 1160 -183 -15.8 7 0 200 4290 3672 -618 -16.8 8 0 299 8413 7505 -908 -12.1 9 0 400 13490 13739 249 13.

Average -365 -10.7

Horizontal - Down Surface: 11 0 100 968 865 -103 -11.9 12 0 200 3204 2850 -354 -12.4 13 0 300 6871 6279 -592 -9.4 14 0 400 13520 11775 -1746 -14.8

Average -699 -12.2

Horizontal - Down Surface: 5 1.021 105 2154 1399 -756 -54.0

10 1.198 101 1807 1388 -419 -30.1 15 1.021 110 1926 1446 -480 -33.2 16 1.021 200 5391 3525 -1866 -52.9 17 1.021 297 10018 7029 -2988 -42.5 18 2.325 100.5 1413 1612 199 12.3 21 2.584 101.0 2768 1878 -890 -47.4 19 2.4 199.5 3430 4197 767 18.3 22 2.584 204.0 7112 5088 -2024 -39.8 20 2.584 300.0 11053 9441 -1611 -17.1 23 2.584 300.0 11053 9441 -1611 -17.1 24 2.584 400.0 14846 15821 975 6.2 25 4.566 100.0 2834 2552 -282 -11.1 26 4.566 202.0 8662 6478 -2184 -33.7 27 4.566 300.0 13845 11732 -2113 -18.0 28 4.566 400.0 16143 19071 2928 15.4 29 6.634 101.0 4888 3495 -1394 -39.9 30 6.634 201.0 9486 7937 -1549 -19.5 31 6.634 300.0 13584 14110 526 3.7 32 6.634 400.0 16920 22364 5444 24.3

Average -466 -18.8

134


TABLE II. HEAT FLUX MEASUREMENTS - NO. 2 Air HEAT FLUX

Velocity Temp. (HFM) Calc. DIFFERENCE m/sec °c W/m W / m

1 W/m

2 %

Vertical Surface: 1 0 101 961 994 33 3.4 2 0 200 4421 3173 -1248 -28.2 3 0 300 8624 6812 -1812 -21.0 4 0 400 14135 12547 -1588 -11.2

Average -1154 -14.3

Horizontal - Up Surface: 6 0 102 1169 1160 -9 -0.7 7 0 200 3780 3672 -108 -2.8 8 0 300 8117 7505 -611 -7.5 9 0 400 14937 13739 -1198 -8.0

Average -481 -4.8

Horizontal - Down Surface: 11 0 101 700 865 165 23.6 12 0 202.5 2754 2850 96 3.5 13 0 300 6590 6280 -310 -4.7 14 0 400 12647 11775 -872 -6.9

Average -230 3.9

Horizontal - Down Surface: 5 1.021 106 1553 1399 -154 -9.9

10 1.198 98.5 1457 1388 -69 -4.7 15 1.021 110 2047 1445 -602 -29.4 16 1.021 205 5980 3672 -2308 -38.6 17 1.021 299 8674 6926 -1748 -20.2 18 2.325 100.5 1402 1578 176 12.6 21 2.584 101.0 2884 1878 -1006 -34.9 19 2.4 198.0 3454 4313 860 24.9 22 2.584 207.5 6834 5219 -1615 -23.6 20 2.584 297.0 13105 9290 -3815 -29.1 23 2.584 298.0 13105 9290 -3815 -29.1 24 2.584 400.0 17247 15819 -1428 -8.3 25 4.566 100.0 3887 2552 -1335 -34.3 26 4.566 202.0 8334 6478 -1856 -22.3 27 4.566 300.0 16246 11723 -4523 -27.8 28 4.566 400.0 22473 19070 -3403 -15.1 29 6.634 101.0 5298 3479 -1819 -34.3 30 6.634 201.0 9764 7937 -1827 -18.7 31 6.634 300.0 15721 14106 -1615 -10.3 32 6.634 400.0 20597 22364 1767

Average -1507 -17.2

135

FIG

URE

1. H

EAT

FLUX

AT

ZER

O V

ELO

CITY

.

HEAT

FLU

X M

ETER

50

100 150 200 250 300 350 400 450

SURF

ACE

TEM

PERA

TURE

, C

FIG

UR

E 2.

HEA

T FL

UX A

T ZE

RO

VEL

OC

ITY.

CAL

CU

ATED

FR

OM

MEA

SUR

EMEN

TS

ffl

16>°

°0 n

1

1

1

1

1

1

rri

%

50

100 150 200 250 300 350 400 450

SURF

ACE

TEM

PERA

TURE

, °C

W/m'xrnd 1V3H

VE

RTI

CA

L

—Q

—

HO

RIZ

ON

TA

L -

UP

HO

RIZ

OTA

L -

DO

WN

VE

RT

ICA

L

—B

—

HO

RI2

0N

TA

L-

UP

HO

RIZ

ON

TA

L -

DO

WN


FIG

UR

E 3

. HE

AT

FL

UX

WIT

H V

EL

OC

ITY

A

ND

TE

MP

ER

AT

UR

E.

FIG

UR

E 4

. H

EA

T F

LU

X W

ITH

VE

LC

ITY

A

ND

TE

MP

ER

AT

UR

E.

50

10

0

15

0

20

0

25

0

30

0

35

0

40

0

45

0

50

1

00

1

50

2

00

2

50

3

00

3

50

4

00

4

50

SURF

ACE

TEM

PERA

TURE

, C

SURF

ACE

TEM

PERA

TURE

,°C

HE

AT

FLU

X

ME

TE

R

CA

LC

UL

AT

ED

F

RO

M

ME

AS

UR

EM

EN

TS

X

H

2

o z s TI

z z o o i

o o o r O ac

H I

LU/AA 'vrn-i iv^H

?LU/M'Xm=!lV3H

FIG

UR

E 5

. HE

AT

FL

UX

AT

ZE

RO

VE

LO

CIT

Y

FIG

UR

E 6

. HE

AT

FL

UX

AT

ZE

RO

VE

LO

CIT

Y

c

HE

AT

FL

UX

ME

TE

R

CA

LC

UA

TE

D F

RO

M M

EA

SU

RE

ME

NT

S

SU

RF

AC

E T

EM

PE

RA

TU

RE

,*C

S

UR

FA

CE

TE

MP

ER

AT

UR

E,

C

HEAT FLUX, W/m z

E

X"

LL

£<

I

U'H

UB

t:

VE

RT

ICA

L

—{

3—

HO

RIZ

ON

TA

L -

LP

--

A--

HO

RIZ

ON

TA

L-D

OW

N

....

0..

.

VE

RT

ICA

L

—E

3—

HO

RIZ

ON

TA

L -

UP

HO

Ri;

:ON

TA

L-D

OW

N


FIG

UR

E 7

. H

EA

T F

LU

X W

ITH

VE

LO

CIT

Y

AN

D T

EM

PE

RA

TU

RE

.

HE

AT

FL

UX

ME

TE

R

FIG

UR

E 8

. H

EA

T F

LU

X W

ITH

VE

LO

CIT

Y

AN

D T

EM

PE

RA

TU

RE

.

CA

LC

UA

TE

D F

RO

M

ME

AS

UR

EM

EN

TS

25,0

00

25

,00

0

50

100

150

20

0 2

50

30

0 3

50

40

0 4

50

SURF

ACE

TEM

PERA

TURE

,̂ 5

0 1

00

15

0 2

00

25

0 3

00

35

0 4

00

45

0

SURF

ACE

TEM

PERA

TURE

, C

I

H

o z s z z o I t-H

O

z o r O

DC

H

w

r en

5 w/M 'Xm=l 1V3H

w/M 'Xmd 1V3H

FIGUR

E 9.

HEA

T FL

UX C

OMPA

RISO

N FIG

URE

10. H

EAT

FLUX

COM

PARI

SON

WITH

VEL

OCITY

AND

TEM

PERA

TURE

. W

ITH V

ELOC

ITY A

ND T

EMPE

RATU

RE.

HEAT

FLU

X (H

EAT

FLUX

MET

ER),

W/m

2 HE

AT F

LUX

(HEA

T FL

UX M

ETER

), W

/m*


HEAT FLUX (CALCUALTED FROM MEASUREMENTS), W/m 2

FIGUR

E 11

. HEA

T FL

UX C

OMPA

RISO

N AT

ZERO

VEL

OCIT

Y.

FIGUR

E 12

. HEA

T FL

UX C

OMPA

RISO

N AT

ZERO

VEL

OCIT

Y.

2,000

6,000

10,000

14,000

2,000

6,000

10,000

14,000

HE

AT

FL

UX

(H

EA

T F

LU

X M

ET

ER

), W

/m*

HE

AT

FL

UX

(H

EA

T F

LU

X M

ET

ER

), W

/mz


HEAT FLUX (CALCULATED FROM MEASUREMENTS), W/rr

HEAT FLUX (CALCULATED FROM MEASUREMENTS), W/m *


Power Regression:

y = axb

Average Heat Flux Surface Velocity Difference. %

Probe No. 1 Probe No. 2

Vertical 0 -20.1 -14.3 Horizontal-Up 0 -10.7 -4.8 Horizontal-Down 0 -12.2 -3.9 Horizontal-Down 1.02 to 6.634 -18.8 -17.2

Surface Position Method R-Square (b) (a)

Probe No. 1:

1. Vertical HFM 0.988 1.708 0.471 2. Horizontal-up HFM 0.999 1.703 0.507 3. Horizontal-down HFM 0.996 1.877 0.164 4. Vertical Calc. 0.998 1.843 0.192 5. Horizontal-up Calc. 0.999 1.807 0.152 6. Horizontal-down Calc. 0.997 1.869 0.152

Probe No. 2:

1. Vertical HFM 0.994 1.954 0.124 2. Horizontal-up HFM 0.998 1.852 0.217 3. Horizontal-down HFM 0.999 2.098 0.042

1. Vertical Calc. 0.998 1.829 0.207 2. Horizontal-up Calc. 0.998 1.792 0.285 3. Horizontal-down Calc. 0.997 1.885 0.138

Plots of the calculated and measured heat flux data at wind velocities grater than zero on a horizontal-down (bottom) surface are given in Figures 3 and 4 for Probe No. 1 and in Figures 7 and 8 for Probe No. 2. Both sets of heat flux data have an excellent fit with power regression equation as indicated by the high correlation coefficients. However, a considerable divergence is evident in the heat flux data obtained directly with the heat flux meter compared with the calculated heat flux data, as shown in Figures 3 ,4 ,7 and 8.

A direct comparison between the calculated and measured heat flux information, plotted in Figures 11 and 12, demonstrates that there is a good correlation between heat fluxes obtained by two methods at zero velocity for all three surface positions; an increase in divergence in heat flux results (particularly at higher heat fluxes) is evident at air velocity


Velocity, m/sec Method R-Square (b) (a)

Probe No. 1-Horizontal-Down:

1. 1.021 HFM 0.992 1.572 1.300 2. 2.584 HFM 0.997 1.229 9.817 3. 4.566 HFM 0.974 1.293 7.990 4. 6.634 HFM 0.998 0.909 74.59

1. 1.021 Calc. 0.996 1.518 1.192 2. 2.584 Calc 0.997 1.536 1.517 3. 4.566 Calc. 0.997 1.440 3.271 4. 6.634 Calc. 0.996 1.338 7.000

Probe No. 2:

1. 1.021 HFM 0.978 1.655 0.768 2. 2.584 HFM 0.992 1.334 5.987 3. 4.566 HFM 0.992 1.288 9.927 4. 6.634 HFM 0.996 0.997 52.12

1. 1.021 Calc. 0.995 1.448 1.388 2. 2.584 Calc. 0.997 1.535 1.525 3. 4.566 Calc. 0.997 1.439 3.275 4. 6.634 Calc. 0.996 1.342 6.867

greater than zero, as shown in Figures 9 and 10. Heat flux data obtained using the heat flux meter is less than heat flux data calculated from direct measurements.

Discussion

Heat Transfer Calculations

In past reduction cell thermal studies, heat flux from cathode surfaces has been calculated using the classical heat transfer equations (1) involving both natural and forced convection and radiation heat transfer from the outside surfaces to the surrounding air:

Natural Convection: Q/A = [c(Ts-Ta)i25] + [eEF(Ts4-Ta4)]

Forced Convection: Q/A = [(l+0.225V)(Ts-Ta)] + [eEF(Ts4-Ta4)]

where:

Q/A = Heat loss per unit surface area (W/m2) c = Convection constant,

0.20 - horizontal downward surfaces 0.27 - vertical surfaces 0.38 - horizontal upward surfaces

Ts = Surface temperature, (degrees C, R) T ac = Ambient air temperature, (degrees C, R)


E Emissivity: 0.95 - steel

F = Shap< V = Ambi

Shape factor Ambient air velocity, m/sec

Using these equations to calculate heat flux requires the laborious measurement of numerous surface temperatures, corresponding ambient temperatures, and wind velocities for each different cathode surface area. Then a step by step building block approach and meticulous documentation is required for successful measurement of the heat transfer at different locations. The problem is broken down into sufficient steps to provide intermediate solutions to evaluate the thermal effects of one feature at a time.

A problem arising during cell heat loss studies is where to measure the ambient temperature and wind velocity that corresponds to a specific cathode surface. The ambient air and velocity conditions are normally quite diverse and unique around each reduction cell; for example, reduction cells may be enclosed in a pit with forced air circulation or will have an open basement arrangement. The effect of heated air rising from the lower sections diluting the cooler ambient air must be accessed. As hot air rises the ability of this air enveloping the cell to absorb more heat on its ascent also depends on side convective patterns. The rising hot air causes heat transfer restraints which in term cause the exterior shell temperatures to increase without an increase in heat transfer.

The problem of where to measure ambient air temperature and velocity is partially resolved by the calculation of a radiation geometrical shape factor. This factor represents that percentage of heat transfer that occurs from a source to a given ambient "window" and is typically calculated for an area encompassed by the leading edge of the sidewall or bottom cradle flanges of cathodes.

1. The heat flux meter readings for each sensor head stabilized just under 3 minutes for the constant test environment. The results were steady and not varying due to the constant test conditions.

2. The heat flux meter results were continuous in their response to increases in surface temperature and velocity. Thus, the regression calibration formulas showed excellent correlations.

3. With zero air velocities, heat flux meter results demonstrate an excellent correlation with calculated heat flux results. The average heat flux meter results at zero velocity were 563 W/m2 less than average calculated results for Probe No. 1 and 627 W/m2 less than average calculated results for Probe No. 2.

4. With air velocities greater than zero, heat flux meter results demonstrate a higher divergence from calculated heat flux results, particularly at higher heat fluxes. The average heat flux meter data at velocities greater than zero were 466 W/m2 less than average calculated for Probe No. 1 and 1507 W/m2 less than average calculated results for Probe No. 2.

5. The use of the heat flux meter may provide an advantage in some situations as the potroom ambient air and velocity conditions are quite diverse and unique around each reduction cell.

Conclusions


References

K. Ari and K. Yamazaki, "Heat Balance and Thermal Losses in Advanced Prebake Cells." Light Metals. 193 (1981).

H. Tsukahara, N. Ono, and K. Fujita, "Establishment of Effective Operation of Prebake Anode Pots," Light Metals. 471 (1982).

M. P. Taylor, B. J. Welch, and J. T. Keniry, "Influence of Changing Process Conditions on the Heat Transfer During the Early Life of an Operating Cell," Light Metals. 437 (1983).

T. Ohta and T. Matsushima, "Thermal Analysis of Soderberg Pots," Light Metals. 689 (1984).

J. J. Marino, "Thermal Design of Refractory and Insulating Systems," Industrial Heating (May 1980).

1.

2.

3.

4.

5.

147

Strontium extraction by aluminothermic reduction

J. Langlais, R. Harris Department of Mining and Metallurgical Engineering, McGill University, Montreal, Quebec, Canada

ABSTRACT

Experiments have been performed in a mechanically agitated molten metal bath to extract strontium from strontium carbonate by aluminothermic reduction. The molten bath was formed by an excess of molten aluminum reductant with additions of magnesium or bismuth. Magnesium was added to the melt for two reasons: 1) changing the melt chemistry to increase the oxygen affinity and 2) lowering the surface tension. Bismuth was added to lower the surface tension without increasing the oxygen affinity. The experimental conditions were: temperature; 1000 °C, pressure; 1 atm and time; 1 hour. Results showed that with a Mg:Sr molar ratio of 10:1, strontium recovery to the excess aluminium was greater than 50%. Comparison between experiments with magnesium and bitsmuth showed that the chemical effect of magnesium was predominant over the surface tension effect of bismuth or magnesium.

KEYWORDS

Strontium, extraction, strontium carbonate, extraction, aluminothermic reduction, excess molten reductant, magnesium, bismuth addition, reaction mechanism, surface tension, oxygen affinity.

INTRODUCTION

Strontium metal has found an increased application in the automobile and aerospace industries since 1970's which has resulted in a growing demand for the metal. The present method of strontium production uses the vacuum retort process. The inherent disadvantages of the retorting process, including high labour requirements and low production rates due to the slow heating and cooling of the vacuum retort, vouch for the developement of an alternative, economical and simpler process for strontium extraction. An alternative process has been developed in our laboratory for the extraction of strontium from

strontium carbonate, SrC03, by metallothermic reduction using the "melt-leach" technique

111.

The melt-leach method consists of contacting the ore or concentrate with an excess molten metal acting as a reductant and a lixiviant. The metal extracted from the source material is dissolved into the molten metallic solvent. The dissolved metal is extracted as a vapour from the excess molten reductant by vacuum distillation

1 2 , 3 , 4 , 5 , 6 , 7 1. Aluminium was selected as the metallic

reductant to extract strontium from strontium carbonate according to the following reaction:

3SrCO. + XSAl -> { 3Sr + (XS - 2)Al ) A u nY + ALO, + 3 CO, (1) 3 v ' ALLOY 2 IsOLID

2GAS


Previous investigations on Li extraction from spodumene, Li02

,Al2034Si02

[ 8 1, found that

Mg enhanced the rate and extent of extraction by excess molten Al. It was concluded that the beneficial effect was due to Mg lowering the surface tension and hence improving the wetting of the melt. However, this conclusion was not certain, thus during the process development, magnesium and bismuth were added separately in two sets of experiments in order to change the physical and/or chemical properties of the excess molten reductant. Magnesium was added to lower the surface tension of the melt and to increase the oxygen affinity. Bismuth was added to the melt to lower the surface tension without increasing the oxygen affinity.


The reactants used in the experiments were as follows: Aluminum*: Commercial grade at 99.97% purity; Magnesium

b: Commercial grade at 99.9% purity; Bismuth

0: Pure granules

of bismuth with nominal composition of 99.9%; SrCOa

d: Pure precipitated SrC03, Grade

AnalaR. The basic procedure for each experiment consisted of reacting 50 grams of strontium carbonate with approximately 915 grams of aluminum at 1000 °C, 1 atm for 60 minutes. The corresponding aluminum to strontium molar ratio was 100:1 for each experiment. Parameters which were changed from one experiment to another were the amount of magnesium and bismuth added. The amount of magnesium added was increased gradually and was reported as the molar ratio of magnesium to strontium (0, 2, 4, 6, 8, 10 Mg:Sr). The amount of bismuth added was determined by the Figure 1 to lower the surface tension to the same extent as the magnesium additions. Additionally, one experiment was performed with a large amount of bismuth, 6 wt%, to lower the surface tension much below the value reached with the highest magnesium addition. These experiments are listed in Table I. A schematic representation of the experimental setup for strontium carbonate reduction is shown in Figure 2. Charging for the first set of experiments consisted of loading the crucible with the aluminum cut in small pieces for faster melting. The furnace power was set to 30 kW and the material was completely melted after approximately 30 minutes. The time to reach the target temperature of 1000 °C took another 30 minutes. Once the target temperature was reached, small pieces of magnesium were added and the impeller with the crucible cap were put in place and the stirring started. During melting, argon was continuously flushed inside the reactor to minimize oxidation. Immersion of the cold impeller into the melt slightly lowered the melt temperature. Once the temperature reached 1000 °C again, the furnace was set at about 19 kW, the strontium carbonate was added via the utility hole. At this point, the reduction experiment began for one hour. To keep the temperature stable, measurements were taken every two to three minutes via a thermocouple imbedded in the base of the crucible and checked with a handheld thermocouple each 10 minutes. The furnace power was periodically adjusted in order to keep the target temperature. For the kinetic studies, samples were taken at twenty minute intervals via the utility hole. Sixty minutes after the strontium carbonate addition, stirring was stopped, the power shut off and the crucible cap, the impeller and the bogey removed. The crucible cap was reinstalled and argon gas kept flowing until the temperature reached 750 °C. At the end of stirring, the upper part of the crucible was found to be covered by a powdered reaction product and other viscous materials which were scrapped out

a Supplied by ALCOA, Pittsburg, U.S.A. b Supplied by Timminco, Haley, Ontario 0 Supplied by American Chemicals Ltd. d Supplied by BDH Chemicals Inc., Montreal, Qc


Table I - Experimental Program

REACTANTS (g) REACTANTS (mole) Molar Add. Ratio

Exp.# Al Mg SrC03 Al Mg SrC03 Mg:Sr Wt%

0 914.6 0 50 33.9 0 0.34 0 0 1 914.6 16.5 50 33.9 0.68 0.34 2 1.8 2 914.6 32.9 50 33.9 1.35 0.34 4 3.5 3 914.6 49.4 50 33.9 2.03 0.34 6 5.1 4 914.6 66.0 50 33.9 2.72 0.34 8 6.7 5 914.6 82.4 50 33.9 3.39 0.34 10 8.3

Exp.# Al Bi SrC03 Al Bi SrO03 Bi:Sr Wt%

6 914.6 9.2 50 33.9 0.04 0.34 0.12 1 7 914.6 27.7 50 33.9 0.13 0.34 0.38 3 8 914.6 55.4 50 33.9 0.26 0.34 0.76 6

700

i Cu A z n

n

s b

P b

1 Bi 1 1 3 4 5

Solute (wt%)

1.INDUCTION FURNACE 2.IMPELLER 3.CAP 4.ARGON 5.IMPELLER SHAFT 6.DRIVING CHAIN

7.MOTOR 8.GEARBOX 9.CONTROL 10.SUPPORT 11.CRUCIBLE 12.THERMOCOUPLES

Figure 1 - Surface tensions for aluminum with solute metals at 50 to 80 °C above liquidus

191 Figure 2 - Schematic representation of the experimental setup

and placed in a container. The melt was then poured into a steel mold. After cooling of the crucible, thin metal films which had clung on the crucible wall and on the impeller and solidified were removed and added to the products. Before running another experiment, the impeller was repaired with alumina cement.


The reaction products under the present experimental conditions, fall into four main product categories:

- Ingot - Dross - Powder Residue - Condensate

The reaction products were analyzed to determine their strontium content. The analytical instruments used were: X-Ray Diffractometer, Atomic Absorption Spectrophotometer, Optical Microscope and Scanning Electron Microscope. Wet chemical analysis was the principal method of analysis. For kinetic evaluation, samples were taken from the melt every 20 minutes. Chemical analysis were performed on these samples for strontium content.

SAMPLE PREPARATION

The sampling method was different for each of the four reaction products obtained namely: Ingot, Dross, Powder Residue and Condensate. The samples from the ingots were obtained by drilling to produce shavings from various areas in order to prevent or minimize segregation. Pieces of dross were taken randomly and cut into smaller ones to form the sample. The powder residue was mixed and a small sample withdrawn randomly and cumulatively to obtain the sample. The condensate was collected from the crucible cap (condenser) by brushing off the very fine dust.

ATOMIC ABSORPTION ASSAYS

Atomic absorption spectrophotometry was used to chemically assay all reaction products. This method is well suited to the determination of strontium and many other metals and is very sensitive although the spectral and chemical interferences for the element to be analyzed must be known. Literature reveals that the results obtained are rigorously dependent on the analytical conditions. Possible interferences encountered with atomic absorption are; spectral interferences, ionization interferences, refractory compound formation, matrix effects and physical characteristics of the solution. Therefore, it is important to know the characteristics of each element to prepare standard and sample in order to obtain optimum results. During strontium analysis, care had to be taken in order to minimize interferences and the depression due to the absorbance by aluminum. Previous research

[10] found that alkali metals added to a strontium

solution will enhance the signal. All solutions were prepared with HC1 and H2S 04 which showed no effect on the absorbance of strontium

1101. The interference caused by aluminum was controlled

by adding 1% w/v (2.67 g of LaCl3-7H20 per 100 ml of solution) lanthanum

110'

11'

121. The

presence of magnesium had positive interference for strontium analysis

[ 1 U 2'

1 3 , 1 4 1. Bismuth had no

effect on the strontium assays

1151. The atomic absorption results for all reaction products and for

kinetic evaluation are listed in Table II to V.


Table II - Assays of Ingot

Exp. # Mg:Sr cSr ACSr Rel. Err. ACMg Rel. Err.

Molar (wt%) (wt%) (%) (wt%) (wt%) (%) Ratio

0 0 0.08 0.007 8.3 _ - _

1 2 0.53 0.047 9.0 1.24 0.169 13.6 2 4 0.93 0.047 5.1 2.79 0.169 6.0 3 6 1.09 0.047 4.3 4.46 0.178 4.0 4 8 1.26 0.048 3.8 5.33 0.169 3.2 5

10 1.97 0.068 3.4 7.49 0.179 2.4

Exp.# Bi Add. cSr ACSr Rel. Err. cBi

ACB1 Rel. Err. (wt%) (wt%) (wt%) (%) (wt%) (wt%) (%)

6 1 0.95 0.048 5.0 0.87 0.194 22.3 7 3 0.48 0.048 9.8 2.61 0.153 5.9 8 6 0.85 0.048 5.6 5.08 0.153 3.0

Table in - Assays of Dross

Exp. # Mg:Sr cSr ACSr Rel. Err. ACMg Rel. Err.

Molar (wt%) (wt%) (%) (wt%) (wt%) (%) Ratio

0 0 5.15 0.134 2.6 _ _ _

1 2 6.34 0.066 1.0 3.33 0.333 9.9 2 4 6.66 0.122 1.8 4.34 0.345 7.9 3 6 6.46 0.095 1.5 4.83 0.338 7.0 4 8 7.04 0.107 1.5 5.81 0.326 5.6 5 10 7.39 0.152 2.1 6.44 0.342 5.3

Exp.# Bi Add. cSr ACSr Rel. Err. cBi

ACBi Rel. Err. (wt%) (wt%) (wt%) (%) (wt%) (wt%) (%)

6 1 5.27 0.094 1.8 0.62 0.055 8.9

1 3 8.56 0.110 1.3 0.83 0.070 8.4

1 8 6 12.38 0.110 0.9 0.76 0.070 9.2


Table IV - Assays of Powder

Exp. # Mg:Sr c sr ACSr Rel. Err. ACMg Rel. Err. Molar (wt%) (wt%) (%) (wt%) (wt%) (%) Ratio

0 0 11.26 1.40 3.7 _ _ _

1 2 37.42 1.39 3.8 2.98 0.421 14.1 2 4 42.21 1.61 2.3 4.60 0.410 8.9 3 6 50.83 1.18 5.7 4.36 0.422 9.7 4 8 24.59 1.40 11.0 7.22 0.422 5.8 5 10 33.29 3.65 6.4 13.52 1.127 8.3

Exp.# Bi Add. cSr ACSr Rel. Err. cBi ACBi Rel. Err.

(wt%) (wt%) (wt%) (%) (wt%) (wt%) (%)

6 1 21.72 1.39 6.4 5.27 1.05 19.9 7 3 24.58 1.18 4.8 5.12 1.07 21.0 8 6 37.96 6.98 18.4 16.26 3.88 23.9

Table V - Strontium Assays from Kinetics

Exp.# Mg:Sr 20 minutes 40 minutes 60 minutes Molar Ratio

CSr (wt%) CSr (wt%) CSr (wt%)

0 0 0.12 0.17 0.08 1 2 0.29 0.37 0.53 2 4 0.25 0.34 0.93 3 6 0.24 0.49 1.09 4 8 0.15 0.57 1.26 5 10 0.33 0.72 1.97

Exp.# Bi Add. (wt%)

6 1 0.29 0.62 0.95 7 3 0.26 0.73 0.49 8 6 0.36 0.57 0.85


XRD ANALYSIS OF POWDER RESIDUE

The powder residue was analyzed by X-Ray Diffraction (XRD) for qualitative identification of the crystalline compounds. The automated Phillips APD 1700

[ 1 61 system

searched various combinations of the more intense diffraction lines until a match was found. All species that "scored high enough" to be positively identified by the Phillips system are noted by an asterisk* and are tabulated by name, compound formula and identification number, Table VI. The XRD analysis showed that the powder residue was composed of various compounds.

Table VI - XRD Results for Powder Residue

EXPERIMENT #

SPECIES 0 1 3 5 6 8

Aluminum, Al 4-787 * * * * * *

Strontianite, Syn. SrC03 5-418

* * * * * *

Corundum, Syn. A1203 10-173

* * * * * *

Aluminum Carbide A14C3 11-629

* * * * * *

Spinel, Syn. MgO-A1203 21-1152

* * *

Strontium Oxide SrO 27-1304

* * *

Periclase, Syn. MgO 4-829

*

Bismuth Oxide Bi203 18-244

* *

* Positively Identified

The compounds present in the powder residue included: Strontium Carbonate, Aluminum, Alumina, Aluminum Carbide, Spinel, Strontium Oxide, Periclase and Bismuth Oxide. XRD analysis was also performed on the condensate. Results showed only the presence of synthetic periclase and bismuth oxide in Experiments 1 to 5 and 6 to 8, respectively. The condensates were obviously formed by evaporation of the respective metal followed by condensation and oxidation. Some vapour escaped through various openings in the crucible cap. There was no evidence of strontium metal or strontium compounds in the condensate.


OPTICAL MICROSCOPE EXAMINATIONS

Polished ingot samples were examined under an optical microscope in order to visualize the change in the microstructure with the change in the strontium content. Figures 3 to 6 show the microstructure of the ingots. It can be observed that more intermetallic phases were present in the sample as the Mg.Sr molar ratio increased. The presence of intermetallic phases was also observed for the bismuth additions but in very small quantity. SEM analysis confirmed the presence of strontium in the intermetallic phases (see next Section). This implies that strontium carbonate was reduced to strontium which dissolved into the aluminum-magnesium reductant. It can be seen that increasing the MgrSr molar ratio improved the extraction while bismuth additions did not.

Figure 3 - Microstructure of the ingot for Figure 4 - Microstructure of the ingot for Experiment 0 (OMg.Sr) Experiment 3 (6Mg:Sr)


Figure 5 - Microstructure of the ingot for Figure 6 - Microstructure of the ingot for Experiment 5 (10Mg:Sr) Experiment 8 (6 Wt% Bi)

SCANNING ELECTRON MICROSCOPE ANALYSIS

The scanning electron microscope used was a JEOL JSM-T300[ 1 7] linked to the SQ program of the Tracor Northern TN-5400 software analysis system[ 1 8 ]. This system permitted analysis of small areas on the sample surface. The major drawback of this instrument is the detector which is protected with a beryllium window making oxygen detection impossible^91. Therefore, it was not possible to differentiate between pure species and compounds formed with oxygen. The SEM was used to examine the reaction products in order to find evidence of reaction between the strontium source material and the molten reductant. Qualitative and semi-quatitative results have been obtained from the study. The analysis of the ingot from Experiment 5 was performed to obtain more information about the intermetallic phase observed under the optical microscope. Table VII gives the average semi-quantitative analysis results obtained.

Table VII - SEM Assay of Intermetallic (Weight %)

ALUMINUM MAGNESIUM STRONTIUM

76.5 1.2 22.3


The phase observed was presumably intermetallic since it created relief during polishing (intermetallic usually shows poor ductility) permitting observation according to the optical microscope principles1201. It was found that some strontium was present in the intermetallic but the intermetallic itself was not identified. Possible intermetallics which could be formed with stontium in presence of aluminum and magnesium are identified in phase diagrams.

The powder residue examinations did not show details of the possible interface between the solid and the liquid phase. Observation of distinct reaction zones between the strontium source material (solid) and the reductant (liquid) would have provided valuable insight concerning the reaction mechanisms. Nevertheless, the dross residue showed interesting details. Figure 7 shows the surface of a piece of dross. Small particles averaging 5 |im in diameter were observed on the dross surface. Spot analysis on the SEM showed that the particles were aluminum. These particles could have been formed during the mixing and under the magnetic field of the furnace. Furthermore, these droplets were certainly oxidized during the process. The aluminum oxide layer on the droplet surface and/or the melt surface prevented wetting of the droplets at this temperature1211. These particles or nodules were found on the surface of the dross, in its cavities and folds and trapped inside. The particles which were free became part of the powder product and were recuperated at the end of the experiment. Figures 8 and 9 show the condensate obtained with magnesium addition (Experiments 0 to 5) and bismuth addition (Experiments 6 to 8), respectively. The condensate obtained with magnesium and with bismuth addition presented the same physical appearance when observed with the SEM. The condensates were made of very fine particles.

Figure 7 - Dross Surface (1500 X)

Figure 8 - Condensate for Experiments 1 to 5 (350X)

Figure 9 - Condensate for Experiments 6 to 8 (350 X)



The experimental and the thermodynamically predicted values for strontium recovery into the molten excess reductant with magnesium and bismuth additions are shown in Figure 10 and 11. The thermodynamic prediction was performed using the EQUILIB routine of the F*A*C*T program

[ 2 2 ].

Mg : Sr MOLAR RATIO Bi ADDITION (Wt %)

| EXPER. PREDICT. | | EXPER. PREDICT?

Figure 10 - Strontium Recovery vs. Mg.Sr Figure 11 - Strontium Recovery with Molar Ratio Bismuth Addition

It can be observed that the predicted values agree with the experimental values. The strontium recovery increased with the Mg.Sr molar ratio. It reaches a maximum of 51.3% for a Mg.Sr molar ratio of 10:1. It can be observed that the strontium recovery was improved for one wt% bismuth addition. However, the strontium recovery stays almost unchanged up to 6 wt% bismuth addition. Generally, the strontium recovery was not improved significantly with bismuth addition over the range one to six weight percent. However, strontium recovery was greatly improved by increasing the Mg:Sr molar ratio and could be explained by the following: a) Lower melt surface tension; b) Higher oxygen affinity of the melt; c) Lower activity coefficient of strontium; d) Lower activity of A1203 in the reaction products due to spinel formation.

By increasing the amount of magnesium (solute metal), the surface tension of the melt (solvent metal) was lowered significantly, Figure 1, and it can be summised that the lower surface tension favoured wetting and so improved the solid/liquid contact and enhanced reactions taking place at the interface and increased the amount of strontium dissolved into the excess molten reductant. However, Figure 11 shows that no significant improvement in strontium recovery was obtained for a bismuth addition above one wt%. Figure 1 also showed that a lower melt surface tension was obtained with bismuth content of 6 wt% than with a Mg:Sr molar ratio of 10:1 (8.3 wt% Mg). Therefore, the surface tension was not the only factor responsible for the

Sr R

EC

OV

ER

Y (%

;

Sr R

EC

OV

ER

Y (%

)


improvement in strontium recovery with magnesium addition. The magnesium dissolved in the melt had a greater affinity for oxygen than aluminum

while the oxygen affinity for bismuth was very low according to the following reactions1221.

2 Mg + 02 -> 2 Mg O AG - -923 kJ @ 1000 °C (2)

i Al + 02 -» 1 AL O, 3 2 3 2 3

AG - -847/:/ @ 1000 °C (3)

iBi + 02 -> 1 BL 0 ,

3 2 3 2 3

AG - -161 kJ @ 1000 °C (4)

SrO

IMPERVIOUS LAYER

OF ALUMINUM OXIDE

Therefore, the reaction mechanism with magnesium in the melt might involve the reduction of the impervious aluminum oxide film, Figure 12, which would accelerate the wetting of the strontium oxide. Although bismuth lowered the surface tension of the melt, it would not reduce the aluminum oxide film present in the system. In addition, a larger amount of aluminum oxide is more likely to be present in the melt toward the end of an experiment than at the beginning due to oxygen contamination. This could explain the increase in strontium recovery, promoted by the addition of bismuth which resulted in a lower surface tension, at the beginning of the experiment. But after the first twenty minutes, the reaction rate diminished to a plateau value due to the aluminum oxide film on the liquid surface, impossible to be reduced by bismuth, preventing further reaction to take place.

Magnesium could have also reduced some strontium compounds present in the system which contributed to the final strontium recovery at lOMg.Sr molar ratio.

Another consideration was the activity of strontium which was lowered by the excess of aluminum reductant (dilution effect) and the presence of a solute metal (interaction effect). The dilution effect would be simple to verify e.g., change the amount of excess Al, but it would be more difficult to verify the interaction effect since no data are available for the following interaction coefficients, e"? or cS, with aluminum as the solvent. The Strontium-Aluminum

SOLID/LIQUID

CONTACT

SrO

NO SOLID/LIQUID

CONTACT

(NO REACTION)

Figure 12 - Schematic of the Possible Reaction Mechanism with Magnesium present in the Aluminum Reductant

system1 ,[24.25,26] was found to show large deviations from ideality which suggested a considerable stability of the intermetallic compounds formed in solid and liquid state. Experimental evidence1261 shows that the Al-Sr system exhibits negative deviations from ideality which imply that Ysi « 1. The following basic equation for aluminothermic reduction presents briefly the role of the activity coefficient on the overall reaction Gibbs free energy.


3 SrO + 2AI -» 3Sr + Al203 (5)

a

3

AG - A G0 + RT In _ i l (6)

2

According to thermodynamic definitions for the activity, and considering that Al may behave ideally, y A 1» 1, at equilibrium the Equation 3 may be rewritten,

A C - -RT In

f s A S rf (7)

[Al]

2

Here, the activity of the aluminum is approximately equal to its concentration. Thus, an increase of the strontium activity coefficient, yS r, would offset the reaction to the right ( AG becoming more positive). Furthermore, the addition of a solute metal such as magnesium or bismuth would affect the equilibrium by changing the activity coefficient of Sr. A knowledge on the interaction parameter values would give further information on the behaviour of the strontium activity coefficient in presence of solute metals. A positive value for e implies an increase in YSr which would not be favourable, while a negative value for e

M

s

9

r implies a decrease in ySr which would be favourable for the reaction.

KINETICS OF STRONTIUM REDUCTION

The kinetics of the reactions were determined by plotting the strontium recovery in the kinetic samples versus the time at which they were withdrawn from the melt. The strontium recovery to the kinetic samples was determined by the following equation:

KIN., REC. ofSr in KIN. (%) - - x 100 (8)

(Mass SrlMass m)

where KINSr : Strontium contained in the kinetic sample MassS r: Initial mass of strontium from reactant Massm : Mass of the melt

Figure 13, indicates that the rates of strontium recovery to the melt with magnesium additions increased steadily with time for the Experiments 2 to 5. It can also be observed that the rate of extraction increased with increasing the Mg:Sr molar ratio. Experiment 0 showed that the rate of extraction diminished with time. The extraction in Experiment 0 could have been constrained by the formation of an impervious aluminum oxide film already discussed in a previous section. For Experiment 1, it can be seen that the rate of extraction reached a plateau after 40 minutes. Magnesium depletion could be the cause for the reaction to stop around the 40

th minute for Experiments 0 and 1.

Figure 14 shows that the rate of extraction did not increase significantly with bismuth additions. After 40 minutes, it can be seen that there is no or little increase in the rate of extraction. The determination of the factors which are responsible for the change in the strontium extraction rate is beyond the scope of this thesis. However, the decrease in the rate of extraction after 40 minutes could have been the result of the following: a) Equilibrium of the system is


reached; b) A rate limiting step has been reached; c) Depletion of strontium from the material source; d) Depletion of magnesium.

- • - 0 Mg:Sr 2 Mg:Sr - * - 4 Mg:Sr 6 Mg:Sr 8 Mg:Sr 10Mg:Sr

20 30 40 TIME (min.)

Figure 13 - Strontium Recovery to the Melt Figure 14 - Stontium Recovery to the Melt as as a function of Time and Mg Additions a function of Time and Bismuth Additions

CONCLUSIONS

In this work the extraction of strontium from strontium carbonate by metallothermic reduction using an excess of molten aluminum reductant has been investigated. The effects of magnesium and bismuth additions to the molten aluminum have also been studied. Magnesium was added to the melt to lower the surface tension of aluminum and to increase the oxygen affinity of the melt. Bismuth was added to the melt to lower the surface tension without increasing the oxygen affinity.

The conclusions drawn from the investigation are outlined below.

1. The main finding was that magnesium addition to the molten aluminothermic reduction process improves the strontium extraction from SrC03.

2. The addition of bismuth did not improve significantly the strontium extraction. 3. The effect of magnesium and bismuth addition on the strontium extraction obtained

experimentally was similar with the thermodynamic predictions using F*A*C*T. 4. Magnesium and bismuth addition lowered the surface tension of the melt which improved

the solid/liquid contact (wetting). 5. The surface tension was not the only factor responsible for the improvement in strontium

extraction from SrC03 with magnesium addition. 6. The wetting process with magnesium could have involved the reduction of an impervious

aluminum oxide film (aluminum oxide film reduction not possible with bismuth) improving the solid/liquid contact (indispensable for reduction and dissolution process).

7. Strontium was not found in the condensate. This suggests that lower pressure would be required to volatilize strontium at the temperature of the experiments (1273 K). It also

TIME (min.)

g

! !


suggests that magnesium and strontium can be preferentially recovered. 8. The rate of reaction increased with increasing the Mg.Sr molar ratio.

ACKNOWLEDGEMENTS

The authors would like to express their gratitude to the financial support provided by the NSERC through Strategic Research Grant to carry out this work.

REFERENCES

1. R. Harris, A.E. Wraith and J. Toguri, Producing Volatile Metals, Canadian Patent Application, Ser. No.: 539,058, June 8,1987, USA Patent Application, Ser. No.: 201,446, June 2, 1988

2. Harris, R.L., Vacuum Refining of Molten Steel, Ph.D. Thesis, McGill University, 1980 3. Dimayuga, F., Vacuum Refining of Molten Aluminum, Ph.D. Thesis, mcGill University,

1980 4. Olette, M., Vacuum Distillation of Minor Elements from Liquid Ferrous Alloys, Physical

Chemistry of Process Metallurgy, Part 2, Ed. by G. St-Pierre, Interscience, N.Y., 1961, pp. 1065-87

5. Werner, E., Materials of High Vacuum Technology, Vol. 1, Metals and Metalloids, Pergamon Press, 1966, pp. 576-90

6. Volsky, A.N. et al., Refining of Strontium, Barium, Magnesium and Calcium, Session C-9, P/2050, USSR

7. Ward, A.G., The Purification of Lithium by Vacuum Distillation, J. Appl. Chem., 13, August 1963, pp. 329-34

8. Mast, E., Role of Magnesium in the Aluminothermic Reduction of Spodumene, M. Eng. Thesis, McGill University, 1989

9. Richardson, F.D., Physical Chemistry of the Melts in Metallurgy, Vol. 2, ©1974 Academic Press, London

10. Chemical Analysis, Vol. 25, ©John Wiley & Sons, 1968, pp. 164-72 11. Intonti, R., Spectrochimica Acta, Vol. 23B, 1968, pp. 437-42 12. Condylis, A., Fourth International Conference on Atomic Spectroscopy, Toronto, Ont.,

1973, p. 64, p. 115 13. Heffernan, B.J., Laboratory Methods, Vol. 83, No. 500, 1971, pp. 205-7 14. Soters, J., Atomic Absorption Method Manual, Vol.1, Instr. Lab. Inc., Washington, Ma.,

1979 15. A.A. Model IL357, Instrumentation Laboratory Manual 16. APD (Automated Powder Diffraction) 1700, Operation Manual, First ed., Phillips Corp.,

Netherlands, 1984 17. JEOL (U.S.A) Inc. 11 Dearbon Rd., Peabody, Mass., 01960 18. Tracor Northern, 2551 W. Beltline, Middleton, Wis., 1985 19. Scanning Electron Microscope Analysis of Materials, Professional Development Seminars,

McGill University, Dept. of Mining and Metallurgical Engineering, 1990 20. Havner, S.H., Introduction to Physical Metallurgy, Second Edition, McGraw-Hill, 1974,

pp. 15-24 21. Carnahan, R.D., Some Observations on the Wetting of A1203 by Aluminum, Journ. Amer.

Cer. Soc., Vol. 41, No. 9, 1958, pp. 343-7 22. F*A*C*T (Facility for the Analysis of Chemical Thermodynamics), W.T. Thompson,

A.D. Pelton and C.W. Bale, CRCT, Ecole Polytechnique, Montreal, 1988

163

Heat and mass transfer between liquid bath and solid cryolite ledge

T.A. Utigard, A. Warczok Department of Metallurgy and Materials Science, University of Toronto, Toronto, Ontario, Canada

P. Desclaux Arvida Laboratories and Experimental Engineering Centre, A lean International Ltd., Jonquiere, Quebec, Canada

ABSTRACT

In the Hal l-Heroult process for the production of aluminium, a ledge of frozen c r y o l i t e separates the l i q u i d e lec t ro ly te from the side of the c e l l . This ledge is of outmost importance in protect ing the ce l l l i n ing from bath corrosion and for stable operation. However, the ledge growth mechanism and the bath-ledge heat t ransfer are not well known and a wide range of heat t ransfer coef f ic ien ts are reported.

During ledge growth and/or melt ing, temperature and composition gradients develop at the ledge/bath in ter face. These changes may resul t in a buoyancy induced boundary layer flow which w i l l a f fect the heat transfer coe f f i c i en t . Therefore, i t becomes important to control the overal l heat f lux and to carry out the measurements under steady state with no net mass t ransfer .

The ' co ld - f i nger ' technique used in these laboratory experiments was developed in order to control the heat f lux from the bath. A data acquis i t ion system is used to record the temperature p r o f i l e wi th in the ledge and the bath during steady state and non-steady state heat t ransfer condit ions.

Keywords

Heat t ransfer coe f f i c ien t , Heat f l u x , Liquidus temperature, Bath density, Boundary layer f low, Cold-f inger technique.


1.0 INTRODUCTION

In the Hall-Heroult process for the production of aluminium, a side ledge forms along the side wall of the c e l l . This ledge protects the ce l l l i n i n g and is of outmost importance for long service l i f e of the c e l l . However, during anode e f fec ts , the ce l l heats up and the ledge p a r t i a l l y melts/dissolves away. During under-cooling of the c e l l , t h i s ledge grows and i t may extend down along the bottom of the ce l l causing i n s t a b i l i t i e s . To improve the energy e f f ic iency of the Hall-Heroult process, the trend is to lower the operating bath temperatures by adjusting the A1F3 content. However, lower bath temperatures lead to a) decreased alumina s o l u b i l i t y , b) decreased rate of alumina d isso lu t ion , and c) increased dif ference in the melting point of the bath and that of the c r y o l i t e ledge. This makes the control of the ledge even more important.

As reported by Gan and Thonstad(l), measured and calculated heat t ransfer coef f ic ients scatter widely, with values varying between 150 and 1080 W/m

2K. The

f i r s t objective of t h i s invest igat ion is to determine why the reported heat t ransfer coef f ic ients vary so much, and secondly to develop a technique capable of generating reproducible heat t ransfer data under control led condi t ions.

1.1 S e n s i t i v i t y Analysis

In order to determine the importance of the heat transfer coe f f i c ien t on the ce l l operation, a series of one-dimensional steady-state heat f lux calculat ions were carr ied out. Figure 1 is a schematical i l l u s t r a t i o n of the temperature gradient through the ce l l side w a l l . At times, an a i r gap may develop between the graphite l i n i ng and the steel s h e l l . At steady state, the ce l l heat f lux is given by the fol lowing equation(2):

(Eq. 1)

The f i r s t four terms in the denominator represent the sidewall thermal resistance which is independent of the ledge thickness and the f l u i d flow wi th in the c e l l . Assuming a carbon conduct iv i ty of 4 W/m*K and a shell to a i r heat t ransfer coef f ic ien t of 20 W/m

2*K, the sidewall thermal resistance for a 10 cm th ick

carbon wall becomes 0.08 (m2*K/W). However, i f the carbon wall thickness is

increased to 30 cm and assuming that a 3 mm wide a i r gap has developed between the steel shell and the carbon, the resistance increases to 0.19 (m

2*K/W). I t

must be noted that the sidewall properties may change with time in an operating pot(3) .

Figure 2 shows the ledge thickness as a function of the side wall thermal resistance for four combinations of the bath superheat and heat t ransfer coe f f i c ien t ; 5 K and 400 W/m

2K, 5 K and 600 W/m

2K, 10 K and 600 W/m

2K, 10 K and

1000 W/m2K, respect ively. A ledge thickness below zero indicates that instead of

forming a ledge, erosion of the carbon l i n ing w i l l occur. I t is observed that for a given superheat and heat t ransfer coe f f i c ien t , the ledge thickness decreases with increasing wall resistance. I f a large a i r gap develops between the carbon and the steel shell at a speci f ic locat ion in the pot, the ledge thickness w i l l decrease. During periods of increased s t i r r i n g and bath superheat, the ledge becomes unstable and dissolves/melts back rap id ly .


Steel Air Gap

Figure 1. Temperature p ro f i l e through the side wall of a Hal l -Heroult c e l l .

LEDGE THICKNESS vs. Superheat and Heat Transfer Coef.

50 i 1 , , ,

Sidewall Thermal Resistance

-m- 5, 400 —• 5, 600 10, 600 10, 1000

Figure 2. Ledge thickness as a function of the sidewall thermal resistance for four combinations of the bath superheat(K) and heat t ransfer coeff icient(W/m

2*K).

A value below zero indicates erosion of the carbon side w a l l .


The sidewall heat f lux in indust r ia l pots during steady operation may t yp i ca l l y vary between 1.5 and 5 kW/m

2 and the bath superheat may be in the order of 5 to

20 K(4-6). I t is reasonable to assume that during steady operat ion, the bath temperature is uniform and the ledge composition is constant w i th in a given c e l l , leading to a constant bath superheat. The ledge thickness w i l l depend on the local heat t ransfer coe f f i c ien t , with the thinnest freeze in areas with fast moving e lec t ro ly te and/or aluminium. I t therefore becomes essential to determine why such a wide range of heat t ransfer coef f ic ients exist and how i t is affected by the ce l l operation.

2.0 BATH/LEDGE HEAT TRANSFER ANALYSIS

When a l i q u i d is f lowing past a plate of d i f fe ren t temperature, a temperature gradient develops. The thermal boundary layer(6T) generally d i f f e r s from that of the viscous boundary layer. The heat f lux from the plate is given by the fo l lowing equation:

q/A = -kdT/dy^ = h(Tw a ll - T l i q u i d) (Eq. 2)

where k is the thermal conduct iv i ty of the l i qu id and h is the heat t ransfer coe f f i c ien t . The heat t ransfer coef f ic ient can be expressed as fo l lows:

h = -k(dT/dyw a H) / (Tw aH -' liquid / (Eq. 3)

For a given temperature d i f ference, the heat transfer coef f i c ien t is a function of the temperature gradient at the in ter face. The main di f ference between laminar and turbulent flow are:

1 ) ^ T . laminar

> $ T , turbulent

2) the temperature gradient p ro f i l e for laminar flow can be approximated as a 2'nd order polynomial, while for turbulent flow the p r o f i l e may be expressed as a 6 ' th or 7 ' th order polynomial of the distance from the wall

For turbulent f low, th i s leads to a steeper temperature gradient at the in ter face, resul t ing in a higher value of the heat t ransfer coe f f i c i en t . A series of expressions based on dimensional less numbers exist for the ca lcu la t ion of the heat t ransfer coef f ic ient as a funct ion of the f l u i d condit ions. However, during freezing or melt ing, these equations may not be va l i d . I t is well known that during phase changes such as bo i l ing and condensation, the heat t ransfer coef f ic ients may change dramat ical ly.

The growth or melting of the ledge in an industr ia l pot is affected by events such as anode ef fec t , anode change, alumina feeding and metal tapping. Therefore, i t "is important to analyze the ef fects of mass exchange on the heat t ransfer behaviour. In the subsequent analysis, the heat transfer is analyzed according to :

1. Heat f lux 2. Na3AlF6 - A1F3 phase diagram 3. Ledge formation and melting 4. Bath flow


2.1 Heat f l ux

In an indust r ia l ce l l the heat f lux is approx. 1.5 to 5 kW/m2 during steady

operat ion. However, most laboratory experiments have been carr ied out under non-steady conditions and at heat f luxes up to 50 kW/m

2. Therefore, i t is expected

that the heat t ransfer coef f ic ients obtained in laboratory invest igat ions are d i f f e ren t from those found in indust r ia l pots during steady operation and at lower rates of heat t ransfer .

2.2 NaaAlF6 - A1F3 system

Because of the steepness of the l iquidus l i ne at high A1F3 contents, a small increase in the A1F3 content leads to a large drop in the l iqu idus temperature. The control of the A1F3 content becomes very important. Events such as anode change and anode e f fec t have to be closely control led in order to maintain constant operating temperature and bath composition. This is d i f f i c u l t fo r ce l l s with poor bath c i r cu la t ion leading to composition gradients (A1F3 and A1203) and var iat ions in the bath superheat.

To measure the heat t ransfer coe f f i c ien t , i t is required to know the temperature at the bath/ledge in ter face. For A1F3 r i ch baths, a small enrichment of A1F3 at the interface w i l l lead to a large drop in the interface temperature. I t is therefore required to measure the temperature at short distance in terva ls wi th in the ledge and the l i q u i d bath. Addit ional bath and melting point determinations may also be required.

Another important resu l t of the steepness of the l iquidus l i n e , is the behaviour of the bath density with changing A1F3 content. Figure 3 shows the density of the bath as a function of A1F3 content and bath temperature. Under condit ions of constant bath superheat, the density is affected more strongly by changes in the bath composition than by the associated temperature change for basic baths(< 5wt% A1F3). For intermediate compositions, the iso-density l ines are para l le l to the l iquidus l i n e , while for highly acid baths(> 15 wt% A1F3) the density increases with fur ther A1F3 addit ions due to the sharp drop in the l iquidus temperature. This behaviour is i l l u s t r a t e d in Figure 4 where the density of c r y o l i t e melts at the l iquidus temperature, is p lot ted as a function of the A1F3 content.

2.3 Ledge Formation/Melting

Since the ledge contains only small amounts of A1F3, the l i q u i d adjacent to the ledge during freezing w i l l have a lower temperature and higher A1F3 content than the bulk bath. The corresponding density variat ions may lead to intense f l u i d motions and/or to the formation of a mushy zone, strongly a f fec t ing the heat t ransfer coef f i c ien t (7-10) . Composition induced f l u i d motion may be of importance under the fol lowing condit ions:

a) formation of side ledge b) ledge melting during anode effects c) ledge freezing during anode change d) freezing of bottom crust in an aluminum ce l l e) ver t i ca l mixing of the e lec t ro ly te due to

gradients in the alumina content

The density var iat ions may induce convection in a double-di f fusive boundary layer


MassV. AIF3

Figure 4. Density of Na3AlF6-AlF3 melts at the l iquidus temperature as a function of the A1F3 content. A l l melts are assumed to contain 3wt% A1203 and 5wt% CaF2.

Mgure 3. Lines of constant density in the Na3AlF./A1F, and the Na3AlFf 3/A lF3/A l203/ CaF2 systems as a function of temperature and A1F3

5Wt% CaF2 and 3Wt% AI203

Temperature Varies

Tem

per

atu

re .

K

NQ3AIF6 Liquidus

Wt% AIF3


as i l l u s t r a t e d in F ig. 5( 7) . Depending of the re la t i ve buoyancy and the re la t i ve d i f f u s i v i t y , the main flow regime wi th in the boundary layer may be upflow, counterflow or downflow. In order to determine the flow regime, a re la t i ve buoyancy r a t i o Rfchange in density due to a drop in temperature divided by the change in density due to the change in composition) is introduced. Using the fo l lowing physical propert ies of the e lec t ro l y te ,

v = 1.4*10"* m2/s, D = 5*10'

9 m

2/s , cr = 1.0*10"

7 m

2/s

the re l a t i ve d i f f u s i v i t y ( L e = a/D) is 20. Based on the information given in Figure 6(8) , the boundary layer flow behaviour depends on the A1F3 content as given in Table 1.

Table 1. Convective flow behaviour in a double-dif fusive boundary layer versus the A1F3 content.

%A1 F3 R Flow Regime

0 0 Upflow 4 0.15 Inner-dominated counterflow 8 0.37 Outer-dominated counterflow

12 0.73 Downflow 16 1.29 Downflow 20 2.08 Downflow

I t is in terest ing to observe the change in flow behaviour from being upflow dominated for low A1F3 baths, to being downflow dominated for baths containing 8% or more A1F3. Slow-cooling experiments are carried out in order to confirm th is analysis.

2.4 Fluid Flow

The heat t ransfer coe f f i c ien t is strongly dependent on the f l u i d flow wi th in the c e l l . Experimentally, i t is very d i f f i c u l t to determine the f l u i d ve loc i ty in an indust r ia l pot as well as in a laboratory cruc ib le . Further, the f l u i d turbulence also af fects the heat t ransfer coe f f i c i en t . To address t h i s , laboratory experiments w i l l be carr ied under two d i f fe ren t ' con t ro l led ' s t i r r i n g condit ions. The f i r s t is to bubble nitrogen through the melt at a given depth. The energy d i ss i pa t i on ( l l ) in the c r yo l i t e melt due to gas in ject ion is given by:

Power(W) = 0.0123*Q*T(K)*ln(l + 0.002*L) (Eq. 4)

where Q is the gas flow rate in l i t e r / m i n and L is the lance immersion in cm. For small immersion depths, th is s impl i f ies t o :

Power(W) - 2.4*10*5*Q*T(K)*L (Eq. 5)

The energy d iss ipat ion due to the anode gas release in a Hal l -Heroult ce l l is approx. 100 W/ton of e lec t ro l y te . For steelmaking processes with gas in jec t i on , the corresponding energy varies generally from 500 to more than 10,000 W/ton ( l l ) .


Figure 5. Schematic i l l u s t r a t i o n of the three region boundary layer caused by a combination of composition and temperature induced density gradients(7).

Convection in a double-diffusive boundary layer

10 fa | | I 111 l l |—I I I 111 • 11—I I I i m i l — i i i i n n . I i i mg

Downflow

io-«

Outer-dominated -d

RD*~ 1

Inner- H dominated RU~ I

• ftm» » i « mi l l I l l ^ i l l l l i i mii i

to 10* 10* 104

Relative diffusivity Le =oc/D

10*

Figure 6. Convective flow in a double-di f fusive boundary layer (8 ) ,

Distance from the wall (arbitrary units)

Vcr

ticnl

vel

ocit

y

T

V Counterflow

I > Upflow

Relati

ve buo

yancy

R •» Ap t/A/

> e


In order to determine the e f fec t of the bath veloc i ty on the heat t ransfer coe f f i c ien t , a ro ta t ing cruc ib le technique w i l l be developed.

3.0 EXPERIMENTAL

3.1 Design Cr i te r ia

While reviewing the l i t e r a t u r e i t became apparent that in order to generate reproducible heat t ransfer data, the experimental technique had to be care fu l l y designed in terms of the fo l lowing considerations:

3 . 1 . 1 . Heat f lux(cont ro l over a wide range)

In order to study the d isso lu t ion k inet ics of alumina in c r y o l i t e baths, Taylor et al(12) dipped graphite blocks into the baths. They found that the bath/graphite block heat f lux i n i t i a l l y peaked at more than 100 kW/m

2 before

s tab i l i z i ng around 40. This is approx. 10 times higher than the steady state heat f lux between the bath and the ledge in an industr ia l pot. This indicates that th is ' t ype ' of technique is not suitable for steady state heat t ransfer measurements since the heat t ransfer a) is not cont ro l lab le and b) poorly represents the sidewall /bath s i tua t ion .

3.1.2. NagAlFg - A1F3 phase diagram

Because of the steepness of the l iquidus l i ne i t is required to ca re fu l l y measure the temperature p r o f i l e in the melt and through the ledge to determine the temperature at thee in ter face. Taylor et al(12) found that the surface temperature was 18 °C below the l iquidus temperature. Because the thermal conduct iv i ty of the ledge may vary with location and temperature, a series of thermocouples has to be used to ensure that th is information can be acquired.

3.1.3. Ledge formation and melting

Because the heat t ransfer coef f i c ien t may be d i f fe rent during ledge melting than during ledge growth, the technique has to be capable of simulating both s i tuat ions in a contro l led fashion.

3.1.4. Bath flow

The heat t ransfer coe f f i c ien t depends strongly on the f l u i d behaviour. In order to simulate pot s t i r r i n g , a ro tat ing crucible technique is presently being developed. Addit ional experiments w i l l be carried out using gas s t i r r i n g and/or a BN s t i r r e r .

3.2 Experimental Set-up

The experimental set-up used in these measurements is shown in Figure 7.

3.2.1 Cold-Finger

The cold- f inger is made from a 130 mm high graphite cruc ib le wi th an inner- and outer diameter of 19 and 25 mm, respect ively. A thermocouple is located at the interface between the cruc ib le wall and the inner thermal insu la t ion to measure the inner wall temperature. A copper-tube cooling system is located in the centre of the cold- f inger and is thermally insulated from the graphite cruc ib le using


an insulat ing br ick . Water flows down into the ' co ld - f inger ' inside an inner copper tube(6 mm OD) and out through an wider copper tube(10 mm 00). The heat f lux is determined by the dif ference in the water temperature between the i n l e t and the ou t l e t . Two type-T thermocouples are placed inside the copper tubes at the height of the c r yo l i t e bath to measure the water temperatures.

Figure 7. Schematics of the experimental technique. A data acquis i t ion system was used to record a l l experimental data. 1- Graphite cruc ib le , 2- Bath, 3- Ledge, 4- Insu lat ion, 5- Graphite co ld- f inger , 6- Insulat ion, 7- Stainless steel support, 8- Inner copper tube, 9- Outer copper tube, 10- Arm for ledge thickness determination, 11- Water ou t le t , 12- Water i n l e t , 13- Thermocouple, 14- System of type K thermocouples protected in alumina sheets, BN-st i r rer ( to be added).

3.2.2 Crucible Assembly

The graphite crucible(10 cm diameter and 12 cm height) containing approx. 1.8 kg of bath, is heated to 960 - 990 °C in a ver t i ca l pot furnace. The stainless steel support for the ' co ld - f inger ' is designed to lock into the top of the graphite


cruc ib le . Thermal insu la t ion between the top of the co ld- f inger and the stainless steel support minimizes the heat loss from the melt. A series of type-K thermocouples are pos i t ion at determined distances from the graphite co ld- f inger . These thermocouples are protected using a 1 mm OD inconel sheet placed inside alumina tubes. A stain less steel rod is used to measure the ledge thickness.

3.3 Procedure

The melt is heated to the required temperature while the stainless steel support and the cold- f inger are kept above the melt for approx. 10 minutes in order to heat up before being immersion in the bath. The water flow and the thermocouple readings are continuously determined and stored using a data acquis i t ion system. Typ ica l ly , the experiments are carr ied out at one f ixed bath temperature un t i l steady state is achieved as determined by the temperature readings and the ledge thickness. The experiment may then be continued for a duration of up to 4 hours at various furnace temperature or bath s t i r r i n g condit ions.

3.4 Variables

Although only prel iminary resul ts are presented in th is paper, the variables to be studied include:

gas s t i r r i n g / c r u c i b l e ro ta t ion bath alumina and aluminium f luor ide contents bath temperature heat f lux

3.5 Ledge properties

The ledge formed on the cold- f inger w i l l be analyzed chemically and mineralogical ly. Since i t is important to know the bath/freeze in te r fac ia l temperature, the melting point of the ledge w i l l be determined using DTA analysis. In separate slow cooling tes ts , experiments are carr ied out at d i f fe ren t c r yo l i t e ra t ios in order to determine i f melt s t r a t i f i c a t i o n takes place or not.

4.0 RESULTS

Figure 8 shows the measured temperatures and ledge thickness over a period of 2 hours for a bath with 8 wt% A1203. In th is experiment, the furnace temperature was lowered in 10 °C steps from 980 to 960 °C. No s t i r r i n g was appl ied.

I t is in terest ing to observe the rapid increase of the temperature inside the graphite f inger (T l ) upon immersion in the bath. As a ledge forms on the f inger , th is temperature drops quickly and stabi l izes around 770 to 800 °C depending on the furnace temperature. The thermocouples located inside the melt(T2, T3 and T4) show a rapid increase upon immersion in the melt before adjust ing according to locat ion and furnace temperature. During th is tes t , the cooling water was changed from 2 1/min down to 0.5 1/min in order to raise the temperature increase of the water. As observed in Figure 8, i t took approx. 40 minutes to reach steady state conditions af ter changing the furnace temperature.

The ledge thickness was measured by a stainless steel arm. At the completion of the experiment, the co ld- f inger was l i f t e d out of the bath while the bath was s t i l l l i q u i d . Because of the i r regular surface of the freeze(Figure 9) , the ledge

T I M E . min

Figure 8. Temperature and ledge thickness as a funct ion of time for a bath containing 8 wt% A1203. The water flow rate was 2.0 and 0.5 1/min with a furnace temperature of 980, 970 and 960 °C, respect ively. Thermocouple Tl is located inside the graphite f inger and thermocouple T5 is kept above the bath surface. Thermocouples T2, T3 and T4 are in the bath and located 2, 5.2 and 11.6 mm from the cold- f inger w a l l .

Figure 9. Picture of the frozen ledge af ter removal from the bath.

TE

MP

ER

AT

UR

E ,

oC

ww

' SS

3N

MO

IH139031

174 EXTRACTION, REFINING AND FABRICATION OF LIUH1 Mb 1 ALo


thickness measurements are less accurate and also d i f f i c u l t to i n te rp re t .

In a second experiment, the alumina content was decreased to 3 wt%. The bath and ledge temperatures are given in Figure 10 as a function of t ime. I t must be noted that the furnace temperature was changed frequently in order to determine the time delayed ef fects on the bath - , ledge - and cold f inger temperatures. The observed behaviour was very s imi lar to that in the f i r s t experiment providing assurance of the rep roduc ib i l i t y of the technique.

The temperature p r o f i l e in the f inger and the ledge is shown in Figure 11. The ledge thickness is also indicated. However, steady state condit ions were not reached due to frequent changes in the furnace temperature.

5.0 CONCLUSIONS

I t has been demonstrated that several physical and chemical parameters may af fect s i gn i f i can t l y the bath/ledge heat t ransfer . Based on changes in the bath density with varying temperatures and A1F3 contents, f l u i d flow is induced wi th in the ledge/bath boundary layer . This f l u i d motion may have a great e f fec t on the heat t ransfer behaviour. In order to measure heat t ransfer coef f i c ien ts relevant to the steady state operation of an indust r ia l pot, an experimental technique capable of carrying out heat t ransfer measurements both during t ransient and steady state condit ions was developed.

The prel iminary experiments indicate that reproducible heat t ransfer data are obtained. The resul ts show that i t takes approx. 40 min. to reach steady state when no s t i r r i n g is employed. Using an alumina saturated bath, the ledge formed on the cold f inger show a very i r regu lar shape.

Further improvements and modif ications to the technique are required in order to carry out the heat t ransfer measurements with forced convection.

ACKNOWLEDGEMENT

The f inanc ia l support provided by Alcan International Limited and NSERC is great ly appreciated. The authors are thankful to Alcan Internat ional Limited for the permission to make th i s work publ ic .

REFERENCES

1. Y.R. Gan and J . Thonstad, Light Metals, Feb. 1990, pp. 421-29. 2. J . Thonstad and S. Rolseth, Light Metals, Feb. 1983, pp. 415-24. 3. M.P. Taylor, B.J. Welch and J.T. Keniry, Light Metals, Feb. 1983,

pp. 437-47. 4. H. Tsukahara, N. Ono and K. Fu j i t a , Light Metals, Feb. 1982, pp. 471-82. 5. T. Ohta and T. Matsushima, Light Metals, Feb. 1984, pp. 689-99. 6. J.G. Peacey and G.W. Medlin, Light Metals, Feb. 1979, pp. 475-92. 7. H.E. Huppert, J . Fluid Mech., V212, 1990, pp. 209-40. 8. R.H. Nilson, J . Fluid Mech., V160, 1985, pp. 181-210. 9. F.J. Spera, D.A. Yuen and D.V. Kemp, Nature, V310, 1984, pp. 764-6. 10. H.E. Huppert and M.G. Worster, Nature, V314, 1985, pp. 703-07. 11. N.J. Themelis and P. Goyal, Canadian Meta l l . Quart., V22, 1983, pp. 313-20. 12. M.P. Taylor, B.J. Welch and R. McKibbin, AIChE, V32, 1986, pp. 1459-65.


Figure 10. Temperature and ledge thickness as a function of time for a bath containing 3 wt% A1203. The measurements were carr ied out during booth cooling and heating of the bath. Thermocouple T2 stopped functioning af ter 95 min.

940-!

920-

90OJ

6 8 CH

I

860-

840-

820-

800-

Insuiat-1 Grap^nte

Don . W a i

7804

Ledge Bam

. . . . .€3"'"

10 mm , 1 mm

20 mm,, 3 mm

40 mm., 4 mm

80 mm , 8 mm I

110 min , 10 mm' i

0 5 10

DISTANCE FROM GRAPHITE CRUCIBLE . m m

15

Figure 11. Temperature p ro f i l e versus distance from the wall of the cold f inger for a bath containing 3 wt% A1203. The temperature p ro f i les are based on the results in Figure 10.

TE

MP

ER

AT

UR

E ,

oC

TE

MP

ER

AT

UR

E ,

oC

0 20 40 60 80 100 120 140 160

177

Agglomeration and dissolution of alumina in cryolite baths

S. Rolseth, J. Thonstad The Foundation for Scientific and Industrial Research, and the Laboratories of Industrial Electrochemistry, The Norwegian Institute of Technology, Trondheim, Norway

ABSTRACT

The behavior of alumina immediately after addition to cryolitic baths was studied in laboratory experiments. Both primary (virgin) and secondary (reacted) aluminas from various sources were tested. It was found that the secondary aluminas dissolved faster than the corresponding primary aluminas when the amount added was large enough to form agglomerates of alumina and frozen bath. The aggregates formed when using reacted aluminas were more porous than those formed from primary aluminas. Among the primary aluminas the dissolution rate was highest for those which showed the least tendency to agglomeration. However, a different ranking was found when the feeding was performed slowly in a continuous manner so that agglomeration was avoided.

KEYWORDS

Aluminum electrolysis; Alumina; Crusting; Dissolution

INTRODUCTION

The alumina and its behavior is one of the decisive factors in operating and controlling modern aluminum electrolysis cells. Its ability to form a crust on top of the bath and its insulating properties are essential for the thermal balance of the cell. Its ability to adsorb fluorides is used in dry scrubbing systems to reduce the fluoride losses from the cells and the emission to the atmosphere. Over the years the alumina manufacturers have adapted their production to meet these requirements using various technologies. The technology will also to some extent depend on the properties of the raw material (bauxite), and therefore a certain variety exists in industrial aluminas of today. This creates problems for aluminum plants which often are forced to change their supply of alumina. Demands for experimental methods which can predict the behavior of various aluminas have often been raised. Therefore, a


number of works can be found in the literature treating this problem, especially the crusting and dissolution of alumina in cryolitic baths.

The most crucial problem is the rate of dissolution of alumina in the bath. It has been shown that well dispersed alumina dissolves in the matter of seconds (1),(2). However, with batchwise feeding the alumina forms agglomerates embedded in frozen bath. As the alumina heats up a phase transformation occurs from the various intermediate alumina phases ( 7 etc) to the stable a phase (3). The phase transformation is accompanied by sintering, so the lumps do not necessarily fall apart when the frozen bath melts away.

The agglomeration of the alumina is particularly pronounced in traditional sidebreaking or center bar-breaking cell where large quantities of alumina is fed together with pieces of broken crust at infrequent intervals of two to four hours. In this case only a fraction ( -1 /3 ) of the alumina dissolves immediately (the first minutes) (4). The undissolved alumina rests in the top crust, along the side ledge, or underneath the metal, in the latter case forming so called bottom sludge (muck). This alumina dissolves slowly into the bath, and this process is called self-feeding. This way of feeding makes it difficult to control the concentration of alumina in the bath at any time.

Modern point-feeding systems which add small amounts of alumina at short intervals (every minute or so) greatly alleviate these problems and allow for a better control of the alumina in the bath. However, the dissolution process is still rather slow because agglomeration and freezing of bath around the agglomerates take place when cold alumina hits the surface of the bath. As long as we have no practical way of effectively dispersing the alumina grains in the bath, it appears that the total dissolution process is governed by the events occurring each time when a batch of alumina is dropped into the bath.

Decisive factors may be the size of the alumina batch, the way the alumina is added and spreads on the surface, the size of the lumps formed and their tendency to break up into smaller pieces before they dissolve or settle at the bottom of the cell.

The purpose of the present work was to conduct a laboratory study of the agglomera-tion of alumina during feeding and the rate of dissolution during semi-continuous feeding.

EXPERIMENTAL AND DISCUSSION

Crusting and wetting while alumina rests on the surface of the bath

Visual observations of alumina additions to cryolitic baths showed a striking difference between the behavior of primary and reacted aluminas. In order to elucidate this phenomenon some preliminary laboratory experiments were conducted where samples of newly formed crusts were collected. These experiments were carried out with aluminas of two different origins: Interalumina (Venezuela) and Aljam (Jamaica). In order to simulate certain point feeders the aluminas were added through a steel tube where the lower part was bent at an angle of 4 5

0 in order to give the alumina a horizontal velocity component when

hitting the bath. There were no other constraints to the movement of the alumina on the bath surface than the walls of the graphite crucible (I.D. 100 mm). The temperature of the bath was kept 20 °C above its eutectic temperature for precipitation of cryolite and alumina.


Figure 1. Vertical sections cut from samples of crust collected immediately after addition of alumina to the bath. A - Primary Aljam, B - Secondary Aljam, C -Primary Interalumina, D - Secondary Interalumina,

A B

c D

1 m m


The composition was cryolite saturated with alumina and 8 wt-% AlF3and 4wt-% CaF2.The cas t samples were collected by having a steel sieve immersed in the bath before addition. The samples were collected immediately before the alumina was completely wetted by the bath, i.e before it would sink in the bath.

Visual observation showed that when the alumina batch size was 10 g there was hardly any tendency to form lumps (aggregates) of crust when secondary alumina was added. Primary aluminas formed lumps which could be removed from the bath after addition. When the batch size was increased to 20 g both primary and secondary aluminas formed lumps upon addition to the bath, and for both types the mass of the lumps was approximately three times the amount of alumina added. However, a pronounced difference in the appearance of the lumps was found. The lumps collected after addition of secondary alumina exhibited a grayish and open structure, while the primary aluminas gave a white and more dense morphology. Figure 1 shows vertical sections of samples from lumps collected after addition of primary and secondary alumina for aluminas originating from two different producers (Aljam and Interalumina).

In connection with anode changing in an industrial cell we were able to collect floating samples of newly formed agglomerates immediately after alumina was discharged from the adjacent point feeder. The secondary alumina discharged from the point feeder gave a porous grayish agglomerate similar in appearance to those obtained in the laboratory experiments. A white and more dense structure was found in the samples collected when "a batch" of primary alumina was added in the same position.

Because the alumina was added in a standardized way in these laboratory tests the bulk of the alumina tended to reach the same position each time. When a thermocouple was placed 1-2 mm above the bath surface in this position a lower drop in temperature upon addition was found when reacted alumina was added compared to primary alumina. This was mainly ascribed to the fact that reacted alumina did spread and cover a larger area than the primary alumina, thus giving a thinner crust. To counteract the tendency of the aluminas to spread differently on the bath surface, a constriction was made as shown i Figure 2. It consisted of a piece of sintercorundum tube of 56 mm i.d. submerged 10 mm into the bath. The size of the charges of alumina added were 20 and 30 g. The thermocouple was centered in the tube and positioned in such a way that the tip was barely immersed in the alumina after addition. This position was determined by trial and error, and it varied from 5 to 8 mm above the bath depending on the amount and type of alumina added. Results obtained with primary and secondary Interalumina will be reported.

Figure 3 shows a typical temperature curve obtained in these experiments. An abrupt drop in temperature occurred when the cold alumina was added. This was followed by a rapid increase in temperature which slowed down until two inflections in the curve occurred 1-3 minutes after addition. It is evident that these inflections were associated with the arrival of a liquid bath phase at the tip of the thermocouple. When the tip was completely surrounded by liquid bath the increase in temperature slowed down again. The breaks in the curve occurred on the average after 80 % of the time required for the alumina to be completely soaked by the bath had elapsed. It was possible to determine visually the time for complete soaking by illuminating the top of the alumina using a white light source. The scatter in these experiments was relatively large, but to some extent it was possible to detect


Figure 2 Experimental arrangement used for the study of the crusting behaviour of cold alumina added to a confined bath area.

the following differences in the behavior of the primary and the reacted aluminas:

Primary alumina was wetted more rapidly than reacted alumina

The two inflection points indicated in Figure 3 occurred at a lower temperature when primary alumina was added than when secondary alumina was used. The mean values (and standard deviation) of the two inflection point calculated from three replicate runs for the primary alumina were 705 °C ( ± 57 °C) and 805 °C ( ± 21 °C) compared to 840 °C ( ± 42 °C) and 866 °C ( ± 24 °C) for the secondary alumina.

These results indicate that the bath which rises in the bed of alumina has a com-position different from that of the the bulk of the bath. The temperatures at the inflection point are below the liquidus temperature of the bath's bulk composition, which means that the composition of the liquid phase that reaches the thermocouple is different from that of the bulk. One can envisage such a change in the liquid bath composition by considering the phase diagram of the system (5). The bath which penetrates the colder alumina will be cooled down, and cryolite and alumina will freeze out first, leaving a liquid phase enriched in the other components of the bath, and it may rise further until it finally reaches the eutectic temperature of the system. This shift in bath composition is also supported by the findings of Becker (6) who analyzed crusts from industrial cells and found that the bath phase in the upper part of the crust was enriched with A1F 3. The fact that higher tempera-tures were found for the secondary alumina can be ascribed to the exothermic transformation of transition modifications ( 7 etc) to the stable a-phase (7). It has been shown that this conversion is catalyzed by the presence of fluorides (8), and the very intimate mixture between the alumina and the adsorbed fluorides which exists in the secondary alumina will promote this reaction.

Charge tube

Bath saturated

with alumina

- Crust

Thermocouple Sintercorundum

ring


1000

800-

o

^ 600-

f-

400-

200 -J . . . . 1 -20 0 20 40 60 80 10

T i m e / s

Figure 3 Typical temperature curve obtained with a thermocouple located so close to the bath surface that it becomes embedded in alumina when cold alumina is added to the bath (see Fig. 2).

Permeation of water in alumina.

We have also compared the penetration of water into primary and reacted aluminas by measuring the rise of water in glass tubes filled with the different aluminas. Four different primary aluminas and their corresponding secondary aluminas were studied. Without exception the wetting occurred more rapidly in the primary aluminas. The rise velocity varied between 0.2 and 0.46 mm/s, the extremes being secondary and primary aluminas of the same origin (Aljam). These velocities are more than one order of magnitude higher than those observed for rise of cryolitic baths (9). This difference demonstrates that the freezing phenomena that occur with cryolitic baths are a decisive factor for the rate of penetration of the bath into the alumina, and hence they play an important role in the crusting process.

Dissolution of newly formed aggregates of crusts

In these experiment the dissolution of alumina was monitored by applying critical current density (ccd) measurements to determine the concentration of dissolved alumina in the bath. A gradually increasing anodic potential is applied to a carbon electrode, and the current increases until a maximum, the ccd, is reached. In cryolite-alumina melts the ccd increases strongly with increasing alumina content. Since the ccd is also affected by other parameters, the actual relationship (ccd vs wt-% A I 2 Q 3 ) must be established by calibration. Various versions of this method have been used as described in several papers (10),(11),(12).


The experimental arrangement used in this work is shown in Figure 4. The bath was contained in a 100 mm i.d. graphite crucible. Because the pieces of crust formed had a tendency to adhere to the walls of the crucible, a special gas induced stirring was employed. The stirrer consisted of a steel tube which was positioned in a loop around the bottom perimeter of the crucible. Three holes with a diameter of 0.5 mm were drilled in the steel tube with 120° spacing. A gas flow of 140 Nml/min was applied which was the minimum flow needed to obtain even distribution of gas from the three holes in the steel tube. The convective pattern set up by this arrangement prevented sticking of crust to the walls and it also set up ripples on the bath surface similar to industrial cells. The bath temperature was measured with a Pt/PtRh thermocouple protected by at stainless steel tube immersed in the bath. The bath was made of 1 kg of synthetic cryolite containing 6 wt-% A1F3 and 6 wt-% CaF2. The temperature was 955 ± 2 °C. Alumina was added in the center of the crucible. The amount was 20 g which should give a 2 % increase in the concentration of dissolved alumina. Such behavior was not always observed which can partly be attributed to uncertain-ties in establishing the calibration curve for the ccd measurements. The ccd method is sensitive to impurities in the bath, for example, a new calibration curve had to be established when the we changed from one brand of synthetic cryolite to another. When handpicked natural cryolite was used a decrease in the critical current density was found by adding 0.1 wt % iron carbonate to the bath. Therefore, when the dissolution behavior for the various aluminas was compared, the time needed to reach steady state after addition was regarded as an equally important factor as the observed increase in alumina concentration. The dissolution curves of primary and reacted Interalumina and Aughinish (Ireland) alumina are shown in Figure 5 and Figure 6.

•Cathode lead Gas tube

Crucible.

ccd probe (anode)

Bath

Figure 4. Sketch of the arrangement used to study crusting and dissolution of various aluminas.

Hopper -

Ball valve-

Charge tube-


2.5

> 2.0

t

! 1.5

| 1-0

<0 .5

0.0

• • • • •

Secondary alumina • rrn

q^ctpn 3^ •

•

Y wk^tJ^m Primary alumina

r/ •

]

10 15 2 0

T i m e / m i n

25 30

Figure 5. Dissolution curves for primary and secondary Interalumina alumina when 2 wt-% is added to a stirred melt.

3.5

3.0

% 2.5

1.5

1.0

0.5

0.0

Secondary alumina C P zxan

fS^^ Primary alumina

J?

10 15

T i m e / m i n 2 0 25

Figure 6. As Figure 5 for Aughinish alumina.

For both types of alumina a more rapid dissolution was found for the reacted type. Visual observations showed that the aggregates formed when secondary aluminas were added were more fragile, and they broke up into smaller pieces more easily than aggregates formed from primary aluminas. On the assumption that the process is mass transfer controlled, the difference in dissolution rates for primary and secondary aluminas can be explained by the difference in contact area between aggregates and bath for large and small aggregates.

In another series of experiments with slightly different experimental conditions, such as higher bath temperature (1007 ±2 °C) etc, the dissolution of three different aluminas was compared. The aluminas came from other shipments than those mentioned above. The


aluminas tested were Aljam, Aughinish and Interalumina. The differences between these aluminas were not so pronounced as when primary and reacted alumina of the same type were compared. Due to the experimental scatter a different approach was used to compare the dissolution behavior. Three replicate runs were made with each type of alumina, and from experimental results similar to those shown in Figure 5 and Figure 6, dissolution curves were drawn using a common sense approach as indicated above. From these curves the amounts dissolved were determined after certain time periods. In Table I the results are presented as the mean value and the standard deviation for each set of observations.

The data in Table I indicate that the Aljam alumina dissolved faster than the other two types. There is no significant difference in the dissolution behavior between Interalumina and Aughinish.

Table I - Mean and standard deviation of the observed increase in alumina concentration in the bath at certain times after addition. The added amount represents a 2 wt % increase in the concentration of dissolved alumina. The values shown are based on three repli-cate runs.

Type of Time after addition of alumina to the bath

alumina 1 min 3.5 min 8 min 16 min 20 min

Interalumina 0.70±0.10 1.33±0.15 1.57+0.29 1.63 ±0.32 1.79±1.67

Aljam 1.50+0.10 1.85 ±0.05 1.95 ±0.05

Aughinish 0.63+0.15 1.16+0.29 1.53±0.31 1.76+0.45 1.83 ±0.45

The dissolution behavior was also compared by simple experiments where aggregates were collected after 1 and 3.5 minutes and weighed. Except for the presence of the ccd probe (graphite anode) and a sieve immersed in the bath before addition the experimental conditions were identical to the dissolution experiments mentioned above. One minute after the alumina was added samples of aggregates were retrieved using a sieve made of 100 mesh nickel netting. Due to corrosion a coarser netting made of steel wire and with a mask size of 3 mm was used when the aggregates were collected 3.5 minute after addition. The results are shown in Table II.

The samples were analyzed for alumina by the wet chemical method (13). The balance between the amount of alumina found in the aggregates (A) and the added amount was taken to represent alumina dissolved in the bath (D).

The results shown in the tables above agree well in the sense that both methods rank the Aljam alumina as the fastest dissolving alumina, and there is no significant difference in the dissolution rate between Interalumina and Aughinish. For the Aljam alumina the dissolved amounts are in reasonable agreement for the two methods. For the other two types the dissolved amounts shown i Table II are higher than given by the critical current density measurements. This discrepancy can be due to failure to collect all of the aggregates on the sieve.


Table II Samples of alumina-bath aggregates removed from the bath at 1 and 3.5 minutes after that a batch of 20 g alumina was added to a stirred bath. T - total weight of sample, A - alumina content of sample, D - calculated amount of dissolved alumina; all units in grams.

Type of 1 minute after addition 3.5 minutes after addition alumina

T A D T A D

Interalumina 19.9 8.7 11.6 12.0 4,0 16,3

Aljam 13,6 5.9 14.4 9.8 3.3 17.0

Aughinish 20.6 9.0 11.3 6.6 2.0 18.3

The initial concentration of alumina in the bath was 1.5 wt-% for the experiments shown in Table I and Table II. The rate of dissolution decreased markedly if more additions were made after the first batch had dissolved. The scatter increased and the difference in dissolution rate between the Aljam and the other aluminas became less apparent.

Semi-continuous feeding.

The idea behind these experiment was to study the dissolution behavior of the aluminas when the amount added was so small that formation of aggregates became negligible. Experiments where small amounts of alumina were added to a vigorously stirred melt where the particles immediately became dispersed in the bath had shown that the dissolution was completed within 10 seconds (1). This observation is in agreement with theoretical calculations assuming mass transfer control (14).

We were not able to maintain an absolute continuous feed, so a semi-continuous procedure was chosen. This was obtained by means of a pneumatically operated feeder where small volumes of alumina were added at controlled intervals. The batch size was 0.23 g and the intervals between the batches were varied from 4.5 to 12.4 seconds. The other experimental conditions were as described above (1 kg bath, 1007 °C, gas induced stirring, initial alumina concentration 1.5 wt-%). Three series of experiments were run using the sieve fractions between 60 ^m and 160 /im of the three aluminas mentioned above. The alumina concentration was determined by frequent ccd measurements during the run as shown in Figure 7.

The squares are the measured values obtained from the ccd measurement and the line is the theoretical concentration given by the feeding rate of alumina. One notices that in the initial period the measured values follow the theoretical concentration which indicates that the dissolution is as fast as the feed rate. After some time a deviation occurs, which means that the rate of dissolution can no longer match the feed rate. At the time when the


deviation occurs the following relation exists between the feed rate, r, and the concentration of dissolved alumina in the melt, c d

r = K- (c*-cd) (1)

where K can be regarded as a rate constant for this particular system and c is the saturation concentration.

40

Time/min 80

Figure 7. Dissolution of alumina added at a constant feed rate. Squa-res: measured concentration of dissolved alumina in the bath, line -calculated concentration assuming instantaneous dissolution.

In Figure 8 the c d values are plotted as a function of the feed rate. The saturation concentration of alumina in this melt is calculated to be 9.2 wt-% (15). The straight lines in the figure are regression lines obtained by using the saturation concentration as the point corresponding to zero feed rate in the set of data. In this treatment the data for Interalumina and Aughinish are lumped together because no significant difference could be found. Aljam on the other hand, gave markedly lower rates of dissolution as shown in Figure 8.

The rate constant can be found from Figure 8 as the inverse absolute value of the slopes of the regression lines. The intersections between the abscissa and the extrapolated regression lines give the maximum dissolution rate under these experimental conditions. For Aljam this maximum is 0.35 wt-%/min and for the other two aluminas 0.7wt-%/min. These values are one order of magnitude lower than the dissolution rates observed when alumina is dispersed in vigorously stirred melts as mentioned above, where 1 wt-% alumina dissolved within 10 seconds. This difference can be explained by small aggregates being formed also in our case where the batch size was 0.23 g (0.023 % of 1 kg bath). The constant K in equation (1) can be expressed as the product of the active surface area of the aggregates, A, and the mass transfer coefficient, k. Ideally is should be possible to estimate the size of the

I

t \ o 0

o

cd

a

Continous feed rate 0.14 wt%/min


aggregates when the other parameters are known. However, the relationship between the mass transfer coefficient, the convection and the aggregate or particle size is quite complex. This makes it extremely difficult to compare results obtained under different experimental conditions. For example when the particles are fully dispersed in the bath, stirring will only have a minor effect on the dissolution rate for particles with diameter less than 50 jjm (14). For aggregates with larger diameter the convective pattern will affect the mass transfer coefficient in addition to the effect it may have on breaking up the aggregates into smaller pieces.

10

0 -I , : , , , 1 0 0.1 0.2 0.3 0.4 0.5

Feed rate/ (wt -%/min)

• Aljam

13 Aughinish

1 =1 Interalumina

Figure 8. Results from experiments with semi-continous feeding of alumina to a stirred bath. The alumina concentration at the time the dissolved amount deviate from the amount added is plotted as a function of the feed rate.

Both in the experiments with semi-continuous feeding and the experiments where dissolution of newly formed aggregates of crust were studied the Aljam was the alumina which showed odd behavior. With semi-continuous feed Aljam was the slowest dissolving alumina and with batch wise addition it dissolved faster than the other aluminas. The Aljam alumina contained 20 wt-% of the stable a-modification, while the a contents in Aughinish and Interalumina were one order of magnitude lower, 2% and 2.5 wt-% respectively. This difference is probably the reason for the divergent behavior of the Aljam alumina. When addition is performed batchwise allowing agglomeration and sintering to occur, a high a content gives weaker crusts (16) which break up more easily and hence dissolves faster. On the other hand aluminas high in a dissolve more slowy when the individual grains are dispersed in the bath (1). In our case this might have been the dominant factor in the semi-continuous feed experiments even if some aggregate formation probably took place also in this case as mentioned above.


This means that the method of addition is as important as the alumina properties for the dissolution behavior, the crucial point being formation of aggregates. Aggregates are formed because bath penetrates, freezes and locks the alumina particles while the 7 to a transformation takes place, leading to a sintered network of alumina particles (8). Factors which counteract aggregate formation are high superheat, convection in the bath where the alumina is added, preheating of the alumina, high content of volatiles in the alumina and good flowability (2).

ACKNOWLEDGMENT

The authors wish to acknowledge financial support and the permission to publish this work from Hydro Aluminium, Ardal Plant. A valuable contribution to the experimental work was due to Mona S.Emanuelsen as apart of her graduation thesis at The Norwegian Institute of Technology, Trondheim, Norway.

REFERENCES.

1. J. Thonstad, F. Nordmo and J.B. Paulsen. Dissolution of Alumina in Molten Cryolite. Met. Trans., 3, 1972, 403-408.

2. G.I. Kuschel and B.J. Welch. Further studies of alumina dissolution under conditions similar to cell operation. Light Metals 1991, Proceedings of the 120th TMS Annual Meeting, New Orleans, February, 1991, p.299-305.

3. L.N. Less, The Crusting Behavior of Smelter Aluminas . Met. Trans. B, 8B, 1977, pp.219-225.

4. J. Thonstad, Semicontinuous Determination of the Concentration of Alumina in the Electrolyte of Alumina Cells. Met. Trans B, 8B, 1977, pp. 125-130.

5. K. Grjotheim, C. Krohn, M. Malinovsky, K. Matiasovsky and J. Thonstad. Aluminium Electrolysis. 2nd, ed. Aluminium-Verlag. Dusseldorf, FRG, 1982,24-80.

6. Aron Becker, Properties of Crust. Internal ALCOA report, 1988.

7. J. Gerlach, Dissolution and Interaction between Alumina and Cryolite Melts. Proceedings of the First International Symposium on Molten Salt Chemistry and Technology, April 1983, Kyoto, Japan, 89-92.

8. R. Oedegard, S. Roenning, S. Rolseth and J. Thonstad. Crust formation in aluminium cells. J. Metals, 37, N o . l l , 1985, 25-28.

9. S. Rolseth, K. Rye, J. Thonstad, Ongoing work to be published

10. K.A. Rye, S. Rolseth, J. Thonstad and Z. Kuang, Behaviour of alumina on addition to cryolitic baths. Proceedings of the Second International Alumina Quality Workshop, Perth, Australia, October 1990, pp.24-37.


16. S. Rolseth, S. Raining, A. Solheim, J. Thonstad and R. 0deg3rd, Non-convential methods for alumina testing. Aluminium, 62, 1986, pp.286-290.

11. R.K. Jain, S.B. Tricklebank, B.J. Welch and D.J. Williams, Interaction of aluminas with aluminium smelting electrolytes . Light Metals 1983, Proceedings of the 112th AIME Annual Meeting, Atlanta, USA, March 1983, pp.609-622.

12. A.T. Tabereaux and N.E. Richards, An improved alumina concentration meter. Light Metals 1983, Proceedings of the 112th AIME Annual Meeting, Atlanta, USA, March, 1983, p.495-506.

13. K. Grjotheim and B.J. Welch, Aluminium Smelter Technology. 2nd edition, Aluminium-Verlag, Dusseldorf, FRG, 1988,256-257.

14. J. Thonstad, A. Solheim, S. Rolseth and O. Skar, The Dissolution of Alumina in Cryolite Melts. Light Metals 1988, Proceedings of the TMS 117th Annual Meeting, Phoenix, Arizona, January 25-28, 1988, pp.655-661.

15. E.W. Dewing, Liquidus Curves for Aluminium Cell Electrolyte . J. Electrochem. S o c , 117, 1970,pp.780-781.

191

Thermodynamic properties of molten chloride electrolytes for magnesium production MgCl2-LiCl-NaCl system

B.R. Davis Department of Materials and Metallurgical Engineering, Queen's University, Kingston, Ontario, Canada

W.T. Thompson Department of Chemistry and Chemical Engineering, Royal Military College of Canada, Kingston, Ontario, Canada

ABSTRACT

The ef f ic ient production of magnesium via the electrolyt ic decomposition of anhydrous magnesium chloride requires an accurate knowledge of the thermodynamic behaviour of magnesium chloride dissolved in other more stable molten chlorides. L i th ium chloride has been considered as a possible additive to the electrolyte to lower i ts density. This has the e f fec t of forc ing the elemental magnesium to sink to the bottom of the cell and thereby reduce the recombination of magnesium and anodically evolved chlorine in appropriately redesigned cells. The ternary system of MgCl2-LiCl-NaCl has been accurately studied w i t h the use of an electrochemical format ion cell to measure the emf associated w i th the format ion of magnesium chloride. Activit ies and other thermodynamic propert ies derived f r o m these measurements were compared to predictions using interpolat ion methods based on an understanding of the binary systems. This work, in conjunction w i th previous studies, is expected to lead to a general thermodynamic model fo r multicomponent MgCl2 electrolytes.

KEYWORDS

Magnesium chloride, l i thium chloride, sodium chloride, electrolysis, molten salts, thermodynamic propert ies, electrochemistry

INTRODUCTION

This work continues previous investigations of MgCl2 molten electrolyte systems d - 5 ) . Through the use of a careful ly designed reversible electrochemical format ion cel l , accurate measurements of the act iv i ty of MgCl2 in various multicomponent salt mixtures can be collected. The aim is to amass suff ic ient data so as to enable computation of thermodynamic properties (6 ) f o r melts of MgCl2 w i t h other alkal i or alkaline ear th chlorides in any proportion. In connection w i t h electrolyt ic magnesium production, these thermodynamic properties permit the precise calculation of

i) minimum energy/voltage fo r electrolysis i i ) pa r t i a l pressures of chloride vapours i i i ) thermal ef fects associated w i th feeding a cell iv) liquidus temperatures and associated "freeze" composition v) conditions to suppress hydrolysis

The present investigation is confined to MgCl2-LiCl-NaCl melts. In the LiCl r ich corner, these par t icu lar melts are of interest in the design of cells in which magnesium does not f loa t on the electrolyte surface. This may o f f e r an improvement in current eff iciency and yield other operating advantages.


EXPERIMENTAL

The electrochemical cell may be schematically represented as:

- C, Mg-B i( 1 )/ M G C L 2- L l C L - N A C L ( 1 )/ C L 2 ( g ), C +

The electrolyte was prepared from anhydrous MgCl2 (MgO less than 0.4%), anhydrous LiCl (Li20 less than 0.1%) and oven dried, 99.9% pure NaCl. All handling was performed in a glove box with a dry nitrogen atmosphere.

The cell assembly is shown in Fig. 1 (7). Graphite was used for both electrodes and crucibles. The chlorine electrode assembly was designed to allow chlorine saturation of the melt in contact with the positive electrode. This arrangement effectively separated the positive electrode from the negative one formed by the pool of magnesium alloy. With the taper closed on a homogeneous melt, the thin film trapped in the annular space was sufficient to permit an electrical connection but limit chlorine diffusion to the Mg electrodes. Since no current passes during a measurement of the emf, the cell resistance does not influence the voltage reading.

For those melts low in LiCl content, magnesium metal will float on the electrolyte complicating the construction and set up of the cell. Therefore the magnesium was pre-alloyed in all cases with 0.15 atom fraction Bi. Since Bi is much more noble than Mg, it acts as a simple diluent; a small correction can be made to the voltage measurements since the activity of Mg in this alloy system is known (11,12).

The Mg alloy rests in a magnesia crucible which electrically isolates the alloy pool from the graphite crucible during measurements. A quartz sheath about the graphite rod contacting the alloy pool completes this isolation at the time measurements are actually made. The quartz sheath does not remain in continual contact with the magnesium. With these essential precautions, the potential of the magnesium is solely determined by the alloy/electrolyte interface. Without these precautions, voltage measurements are unstable and lowered by about 20 mV.

Following the admission of chlorine to the cell in preparation for a measurement, there was always a slow rise to the steady potential especially above 1.5 V. This was attributed to slow chlorination of minor amounts of MgO in the electrolyte which was unavoidable in view of the strong tendency of MgCl2 to react with water vapour. By minimizing the extent to which the graphite electrode penetrated the electrolyte, reproducible potentials could be reached within an hour of admitting the chlorine gas. The pressure of chlorine was periodically measured with a manometer to enable small corrections of the emf to be made to that corresponding to a pressure of chlorine of exactly one atmosphere.

Throughout an experimental run lasting several hours, dry, pre-purified argon was circulated through the cell assembly to protect the electrolyte from hydrolysis. The temperature was measured with a calibrated chromel-alumel thermocouple. The temperature measurements are considered accurate to within 1 C . The emf of the cell was measured using a calibrated Fluke 8060A voltmeter. Tests of reversibility were occasionally made by polarizing the cell with a small applied current and observing the return (over a few seconds) to a reproducible open circuit voltage (stable to 0.1 mV over at least 10 minutes).


(-)

Teflon O-Ring Mg-Bi Alloy Seals Electrode^ Thermocouple

Copper Lid

Cooling Water

Quartz Sheath

(lowered as shown during a measurement)

MgO Crucible

Chlorine , * Electrode v)

Chlorine Inlet

Chlorine Outlet Argon Inlet

Argon Outlet

Quartz Vessel

Graphite Crucible

Japered Joints

(closed during measurements)

Fig. 1 - Schematic drawing of the electrochemical format ion cell


RESULTS

The overall spontaneous reaction which would occur at closed circuit is simply: M gU>

+ C 12«g, "

M*

C 12

(V

for which the Nernst equation can be written : aM g c i

Em = E° - £ i in - ^ — (2) ZF a • r

M g C l z where Em is the measured emf of the cell at temperature T arising from the formation of magnesium chloride from its elements and E° is the standard emf for this reaction when both reactants and their products are in their standard states (activities, a, and partial pressure, P, are unity). R and F are the gas constant and Faraday respectively.

From published thermodynamic data on the Mg-Bi system ( 8 ) , the relative partial Gibbs energy, AG , of magnesium in Mg-Bi melt at X = 0.85 is independent of temperature

Mg Mg

AG = R T in a = -2900 J/mol (3) Mg Mg

Thus each emf reading must be raised by 15 mV to correct for the presence of bismuth. Similarly a small correction can be made to bring the chlorine pressure to one atmosphere.

For a particular composition it was found that the corrected emfs varied linearly with temperature. The linear regression analysis for each experimental composition studied is given in TABLES 1.1 and 1.2 for two proportions of LiCl/NaCl. Similar equations for the MgCl2-NaCl binary and the MgCl2-LiCl binary were previously reported ( 1 , 5 ) . For all the measurements, the uncertainty is in the order of 1 mV. The voltages in TABLES 1.1 and 1.2 are proportional to the Gibbs energy of formation of magnesium chloride at various levels of dilution in LiCl-NaCl melts, the proportionality constant being two Faradays. Accordingly, by difference, the relative partial Gibbs energy (AG) associated with the dissolution of pure liquid MgCl2 into MgCl2-LiCl-NaCl melts can be determined for a range of temperatures and compositions. It also follows from the temperature dependence of the_ relative partial Gibbs energy that the partial entropy (AS) and partial enthalpy (AH) can be determined relative to liquid MgCl2.

The salient relationships are as follows:

AG = R T in a = -2 F (E - E°) (4) M g C l 2

M g C 1

2 AG AH - T AS (5)

M g C l 2 M g C l 2 M g C l2 The activity of MgCl2 is displayed on the Gibbs triangle in Fig. 2 for a temperature of 1000 K. The activity of MgCl2 is lower than the corresponding mole fraction, X M g C i 2 , at all compositions. The negative deviation is most pronounced in the NaCl rich melts.

The activities of NaCl and LiCl may be determined by the Gibbs-Duhem equation:

X d In a + X d In a + X d In a = 0 (6) M g C l 2

M g C 1

2

N a C1 N a C1 L l Cl L l C1

The integrations were carried out analytically by expressing the partial excess Gibbs energy of MgCl2 as a function of composition using a multiple linear regression technique (7 ,9) . The resultant activity isotherms for NaCl and LiCl are shown in Figs. 3 and 4 for 1000 K.


M g C l

TABLE 1.1

Regression l ines f o r t e m p e r a t u r e dependence

o f e m f i n M g C l 2- L i C l - N a C l me l ts

XL i C l _ . 0.333

Y + V N a C l L i C l S t a n d a r d

X T e m p . Range E = a - b x l O " T(K) D e v i a t i o n

(K) (V) d o "

3 V)

0 . 1 0 0 9 4 9 • - 1 1 1 4 3 . 2 5 1 2 7 0 . 5 5 6 3 1. , 6 8

0 . . 2 0 0 9 4 3 • - 1 1 0 6 3 . 1 9 8 4 7 0 . 5 6 7 2 1. , 14

0 . , 3 5 0 9 1 7 • - 1 0 9 0 3 . 1 5 7 7 6 0 . 5 9 0 6 1. 0 2

0 . , 5 0 0 9 2 0 • - 1 0 8 6 3 . 1 3 9 8 0 0 . 6 0 8 7 0 . . 8 3

0 . , 7 0 0 9 7 8 -- 1 0 8 7 3 . 1 3 5 4 1 0 . 6 3 7 7 0 . . 3 7

1 . , 0 0 0 1 0 0 4 • - 1 0 9 8 3 . 1 2 5 8 6 0 . 6 4 5 1 1. . 3 3

M g C i 2

TABLE 1.2

Regression l ines f o r t e m p e r a t u r e dependence

o f e m f in M g C l 2- L i C l - N a C l me l ts

X L i C1 - = 0.667

X + X N a C l L i C l

S t a n d a r d X T e m p . Range E = a - b x l O " T(K) D e v i a t i o n

(K) (V) d o "

3 V)

0 . 1 0 0 9 3 7 • - 1 0 8 4 3 . 2 4 5 7 4 0 . 5 7 8 3 0 . , 9 2

0 . 2 0 0 8 8 5 • - 1 0 6 8 3 . 1 7 4 2 6 0 . 5 6 3 4 0 . , 9 8

0 . 3 5 0 9 1 6 • - 1 1 2 0 3 . 1 4 8 1 9 0 . 5 9 1 0 0 . , 2 4

0 . 5 0 0 9 1 6 • - 1 0 7 5 3 . 1 4 2 8 1 0 . 6 1 6 9 0 . , 7 1

0 . 7 0 0 9 5 2 -- 1 0 8 1 3 . 1 3 6 9 4 0 . 6 3 6 5 2 . , 5 0

1 . 0 0 0 1 0 0 4 • - 1 0 9 8 3 . 1 2 5 8 6 0 . 6 4 5 1 1. . 3 3

S i m i l a r e x p e r i m e n t a l d a t a f o r t h e MgCl2-NaCl and MgCl2-LiCl b i n a r y m e l t s a r e t o be f o u n d in t h e w o r k o f K a r a k a y a and Thompson (i) and B a r n e s and T h o m p s o n ( 5 ) .

E r e f e r s t o m e a s u r e m e n t s c o r r e c t e d t o an a c t i v i t y o f m a g n e s i u m o f 1 and a p a r t i a l p r e s s u r e o f c h l o r i n e o f 1 a t m o s p h e r e .


NaCl 0.2 0.4 0.6 o.8 LiCl

'YiCi

Fig. 2 - Isoactivity lines fo r MgCl2 in MgCl2-LiCl-NaCl solutions at 1000 K

Fig. 3 - Isoactivity lines fo r NaCl in MgCl2-LiCl-NaCl solutions at 1000 K


MgCL

a-»

1000 K

0.8 LiCl

YiCI

Fig. 4 - Isoactivity lines for LiCl in MgCl2-LiCl-NaCl solutions at 1000 K As mentioned previously (equations (4) and (5)), the relative partial enthalpy of MgCl2, AHMgCi2, can be obtained from the experimental emf data. The relationship between the activity and the partial enthalpy may be expressed as:

d(ln aMgCi2) d(l/T)

AHMgC12 R

(7)

This equation applies to any component. Thus, from a knowledge of the temperature dependence of the activity isotherms in Figures 2, 3 and 4, it is possible to establish the integral enthalpy of mixing, AH.

AH = X -AH + X -AH + X -AH

M g C l M g C l N a C l N a C l L1C1 L i C l

(8)

Lines of constant integral enthalpy of mixing are plotted in Fig. 5. This enthalpy change corresponds to the heat effect when the pure liquid component salts are combined isothermally to form a solution.

MgCl, AH (kJ/mol)

NaCl 0.2 o.4 o.6 o.8 LiCl XLiCI

Fig. 5 - Constant lines of AH in ternary MgCl2-LiCl-NaCl solutions

NaCl


DISCUSSION

Various empirical methods are available to estimate the activities of components in ternary and higher order solutions. It is instructive to compare the measured activities of MgCl2 with those estimated by two of the most frequently used procedures (10,11). In both cases, the estimated activity in the ternary solution is based upon an accurate knowledge of the properties in the component binary systems. It is convenient to discuss the estimation procedure in terms of the excess Gibbs energy. For MgCl2, the partial excess Gibbs energy may be defined as:

G

E = RTZn a - RTZn X = RTin ? (9)

MgCl MgCl MgCl MgCl 2 2 2 2 where r̂, the activity coefficient, is a/X. The integral excess Gibbs energy is defined as:

G

E = X • G

E + X • G

E + X • G

E (10)

M g C l 2

M8

C 1

2

N a C1 N a C1 L i C1 L i C1

With reference to Figs. 6 and 7 and the equations which follow, the integral excess Gibbs energy may be estimated from the bounding binary systems by one of two methods.

KOHLER METHOD do)

G

E = (1 - X )

2 G

E + (1 - X )

2 G

E + (1 - X )

2 G

E (11)

p

M g C 1

2

a N a C1 b L 1 C1 C

TOOP METHOD (n)

GE = (1-X )

2 G

E

+ y

X l i" G

E

+ y Xn

:CA G

E (12)

p M g C l a X + X b X + X c 2 N a C l L i C l N a C l L1C1 E E E E

Gp is the estimated integral excess Gibbs energy at point p and Ga, Gb and Gc are the known integral excess Gibbs energies in the bounding binaries at the points shown in the figures.

The integral excess Gibbs energies in the bounding binary systems are known from the previous experimental work on MgCl2-NaCl (i), MgCl2-LiCl ( 5 ) , and the assessment of LiCl-NaCl (12).

MgCI 2

NaCl a LiCl Fig. 6 - Illustration of the Kohler estimation method


Fig. 7 - Illustration of the Toop estimation method

Since it is possible to estimate G for any ternary_ composition, the following equation may be used to provided the consequent estimated G M g C 1

2

G

E = G

E + (1 - X ) ' ( - i v - 1 <

13>

MgCl MgCl I dX I 2 2

V MgCl 'LiCl /NaCl 2

The partial derivative is evaluated along a line of constant LiCl/NaCl proportion. From the estimated partial excess properties from MgCl2, the consequent activity can be calculated using equation (9). TABLE 2 shows experimental and estimated activities for a range of compositions likely to be of possible interest in electrolysis. Clearly, there is a significant difference between the measurements and estimates. The extreme difference between the activities at 0.05 mole fraction MgCl2 at the equimolar property of NaCl to LiCl corresponds to a voltage effect in the experimental cell of 16 mV as compared to experimental uncertainty in the order of 1 mV.


TABLE 2

Act iv i t i es o f MgCl^ obta ined f r o m e x p e r i m e n t a l measurements

compared w i t h p red ic t ions using t h e Koh le r do) and Toop (11) methods

X + X N a C 1 L i C 1

MgCl

0 0 .2 0.333 0 . 5 0.667 0 .8 1.0

E x p . 0.0120 0.00659 0 .00456 0, . 0 0 2 9 9 0 .00204 0.00155 0.00108 0.05 Toop 0.0120 0.00794 0 .00596 0, . 0 0 4 0 6 0 .00269 0.00189 0.00108

Koh. 0.0120 0.00865 0, .00647 0. . 0 0 4 3 6 0 .00284 0.00197 0.00108

E x p . 0.0287 0.0181 0, .0135 0. , 0 0 9 4 6 0 .00675 0.00521 0.00360 0.10 Toop 0.0287 0.0201 0. .0157 0. , 0 1 1 3 0 .00792 0.00585 0.00360

Koh. 0.0287 0.0229 0. .0180 0 . , 0 1 2 7 0 .00873 0.00629 0.00360

Exp. 0.0507 0.0358 0. ,0284 0 . , 021 1 0 .0157 0.0124 0.00866 0.15 Toop 0.0507 0.0376 0. ,0305 0 . 0 2 3 0 0 .0170 0.0131 0.00866

Koh. 0.0507 0.0435 0. ,0358 0 . 0268 0 .0192 0.0143 0.00866

E x p . 0.0778 0.0608 0. 0506 0 . 0397 0 .0308 0.0249 0.0178 0.20 Toop 0.0778 0.0613 0. 0515 0 . 0407 0 .0315 0.0253 0.0178

Koh. 0.0778 0.0707 0. 0610 0 . 0479 0 .0360 0.0278 0.0178

It is evident from TABLE 2 that the activities estimated by either method are somewhat greater than the experimentally determined values. The Toop method, which treats the MgCl2 differently than the LiCl and NaCl, gives a somewhat better result. It appears that for calculations where the best accuracy is not required, it may be sufficient to use one of the above estimation methods as a substitute for direct experimental measurements in systems similar to the one studied in the present investigation.

SUMMARY

Activities and heats of mixing have been measured in the MgCl2-LiCl-NaCl molten salt system using a carefully designed reversible electrochemical cell. The activities for all components at all compositions are less than the corresponding mole fractions. For MgCl2, the largest deviations occur near the NaCl corner. The integral heat of mixing, extracted from the Gibbs-Duhem equation by integration, is negative at all compositions with the largest negative value at approximately 0.45 mole fraction of MgCl2 in the MgCl2-NaCl binary. Comparisons were made of measured activities with those based on the empirical interpolation methods of Kohler do) and Toop (ii). The measured values were lower than the estimated values in each case.

ACKNOWLEDGEMENT

The authors acknowledge with thanks the financial support of the Natural Sciences and Engineering Research Council of Canada.


REFERENCES

1. I. Karakaya and W.T. Thompson, J. Electrochem. S o c , p. 702, 133 (1986).

2. I. Karakaya and W.T. Thompson, J. Chem. Thermodynamics, p. 859, 18 (1986).

3. I. Karakaya and W.T. Thompson, Can. Met. Quarterly, p. 307, 25 (1986).

4. J. Zhu, I. Karakaya and W.T. Thompson, J. Electrochem. S o c , p. 122, 135, #1 (1988).

5. C. Barnes and W.T. Thompson, Can. Met. Quarterly, p. 267, 27, #4 (1988).

6. W.T. Thompson, C.W. Bale and A.D. Pelton, "F*A*C*T Facility for the Analysis of Chemical Thermodynamics", McGill University, Montreal, Canada (1991).

7. B.R. Davis, M.Sc.(Eng.) Thesis, Dept. of Mats. Sci. & Met., Queen's University, Kingston (1990).

8. J.J. Egan, Acta Metallurgica, p. 560, 7 (1959).

9. A.D. Pelton and S.N. Flengas, Canadian Journal of Chemistry, p. 2808, 47 (1969).

10. F. Kohler, Monatsch, Chemie, p. 738, 91 (1960).

11. G.W. Toop, Trans. TMS-AIME, p. 850, 233 (1965).

12. A.D. Pelton and J. Sangster, Journal of Physical and Chemical Reference Data, p. 519, #3, 16 (1987).

201

203

Recovery of cryolite from spent potlining of Al reduction cells by flotation method

Zhai Xiujing, Qiu Zhuxian Department of Non-ferrous Metallurgy, Northeastern University of Technology, Shenyang, Liaoning, China

ABSTRACT

The possibility of cryolite recovery from spent potlining by flotation method has heen investigated in the laboratory. The results of experiment showed that the acidity and concentration of liquid and particle size of material affected the recovery rate of flotation. Suitable collecting agents, modifiers and flothing agents were examined in ex-periments. In suitable conditions,the recovery of cryolite reached over 9 0 % . The treat-ment of fluoride—containing waste water was examined and the results were satisfac-tory.


INTRODUCTION

Spent potlining (SPL) from Hall cells used in the aluminum reduction process has long been regarded as a valuale resource by virtue of its chemical values and energy content (1) . At same time,the treatment of spent potlining was an important part of en-vironmental controls in the aluminium production(2).

The reaults of analysis showed that there were fluoride 3 0 % and carbon 70% in spent potlining. The recovery of fluoride and cryolite was the major aim of many stud-ies. J. P. McGeer ,e ta l (3 ) . explored the possibility of recovering fluorides from potlin-ing by vacuum distillation. M. M. Will iams(4) investigated treatment of carbon lining, crushed potlining was exposed to steam at approximately 15 psi or higher. B • Gnyra ( 5 ) examined the sulfuric acid method to treat the spent potlining. This process includ-ed a separation of silicon tetrafluoride from the hydrofluoric acid, lime was added to re-act with the potlining chemicals by B • Gnyra,R • R • Sood,and J • D • Zwicker(6) It is obvious that the carbon and cryolite have different surface character,such as ,wet-tability. S o , we used flotation method to separete carbon from the salt fluoride in spent potlining,in order to realize comprehensive utilization of spent potlining material.

EXPERIMENTAL

Recovery of salt fluorides by flotation. In essence the flotation cell consists of a stiring device made of steel with an air

device and a flotation trough which was made of plastics.

Materials

The spent potlining was taken from one aluminium plant. It was ground in ball mill to sizes of 0. 10mm, 0. 15mm, 0. 18mm, 0. 2 8 m m , 0. 45mm. Eight kinds of col-lecting agents were prepared as alcohol solutions. Two kinds of flothing agents were made into the aqueous solution. Five kinds of modifiers were made into the alcothol so-lution.

Procedure

The ground spent potlining and water were added into the flotation trough, then, adjusted the concentration of solution as needed and operated the flotation machine. Af-ter five minutes of operation, modifiers, collecting agents and flothing agents were added in order.

when some foams were formed, operated the air device and began to float. The flotation time was about ten minutes.


Analysis

Two products of flotation were obtained. The foam which contained carbon main-ly and the fluorides remained in water. The content of carbon and fluorides in the prod-ucts was determined by chemical analysis method. The content of fluoride in waste wa-ter can be determined by F-ion selecting electrode.

The treatment of fluoride waste water It is important to treat fluoride waste water. Otherwise secondary pollution will be

generated. The precipitate flotation method was used to examine the removal of fluo-ride.

The solution of CaO, MgCl2 and A1C13 of different concentrations were added to the waste water and there formed precipitate in a suitable acidity. The collecting agent, sodium dodecyl sulfate, was added and floated in a flotation column.

The results about flotation recovery of salt fluoride were given in Figure 1 to 1 1 , and that of the treatment of fluoride waste water in Figure 12 to 14. Effect of pH

RESULTS A N D DISCUSSIONS

—o— sodium dodecyl sulfate — » ^ hexadecyl acid —<*>— salicylate

0 3 5 7 9 11 13

Figure 1—Effect of acitity[pH]|on recovery rate

Figure 1 gave the effect of pH with three kinds of collecting agents, when pH val-ue increased, the recovery rate slowly increased, and when pH value reached over 7 , the recovery rate did not change any longer.

pH


The HCI H 2S 0 4 and H N O 3 were used as modifiers of pH (see Figure 2 ) , i t was shown that different acids will.affect the result and HCI was ideal modifier of acidity.

pH

Figure 2 —Effect of acidity [ p H ] with three kinds of acids on recovery rate

Effect of concentration

lOOh

80

60

40 > O

o 20

o— - •

0. 10mm

10 15 20

Figure 3—Effect of concentration of solution [ W % ] on recovery rate


The experiments examined the influence of concentration of solntion for recovery rate. Figure 3 showed that change of concentration had not obvious effect. In the flota-tion, the concentration of solution should be chosen as high as possible.

Effect of particle size In the experiment, the size range changed from 0. 10mm to 0. 45mm. Figure 4

showed that the effect of size was determined by different collecting agents. As carbon oil was used as collecting agents, recovery rate incveased with increasing of size. Con-versely, with increasing of size, recovery rate decreased when sodium dodecyl sulfate was collecting agents. For cetyl pyridinium bromide,suitable size was from 0. 15mm to 0. 20mm.

100

« 80 I I

t3 60

o o w 20 Pi

0. 10

• sodium dodecyl sulfate

. carbon oil

— cetyl pyridinium _J I I L_

0. 20 0. 30

[ m m ]

0.40

Figure 4—Effect of grain size of sample (mm) on recovery rate

In general,the grain size of sample determined the flotation performance. If size was large enough,it was not able to be floated by foam. The carbon oil can float carbon grain of less than 0. 45mm,only sodium dodecyl sulfate was appropriate to floating the size of less than 0. 15mm.

Effect of collecting agents

Eight kinds of collecting agents were used to examine the recovery rate of flota-tion. Figure 5 shows the effect of salicylate, sodiUm dodecyl sulfate, hexadecyl acid and sodium dodecyl sulfonate.

It was obvious that sodium dodecyl sulfonate was a good collector, and salicylate was poor one. It was found that salicylate generated poor foam and sodium dedcyl sul-fonate gave much foam in the Flotation. It showed carbon grains needed mnch foam so


that it may be floated. But In Figure 6. the effect was obviovs. An overamount of cetyl pyridinium bromide may control the separation of carbon and salt fluoride. On the con-trary , with increasing amount of naphthyl hydroxamic acid, the recorevy rate maybe

100

Qi 80 i i

eS 60

£ 40 O

I 2 0 k

- o sodium dedecyl sulfonate

hexadecyl acid

—o> sodium dodecyl sulfate

e salicylate

0. 1 0. 2 0. 3

[ k g / t ]

0 .4

Figure 5 —Effect of weight radio of collector to sample [ k g / t ] on recovery rate

100

>>

O o

60^

40F

20H

o — hexadecyl amine

— • — naphthyl hyridinium acid

cetyl pyridinium bromide

0. 1 0.2

[ k g / t ]

0. 3 0. 4

Figure 6—Effect of weight ratio of collector to sample [ k g / t ] on recovery rate


raised. It may be explaned that high amount of naphthyl hydroxamic acid can generate much foam so as to float the carbon grains. Figure 7 shows the effect of carbon oil. car-bon oil was a collector using to recover carbon, coal, e t c It was cheap as compared to other collectors. But its recovery rate was lower than that of hexadecyl amine and sod-inm dodecyl sulfonate.

i i

.2 *+->

1 0 0 h

80

60

% O a>

40h

20

— o — carbon oil

[ k g / t ]

Figure 7 —Effect of weight ratio of collector to sample [ k g / t ] on recovery rate

Effect of modifiers

Figure 8 and 9 show effect of five kinds of modifiers, sodium silicate and oxalic acid were ordinary modifiers. In the flotation. Figure 8 showed that with increasing amount of sadim silicate, the recovery rate increased. Howerer, oxalic acid did not change recovery rate when its amount was increased. Oxacic acid was prior to sadium silicate in this experiment.

Complex agents were used as modifiers in recent years. They can not only make the salt fluoride increase wettablity but also form complex with ions in solution. Fignre 9 shows that citric acid is a good modifier in this experiment. Cyauest 40 is not ideel modifier and when Its amount increased, the recovery rate is decreased. It may be ex-plained that cyauest 40 can complex some ions.

It was said that amylum is a good modifier in carbon flotation. In fact, it re-strained flotation. It was found that amylum played a role of flocculant in the experi-ment.


& 80 i i

e3 60

40 > O o £ 20

oxalatic acid

sodium silicate

-J I I L_ 1

[ k g / t ]

Figure 8 —Effect of weighe radio of modifier to sample [ k g / t ] on recovery rate

100

80

c3 60

S 40 > o o

20

- o — citric acid

- • — cyauest 40

- c d — amylum

[ k g / t ]

Figure 9 —Effect of weight radio of modifier to sample [ k g / t ] on recovery rate

Effect of frothing agents


[ k g / t ]

Figure 10—Effect of weight ratio of flothing agent to sample [ k g / t ] on recovery rate

The terpenol was an important frothing agent. It was widely used in flotation. Fig-ure 10 shows that terpenol was prior to isoamyl alcohol. It was found that terpenol gave much and tiny foam. At the sane time,the amount of frothing agent should be contolled in the experiment, because it became depressant when its amount was increased to more than l k g / t in this experiment.

Effect of circulationg water

It was necessary that circulating water was used and controlled in the flotation. There were some kinds of salt fluoride in potlining,they had some solubility in water. The results of experiment indicated that the use of circulating water can affect the re-covery rate and content of F-ion with increasing circulating frequency. Figure 12 shows that the content of F-ion increased to 0. 25g / l from 0. 115g/ l when circulating frequency was six. Then the change became smaller and smaller. It was noted that fluo-ride promoted the floataion rate (see Figure 1 1 ) . Hence F—ion itself may be a modifi-er.

The treatment of fluoride waste water. Figure 13 shows the effect of mol. ratio of precipitant to F-ion on the flotation

rate,and Figure 14 shows the effect of pH value. when cao was precipitant, its range of pH was from 1 to 14. It was explained that

C a

2+ and F formed CaF2 precipitate in solution, Because it was known that pH ralue

reached more than l l , C a ( O H ) 2 precipitate can be formed. The suitable pH range of A1C13 was from 5 to 8 and MgCl2 was from 10. 5 to 12. This result explained that M g

2 ++ and Al

3+ formed M g ( O H ) 2 and A l ( O H ) 3 precipitate, separately, and F~ were


removed from the solution by adsorption. It was understood that CaF2 and A 1 20 3 obtained from flotation can he used to alu-

minum production,and M g ( O H ) 2 can form M g C 03 with C 0 2. M g C 03 is also an additive in aluminium production.

ctf

100

80

60

40 > o o

g 20

0. 45mm

0. 15mm

I 1 2 3 4 5 6

[ frequency]

Figure 11 —Effect of circulating water freguency on recovery rate

Figure 12 —Effect of content of F - i o n [ g / l ] on circulating freguency


M Figure 13 —Effect of mol ratio of precipitant to F-ion[<J>]

on removing rate

10 12 14

[ p H ]

Figure 14 —Effect of acidity (pH) on removing rate

rem

ov

ing

rat

e [%

]

213

— o — CaO

— • — A1C13 — M g C l 2

— o — CaO

— #_ A1C13 —®— MgCl2


CONCLUSION

It was found that the flotation method can be used to separate the carbon and cry-olite from spent potlining of aluminium reduction cells. The recovery rate of cryolite can reach more than 9 0 % 0

The result of experiment gave the optimum acidity of solution,the concentration of solution and the grain size of sample for the treatment.

Eight kinds of collecting agents were compared in the experiment. It showed that most of them could be used in flotation of potlining. Five kinds of modifiers and froth-ing agents were also examined. The result showed that two kinds of modifiers and one kind of flothing agent were suitable to this experiment.

The experinents examined the use of circulating water and content of F ion in wa-ter. After treatment,99. 5% F ion was removed by using CaO,MgCl2 and A1C13 as pre-cipitants.

REFERENCES

Lee C. Blayoden, Alcoa and Seymour, G. " Spent Potlining Symposium ", Journal of Metals,No. 4 , 1 9 8 4 , 2 2 — 3 2 . Jack H. Goldman,Ph. D. ,"Regulatory Reguirements for spent potlinig",Lingt Metals, 1 9 8 7 , 6 4 7 - 6 5 8 . J. P. McGeer, V. V. Mirkovich and N. W. F. Phillips, "Recovery of Fluoride from potlining, "U. S. Patent,No. 2 , 8 5 8 , 1 9 8 , 2 8 October, 1959. M. M. Williams, "Treatment of carbon lining from Reduction cells, "U. S. Patent, 3 ,

6 3 5 , 4 0 8 , 1 8 January, 1972. B, Gnyra, "Processing Spent Cell Lining Materials from Electrolytic Reduction Cells for prlduction of Aluminun,"U. K. Patent,Applicable G B 2 , 0 5 6 , 4 2 2 . B. Gnyra, R. R. Sood,and J. D. Zwicker, "treatment of wastes containing water — leachable Fluorides,"U. K. Patent,Application GB 2 , 0 5 6 , 4 2 5 .

217

Vacuum casting of aluminum alloys

J.L. Dion, M. Sahoo Metals Technology Laboratories, CANMET/EMR, Ottawa, Ontario, Canada

ABSTRACT

The vacuum casting technique has been applied to two types of aluminum alloys (356.2 and 319) to produce test castings using both green sand and C02-bonded sand moulds. For control pur-poses, castings were also poured by the conventional gravity-casting practice in these two types of sand moulds. The various foundry characteristics such as fluidity, mould-filling time, porosity distribution etc. and the mechanical properties have been described. It is shown that the increase in fluidity and decrease in mould filling time observed in the vacuum casting process would be an asset in producing thin-wall castings. The degree of vacuum in the vacuum box and, to a lesser extent, the casting temperature control the distribution of porosity for gassy melts.

KEY WORDS

Vacuum casting, sand casting, aluminum alloys, fluidity, mould filling time, sand penetration, cold shut.


INTRODUCTION

Recent publications have described the vacuum casting process in detail (1-3). Some of the advantages of the castings produced by this process are thin-wall, near-net shape, complex shapes and metallurgical integrity. As described in an earlier publication (4), this process was also developed at this laboratory from our in-house work on low-pressure die casting and lost foam casting processes. In this method, the green sand or C02-bonded sand mould is filled by drawing the metal up a hollow tube from below the exposed molten metal surfaces. The pressure differential between the atmospheric pressure on the molten metal held in the ladle and the partial vacuum produced in the mould helps to fill up the mould cavity. This technique has the potential for minimizing oxide formation as well as entrapment.

Although some U.S. foundries are using the vacuum casting process, no Canadian foundry has yet ventured to produce castings by this process. In order to develop the Canadian capability to utilize this advanced casting process, additional work has been carried out on aluminum alloys. This paper focuses on the foundry characteristics and mechanical properties of two aluminum alloys produced by vacuum casting and compares them with those produced by the gravity casting process.

EQUIPMENT

A schematic of the vacuum casting unit is shown in Fig. 1. The disposable feed tube extends from below the molten metal surface to the sprue of the mould. A fibrefax gasket is inserted between the flange of the tube and the flange of the container. The container is 92 cm (36 in.) x 92 cm (36 in.) and 76 cm (30 in.) deep. A restraining bar is used to prevent movement of the mould during pouring or handling of the container. The top of the container which rests on a rubber gasket is held in place by vacuum during casting. The vacuum was set in the container by opening a large valve on the right side. This valve was connected to a large vacuum furnace by a 9 cm (3.5 in.) hose. The whole assembly, consisting of the container, valve, hose, etc., is seen better in Fig. 2. A large vacuum furnace was used as a vacuum reservoir as the regular vacuum pump was unusable because of insufficient capacity. The degree of vacuum in the casting unit was controlled by the degree of vacuum in the reservoir (larger furnace).

Figure 1 - Schematic of the container for vacuum casting of regular sand mould, (neg 20085)

VACUUM GAUGE -

- VALVE RELIEF VALVE

VACUUN OUTLET

- RESTRAINING BAR

- MOLD

HOLD CAVITY

DISPOSABLE TUBE - FLANGE

LIQUID METAL

LADLE


Figure 2 - Container used for vacuum casting of regular sand mould, (neg 19249-40)

EXPERIMENTAL

AHQYS and Test Castings

In the present work, the casting parameters and mechanical properties of two types of aluminum alloys (alloy 319 or the Al-Si-Cu alloy and alloy 356.2 or the Al-Si-Mg alloy) were studied using both green sand (McConnellsville AFS 120) and C02-bonded sand (silica 45) moulds. The chemical compositions of the alloys are given in Table I.

The following three test castings were cast to evaluate the casting parameters, foundry characteristics and mechanical properties associated with the vacuum casting process:

(i) fluidity spiral to determine the casting fluidity;

(ii) a step block (Fig. 3) to determine the mechanical properties as a function of casting thickness. The overall dimensions were 20 cm (8 in.) long by 15 cm (6 in.) wide by 2.5 cm (1 in.) and 6 cm (2 1/2 in.) thick. The pouring sprue was 16 mm (5/8 in.) in diameter;

(iii) an elbow fitting (Fig. 4) to evaluate the casting parameters including soundness. The symmetrical nature of the elbow permitted the use of two different types of risers, e.g., cylindrical side riser and trapezoidal top riser as shown in Fig. 5.

Table I - Chemical analysis

Alloy Melt Element %

type No. Cu Fe Mg Mn Si Sr Ti Zn

319 285 1.6 0.10 0.002 0.002 5.90 0.016 0.020 0.045

319 286 1.3 0.09 0.009 0.003 5.66 0.016 0.022 0.044

356.2 294 0.003 0.11 0.35 0.002 6.80 0.012 0.064 0.003

356.2 283 0.002 0.11 0.32 0.002 7.02 0.006 0.098 0.005 356.2 284 0.002 0.12 0.31 0.002 6.72 0.005 0.082 0.003


Figure 3 - Step block used for tensile test specimens, (neg 18226)

r (22.2)

L 2JrDIA

(63.5)

7 j PI A

(181.0)

DIMENSIONS P^-. (mm)

Figure 4 - Dimensions of the elbow casting, (neg 275)

Every vacuum casting was accompanied by a gravity-poured casting with similar gating and risering systems except that the sprue was inverted for vacuum casting as shown in Fig. 3. Both the step block and the elbow fitting were cast under two different melt conditions (gassy as well as fully degassed).

3 £ DIA (81.0) s


Figure 5 - Elbow casting with two types of risers, (neg 19271)

Casting Procedure

The various trial runs that led to an optimization of the vacuum casting process have been described in Ref. 4. The following casting procedure was used in the present work. The vacuum was set-in only after the feed tube, 320 mm (12.75 in.) long, was plunged into the melt. This practice was followed in order to minimize the drying of the green sand mould under vacuum. This drying action was even faster with a leak in the system like the one provided by the feed tube, resulting in rapid erosion of the mould and leading to its eventual collapse. Such erosion problem was also noted with the CO2 sand mould, although at a much lower rate. At the end of the cast time, the valve was closed, the relief valve opened and the whole unit lifted from the ladle.

The cast time defined as the interval of time the vacuum is applied to prevent the metal from leaking down the tube was much too long to permit the risers to act. It could be shortened in the present work by providing freezing spots, 51 mm (2 in.) wide by 5 mm (3/16 in.) thick, on the runners as shown in Fig. 6 for the elbow. In the case of the step block, the freezing spot was connected to the riser (Fig. 3) and was 63 mm (2.5 in.) wide and 3 mm (1/8 in.) thick.

Figure 6 - Runner system for elbow fitting showing a freezing spot (gradual flattened area) on each side of the sprue, (neg 19270)


Figure 7 - Comparison of the fluidity of alloy 319 in green sand moulds for both gravity and vacuum casting processes, (neg 21182)

Aluminum Alloy 319

30 I J ' i i i 1 —' 1 > 1 ' 1 J 1 630 650 670 690 710 730 750

Temperature deg. C

Figure 8 - Comparison of the fluidity of alloy 319 in C02-bonded sand moulds for both gravity and vacuum casting processes, (neg 21179)

Fast-response thermocouples were located in the mould to measure the drop in metal temperature from the ladle to the sprue entrance in the vacuum box.


Fluidity

Fluidity data obtained by vacuum and gravity casting of fluidity spirals in both green sand and C02-bonded sand moulds are shown in Fig. 7 and 8 for alloy 319 and in Fig. 9-11 for alloy 356.2. The

• Gravity, green sand

+ Vacuum, green sand

Sp

fral

len

gth

(cm

) S

pira

l le

ngth

(cm

)

• Co2+Zr wash, gravity

+ Co2*Zr wash, vacuum


Figure 9 - Comparison of the fluidity of alloy 356.2 in green sand moulds for both gravity and vacuum casting processes, (neg 21181)

Aluminum Alloy 356.2

60

55

50

45

?

1 40

±

I 35 -

30

25

20

15

• Gravity, Co2

+ Vacuum, Co2

O 680

Temperature deg. C

Figure 10 - Comparison of the fluidity of alloy 356.2 in C02-bonded sand moulds for both gravity and vacuum casting processes, (neg 21180)

pouring temperatures varied between 640 and740°C for alloy 319 and between 630 and730°C for alloy 356.2. The liquidus temperatures for these two alloys are respectively 604°C and 613°C.

In each case, there is an increase in the spiral length in the vacuum casting condition. It must be remembered that, in vacuum casting, the molten metal has to travel 320 mm up the feed tube before it gets to the sprue entrance. Thus, the temperature of the molten metal at the sprue entrance is less than that measured in the ladle. This was confirmed by measuring the temperature at the ladle and the sprue entrance using fast-response thermocouples. For alloy 356.2, temperature drops of 20, 29 and 37°C were noted for ladle temperatures of 690,665 and 655°C, respectively. Although such measurements

• Green sand, gravity

+ Green sand, vacuum

Spira

l le

ngth

(cm

)


Figure 11 - Comparison of the fluidity of alloy 356.2 in C02-bonded sand moulds with a zircon wash for both gravity and vacuum casting processes, (neg 21183)

were not repeated to check the reproducibility, these three sets of data show that the temperature drop at the sprue entrance increases as the ladle temperature decreases.

The effects of moulding sand and mould wash on casting fluidity were studied in detail for alloy 356.2. As shown in Fig. 9-10, better casting fluidity is obtained for green sand than for C02-bonded sand for both the casting processes. Use of zircon mould wash in the latter produces significant increases in the casting fluidity (Fig. 11) which is also higher than that obtained for green sand.

Previous work at CANMET (5) on Zn-Al alloys has shown that the C02-bonded sand produces relatively more chilling action on the casting than the green sand. As a result, the fluidity is supposed to be reduced for the C02-bonded sand mould. However, the application of a zircon wash acts as an insulating layer and reduces the chilling action. This is probably the reason for the increased fluidity in C02-bonded sand moulds with a zircon wash.

Mould Filling Time

The mould filling time was measured on the elbow casting with sensors positioned at the sprue entrance and on top of the trapezoidal riser. As shown in Table II for alloy 319, it decreased with increases in the degree of vacuum. Similar observations had been made for the higher density monel (4).

Table II - Mould filling time with Al-319

Degree of vacuum Time

Process (mm) (s)

Vacuum 210 1.9

Vacuum 380 2.8

Vacuum 510 3.1

Vacuum 550 3.0

Gravity - 6.9

• Gravity, Co2+wa«h

+ Vacuum, Co2*wa»h

?

.c

m

Temperature deg. C


The results of increased casting fluidity and reduced mould filling time associated with the vacuum casting process show that it should be possible to cast thinner sections by vacuum casting than by gravity casting.

Casting Quality

The main problem in vacuum casting was sand penetration which was a function of the type of sand, metal temperature and the degree of vacuum. In the case of C02-bonded sand moulds, it could be eliminated by applying a zircon wash on the mould cavity provided the amount of vacuum was >275 mm and the pouring temperature was maintained between 680 and 690°C for each alloy. It may be noted that, in the case of aluminum alloys, it is not necessary to have such high vacuum since the mould cavity could be filled even at a pressure of 550 mm.

The degree of vacuum had a greater effect on sand penetration than the casting temperature. At much lower pouring temperatures, a few spots of sand penetration did appear. Lowering the amount of vacuum (>275 mm) corrected the situation. In green sand moulds with a somewhat finer sand, the maximum amount of vacuum was 350 mm with the pouring temperature in the range of 670-680°C to eliminate the sand penetration. Here again, the control of the degree of vacuum was the most important factor. The above data were determined with the elbow casting. However, they are equally applicable to the step block casting.

The cylindrical and trapezoidal risers were both effective in producing sound castings for both gravity and vacuum casting. Corrective measures such as an increase in riser size, use of insulating sleeve for risers and a chill on the bottom part of the flange had the same effect on casting quality for both vacuum and gravity casting. A chill on the bottom flange is used to overcome a shrinkage defect. The advantage of the side riser is that it reduces the amount of sand in the mould. The quality of the casting indicates that the top riser would be preferred at low pouring temperatures.

Figures 12 and 13 show the elbow fittings vacuum cast at 665°C and 655°C, respectively with alloy 356.2. A cold shut or misrun defect appears on the bottom of the flange with the side riser. By contrast, the cold-shut defect appears on the top of the flange in the gravity-cast elbow poured at 645°C (Fig. 14).

Figure 12 - Vacuum-cast elbow at 665°C showing cold-shut defect on flange with side riser, (neg 21096)


Figure 14 - Gravity-cast elbow at 645°C showing cold shut defect on top of flange, (neg 21095)

A steel chill 25 mm (1 in.) thick was used on the bottom half of each flange for both vacuum and gravity casting. Vacuum casting at 655°C without the use of the chill did not exhibit such cold shut defect. In view of the fact that a 35°C drop in metal temperature at the sprue entrance can be expected at a ladle temperature of 655°C and with a liquidus temperature of 613°C, the appearance of the cold shut defect can be attributed to the severe chilling action caused by the metal chill. Despite such a defect, this study shows that it is possible to fill the mould cavity at a low superheat in vacuum casting. It may be noted that all the castings were found to be sound internally.

Figure 13 - Vacuum-cast elbow at 655°C showing cold-shut defect on flange with side riser, (neg 21094)


Figure 15 - Flange section of the elbow produced by vacuum casting. Flange with side riser is on the left, (neg 21172)

n|wi{i CENTIMETRES

Figure 16 - Macrographs of the upper half of the thick section of the step block casting, (a) vacuum cast (neg 21174)

All the above castings were produced from a fully degassed melt. With a gassy melt, however, there was a difference in the quality of the castings produced by gravity and vacuum casting. Figure 15 shows the sections of both flanges of a vacuum-cast elbow. In this figure, a large gas hole is observed at the top of the flange with a side riser. The absence of the gas hole in the flange with a top riser indicates that the gas has probably escaped through this top riser. The gravity-cast elbow, on the other


Figure 16 - Macrographs of the upper half of the thick section of the step block casting, (b) gravity cast (neg 21173)

hand, did not exhibit any such gas defect. Instead, well-dispersed porosity was observed in each flange regardless of the type of riser. Similar observations were made with the step block also. Figure 16(a), which is the macrograph of the upper half of the thick section of the vacuum-cast step block, shows a network of gas holes at the top. By contrast, the gravity-cast step block exhibits well-dispersed porosity [Fig. 16(b)]. This behaviour is similar to the observations made during the reduced pressure test of copper- and aluminum-base alloys (6). The presence of the gas bubbles in the top of the thick section of the step block indicates that, for gassy melts, the gas pressure inside the molten metal is greater than the pressure in the vacuum chamber. The gas tries to escape from the melt because of the difference in pressure and is trapped in the top section due to the formation of the solid surface skin. Reducing the degree of vacuum should minimize or eliminate this defect in vacuum casting.

Table III - Mechanical properties of Al-319* produced by both gravity and vacuum casting

0.2%

Section offset 0.5%

Melt thickness UTS YS YS % % Specific

No. Sand Process (mm) (MPa) (MPa) (MPa) Elong RA gravity

285-1 CQ2 sand Gravity 25 145 97 106 2 2 2.70

63 132 95 103 2 2 2.70

285-2 CO2 sand Vacuum 25 147 95 106 2 2 2.70

275 mm 63 132 92 103 2 2 2.69

286-1 Green sand Gravity 25 137.2 98 109 2 1 2.69

63 118 92 102 1 1 2.68

286-2 Green sand Vacuum 25 140 99 109 2 2 2.69

350 mm 63 118 89 100 2 2 2.67

*Both melts were gassy.


Table IV - Mechanical properties of Al-356.2 produced by both gravity and vacuum casting

0.2%

Melt No. Sand

Melt condition Process

Section thickness

(mm) UTS (MPa)

offset 0.5% YS YS %

(MPa) (MPa) Elong %

RA Specific gravity

283-1 CO2 sand Degassed Gravity 25 136 78 87 3 4 2.67

63 125 74 83 2 3 2.66

283-2 CQ2 sand Vacuum 25 138 80 86 4 4 2.66

275 mm 63 128 763 84 3 3 2.65

284-1 Green sand Degassed Gravity 25 133 88 97 2 3 2.66

63 124 89 96 2 3 2.66

284-2 Green sand Vacuum 25 124 85 93 2 2 2.64

350 mm 63 113 84 92 2 2 2.62

294-1 CQ2 sand Not Gravity 25 137.5 81 90 4 5 2.65

degassed 63 126 77 87 3 3 2.64

294-2 CO2 sand Not* Vacuum 25 133 82 90 4 5 2.62

gassy 275 mm 63 122 77 85 3 4 2.62

* Not degassed, but examination of sectioned surfaces indicated it to be not gassy.

Mechanical Properties

The tensile properties of the test bars machined from the step blocks are shown in Tables HI and IV for alloys 319 and 356.2, respectively. The type of sand mould and the melt conditions are also included in these two tables. These results show that the mechanical properties are unaffected by the casting process considering the type of mould and the melt condition. However, in each case, slightly better tensile properties were obtained in the thinner sections.

CONCLUSIONS

1. The vacuum casting process has been applied to two types of aluminum alloys (alloys 319 and 356.2) to evaluate the casting fluidity, mould filling time, mechanical properties and casting soundness.

2. The fluidity of each alloy increases during vacuum casting. Relatively higher fluidity is observed in green sand moulds than in C02-bonded sand moulds. A zircon wash applied to the CO2 sand mould produces significant increases in fluidity.

3. Mould filling time is decreased with vacuum. It also decreased in proportion with the amount of vacuum.

4. It should be possible to produce thin-wall castings in vacuum casting because of the increase in fluidity and decrease in mould filling time.

5. Sand penetration, which is a major problem in vacuum casting, could be minimized by controlling the fineness of sand, mould wash, amount of vacuum and the casting temperature.

6. It is possible to fill the mould cavity at very low superheats in vacuum castings. However, cold shuts may be observed if metal chills are used.


7. Gas holes can be observed in the upper part of a casting if a gassy melt is used. However, reducing the amount of vacuum could eliminate/reduce this defect.

8. The mechanical properties are not improved in vacuum casting irrespective of the type of sand mould, section thickness and melt condition.

ACKNOWLEDGEMENTS

The technical assistance provided by the foundry support staff is gratefully acknowledged. Our special thanks are extended to Mr. R. Matte for his support on data logging and various mechanical designs.

REFERENCES

1. R.D. Blackburn, "Vacuum Casting Goes Commercial", Advanced Materials and Processes. Feb. 1990,17-21.

2. J. Cook, "Vacuum Casting Opens up Design Possibilities", Canadian Machinery and Metal-wprlring, August, 1990,16.

3. S.P. Thomas, "CWC Textron Invests in the Promise of Advanced Vacuum Casting", Modem Casting. Dec. 1990,22-25.

4. J.L. Dion, R.K. Buhr and M. Sahoo, "Vacuum Pouring", F. Weinberg International Symposium on Solidification Processing. The Metallurgical Society of CIM, 29th Annual Conference of Metallurgists, Hamilton, Ontario, Aug. 1990, 90-103.

5. M. Sahoo and L. V. Whiting, "Foundry Characteristics of Sand-Cast Zn-Al Alloys", Trans. AFS. Vol. 92,1984, 861-870.

6. "Casting Copper-Base Alloys", Published by the American Foundrymen's Society, Des Plaines, IL, 1984, 88-89.

231

Effect of the addition of 1.3 wt% Fe on the microstructure and mechanical properties of a powder atomized Al-2.5 wt% Li alloy

J.H. Torres, G. Champier

Laboratoire de physique du solide, Ecole des Mines, Nancy, Cedex, France

P.H. Samuel Departement des sciences appliquees, Universite du Quebec a Chicoutimi, Chicoutimi, Quebec, Canada

Ph. Arcade Laboratoire de physique du solide, Ecole des Mines, Nancy, Cedex, France

ABSTRACT

In addition to 6'-Al3Li phase particles in binary Al-Li alloys, Fe leads to the formation of two other phase precipita-tions. These are A l3F e (needle-like) and A l6F e (globular) parti-cles. The former phase appears mainly in the grain interiors whereas the latter phase occurs at the grain boundaries. The volume fraction of A l3F e dispersiods is one tenth that of the A l3L i . The formation of this phase i.e. A l3F e is not affected by the presence of grain boundaries, and is enough to strengthen the PFZs. Thus, the presence of A l3F e dispersiods in Al-Li-Fe alloys leads to a relatively better combination of strength and ductility when compared to Al-Li binary materials.

KEY WORDS

Rapid solidification. Centrifugal atomization. Power metallurgy, Al-Li-Fe alloys, Al-Li alloys. Hot extrusion. Ageing.


INTRODUCTION

An effective wav of improving the strength and ductility of binary Al-Li alloys is to introduce incoherent dispersoid parti-cles and to refine the grain size using the rapid solidification technique. During the annealing of Al-Li binary allocs, soft continuous precipitate-free zones form along the grain boundaries leading to early cracks. The importance of the dispersoid particles is to change the mechanism of deformation from a heterogeneous into a more homogeneous one, which, in turn, improves the alloy ductility [11.

From the Al-Fe binary diagram \2~]r four intermetallic phases are expected to occur, depending on the Fe concentration. These are: •

Al6Fe orthorhombic, metastable A l9F e2 monoclinic, metastable Al3Fe monoclinic, stable A l 5F e 2 monoclinic or orthorombic

In the present alloy, Al-2.5 wt% Li-1.3 wt% Fe, Al3Fe appears to be the essential phase. It should also be noted that the solubility of Fe in molten Li is very low i.e. 0.014 at % at 930°C. However, the ternary Al-Li-Fe phase diagram is not yet known.

The present study deals briefly with the following aspects:

1) cetrifugal pulverisation 2) degassing 3) hot extrusion 4) heat treatment (solution treatment, quenching, ageing) 5) alloy characterization at each step 6) evaluation of mechanical properties 7) determination of the mechanism controlling the plastic

deformation

The technique of powder atomization by certrifugal pulveriza-tion has been developed by Pratt and Whitney Aircraft-United Technologies Corporation, U.S.A.. Publications concerning this technique are mostly proprietaries and patents T3-51.


Powder Pulverization

Two alloys were used in the present study:

a) L3, containing 2.5 wt% Li

b) L3F2, containing 1.3 wt% Fe, 2.5 wt% Li

Rapidly solidified powders were produced by the conventional certifugal atomization process. Ingots of 400g, were melted in graphite crucibles (50 mm internal diameter, 10 mm thickness) having an orifice of 1 mm diameter and coated with a thick layer of boron nitride refractory. The atomization chamber was eva-cuated and filled with helium up to 1 bar. The liquid metal was


ejected onto a rotary atomizer with helium overpressure.

The temperature of the liquid droplets that leave the atomi-zer should be above the liquidus temperature. It was experimen-tally found that:

1) Increasing the superheat temperature led to a reaction between the molten metal and the crucible walls. Therefore, the inital pour temperature for L3 alloy was chosen to be in the range 750-850°C, whereas for the L3F2 alloy it was close to 800°C.

2) On using a thermally isolated rotary atomizer, the liquid metal was in contact with a low calorific capacity spot.

3) Heating the atomizer at high temperature to reduce the thermal changes, required that the material of the atomizer should have a good strength at elevated temperature. For example, an atomizer of 44 mm diameter rotating at a velocity of 30 000 r.p.m., the acceleration at the periphery would be about 217 m

2/ s , which is close to 2.7 x 1 0

5, the gravity

acceleration.

Several sets of experiments were carried out using atomizers (44 mm diameter) with different forms. Fig. 1. Also, different materials for atomizer fabrication were tested, viz. stainless steel 18-10, refractory steel 20-25, aluminum graphite and gra-phite protected by a layer of boron nitride. For L3 alloy most experiments were done usinq the atomizer (a) in Fig. 1, made of graphite, rotating at 27,300 r.p.m. and heated at a temperature ranging between 720 and 810°C. For L3F2 alloy, the same atomizer configuration was used, made of graphite covered with boron nitride and heated up to 750°C.

Powders were collected in aluminum cans and degassed under vacuum (pressure lower than 2 x 1 0 ~

6 torr) for a period of about

50 h at 200°C. At the end of degassing, the cans were sealed tightly. The cans were preheated at 420°C prior to hot extrusion. The diameter of the extruded rods was about 6 mm. The relation between the final length of these rods and the initial container height is given by:

Hf =

Hc

C Dc e

+ <?7 -

1 »

Dc i

1 I

RL

length of the extruded rods height of the powder containers (100 mm) container external diameter (43 mm) density of loose powders (.1.5 g/cm

3)

density of extruded rods (2.7 g/cm

3)

container internal diameter (33 mm) diameter of extruded rods (6 mm)

where

H

D

C

I * ' f

XT . fe


Microstructure and Properties Evaluation

The microstructure of powder particles (with sizes ranging between 25 and 200 pm) as well as extruded materials were examined by light microscopy, scanning electron microscopy (SEM) and transmission electron microscopy (TEM operating at 200 kV) . The precipitates were identified by X-ray diffraction using a C K a source. °

Tensile tests were carried out with an Instron type tensile machine. The gauge length of the specimen was 18 mm and the diameter was 4 mm. The strain rate was 2.5 x 10

4 s

l.


Microstructure Characterization

Loose Powders

The powders are a mixture of spherical and quasi-spherical particles for diameters less than 100 ym and of elongated shape for larger particles. Observations made by SEM on the outer surfaces of the particles of L3 loose powders reveal a well-defined dendritic structure as shown in Fig. 2a. Figure 2b is a typical microstructure of L3F2 powders. The absence of a growth direction makes us believe that the microstructure is mainly comprising of cells or grains. The sides of these grains are not always straight.

Figure 3a is an electron microgarph obtained from L3 powders showing a dendritic-like structure with an arm spacing ~ 0.3 pm in colonies of 2 to 3 ym diameter. Selected area diffraction pat-terns and dark-field microscopy identified the dendritic-like phase as a-Al solid solution with a common orientation throughout each colony. The interdendritic network phase could be indexed on the basis of Al3Li (6

1) phase.

The TEM micrograph in Fig. 3b is produced from the powder of L3F2 alloy. Near the centre of each grain, a coarse particle of Al6Fe (as could be identified from analyzing the corresponding electron diffraction pattern) can be seen. The nucleation and growth of primary AleFe is followed by the subsequent nucleation and growth of primary aluminum dendrite arms. The primary alu-minum is found to grow radially outward from the central inter-metallic particle. The secondary dendrite arms are delineated by the precipitation of both Al3Li and Al3Fe particles.

Jones [6] has reported a similar observation in a rapidly solidified Al-11 wt% Fe hypereutectic alloy made by melt spinning. He refers to this microstructure as zone B, and it is characteri-zed by coarse intermetallics surronding primary aluminum dendri-tes. It should be noted that the intermetallics appear to be located at the centres of local solidification fronts. This would suggest that the intermetallic was the first solid to nucleate and grow, followed by primary aluminum and not the reverse.


Extruded Powders

Figures 4a and 4b represent longitudinal sections. Some of the grains are elongated in the extrusion direction, probably due to the drop in the extrusion temperature towards the end of deformation. The microstructure of L3F2 alloy. Fig. 4b, shows the presence of many dark particles. These particles are absent in Fig. 4a obtained from L3 alloy. These spots were still persisting after the solid solution heat treatment (540°C for 30 min. follo-wed by water quenching). They are believed to be due to Al3Fe and Al6Fe precipitate particles.

Figure 5a is a TEM micrograph obtained from L3 as-extruded powders (transverse section). The microstructure is comprising of fine equiaxed grains (2-5 ym). Coarse precipitates due to 6 (AlLi) phase at the grain boundaries as well as within the grains are viewed. Figure 5b represents the microstructure of as-extru-ded L3F2 powders. The microstructure is microcrystalline with grain sizes in the range 1 to 3 ym. Dense precipitation due Al3Fe and Al6Fe phase particles is marked. It is believed that these are the black spots seen in Figure 4b.

The TEM micrograph shown in Fig. 6a corresponds to L3 alloy after solid solution treatment. The microstructure is characte-rized by the absence of AlLi(6) phase particle precipitates. The grain size varied between 2 and 5 ym. Homogeneous precipitation of 5' (Al3Li) phase particles with diameters up to 4 ym was first observed immediately after quenchinq and refriqerating in liquid nitrogen. Fig. 6b. The presence of these precipitates was not affected by the grain boundary.

When an alloy containing sufficient Li is quenched from the a-single phase field, decomposition of the supersaturated solid solution takes place by homogeneous precipitation of the ordered 6' phase. Nozato and Nakai [7] have deduced from thermal analysis that 6' formation can not be supressed by ice-water quenching in alloys containind more than 1.7% Li. The 6' (Al3Li) phase has an L I 2 type superlattice structure and a spherical shape possessing a cube/cube orientation with respect to the matrix.

Figures 7a and 7b are produced from L3F2 alloy after solution heat treatment. Two types of precipitates could be identified:

a) coarse spherical particles (> 100 nm) due to Al6Fe phase (open triangle) placed at the grain boundaries as well as at their interiors;

b) Needle-like particles due to Al3Fe phase (closed triangle) distributed mainly intragranularly.

The solution heat treated materials were aged artificially at 200°C (under helium atmosphere) for different ageing times. The presence of Fe does not seem to have any significant effect on the kinetics of either the precipitation-free zone (PFZ) formation or 6* (Al3Li) phase particle coarsening.

Figure 8 shows the occurence of a wide PFZ (280 nm) in the vicinity of the grain boundary when the L3 alloy was aged at 200°C


for 16 h. The average 6

1 particle diameter is about 40 nm.

Figures 9a and 9b are bright and dark field micrographs obtained from the L3F2 alloy treated similarly. These figures emphasize the role of rapid solidification in increasing the extent of solid solubility. An important fraction of iron was in solid solution after ice-water quenching and it decomposed uniformly throughout the grain after ageing. The process is expected to enhance strengthening of the soft PFZs observed in L3 alloy. Fig. 8.

Tensile Properties

Tensile test specimens from both alloys were aged at 200 and 220°C for 16, 20 and 24 h at each temperature. In each case, four specimens were tested and the average values are listed in Table 2. Typical load-displacement curves produced from L3F2 allov are presented in Fig. 10.

During the solution heat treatment process (540°C/30 min) , all the 6-AlLi particles as well as a fraction of Al3Fe and Al6Fe particles were dissolved in the matrix. This process, also, was associated with annealing out of the internal stresses.

Age hardening at 200°C took place after an incubation period of about 14 h and reached its maximum after 20 h. Increasing the ageing temperature results in accelerating the time required to reach maximum strength.

As expected, the presence of Al3Fe dispersoid particles leads to a noticeable improvement in both the strength and ductility of L3F2 alloy as compared to L3 material (see Table 1 ) . Considering an alloy containing in weight pet: 96.2 % Al, 2.5 % Li and 1.3 % Fe, the corresponding atomic composition is: 90.6 % Al, 9.16 % Li and 0.6 % Fe. Assuming that all lithium will be in the form of 6

1-AlLi (density ~ 2.33 g / c m

3) , and all iron will be in the form

of Al3Fe phase (density ~ 3.45 g / c m

3) , under these conditions, the

volume fraction of 6' will be close to 36.5 % and that of Al 3Fe will be about 3.65 % i.e. one tenth. It was found that such volume fraction of Al3Fe is enough to balance the loss of strength of the PFZs, due to migration of 6' particles towards the grain centre.

It is interesting to compare our results with those previously published for alloys containing the same elements, viz. Fe, Co and Ni. These are tabulated in Table 2. It is seen that fo Al-Li binary alloys, materials made by powder metallurgy (PM) possess somewhat higher strength parameters compared to those obtained from ingots (IM) made by classical casting. The observed low ductilitv for PM alloys is explicable in terms of the remarkable difference in the strength level between the PFZs and the rest of the grain. It should noted that for these alloys, when the material is aged at 200°C, the volume fraction of the PFZs is approximatelv 24% for 16 h and 50% for 100 h ageing time TCI. These PFZs lead to intergranular brittle fracture. Thus, the presence of Al3Fe dispersoids greatlv enhances transgranular ductile fracture through strengthening of the PFZs [9].


CONCLUSIONS

1. By controlling

a) the initial temperature of the molten alloy,

b) the atomizer material, temperature, speed and configuration, and

c) the cooling media,

spherical and quasi-spherical powders with particle size in the range 25 - 200 ym, suitable for hot extrusion can be produced on a large scale (in terms of Kgs).

2. For Al-2.5 wt% Li- 1.3 wt% Fe, in addition to 6'-Al3Li phase precipitates, Al3Fe (needle-like) and Al^Fe (globular) phase precipitates are also present. The former phase appears mainly in the grain interiors whereas the latter phase occurs at the grain boundaries.

3. The volume fraction of Al3Fe dispersoids is one tenth that of the A l3L i . The formation of this phase (Al3Fe) is not affected by the presence of the grain boundaries, and is enough to strengthen the PFZs.

4. The presence of Fe and hence Al3Fe particles leads relatively to a better combination of strength and ductility when compared to Al-Li binary alloys.

Acknowledgments

One of the authors (FHS) would like to acknowledge with gratitude the financial support received from the Natural Sciences and Engineering Reseach Council of Canada, the Fondation Sagamie de 1*University du Quebec a Chicoutimi and the Socie"te" d

1 Electro-

lyse et de Chimie Alcan (SECAL).


REFERENCES

1. T.H. Sanders Jr., E.A. Ludwiezak and R.R. Sawtell, Mater. Sci. Eng., Vol. 43, 1980, 247.

2. M. Hansen and K. Anderko, Constitution of Binary Alloys, McGraw-Hill Co. Inc., New York, 1958; and also supplement, 1965.

3. R.G. Bourdeau, "Spin cup means for the production of metal powders", U.S. Patent 4217082, August 12, 1980.

4. J.A. King, "Apparatus for making metal powders", U.S. Patent 4025249, May 4, 1977, and U.S. Patent 4053264, October 11, 1977.

5. P.R. Holiday and R.J. Patterson, "Apparatus for producing metal powder", U.S. Patent 4078873, March 14, 1978.

6. H. Jones, Mater Sci. Eng., Vol. 5, 1970, 1.

7. R. Nozato and G.R. Nakai, Trans. J.I.M., Vol. 18, 1977, 679.

8. F.H. Samuel, Ibid, Vol. 21, 1986, 3097.

9. F.H. Samuel and G. Champier, Mater Sci., Vol. 22, 1987, 3851.

10. K.K. Sankaran, S.M.L. Sastry and J.E. O'Neal, Proc. Conf. on Aluminum-Lithium I, Stone Mountain, 19-21 May 1980, TMS-AIME 1982, p. 189.

11. T.H. Sanders Jr., E.A. Ludwiezak and R.R. Sawtell, Mater. Sci. Eng., Vol. 43, 1980, 247.

12. B. Nobel, S.J. Harris and K. Dinsdale, J. Mater Sci. , Vol. 17, 1982, 461.

13. S.M.L. Sastry and J.E. O'Neal, Proc. Conf. on Aluminum-Lithium II, Monterey, April 12-19, 1983, TMS-AIME 1984, p. 79.

14. I.G. Palmer, R.E. Lewis and D.D. Croaks, Proc. Conf. on Rapid Solidification Processing: Principles and Technologies II, Reston 23-26 May 1980, Claitor 1980, p. 347.


TABLE 1 - Tensile properties of the L3 and L3F2 extruded powders

L3 L3F2

Treatment a0.2 MPa

am MPa

El %

a0.2 MPa

am MPa

El %

A) 540°C/30 min/WQ 72 196 19 75 178 19

A + 200°C/16h 319 395 4.5 372 440 5.5

A + 200°C/20h 312 394 5 366 432 6.2

A + 200°C/24h - - - 360 430 7.5

A + 220°C/16h 281 348 6.0 310 387 8.0

A + 220°C/20h 254 318 7.0 311 386 9.0

A + 220°C/24h - - - 266 354 10.5

Table II - Tensile Properties differently at 200

of some Al-Li-°C.

-X alloys. aged

Alloy Composition a0. 2 MPa

am MPa

El %

Ref.

Al-2.5 Li o 312 394 5 P.W.

o 296 396 5.5 10

256 385 8.0 11

264 376 5.0 11

Al-1.9 Li (aO.l) * 177 213 2.0 12

Al-3 Li o 295 - 5.0 13

Al-2.7 Li - 1.3 Fe o 366 430 6.2 P.W.

Al-2.95 Li - 0.64 Co o 296 409 8.3 10

Al-3.8 Li - 0.48 Fe - 4.3 Co o 391 468 4.5 10

Al-356 Li - 0.36 Fe - 0.48 Ni o 378 464 4.8 14

o Rapidly solidified alloys * Classically solidified alloys P.W. Present work

Table II - Tensile Properties of some Al-Li-X alloys, aged differently at 200°C.


H ( 0 ) n N ( b ) A r ( C ) x [

Fig. 1 Different forms of the rotary atomizer used in the present work: a) flat bottom with straight edges, b) flat bottom with inclined edges, c) spherical bottom.

Fig. 2 SEM micrographs of surface of as-quenched powders of: a) L3, b) L3F2 alloys.

240


Fig. 4 Optical micrographs of as-hot extruded powders: a) L3, b) L3F2 alloys.

241

Fig. 3 TEM micrographs of as-quenched powders of: a) L3, b) L3F2 alloys.


Fig. 6 TEM micrographs of hot extruded L3 alloys following solu-tion heat treatment: a) bright field, b) dark field using a <S'-spot diffracted beam.

Fig. 5 TEM micrographs of as-hot extruded powders of: a) L3, b) L3F2 alloys.

Fig. 8 TEM micrographs of L3 alloy aged at 200°C for 20 h.

Fig. 7 Two TEM micrographs of hot extruded L3F2 alloy following solution heat treatment, A Al6Fe, A Al3Fe phase precipita-tes.


2/im

0,6fim


Fig. 10 Load - displacement curves produced from L3F2 alloy following various heat treatments.

Fig. 9 TEM micrographs of L3F2 alloy aged at 200°C for 20 h: a) bright field, b) dark field using a 61-spot diffracted beam.

A*-Extrud»d alloy

«ft»r Solution Haat Traat»»nt

Alloys aged at 220°C

Alloys agvd at 200°C

245

Permeability of refractory coating in evaporative foam pattern process

C. Ravindran, B. Jue, J. Karpynczyk Centre for Near-net-shape Manufacturing, Ryerson Polytechnical Institute, Toronto, Ontario, Canada

ABSTRACT

Permeability of refractory coating in evaporative foam casting process is of extreme importance in that it affects the metal-foam interactions and ultimately the quality of the castings,

A novel method for determining coating permeability was developed and verified. Effects of coating density, coating thickness and casting temperature on coating permeability are discussed. These data are correlated with microscopic observations of suspended particles and porosity effects.

Keywords

Refractory coating wafer, permeability, flowrate, porosity, evaporative foam pattern casting.

INTRODUCTION

The refractory coating which is applied to the surface of the completed pattern assembly cluster plays an important role in the evaporative foam pattern casting (EFPC) process (1). The coating is used not only to produce a smooth, acceptable surface finish to the castings, but also to contribute to the rigidity of the pattern assembly cluster. This rigidity aids in minimizing distortion through bending and twisting of the pattern shape during sand compaction.

The coating must also control the release of gases and liquids from the decomposing foam pattern into the unbonded molding sand. The mechanism of mold filling as a result of gasification of the evaporative foam pattern (EFP) has been fully described and demonstrated by Butler and Pope (2).

Dr. C. Ravindran and Dr. B. Jue are Professors and Mr. J. Karpynczyk is Senior Technologist with the Department of Mechanical Engineering, Faculty of Engineering and Applied Science, Ryerson Polytechnical Institute, Toronto, Canada M5B 2K3


The permeability of a coating may be decreased by reducing the size of the refractory particles and other powder constituents used (3) • Smaller particles are known to reduce the voids between the grains, thus resulting in smoother casting surface as well. If the particle sizes are left unchanged, the permeability of the refractory coating can be decreased by increasing the deposit thickness of dried coating.

It has to be recognized that the coating permeability is only a part of the EFPC system permeability (3) . The system permeability is the net effect of the coating and unbonded sand permeabilities. Nonetheless, refractory coatings with a wide range of dry permeabilities are available and therefore a reasonable understanding of changes to permeability levels and the corresponding porosity levels is of immense value for the researcher and the industry.

Several test methods for permeability of refractory coating use an electric permeability tester, often with a screen to support the coating (4,5). Some other methods have used an inert gas such as argon, helium or nitrogen to relate flowrates to permeability over a range of temperatures (5).

Goria et al (6) developed a method to measure the permeability of coating by using coating flakes taken directly from special foam specimens or actual patterns.

In this paper, a method for preparation of coating wafers of controlled thickness is described. The wafers are then subjected to permeability tests in a dynamic forced air environment with heating and cooling as in actual production operation.


Preparation of Refractory Coating Mix

The refractory coating slurry was mixed in a 600 litre tank with a recirculating pump and an auxiliary propeller-mixer. A 1000ml beaker sample was obtained from this tank. After allowing a settling time of 1 minute, a foundry hydrometer (Dietert No.625) was inserted into the beaker. A Baume reading was taken 30 seconds thereafter.

A range of specific gravities of different types of coating was achieved with controlled additions of water prior to mixing.


Preparation of Coating Wafers

Predetermined weight of refractory coating was obtained by careful transfer of slurry to a preweighed standard petri dish using a plastic syringe. In effect, for a refractory coating of known specific gravity (Baume), the thickness was controlled by weight. The coating wafers were 80mm in diameter and the thicknesses ranged from 0.40mm to 1.25mm and were controlled to + 0.05mm.

Permeability Measurement

A permeability apparatus was designed (Figure 1) . A 45cm long stainless steel (Type 304) tube with a wall thickness of 10mm and bore opening of 45mm was fabricated. The vacuum was measured with a U-tube manometer. The steel tube was connected to a flowmeter, and then to a pressure control valve and a vacuum pump.


Flowmeter

T h e r m o c o u p l e

% / — Test Wafer

Pressure Control Valve

> Vacuum Pump

U-Tube Manometer

Fig. 1. (a) Schematic of permeability apparatus

Fig. 1. (b) Permeability apparatus

248


The coating wafer (approximately 75mm in diameter) was firmly attached to the open end of the steel tube with a high temperature cement (stable up to 800°C) and tested for any leakage through the cement.

The entire apparatus was mounted on a mobile platform so that the coating wafer end of the steel tube could be positioned in the constant temperature zone of a furnace. The temperature of the wafer coating (+ 2°C) was monitored up to 750 C with a type J thermocouple positioned in close proximity to the wafer. The flowrate of air (i.e., permeability) through the coating wafer was monitored with a flowmeter as air was drawn at constant vacuum.

Optical Microscopy

Optical microscopic examination of all the flowrate test samples was carried out after testing. Also, several wafer samples were heated in air at atmospheric pressure to various temperatures up to 750°C and subjected to optical microscopy. Both transmitted and reflected lights were used for these examinations, and it was determined that dark field microscopy with reflected light provided the most favourable condition for observation of the suspended particles and the pores. These qualitative observations were then correlated with the permeability data.

RESULTS

Permeability Tests

All the flowrate tests were carried out at a constant vacuum of 100mm water. Small changes in vacuum, if any, due to changes in porosity of the coating wafer during the experiments as indicated by the manometer were continuously corrected by readjustment of the vacuum control valve. Permeability values for various thicknesses and specific gravities at various temperatures are shown in Table 1.


Table 1. Permeability tests

Coating Wafer

~ H Specific Gravity Baume

Thickness mm

Flowrate at lOOmm water cc/min

Temperature, °C

25 300 500 600 700

Al

i )

62 i 0.625 111.5 53.4 50.0 - -A5 62 j 0.625 104.7 54.5 53.4 - -A12 62 » 1.000

1 78.0 41.1 39.7 - -

A15 62 1 1 1.000

78.0 39.7 39.7 - -C4 46 1 0.625 109.2 50.0 50.0 50.0 -C7 | 46 \ 0.625

t 109.2 50.0 50.0 50.0 -

Dl j 62 I j 0.625

111.5 51.1 44.5 43.7 -D7 \ 62 1 0.625 111.5 56.7 47.8 43.7 37.1

D8 62 I 0.625 1

113.7 59.1 47.8 - -D9

\ ; 62

I ! 0.425

159.0 73.0 69.6 67.9 54.5

D14 1 62 | 0.425 153.0 74.7 59.1 - -F2

i ! 47 j

| 0.425 252.5 118.6 : 95.7 86.3 69.6

G9 ! 47

1 | 0.625 i

164.3 89.6 ; 66.1 60.1 57.4

Figures 2 and 3 show typical permeability curves for two different stock coatings. In general, at constant vacuum the permeability of the coating wafer was seen to decrease with increase in temperature. In some instances, the permeability reached a minimum and remained constant thereafter. In some other instances, the permeability continued to decrease up to the maximum experimental temperature of 750°C, or even showed a slight increase at higher temperatures.

250


FLOW RATE, C C / M IN 120

100

1 A1 0.625mm thick: 62Be A12 1.000mm thick: 62Be C4 0.625mm thick: 46Be

25 5 0 100 150 2 0 0 250 3 0 0 3 5 0 4 0 0 4 5 0 5 0 0 5 5 0 6 0 0 6 5 0 7 0 0 750 8 0 0 T E M P E R A T U R E, DEO C

Fig. 2. Typical permeability curves

3 0 0

250

2 0 0

150

FLOW RATE. C C ' M IN

100

F 2 0 . 4 2 5 m m th ick 4 ? B o 3 9 0 6 2 5 m m t h i c k 4 7 8 e D 9 0 4 2 5 m m t h i c k 6 2 8 e D 7 0 . 6 2 5 m m t h i c k 6 2 B t

G9 * D9

25 5 0 100 150 2 0 0 250 3 0 0 3 5 0 4 0 0 4 5 0 5 0 0 550 6 0 0 6 5 0 7 0 0 750 8 0 0 T E M P E R A T U R E. DEO C

Fig. 3. Typical permeability curves

251

50


Coating types A (62 Baume) and C (46 Baume) were from the same stock. As is evident, with increasing wafer thickness permeability decreased at all temperatures. However, for the same thickness, as the specific gravity decreased, the permeability values remained approximately the same (Fig. 2).

Coatings from a second stock (Fig. 3) show slightly different permeability behaviour. Coating Type D also showed increased permeability as the thickness was decreased. However, with this refractory coating, reduction in specific gravity resulted in increased permeability for the same thickness.

It is also of interest to note that the rate of decrease in permeability with temperature is higher for one stock (Fig. 3) than for another stock (Fig. 2).

Efforts at subjecting a third stock material. Type H, to permeability tests were unsuccessful as the flowrates of these coating wafers exceeded the capability of the equipment. However, these obviously high permeability wafers were subjected to microscopic examination.

Optical Microscopy

The coating wafers heated in air at atmospheric pressure to various temperatures were subjected to optical microscopy (Table 2) . Pore size and number appear to vary with stock. For example. Type H samples exhibited roundish particles, and large number and size of pores compared to Types D and A (Figs. 4 and 5) . The latter two types with flakier particles had less number of pores of smaller size. The similarity in pore structures of Types A and D is reflected in the permeability curves for samples of same thickness and specific gravity.


Table 2. Optical microscopy of coating wafers

Coating Wafer

Specific Gravity Baume

Thickness mm

Max Temp. °C

Observation

H8 84 0.600 25 Roundish particles, large number of

deep pores

A8 62 0.650 25 Flake-like particles,

less pores of smaller size relative to

H8

D5 62 0.575 25 Flake-like particles, less pores of smaller size relative to

H8

H2 84 0.575 200 Less pores than H8, some evidence

of agglomeration of

particles

A4 62 0.650 200 Smaller pores than H2 but seemingly

more of them, less agglomeration than

H2

D2 62 0.600 200 Some fusion, less porous than H2

H4 84 0.525 500 Platelets of fused masses, large pores

A8 62 0.575 500 Less, smaller pores than H4 and A4

D5 62 0.700 500 Similar to A8

H7 84 0.650 700 Large pores, large fused masses

A7 62 0.575 700 Less pores than A4, agglomeration

of flakes

D4 62 0.600 700 Some evidence of agglomeration similar to A7


(a) Type H8 (a) Type H7

(b) Type A8 (b) Type A7

(c) Type D5 Fig. 4. Optical micrographs of unheated wafers

(c) Type D4 Fig. 5. Optical micrographs of wafers heated to 700°C


Effect of temperature on pore structure is not clearly discerned from optical microscopy. There is evidence however, of some aggregation of particles at higher temperatures and a reduction in the number and size of pores (Fig. 5).

DISCUSSION

Results from flowrate tests all show that regardless of thickness or specific gravity, permeability decreases with increase in temperature to some limit. This suggests occlusion of the through holes in the coating wafer. Reduction in permeability may also be attributed to some form of agglomeration of particles resulting in more tortuous paths for passage of air or gas. This appears to be substantiated by micrographs of thermally treated wafers (Fig. 5) in agreement with the earlier observations of Goria et al (6).

The fact that H-type was observed to be of high permeability from the aforementioned unsuccessful test suggests that porosity is more a function of initial composition and particle size or shape distribution than dilution or Baume effect.

The observed increase in permeability at higher temperatures in some instances may be attributed to the onset of cracking in the wafer. Attempts were made to remeasure the permeability of wafers during cooling from 750°C. Invariably cracking of the wafer and increased permeability were experienced before reaching room temperature. However, before the onset of cracking, it appeared that permeability has a reversible tendency. Further experimental work is in progress to gain a further understanding of these changes in permeability. In addition, we are devising experiments in an attempt to observe in-situ the effect of temperature on size, shape and distribution of pores and particles under forced air flow conditions.

CONCLUSIONS

A novel method for determination of coating wafer permeability with forced air was developed and tested. This will assist the foundrymen to gain a further understanding of the permeability and porosity effects and hence to choose a coating for specific operating conditions.

Controlled coating thicknesses were obtained to result in repeatable experimental data.


Changes in permeability are explained through microscopic observations. Occlusion of pores, agglomeration of particles and layering or flaking effects appear to affect permeability. Initial composition and particle morphology of the refractory coating appear to more than offset changes to permeability due to reduced specific gravity.

REFERENCES

1. R.B. Ballman, AFS Transactions, 77, 465 (1980).

2. R.D. Butler and R.J. Pope, The British Foundryman, 4, 178 (1964) .

3. O.A. Martinez, AFS Transactions, 21, 241 (1990).

4. AFS Publication: Prepared Mold and Core Coatings Manual (1982) .

5. AFS Division 11-D Subcommittee report, February 1991.

6. C.A. Goria, G. Serramoglia, G. Caironi and G. Tosi, AFS Transactions, 101, 589 (1986).

257

An experimental study of twin-roll casting

D.V. Edmonds, J.D. Hunt, D.J. Monaghan, X. Yang, M. Yun Department of Materials, University of Oxford, Oxford, U.K.

D.J. Browne, R. Cook, P.M. Thomas Davy McKee (Poole) Ltd., Wallisdown Road, Poole, U.K.

ABSTRACT

The twin roll casting of strip and sheet is an established commercial process for aluminium. However, it is limited in terms of casting speed, product thickness and alloy applications. Commercial twin roll sheet casters do not normally operate at casting speeds greater than approximately 1.5 m/min and sheet thicknesses less than 6 mm. Commercial application of twin roll casters has also been restricted to aluminium, and moreover, limited to common grade alloys for thin gauge products. This paper will report progress being made with a project evaluating the potential to widen the commercial application of twin roll casting.

In excess of 400 experimental casts have been carried out on an experimental twin-roll caster. Good quality material has been produced at casting speeds up to 15 m/min and thickness gauges down to 1.25 mm, and significant productivity increases have been achieved. A number of aluminium alloy systems have been successfully cast ranging from ultra-high purity aluminium to alloys containing significant amounts of alloying elements. Other non-ferrous alloy systems of commercial interest have also been cast successfully. The results of the programme so far have been encouraging in terms of demonstrating the potential for the further exploitation of the twin roll casting process.

KEY WORDS

Twin roll casting, Experimental caster, High productivity casting, Aluminium alloys, Segregation, Microstructures, Alloy development.


INTRODUCTION

Twin roll casting is an established commercial process for the production of sheet and strip products. The earliest patents date back to the nineteenth century(l) but it was not until the middle of this century that the process was exploited commercially. Since this time something like 200 machines have been built and put into commercial operation.

The earliest machines were built with 600 mm diameter rolls and configured so that the cast sheet emerged from the caster travelling vertically upwards. In later years, the diameter of the caster rolls was increased - typically to about 1000 mm diameter - and also the attitude of the caster was changed so that the cast strip emerged from the caster either horizontally or at some small angle to the horizontal. Apart from the above changes, there has been very little development of the process.

The equipment has remained simple but in spite of this the material produced from twin roll machines is suitable for subsequent state-of-the-art downstream processing and many facilities regularly produce finished foil products at gauges less than 7 microns.

Compared with the other major casting processes, twin roll casting is a slow, low productivity process, the production rate for a 1000 mm diameter roll caster being approximately one tenth that of a belt or block caster producing the same alloy. Moreover twin roll casting has been limited, at least commercially, to a narrow range of aluminium alloys. The foilstock alloys - lxxx and 8xxx - are regularly twin roll cast in large quantities as are AA3003/AA3103 and some dilute 5xxx series alloys. Other aluminium alloys are cast in more limited amounts, but the majority of material produced by this method is from the narrow freezing range alloys mentioned.

Further exploitation of twin roll casting has been plagued by characteristic defects such as heat lines, sticking and the presence of centre line (channel) segregates(2,3).

The product from commercial casters is normally 6 - 1 0 mm thick and, although the earliest machines were usually narrow, the most recent machines are capable of producing sheet approaching 2 metres wide. The surface quality of as cast sheet is normally good, as are the gauge and profile tolerances.

THE PROCESS

A twin roll caster is nothing more than a slow, 2-high rolling mill where the feedstock is molten aluminium and the end product cast aluminium sheet. If the process is compared with the other major casting techniques such as DC or thick strip casting, it becomes apparent that because of the integral rolling stage the mode of solidification is fundamentally different.


Although there are differences in the detail of, for example, DC casting and thick strip casting as exemplified by the Hazelett and Lauener (Alusuisse Caster II) processes, particularly in terms of the available mould length, the mode of solidification is virtually identical. In each case, solidification begins when the molten metal comes into contact with the mould. Once a shell has solidified there is a tendency for this thin shell to contract away from the mould and at this point there is a dramatic reduction in the rate of heat transfer into the mould.

In the case of DC casting secondary cooling is applied to the solidifying shell once it leaves the mould cavity. Because of the significantly longer effective mould lengths employed in the thick strip casters much more of the heat is removed through the mould wall and secondary cooling is not used. The DC and thick strip casting processes can be classified as 'solidification only' techniques and it is the loss of contact with the mould after solidification has started that is responsible for the surface and sub-surface defects that are characteristic of these processes.

In twin roll casting molten metal is fed into the bite of a pair of counter-rotating rolls and solidification starts when the molten metal contacts the rolls. Due to the progressively reducing dimension of the roll bite there is little possibility of the solidifying shell contracting away from the 'mould'. It is essential that solidification is completed before the plane of the centre line of the casting rolls and, as a result, the as - cast sheet is subjected to a hot rolling operation before leaving the rolls. This gives the opportunity for the more rapid removal of heat as well as ensuring good surface quality and dimensional accuracy.

Bagshaw et al(3,4) modelled the twin roll casting process for a series of dilute alloys and defined three regimes of heat transfer in the roll bite area. On initial contact with the caster rolls an intermediate value of heat transfer coefficient operates. At some stage the solidified shell starts to 'buckle' and a second, lower heat transfer coefficient applies until, as a result of the hot rolling operation, a third, much higher, heat transfer regime comes into effect.

Because of the presence of the integral hot rolling operation, twin roll casting can be described as a 'solidification/deformation' technique and as such is unique amongst the casting processes used for the production of wrought products. The high loads generated during the twin roll casting operation are responsible for the good surface quality and good gauge and profile tolerances that characterise the process. The loads generated during twin roll casting are significant and even for a foilstock alloy such as AA1145 cast at a width of 1500 mm on a 675 mm roll diameter machine, loads in excess of 600 tonnes have been measured(5).

Although other metals such as zinc have been cast via the twin roll process only aluminium is processed in any meaningful quantity at present. There is, however, considerable interest in the twin roll casting of steel and several prototype machines exist.


PRESENT INVESTIGATION

It is believed that, providing the conventional limitations of the twin roll casting process can be overcome, there exists considerable potential for further development of what is an inherently good casting process. Consequently, a collaborative research and development programme has been established between Davy McKee and Oxford University with initially five main areas of activity:

Mathematical Modelling

Experimental Caster

Casting Trials

Cast Product Evaluation

Alloy Development Programme

L Mathematical Modelling

Two numerical models have been developed(6), the first being time dependent (transient) and the second being steady state. Both models use the finite difference control volume method to solve the equations of heat flow and use the Scheil equation(7) and similar assumptions to those of Bagshaw et al(3,4) to describe the evolution of latent heat and the different regimes of heat transfer.

The transient model describes how the caster behaves under start up conditions and how steady state is approached as a function of time. This has allowed prediction of start up procedures and casting parameters for a wide range of materials. The steady state model predicts the behaviour of the caster using assumed values of heat transfer coefficient along the contact length. A variant of the steady state model uses measured values of the roll sub-surface temperature (obtained from an instrumented caster roll) to predict the variation of heat transfer coefficient as a function of position in the roll bite area. The steady state models have been used to predict the onset of defects such as heat lines in narrow freezing range alloys and the effect of machine and operating parameters on caster performance.

The steady state model has also been used to predict the effect of strip thickness on caster productivity and an example of this is given in Figure 1. From this it can be seen that as the thickness of the cast strip is reduced there is a significant increase in caster productivity. Twin roll casting has traditionally been regarded as a 'constant mass flow process' and this is seen to be the case for the normal range of cast thicknesses (6 - 10 mm). By reducing the cast thickness to less than 2 mm the caster productivity is approximately doubled.


10

£ e

CO CO LU

z (J X h-Q l

rr H CO

Roll Diameter 675mm

_ J

Alloy AA1145

' 1

0.5 1 1.5 2 2.5

PRODUCTIVITY (t/h/m width)

Figure 1. Prediction of the effect of strip thickness on caster productivity using the steady state model (6)

In addition to this significant gain in productivity it is to be expected that, as a result of the increased solidification rates, the cast product will have a refined microstructure and conventional solid solubilities could be extended.

2. Experimental Caster

2.1 Machine Design

During the conceptual stage of design of the experimental caster it was decided to incorporate as much versatility as possible. For this reason the machine was designed to operate at two speed/torque ranges and the frame was made adjustable so that the caster could operate


in a range of attitudes. For simplicity and to reduce costs the machine was designed without a conventional housing, the chocks being held together using tie rods. The roll gap was set using a series of shims positioned between the chocks and changed by adjusting the nuts on the tie rods. The caster rolls are internally water cooled but provision has been made for external roll cooling. Table I summarises the design features of the caster and Figure 2 is a photograph of the caster built to this original specification.

Figure 2. The experimental caster


Table I - Specification of the Experimental Caster

Roll Diameter (maximum) Roll Diameter (minimum) Roll Facewidth Strip Width (maximum) Casting Speed:

600 mm 240 mm 300 mm 250 mm

Low Speed High Speed

0-10 m/min 0-150 m/min

Torque (maximum) Low Speed High Speed

17.5 kNm 1.2 kNm 75 tonnes Roll Separating Force (maximum)

Instrumentation is a key feature of the caster and during casting a large number of parameters are recorded using a computer based data logger.

For the initial trials the caster was configured so that the cast strip emerged horizontally from the caster and this is the only configuration examined to date. A number of machine modifications have been made, however, and the most significant of these are as follows:

a. The addition of a top roll drive. The addition of an independent drive train to the top roll overcomes the operational problems of having only the bottom roll driven and eliminates the odd friction effects this causes.

b. The addition of hydraulic gap control. The closed loop hydraulic gap control system allows the roll gap to be set and controlled accurately before and during casting. A tilt (steer) control allows an asymmetrical roll gap to be set.

c. The addition of a tip actuating mechanism that adjusts the set back as a function of roll gap. By linking the tip table to the hydraulic gap control system it is possible to ensure that the tip is maintained in the same relationship to the caster rolls for any roll gap. This means that the caster can be started using one set of conditions and can then be adjusted to quite different steady state conditions. This is particularly important for high speed/thin strip casting.

d. Increasing the top speed of the low range to 15 m/min. This increased range of speeds allows a wider range of cast gauges and end products to be produced.

e. The addition of a coiler. The addition of the coiler allows strip tension to be used as a control variable and also conveniently collects the cast material.


2.2 Machine Operation

Figure 3 shows the caster during operation. The charge is melted in a clay graphite crucible housed in an electric resistance furnace. The capacity of the crucible is nominally 25 kg of aluminium. Once molten, the charge is degassed in situ using a nitrogen lance and prior to casting the surface of the melt is skimmed after treatment with a proprietary flux. Once the metal is at the correct temperature it is transferred to a pouring gantry. To minimise heat losses during pouring, the crucible is encased in a refractory box lined with an insulating blanket. Using this technique the temperature drop during casting is limited to typically 2°C/minute. For casts where the material is to be grain refined, Tibor rod is added to the crucible after it has been removed from the furnace.

Figure 3. The modified caster in operation.


The refractory feed tips are pre-heated prior to use and are removed from the furnace and installed in the machine when the molten metal charge reaches a predetermined temperature. For operational convenience the feed tips used in the experimental programme incorporate an integral head box but the refractory side dams that would normally be part of the tip are permanently fixed to the tip table and are used for location purposes. Metallostatic head is controlled by the use of overflows in the head box.

To start the cast an appropriately sized starter bar is used to block the roll gap and metal is poured into the feed tip. Once the cast sheet is through the rolls it passes over a single pass line roll before being fed into the coiler mandrel. The speed of the coiler is linked to the caster roll speed and tension control is possible.

Depending on the thickness of the cast strip and the casting speed the duration of the cast can vary between 3-4 minutes and 10 minutes. It is accepted that these casting times are short but from roll and strip temperatures and cooling water exit temperatures it is apparent that steady state conditions are being achieved. However, plans are in progress to scale-up the casting operation in order to produce larger quantities of material for subsequent processing and so a second, much larger furnace is currently being installed. This will give casting times in excess of 30 minutes and coil weights up to 150kg.

3. Casting Trials

At the time of writing over 400 experimental casts have been carried out. The vast majority of the trials have used conventional wrought aluminium alloys but, in addition, trials have been conducted with other non ferrous alloys, metal matrix composites, Al-Sn alloys and a series of Al-Cu binary alloys (0-33 %Cu). The effect of grain refining additions on twin roll cast microstructures, and casting conditions on segregation, are also being investigated in parallel programmes.

In the majority of casting trials a series of casts at various thicknesses have been carried out. Material is usually produced at a target gauge of 6 mm to compare with normal commercial production as well as at target gauges of 4 and 2 mm. Some materials have been produced at even thinner gauges; AA 1070, for example, has been produced at 1.25 mm.

The following is a summary of the alloys that have been cast successfully:

3.1 Aluminium wrought alloys

The specific aluminium alloys that have been cast are given in Table II. With the exception of the high purity materials the alloys are classified using the Aluminum Association system.


Table II - Aluminium Alloys Cast Experimentally

High Purity 99.993 %A1 99.98%A1 99.93 %A1

lxxx AA1070 AA1100

3xxx AA3003 AA3004 AA3104

5xxx AA5052 AA5182

7xxx AA7072

8xxx AA8111

A significant number of casts have been carried out on alloy AA5182. This alloy is a high strength, non heat-treatable alloy that is currently the focus of much attention. The alloy is used universally for can ends but there is also a view that aluminium alloys will be used for automotive panels in the near future and AA5182 may be a candidate for this application.

The traditional process route for this alloy is DC casting followed by hot and cold rolling and until recently it was generally accepted that twin roll casting was not a suitable process for this alloy, particularly if the application was demanding. Centre line (channel) segregates are acknowledged as a major problem in long freezing range alloys and AA5182 with its freezing range of 61 °C. has caused particular problems. This, together with the slow rates of production and extremely high roll separating forces associated with this alloy, has made the twin roll casting of this alloy technically unattractive.

The economics of twin roll casting are such, however, that there would be considerable benefits if the technology could be developed to produce AA5182 with equivalent or superior properties to conventionally produced material.

The results of the current investigation have been extremely encouraging. AA5182 has been produced at casting speeds of 15 m/min at gauges down to 1.40 mm. At these high casting speeds the as cast microstructures are markedly different from conventional twin roll cast structures with a modification to the normal segregation pattern and an absence of channel type segregates.

The productivity figures are high, typically 3-5 tonnes/hour/metre of cast width which, if scaled up to commercial diameter caster rolls, is equivalent to approaching a 5 fold increase in productivity over conventional twin roll casting.


3.2 Al-Cu Binary Alloys

In addition to the inability of the model to predict accurately the productivity gains in wide freezing range alloys referred to above, it has become apparent that the tendency to form heat lines and for sticking to occur are both a function of alloy content and freezing range. In order to gain a fuller understanding of the differences between the solidification behaviour and the onset of defects in narrow and long freezing range alloys, a series of Al-Cu binary alloys with copper contents of 2,4,8,16, and 33 weight per cent have been investigated.

The results from this series of trials shows that castability, as measured by the ability to cast the alloy without sticking, increases as the alloy content increases. The results also show that, at least for this family of alloys, there is no simple relationship between freezing range and sticking tendency as the Al 4%Cu, which has the widest freezing range, sticks at thin gauges whereas higher alloy contents are castable. Under normal casting conditions there was no evidence of localised heat lines in these alloys.

3.3 Al-Sn Alloys

A range of Al-Sn alloys were examined but many of the alloys proved difficult to cast due to the extremely long freezing ranges and low eutectic temperature of the alloys examined; typically 430 and 228°C, respectively.

As a result, it was not possible to produce strip that had completed solidification before leaving the rolls and the strip that was produced exhibited tin rich surface layers. For some applications this is not regarded as a serious problem as the surface layers are removed by machining at an early stage of the manufacturing process.

3.4 Grain Refining Trials

The efficiency of Al-5Ti-lB grain refining rod on twin roll cast A A1070 and AA3004 has been examined(8). For AA1070 the structure of grain refined material is much finer than untreated material by a factor of about 0.05 to 0.1. However, grain refinement is more effective when casting at thicker gauges. Although the grain refiner refines the microstructure it does not change the grain morphology; up to an addition of 0.06% titanium the structure remains columnar. For AA3004 the addition of grain refiner also leads to refinement, in this case by a factor of about 0.15 to 0.25. However, for this alloy the structure changes from columnar to equiaxed.

3.5 Lead Alloys

A number of experimental casts have been carried out on a series of lead rich alloys. The casting trials themselves were extremely encouraging once the differences between lead and aluminium were fully recognised and sheet 2.15mm thick was produced. Because of the


problems associated with the rolling of lead there is a considerable attraction in casting at or near the finished gauge.

3.6 Metal Matrix Composites (MMCs)

The potential for application of the twin roll casting process to the production of MMC materials in sheet or strip form was examined by preliminary experiments on three types of material including both particulate and fibrous reinforcements:

(Al - Si) + 20% by volume SiC particles Al + silica fibres

Al + carbon fibres

The results of the MMC casts were encouraging, particularly the casts of Al-Si alloy containing silicon carbide particles. The Al-Si-SiC material cast extremely well at nominal gauges of 3 and 4 mm and the structures showed that the SiC particles were distributed reasonably uniformly throughout the cast strip.

An interesting feature of the as cast microstructures is the extremely fine eutectic of the matrix, attesting to the very high solidification rates achieved in the process.

The feasibility of continuously feeding a fibrous mat through the tip to become incorporated into the solidifying matrix was also demonstrated but much more work is needed in this area before it can be proved that this process could be used to produce fibre reinforced strip or sheet. The preliminary results are reported more fully elsewhere(9).

It is apparent that twin roll casting is a viable method for producing composite materials at near net shape and this area warrants further attention.

4. Cast Product Evaluation

A considerable amount of metallographic work has been carried out and certain alloys have been further processed into finished products. For obvious reasons this phase of the work is lagging behind the casting trials but some interesting results and lines of further study are emerging. A number of solidification defects and microstructures specific to the twin roll casting process have been identified(lO). These include both macro and microstructural features located both internally and at the surface and in some cases are nonspecific to the alloy being cast, resulting purely from the twin roll casting mode. For example, Figure 4 shows the development of the subsurface microstructure resulting from a particular roll surface finish, and Figure 5 is an example of a segregation pattern typical of long freezing range alloys cast at high speeds.


Figure 5. Typical segregation pattern observed in a long freezing range alloy cast at high speed.(Alloy AA5182 cast at 9 m/min)

Figure 4. The development of subsurface microstructure in AA5182


5. Alloy Development Programme

The present programme has not yet begun activities in this area. However, some general background comments can be made. Although it is clear that at some stage a new family of twin roll casting alloys will be developed, to date only a limited number of 'designer' alloys have been formulated. Alcan(ll) have developed a family of high strength foil alloys based on Al-Fe-Mn and Al-Fe-Si that rely on the increased cooling rates achievable in twin roll casting to promote novel microstructures with enhanced mechanical properties.

A major problem for twin roll caster users is typified by the beverage can making industry which is dominated by two alloys; 3004 for the body, 5182 for the end. Both of these alloys were developed specifically for the DC casting process and are not suited to conventional twin roll casting. Significant time and effort has been spent trying to produce cans from twin roll cast material and although it is claimed to be technically possible nobody is producing cans commercially by this route. The problem is not in casting AA3004 which is relatively straightforward, but in the can making operation where the can maker is faced with a product that is different in character to DC cast/hot rolled material and is therefore deemed unsuitable.

By casting thinner and faster, materials with significantly different structures and properties will be produced and there is at least the possibility that properties that can only be produced by twin roll casting will result. This, together with new uses for cast products, may well be the driving force for a major alloy development programme.


REFERENCES

1. Sir H. Bessemer, British Patent No 11317, 1846

2. I. Jin, L.R. Morris, and J.D. Hunt, Journal of Metals, June 1982, p70.

3. M.J. Bagshaw, J.D. Hunt and R.M. Jordan, "Casting and Welding Processes", Santa Barbara, California, 1986, p. 135.

4. M.J. Bagshaw, J.D. Hunt and R.M. Jordan, Cast Metals, Vol. l ,No. 1, 1988, pl6.

5. Davy McKee (Poole) Ltd., Internal Report, 1989.

6. D.J. Browne, "The Measurement of Heat Transfer Coefficients in Roll Casting", M.Sc.Thesis, Oxford University, 1989.

7. E. Scheil, Z.Metallk.,42. 1942, p70.

8. X. Yang, Oxford University, unpublished research.

9. P. Griffiths, "The Twin Roll Casting of Metal Matrix Composites", Part II Thesis, Oxford University, 1990.

10. D. J. Monaghan, Oxford University, unpublished research.

11. L.R. Morris, "Solidification and Casting of Metals", The Metals Society,1979, p218.

ACKNOWLEDGEMENTS

Professor Sir Peter Hirsch FRS is acknowledged for the provision of laboratory facilities. The research programme is jointly funded by SERC/DTI and Davy McKee (Poole) Ltd under the auspices of the TCS Directorate. Some of the aluminium alloys have been kindly supplied by Alcan International, Banbury, the grain refining rod by Anglo Blackwells, Widnes, and the refractory tip sections by Granges Luxembourg Aluminium Company, Luxembourg. We are also indebted to Dr J. Jang for experimental assistance, and the Davy McKee Training Officer for the allocation of apprentices to assist with casting trials. The permission of the Technical Director, Davy McKee (Poole) Ltd, to publish this paper is gratefully acknowledged.

273

Aluminum alloy castings — production routes for optimum fracture toughness

G. Bellamy Ontario Hydro Research Division, Toronto, Ontario, Canada

ABSTRACT

The importance of pouring temperature and cooling rate through the freezing range have been established and quantified for optimization of tensile and fracture toughness properties in a number of commercially popular cast aluminum alloys. Production techniques developed are applicable to processes using sand, plaster or permanent metal moulds. Soundness and fracture toughness have been greatly improved with no sacrifice in tensile properties.


INTRODUCTION

Many cast aluminum alloy components are utilized in the construction, operation and maintenance of transmission and distribution structures. Typical examples are overhead conductor suspension clamps and grounding cable clamping and connecting devices. Unfortunately specifications covering cast aluminum components are typically undemanding on requirements for tensile ductility and the question of fracture toughness is not addressed. Also in most cases specified tensile properties need only be met by separately cast non-machined test bars which rarely reflect the properties of actual production castings. Consequently, the customer receives castings which often experience brittle fracture under the influence of normal installation stresses.

It is encouraging to note that ASTM now recognize this problem and have introduced a more demanding specification, ie, B686 which strongly recommends the use of chills in sand castings and demands somewhat higher tensile ductility values in specimens machined from actual sand castings. Our research work intended to progress further in this direction and to determine control parameters for production of optimum tensile ductility and fracture toughness in cast aluminum alloy components.

HISTORICAL BACKGROUND TO THE PROBLEM

Because of their excellent fluidity and good overall foundry handling qualities the alloys most favoured by the suppliers are based on the aluminum-silicon binary system. The heat-treatable 7 wt% silicon alloy 356 or its premium quality variant A356 represents the most favoured composition. The requirements of various specifications for the premium grade are listed in Tables 1 and 2.

Where considerations of possible warpage apply or a requirement for optimal tensile ductility exists, the manufacturer will typically utilize a non-heat treatable alloy from the aluminum-magnesium system. Tables 1 and 3 present specified requirements for the commonly chosen type 535.0 alloy.

Regardless of the type of alloy selected, it has been repeatedly demonstrated that tensile properties of specimens machined from failed castings fall considerably below the very modest minimum values tabulated in the purchase specifications.

Both of the aforementioned popular alloys have wide freezing ranges, ie 56°C for A356 and 80°C for type 535.0. Consequently, when foundry practice results in relatively slow cooling through the casting solidification range there is a marked tendency toward development of scattered interdendritic porosity in the aluminum-silicon eutectic type system and gross intergranular unsoundness in the peritectic type aluminum-magnesium alloy system. The stress concentrating effects and loss of load bearing area resulting from such discontinuities contribute substantially toward the problem of brittle fracture under very modest loading conditions. Also resulting from slow solidification rates is the excessive growth of coarse microstructural constituents. When the brittle silicon phase in alloy A356 exists as a semi-continuous network of large platelets the overall toughness of the casting is greatly compromised. Whilst the matrix


properties of the 535.0 alloy are generally acceptable in terms of tensile strength, ductility and toughness the overall determinant of casting behaviour is usually the degree of grain boundary unsoundness which is generally sufficiently severe to guarantee brittle fracture.

THE INFLUENCE OF FOUNDRY PRACTICE ON CASTING QUALITY

Attainment of a suitably high solidification rate has been stressed as crucial if widely scattered defects attributable to either interdendritic or intergranular shrinkage are to be avoided. Attendant benefits to the overall microstructure of the aluminum-silicon alloys were also discussed. Further refinement of the silicon phase is possible by incorporation into the alloy composition of a trace (0.005-0.015 wt%) amount of a suitable modifying element. Sodium has been the most widely used with strontium as a popular alternative and antimony showing some interesting possibilities.

Rapid solidification clearly depends on heat being swiftly conducted from the mould. With permanent metal moulds their high conductivity favours this desirable objective. Sand and to an even greater degree plaster have highly insulating properties and it is generally necessary to incorporate strategically located metal chills to act as heat sinks. Solidification rates in the vicinity of chills are high and undercooling below the alloy solidus avoids scattered shrinkage problems. The judicial placement of chills is designed to provide both a sound structure in critical areas of the casting and to favour progressive planar (ie, non-dendritic) solidification towards risers and feeders, thereby confining shrinkage to disposable appendages beyond the boundaries of the casting proper.

Undercooling and rapid planar solidification in a controlled direction also depend upon establishment and maintenance of steep thermal gradients. Toward this end alloy pouring temperatures and mould surface temperatures should be as low as possible and in the case of the mould should ideally remain so throughout the solidification process.

Most of ou • suppliers use either oil or gas fired furnaces of the crucible or reverberatory type. Significant disadvantages are inherent in the use of such equipment. Slow melting rates and in the case of reverberatory furnaces, particularly long holding times both result in unacceptable loss of modifying elements by volatolization. This is most serious when sodium is employed. Long term exposure to the atmosphere and combustion products introduces undesirable contaminants which may result in dross entrainment or gas porosity in finished castings. Degassing often removes modifying elements necessitating further ladle modification additions. Finally, difficulty in temperature control results in variable pouring conditions which adversely affect production of sound castings. Use of medium frequency electric induction furnaces avoids all of the aforementioned difficulties and provides not only uncontaminated, compositionally homogeneous, alloy of precise pouring temperature but also results in a clean, cool, quiet working environment and high productivity.

The desire to minimize thermal shock and attendant possible premature failure of permanent metal moulds may prompt a supplier to use an excessively high pre-heating temperature which has an adverse effect on establishment and maintenance of desirable temperature gradients during casting solidification. Non-metallic moulds and generously proportioned strategically located associated metal chills should be held at a temperature no higher than that necessary to maintain safe pouring conditions by expulsion of moisture.


EARLY RESPONSES TO THE LOW STRENGTH AND EMBRITTLEMENT PROBLEM

Recognizing the variations in technical sophistication among our suppliers, the prohibitive cost of permanent metal moulds for small orders and a general reluctance to use chills in non-metallic moulds, a decision was made to specify only alloys with very small freezing ranges. These are not susceptible to development of widely scattered shrinkage defects. Unfortunately, available alloys of this type fall into two distinct groups each of which has its own disadvantages. The simple non-heat treatable binary eutectic type such as the aluminum 10-12 wt% silicon composition is purchased to specifications guaranteeing tensile yield strength which may be as low as 60 MPa. Higher strength precipitation hardened aluminum-silicon alloys of approximately eutectic composition are available but their specified minimum tensile elongation values are less than 1 % on a gauge length of 4D and they tend to be extremely brittle. Despite the foregoing, the low strength eutectic alloy has proved to be very useful in certain applications.

Rationale for Possible Application of a Low Strength Eutectic Alloy

Figure 1 illustrates failure by brittle fracture of a typical switchgear component. The component was originally fabricated in alloy 535.0 and supplied to ASTM specification B26 for sand castings. Considering the loss of load bearing area represented by the grain boundary shrinkage defects of Figure 2, it is not surprising that the casting failed prematurely under normal loading conditions. A replacement, sand cast in alloy 356-T6 fared no better. Hardness tests on this material gave results which correlated well with acceptable tensile strength values. However,, as noted in Figures 3 to 6, the loss of load bearing area due to the presence of interdendritic shrinkage porosity is considerable. Measurement of this area on the fracture face of the broken tensile test specimen of Figure 6 allowed a correction factor to be applied to tensile strength results which then placed the ultimate tensile strength in excellent correlation with the hardness of the alloy.

Attempts to eliminate the scattered shrinkage pores by having recourse to an alloy of near eutectic composition were successful. A desire to maintain yield and ultimate tensile strength properties at or above the minima specified for the previously used 535.0 and 356-T6 alloys resulted in selection of alloy 336.0-T65. This alloy is normally poured into a permanent metal mould and there is no specified minimum tensile elongation value. Sand castings in this heat treated alloy had the desired tensile strength but essentially zero ductility and very poor fracture toughness. Brittle fracture resulted from normal bolt tightening stress during component installation.

Considering the evidence displayed in Figure 6, it is clear that a casting which loses half of its load bearing area due to interdendritic shrinkage porosity is no better than one having half the material yield strength and no porosity. Using this rationale a number of successful substitutions have been made using the relatively tough ductile aluminum-silicon non-heat treated eutectic composition alloy S12N ordered to CSA specification HA9. A related significant advantage of this alloy is excellent weldability with no tendency toward development of heat-affected zone unsoundness.

DEVELOPMENT OF A PROGRAM FOR OPTIMIZATION OF DUCTILITY AND FRACTURE TOUGHNESS IN HIGH STRENGTH ALUMINUM ALLOY CASTINGS

Preliminary responses to unacceptably high failure rates had noted the suppliers tendencies to use high freezing range alloys and to utilize foundry techniques which invariably result in gross and widespread interdendritic porosity or intergranular unsoundness. Compounding the


Figure 2 Sand Cast Alloy 535.0 with Gross Grain Boundary Unsoundness. 100 M

277

Figure 1 Brittle Fracture of an Alloy 535.0 Sand Cast Switchgear Component X 0.25 mag.

Figure 4 Gross Interdendritic Shrinkage Porosity in Sand Cast Alloy 356-T6. 100 M

Figure 3 Gross Interdendritic Shrinkage Porosity in Sand Cast Alloy 356-T6. 100 M.



Figure 5 Interdendritic Porosity and Coarse Eutectic Structure in Sand Cast Alloy 356-T6.

279

a) Interdendritic Shrinkage Porosity, 1000 M

b) Coarse Eutectic Structure, 50 M


Figure 6 Tensile Test Specimen Showing Substantial Loss of Load Bearing Area Due to Interdendritic Shrinkage Porosity

1000 M


problems of stress concentration and loss of load bearing area introduced by these defects was a general failure to provide a suitably modified tough, ductile matrix microstructure. Substitution with low freezing range alloys of near eutectic composition was successful in eliminating unsoundness and where low yield strength was acceptable this approach provided an effective solution to the problem of brittle fracture. However, for many applications a combination of both high tensile strength and fracture toughness is preferred. A research program was devised to determine the production parameters necessary to achieve this goal. The controllable variables in the casting process are alloy composition; melting technique; pouring temperature, mould material and temperature and heat treatment parameters.

A decision was made to utilize alloy compositions most favoured by our suppliers. These are listed in Table 1 as alloys A356 and 535.0. The benefit of also including a low freezing range alloy of eutectic composition was recognized and a strontium modified version of the West German DIN 1725 specification for alloy wa/3.2381.61 was selected. Compositional details of this sand casting alloy and a related one for permanent metal mould casting are included in Table 1.

Available equipment dictated that melting would be carried out in electric resistance heated refractory crucibles. A 0.5 kg charge could be heated to the desired pouring temperatures in 30 minutes. The lowest feasible pouring temperature was considered to be 50°C above the liquidus. Pouring temperatures of 700°C and 800°C were used to assess the effect of this variable on casting properties. The moulds had tapered cylindrical cavities 132 mm high. Diameters were 31.5 mm at the top and 25.4 mm at the bottom. Mould materials were selected to offer a wide range of thermal conductivities and heat transfer properties. In descending order of efficacy the following were selected:

a) Copper; b) AISI type 1020 carbon steel; c) Plaster of Paris with a 523 mm thick copper chill in the base; and d) Plaster of Paris without the chill.

Sand moulds for an actual production part were obtained from a casting supplier. Details of the part appear in Figure 7.

Laboratory produced heat treatable castings were water quenched and artificially aged to the T6 condition.

Effect of Mould Material

Cavities of the aforementioned shape and size were machined in copper and steel barstock of 75 mm diameter and 150 mm height. Plaster moulds with and without chills were of the same shape and size.

Chromel-Alumel thermocouples were positioned along the centre axes of the mould cavities with their hot junctions located 25 mm from the bottom. For the commercially fabricated sand mould of the air-set type a similarly dimensioned feeder cavity was selected to house the thermocouple. The walls of this cavity were backed with 250 mm to 450 mm of sand. To expel moisture and thereby ensure safe pouring conditions the moulds were heated to 100° prior to pouring. Figure 8 provides cooling curves and alloy solidification rates obtained with the various mould materials. The curves were obtained by monitoring the cooling of type A356 alloy which was held at 700°C for 10 minutes prior to pouring. Similar solidification rates were obtained

X 0.2 mag b) Underside

Figure 7 Details of the Gating, Risering & Casting Layout for a Cable Clamp Component


a) Plan View X 0.2 mag

800

800

800

20

40

60

80

Time -

Seconds

a) Copper

2

4

6

Time -

Minutes

d) Sand

800

500

400 H

300

258°C Per

Minute

20

40

60

Time - Seconds

b) Steel

U°C Per

Minute

2

4

6

Time - Minutes

e) Plaster

800

700 V

47°C Per

Minute

2

4

6

Time - Minutes

c) Plaster • Chill

FIC

UR

E

8

CO

OL

ING

C

UR

VE

S

OB

TA

INE

D

DU

RIN

C

SO

LID

IFIC

AT

ION

O

F

AL

LO

Y

A3

56


Temperature °C Temperature

Temperature c

Temperature '

o T3

S- Q)

CL

E <D


with the type 535.0 and DIN 1725 alloys. The heat treatable alloys were solution treated and aged to the T6 condition. Hounsfield No 14 tension test coupons and standard 10 mm x 10 mm un-notched Charpy test bars were machined with their longitudinal centre axes approximately coincident with casting section mid-radial positions. Tables 2 to 4 compare test results with those obtained from material removed from commercial castings. As anticipated there is a general trend showing marked improvement in mechanical properties with increasing solidification rate. A critical value was achieved with steel moulds, therefore the even faster cooling rate achieved with copper produced essentially no further improvement.

Effect of Pouring Temperature

Clearly this variable has the potential to influence solidification rates and associated mechanical properties in the aluminum casting alloys under investigation. Data presented in Table 5 show this effect in alloy A356 poured into metal moulds. Reduction in solidification rate occasioned by an increase in pouring temperature from 700 to 800°C was minor. The effect on mechanical properties was negligible. However, it may be confidently anticipated that much higher pouring temperatures would have a more deleterious effect on alloy strength and toughness.

Effect of Mould Temperature

For rapid alloy solidification it is essential that a steep temperature gradient be maintained between the mould material and the cast alloy. This is difficult to achieve in sand and plaster moulds without recourse to metal chills. Extending the life of permanent metal moulds by use of excessive preheat to minimize thermal shock cycle damage may be expected to adversely affect casting solidification rates. The effect is significant as evidenced by results appearing in Tables 5 and 6.

Effect of Alloy Composition

Data displayed in Table 4 indicates that early efforts to produce tough and ductile high yield strength castings in low freezing temperature range heat treatable aluminum-silicon-magnesium alloys were only partially successful. At a yield strength of 235 MPa, tensile elongation values of 4% and unnotched Charpy impact results of 9J absorbed energy represented only a modest combined strength and toughness improvement over properties obtainable from the non-heat treated eutectic composition aluminum-silicon alloy S12N sand cast to CSA specification HA9. Since the increased strength and accompanying reduction in toughness and ductility are attributed to the precipitation hardening potential of magnesium combined with silicon in the heat treatable alloy, it was considered useful to reduce the magnesium concentration by approximately 50%. Results presented in Table 5 indicate that with a magnesium concentration of 0.22 wt% tensile strength values are maintained whilst ductility and toughness properties are substantially improved.

Correlation Between Mechanical Properties and Microstructure

A strong correlation has emerged between alloy solidification rate and resultant strength and toughness properties. Microstructural features responsible for these developments are illustrated in Figures 9 and 10. Solidification times of less than 0.25 minutes in permanent metal moulds produced relatively large volumes of aluminum-silicon eutectic in which the hard silicon phase was present as small well dispersed spheroids or short rounded rod-like particles. Details are displayed in Figures 9a and 10a respectively for the 7 wt% silicon A356 alloy and the near


a) 258°C per min (Steel Mould)

b) 47°Cpermin (Plaster Mould with Chill)

c) 25 °C per min (Sand Mould) d) 14°C per min (Plaster Mould)

Figure 9 Effect of Solidification Rate on the Microstructure of Alloy A356-T6, 20 M.


a) Steel Mould b) Plaster Mould with Chill

e) Sand Mould d) Plaster Mould

Figure 10 Effect of Solidification Rate on the Microstructure of Alloy wa/3.2381 20


eutectic composition wa/3.2381. Plaster moulds with a copper chill produced less pronounced undercooling in the poured alloys. Consequently, the microstructure in Figure 9b has a lower, nearer equilibrium concentration of the eutectic phase. The microstructural effects of undercooling were less pronounced with the strontium modified near eutectic alloy as illustrated by a comparison of Figures 10 a and b. Eutectic concentration together with silicon particle shape, size and distribution are similar in castings from both permanent metal moulds and plaster moulds with metal chills. Casting into sand or plaster without chills leds to the development of very coarse eutectic microstructures as observed in Figures 9c and d together with 10c and d. From Figures 9 and 10 together with Tables 2 and 4 a clear correlation emerges in which for a given tensile strength level toughness and ductility increase markedly with eutectic refinement.

A comparison between Figures 2a and 11 shows the effect of increased solidification rate on the microstructure of alloy 535.0. The commercially sand cast alloy has a relatively coarser primary solid solution grain size and shows a marked tendency toward development of grain boundary shrinkage cavities in poorly fed Al/Mg2 Al3 peritectic solidification zones. A dark etching microconstituent Fe Al3 is also visible in Figures 2a and 11. The latter Figure demonstrates the effectiveness of undercooling and controlled thermal gradient development in both refining the primary phase grain size and eliminating grain boundary unsoundness. The benefits show in Table 3 as a markedly improved toughness level with no sacrifice in either tensile strength or elongation values.

CONCLUSIONS

1. Commercial suppliers of aluminum alloy castings for application in electrical transmission line components frequently encounter difficulties in meeting the very modest mechanical property requirements of specifications such as ASTM B26 and B108 for sand and permanent mould castings respectively. Associated with marginal tensile strength and low ductility the castings also display very poor fracture toughness properties.

2. Poor performance results from stress concentrating interdendritic or in some cases intergranular shrinkage defects and the coarse lenticular form of brittle secondary phases. These can be avoided by control of alloy composition; pouring temperature, mould material, mould temperature and heat treatment. For sections up to approximately 30 mm thick this work has identified the critical parameters and demonstrated their effectiveness in producing sound, strong and tough components in alloys A356, 535.0 and both wa/3.2381 and wa/3.2383.

3. To achieve the desired properties, pouring temperature may be between 700 and 800°C. The temperature of either refractory or permanent metal moulds should be controlled in the range of 100-125°C. Use of metal chills (eg copper) is essential when using sand or plaster moulds.

4. Although the introduction during 1985 of ASTM specification B686 is a welcome step in the right direction, it is still relatively undemanding and does not address fracture toughness. Based on the current work it would seem not unreasonable to draft specifications requiring minimum properties outlined in Table 7. These should be obtained on test specimens machined from actual castings.


b) 50 M

Figure 11 Microstructure of Permanent Metal (Steel) Mould Casting in Alloy 535.0

a) 1000 M


ACKNOWLEDGEMENTS

Casting, heat treatment, mechanical testing and metallographic work was carried out by Mr. T.R. Ryans. We are also indebted to a number of suppliers for useful discussion and provision of commercial scale sand moulds.

TABLE 1 - SPECIFIED COMPOSITIONAL REQUIREMENTS

ELEN 1ENT -

1 WT% C( DNCENTRATION

SPECIFICATION ALLOY Si Fe Cu Mn Mg Zn Ti

ASTM B26-86 & A356 6.5- 0.20 0.20 0.10 0.25- 0.10 0.20 ASTM B108-85A A356 7.5 max max max 0.45 max max

DIN 1725 wa/ 9.0- 0.50 0.05 0.40 0.20- 0.10 0.15 G A l Si 10 Mg 3.2381.61 11.0 max max max 0.50 max max

DIN 1725 wa/ 9.0- 0.60 0.30 0.40- 0.20- 0.10 0.15 GK Al Si 10 Mg 3.2381.62 11.0 max max max 0.50 max max

Hydro Low Mag 10.2- 0.60 0.10 0.10- 0.23- 0.10 0.05 12.2 max max 0.30 0.27 max max

ASTM B26-86 535.0 0.15 0.15 0.05 0.10- 6.2- 0.05 0.10-max max max 0.25 7.5 max 0.25

TABLE 2 - MECHANICAL PROPERTIES OF A356-T6 ALLOY CASTINGS

UNNOTCHEDI MOULD SOURCE UTS 0.2% PS % EL ON % HARDNESS CHA fcPY

MPa MPa 4D RA HV J % LE

Steel Commerical 240 175 4 _ 80 6 3.0 Steel Laboratory 311 200 20 20 77 85 20.3

Plaster Laboratory 266 179 25 18 66 37 10.7 & Chill Sand Commercial 171 171 1 - 99 2 0.1 Sand Laboratory 240 166 6 5 60 9 3.0

Plaster Laboratory 174 168 2 3 45 3 0.1

ASTM B26-86 234 166 3.5 _ _ _ _

Spec min min min

ASTM B108-85a 228 179 5.0 _ _ _ _

Spec min min min

ASTM B686-86 262 193 5.0 _ _ _ _

Spec min min min


TABLE 3 - MECHANICAL PROPERTIES OF TYPE 535.0 ALLOY CASTINGS

UMNOTCHED MOULD SOURCE UTS 0.2% PS % EL ON % HARDNESS CHARPY

MPa MPa 4D RA HV J % LE

Sand Commercial 224 125 13 10 80 29 9.7

Steel Laboratory 289 140 14 10 72 98 20.3

ASTM B26-86 241 124 9 _ _ _

Spec min min min

ASTM B108-85a 241 124 8 _ _

Spec min min min

TABLE 4 - MECHANIC AL PROPERTIES OF ALLOY wa/3.2383.61/62-T6 ALLOY CASTINGS

MOULD UN NOTCHED

MOULD SOURCE UTS 0.2% PS % El ON % HARDNESS CHARPY MPa MPa 4D RA HV

J %LE

Steel Commercial 240 210 1 _ 90 5 < 1 . 0 Steel Laboratory 278 235 4 3 69 9 2.3

Plaster Laboratory 228 178 4 2 56 6 1.0 & Chill Sand Commercial 260 200 4 3 102 3 < 1 . 0

Steel* Laboratory 285 196 9 6 70 25 5.1

DIN Spec 220 180 1 __ 80-100 _ _

1725 min min min GA1 Si 10 Mg (sand castings)

DIN Spec 240 210 1 85-115 _ _

1725 min min min GK Al Si 10 Mg (permanent metal mould)

Magnesium concentration reduced from 0.50 wt% to 0.22 wt% in the wa/3.2383.61-T6 alloy.


TABL1 3 5 - ALLOY SOLIDIFICATION KIN [ETICS

ALLOY FREEZING MOULD POURING MOULD COOLING SOLIDI-

RANGE MATERIAL TEMP TEMP RATE FICATION

( ° Q (°C) (°C) THROUGH TIME-THE MINS

SOLIDIFI-CATION RANGE

(°C) PER MIN

A356 56 Copper 700 100 467 0.12

A356 56 Copper 800 100 431 0.13

A356 56 Copper 800 350 233 0.24

A356 56 Steel 700 100 258 0.22

A356 56 Steel 800 100 243 0.23

A356 56 Steel 800 350 78 0.72

A356 56 Plaster & 700 100 47 1.19

Chill

A356 56 Sand 700 100 25 2.24

A356 56 Plaster 700 100 14 4.00

wa 3.2383 8 Steel 700 100 - 0.17

wa 3.2383 8 Plaster & 700 100 - 1.70

Chill

TABLE 6 - EFFECT OF PERMANENT METAL MOULD TEMPERATURE ON MECHAN HCAL PRO PERTIES OF ALLOY A356-T6

UNNOTCHED POURING MOULD MOULD UTS 0.2% PS % EL ON CHARPY

TEMP MATERIAL TEMP MPa MPa 4D

(°C) ( ° Q J % LE

800 Copper 100 302 215 16 59 15.7 800 Copper 350 222 186 5 30 7.6

800 Steel 100 311 200 20 85 20.3 800 Steel 350 245 173 7 24 7.1

TABLE 7 - SUGGESTED SPEC FICATIONS - MECHANICAL PROPERTIES UNNOTCHED CHARPY ALLOY UTS

MPa 0.2% PS

MPa % EL ON

4D HARDNESS

HV J % LE

A356-T6

535.0

Hydro Low Mag -T6

250 min

241 min

225 min

170 min

124 min

170 min

10 min

10 min

7 min

61-77

59-74

55-75

25 min

25 min

20 min

7 min

7 min

5 min

293

Distribution de la temperature a Pinterface moule-metal lors de la solidification de P aluminium

G. Fortin, F.H. Samuel Departement des sciences appliquees, Universite du Quebec a Chicoutimi, Chicoutimi, Quebec, Canada

RESUME

L

1evaluation du coefficient de transfert de chaleur a 1'interface

moule-metal demande de connaitre les temperatures a la paroi du moule et du metal. Un moyen simple pour les evaluer est de determiner la distribution de la temperature a 1

1interieur du moule

et du metal. Mais le changement de phase et la connaissance imparfaite de l'etat a 1»interface moule-metal, necessite des simplifications des mecanismes en jeu, ce qui engendre des erreurs sur les resultats.

La realisation d'un modele representatif est possible, mais la sensibilite associee a 1

1 interpretation des resultats dependra des

simplifications apportees lors de sa confection. Les parametres qui seront etudies determineront les simplifications permettant au modele de faire ressortir leur effet sur le systeme.

Introduction

Le transfert de chaleur a 1'interface moule metal est un probleme tres important dans le domaine de la fonderie, car £l influence le taux de solidification affectant la qualite des pieces de fonderie. Plusieurs parametres influencent le transfert de chaleur, les principaux sont la distributions spatiale et temporelle de la temperature, les proprietes thermiques et physiques de l'alliage et du moule, les etats a 1'interface moule metal, le revetement de surface, l'oxydation et la temperature initiale du metal et du moule. La complexite des mecanismes de transfert de chaleur lors de la solidification et le nombre de parametre accroit les difficultes pour une generalisation du probleme.


Definition du probleme

Le probleme consiste a calculer le coefficient de transfert de chaleur a l

finterface moule-metal. La connaissance des

temperatures a la paroi du moule et du metal est essentielle pour y parvenir. Une methode, qui consiste a calculer la distribution spatiale et temporelle de la temperature a 1

1interieur du systeme

moule metal permet de les determiner.

Le calcul de la distribution de la temperature dans le moule peut etre effectuee a partir d'une simple methode de conduction solide, car le moule ne subit aucun changement de phase. Plus la connaissance des proprietes thermique sera precise, meilleur sera 1'evaluation de la distribution de la temperature dans le moule.

Pour ce qui est du metal, comme celui-ci passe d'un etat liquide a un etat solide, il subit un changement de phase qui vient compliquer le calcul de la distribution de la temperature. Ce changement de phase apporte une variation des proprietes thermiques du metal et est caracteiise par un front de solidification a l

finterface solide liquide. De plus lorsque l'on est en presence

d'un alliage, le passage de l'etat liquide a l'etat solide sera caracterisee par une zone pateuse, dont les proprietes thermiques sont plus ou moins connues. Dans les pages suivantes, trois methodes seront evaluees sommairement pour le calcul de la distribution de la temperature dans le metal, ces methodes tiennent compte du changement des proprietes thermiques du au changement de phase et du front de solidification.

mfctal liquide

M

Figure 1: Representation de l

1interface moule metal pour l'etat

conforme

Les differents etats que l'on rencontre a 1»interface moule metal lors de la solidification (voir figure 1), soit l'etat de transition caracterise par la convection naturelle, l'etat conforme constitue de point de contact solide et de zones gazeuses et

Parol du moule

mfctal

Parol du moule

mttal [cl


caracterise par la conduction solide, l'etat non conforme constitue de point de contact solide peu nombreux et de grande zone gazeuse et caracterise par la conduction gazeuse et la formation de l'espace interstitiel amenant la separation de la paroi du moule et du metal et caracterisee par la conduction gazeuse et le developpement de l'espace entre les parois du moule et du metal vont influencer la valeur du coefficient de transfert de chaleur a 1'interface moule metal. Mais ces etats n'influenceront pas le calcul des temperatures aux parois du moule et du metal, car on les calcul a partir des changements se produisant dans le moule et dans le metal. Cette approche par contre a besoin d'une condition frontiere a 1'interface moule metal pour permettre la continuity entre le moule et le metal. Une condition frontiere ne dependant pas de 1' interface moule metal, mais du flux de chaleur a 1'interface sera propose pour relier la distribution de la temperature du moule a celle du metal.

Evaluation de la distribution de la temperature

ecran table tragante

micro-ordinateur avec logiciel pour

1'acquisition

interface thermocouple

interface sonde

moule

Figure 2: Systeme utilise pour recueillir les temperatures


Pour evaluer la resistance thermique a 1*interface moule metal, la connaissance des temperatures a la paroi du moule et du metal est essentielle. Comme on ne peut mesurer la temperature a 1*interface de chaque zone et ni la distribution de la temperature a l'interieur du systeme moule metal, on doit utiliser une methode permettant de contourner le probleme. La methode utilisee est le probleme inverse de conduction de chaleur

1 qui permet de determiner

la distribution spatiale et temporelle de la temperature a partir de temperature mesuree en laboratoire et des proprietes thermiques des materiaux consideres. Des thermocouples sont inseres a certains endroits du moule et du metal et grace a un systeme d

1 acquisition de donnee, on enregistre la temperature en fonction

du temps de solidification (voir figure 2).

Deux methodes existent pour resoudre le probleme inverse de conduction de chaleur entre deux temperatures connues, la premiere se base sur un maillage mobile permettant de connaitre en tous temps 1

1 emplacement du front de solidification et la distribution

de la temperature. Le flux de chaleur balance l'equation appliquee le long de l

1 interface solide liquide dans un espace de point

constant dans le temps. Les equations regissant les methodes a front de solidification sont:

T, Tf T8

liquide j solide > S(t)

> (x,y,z)

pour Tj>Tf>Ts 1- pour la partie liquide

(pCp). = V (K.VT.)

dt 2- pour la partie solide

(PCP)S = V (KSVTS) a t

3- Bilan d'energie a 1*interface solide liquide dS(t)

-K.VT^VT, = pL dt

Ou Cp est la capacite thermique, t le temps, L la chaleur latente, p la densite, K la conductivity thermique et S est la position du front de solidification, tandis que s et 1 sont les indices pour la phase solide et la phase liquide respectivement.

La deuxieme se base sur un maillage fixe et elle permet de connaitre la distribution de la temperature, tandis qu'une


evaluation doit etre effectuee pour determiner la position du front de solidification. L'effet du changement de phase est automatiquement estime par la relaxation entre l'enthalpie et la temperature. Les equations regissant les methodes enthalpiques sont:

T, Tf Ts

liquide |j solide pour T1>Tf>Ts

3H(T) = V (KjVT)

a t

H(T) =J ou H(T) pc dT + pLf^T)

Ou f{ est la fraction liquide, a est la diffusivite thermique et c est la capacite.

pour tenir compte du changement de phase du metal dans 1'equation enthalpique, on utilise la transforme de Kirchoff

2 avec la diffusivite thermique a, soit: (T K; T = J a (T) dT ou a (T) =

Tc pc

La transformation de Kirchoff est applicable seulement si le changement de phase s'effectue lentement avec la temperature et que le pas de temps utilise permet de les considerer comme constant.

La premiere methode est plus complexe que la seconde, mais elle donne un renseignement supplementaire qui n'est pas toujours necessaire de connaitre. Le choix de la methode utilisee dependra seulement des parametres qu'il faudra determiner.

La resolution de ces equations peut etre effectuee par des methodes de differences finies

2^ ou d

1 elements finis

3'

4 e t 5. Les methodes de

differences finies sont faciles d'utilisation, mais lorsque les geometries deviennent complexes, il peut s'averer ardu de les programmer. Les methodes d'elements finis sont tres difficiles a programmer, independamment de la geometrie du systeme etudie, elles deviennent avantageuses seulement lorsque les geometries sont complexes.

Plusieurs methodes de front de solidification et enthalpique ont ete developpees aux cours des annees pour evaluer la distribution de la temperature lors de la solidification d'un metal

6 , 7*

8. Les

plus connus pour les methodes enthalpiques sont celles de Shamsundar, de Furzeland et les schemas utilisant des matrices


tridiagonales que l'ont retrouve dans l

1article de V.R. Voller ou

encore la methode du q de J.V. Beck. Pour les methodes de front de solidification, elles ont ete developpees surtout avec les elements finis.

Une nouvelle technique existe, permettant de resoudre le probleme inverse de conduction de chaleur

9. Elle provient des nouveaux

developpements dans les methodes numeriques et repose sur une combinaison d

1equation integrale et d'elements finis. Elle permet

de determiner la temperature sur la surface externe du moule sans connaitre la distribution de la temperature a 1

1interieur du

systeme, ensuite a partir d'une methode de difference finie, la determination de la distribution de la temperature a 1

1interieur du

moule est effectuee. L'avantage de cette methode reside en une formulation numerique plus simple des equations du domaine considere, permettant de diminuer les temps de calcul, mais 1

1 inconvenient majeur est contenu dans sa complexity mathematique.

Proprietes dans la zone metallique

Pour resoudre la distribution de la temperature dans le metal liquide, il faut considerer trois cas, soit la zone liquide, la zone pSteuse et la zone solide. Le changement de phase du metal lors du refroidissement, apporte une variation dans les proprietes thermiques et mecaniques du metal. Comme les equations permettant de resoudre le systeme dependent des proprietes thermiques et que celles-ci varient avec la temperature (voir graphique 3) , leur dependance se doit d'etre connue.

Figure 3: Schematisation de la variation de la conductivity thermique en fonction de la temperature

Lorsque la relation propriety et temperature n'est pas connu precisement, les proprietes sont definies a 1*interieur de zones de temperature:


Pour la zone liquide K=K, pour T > Tx

Pour la zone p&teuse K=(K,+K,)/2 pour T, < T < ^

Pour la zone solide K=Kg pour T < T8

Ou K represente une propriete, T la temperature, 1 et s sont les indices pour la partie liquide et solide respectivement.

Conditions frontieres

Resoudre la distribution de la temperature a 11interieur du moule et du metal entre deux temperatures ne pose aucun problemes serieux, mais lorsque lfon arrive a la zone d'interface moule metal, on doit connaitre le taux de transfert de chaleur entre les deux surfaces pour etre en mesure de determiner correctement les temperatures de la paroi du moule et du metal (voir figure 4).

||: moule metal liquide ou solide

||: interface moule metal L:Distance entre les thermocouples Tj:Temperature a la paroi du moule T2:Temperature a la paroi du metal

Figure 4: Perte de chaleur a 1'interface moule metal ou T sont les temperatures mesurees au laboratoire, T1 sont les temperatures a determiner et r indique la position des thermocouples par rapport a la paroi externe du moule.

Pour determiner la distribution spatiale et temporelle de la temperature dans le moule et le metal, on doit done separer le probleme en deux parties, soit resoudre pour le moule et ensuite pour le metal avec comme conditions frontieres:


pour r=R4 ==> T=T4 valide dans le metal et

pour r=R ==> q = hA(Tj-Ti) = q} valide a 1' interface moule metal

Done, en utilisant les deux temperatures mesurees, soit T3 et T4 et la condition frontiere a r=R, on est en mesure de determiner la distribution spatiale de la temperature pour le metal. Ensuite en utilisant les deux temperatures mesurees dans le moule, soit Tj et T2 et en utilisant la condition frontiere a r=R, on determine la distribution spatiale de la temperature pour le moule. La distribution temporelle de la temperature correspond au pas de temps considere.

La condition frontiere q{ est une inconnu du probleme, mais si l'on fait un bilan d'energie a 1'interface moule metal, on aura

4 = Kmouie A (Tj-TiJ/^-Lj) et q~= A (T$-T3) / (L3-L4)

Done ( T ^ / ^ - L , ) = (Tj-T3) / (L3-L4)

La condition frontiere a r=Rj sera une inter-relation entre le moule et le metal, lorsque les temperatures evaluees a 1'interface balanceront 1'equation, la distribution spatiale de la temperature dans le moule et le metal obtenue correspondra a la distribution reelle.

Cette condition est necessaire, mais non suffisante pour assurer une bonne evaluation de la temperature a 1'interface moule metal. Done pour resoudre le probleme, une deuxieme condition frontiere devra etre utilisee. Car le flux de chaleur en theorie est identique pour le moule et le metal, mais en pratique, on traite les deux zones comme etant distinctes l'une de 1'autre, et ensuite on applique la condition d'interface. Si le flux de chaleur dans le moule et le metal n'est pas identique, on doit done le corriger, mais cette correction se doit de fausser le moins possible les resultats. Cette difference dans les resultats peut etre due a deux phenomenes, soit un effet de choc thermique dans le systeme, cause par la methode experimentale ou a la zone d'interface moule metal qui se comporte comme un condensateur, cette zone que l'on peut appeler zone tampon emmagasine de la chaleur durant une partie du temps de la solidification pour ensuite la liberer.

Comme deuxieme condition, trois approches peuvent etre considerees, soit:

La premiere utilise la distribution de la temperature dans le moule. Si l'on considere que le moule possede une diffusivite thermique qui subit peu de variation tout au long du processus de solidification, 1'evaluation de la distribution de la temperature a 1'interieur du moule pourra etre tres precise compte tenu qu'il ne subit aucun changement de phase. En extrapolant la temperature T{ a partir de Tl et T2, la condition frontiere q{ pourra etre calculee, ce qui permettra


d'evaluer la temperature T

{

2 a la surface du metal. Ensuite, en imposant se flux de chaleur comme condition frontiere au metal, la distribution de la temperature pourra etre evaluee.

La deuxieme approche consiste a proceder d'une maniere similaire a la premiere, sauf qu'elle est effectuee a partir du metal et on impose le flux de chaleur au moule.

La troisieme approche consiste a evaluer la distribution de la temperature dans le moule et le metal, ensuite, le flux de chaleur est calcule dans les deux cas, si ils sont differents, on prend la moyenne entre les deux flux de chaleur et on l

1impose comme condition frontiere, et une reevaluation de la

distribution de la temperature est effectuee.

Comme le moule impose le flux de chaleur, il serait logique de l'utiliser, mais cette methode ne tient pas compte de l'effet du choc thermique ni de la zone tampon. Comme l'effet de choc thermique se produit dans le metal, la deuxieme approche en tient compte, mais elle ne considere pas la zone tampon, ni du flux chaleur impose par le moule. La troisieme approche permet de considerer en partie l'effet du choc thermique, du flux de chaleur impose par le moule et de la zone tampon.

Mais, comme le gradient thermique du moule est tres important, que la mesure des temperatures dans le moule est plus precise que celle dans le metal et qu'il n'y a aucun changement de phase dans celui-ci, les resultats calcules a partir des temperatures mesurees dans le moule seront plus pertinents que ceux obtenus avec le metal. En considerant que l'effet de choc thermique est amoindri lorsque l'on utilise des thermocouples tres petits et en considerant que la zone tampon ne peut emmagasiner qu'une tres faible quantite de chaleur par rapport au flux de chaleur a l'interieur du systeme, la premiere approche sera retenue comme deuxieme condition d' interface.

Ensuite, a partir des temperatures d'interface TJ et T|, et du flux de chaleur qif le coefficient de transfert de chaleur h sera evalue comme suit:

h = cfe/(A(Tj-TJ))

Ensuite, a partir des differents etats a l

1interface moule metal et

en considerant la contribution de chaqu'un des mecanismes present, la resistance thermique correspondant au revetement, a l'oxyde, a l'espace interstitiel pourra etre evaluee par rapport a la valeur du coefficient de transfert de chaleur qui aura ete obtenu. Comme l'epaisseur de la zone d

1interface moule metal est tres petite, une

approximation lineaire de la temperature peut etre effectuee entre la paroi du moule et la surface du metal. Ensuite, on evalue la resistance thermique de chaqu'un des parametres pour l'etat considere a 1'interface moule metal.


Conclusion

L'evaluation de la distribution de la temperature entre deux temperatures connues est un probleme qui est simple, mais le changement de phase du metal lors de la solidification et 1'interface moule metal apportent une complexity qui peut etre contournee en simplifiant les mecanismes en jeu. Mais la simplification du probleme engendre des erreurs sur les resultats. Done la realisation d'un modele representatif est possible, mais la sensibilite associee a 1• interpretation des rysultats dependra des simplifications apportees lors de sa confection. Les choix effectuys pour la simplification du modele devront etre fait en fonction des parametres qui seront etudiys, soit: le modele utilise doit etre en mesure de faire ressortir l'effet des parametres sur le systeme.

Le choix d'une methode pour calculer la distribution de la temperature doit etre fait en fonction des parametres etudier, de la minimisation du temps de calcul et de sa simplicity. La methode qui s'adaptera le mieux aux besoins sera toujours la plus adequat.

La technique utilisee pour resoudre la methode choisie dependra de la geometrie du systeme a solutionner. Pour des geometries simples, une methode de difference finie sera plus adequat, tandis que pour des geometries complexes, une methode d

1element fini sera

preferable.

Le modele considere ne represente pas le phenomene parfaitement, mais il peut etre considere comme etant une bonne approximation pour evaluer les effets produit par la variation des parametres initiaux.

Reference

1 - J.V. Beck & B. Blackwell. "Inverse Problems". ,pp. 787-833 2 - V.R. Voller. "Fast Implicit Finite-Difference Method for the

Analysis of Phase Change Problems". Numerical Heat Transfert, Part B, vol. 17, pp. 155-169, 1990.

3 - N. Zabaras. "Inverse Finite Element Techniques for The Analysis of Solidification Process". International Journal for Numerical Methods in Engineering, vol 29, pp. 1569-1587, 1990.

4 - R.W. Lewis & P.M. Roberts. "Finite Element simulation of Solidification Problems". Applied Scientific Research, vol 44, pp. 61-92, 1987.

5 - T.S. Prasanna Kumar, S.D. Pathak & 0. Prabhakar. "Finite Element Formulations for Estimating Feeding Efficiency Factors". AFS Transactions, pp. 789-800, 1985.

6 - J.V. Beck. "Nonlinear Estimation Applied to the Nonlinear Inverse Heat Conduction Problem". Journal Heat Mass. Transfer, vol. 13, pp703-716, 1970.

7 - C. Wei, P.N. Hansen & J.T. Berry. "The O Methode-A Compact Technique for Describing the Heat Flux Present at the Mould-Metal Interface in Solidification Problems". Numerical

302


Methods in Heat Transfe, vol. 2, pp. 461-472, 1983. 8 - D.A. Murio & C.C. Roth. "An Integral solution for the Inverse

Heat Conduction Problem after the Method of Weber". Comput. Math. Applic, vol. 15, No 1, pp. 39-51, 1988.

9 - C P . Hong, T. Umeda & Y. Kimura. "Numerical Models for Casting Solidification: Part I. The Coupling of the Boundary Element and Finite Difference Methods for Solidification Problems". Metallurgical Transactions, vol. 15B, pp. 91-99. 1984.

307

The use of computational fluid dynamics in modelling complex metallurgical processes

R.T. Bui Universite du Quebec a Chicoutimi, Chicoutimi, Quebec, Canada

Abstract

In the metallurgical industry, a host of thermohydrodynamic processes needs or deserves to be analyzed and/or designed using mathematical and numerical models based on the science and techniques of computational fluid dynamics. With the advent of high-speed computers, this opens whole new horizons to academic and industrial workers alike. By working together, the two groups can generate exciting synergetic effects, providing that the model builder understands the needs of the model user, and the model user understands the constraints and accepts the limits to which the model builder is submitted. This paper discusses these needs and constraints, arising from the complexities of the processes on one hand, from the stringent requirements on the discretizations or numerical treatments on the other hand, or a combined effect of both, thus generating compounded difficulties and requiring exorbitant computing times. Based on examples taken from recent university-industry joint research, solutions are suggested including simplified models, sub-models, correlative models, multiple grids, parallel computing. The discussion highlights the need for a close collaboration between model builder and model user at each stage of the work, leading to a trade-off which ideally should yield a model well balanced between representativity and workability.

Keywords

Mathematical modelling, metallurgical process, computational fluid dynamics.

Introduction

Industrial processes involving heat, momentum, mass transfers and chemical reactions are legion in the metallurgical industries, from the primary production of metals to the subsequent stages of transformation. These processes are physically complex. Even each individual phenomenon underlying these processes is, more often than not, complex enough to qualify as a research subject on its own merit. These phenomena also occur simultaneously and interactively in a cause-to-effect relationship. What is more, these processes are of large dimensions and often difficult to access, and as a result it is impractical, costly or sometimes impossible to carry out experimental measurements or plant studies on them. Also, these processes are sensitive, and even small changes in their operational parameters can seriously affect the quality of the product, sometimes in an irreversible manner. These processes are mostly slow, with time constants ranging from days to months. Finally, they are huge energy consumers with not so high energy efficiencies, so that even an improvement of a few percentage points in their behaviour can result in appreciable money savings. A better understanding of these processes,


and a better design, are thus technically and economically justified. As experimental studies are slow, costly, even risky and necessarily limited in scope, an alternative is to build mathematical models and do numerical simulations on the computer.

Considerable advances have been achieved recently in mathematical modelling of thermohydro-dynamic processes, owing to the progress made in algorithm development and code building in computational fluid dynamics on the one hand, and the advent of high-speed computers on the other. Yet it seems like we are chasing our own tail. The more we progress, the more we need accurate analysis, the more we want to drop the simplifying assumptions, and the more we need powerful computers to solve the sophisticated models. At some point in time, we may end up with a good model that is of little or no practical use because too cumbersome or too long to run even on the best computers. We will have replaced an all too complex physical process with an all too complex mathematical model. To break away from the vicious circle, it takes a short cut, providing that the implications of the short cut are well understood and accepted by both the model builder and the model user.

Model User's Needs versus Model Builder's Constraints

In the context of the light metal industries, a number of thermohydrodynamics-based processes can readily be illustrated by examples found in the aluminum industry: the baking of carbon electrodes used in the electrolysis of alumina, the melting of solid aluminum or the remelting of scrap, the holding and preparation of liquid aluminum or of alloys, the calcination of alumina, the calcination of petroleum coke used in the fabrication of carbon electrodes, the extraction of aluminum from alumina in the electrolytic cells... In these processes we recognize several intertwining phenomena such as convective, conductive, radiative heat transfer, phase change, chemical reactions, reactive flow, dispersed flow, multiphase flow and magnetohydrodynamics. Convective transfer is present in the form of natural, forced, mixed or solutal convection.

Model users are mostly found among industrial researchers and production engineers, while model builders may be academic or laboratory workers. Model users need the mathematical models to help them do one or several of the following things:

— study the various aspects of the operation of the process and their effects on productiv-ity, product quality, energy efficiency, and also on other aspects such as equipment life expectancy or working conditions.

— study the various aspects of the design of the process and related equipment. The problem at hand may be one of overall design or partial design.

— undertake parameter studies to analyze the effect of changes in a given parameter (or parameters) of the process; this may be a design parameter or an operational parameter.

Beside the usual requirements that the model should be fairly transparent and have a good degree of flexibility and user-friendliness, model users want the models to take no more than a reasonable amount of time to run on their computer and produce outputs. This is all the more important for parameter studies, where a large number of simulations needs to be made. Yet with all the constraints facing the model builder, this last requirement may not be easily satisfied.

Among the constraints imposed on the model builder, the most fundamental ones come from the complexity of the process itself and of the many underlying physical phenomena, and also from their interactions, as already mentioned. To capture the much needed fine details of such phenomena as a boundary layer, a gaseous leakage or a small-scale recirculation, fine discretizations are required, and this leads to high computing times. A coarser grid often leads


to convergence difficulties, especially when complex phenomena such as combustion are present. This difficulty mostly occurs either at the control volume boundaries, or at the interfaces such as the surface of a solid bed exposed to a freeboard gas.

The higher the number of physical phenomena in the process we must account for, the more variables have to be solved for, at each of the time- and space-discretizing points of the control volume: velocity components, pressures, temperatures, turbulence quantities, radiative quantities, densities, concentrations, chemical species. In some fairly sophisticated models, the number of variables can easily reach the dozen or more. This is still another element that boosts computing time even further. Last but not least, in many processes the operational procedure requires that the geometry and the limits of the control volume change with time. This is the case, for example, with a furnace that is gradually filled with (or emptied of) liquid metal arriving in successive batches, or a melting solid block with a decreasing solid volume while the liquid volume increases correspondingly and fills in the voids as it appears.

Solutions to the Dilemma

1. Simplified models, sub-models, correlations

A problem that occurs to all builders of complex process models at some point is the high computation time. Model builders should look at possibilities of building simplified models with less than three dimensions. Although this seems evident, model builders who are absorbed in pursuing the challenge of three-dimensional (3D) dynamic models, often overlook the many possibilities offered by one- or two-dimensional models (ID or 2D) in a given situation. Less-than-three dimensional models are often good enough as tools for the study of the operation of processes, while fully three-dimensional models are usually required as design tools. One dimensional models are built starting from simplified ID governing equations, in which the process variables such as temperatures, pressures, velocities, concentrations are "collapsed" onto one axis instead of being defined at every point in space. An intuitive physical interpretation is to see the ID value of the variable as its average value over the whole cross-section. The resulting model yields only a limited amount of data, but it is simple and takes less time to run on the computer. A profile of temperatures, velocities ... simplified to ID is often adequate to analyze several important aspects of the operation of the process. What is more, as it takes relatively little time for each simulation, the ID model lends itself better to sensitivity studies or parameter studies, in which simulations have to be run repeatedly in order to analyze the effects of changes in a given parameter or group of parameters.

Two-dimensional models fit in between the ID-model and the 3D-model in terms of costs and benefits. Whenever the distribution of temperatures, velocities ... over a given cross-section proves to be an important factor in the assessment of the operation of the process, 2D-models are required. If this requirement is added to the need to have a ID-profile in the other direction, this means a 3D-model is called for. However, 2D-models may be partial, in the sense that 2D-distributions are calculated only at a selected number of locations considered as critical or representative.

The same remark applies to 3D-models. A complete 3D-model is indeed an ideal tool for the design of the process. Yet quite often it is not much of practical use due to its volume and the resulting long computing time. In this case, 3D partial models or 3D sub-models should be envisaged. The former term refers to 3D-models built for one or some parts of the process only, while other parts remain ID or 2D. This allows the model user to "zoom in" on some important


details. The term sub-model refers to 3D-models built separately, one for each of the main components of the process. This makes the overall 3D-model better structured, more organized, more transparent, less cumbersome to handle and also less error-prone to use. During the model building stage, it is simpler to build a separate sub-model for each process component, and it is easier and faster to verify it and to pinpoint errors or imperfections, as it is not necessary to run the whole model at each test

There is the need, however, to couple the various 3D sub-models together through appropriate interfaces to obtain the overall process model. A sub-model "sees" the one adjacent to it through an "equivalent surface", that must be described, possibly with the help of appropriate simplifying assumptions. This is referred to as the physical interface. Besides, a sub-model interacts with the one next to it by exchanging numerical data, the output of one serving as input for the other and vice-versa. This data exchange is done according to a protocol and can be referred to as the numerical interface.

Still another advantage of sub-models is found when the model is put to use. Each sub-model can be used separately, independently of the others, for the solution of partial design problems. In such circumstances, the physical interfaces between the sub-model and its neighbors must be specified, and other boundary conditions must be fixed.

Sub-models can be built in such a way, not only to partition the process into various components as was said previously, but also to isolate the various underlying phenomena from one another, such as isolating the heat transfer from the fluid flow aspect, thus breaking down the process in order to study the separate effects of one phenomenon at a time. As a result of such study of isolated effects, it may occur that one phenomenon is appreciably more time-varying than another, which then leads to the decision of calculating the latter at a much lower frequency than the former, and this results in additional time savings. This is the case, for example, when heat transfer varies with time while the fluid flow pattern is only slightly influenced by temperatures.

Concrete examples are readily found in various industrial processes. In the operation of ring furnaces used for the baking of carbon anodes or cathodes in aluminum plants, the axial temperature profiles (along the furnace) of the gas and the anodes are important, so a ID-model turns out to be a useful tool in the study of furnace operation (1). When the temperature distribution in a given section of the furnace needs to be analyzed, a partial 2D-model is required (2). On the other hand, to build a comprehensive design tool for an aluminum melting furnace, in which the geometries of the combustion chamber and of the solid charge are critical elements, a full 3D-model is needed (3). The same can be said for a rotary kiln used for the calcining of petroleum coke or the calcining of alumina: a ID-model, with the cylinder axis as the dimension, proves to be a useful tool for analyzing its operation (4), but a 3D-model is required if the effects of gas flow, combustion, air injection, bed flow, bed shape etc. are to be studied (5). In such cases, a sub-model approach is useful or even necessary. For example the melting furnace can be modelled as two self-contained sub-models, one for the combustion chamber, one for the metal charge (see Figure 1). The two must be coupled through an "equivalent surface" characterized by a number of heat transfer coefficients and radiative properties, often determined at the cost of some well chosen simplifying assumptions (6). The refractory walls may play a less critical role except for their surface temperatures and radiative properties, so that they may be approximated in one dimension (thickness) while the chamber and metal sub-models are built in 3D. Similarly, the rotary kiln can be represented by a sub-model of the bed and a sub-model of the freeboard gas, with appropriate coupling between the two.

PAR

AM

ET

ER

S O

F T

HE

BA

TCH

loa

din

g o

f s

cr

ap

me

tal

: n

um

be

r o

f b

loc

ks

po

sit

ion

of

blo

ck

s -k

ind o

t a

llo

y le

ve

l an

d

ma

ss

of

th

e h

ee

l d

istr

ibu

tio

n o

f t

he

arr

iva

ls

of

cru

cib

les

ch

ara

cte

ris

tic

s o

f th

e b

ur

ne

r (f

ue

l.m

as

s f

low

) c

ha

ra

cte

ris

tic

s o

f s

tirr

ing

(c

on

tin

uo

us

or

in

te

rm

it

te

nt

.in

te

ns

it

y)

IPH

OE

NlC

Sl

he

at

tra

ns

fer

(c

om

bu

sti

on

.ra

dia

tio

n,

co

nd

uc

tio

n.c

on

ve

cti

on

) in

un

ste

ad

y s

ta

te

co

nv

ec

tio

n :

ve

loc

ity

fie

ld

pre

-ca

lcu

la

te

d

ca

lcu

lat

e t

he

eff

ec

-ti

ve e

mis

siv

itie

s i

eff

ec

tiv

e t

em

pe

ra

- ,

I**— J

ture

s a

t t

he

eq

uiv

a-

j i

len

t p

lan

e i

he

at

flu

x a

t th

e

co

ntr

ol

vo

lum

e s

ur

-fa

ce o

f m

eta

l

• g

eo

me

tr

ic

al

,^

[ p

ar

am

et

er

s 1

|PH

OE

NIC

S|

he

at

tra

ns

fer

by

co

nd

uc

tio

n w

ith

ph

as

e c

ha

ng

e a

nd

au

gm

en

ted k

to

re

-

pr

es

en

t c

on

ve

cti

on

ve

rif

y th

e s

ett

ing

of

th

e b

loc

ks

sim

pli

fy

mo

de

l to

a

dd

liq

ui

d m

eta

l fr

om c

ru

cib

les

Flo

w

ch

ar

t o

f th

e

co

mp

ute

r m

od

el

of

th

e

alu

min

um

m

eltin

g

fu

rn

ac

e.

Th

e

mo

de

l is

o

rg

an

ize

d

into

tw

o

su

b-m

od

els

, o

f th

e

co

mb

us

tio

n

ch

am

be

r a

nd

th

e

me

ta

l c

ha

rg

e

re

sp

ec

tiv

ely

, a

nd

c

ou

ple

d

th

ro

ug

h

an

e

qu

iva

len

t p

lan

e.

Th

e

left

ha

nd

s

ide

sh

ow

s

th

e

pr

ep

ar

atio

n

of

th

e

da

ta

file

s,

th

e

rig

ht

ha

nd

s

ide

sh

ow

s

th

e

re

cu

rre

nt

co

up

lin

g

pro

ce

ss

. M

es

h

ge

ne

ra

tio

n

is

do

ne

us

ing

P

AT

RA

N

an

d

co

mp

uta

tio

n

us

ing

th

e

ge

ne

ra

l p

ur

po

se

CF

D

co

de

PH

OE

NI

CS

.

So

urc

e:

Re

f. (

6)

Fig

ur

e

1 G)

©


0"

^c

ha

mb

er

} ^

ME

TA

L

jex

tern

al

co

de

j

inp

ut o

f d

at

a a

nd

pre

pa

rati

on

of th

e m

es

he

s :

rad

iati

on

an

d

refr

ac

tori

es

ca

lcu

lati

on

of

th

e s

ha

pe

fnrtn

rs

(M

on

te

Ca

rlo

)

jext

ern

al

cod

ej

(ge

ne

ra

tio

n o

f th

e m

es

h w

ith

PAT

RA

N [P

IL

cod

es |

pr

eh

ea

tin

g a

rriv

al

of

cru

cib

le

#1

arr

iva

l o

f c

ruc

ibl

e #

2

arr

iva

l o

f c

ruc

ibl

e #

n

he

ati

ng

|ph

oe

nic

s|

flu

id

flo

w w

ith

he

at

tra

ns

fer

(c

om

bu

sti

on

,ra

dia

tio

n.c

on

du

cti

on

. c

on

ve

cti

on

)in

ste

ad

y s

tat

e


Finally, correlations play a crucial role and their importance cannot be overstated. Used as part of the model or sub-models, they shorten considerably the computation without sacrificing the representativity of the model. Classical correlations such as flows on plates or pipes, flows in packed beds, boiling heat transfers, friction factors, drag forces and the like, are reported in the literature as results of experimental or semi-experimental research, and widely accepted. Other correlations are less widely accepted but should be considered if they are helpful in simplifying the model and if they can be shown, e.g. by comparison of known cases, to be realistic for the situation at hand This is the case for example with the use of a turbulent viscosity as a function of the known laminar viscosity to avoid the more rigorous calculations of turbulence quantities.

In still other cases, correlations may be arrived at after a long and complete analysis leading to relatively simple results that can be considered as typical of a given physical situation. An example is found in the determination of the internal transverse motion of a particulate bed inside a rotary kiln, treated as a non-Newtonian flow. The bed behaves very much like a superposition of two distinct layers, the thin top layer, cascading downward at high speed, and the bottom layer, moving slowly upward in a near plug flow. The calculation can be considerably simplified in practice by assigning to each layer a different value of Newtonian viscosity and treating each layer as a separate Newtonian fluid. Still another example is seen in the use of an augmented value of thermal conductivity (augmented-k) to account for the effect of natural convection in some convective phase-change problems. Such a simplification is not always applicable, but in situations where it is, is reduces the computation needs considerably.

2. Multiple grids, variable grids

The need for a fine and costly discretization can often be diminished to an appreciable extent by the use of multiple grids or variable grids. The term multiple grids refers to the use of two or more different grids to discretize the same control volume, each grid being chosen to suit the needs of a given set of calculations. As an example, in a reverberatory furnace chamber, a fine grid may be used for flow calculations, while a superimposed coarser grid serves to calculate the radiative heat transfer (7). This results in time savings, as the calculation of radiative transfers requires the determination of shape factors among and between surface elements and volume elements, which is time consuming. The accompanying constraint is that the fine grid, in each direction, must be a multiple of the coarse grid, so that the results obtained on one can be transposed onto the other without ambivalence. On the other hand, the term variable grid refers to the use of one grid system to discretize the whole control volume, but the degree of fineness varies from place to place to suit local needs in accuracy or convergence. As an example, in the rotary kiln with freeboard gas on top and granular bed below, the bed surface and sharp boundaries such as bed corners often require a finer grid than elsewhere in the control volume.

The multiple grid solution often leads to considerable time savings but also requires more care from the model user.

3. Parallel computing

Even after the foregoing measures have been taken, the computing time required by the overall model may still remain so high as to discourage its use in practice, in view of the fact that workdays are only 8-hour long, and model users, as all human beings, are subject to impatience, demotivation and weakening of mental follow-up due to long delays between piecemeal results. In such case, parallel (sometimes referred to as distributed) computing should be envisaged.


Each sub-model is run on a separate processor, and outputs are traded between sub-models. For example in a melting furnace model, the sub-model of the combustion chamber calculates the gas flow, combustion and heat transfer in the chamber and sends the resulting heat flux distribution to the sub-model of the metal charge. The latter calculates the heat transfer, fluid flow and phase change in the charge and sends the resulting surface temperature distribution back to the chamber sub-model, which makes use of such temperatures to do its own calculations again, and the feedback-feedforward cycle starts anew (see Figure 1). In case a sub-model takes less time to run on its own processor than its companions do on theirs, provisions are made to deactivate it for the duration required to have all sub-models synchronized again for the next data exchange. To coordinate such a scheme, a master program may be needed as task manager for the purpose of orchestrating the activities of the various processors.

In order to guide the model builder toward the best possible choice of a modelling strategy, a thorough analysis of the model user's needs must be made, so as to avoid either a costly "overkill" or a simplistic tool unfit for the job. Thus it is essential to determine whether there is the need to study the various alternatives of process operation, or to carry out sensitivity and parametric studies on some given parameters or groups of parameters, or rather to solve process design problems, and whether the design problems envisaged are partial or global. The hardware aspect should not be overlooked: a model built to be run on PC's should not be conceived the same way as the one intended to go on powerful computers.

Conclusion

The mathematical modelling of industrial thermohydrodynamic processes is such a complex and time-consuming undertaking that today the use of general purpose computing codes has become commonplace. Many of the problems arising at the interface between model builders and model users, as analyzed in this article, are also found at another level, namely at the interface between code builders and code users. Also, it often happens that code users are also model builders.

Discussing the reasons behind the apparently slow progress made by general purpose codes within industry, Singhal (8) mentioned the fact that general purpose codes, due to their being general, require a good amount of training to enable the code user to make efficient applications, and as a result, code users often blame code builders for underestimating users' needs, while code builders in turn blame code users for underestimating the importance of training, and for overexpecting from current algorithm development and code building technologies. These problems have their equivalent forms at the level of model builder-model user relationship, with the difference that in model building, the aim is to address a specific application or class of applications, and therefore it is easier to have a close and organic collaboration between builder and user at every stage of the work. Industrial processes are complex, their modelling is not simple, and there is no snapshot answers to sophisticated questions. It takes time and effort to do the work, but it is worth the cost. It is only by working with the model user and carefully analyzing his needs that the model builder can best exercise his choice of a modelling methodology, keeping in mind that ID, 2D, 3D models, or sub-models, or correlative models are all there to serve different purposes and should be taken into consideration, so that the final model or models should reflect a good balance between representativity and workability.

Acknowledgements

The examples quoted in this article are drawn from recent or current joint research projects between the Groupe de recherche en ingenierie des proced6s et systemes (GRIPS) of the


University du Quebec k Chicoutimi (UQAC), the Arvida Research and Development Center (ARDC) of Alcan International Limited, and the Jonquifcre Works of the Soci&6 d'&ectrolyse et de chimie Alcan Limine (SECAL).

References

1. MA. Thibault, R.T. Bui, A, Charette, and E. Dernedde, "Simulating the dynamics of the anode baking ring furnace". Light Metals, AIME, pp. 1141-1151 (1985).

2. R.T. Bui, A. Charette, T. Bourgeois, E. Dernedde, "Performance analysis of the ring furnace used for baking industrial carbon anodes." Can. J. Chem. Eng., 65, 1, pp. 96-101 (1987).

3. R.T. Bui, "Aluminum casting furnace modelling". J. of Metals, 41, 2, pp.43-47 (1989). 4. J. Perron, V. Potocnik, R.T. Bui, "Modelling of the coke calcining kiln", Proceedings of the

Metallurgical Society of CIM, Vol. 8, pp. 87-98 (1988). 5. R.T. Bui, G. Simard, A. Charette, M. Lacroix, Y. Kocaefe, J. Perron, A.L. Proulx, and P.V.

Barr, "3D-simulation of thermal performance of the coke calcining kiln". Proceedings of the Metallurgical Society of CIM, in press (1991).

6. R.T. Bui, A. Charette, G. Simard, A. Larouche, Y. Kocaefe, E. Dernedde, and W. Stevens, "Performance prediction of the aluminum casting furnace". Proceedings of the International Symposium on Production and Electrolysis of Light Metals, Halifax, Pergamon Press, pp. 3-13. (1989).

7. T. Bourgeois, R.T. Bui, A. Charette, Y. Kocaefe, and E. Dernedde, "Simulating the combustion chamber of an aluminum casting furnace". Light Metals, AIME, pp. 375-380 (1988).

8. A.K. Singhal, "A critical look at the progress in numerical heat transfer and some suggestions for improvement". Numerical Heat Transfer, 8, pp. 505-517 (1985).

315

Finite element method in die filling simulation of non-ferrous industrial castings

D. Frayce, G. Salloum, C.A. Loong Industrial Materials Institute, National Research Council of Canada, Boucherville, Quebec, Canada

ABSTRACT

A research program on computer and experimental modelling of the die casting process

is presently being carried out at National Research Council of Canada. In die casting, metal flow

and filling in the die cavity are known to have a considerable bearing on the part quality even

before solidification takes place. This paper describes recent progresses achieved in the

mathematical modelling of the flow fronts under turbulent conditions. In the first phase of this

work, free surface flow subject to a number of constraints is tracked by means of a 3-D finite

element mesh. With clear knowledge of the fluid domain evolution and boundary conditions, the

second phase of the program will lead to the generation of velocity, pressure and temperature

fields. Some examples of filling simulations of industrial die cast aluminum parts are given. The

results of validation of a few experiments are described.

KEYWORDS

Die Casting, Light metals alloys, Finite Element, Flow fronts, Mathematical Modelling, Filling,

Free Surface.

INTRODUCTION

The die casting process is a complex operation during which molten metal is injected

under high pressure into a die cavity. The objective is to manufacture a large number of intricate

and identical components with high precision. The high pressures involved are essentially used

to minimize risks of micro and macro-porosities. Due to the high metal velocity, these pressures

lead to turbulent mechanisms in fluid flow phenomena from the injection point to the gate and

the cavity. This process is non isothermal since it results in the formation of three phases (liquid,

solid and mushy zones) and flow is generally through a complex geometry of varying cross-


sectional thicknesses. The model is particularly suitable to thin wall castings, which represent the

majority of castings produced in industry.

Most investigators in the past have concentrated their efforts on the thermal aspect of

casting (1-3), including the application of very sophisticated methods such as the Boundary

Element Method (4). Nevertheless, it is commonly known (5-6) that the flow behaviour of molten

metal prior to solidification has a considerable influence on casting quality. A clear understanding

of transport phenomena involved in die casting will clearly lead to improved die design and to

the reduction of the high capital cost associated with the machining of the steel die cavity. For

instance, stringent tolerances are often required in the final product and the effect of shrinkage

during cooling must be taken into account. Die fabrication by trial and error may often have

dramatic effects on the yield of the manufacturer's investment.

At the National Research Council of Canada, mathematical and experimental modelling

of metal flow in die cavity, within the frame of an integrated CAD system, is used to help die

casters to design dies and reduce manufacturing costs . In respect of filling, the process in 2D

has been partially solved by some investigators (7-9) with some success. The work presented

herein pertains to three dimensional modelling. It is to be noted that the particular difficulties

mentioned below represent major obstacles to analytical or so-called "exact" solutions:

- The inherent difficulty of solving the Navier-Stokes equations for incompressible flows;

in particular, convective terms of a non linear nature always require extensive numerical iterations

after discretization of the equations.

- The turbulent character of the flow involves time averaging of the equations and

modelling the Reynolds stresses by solving additional partial differential equations.

- The domain boundaries are unknown a priori: the flow front of molten metal entering

the die cavity needs to be tracked before the flow field and other primary variables can be solved.

Constraints due to in-core obstacles in castings must be accounted for. Die casting modelling is

thus a non steady process and transient simulations impose a high demand on computational

resources: the solution at a given time step must be completed before proceeding to the next time

step.

- The three dimensional aspect of the problem entails a significant number of additional

degrees of freedom and a large computing time as well as a powerful CPU. Because of the often

complex geometry involved the finite element method is generally more suitable than the finite

difference method even though, for the same number of degrees of freedom, this method is more

expensive in terms of computing resources.

Several approaches can be used for solving the filling problem. The so-called "exact" and

approximate methods are often adopted (9): the "exact" methods attempt to track the free surface

of the domain exactly and more often use moving grids as well as lagrangian approaches while

the approximate methods take advantage of a fixed mesh over which the moving front is

E X T R A C T I O N , R E F I N I N G A N D F A B R I C A T I O N O F L I G H T M E T A L S 317

approximated. In the latter option, techniques like SOLA (solution algorithm) and MAC (marker

and cell) have been extensively used in the past (10) and coupled to the finite difference method

for resolution of the continuity and momentum equations. Some authors (7) have even applied

theses methods to metallurgical processes such as sand casting with some success.

MATHEMATICAL MODEL AND IMPLEMENTATION

In the algorithm proposed, the casting (including the gate and runner) is approximated by

a fixed three dimensional finite element mesh. The thickness of each element is small as

compared to the other dimensions of the part and not necessarily uniform over the domain. Row

propagates from the inlet of the runner through the lines of the finite element network, these lines

being regarded as cylindrical pipes. The progression of the flow front is determined by the

computation of a line variable obtained from the general one-dimensional Darcy-Weishbach

equation (11); this equation relates the pressure drop to the flow rate as follows:

Where:

• APjj= Pressure drop in cylindrical line (ij) from node i to j

• Vy = Average velocity in line (ij)

• Ltj = Length of line (ij)

• D H jj = Hydraulic diameter of line (ij)

• D m 4 j = Modified hydraulic diameter (ij)

• Ky = Overall minor loss factor

• p = Molten metal density

• f = Friction factor

The second term in equation (1) containing incorporates the effects of flow resistance

caused by abrupt changes of section or changes of direction from one line to the next one.

Although in the laminar regime these terms can be sometimes neglected, they are really

significant in the turbulent regime. In the laminar regime, the hyperbolic dependence of the


friction factor on the Reynolds number can be used:

(2) J Re

for Reynolds numbers up to 2500. For turbulent flow, f is determined by more complex

correlations and is also a function of the relative roughness factor. For convenience, the Moody

diagram (11) can be used. Practical values of the friction factor used for the simulations are given

in Table I.

j Roughness Factor Reynolds Number Friction Factor

1.10"

6 3,000 0.045

1.10* 10,000 0.031

1.10

6 100,000 0.018

1.10

6 1,000,000 0.0115

0.0002 3,000 0.047

0.0002 10,000 0.032

0.0002 100,000 0.019

0.0002 1,000,000 0.0120

0.01 3,000 0.052 1

0.01 10,000 0.043 1

0.01 100,000 0.038 1

0.01 1000,000 0.038

Table I : Friction factor - turbulent regime

Mathematically, the flow front can be computed by considering the pressure drop from

the inlet to a given location. After computing the optimal flow paths of the fluid through the

finite element mesh, the global pressure drops of these paths are computed from equation (1) and

compared. Subsequently, the global line value enables the interpolation of the flow fronts on the


three dimensional mesh. Knowledge of the flow rate at inlet then allows the computation of the

flow pattern versus time. The main advantage of this method lies in its computing speed as well

as in its ability to produce realistic results as will be shown in the next section. In a second phase

of the program, the model will be coupled with a module able to predict both the velocity and

pressure fields, by the resolution of the Navier-Stokes equations at each time step. Conjugate heat

transfer will be also addressed.

SIMULATIONS AND RESULTS

The algorithm presented above has been implemented in the CASTFILL-3D code (module

PROMETHEUS-3D) and tested on simple geometries as well as industrial castings. Experimental

validation of the predictions is now outlined in three test cases considered below:

1. "Die Test" - Double row box - Noranda

Of relatively simple geometry (Figure 1), this part was produced in zinc aluminum alloy

(ZA-8) at the Noranda Research Centre in a 250 ton cold chamber die casting machine. The

casting is approximately 200 mm long and 75 mm wide and has a thin wall (average

thickness=2.5 mm). The runner has about the same length and ends in a 1.5mm thick tangential

type gate. This part was meshed in a CAD system and required 1840 elements and 960 nodes,

including those of the runner and gate. The mesh is shown in Figure 2. Simulations were

undertaken on a VAX 3500 system and required about 15 minutes of CPU time; graphic display

was significantly longer because of hidden lines removal. The flow contours given in Figure 3

show unidirectional propagation, as expected. Also, flow in the row containing the obstructing

boss takes a little longer to reach the end of the cavity. This part was chosen for preliminary tests

and had limited industrial interest.


Figure 1: "Die Test" - Double row box

Figure 2: Mesh of the "Die-Test Casting"


2. Electrical Housing - Eastern Die Casting Cast in aluminum alloy (380), this electrical housing is mass produced at Eastern Die

Casting in Montreal, Quebec. More complex than the previous casting, this geometry consists of

two symmetrical parts with the following dimensions (Figure 4): 205mm x 60mm x 60 mm. For

reasons of geometrical and feeding symmetry, only half of it was considered. The mesh contained

8060 elements and 4132 nodes from the gate (Figure 5). The horizontal gate, relatively large for

this part, was represented by 11 nodes which were the starting points for the initialisation routines

and flow pattern calculations. Simulations were carried out on a VAX 3500 computer and

required two hours of computing time. It should be mentioned that a new version of the code

CASTFDLL-3D runs much more efficiently on personal computers (80386 and 80486), especially

for graphic displays. For example, the CPU time needed for postprocessing these results was

three times less with a PC 486-33Mhz compared with the VAX minicomputer.

Figure 3: "Die Test" box - Flow Patterns


Figure 4: Electrical Housing - Eastern Die Casting

Figure 5: Mesh of the Electrical Housing

322

F GATE


Figure 6: Electrical Box: Flow patterns

The results, as indicated by three dimensional contours with hidden lines removed are

shown in Figure 6. These contours represent the flow front progression with time and can be used

as moving boundary conditions for other analyses. The line crossing the set of contours represents

the longest flow path from the gate. As observed in Figure 6, the strong flow originating from

the gate leads to the convergence of the flow paths at the top surface. The risk of macro-porosity

at this level corresponds to the presence of concentric contours. This feature of the flow was

actually experienced in practice as exemplified by a partial shot in Figure 7. The shot also

showed metal recirculation and air entrapment. The same simulation without the turbulent terms

of equation (1) did not produce the feature observed in actual castings but rather lead to a

smoother propagation of the flow all around the electrical box. For this reason, turbulence

modelling needs to be precisely taken into account for the die casting process. This specific topic

is presently being studied at NRC.


Figure 7: Electrical box: Short shots

3. Transmission Axle tube - Parker White Metal Company

Figure 8: Transmission Axle Tube


Figure 9: Axle Tube Mesh Figure 10: Axle Tube - Partial Filling

This part is a more complex automotive component in zinc aluminum (ZA-27) alloy.

Dimensions are: length=330 mm, max width=130 mm, cylinder diameter=60 mm. Heavier than

the castings analyzed above, this part (Figure 8) is fed by two gates, thus requiring additional

computational efforts.

The mesh consists of 986 elements and 1792 nodes including these of the two runners and gates

(Figure 9). Results of simulations closely resembled the flow fronts in short shots with most of

the metal going from the longer gate to the main cylinder before proceeding to the left lower part

of the bracket, and terminates at the left upper part and the secondary cylinder. Complete and

partial filling simulations are shown in Figures 10 and 11 and an example of a partial shot in

Figure 12. The voids observed at both the secondary cylinder (bottom right) and upper part of

the bracket matched those seen in the simulations.

CONCLUSIONS

By applying the Darcy-Weishbach equation to metal flow in a three dimensional finite

element mesh, it is possible to predict filling in a die cavity accurately. This method results in

a low demand on computing time and provides a practical and effective solution to the detection

of defects such as porosity or cold shuts in die casting.

The CASTFTLL-3D module is a significant improvement over CASTFILL-2D using a lay-

flat approach (8). Propagation of molten metal flow in a complex geometry is tracked by means

of a series of isocontours. Flow fronts and regions with potential filling problem are observed

and identified by studying these contours.

The model is presently supported on VAX mini-computers as well as PC micros. Versions


on some Unix workstations will be available in the near future.

REFERENCES

1. L.E. SMILEY: "Use of a personal computer to predict casting heat flow and

solidification", A.F.S. Transactions, vol 92, pp 689-696, 1988.

2. P.V. DESAI & C.W. KIM: "Heat losses in runner channels", Engineering foundation,

New-York, 1983.

3. E.R.G. ECKERT: "Similarity analysis applied to the die casting process", Modeling of

material processing, A.A.Tseng, 1987.

4. K. DAVEY & S.HINDUJA: "Modelling the transient thermal behavior of the pressure

die-casting process with the BEM", Applied Mathematical Modelling, vol 14, pp 394-409,

Aug. 1990.

5. R.W. HEINE & P.C. ROSENTHAL: "Principles of metal casting", McGraw Hill, New

York, 1955

6. BARTON: "The pressure die casting of metals", Metallurgical Reviews, vol 9, 36, pp

305-314, 1964.

7. W.S. HWANG & RA. STOEHR: "Computer aided fluid flow analysis of the filling of

casting systems", Proceedings of Numiform Conference, Aug. 1986.

8. V.V. KAPPEL, G. SALLOUM and C.A.LOONG: "Modelling and experimental

verification of aluminum and zinc die casting", CSME Mechanical Engineering Forum,

1990.

9. D. FRAYCE & CA. LOONG: "Mathematical modelling of the die casting process, Light

metals, pp 893-902, Elwin L. Rcoy ed., 1991.

10. C.W. HIRT & NICHOLS: "Volume of fluid VOF - Method for the dynamic of free

boundaries", J. Comp. Phys., vol 39, 1981.

11. M. HUG: "Mecanique des fluides appliquee", Cours de 1' Ecole Nationale des Ponts et

Chauss6es", Eyrolles Ed., Paris, 1975.


Figure 11: Axle Tube - Flow patterns

I Figure 12: Axle Tube - Short shot

329

Simulation of different alumina feeding strategies on a training workstation

L. Tikasz, R.T. Bui Universite du Quebec a Chicoutimi, Chicoutimi, Quebec, Canada

V. Potocnik Alcan International Ltd., Jonquiere, Quebec, Canada

M. Barber Alcan Smelters and Chemicals, Jonquiere, Quebec, Canada

Abstract

A process simulator of the electrolytic cell has been built and embedded in a training workstation. It simulates a cell under its own automatic control, and can accommodate the regular as well as the exceptional operating procedures. Different alumina feeding strategies are simulated, including batch feeding and point feeding, and results are evaluated.

Keywords

Modelling, simulation, control, aluminum electrolysis, alumina feeding.

Introduction

For a good operation of the electrolytic cell, it is essential to keep the alumina concentration in the bath as constant as possible, at a value determined as optimal. A too low concentration leads to an unstable situation where cell resistance increases drastically, a too high concentration causes sludge problems and other side effects. So it is important to keep the concentration within the stable zone of operation. The problem is, that the concentration is not a process variable that is monitored on a continuous basis. Some information about the concentration can be obtained from the cell electrical resistance, which is monitored continuously.

Concentration varies in time depending on different alumina feeding strategies and cell condition. Feeding can be done in the traditional way by breaking the crust and feeding by batch, or in a semi-continuous manner by introducing smaller amounts at a time, a method known as point feeding.

A process simulator was built and presented earlier, as part of a joint project between ALCAN and UQAC. It is tailored to run on a training workstation where the operation of the cell can be studied. In this paper, the simulator is used to simulate and evaluate different feeding strategies.

Structure and use of the simulator

The simulator has a modular structure. The main module fulfilling different tasks and the use of the simulator are briefly described below.


Process Model

The modeling work involved is no simple task. However, the alternative would be to conduct plant experiments which are long, costly and sometimes risky, without giving all the answers that a mathematical model can give. For the present purpose, a lumped-parameter model was selected, based on physico-chemical calculations of mass and energy balance and chemical kinetics. The relationships were expressed by a set of differential and algebraic equations, derived on the basis of a thorough analysis of published work, of laboratory and real-plant measurements and plant experience (1-4). The present structure and parameters of the model are the results of several validations and modifications. This part of the study is not closed, further adjustment will likely be made in the future as the need arises.

When dealing with a complex process, it is often helpful to split it into two or more parts and model each of them separately. Each model can be conceived as a self-containing simulator that can be used separately for partial problems. In the model, the cell is divided into anode, cathode and liquid zone. Each of these is in turn divided into elements, within which the mass and energy balance is performed. In certain parts of the cell only heat balance is necessary. In a typical simulation, 18 ordinary nonlinear differential equations are solved, together with hundreds of algebraic equations, that describe cell parameters and material properties.

The set of equations is coupled, time-variant, and nonlinear. Significant linearization led to serious stability problems. The model provides no explicit knowledge as to how to perform analyses or to interpret results. This task remains with the operator, or the expert system, if one is used for cell control or supervision.

The process model regularly cooperates with other modules built into the simulator such as control system emulation, current submodel, etc The simulator is designed to act as a stand-alone artificial cell generating and accepting all the signals which are present in a real cell. Instead of using the simulator's control emulation module, the simulator can be hooked directly to the real cell control system from which it obtains the signals for feeding, anode adjustment, tracking, anode effect killing, etc.

Current Submodel

The current submodel is a special one, differing from the other submodels because the line current is driving the electrolysis. Therefore, the selection of the line current data can be made more freely than the other process parameters. The simulator can be fed with current data of different sources depending on the situation and purpose at hand.

For long-term simulation, when energy balance and temperature calculations are in the focus, the line current can be taken as a constant representing its average value for the whole duration covered by the calculation.

For a shorter period of time, line current variations may have significance. In this case, a handy solution is to use real, on-line current measurements for the simulation purpose. Unfortunately, the actual current might contain serious disturbances which make the simulation difficult. A better solution is to store sets of real current measurements and choose a typical one, free of too abrupt disturbances.

Another option is to identify the properties of the real line current and based on the results, generate series of data with the same statistical properties and select a proper set for the purpose of simulation. In other words, a statistical model of the line current is used.


The current submodel supports all these alternatives and the simulation can be done with the pre-selected source of line current data.

Control System Emulation

To simulate a cell under control, the simulator must have some kind of control emulation. To define and develop the control emulation part of the simulation, the goal of the simulation has to be determined with great care.

One possibility is to build a perfect replica of an existing or planned control system inside the simulator. In reality however, this is an extremely difficult and nearly useless job. It is difficult because of the uncountable incompatibilities between different computational platforms. It is also useless, because the real cell control system can be used instead. Such an arrangement may be useful for debugging the control system.

Another idea is to implement the original control philosophy in a simplified form within the simulator. The emulated control routines differ from the original ones but they can nevertheless be very useful for the search of new control ideas.

In whatever option, the simulator can also be used for testing cell control logic The process model serves as an artificial cell tuned to average or extreme states as required. The tested control routines can be checked again in real-plant tests.

The present control emulator can be set-up in two manners: it can serve as a simple, low-level resistance controller connected to a batch-feeding operational mode, or it can provide the typical functions of a distributed, hierarchical system with resistance control and point-feeding control.

Interfaces

The simulator runs on several hardware platforms. It is fairly portable, because commercial mathematics and graphics libraries are used. The versatility of the simulator can be increased further by adding more interfaces to the system. One way is to hook it on a standard data base allowing full access to the existing plant data. Another is to add analog-digital I/O facilities and make it a nearly independent cell-like signal generator.

Working Environment

The simulator provides an integrated working environment for the user. Three different types of potential users are considered and the access to the knowledge built into the simulator is determined accordingly. The users are separated by private passwords. User environments can be pre-set by upper-level users for lower-level ones.

All the main modules can be reached via a menu-driven, interactive interface. This facility is essential in the initial tuning and also very useful during the simulation. Moreover, the user can choose his imagined position or status. For example, he can move from the "computer control room" where he has just set the target resistance, to the "potroom" in order to switch the state selector at the cell-side to a different state. These facilities will be demonstrated in the next sections.


Initial Tuning

To make the simulation efficient, the adjustable parameters are carefully analyzed, selected and grouped in advance. These parameter groups describe a cell in a certain state with special emphasis on its age and operating characteristics. Similarly, the scheduled work routines are also selected and grouped. Reasonable sets of time diagrams (line current, resistance, alumina concentration, etc.) are also provided.

To start a simulation, adequate groups of parameters have to be selected from the actual user environment. Based on these data, a steady-state calculation is done. It can be considered as part of the initial tuning because the calculated results are applied to verify the given measurements. The dynamic simulation is started from a steady state with parameters that have physical meaning.

Dynamic Simulation

The user has a possibility to interrupt the calculation process and to interact with the simulator. He can generate reports, display different functions or numerical values of the actual simulated state and he can modify the operational actions (add or delete actions, change the amount). The results can be examined on the simulated time diagrams and on the reports. The experience gathered this way helps the operator to compare various cell operation routines and to determine the proper cell parameters.

Simulation of alumina feeding strategies

There are two significantly different feeding strategies applied in the aluminium electrolysis. In one of them, the feeding actions are well separated, rare events (done every 2-5 hours) and the amount of alumina added by a feeding action is relatively high (50-200 kg). In the other, the feeding frequency is high (on the order of 1-10 minutes) and the added amounts are low (in the range of 0.5-5 kg per addition). The former feeding strategy is typical for older cell technologies, the latter is common in modern cells. Both strategies can be simulated and evaluated on the simulator described herein.

The simulation of alumina balance is a rather complex task. The dissolution of smelter-grade alumina in industrial electrolyte has been studied over the past decades. Depending on the applied feeding technique, the electrolyte condition and the alumina properties, the dissolution process widely varies. Real-plant calculations or measurements are mainly based on extrapolation of laboratory experiments. An excellent description of alumina dissolution calculation can be found in (5).

With the present simulator, there are several possibilities to approximate a desired cell state. The user can prescribe the alumina properties, set-up the cell state, determine the feeding characteristics, schedule the cell operation routines and select the applied control logic.

The alumina properties

Smelter alumina should have the following characteristics: be easily soluble in the bath; form a crust and cover the top of the cell; provide heat insulation; protect anodes against burning; be suitable for the dry scrubber process. A detailed description of these requirements is beyond the scope of this article. We concentrate on parameters which directly influence the alumina solubility and crust formation. In the initial tuning phase, the user can prescribe the density,


solubility, heat transfer coefficient and thermal conductivity of alumina. When alumina gets into contact with the bath, there are three possibilities: crust formation, agglomeration followed by settling and dispersion of single grains followed by rapid dissolution. According to these forms, the alumina balance model handles the alumina inside the electrolyte in dissolved, dispersed and settled forms. Solubility parameters for these forms can also be adjusted. A partial model of this process used in the simulator is given in (2).

The cell state

Among the cell's internal submodels, the electrolyte part is the most important for the evaluation of feeding strategies. Here the physico-chemical properties of the bath have to be set up and dynamically calculated. These include the calculation of the electrolyte composition (bath ratio, AIF3, MgF2, CaF2, LiF additives), calculation of electrolyte temperature, liquidus temperature, etc. The user can set the initial fraction of additives and the starting bath ratio as well as the estimated components of the initial alumina balance: alumina concentration in the electrolyte, mass of dissolved, dispersed and settled alumina. Initial temperatures of electrolyte, crust surface, alumina on the top of the crust, etc. can be set as well. Alternatively, these data can be picked up directly from the plant data base.

The feeding characteristics

The two operational feeding routines (batch feeding and point feeding) have different charac-teristics that can be accommodated by the simulator. The user can select the routine to be used during the computation as regular, automated feeding mode. However, there are different possibilities to introduce manual feeding. In batch feeding mode, the manual feeding has the same characteristics as the automated one. In point feeding mode, the user can start an addi-tional manual point feeding (with the same parameters as an automated one) or can prescribe a side-break type feeding. The latter is used to describe an emergency feeding or to approximate the auxiliary feeding effect which follows an anode change.

In batch feeding mode, there is a possibility to introduce uncertainties in the exact feeding time, the added amount of alumina, the broken area of the crust or the average alumina height on the top of the crust. These facilities make the simulation more realistic. In point feeding mode, where the feeding action is more precise, these uncertainties are not supplied. But a fault breaker or an empty alumina container can be simulated on demand.

The control logic

Two types of control logic are provided by the simulator as it was mentioned in connection with the control emulation. The number and the characteristics of these routines can be increased and modified as required. The simple resistance control superposed on a scheduled batch feeding type operation represents an old, traditional solution while the other one (resistance control connected to a point feeding mechanism) is closer to the present-day applications. Both these routines are represented within the simulator only in their basic characteristics and are not necessarily related to a particular control arrangement.

The operational events

Regular, scheduled operational events can be introduced during the simulation. These include metal tapping, anode change, anode position adjustment, feeding actions and anode effect


handling. The user can determine in advance a set of scheduled events (type, action time and amount) and can prescribe events during the calculation. To do this, the calculation has to be interrupted and, with the aid of the utilities provided, the user rewrites the list of scheduled events (cancel, add, modify) or simply insert quick actions similar to the manual anode adjustment or manual feeding. In the following examples, only the feeding and the anode adjustment events are kept in order to focus on the effects of the alumina feeding routines

Examples

The following examples are taken from different phases of the simulation. There are no examples taken from Login, Data Management, Initial Setup and Steady-State Calculation phases. We concentrate on the demonstration of feeding routines. Nevertheless, the general screen arrangement and the accessible utilities shown in the examples are explained.

Figure 1 represents a typical screen format. A day-long simulation was done, the important diagrams (line current, resistance, anode position, alumina concentration and electrolyte tem-perature versus time) are displayed in the center of the screen in the indicated order, from top to bottom of the screen. In the headings, a User-Logo, the name of the actual menu page and the classification of the user can be seen, from left to right. In the footing, the simulated time, the instantaneous numeric values of the diagrams and a message window (presently empty) are shown in the same order. In the right-hand side column, the menu is shown. Available at this particular stage are: Report to display the final results, Display events to check all the events done during the simulation, and a Hardcopy option (which was selected to generate Figures). Quit is used to leave the present stage and continue with other facilities.

U S ER - LOGO D Y N A M I C S I M U L A T I O N U S E R: E X P E RT

7 5 .0 -

I (*A) I

5 5 . 0-

" T Y l T T ^ — i (

vi V T T r T i i r ~ ~ ^ i i u t o

imo

- 24 h

R e p o rt 7 5 .0 -

I (*A) I

5 5 . 0- n.

imo

- 24 h D i s p l ay « v « n ts

5 0 .0 -

R (uOhm

H a r d c o py 5 0 .0 -

R (uOhm

2 . 0-

D (cm) "

7 .0 -

C (%) "

2 .0 -

——' > ̂ N. / ^ v/ Time

1 0 0 0 .0 -

T (C) "I 950 .0 -

1 0 0 0 .0 -

T (C) "I 950 .0 -

Q U IT

T i me I R D C T

24 : 0 7 1 . 59 4 4 . 86 0 . 20 4 . 99 9 7 3 . 86

Figure 1 Batch feeding, overfed operation

Figure 1 shows a traditional batch feeding operation with a simple resistance control. The results of this control can be readily seen on the resistance and the anode position diagrams, while the feeding effects can be studied on the concentration and bath temperature diagrams.


The parameters of this simulation period were very close to reality: the cell is slightly overfed, concentration is high, no anode effect during the simulation. The applied line current reflects real disturbances. A similar simulation can be seen in Figure 2, where the timing of the scheduled batch feeding was kept but the alumina cover on the top of the crust and the broken area were reduced. The concentration varies between wide limits, but in its general trend, it moves to a lower level. Two anode effects can be observed: the first at about 12 h and the second at 23 h. According to the applied control logic, the scheduled feeding was skipped after the first anode effect treatment

USER - LOGO DYNAMIC S I M U L A T I O N USER: EXPERT

7 5 .0 -

I <kA> ~

5 5 .0 -

hTY r ~~l r t Y T 1^^ T i m*

R e p o r t 7 5 .0 -

I <kA> ~

5 5 .0 -

hTY r ~~l r t Y T 1^^ T i m*

D i s p l a y e v e n t *

5 0 .0 -

R (uOhm

4 0 .0 -

Hardcopy 5 0 .0 -

R (uOhm

4 0 .0 -

2 .0 -

D (cm) "

- 2 .0 -

- r ^ ~ W J ^ ~ i — t - J S _ _ , _ _/ ^ - t> T i m*

7 .0 -

C (%) \

2 .0 -

1 0 0 0 .0 -

T (C) ~

950 .0 -

1 0 0 0 .0 -

T (C) ~

950 .0 -QUIT

5.67 967.32

Figure 2 Batch feeding, underfed operation with effects

Figure 3 shows a point feeding operation with a constant feeding rate. The simulation was done from the top-level simulation menu. At this level, the certified user has the right to reach selected Cell Controller, Group Computer, Central Computer and Training utilities. For example, in order to modify the pre-set feeding rate, the user selects the Group Computer menu, and by moving to the Group Computer environment the required modification be can carried out.

For this simulation a constant feeding rate, corresponding to slight underfeeding was choosen in order to test the correctness of the normal feeding rate. The simulation shows that the original feeding rate was concluded to be slightly high. After an anode effect, scheduled fast feeding was executed and once this period was over, the original feeding rate was adjusted downwards. The anode adjustment is done on the basis of resistance control band, no anode movement prohibition is in place.

The last set of figures from Figures 4 to 6 attempts to demonstrate the simulation process. The simulated cell has still point feeding, but the overfeeding and underfeeding periods were alternated systematically. This is the way to keep the alumina concentration as low as possible.


USER - LOGO DYNAMIC S I M U L A T I O N USER: EXPERT

7 5 .0 -

I <kA) '_

Call Controller 7 5 .0 -

I <kA) '_

Group Computer

5 0 .0 -

R (uOhm 1

- 24 h

Central Computer 5 0 .0 -

R (uOhm

- 24 h Training util.

2 .0 -

D (can) "

- 24 h

Hardcopy 2 .0 -

D (can) "

- 24 h

7 .0 -

C (%> I " ~ — " " - ^ ^ ^ ^ ̂ Time

1 0 0 0 . 0-

T (C) . L — — — — - 24 h

1 0 0 0 . 0-

T (C) .

- 24 h QUIT

Time I R D C T

23 : 30 66.96 44.46 - 0 . 1 0 2 .66 981.32

Figure 3 Point feeding with constant underfeeding rate

For demonstration purposes, the width of the concentration band was kept wide enough to see all the changes.

In Figure 4, the user simulates the cell using the cell controller utilities. To him, it seems that he is at the side of a real cell. He can reach all the functionalities of a cell controller: turn the State Selector, set the Feeding Switch from Auto to Manual, Move the Anode Up or Down and Reset the controller. The two last menu functions are provided in order to keep the contact with

USER - LOGO DYNAMIC S I M U L A T I O N CELL CONTROLLER UTILITIES

7 5 .0 -

I (kA) ^

5 5 .0 -

State sel. : A 7 5 .0 -

I (kA) ^

5 5 .0 -Feeding : AUT

5 0 .0 -

R (uOhm

4 0 .0 - -24 h

Anode UP/DOWN 5 0 .0 -

R (uOhm

4 0 .0 - i j u

n - 24 h Reset

2 .0 -

D (cm) [

- 24 h

Hardcopy 2 .0 -

D (cm) [

<-> - 24 h Continue

7 .0 -

C (%) '_

1 0 0 0 .0 -

T (C) '

950 .0 - -24 h

1 0 0 0 .0 -

T (C) '

950 .0 - • i

n - 24 h QUIT

Time I R D C T

15 : 35 67 .56 44.29 0.10 4 .39 972 .71

Figure 4 Point feeding with underfeeding-overfeeding periods Cell Controller utilities


the upper-level menu pages of the simulator. The divided gray message window represents two lights, flashing according to the feeding and the anode adjustment actions.

Figure 5 shows a later stage of the same simulation but from the Group Computer's viewpoint. The access to adjust the resistance target and the feeding rate is located at this level. The right to do the adjustments is given to the foreman.

USER - LOGO DYNAMIC S I M U L A T I O N GROUP COMPUTER UTILITIES

7 5 .0 -

I (kA) \

5 5 . 0-

Set R target 7 5 .0 -

I (kA) \

5 5 . 0-Set feeding rate

5 0 .0 -

R (uOhm

4 0 .0 -

An n —i— -1-1-1 i n —i i -i ^ , , , .n —r~\, v ru,;,!- T i me

Hardcopy 5 0 .0 -

R (uOhm

4 0 .0 -

An n —i— -1-1-1 i n —i i -i ^ , , , .n —r~\, v ru,;,!- T i me

Continue

2 .0 -

D (cm) "

- 2 .0 - - 24 h

2 .0 -

D (cm) "

- 2 .0 - - 24 h

7 .0 -

C <%) "

2 .0 -

" ^ — T i me

1 0 0 0 .0 •

T (C) \

9 5 0 .0 -

i me

- 24 h

1 0 0 0 .0 •

T (C) \

9 5 0 .0 -

i me

- 24 h QUIT

Time I R D C T

22 : 1 7 7 0 . 22 4 3 . 38 - 0 . 1 0 3 . 1 0 9 7 5 . 95

Figure 5 Group Computer utilities

Finally, Figure 6 shows the results of the whole simulation session. The hardcopy was made

USER - LOGO DYNAMIC S I M U L A T I O N TRAINING UTILITIES

7 5 .0 -

I (kA) "

5 5 .0 -

Meters On/Off 7 5 .0 -

I (kA) "

5 5 .0 -Report

5 0 .0 -

R (uOhm

- 24 h

Events 5 0 .0 -

R (uOhm

n - 24 h Set speed

2 .0 -

- 2 .0 - - 24 h

Hardcopy 2 .0 -

- 2 .0 -rrr 1

• ^

T

- 24 h Continue

7 .0 -

C (%) ~

2 .0 -

-~ s^" — y T ime

1 0 0 0 .0 -

T (C)

9 5 0 .0 •

_ _ _ _ _ _ _ _N ^ ^ ^

- 24 h

1 0 0 0 .0 -

T (C)

9 5 0 .0 • j n - 24 h QUIT

Time I R D C T

22 : 5 1 7 0 . 35 4 4 . 89 0 . 5 0 3 . 6 6 9 7 3 . 06

Figure 6 Training utilities


shortly before the end of the simulated day. Now the user deals with the Training utilities: he has access to switch on or off the Meters (footing, center position), to generate intermediate Reports, update Events, Set the Speed of the simulation and make a Hardcopy.

Conclusion

The training workstation is especially suitable for evaluating the different operating routines and possible feeding strategies. It can also help analyze the effect of the anode position control. It proves to be a good research tool that can open new horizons in cell operation, such as combining anode position control with alumina concentration to determine an optimal feeding strategy for the cell. Benefits are expected in terms of adequate process handling, early fault detection and prompt process recovery.

Acknowledgement

The final support comes from Alcan International Limited of Jonquifcre, Quebec and NSERC of Canada. The interest and support of Alcan Smelters and Chemicals, Jonqutere, are gratefully acknowledged.

References

1. L. Tikasz, R.T. Bui, V. Potocnik. Expert Systems Applied to Control of Aluminium Smelters AIME Annual Meeting Proceedings, Light Metals 1990, pp. 197-202.

2. L. Tikasz, R.T. Bui, V. Potocnik, M. Barber. A Process Simulator of Aluminium Cells for Expert System-Based Supervision, 9th IFAC/IFORS Symposium on Identification and System Parameter Estimation, Budapest, Hungary, 1991. (in press).

3. A.Ek, G.E. Fladmark. Simulation of Thermal, Electric and Chemical Behaviro of an Aluminium Reduction Cell on a Digital Computer AIME Annual Meeting Proceedings, Light Metals 1973, pp. 85-104.

4. E.A. Sorheim, P. Borg. Dynamic Model and Estimator for Online Supervision of the Alumina Reduction Cell AIME Annual Meeting Proceedings, Light Metals 1989, pp. 379-384.

5. Xiaoling Liu and al. Measurement and Modeling of Alumina Mixing and Dissolution for varying Electrolyte Heat and Mass transfer conditions AIME Annual Meeting Proceedings, Light Metals 1991, pp. 289-298.

339

Modelling of bed filters — separation mechanisms

F. Frisvold Sintef Metallurgy, The Foundation for Scientific and Industrial Research, The Norwegian Institute of Technology, Trondheim, Norway

T.A. Engh Division of Metallurgy, The Norwegian Institute of Technology, Trondheim, Norway

ABSTRACT

The result of a water model simulation of filtration is presented. A two-dimensional filtration model consisting of rods with both cylindrical and quadratic cross-sections have been used. In order to study the interception mechanism "neutrally" buoyant polystyrene particles in water were employed. The position of deposited particles on collectors was registered.

The experimental results are explained as follows. In the lower velocity range Stoke's settling applies together with a correction factor caused by wall effects. To determine the collision efficiency due to direct interception the boundary layer is taken into account. The fit between experiments and theory is good.


INTRODUCTION

The inclusions present in molten metal prior to casting are an inevitable feature of the production route. It is found from industrial observation that the majority of the inclusions which are deleterious to product quality lie in the range of 1-3 0 microns and are dilutely suspended.

In a depth filter the inclusions are dispersed through part or all of its volume (depth) . It thus has the advantage of having a large surface area for entrapment and can trap particles much smaller than the pores present in the filter bed.

The deep bed filter consists of a packed bed of refractory particles (usually tabular alumina) through which the molten aluminium flows. The inclusions deposit onto the grains of the filter medium due to direct interception, gravity, surface forces, or diffusion.

A two-dimensional filter model showing flow and particle deposition processes inside a filter is presented. The results are used to determine the relative importance of settling and interception. Flow was in the laminar, transition or turbulent range.

FILTRATION THEORY

The collision efficiency, r\, is defined as the fraction of particles approaching the collector that will touch the collector. If an adhesion efficiency of 100% is assumed the collection efficiency equals the collision efficiency. Given a collection efficiency for the single collectors, the filtration efficiency, E, for the entire bed may be calculated.

A simple statistical approach is used. The probability for

a particle of escaping the first collector is given by (1—n i) and

the probability of escaping both the first and the second collector

is given by (1-n ,) (1—r| 2) assuming statistical independence i.e.

r|2 is not affected by the presence of the first collector. The

probability of escaping the entire filter with 5 collectors

in-line is then given by

( 1 " E) = ( 1 " " n 2)4 (1)

where the same collision efficiency has been assumed for the last four collectors.

To determine the collision efficiency due to gravity or buoyancy, the velocity difference between fluid and particle, Us, is introduced. Us is Stoke's settling velocity

A p o d2

u- - - r f r

( 2)

When the particles get close to a wall, it will affect their settling velocity. O'Neill (1) calculated the drag force on a sphere in contact with a plane wall in a slow linear shear flow and got a correction factor 1.7009 for the Stokes drag force. Using this correction factor the settling velocity becomes 0.03 5 cm/s. This is strictly valid for spherical particles only. It


was observed that some of the particles had been deformed due to mechanical wear since the same particles were used in all experiments. Some of the particles had a shape rather more like a disc. Particles of this shape will have a lower settling velocity.

If the vertical projection of the collector is Av and the projection normal to the mean particle flow direction is A then (2):

Av'

Us

U„ is the velocity of the particles. The model geometry gives

c / o o / e + u s

Here e is the bed porosity; e = 0.61 for the cylindrical collector

bed and e = 0.5 for the bed with quadratic collectors.

To determine the collision efficiency due to direct interception we must take the boundary layer into account. We may derive the influence of boundary layers by calculating the ratio between the collision efficiency for boundary layer flow and the collision efficiency due to potential flow as a function of a dimensionless parameter (Reynolds number). This ratio is plotted on Figure 2 as discrete values. (In order to calculate the ratio for any velocity value, a linear interpolation between the calculated discrete values is used) . The filtration efficiency resulting from these calculations is given in Figure 5. It is zero at zero velocity and rises to approximately 0.18 at 1.2 cm/s.

In Figure 5 we have plotted the filtration efficiency due to the sum of the two collision efficiencies, i.e.

This way of summing the mechanisms gives a too high filtration efficiency because implicitly some particles are removed by both mechanisms, i.e. twice. Also the calculations do not take into account that there is a separation zone behind the collectors that would give rise to increased residence times, and thereby increased settling.

The influence of boundary layers

Above we assumed potential flow which is valid only at Reynolds numbers much greater than 1. A boundary layer will take care of the no-slip condition by reducing the velocity from the velocity, U, in the potential flow in the core to zero at the wall. This boundary layer will affect particle deposition. Schlichting (3) gives the velocity profiles in the boundary layer around both a sphere and a circular cylinder. From this information we calculate the reduction in deposition due to the boundary layer.


Circular cylinder

The ideal velocity distribution in potential irrotational flow past a circular cylinder of radius R and free-stream velocity U* is given by

U(Q) = 2Ua>smO ( 6 )

where the polar angle 0 is zero at the stagnation point. The distance from the surface of the cylinder is

in terms of the dimensionless distance, P c, from the cylinder

surface, v is the kinematic viscosity. Differentiation gives

Introducing the velocity u(0) in the boundary layer and the stream

function ip ( a = •£) we get:

a 1 i2U„

U ~ Z7 * Jf3~ " V

Rewriting Eq. (9) and inserting Eq. (6):

( 9 )

| J = 42vRUm-±-s\nQ ( 1 0 )

Integrating over p c:

ip = yl2y/RUmsinB J ~ d p c - ^2vRU I($ctQ) ( 1 1 )

o

u/U as a function of (3c is tabulated by Schlichting ( 3 , p.

154). I ( P C, 0) has been calculated as a function of 0 for some |3C values shown in Figure 1.

I ( P C, 6) i

s a measure of the volume flow at distance ftc at

different angles 0. We see that a maximum angle ( 0m a x) exists.

0 m ax increases with increasing |3C. The physical implications of

this is that inclusions of different sizes have maxima in removal

by pure interception at different angles. All calculated 0 m ax are less than 90°.

Then the collision efficiency due to interception is

2 £ y ( y = a ) M_ n<

= ZRLU.

( 1 2)

where the boundary layer has been taken into account through the stream function. Potential flow gives


2La(2U„) _ 2a T ] l'

p ot ' 2RLU„ ~~R

( 1 3)

where a is the inclusion radius and L the length of the cylinder.

Inserting Eq. (11) in Eq. (12) and dividing by Eq. (13):

nt / ( P c( a ) , 9 m a x)

T l i . po i Pc(a) ni

/ rl i . p o t i

s plotted in Figure 2 as a function of (3C.

(14)

0 20 40 60 80 100 120

Figure 1 I((BC/ 0) as a function of 0 for some ftc y I2UmR

( = * \ / ~ T - ) values.


0.8

0.7 I--

O

'Jp 0.6

O

C 0.5

a ; o

•i—• 0>

0.4

O 0.3

0.2

0.1

Figure 2 n./Tlt.po.

as a function of f3c ( = ^ y —

WATER MODEL EXPERIMENTS

Description of two-dimensional filter model

A two-dimensional plexiglas model has been built (Figure 3 ) . It has a total length of 0.99 m with an inlet section of 0.12 m to distribute the water and a 0.54 m open channel to establish a uniform flow in front of the filter section. The cross-section available for flow is 0.2 m * 0.2 m. The length, height, and width of the filter section are all 0.2 m. The filter section consists of fifty plexiglas rods placed in a matrix. The centres are 2 cm apart vertically and 4 cm apart horizontally. There are ten horizontal rows of rods each with five rods in depth facing the flow direction. Every second row is displaced 2 cm (see Figure 3) . The rods are fixed to the walls by screws through their centres. The outlet and inlet sections are separated from the section with the filter by perforated plexiglas plates. The perforation consists of holes with diameter 3 mm. The outlet separation plate is positioned 4.5 cm behind the filter section.


Two types of collectors were used. One type was cylindrical rods with diameter 2 cm. The other was quadratic rods with sides 2 cm. The quadratic rods were placed with an edge facing the flow. This means that they somehow acted as cylindrical collectors with protrusions. Both types of collectors were placed with their centres in the same positions. This leads to a porosity of 0.61 for the cylindrical collectors and 0.5 for the quadratic.

I n l e t P l e x i g l a s r o d s F i l t e r O u t l e t

o o o o o o o o o o o o o o o

o o o o o o o o o o

o o o o o o o o o o

o o o o o o o o o o

o o o o o

P e r f o r a t e d p l a t e s

Figure 3 Water model (schematic side view)

Water was used to simulate aluminium melt. Polystyrene particles were employed to simulate inclusions. To attain a particle density as close as possible to the density of the liquid these particles had to be expanded by heating. This is described below.

Water was taken from the water tap and led from the outlet of the model into the drain. Flow volume was measured by a rotameter on the inlet side. Water temperature was approximately 9°C.

Preparation of particles

The "neutrally" buoyant particles were produced from polystyrene particles by heating them in water in a heating cabinet. Before heating their density was about 1040 kq/m

3 so

they settled rapidly in water. After heating a large number of particles, 100 suitable particles were chosen by observing their settling behaviour in a water tank. The particles had a diameter of approximately 1 mm. Measurements showed that the settling velocity was 0.06 cm/s in mean. These particles were used in all the experiments. A very fine masked filter at the outlet of the model ensured that the particles were retained. But even with these precautions some particles were lost and eventually the last experiments were run with only 75 particles.

In order to get good adhesion between particles and collectors, the particles were suspended for half an hour in Magnafloc 292 (Allied Colloids Ltd., Low Moor, Bradford, Yorkshire BD120JZ, England) before each experiment. Observations confirmed that this gave a very good adhesion.

Before introducing the particles into the model they were suspended in water and mixed completely. This suspension was then carefully poured into the model in a position just behind the inlet section dividing plate.


Particle removal - experimental results

Experiments with cylindrical and square collectors were performed. For both collectors several tests with (approach) velocities ranging from nearly zero to slightly above 1 cm/s were run. Every test was run twice. For each test the following data were recorded:

- Type of collector.

- Flow velocity.

- Number of particles injected and number of particles "settling" out in front of the filter. The difference gives the number of particles entering the filter.

- Number of particles deposited on each of the fifty rods. The sum for each row is given in (5).

- For all of the square collector experiments the position of the deposited particles on the collectors were noted. This was also done for some of the cylindrical collector experiments.

The filtration efficiency is defined as the number of particles deposited over the number of particles entering the filter. Using this definition the filtration efficiency has been calculated from the experiments in Table I.

It should be noted that the velocity given is the approach (superficial) velocity. Since the porosity for the cylindrical collectors is 0.61 and 0.50 for the square collectors, the interstitial velocities are different. The simplest possible way to estimate the interstitial velocity is to use Dupuit's law:

interstitial velocity = approach velocity divided by porosity

TABLE I Filtration efficiency

Um [cm/s] Cylinder Square

0.03 0.698 0.586 0.900 0.869

0.06 0.581 0.643 0.695 0.820

0.10 0.557 0.514

0.11 0.646 0.671

0.19 0.321 0.316 0.543 0.500

0.37 0.172 0.140 0.171 0.192

0.56 0.047 0.070 0.090 0. 096

0.67 0.133 0.073 0.038 0. 110

0.74 0.069 0.159 0.136 0.122

0.82 0.250 0.162 0.136 0.181

0.89 0.205 0.197 0.155 0.145

0.96 0.122 0.167 0.242 0.327

1.03 0.194 0.133 0.224 0.254

1.10 0.222 0.186


This means that flow velocities inside the cylindrical collector filter is only 82% of the flow velocities inside the square collector filter for the same approach velocities.

In Figure 4 the depth distribution of particles is shown. To produce this figure the number of particles deposited on rows number 1,2,..,10 for all experiments have been added. Then these numbers have been summed up to give a total number of deposited particles (square: 537, cylinder: 483). The mean percentage deposited on each row has then been calculated. The distributions seem to be exponential functions of distance through filter.

25

[%]

20

0

A Square col lectors ; a ! •

Cylindrical col lectors i g [_

O

A O

" ° O

o - A

A O

o 0

o

- -cr -

-

A ^

0 2 4 6 8 10

Row

Figure 4 Depth distribution of particles. Percentage of particles on each row.

As mentioned above, the position of the deposited particles was registered for all experiments with square collectors (24) and for the last 10 (of 26) of the experiments with cylindrical collectors. The following information can be extracted:

- The number of particles over and under the collectors.

- The number of particles in front and behind the collectors.

- For the square collectors it is also of interest to know whether the particles were sitting on an edge or not.


- Two sides of the square collectors were rough and two were smooth i.e., every second face of the collector were smooth. The first row had a rough side upwards facing the flow and the second row a smooth side upwards. This pattern was continued throughout the filter. The number of particles on rough sides versus the number on smooth sides have been counted.

This information has been tabulated in Table II. All numbers are mean percentages over all experiments. The percentages for square and cylindrical collectors individually fulfil: (over) +

(under) = 100%, (in front) + (behind) = 100%, (on edge) + (not on edge) = 100%, and (rough) + (smooth) = 100%. Particles on top are counted as behind.

TABLE II Distribution of particles on collectors

Position Cylinder [%] Square [%]

over / under 84.7 / 15.3 90.4 / 9.6

in front / behind 41.6 / 58.4 34.6 / 65.4

on edge / not on edge - 31.8 / 68.2

rough / smooth - 50.6 / 49.4

The roughness of the collector surface has no measurable effect on deposition. A major part of the inclusions deposit on the top side of the collectors due to their being slightly heavier than the carrier fluid.

A larger number of particles deposit at the back of the collectors than at the front. An explanation for this is that there is a separation zone at the back of the collectors. In this separation zone the particle residence time over the collector increases and thereby enhances the probability of deposition due to settling.

For both square and cylindrical collectors it was found that the flow was laminar at the lower flow velocities and developed into a flow that could be described as turbulent. However there is no evidence that the flow is not just an unstable laminar transition flow regime.

Figure 5 shows the filtration efficiency, E, for these particles as a function of approach velocity v. Figure 5 also shows the filtration efficiency as a function of Reynolds number based on interstitial velocity, i.e.

Rq = v

It is seen that filtration efficiency drops with velocity in the lower velocity range below 0.6 cm/s. For this velocity range the filtration mechanism may be explained by buoyancy effects -in spite of the use of particles very close to neutrally buoyant. Using v = 0.035 cm/s gives the "exponentially" decreasing filtration efficiency curve shown in Figure 5 for cylindrical


collectors. The increasing section of the filtration efficiency curve is the theoretical curve for removal by direct interception, where the effect of the boundary layer has been taken into account. On Figure 5 also the measured filtration efficiencies for cylindrical collectors and collectors with a quadratic cross-section are shown. It is seen that the measurements lie below the calculated curve.

Re (cylinder) 0 6 8 1 3 3 2 0 0 2 6 7 3 3 3

Re (square) 0 8 0 1 6 0 2 4 0 3 2 0 4 0 0

4 0 0

4 8 0

0 . 2 0 . 4 0 . 6 0 . 8 1

A p p r o a c h v e l o c i t y [ c m / s ] 1 . 2

Figure 5 Filtration efficiency versus approach velocity and Reynolds numbers.

In the higher velocity range there is a transition to unsteady flow. This was observed from colour tracer experiments. In this velocity region the interception mechanism starts to play a role. Due to the decreasing boundary layer thickness the collision efficiency increases with increasing velocity.


For the lower velocity region it is seen that the filtration efficiency is very high at low velocities. The explanation is that the particles have a very small, but still finite settling velocity. Thus at velocities of 0.1 cm/s removal is impressive in spite of the use of near-buoyant particles. According to Netter and Conti (4) most industrial units operate in the velocity range 0.03 - 1.00 cm/s (approach velocity) . Our water model experiments indicate that in the lower half of this velocity range removal by settling will be the dominant mechanism while in the higher half removal by interception may be the dominant mechanism.

In Figure 5 we see that there is a minimum in the measured filtration efficiency curve at approximately 0.5 cm/s. Our theory does not foresee such a minimum. The explanation seems to be that this is an effect of the symmetry of the water model. For a certain velocity range it may give rise to a situation where some particles that should have been removed by sedimentation miss the collectors repeatedly. Such an explanation should also give rise to a smaller minimum at higher velocities that are multiples of the first. For the cylindrical collectors a tendency for this appears at approximately 1.0 cm/s.

An example of this effect is given. Let the approach velocity be 0.5 cm/s. With a porosity of 0.5 the interstitial velocity is 1.0 cm/s according to Dupuit's law. If a particle is exactly between the two first collectors on the line through their centres and settles with a velocity of 0.06 cm/s it will have settled a distance of 1.0 cm when it comes to the last collector. This last collector is exactly 1 cm lower so that the particle barely passes above it.

It is also observed that the square collectors give the highest filtration efficiencies at low flow velocities. This might be an effect of the sharp edges of the square collectors. They give rise to a separation zone where settling will be increased. On the other hand it should be noted that the adjusted standard error is larger for the square collectors than for the cylinders (see Frisvold (5)).

CONCLUSIONS

The water model experiments emphasize the importance of the settling mechanism - especially at low velocities - in industrial bed filters. The explanation for this effect is found in the short settling distances and the long residence times in such filters. To determine settling velocities, it is neccesary to take into account a correction factor due to wall effects.

At slightly higher velocities the interception mechanism dominates. Here one must consider the effect of the boundary layer around the collector.

ACKNOWLEDGMENTS

The authors wish to acknowledge the financial support of this work by the Royal Norwegian Council for Scientific and Industrial Research under project number MT10. 01.18513 and by Hydro Aluminium a.s and Elkem Aluminium ANS.


REFERENCES

1. M. E. O'Neill, "A sphere in contact with a plane wall in a slow linear shear flow", Chem. Enana. Sci.. Vol. 23, 1968, 1293-1298

2. T. A. Engh, Principles of Metal Refining. Oxford University Press, Oxford, England, To be published

3. H. Schlichting, Boundary Laver Theory. Sixth Edition, McGraw-Hill Book Company, New York, USA, 1968

4. P. Netter and C. Conti, "Efficiency of industrial filters for molten metal treatment evaluation of filtration process model". Light Metals 1986. Vol. 2, 1986, 847-860

5. F. Frisvold, "Filtration of aluminium - theory, mechanisms, and experiments", Dr.ing. Thesis, The Norwegian Institute of Technology, 1990

353

Thermodynamic considerations of aluminum refining and fluxing

T. Utigard Department of Metallurgy and Materials Science, University of Toronto, Toronto, Ontario, Canada

ABSTRACT

The free energy of formation for several compounds encountered during aluminum refining and fluxing, is compared at 1000 K. Temkin's model for molten salts shows that when a flux is added to a molten salt mixture, a series of cation-anion salt combinations have to be considered. A molten NaCl-KCl flux has no tendency to react with metallic aluminum. The use of metal fluorides leads to enhanced interactions with liquid aluminum and aluminum oxide.

Based on surface and interfacial tension considerations, the removal of the aluminum oxide skin on the surface of aluminum particles becomes increasing favourable with the addition of salts such as NaF and KF. However, this removal gets more difficult for aluminum contaminated with surface active impurities such as Mg, Na, Li and K, leading to a reduction in metal recovery.

The solubility of A 1203 in NaCl-KCl is insignificant at 1000 K, retarding the dissolution and coalescence processes. However, the addition of Na3AlF6, CaF2 or NaF leads to increased alumina solubility and improved coalescence.

1.0 INTRODUCTION

Over the last 25 years, the recycling of consumer aluminum has grown significantly and billions of beverage cans are recycled annually. The aluminum scrap is normally remelted under a cover of molten salts in order to prevent oxidation of the metal and to enhance the coalescence of the molten aluminum. The flux should also dissolve and remove any unwanted material from the aluminum metal. Most fluxes are made up from a mixture of sodium chloride and potassium chloride with the addition of small amounts of fluoride compounds such as NaF, CaF2 and Na3AlF6 to promote the coalescence and melting process. Some of the important properties of the fluxes have recently been reviewed by Peterson(1) and by Ho and Sahai(2). Peterson(1) noted that there are over 120 non-patented and 30 patented fluxes listed.

In spite of this commercial success, there are still problems with the remelting and fluxing process. Sully et el(3) noted that during remelting of aluminum alloys, the flux becomes opaque, dense and viscous and with time it has to be replaced with a new flux.


West(4) explained this by the formation of a suspension of metal oxides in the chloride based flux. Johnston and Peterson(5) found that magnesium impurities makes the recovery of recycled aluminum more difficult. They attributed this to the formation of a MgO*Al203 spinel, increasing the density and the viscosity of the flux. The cleaning and/or disposal of this flux is both costly and environmentally problematic, promoting research and development into new dross treatment processes(6,7).

The objective of this investigation is to review and analyze the thermodynamic and interfacial tension phenomena affecting the refining of aluminum with molten salt fluxes.

2.0 THERMODYNAMICS OF MOLTEN SALT MIXTURES

When a salt such as2 +CaF2 +is added to a molten mixture of NaCl and KC1, the cations(Ca

+, Na

+, K

+) and the anions(Cl" and F ) are free

to move around in the melt. It is common to separate the melt into a cation lattice and an anion lattice(Temkin

1s model), and the

activity of any salt(assuming ideal solution) is given by its cation mole fraction multiplied by its anion mole fraction. To illustrate this, the activity of CaF2 is given by the following expression:

< 3 C aF ^Ca__ x ( )2 E q> 1 2 XCa

+ XNa

+ XK

XF

+ XCl

where xCa is the mole fraction of Ca

2+ within the cation lattice.

As an example, when small amounts of CaF2 is added to a 50wt% NaCl - 50wt% KC1 mixture the following salt species have to be considered:

NaCl, KC1, CaCl2, NaF, KF, and CaF2

By assuming ideality, the activity decreases in the following order:

aN a C l ~

aK C l >

aN a F ~

aK F ~

aC a C l 2

> aC a F 2

It is interesting to note that the compound added [CaF2] has the lowest activity in the melt, and any reaction between the aluminum and the melt is more likely caused by KF and NaF. With the addition of compounds such as A 1 F 3 and MgCl2, it becomes more complex due to the formation of stable complexes such as A I F 4 and MgCrV . To make a complete analysis of the reactions between liquid Al and a salt mixture, all species have to be considered and their activities determined.

3.0 THERMODYNAMICS OF PURE MOLTEN SALTS

The Gibbs energy of formation of the species considered in this review is given in Figure 1. The corresponding decomposition voltages are shown in Figure 2. With a few exceptions, the thermodynamic stability and the decomposition voltage decrease in the following order:


0

Ca Ba LI Sr Mg Na K Al Si Mn Zn Fe Cu Metal

Sulphide Oxide Chloride Fluoride

Figure 1. A comparison of the Gibbs energy of formation of several sulphides, oxides, chlorides and fluorides. The data are given at 1000 K, per mole of S, 0, Cl2, and F2, respectively.

° Ca Ba Li Sr Mg Na K Al Si Mn Zn Fe Cu Metal

Sulphide —+— Oxide Chloride - e - Fluoride

Figure 2. Reversible decomposition voltage of pure metal sulphides, metal oxides, metal chlorides and metal fluorides at 1000 K.

355


fluorides > chlorides > oxides > sulphides

Among the key exceptions, are the industrially important substances A 1203 and MgCl2. Due to the extreme stability of A1203, only a fluoride based 'inert

1 electrolyte can be used. The decomposition

voltage of alumina is significantly lower than that of the metal fluorides. Because of the low stability of MgCl2 relative to the alkali chlorides, a NaCl-KCl based •inert

1 electrolyte can be used

for the electrolytic production of magnesium from MgCl2.

As shown in Figure 3, an alkali/alkali-earth chloride electrolyte has no tendency to react with aluminum. The corresponding metal fluoride electrolyte is slightly more reactive. A chloride electrolyte is therefore suitable for the refining of aluminum since it will promote the removal of alkali/alkali-earth metal impurities while maintaining high aluminum recovery. The removal of other impurities such as Zn, Si, Fe, and Cu by chlorine or fluorine treatment is significantly more difficult.

As demonstrated in Figure 4, magnesium impurities can not be removed from aluminum using an alkali chloride mixture. On the other hand, A1C13 and SiCl^ promote the removal of magnesium from the metal. By adding NaF and/or KF to the chloride flux, the removal of Mg from aluminum scrap is enhanced. However, this leads to the contamination of the aluminum with Na and/or K as given by the following exchange reaction:

2NaF + Mg = 2Na + MgF2, K-eq = 2.74, 1000 K Eq. 2

Assuming aN aF = 0.05, aM g F2 = 0.01, activity coefficients of Na and Mg in aluminum are 426 and 0.15, respectively, the sodium content in the molten aluminum depends on the magnesium content as follows

Na-ppm = 125 * [%Mg in A l ]

1 / 2, [fluoride flux] Eq. 3

This shows that as long as there are fluorides present in the flux and magnesium in the metal, the removal of Na is futile(8). With a chloride flux, the following exchange reaction controls the Na and the Mg contents in the aluminum:

2NaCl + Mg = 2Na + MgCl2, K-eq = 7.4*10'

9, 1000 K Eq. 4

Assuming aN a Cl = 0.5, aM g C l2 = 0.1, activity coefficients of Na and Mg in aluminum are 426 and 0.15, respectively, the sodium content in the molten aluminum is given by:

Na-ppm = 0.012 * [%Mg in A l ]

1 / 2, [chloride flux] Eq. 5

4.0 ALUMINUM OXIDE REMOVAL

All aluminum scrap, when added to a remelting furnace, is covered by a stable surface skin of aluminum oxide. In the subsequent analysis, it is assumed that the removal of this oxide layer is required before the molten aluminum particles can coalesce. The dissolution/removal of this oxide layer is therefore of paramount importance in achieving high metal recovery.


40

- 4 0 — i 1 1 1 1 1 r — 1 1 1 1 1 —

Li Na K Mg Ca Sr Ba Cu Mn Zn Fe Si Metal

Figure 3. Exchange equilibrium between aluminum and different metal chlorides and metal fluorides at 1000 K.

Metal Compound

— h — Oxide Chloride Fluoride

Figure 4. Exchange equilibrium between magnesium impurities in aluminum and different metal oxides, metal chlorides and metal fluorides at 1000 K.

Chloride Fluoride

Al + 3MeX = AIX3 + 3Me

Log

K-eq

uilibr

ium

MeX + Mg = MgX + Me-

Log[

K-e

quilib

rium

]


4.1 Interfacial tension of aluminum in molten salts

The surface tension of aluminum has been measured by several authors and using the recommended va^lue from Turkdogan(9), the surface tension at 1000 K is 890 mJ/m . Since it is impossible to remove the oxide skin from an aluminum drop by a reducing gas, it is thought that this surface tension value corresponds to that of an aluminum drop with a mono-layer thick aluminum oxide skin on the surface(10).

For the purpose of this analysis, it is assumed that the oxide skin removal takes place as follows(see Figure 5 ) :

1. The oxide skin is removed from the aluminum drop and forms a single alumina particle. The aluminum oxide/flux interfacial area decreases from being equal to the surface area of the aluminum drop to approximately zero. As an example, a 5 mm diameter drop covered with a 100 A thick oxide layer will form an alumina particle with a surface area 3000 times smaller than that of the aluminum drop.

2. This exposes the aluminum to the flux with an aluminum/flux interfacial area equal to the initial area of the drop.

Per unit area, the change in the Gibbs energy of this process is:

A G = C T A l / F l ux - ( aA l / A l 2 03 + <7A1203/Flux)

= ^ A l / F l u x - <JAl + c j F I u x* c o s ( G ) Eq. 6

where aAl is the surface tension of aluminum covered with a mono-layer of aluminum oxide and 9 is the contact angle the liquid flux forms on an alumina substrate. Contact angle determinations(11) of chloride fluxes on alumina substrates showed that these fluxes wet the alumina well and contact angles less than 15° were formed. Therefore, per unit area of aluminum, the equation can be approximated as follows:

A G = aA l / F l ux + aF l ux - aAl Eq. 7

For calculation purposes, it is assumed that for small changes in the flux composition, the flux surface tension changes insignificantly as compared with the Al/flux interfacial tension,. When the surface tension of the flux is assumed to be 110 mJ/m , the equations simplifies to:

A G(mJ/m

2) = aA l / F l ux + 110 - c r Al Eq. 8

4.1.1 Pure Aluminum

If the presence of a surface oxide skin prevents impurities to contact the metallic aluminum, then the drop will remain of high purity as long as it is protected by its oxide skin. Therefore, the surface tension of the aluminum remains constant(890 mJ/m ) and the equation simplifies to:

A G(mJ/m

2) = < J A L / F L UX - 7 8 0 Eq. 9


Since the Al/Flux interfacial tension is always less or equal to 780 mJ/m (2,13,15), the removal of the oxide skin is favourable for all the flux systems. However, the addition of fluorides(NaF/KF) decreases the interfacial tension the most and promotes most strongly the oxide skin removal. It must be remembered that the addition of CaF2(and also other fluoride containing salts) leads to a decreased interfacial tension due to the increased NaF and KF activities as outlined in section 2.0, even if Ca by itself is not surface active.

4.1.2 Impure Aluminum

Davies et al(12) found tha^ the aluminum surface tension dropped approximately 150-200 mJ/m with the addition of sodium to the metal. Figure 6 illustrates the decrease in interfacial tension between aluminum and fluoride containing salts(13). For used beverage can alloys(1.91% Mg, 0.79% Mn, 0.03%Si and 0.04% F e ) , it is shown that the interfacial tension between these alloys and a pure NaCl-KCl melt is 150-200 mJ/m lower than the corresponding interfacial tension for pure aluminium(2) .

Since the surface tension of Al with known amounts of Na and K is not known, a surface tension model(14-17) had to be applied to analyze Eq. 8. Figure 7 shows the effect of Pb, Bi and Zn on the surface tension of aluminum with the solid lines representing model calculations(16,17). Using this model, the surface tension of Al and the interfacial tension between Al and NaF-KF melts were calculated as a function of the logarithm of the sodium and potassium activity(Fig. 8) . It was assumed that the sodium and potassium activities were always equal. It is interesting to note that the surface tension and the interfacial tension decrease at approximately the same rate with increasing sodium and potassium activity. However, for Na and K activities above approx. 10" , the difference between the surface tension?and the interfacial tension decreases from approx. 180 to 100 mJ/m .

According to Eq. 8, the removal of the oxide skin is thermodynamic feasible for pure aluminum with a sodium and potassium activity less than 0.01. However, at higher sodium and potassium contents, the Gibbs energy change approaches zero and then becomes positive. This may explain the increased aluminum losses under industrial conditions with impure recycled aluminum and contaminated fluxes.

4.2 Alumina Solubility in Molten Fluxes

In addition to physical removal, the aluminum oxide can also be removed by chemical dissolution. However, molten chloride fluxes have a very low solubility for oxides. As shown in Figure 9, the solubility of A 1203 in cryolite decreases rapidly with increasing NaCl contents and decreasing temperatures. Using neutron activation, Haupin(18) found that in pure LiCl at 700 °C the maximum solubility of oxygen was 0.051 wt%. With increasing additions of A1C13, the oxygen solubility increased to 0.284 wt% for 25 wt% A1C13. Peterson(1) noted that among the chlorides, only LiCl and A1C13 promoted coalescence, which he attributed to the mixing induced by the gas release.


AIR

Figure 5. Schematics of proposed mechanism for the removal of an A1203 skin from the surface of an aluminum drop in a molten flux.

(NoCI-KCH.-f 2, . ,

' £ 1 £ 1 2*> Mol *h f

Figure 6. Interfacial tension between Al and a KCl-NaCl based flux with the additions of metal fluorides(13).

360

Aiaoa

AI203-

0 No,AlFj

o UF

4 KURCrruMOV ttal


1 I . I

Zn-Pb 760 K

| S

I 600

J? 400|

* 200

Zn-Bi

i \ |

J 1 • l

x X J 760 K

J 06 1.2 IJ

Mol Percent Bi 2.4 )

W)

B i

g U7 n PA ° ^ ^

a tn

(?

e xP

e r i r o e n t* l surface tension of Al-Pb, Al-Bi, Zn-Pb, and Zn-Bi alloys(16).

900

-4 -3 Log(a-Na & a-K)

l n t- E n e r

9 V Surf. Energy Surf. - Int. Energy

Figure 8. Calculated aluminum surface tension and aluminum-flux IntUtt* tension as a function of sodium and potassium a^tw^ 1 G S; F <

^

c a l c u l a t i° n purposes, it was assumed that the

activity of sodium and potassium were equal at all tiroes, and that the surface tension of the flux was 110 mJ/m\

Mot Percent Pt> Md Percent Bi

Surface Tension (mN/mj

Surface Tension (mN/m)

Al-Pb 117J K Al-Bi »73 K

Sur

face

Ene

rgy(

mJ/

m ~

2)


One important aspect of adding fluorides to the chloride flux is to enhance the solubility of alumina in the flux. Using solubility data for Na3AlF6, CaF2 and NaF at high temperatures, the solubility of alumina at lower temperatures was estimated using Van H o f f

!s

equation. The resulting solubilities are given in Fig. 10 as a function of temperature for the hypothetical pure liquid fluorides. However, without experimental data, it is difficult to predict the actual enhanced solubility by fluoride additions at these low temperatures.

As demonstrated in Figure 11, alkali chloride electrolytes have no tendency to react chemically with the aluminium oxide. The free energy change for the following reaction is highly positive:

A 1203 + 6MeCl -> 2A1C13 + 3Me20 Eq. 10

The only exception is SiCl4 for which the equilibrium constant approaches unity.

Figure 11 shows that fluoride based electrolytes have a somewhat greater tendency to react chemically with the aluminium oxide:

A 1203 + 6MeF -> 2A1F3 + 3Me20 Eq. 11

This reaction is driven further to the right because the activity of A1F3 is lowered by the formation of cryolite.

A1F3 + 3MeF -> Me3AlF6, where Me = Li, Na, or K Eq. 12

This analysis is also supported by the experimental work carried out by Peterson (1) , who concluded that fluoride salts are much more effective than chloride salts in promoting coalescence.

5.0 SUMMARY

The thermodynamics of aluminum-flux exchange reactions during refining and recycling of aluminum have been reviewed and discussed. It has been demonstrated that a chloride based flux is well suited for the purpose of protecting the aluminum from oxidation and to maintain a high purity metal product. However, due to the inertness of aluminum oxide relative to alkali chlorides, the coalescence and recovery of small aluminum particles are retarded.

The addition of fluoride salts such as Na3AlF6, KF and NaF enhances the removal of A 1203 due to i) favourable thermodynamics, ii) interfacial tension changes, and finally iii) increasing alumina solubility. However, fluoride additions lead to detrimental effects such as increased Na and K contents in the refined metal. Additionally, the removal of Na from aluminum containing Mg is impossible by the use of a fluoride based flux.

E X T R A C T I O N , R E F I N I N G A N D F A B R I C A T I O N O F L I G H T M E T A L S

9 0 0

Alumina Solubility Estimated using Van Hoffs Equation

9 5 0 1000 1 0 5 0 1 1 0 0 1 1 5 0 T e m p e r a t u r e [K]

1 2 0 0 1 2 5 0 1 3 0 0

Na3AIF6 C a F 2 N a F

Figure 10. Estimated solubilities of A1203 in pure hypothetical liquid Na3AlFf, CaF2 and NaF as a function of temperature. The calculations involved the use of Van H o f f s equation and were based on high temperature solubility data.

363

No3AIF6 ( 1 0 0 9 ° ) Wt.% NoCI

20 40 60

7 3 4* 80

Figure 9. Liquidus lines in the Na3AlF6 - NaCl - A1203 system.

mo!

e%

AI2

03


1/3AI203 + 2MeX = 2/3AIX3 + Me20

K Na Ba Li Sr Ca Cu Mn Zn Fe Mg Si Metal

Chloride —+— Fluoride

Figure 11. Exchange equilibrium between aluminum oxide different metal chlorides and metal fluorides at iSoo K?

364

Log

K-eq

uilib

rium


REFERENCES

1. R.D. Peterson; Second International Symposium - Recycling of Metals and Engineering Materials, TMS, 1990, pp. 69-84.

2. F.K. Ho and Y. Sahai; Second International Symposium -Recycling of Metals and Engineering Materials, TMS, 1990, pp. 85-103.

3. A.H. Sully, K.K. Hardy and T.J. Heal; Journal of the Institute of Metals, V82, 1952, pp. 49-58.

4. E.G. West; Trans. Inst. Weld., V3, 1940, p. 93. 5. T.J. Johnston and R.D. Peterson; First International Symposium

- Recycling and Secondary Recovery of Metals, TMS, 1985, pp. 417-28.

6. S. Lavoie, C. Dube, and G. Dube; Second International Symposium - Recycling of Metals and Engineering Materials, TMS, 1990, pp. 451-62.

7. S. Lavoie, C. Dube, and G. Dube; Light Metals, 1991, pp. 981-85.

8. F. Patak; Ph.D. Thesis, Rheinisch-Westfalischen Technischen Hochschule, Aachen, West Germany, 1983.

9. E.T. Turkdogan; "Physical Chemistry of High Temperature Technology", Academic Press, 1980, p. 94.

10. C. Garcia-Cordovilla, E. Louis and A. Pamies; Journal of Materials Science, V21, 1986, pp. 2787-92.

11. T.A. Utigard, D-C. Mo and J.M. Toguri; Journal of Chemical Engineering Data, V31, 1986, pp. 383-85.

12. V.L. Davies and J.M. West; Journal of the Institute of Metals, V92, 1963, pp. 208-10.

13. L. Martin-Garin, A. Dinet and J.M. Hicter; Journal of Materials Science, V14, 1979, pp. 2366-72.

14. T.A. Utigard and J.M. Toguri; Metall. Trans. B., V16B, 1985, pp. 333-38.

15. T.A. Utigard, J.M. Toguri and K. Grjotheim; Canadian Metall. Quart., V26, 1987, pp. 129-35.

16. T.A. Utigard and J.M. Toguri; Metall. Trans. B., V18B, 1987, pp. 695-702.

17. T.A. Utigard; Ph.D. Thesis, University of Toronto, Toronto, Canada, 1985.

18. W.E. Haupin; Light Metals, 1979, pp. 353-61.

365

367

3D-simulation of the thermal performance of a coke calcining kiln

R.T. Bui, G. Simard, Y. Kocaefe, A. Charette, M. Lacroix, S. Jain Universite du Quebec a Chicoutimi, Chicoutimi, Quebec, Canada

J. Perron Alcan International Ltd., Jonquiere, Quebec, Canada

A. Proulx Alcan Smelters and Chemicals, Joinquiere, Quebec, Canada

P. Barr The University of British Columbia, Vancouver, British Columbia, Canada

Abstract

This is the first time ever a fully three-dimensional model of the rotary coke calcining kiln is built, to include all the important phenomena occurring therein. The overall kiln model consists of two separate models, one for the freeboard gas and one for the coke bed, coupled together with intermittent information exchange. The phenomena treated by the model include heat transfer, fluid flow, turbulence, volatiles combustion, third air, effect of kiln rotation. The model, based on the general equations of conservation is solved with the help of the CFD general-purpose code PHOENICS. Due to the huge dimensions of the real kiln, the model is validated on a laboratory size pilot kiln. Results are presented and the potential use of the model is discussed.

Keywords

Coke calciner, rotary kiln, mathematical model, heat transfer.

Introduction

Green petroleum coke must be calcined before being used for electrode manufacturing. Mois-ture and volatiles are removed from the coke during calcination in order to prevent cracking due to shrinkage in the subsequent baking of the electrodes. Calcination is carried out in large rotary kilns which act as counter-current heat exchangers. A typical kiln is about 60 meters in length and 2.5 meters in diameter. Quality of calcined coke is measured by its density and crystalline length, both of which depend strongly on the temperature history of the coke during calcination. The process is shown schematically in Figure 1.

There are many processes occurring in the kiln. After crushing, coke is fed to the kiln through the feed chute. The kiln is slightly tilted (2 to 4 %) and this inclination combined with the rotary action provides the coke movement along the kiln. During calcination, first the moisture and then the volatiles are driven out of the coke as it is heated from room temperature to approximately 1250° C. Also coke undergoes changes in its crystalline structure, and consequently the real density increases from 1400 kg/m

3 for green coke to about 2050 kg/m

3

for calcined coke.

— All correspondence should be addressed to R.T. Bui, Departement des sciences appliquees, University du Quebec a Chicoutimi, Chicoutimi, Quebec, Canada, G7H 2B1


The energy required for the calcination is provided by the partial combustion of volatiles. At the start-up of the kiln the burner remains on until the system reaches the state of energy self-sufficiency. Then only the auxiliary (third) air is fed to the kiln to ensure the combustion of the volatiles under normal operating conditions. Also the coke dust generated by the solid bed along with the fine particles in the feed are entrained by the gas. These fines are burned partially in the solid bed and in the freeboard gas, and the remaining part leaves with the exhaust gas.

The coke encounters different zones as it moves along the kiln. First is the drying zone where the moisture is removed from the coke. Further down the kiln the coke reaches the devolatilization and calcination zone. In this part of the kiln the temperature increases up to 1250° C where the combustion takes place. Also the third air is blown into the kiln in this section. The last part is the calcined-coke zone where the coke starts to cool down. This part also plays an important role in stabilizing the solid bed flow since the calcination zone may move up and down the kiln depending on the operating conditions of the kiln.

To insure coke quality and efficient production, the calcination temperature and coke heating rate should be properly controlled during operation. Difficulties encountered in control are due to the complex nature of the process and a lack of knowledge on the interaction between the controlling variables (gas flow, third air, rotation, kiln inclination, coke flowrate, granulometry, etc.) and the process variables (calcination temperature, heating rate, crystalline structure, real density, coke-dust generation, production rate, etc.). Presently, a three-dimensional model is being developed in order to be able to study such a complicated process and to better understand the interaction between various phenomena occurring in the system.

There are many models available in the literature for cement kilns, titanium dioxide kilns, alumina kilns and drying kilns (e.g. references 1, 2, 3). Also there are studies on various

Figure 1- The coke calcination process.

i WATER

COOLER

- CALCINED COKE

DUST

DUST

THIRD A/R

NATURAL GAS

GAS~

COKE

STACH

INC

INE

RA

TO

R

AIR

SECONDARY AIR -

PRIMARY AIR -

FEEDER

GREEN

k COKE


aspects of calcination kiln modelling such as dust combustion (4) and structural changes in the coke during calcination (5). As for a comprehensive modelling of the kiln, work has been done so far in one dimension only, the dimension being the kiln axis. Even in one dimension, most models built in the past ignored some or several of the basic phenomena prevailing in the process. To date, the most comprehensive one-dimensional model of the coke calcining kiln is found in the work of Perron et al. (6). The current project was undertaken to develop a three-dimensional model incorporating all the physical aspects of coke calcination.

The model is described in the following section. As a first test, it was used for the simulation of the thermal performance of a pilot kiln for which experimental data are available. Description of the pilot kiln, its simulation and the results which compare the model predictions with the experimental data are presented in the subsequent sections.

The model

The coke bed and the freeboard gas exhibit entirely different characteristics. There are many advantages in developing a separate model for each part with a proper interface between them. This structure is not only more manageable but also allows the use of each model exclusively for the solution of partial design or operational problems. Since there are considerable differences in the values of the physical properties of the two media, handling of each part by a separate model eliminates the numerical problems that originate from such differences. Also by using parallel computing technique these models can be run simultaneously on separate CPU's, and the interaction between them can be established by intermittent exchange of data or information during the run. This reduces the real computation time significantly. Due to these reasons, the calcination kiln was modelled in two parts: the coke bed and the freeboard gas.

Both models are based on the solution of the fundamental partial differential equations gov-erning the system. For this purpose, the PHOENICS code was used. PHOENICS solves the general conservation equation using the finite-volume technique. For details of the numerical solution technique the reader is referred to Rosten and Spalding (7).

In the coke bed the flowfield and the heat transfer are solved by PHOENICS. Three velocity components and the pressure field are determined for flow. The coke bed undergoes different modes of flow along the kiln. In general rolling is the dominant mode which is characterized by a core inside and the particles moving around this core in a crescent-shaped path. In the zone where there is significant volatile evolution (around the third air entry), fluidization occurs. Presently the entire bed flow is considered laminar and calculated in the rolling mode. The bed is divided into two layers: a thin top layer which behaves like a light fluid and the bottom layer in which the flow resembles that of a viscous fluid. This is simulated by assigning a low viscosity to the top layer and a viscosity three orders of magnitude larger to the bottom layer. This will further be discussed in the results section. For heat transfer both conduction and convection are considered. An effective thermal conductivity is calculated for the coke bed.

In the freeboard gas, PHOENICS solves for the momentum, heat and mass transfer. The flow is turbulent, and k-e model is used for turbulence. For flow, the three velocity components, the pressure field and the turbulence parameters (k and e) are calculated. Because of the high temperatures radiation is the most important mode of heat transfer in the kiln. The heat transfer calculation includes conduction, convection and radiation. Presently, the six-flux method is being used for the analysis of radiative heat transfer. The kiln length is very much greater than the diameter, and the kiln can be assumed as a long furnace in which case the flux method gives good results. For the simulation of combustion, two options are available. In the first option,


combustion pattern can be specified directly, that is, the heat release by combustion is entered as source terms in the enthalpy equation in the zones where the combustion is considered to take place. In the second option, this can be calculated based on a single-step overall reaction with diffusion as the controlling mechanism.

For both the coke bed and the freeboard gas, heat loss through the wall is calculated from a one-dimensional analysis in the radial direction including the conduction through the refractory, natural convection and radiation outside the kiln shell.

The coke bed and the freeboard gas models are linked by means of the heat and mass balance at the interface. The moisture, volatiles and coke dust generated in the bed enter the freeboard gas and heat is transferred from gas to bed through this interface. During the execution of the computer program, the data between the two models are exchanged at certain time intervals in order to update the conditions at the interface, which serve as the boundary conditions for both parts.

Each model (coke bed or freeboard gas) consists of two sub-models: one for flow and one for energy (plus combustion for the freeboard gas). Therefore, the overall kiln model is made up of four sub-models. Such a modular construction of the overall model serves many purposes. While it is possible to solve the energy and the flow simultaneously for each part, this procedure leads to numerical instability and convergence problems due to the highly complex nature of the process and consequently their simulation. However, separate handling of the flow and energy makes it easier to attain numerical stability. Also, the modular structure allows the analysis of the flow or energy exclusively or interactively. These four sub-models are run on four different CPU's and data transfer between them is managed by a control program called Task Manager. This simultaneous handling of the four sub-models reduces the real computation time significantly. The computation time for the overall solution is equal to that of the longest running sub-model. Figure 2 describes the model structure.

Fluid flow including : - burner and tertiary air - moisture , volatiles and dust

from the bed - kiln rotation - turbulence

velocities

temperatures

Energy transfer with : - conduction , convection and

radiation - heat loss through the refractory - combustion

Heat flux distribution on the bed surface

- temperature distribution on the bed surface

- moisture , volatiles and dust generated in the bed

Freeboard gas model

Interface (physically located at the bed surface)

Granular fluid flow with : (in the rolling mode , laminar flow) - kiln rotation - free surface on the bed

velocities Energy transfer with : - conduction and convection - heat loss through the refractory

Granular fluid flow with : (in the rolling mode , laminar flow) - kiln rotation - free surface on the bed

Energy transfer with : - conduction and convection - heat loss through the refractory

Granular fluid flow with : (in the rolling mode , laminar flow) - kiln rotation - free surface on the bed temperatures

Energy transfer with : - conduction and convection - heat loss through the refractory

temperatures

Coke bed model

Flow sub-models Energy sub-models

Figure 2 Structure of the mathematical model showing the interaction between the various submodels.


The pilot kiln of the Center for Metallurgical Process Engineering of the University of British Columbia (UBC) has a length of 5.6 m and an inside diameter of 0.41 m. A schematic diagram is given in Figure 3. Natural gas is used as fuel It is equipped with numerous thermocouples to measure temperatures in the solid, in the gas and within the refractories. Based on these measurements, the thermal performance of the kiln can be evaluated under a given set of operating conditions. During the green petroleum coke calcination, heat transfer is the most important phenomenon which controls and determines the coke quality. For this reason, the thermal performance of this pilot kiln is being simulated in order to evaluate the model.

The kiln is charged with 100 kg/h of solid material. Due to physical limitations and safety reasons, the charge consists of 40 % green coke and 60 % calcined coke. If the solid flowrate is decreased, the bed becomes too shallow in which case the pilot kiln process would no longer be representative of the real kiln process. On the other hand, an increase in the green coke content of the charge results in large amounts of volatiles in the gas in which case a high heat release rate by the combustion causes instability in the operation as well as fire or explosion hazards.

Green coke inlet

Bed depth (.076 m) '

Stack

p.D. = 0.13m ID. = o.10m

I I >L 0 ^ — Refractories and shell

Calcined coke outlet

-5.6 m-

Burner (natural gas and air)

2=r Tertiary air duct

5 » o

Cn S "

n d er

O.D. = 0.61m ID . = o.41m

Figure 3 Schematic representation of the pilot kiln.

The third (or tertiary) air system on the real kiln is simulated by a lance in the pilot kiln. It protrudes 0.46 m into the kiln from the burner end. Also, the burner operates continuously during the operation. Figure 3 shows the lance and its position in the kiln. All the information on the kiln and its operating conditions are given in Table 1.

The pilot kiln


Table 1- Description of the Pilot Kiln and Operating Conditions

Kiln length 5.6 m Kiln diameter:

Outside 0.61 m Inside 0.41 m

Refractory thickness 0.095 m Insulation thickness 0.006355 m Lance position from

the burner 0.46 m Kiln slope 3 degrees Kiln speed 2.5 rpm Thickness of bed 0.076 m Bed fill 10% Volatile release 9 0 % Volatile combustion 37 % of release Feed rate:

Green coke 40kg/h Calcined coke 60 kg/h Total 100 kg/h

Firing rate: Natural gas 5.95 Nm

3/h

Primary air 34 Nm

3/h

Tertiary air 60 Nm

3/h

Simulation

For the simulation of the pilot kiln, the system was divided into 29, 84, 15 grids in the radial, angular and axial directions, respectively, giving a total of 36540 finite volume cells. A large number of divisions was used in the angular direction for a realistic representation of the bed surface. This is essential in the simulation since the physical location of the interface (between the two models) is the bed surface. Non-uniform grid system was used to put a larger number of cells in the regions where steep gradients are encountered for different variables. The number of cells can be increased as desired at the expense of higher computation time.

The numerical parameters of the simulation and the physical and thermal properties of the media are given in Table 2. The kinematic viscosity of the top layer of the bed is assigned a value close to that of the gas in order to simulate its light fluid behaviour, and the kinematic viscocity of the bottom layer is three orders of magnitude higher. The gas radiation is represented by a gray gas model with the average absorption coefficient calculated from the average emissivity in the kiln. As an option a sophisticated model is available which accounts for the concentration variation of different radiating species of gas and of particles.


Table 2- Numerical Parameters of Simulation and Properties of Gas, Coke and Refractories

1-Numerical parameters:

Grid system Radial divisions Angular divisions Axial divisions Total finite volumes

2-Properties of Gas

Density Kinematic viscosity Thermal conductivity Heat capacity (Prandtl number

3-Properties of Coke Bed

Bulk density Thermal conductivity Heat capacity Kinematic viscosity:

Top layer Bottom layer

(Prandtl number: Top layer Bottom layer

4- Properties of Refractories and Surfaces

Refractory thermal conductivity Insulation thermal conductivity Bed surface emissivity Wall emissivities Gray gas absorption coefficient

Non-uniform 29 84 15

36540

1 kg/m

3

1.95 x 10"

5 m

2/s

0.034 W/mK 1075 J/kg K 0.62) 800 kg/m

3

0.4 W/m K 750 J/kg K

1.0 x 10"

4 m

2/s

1.0 x 10"

1 m

2/s

150 150000)

0.3 W/m K 0.15 W/m K 0.9 0.9 0.3 m"

1

Note: Part of the above data on properties is taken from Reference (8).

The four sub-models are run on 4 separate CPU's (all SUN computers) in parallel as discussed earlier. The computational parameters (number of variables solved, number of sweeps, computation time) are listed in Table 3. The total (real) CPU time is slightly over 670 minutes. If the sub-models were run in a sequential manner, it would have taken 1490 minutes. Thus, the parallel computation technique used here offers a savings of 55 % in computation time. As clearly seen in Table 3, the most time-consuming part of the model is the calculation of turbulent flow in the gas. Inclusion of radiation in the freeboard gas energy transfer also requires substantial increase in computation time.


Table 3- Computation Times on SUN Computers Working in Parallel

Model Sub-Model Number of variables

Number of sweeps

700 500

Computation time (min.)

670 448

Freeboard gas

Flow Energy

6 4

Coke bed

Flow Energy

4 1

500 200

282 90

Results and Discussion

The model predicts the three-dimensional velocity, pressure, temperature and radiative heat flux distributions as well as the distributions of turbulence variables k and e in the freeboard gas in the kiln. Figures 4 to 7 show some of these distributions. In general, the bed velocities in the transverse direction are much greater than those in the longitudinal direction, in the order of 10 to 25 times. The average longitudinal velocity is approximately 0.002 m/s in the bed. The situation is opposite in the freeboard gas where the axial velocities are more important than the transverse ones.

In Figure 4, transverse velocity profiles are presented at different cross-sections for gas and bed flow. Figure 4a shows the velocities in the freeboard gas at approximately 0.1 m away from the burner. There is a recirculation zone in this region created by the third air jet, as also seen in Figure 5a. The effect of rotation is negligible and restricted to a small region near the wall. Recirculation dominates the flowfield. Figure 4b also shows the gas velocities at midlength of the kiln, namely at 2.8 m. from the burner. In this section, the effect of rotation is clearly visible and dominant since the end effects (burner at the gas entrance and stack at the gas exit) have negligible influence at this point. Figure 4c shows the flowfield in the coke bed, near the coke discharge end. The cross-sectional velocity profiles are similar at every axial location due to the nature of flow and also due to the assumption of rolling motion for the entire bed. High velocities within the top layer simulates the fast rolling motion of the particles in the cascading layer near the bed surface. The flow in the bottom layer resembles that of a plug flow which results from the viscous behaviour of the coke bed in this region.

Figures 5a and 5b show the longitudinal velocities and temperature contours within the freeboard gas on a vertical plane passing through the kiln centre. Figure 5a displays the view within the first two meters of the kiln (the burner end) and Figure 5b the last two meters (the stack end). Near the burner, the recirculating flow created by the third air jet is clearly seen in Figure 5a, and the highest velocities are obtained near the third air inlet as expected. Volatiles are released in the region close to the third air entry since the volatile combustion occurs here and consequently the coke bed attains high enough temperatures for devolatilization. Naturally, the highest gas temperatures are also obtained in this zone. Gas temperature decreases along the kiln as the heat is transferred from gas to the coke bed and to the refractories. Figure 5b shows the effect of the stack on the velocity field. Also a low temperature central core surrounded by a hotter region is seen in this figure. This is due to the effect of cold high-momentum third air jet injected into the kiln at the centre of the cylinder.

Figure 6 presents a three-dimensional view of the same region shown in Figure 5a, i.e. the first two meters of the kiln from the burner end. The longitudinal velocity profiles and the


Figure 4- Transverse flowfields of gas and coke bed at various cross-sections of kiln. (a) gas flow close to burner end; (b) gas flow at midlength of kiln (c) granular bed flow near coke discharge end, showing a cascading zone and an upward flowing zone.

: 1.1111E+Q0 m/s,

temperature contours again show the recirculation zone and the combustion zone (combustion of fuel and volatiles).

The temperature contours in the coke bed are presented in Figure 7. Coke and gas flow in opposite directions. Coke temperature increases toward the burner end. The bed and gas temperature profiles along the kiln resemble those of a counter-current heat exchanger. As the contours indicate, the gradient is very small in the coke bed, and this was also observed during the experimental work.

Figure 8 compares the predicted and measured gas temperature profiles in the axial direction, both averaged along a circumference of radius 0.1 m centered on the kiln axis. A fairly good agreement is obtained between the model predictions and the experimental data.

In Figure 9, predicted and measured bed temperature profiles in the axial direction are compared.

> : 3.0000E-01 m/s. > : 5.7971E-02 m/s.

(a! (b)

( c )


: 1 . 5 0 0 0 E + 01 m / s .

T

Figure 5 - Gas flowfield and temperature contours (in

0 C) in vertical plane:

(a) burner end; (b) stack end.

- > : 5.0nrjQE+00 m / s

Figure 6-Three-dimensional view showing: (1) axial velocity field at six different cross-sections starting from burner end, and (2) temperature contours in

0 C.

- > : 1 . 5 0 0 0 E + 01 m / s . V ^ )

(a)


227

— > > > ^ > r Figure 7 - Temperature contours of the bed in

0 C.

Both are taken near the bed surface. However, due to the high level of mixing in the transverse direction, the temperature gradient is very small. Therefore each temperature at a given axial location can be considered as the average cross-sectional temperature at that point. The agreement between the predicted and the measured results is very good. Since coke calcination and coke quality mainly depend on the temperature history of the coke bed, accurate calculation of the coke bed temperatures is important for studying the design, operational or control problems.

In Table 4, the coke residence time and the energy balance are presented. Coke residence time is one of the most important factors in the kiln which determine the coke heating rate. The model accurately predicts this parameter as 46 minutes while the experimental value calculated from real volumetric flowrate of the coke, is 45 minutes. The energy imbalance within the model is less than 1 %. The energy transfer from the freeboard gas to the coke bed is calculated as 31 kW. This is very close to the value determined from the experimental measurements which is 29.2 kW including sensible enthalpy of coke, moisture evaporation and calcination.

With the above comparative study of results, it can be reasonably concluded that the model is validated on the pilot kiln. It can be seen that the results obtained from the model (coke bed temperatures, coke residence time, heat transfer, gas temperatures) are reliable. Therefore,

1400.00

800.00

p q 1200.00

ID 1000.00

H < W

CU 600.00

W 400.00

200.00

< 0.00

S I M U L A T I O N

• E X P E R I M E N T A L ]

0.00 1.00 2.00 3.00 4.00 5.00

AXIAL DISTANCE (m) 6.00

Figure 8 - Gas temperatures, simulated and measured, both averaged over a circumference of radius 0.1 m centered on the kiln axis.


( J 1200.00

w Oh

S w

H

Q W PQ

1000.00

800.00

600.00

400.00

200.00

0.00

S I M U L A T I O N

• E X P E R I M E N T A L !

0.00 1.00 2.00 3 .00 4.00 5.00

A X I A L D I S T A N C E ( m ) 6.00

Figure 9 - Bed temperatures, simulated and measured, both near bed surface.

the model can be used as a tool to analyse the various phenomena taking place in a coke calcination kiln for predictive purpose.

Further Work

Due to the large number of interacting phenomena involved in the petroleum coke calcination process, the model is complex and necessarily some of the phenomena had to be treated in an approximate manner, using appropriate correlations or simplifying assumptions. For these

Table 4- Energy Balance for the Kiln and Coke Residence l ime

Input

a

Model Output

b Experimental

0

Energy balance: Combustion of fuel 58 kW Combustion of volatiles 27.3 kW Gas leaving the kiln through stack 24 kW Heat transferred to coke from gas 31 kW 29.2 kW Heat loss through the refractories 31 kW Total 85.3 kW 86 kW

Coke residence time: 46 min 45 min

Note a: Specified in the simulation based on the operating conditions b: Simulation results from the model c: Calculated from measured data.


phenomena (for example dust generation from the bed and dust entrainment by the gas, or non-Newtonian flow of the particulate bed), more basic research is required which indeed will result in an improvement of the model.

On the other hand, the computational work involved is huge. Even by applying the parallel computation technique using four CPU's working simultaneously, and simulating a small-sized pilot kiln instead of the real one, it takes approximately eleven hours for a single simulation run. Therefore, further optimization of the computational aspects is called for, particularly in grid sensitivity, accuracy and computing time.

The model should not wait long to prove its industrial applicability. Therefore the next item on the priority list will be to apply it to the simulation of a real calcining kiln.

Potential Use of the Model

Although more work can be done to improve this model, in its present form it offers interesting possibilities for the study of the coke calcining process. For example, it can be used to study the effect of kiln rotation on gas flow and bed flow, and the effects of changes in third air flowrate or third air position in the kiln. These changes can generate recirculations of the gas, and rearrangements of the gas temperature contours, and consequently can affect the location of the calcining zone. The UBC pilot kiln does not produce dust, but once the model is applied to a real kiln and the dust sub-model activated, it will further add an interesting aspect to the use of the model.

Conclusion

The work presented here is part of a sizeable three-year modelling project involving several workers operating as a team, as evidenced by the long authors' list of this article. Still other workers participated in the project at one stage of the work or another. Within the limits of a conference article, it is not possible to do justice to all the work involved. Rather, the intention is to present an overview of the project and show how the model is built, how it works and what it can do. As the project is still underway, other technical publications will follow that will better focus on each of the important parts of the model.

Acknowledgements

The authors gratefully acknowledge the joint funding of this undertaking by the Natural Sciences and Engineering Research Council of Canada (NSERC), Alcan International Limited (Alcanint), Jonqui&re, Qudbec and Alcan Smelters and Chemicals Limited (S6cal), Jonquifcre, Quebec. They thank Alcan for authorizing the publication of this text, and the many colleagues for their contributions in setting up and carrying out this project. V. Potocnik of Alcanint was instrumental in starting the project S. Brisson of UQAC contributed the much needed Task Manager.

References

1. Lyons, J.W., Min. H.S., Parisot, P.F., and Paul, J.F. "Experimentation with a Wet-Process Rotary Cement Kiln via the Analog Computer". I&EC Process Design and Development, pp. 29-33, 1962.

2. Manitius, A., Kurcyusz, E., and Kawecki, W. "Mathematical Model of the Aluminum Oxyde Rotary Kiln". I&EC Process Design and Development, 13, pp. 132-142, 1974.


3. Bui, R.T., Tarasievicz, S., and Charette, A. "A Computer Model for the Cement Kiln", IEEE-IA Trans, IA-18, 4, pp. 424-430, 1982.

4. Li, K.W., and Friday, J.R. "Simulation of Coke Calciners", Carbon, 12, pp. 225-231,1974. 5. Whittaker, M.P., Miller, F.C, and Fritz, H.C. "Structural Changes Accompanying Coke

Calcination". Ind. Eng. Chem. Prod. Res. Devel., 9, 2, pp 187-190, 1970. 6. Perron, J., Potocnik, V., Bui, R.T. "Modelling of the Coke Calcining Kiln", Proceedings

of the Int. Symp. on Reduction and Casting of Aluminium, Montreal, pp. 87-98, 1988. 7. Rosten, H.I and Spalding, D.B. "PHOENICS: Beginner's Guide and User Manual". Report

No TR/100, CHAM Ltd, London, 1986. 8. Perron, J. "Module math6matique d'un four de calcination du coke de p^trole" Ph.D. Thesis,

ficole Polytechnique de Montreal and University du Qu6bec k Chicoutimi, 1991.

383

Enhanced corrosion resistance in Al-Zn-Mg-Cu Alloys

R.C. Dorward Center for Technology, Kaiser Aluminum Pleasanton, California, U.S.A.

ABSTRACT

The corrosion resistance of Al-Zn-Mg-Cu alloys has been significantly improved by the development of the overaged T76 and T73 tempers, but strength penalties were incurred relative to the peak-strength T6 temper. Although retrogression and re-aging (RRA) treatments provide a combination of T76/T73 corrosion resistance and T6 strength, these short time, relatively high temperature processes are difficult to achieve in commercial practice. This paper compares the strength-electrical conductivity-corrosion resistance relationships of an Al-Zn-Mg-Cu alloy aged by (1) conventional 2-step (low-high) practices, (2) 3-step (low-high-low) methods in which the second stage is relatively long by RRA standards, and (3) a modified 2-step (high-low) procedure. Differential scanning calorimetry (DSC) was used to characterize the structure of the materials, and for correlation with the property data.

KEYWORDS

Al-Zn-Mg-Cu alloy, aging kinetics, step aging, tensile properties, corrosion resistance, differential scanning calorimetry, microstructure.

INTRODUCTION

Achieving high strength in combination with good resistance to exfoliation and stress corrosion in Al-Zn-Mg-Cu alloys has long been a goal of the producers and users of these materials. Although the development of the overaged T76 and T73 tempers provided great improvements in corrosion resistance, significant strength penalties were incurred relative to the peak-strength (and corrosion susceptible) T6 temper. However, in the mid 1970's it was shown that retrogression (reversion) and re-aging (RRA) treatments applied to 7075-T651 improved corrosion resistance dramatically while maintaining T6 tensile properties(l). (RRA treatments date back to at least 1941(2), but this was the first application for improved corrosion resistance.) The optimum practice coincides approximately with the minimum in the retrogression curve, and provides a T73 level of corrosion resistance in


combination with T6 or higher tensile properties. Unfortunately, the process requires rapid heating rates and soak times of only a few seconds or minutes at temperatures of 200-260°C. These thermal conditions are difficult to achieve in practice, especially with thick section sizes and batch operations.

Other workers have attempted to apply RRA technology using more practical (longer) aging treatments. In 1981, Wallace, et al(3), reported that T6 minimum tensile properties in combination with an electrical conductivity of 38% IACS

a could be achieved

in 7075 plate with retrogression treatments of a few hours at temperatures as low as 160°C. These findings have subsequently been confirmed by other studies(4,5). For these practices, the retrogression treatment is much longer than that corresponding to either the minimum hardness of the secondary hardening point, and therefore is more appropriately considered a second stage T7 type of aging thermal. Similarly, the re-aging treatment is actually the third stage of a 3-step (low-high-low) aging practice.

This paper compares the strength-electrical conductivity-corrosion resistance relationships of an Al-Zn-Mg-Cu alloy aged by conventional 2-step (low-high) practices and by 3-step methods in which the second stage is relatively long by RRA standards. Also included for comparison were modified practices in which the first step of the aging sequence was eliminated, i.e., a 2-step high-low practice. Differential scanning calorimetry (DSC) was used to characterize the structure of the materials, and for correlation with the property data.

MATERIALS AND PROCEDURES

Two 23 mm thick plates of the following compositions were solution heat-treated, quenched in room temperature water, and stretched 1.5 - 2%.

Table I - Chemical Compositions of Al-Zn-Mg-Cu Plates

% b y w t

b

Si Fe Cu Mg Zn Ti Zr

Lot # 1 0.04 0.12 2.21 2.17 6.46 0.06 0.09

Lot # 2 0.04 0.14 2.12 2.21 6.74 0.05 0.09 b By inductively coupled plasma spectroscopy except silicon (Quantometer).

Sections of each plate were stage-1 aged for 12 hr at 120°C, followed by stage-2 ages of 2 to 12 hr between 160 and 180°C. These sections were further divided into two samples each, one of which was stage-3 aged for 12 hr at 120°C. Heating and cooling rates were controlled at 40°C/hr.

The corrosion resistance of Al-Zn-Mg-Cu alloys correlates well with electrical conductivity. The minimum value for 7075-T73 is 38% LACS (International Annealed Copper Standard).


Additional samples were processed using heating and cooling rates more typical of

commercial batch operations:

A) Stage-1 aged at 120°C for 12 hr (30°C/hr heating rate), stage-2 aged at 170°C using a sequenced heating rate (30°C/hr to 150°C, 10°C/hr 160°C, 5°C/hr to 170°C), cooled 15°C/hr to 120°C and stage-3 aged for 12 hr, cooled 15°C/hr to room temperature.

B) Same as (a) above except heated directly from room temperature to 150°C at 30°C/hr, i.e., no stage-1 age at 120°C (a 2-step high-low practice).

The aged samples were tested for long-transverse tensile properties, surface electrical conductivity (MagnaFlux eddy current method), exfoliation corrosion resistance in EXCO solution (ASTM G34) and acidified synthetic seawater (ASTM G43), and short-transverse stress corrosion resistance using C-ring specimens stressed to 210 MPa and exposed (alternate immersion) to 3-1/2% NaCl (ASTM G44)). Selected samples were also examined by differential scanning calorimetry (DSC) using a Perkin Elmer DSC-4 instrument.

RESULTS

Aging Response

Average incremental changes attributable to the re-aging treatment (stage-3 age) were 13 MPa in yield strength and 0.4% IACS in electrical conductivity. As the 170°C isothermal stage-2 aging curves in Figure 1 show, the effects were greatest in the slightly overaged (strongest) condition, with benefits of about 15 MPa in yield strength and 1% IACS in conductivity. The effect of re-aging on strength is not as great as that claimed for 7075 alloy processed by similar treatments(3-5). This may be related to the fact that the alloy used in this investigation undergoes a significant secondary age hardening effect at moderate temperatures (150-165°C) due to its high copper content, i.e., less solute is available for hardening at the re-aging (stage-3) temperature.

Strength-Conductivity Relationships

Since electrical conductivity is an indicator of exfoliation and stress corrosion resistance in Al-Zn-Mg-Cu alloys, the conductivity data were plotted as a function of yield strength as shown in Figure 2. 3-step aging resulted in significantly higher conductivity (better corrosion resistance) for a given strength level. At the T6 (peak) yield strength (550 MPa), the advantage was about 1.5 IACS units. Conversely, 3-step aging provided about a 20 MPa yield strength increment at the T76 level of exfoliation resistance ( - 3 7 % IACS). The beneficial effect of 3-step aging appeared to improve somewhat with increasing strength; effects at the T73 strength level ( - 4 8 0 MPa) were only minor. Stage-2 aging temperature over the range 160 to 180°C did not appear to have a significant effect on the strength-conductivity relationships for each practice.

Also shown in Figure 2 are the strengths and electrical conductivities of the plate


samples that were aged by the simulated batch practices. Both had essentially equivalent properties, falling near the trend line for the 3-step aging practice.

Stress Corrosion Resistance

All the C-ring specimens from plates having a 36% LACS electrical conductivity or greater survived the 30-day alternate immersion exposure to 3-1/2% NaCl at 210 MPa applied stress, i.e., 3-step aged material with yield strengths as high as 550 MPa were resistant to the conditions noted. As shown in Figure 3, the type of attack consisted primarily of pit fissuring with some minor intergranular cracking in the stronger conditions. The more overaged materials underwent pitting only. Failure times for the specimens from the lower conductivity plates (<35.5% LACS) ranged from 1 to 4 days.

Exfoliation Corrosion Resistance

The test specimens exposed either 2 days in EXCO solution (ASTM G34) or for 7 days to the acidified synthetic seawater test environment (ASTM G43) were inspected visually before and after cleaning in concentrated H N 0 3. Selected samples were also cross-sectioned and examined metallographically. Agreement between the two test methods was good when the samples were rated after cleaning; however, there was less discrimination in the EXCO test when they were examined immediately after the exposure.

The results of the acidified synthetic seawater test (SWAACT) are given in Figure 4, in which the ratings are plotted against the electrical conductivity readings. As the metallographic cross-sections shown, the type-of-attack ranged from slight to severe, with a rating of EA (superficial) corresponding to a surface electrical conductivity of 37% LACS, i.e., a yield strength of 540 MPa in the 3-step aged plate. In all but one instance "pairs" of samples aged by 2-step and 3-step practices had the same exfoliation ratings, i.e., somewhat higher strength in 3-step aged material at an equivalent level of corrosion resistance (see Figure 5).

In addition to the visual/metallographic ratings, weight losses were also determined after wire brushing the chemically cleaned specimens. As shown in Figure 6, the correlation between weight loss and SWAACT rating was fairly good, suggesting that weight loss has promise as a quantitative measure of exfoliation corrosion resistance. Similarly, there was a good correlation between weight loss and electrical conductivity.

Differential Scanning Calorimetry (DSC)

Since DSC reveals information about the size and number of precipitates in age hardening systems, representative samples were examined by this technique using a scanning rate of 20°C/min. Precipitate dissolution is an endothermic process and the peak maximum temperature is a measure of the size of the dissolving precipitates (the larger the higher the temperature), and the area of the peak is a measure of the amount of the precipitate.


Figure 7, which compares the response of 2, 4 and 8 hr stage-2 ages at 175°C, shows a broadening of the endothermic dissolution peak and an upward shift of the maximum temperature as aging progresses. The effect of a 12 hr/120°C stage-3 age superimposed on the 4 hr/175°C thermal is shown in Figure 8. The maximum temperature (Tm) shifted back to a lower temperature and the peak broadened further (1.39 to 1.76 cal/g). Similarly, further stage-3 aging applied to material stage-2 aged for 12 hr at 165°C had the same effects.

The relationship between the endothermic DSC maximum temperature and strength is shown in Figure 9. In the underaged condition, T m increased with increasing strength up to peak strength ( T m ~ 214°C). As the alloy overaged (decreasing strength) T m continued to increase. Since electrical conductivity is a monotonic function of aging time, one would expect a simple correlation between DSC response and this property; and, as Figure 10 shows, the relationship appears to be linear. This suggests that the average size of the coarsening precipitates and the amount of alloying elements in solution are directly related as might be expected from the Gibbs-Thompson equation.

Finally, in view of the relationship between DSC response and the tensile/conductivity properties, there is also a correlation between peak dissolution temperature and corrosion resistance as shown in Figure 11. Aging conditions that result in T m values greater than about 215°C appear to confer good corrosion resistance.

DISCUSSION

Most studies of stress corrosion cracking (SCC) susceptibility in Al-Zn-Mg-Cu alloys have been directed toward the role of grain boundary precipitates, precipitate-free zones (PFZs), and matrix precipitates. The reason for the beneficial effect of overaging is still not understood, but it is generally believed that the size and distribution of grain boundary precipitates are the most important factors: improved resistance is associated with a relatively coarse dispersion.

During stage-2 aging, G-P zones dissolve, TJ ' and TJ precipitates nucleate, and grow, and

the grain boundary precipitates coarsen.' But to achieve T6 strength, step-2 aging must be

terminated before the matrix precipitates coarsen and while the matrix is still in a fairly high

degree of supersaturation. Then, upon stage-3 aging at a lower temperature, G-P zones are

reprecipitated and the t\

9 and T) continue to grow. The net result is larger boundary

precipitates spaced farther apart and a greater amount of n/ and r) precipitates along with

a G-P zone dispersion in the matrix. The PFZ width is not affected.

The DSC results showed that as degree of stage-2 aging increases, the dissolution peaks

broaden and shift to higher temperatures. This is due to the coarsening of G-P zones and

an increase in the proportion of r\\ In addition, some stable t) may be forming. When a c There continues to be some controversy over the relative amounts of these various precipitate

types for a given aging practice(6).


lower-temperature stage-3 age is superimposed, the dissolution peak maximum temperature

shifts back to a lower temperature. This is probably because the stage-3 formed G-P zones

and n/ lower the average size of the strengthening precipitates in the final product.

It is noteworthy that the first stage of the sequence is not required for commercial batch

aging operations in which heating rates are fairly slow. The benefits of such a practice, i.e.,

2-step high-low aging, were actually disclosed in 1967(7), a number of years before the first

RRA treatments were reported(l).

REFERENCES

1. B. Cina and B. Ranish, Proc. Int. Conf. on Aluminum Industrial Products, Pittsburgh,

ASM, 1974. 2. D. W. Smith and W. L. Fink, U.S. Patent #2,239,744, Apr. 29, 1941.

3. W. Wallace, J. C. Beddoes and M. C. DeMalherbe, Can. Aeronautics & Space J., Vol.

27 (1981), p. 221. 4. T. Ohnishi and H. Shiota, J. Japan. Inst. Light Metals, Vol. 36 (1986), p. 647. 5. E. S. Tankins and W. E. Frazier, Materials Performance, June, 1987, p. 37. 6. J. K. Park and A. J. Ardell, Scripta Met, Vol. 22 (1988), p. 1115. 7. G. R. Subeltt and M. W. Fien, U.S. Patent No. 3,305,410, Feb. 21, 1967.

YIELD SRENQTH (MPa) ELEC. COND'Y <%IACS) 575 i 142

550

525h

500

475 0 2 4 6 8 10 12 14

STAGE-2 AGING TIME (hr)

Figure 1 - Effect of stage-2 aging time at 170°C on the strength and electrical con-ductivity of plates aged by (open symbols: 2-step; closed symbols: 3-step) 2-step and 3-step practices.

41

39

37

35

33

ELEC- COND'Y (%IACS)

• 2-STEP

B 3-STEP

+ COMM. 3-STEP

x HI-LO 2-STEP

475 5 0 0 525 550 575

YIELD STRENGTH (MPa)

Figure 2 - Strength-electrical conductivity relationships for 2-step and 3-step aged plates. Symbols + and x refer to practices A and B described in text.


Figure 3 - Short-transverse C-ring specimen from 3-step aged plate (535 MPa yield strength) exposed to 3 1/2% NaCl (alternate immer-sion) for 30 days at 210 MPa applied stress.

ELEC. CONDUCTIVITY (%IACS)

ED EC EB EA P

SWAACT RATING (CLEANED)

Figure 4 - Exfoliation corrosion resistance of 2-step (open symbols) and 3-step (closed symbols) aged plates (ASTM G43).

389

100 Mm


580

560

540

520

500

480

460



150 WEIGHT LOSS (mg/sq.cm)

120h


Figure 7 - DSC curves for plates stage-2 Figure 8 - DSC curves showing effect of aged at 175°C. stage-3 aging on peak shift.

TEMPERATURE ( ° C ) TEMPERATURE ( ° C )

Figure 5 - Relationship between strength and exfoliation resistance of 2-step (open symbols) and 3-step (closed symbols) aged plates.

Figure 6 - Relationship between visual SWAACT rating and weight loss.


240

230

220

210

200

190

PEAK TEMPERATURE (C)

OVERAGED

•

UNDERAGED

425 450 475 500 525 550 575


Figure 9 - Relationship between DSC peak maximum temperature and yield strength.

240

230

220

210

200


•

228

224

220

216h

212


\

190

L

25 30 35 40

CONDUCTIVITY (%IACS)

208 45 25 50 75 100

WEIGHT LOSS (mg/sq.cm)

125

Figure 10 - Relationship between DSC peak maximum temperature and electri-cal conductivity.

Figure 11 - Relationship between DSC maximum peak temperature and corrosion resistance.

393

Scrap melting and metallurgical processes employed in aluminum recycling

David V. Neff Metaullics Systems, Solon, Ohio, U.S.A.

INTRODUCTION

The o r ig in of aluminum recycl ing occurred during World War I I as the nation scrambled to produce metal products to meet m i l i t a r y requirements. During the next t h i r t y - f i v e years or so, recycl ing of aluminum was largely l im i ted to the secondary smelters. With ever-increasing energy costs, environmental concerns, resource conservation, and demise of many primary smelters in North America, pa r t i cu la r l y the older United States f a c i l i t i e s , and the accelerated use of the aluminum beverage can and used beverage containers (UBC) as the most important scrap feedstock avai lable, recycl ing of aluminum has gained great new prominence.

Today, recycl ing of both consumer scrap (secondary producer recycl ing) and producer scrap (primary smelters and mi l l products producer recycl ing) provides a very v i t a l source of metal un i ts , and is environmentally, energet ica l ly , and economically good business. Various technologies in melting and molten metal treatment have evolved to render recycl ing to be a technica l ly successful proposi t ion. This paper describes the emergence of speci f ic scrap melting and metals re f in ing techniques - scrap submergence, pumping, demagging, degassing, a l ka l i r e f i n i n g , and f i l t r a t i o n techniques which have made th i s possible.

Character ist ics of Scrap Aluminum

For the purposes of th is paper, three types of scrap aluminum shall be del ineated. The f i r s t of these, consumer scrap, consists of a l l used aluminum a l loy products which have been col lected through domestic usage, are sorted and fur ther processed pr io r to remelt ing. The sort ing is of course indeed v i t a l to avoid mixed a l loy contamination. Secondary operations include de-o i l ing or delacquering, chopping, crushing, shredding, e t c to render the scrap a more usable size for handling and melt ing.

Consumer scrap contamination almost always includes oxide, d i r t , organic residues (pa int , lubr icants , etc. and other surface debr is, even with various treatment preparations), and hydrogen gas as a resu l t of the contaminants. Metal l ic impuri t ies can also arise because of inadequate sor t ing . One meta l l ic impurity is germane to secondary recycl ing and usually inevi table - magnesium. Most wrought product, a major scrap source for secondary recyclers, contains higher magnesium levels than the foundry ingot or die cast a l loy product which secondaries create, hence the magnesium content must be decreased during the remelting process.


The second type, producer scrap, is scrap generated by the producer -e i ther as run-around scrap in his own processing (but ts , c o i l s , s l i t t e r t r im , scalper chips, e tc . ) or as t o l l buy-back (ends, skeletons, e tc . ) from a d i rec t fabr icat ing customer. This type of scrap is usual ly, but not always, f a i r l y clean and composition is cont ro l lab le ; a l loy mixup and metal l ic contamination are usually minimal. Hydrogen gas, oxide inclusions, and often a lka l i metal impuri t ies at very low level may s t i l l be present and must be control led during the recycl ing process. The concentration of such contaminants is overal l less than that associated with consumer scrap in secondary recycl ing, however.

The t h i r d type of aluminum scrap is used beverage containers (UBC). The recycl ing of UBC has evolved into a huge business wherein dedicated f a c i l i t i e s exist handling only th is type of scrap. While a l loy mixups are not a problem, there is s t i l l v i t a l concern for the usual metal lurgical problems - hydrogen con t ro l , a l ka l i metal removal, inclusions, as recycled UBC metal is refabricated into exactly the same product - beverage cans. Can body and l i d a l loy compositions are d iss im i la r , however, often requir ing special handling or metal lurgical con t ro l . Processing to l i g h t gauge by the metal producer, and subsequent forming into cans by the can maker, demands great c r i t i c a l i t y and control over metal defects.

ALUMINUM SCRAP MELTING TECHNOLOGY

Scrap Submergence

Because considerable energy must be put into a pound of aluminum scrap even at i t s melting temperature (nearly 180 out of 500 BTU t o t a l ) , in order to melt, the condit ion and shape of the scrap metal i t s e l f becomes quite important. Heavy sol ids can be melted easi ly given enough thermal energy avai lable and residence t ime. Light gauge, low density scrap such as turnings, borings, shreds, and UBC, with the i r high surface area to volume ra t i os , pose other problems, however, in that such material f loa ts and oxidizes read i ly . Consequently, th i s type of scrap must be fo rc ib l y submerged and melted quickly to achieve high y ie lds and melt e f f i c iency . A var iety of submergence devices and techniques for l i g h t gauge scrap have therefore arisen to overcome these problems(l).

Scrap preparation is usually v i t a l when planning to remelt l i g h t gauge low density aluminum scrap. Where possible, baling and br iquet t ing of homogeneous scrap achieves consolidation such that r e l a t i ve l y large volumes of metal may be charged into side well or the main hearth without undue losses. In the case of chips, turnings, and borings, de-o i l ing and drying are necessary to avoid excessive melt losses (5-10%) through burning and subsequent oxidat ion. In the case of UBC's, the pigments and lacquer coating must be removed in a delacquering process to achieve optimal melting and recovery rates. Addi t iona l ly , at least one recycler uses a thermo-mechanical separation process(2) to e f fec t i ve ly separate d iss imi lar can body and l i d a l loy components from UBC pr io r to melt ing.

Scrap submergence techniques include induction currents, mechanical "puddlers", and "vortexers". Induction furnaces are used in certa in instances, especial ly for remelting chips, where the induction convectional current pattern cont inual ly draws the chips into the bath. A variant on th i s appl icat ion employs the l inear induction motor as a submergence t o o l , although the melting rates which have been demonstrated(3) are quite low and not commercially viable in most recycl ing operations.


Mechanical "puddlers" have been the simplest and most common devices employed, especial ly when melting a var ie ty of scrap forms, almost always in side-well charged reverberatory furnaces. These devices simply force the scrap beneath the melt surface by mechanical means. Scrap submergence rates can be quite high, l im i ted only by the thermal capacity and heat t ransfer capab i l i t i es of the furnace. However, melting e f f i c ienc ies and y ie ld are often compromised by the r e l a t i ve l y prolonged exposure of the scrap on the bath surface pr io r to being submerged fo r c i b l y by the mechanical " f ee t " .

Another class of submergence devices consists of various "wheels" ro ta t ing devices which create a "vortex" or near vortex in which l i g h t gauge, r e l a t i ve l y small scrap forms such as turn ings, chips, shreds, and UBC can be drawn down and mixed into the metal bath. Most of these are radial f low while one is axial which presumably should have inherently better scrap draw-down and submergence c a p a b i l i t y ( l ) . A collage of various vortexing devices is depicted schematically in FIG. 1. Many of these are current ly used in practice covering a var ie ty of feed stocks, charge rates and resul tant melting e f f i c ienc ies . While each has i t s mer i ts , i t should be noted that the item represented in FIG. 1(e) of fers s ign i f i can t advantages. Recently evaluated in production, t h i s device (patent applied for ) is capable of excellent submergence capab i l i t i es of UBC scrap in excess of 20,000 l b s . / h r . , operates at lower ro ta t ional speeds, and is non-clogging, which s i gn i f i can t l y aids product iv i ty and achieves long performance l i f e . The uni t is current ly outperforming several competitive vortexing devices.

Drive systems for a l l such vortexing devices can be simple or complex, overhead f ixed mounted or portable. FIG. 2 depicts a robust portable un i t which can be easi ly moved from furnace to furnace.

The effectiveness of vortexing technology is governed by many fac tors . On-going s tud ies( l ) indicate that highest melt rates and metal recoveries are great ly dependent on furnace geometry and conf igurat ion, furnace thermal capacity, homogeneity of scrap feed, temperature in the charge w e l l , and inherent submergence and mixing character is t ics of the vortexing device i t s e l f . Add i t iona l ly , while vortexing devices used alone do move metal, greatest benefi t has been seen when such devices are coupled with another means to achieve c i rcu la t ion through the charge well and throughout the ent i re furnace, e i ther by s t i r r i n g or pumping, which w i l l be described in the next sect ion.

Heat Transfer in Aluminum Reme1ting(4,5)

Heat t ransfer occurs in aluminum melting conventionally by rad ia t ion , conduction and convection. I n i t i a l heat up of a cold charge is usually accomplished by radiat ion alone, such as in a "reverberatory" furnace.

Most reverberatory furnaces employed in the aluminum remelting industry are e i ther rectangular or round. The l a t t e r are top charged and are pr imar i ly used by the primary producers and m i l l recyclers where scrap co i l s and sheet are charged. Secondary remelters use rectangular furnaces. The high radiat ion heat t ransfer and po ten t ia l l y high thermal heads (roof and sidewall temperatures) o f fe r ample opportunity fo r a bath c i r cu la t ion system to improve the radiat ion heat f l ux in to the bath by providing forced convection wi th in the melt, which overcomes rather small


USPN 4,598,899; 4,930,986

USPN 4,486,228 (Pat. Appl. For)

FIG. 1 - Schematic Collage, Vortexer Devices

396

(a)

(c)

(d)

(b)

USPN 4,884,786 USPN 4,322,245


FIG. 2 - Shrouded Auger Melt ing System (SAMS) (Metaul l ics Systems)

natural convection forces. High ve loc i ty burners do aid somewhat in forced convection, but only as a surface phenomena at the so l id charge and jus t a few mi l l imeters into a molten bath.

In the melting of so l id aluminum, approximately 500 BTU are required per pound to bring the sol id to the melting point and to overcome a rather large l a t e n t heat of fus ion. A forced convectional system grea t ly assists the heat t rans fer in two fundamental ways. F i r s t l y , w i th in the molten bath i t s e l f the greater forced convectional heat t ransfer permits a greater heat f lux into the bath, through the surface, from the radiant heat source, namely the burners and re f l ec to rs (hence reverberatory ) - roof and s idewal ls . Without the a b i l i t y of forced convection to aid in t h i s heat f l u x , the bath surface temperature and the thermal head temperatures would r i s e with continuing heat input . Hot spots wi th in the melt i t s e l f , which re tard fur ther heat t ransfer e f f i c i e n c y , can also be a l l e v i a t e d by a forced convectional system. When melting scrap, th is is especia l ly important in minimizing oxidation and dross losses. Secondly, a forced convectional system wi th in the bath allows sol id charge to be melted much more quick ly , both by heat t ransfer and by r e l a t i v e momentum or motion. This becomes especia l ly important in the so-cal led f l a t - b a t h melting stage, wherein "sunken solids" charge metal coexists with surrounding l i q u i d metal at very nearly the same temperature (FIG. 3 ) . There i s , in t h i s case, l i t t l e thermal gradient to continue the melt-down process at a high r a t e , hence fur ther melting proceeds somewhat slowly with great i n e r t i a . A l i q u i d metal pumping system assists in such "abla t ive" melting by v i r tue of increasing the k ine t ic energy at the s o l i d - l i q u i d i n t e r f a c e , improving the ra te of heat t ransfer and hence the melting r a t e ( 4 , 5 ) .


FIG. 3 - Schematic, Sunken Solids at F la t Bath Melting Condition

The heat t ransfer coe f f i c ien t applied wi thin a molten aluminum bath can be determined by conventional convective heat t ransfer theory which makes use of various dimensionless number re la t ionships . The end resu l t is that the heat t ransfer c o e f f i c i e n t , h, is proportioned to a f rac t iona l power (K) of the induced f l u i d ve loc i ty :

h = f (v) (expK) (1)

For l i q u i d metals, k is usually near 0 . 8 . Providing Forced Convection in Scrap Melting

Thus in melting aluminum scrap i t is necessary to provide a forced convectional "assist" for optimum e f f i c i e n c y . There are several means by which forced convection, i . e . forced bath c i r c u l a t i o n , can be applied to an aluminum melting f u r n a c e - - s t i r r i n g (manually or e lect romagnet ica l ly ) , or e lectro-pneumat ical ly ; or pumping (electromagnet ical ly or mechanical ly) . Other devices such as f lux ing tubes and porous plugs provide only minimal means of true convectional heat t r ans fe r . Indeed, such degassing devices can actual ly lower bath temperatures.

The key to providing the forced convectional heat t ransfer "assist" is to u t i l i z e an external energy system to increase momentum and k ine t ic energy wi th in the molten metal bath i t s e l f . Hence mass (f low r a t e ) and induced ve loc i ty c a p a b i l i t i e s provided by any forced convection device w i l l be s ign i f i can t character is t ics which determine effect iveness of that device.

Electromagnetic pumps and s t i r r e r s have been in use for some time in l im i ted appl icat ions, and work very well in minimizing temperature s t r a t i f i c a t i o n and a l loy inhomogeneity. Discharge ve l oc i t i es are usually l.Om/sec or l ess , however(6). Recently, an electro-pneumatic s t i r r e r has been developed which has been applied to several primary as well as secondary (UBC) melting furnaces. Discharge ve loc i ty of t h i s device is of the order of 3-5m/sec. and at a flow rate of 2000 kg/min(7) .

398


Centr i fugal pumping systems (FIG. 4) have been very useful in secondary smelters and m i l l recyclers because of t h e i r r e l a t i v e l y high momentum and k ine t ic energy inputs. An e l e c t r i c a l l y driven centr i fuga l pump has been measured to have a discharge ve loc i ty 3-8m/sec and a flow ra te of nearly 15,000 Ib /min . (7000 kg /min . ) . Hence the e f f e c t on the heat t ransfer c o e f f i c i e n t (eqn 1) w i l l be considerably greater . Add i t iona l l y , the capi ta l and operation costs ( e l e c t r i c a l energy) are considerably less for mechanical pumping systems versus electromagnetic or electro-pneumatic pumps and s t i r r e r s . Consequently, mechanical forced convection systems, as embodied by the centr i fugal pump, have become the predominant means to provide the melting heat t ransfer "assist" in the major i ty of secondary smelter and mi l l recycler plants operating side well mel ters . Add i t iona l l y , one can also u t i l i z e mechanical pumping systems to advantage for scrap melting in closed hearth, d i rec t charge melters as w e l l ( 8 ) . Current experience indicates that proper pump i n s t a l l a t i o n , maintenance, and operational procedures can y i e l d s imi lar benef ic ia l resu l ts in improvement in charging ra tes , melting e f f i c i e n c y , and furnace cycle t imes.

FIG. 4 - Centr i fugal Pump (Metaul l ics Systems)

Use of Mechanical Pumping Systems in Remelting

The benef i ts of centr i fugal pumps in secondary aluminum recycl ing have been well described r e c e n t l y ( 4 , 5 ) . Improvements in energy e f f i c i e n c y ,


product iv i ty , and melt ra te in secondary smelting of mixed consumer scrap are well documented, b r i e f l y summarized in Table I . Energy e f f i c i enc ies depend of course on the s ta r t ing point . Several years ago i t was not uncommon for energy di f ferences of 3500 BTU/lb. (without a pump) and 2000 BTU/lb. (with a pump) to be experienced.

TABLE I

General Benefits of Adding Centri fugal Pumping Systems to Scrap Remelting Processes*

Charge Rate- - Increased up to 100% Melt Rate Increased 25-50% ( to 16,00-18,000 l b s . / h r . ) * * Energy E f f i c iency - - To 1600-1800 BTU/lb. (20-50% improvement) Product iv i ty As much as 100% improvement Temperature S t r a t i f i c a t i o n - Reduced to 10° F. or less throughout Thermal Heads Roof temperature reduction - 2300-1900° F.

Sidewall temperature reduction - 1900-1650° F. Bath surface temperature reduction - 1600 -1400° F.

*Based on 100,000 l b . or greater furnace s i ze . **Also dependent on burner capacity.

Today even a good energy e f f i c i e n t secondary operation s ta r t ing at 2200 BTU/lb. can generate a reduction to 1600-1800 BTU/lb. (50% hee l , cold charge) by adding a centr i fugal pumping system. Even such modest reduction of 10-15% energy savings make pumping system i n s t a l l a t i o n cost-j u s t i f i a b l e .

Melt ing rates are highly dependent on thermal input capacity to the furnace and on speci f ic operating circumstance. I t is not uncommon to achieve 25-50% increase in melt r a t e . Most secondary aluminum 100,000 l b . well-charge furnaces experience 12,000 to 18,000 l b . / h r . melting rates with a var ie ty of scrap metal feeds.

Most of these foregoing resul ts apply to secondary smelting of consumer scrap, with perhaps some producer scrap mixed i n . The usual heterogeneous nature of secondary scrap, c l i p s , chips, crushed cast ings, archi tectura l scrap, f abr ica t ions , e t c . , plus unknowns, places a burden on pumping operations to handle such a diverse scrap mix, especia l ly when the scrap is not consolidated by br iquet t ing or ba l ing . Nevertheless, pumping systems perform admirably well and are very cos t -e f fec t i ve in secondary scrap smelting operations.

The mi l l products recycl ing f a c i l i t i e s have a higher class of control led scrap and other so l ids , ( i . e . c o i l s , but ts , extrusions, s l i t t e r scrap, p l a t e , sow, RSI, e t c . ) and where these f a c i l i t i e s are located wi thin primary p lants , there is a source of hot (pot l ine ) metal as well to make up the furnace charge. Melt r a t e , energy e f f i c i e n c i e s , e t c . are usually a notch above those for the secondary operations, p r i n c i p a l l y because of the bet ter cont ro l led , narrower scrap feed mix. Likewise, pump performance and longevity are also improved.

Dedicated UBC recycl ing f a c i l i t i e s af ford an even bet ter performance scenario. Table I I describes pumping performance in one such dedicated UBC recycl ing f a c i l i t y , IMSAMET, Hauser, Idaho(5 ,9 ) . Of special note is


that in t h i s type of dedicated remelt ing:

- melting ra te is considerably higher (22,000 l b s . / h r . ) than for conventional secondary mel t ing.

- pump performance l i f e is very good (35 days average) without parts replacement.

- overa l l pumping costs are very modest.

- energy e f f i c i e n c i e s are exce l len t , especia l ly with a hot delaquered scrap charge.

Centr i fugal Molten Metal Pumps in UBC Recycling at IMSAMET, Hauser, Idaho

Thus UBC recycl ing provides an even more successful appl icat ion of molten metal pumping systems than in conventional mixed consumer scrap secondary recyc l ing .

Mechanical pumps are used s i g n i f i c a n t l y in conjunction with a va r ie ty of "vortexer" scrap submergence devices in several recycl ing f a c i l i t i e s . The molten metal c i r c u l a t i o n a b i l i t y and forced convectional heat t ransfer "assist" enables scrap to be in t imate ly mixed with incoming hot metal and to be carr ied through the charging well (with s u f f i c i e n t residence t ime, nevertheless) to permit highest charging and throughput r a t e s ( l ) . Recent experience with one axia l vortexing device generated a melt ra te in excess of 30,000 l b s . / h r . , which should be sustainable given proper furnace thermal design and furnacing cycle management.

There are other functions that must be performed in successful remel t ing / -recycl ing operations. Metals re f in ing - demagging, a l k a l i removal, degassing - are s ign i f i can t melt treatment pract ices often necessary in remelt ing. A centr i fugal pumping system can uniquely provide a l l three in a furnacing batch operat ion, and other technologies are ava i lab le for i n - l i n e treatment.

TABLE I I

Pump Type Number in Continuous Use Melt ing Rate Average Metal Temperature, Charge Well Average Metal Temperature, Furnace I n t e r i o r Furnace F i r ing Capacity Energy Ef f ic iency (Hot UBC) Energy Ef f ic iency (Cold UBC) Pump L i fe (on- l ine time) Average Monthly Pump Maintenance Cost

4-M-42 (Metaul l ics Systems) 2 22,000 l b . / h r . maximum 1375° F. (18,000 l b . / h r . 1400° F. 28,000,000 BTU 1150 BTU/lb. 1650 BTU/lb. 35 days average $4,520 t o t a l (4 pumps)

METALS REFINING IN SCRAP MELTING

Demagqinq

Demagging is commonly practiced in secondary smelter recycl ing of consumer scrap. Often the ava i lab le scrap charge mix includes a predominance of wrought scrap products whose overal l magnesium content exceeds that


required for producing speci f icat ion die cast or foundry a l loy ingot. Hence, the melt often must be "demagged". The use of chlor ine in jec t ion has been found most b e n e f i c i a l ; ch lor inat ing a molten aluminum a l loy containing magnesium produces a reaction which u l t imate ly resul ts in the formation of a l i q u i d magnesium chloride which usually separates to the surface or dross phase given enough residence t ime.

H i s t o r i c a l l y , t h i s react ion has been accomplished by f lux tube i n j e c t i o n , the use of s a l t f luxes (Derham process), or the scrubber be l l process. These techniques were general ly found wanting, e i ther possessing somewhat low react ion e f f i c i e n c y , or environmental and disposal problems. The favored technology for demagging wi th in the secondary industry employed today is the gas in jec t ion pump shown in FIG. 5 ( 1 0 ) . This device is commonly used in reverberatory furnace remelting of consumer scrap. Key benef i ts include operating at near theoret ica l e f f i c iency of chlorine usage with v i r t u a l l y no e f f luent ( f ree aluminum chlor ide and i t s der iva t ives) released to the atmosphere when properly operated. The high demagging e f f i c iency and non-pollut ing nature of th is technology is a resu l t of high operating speeds of the pump (600-1100 rpm) wherein the high ve loc i ty discharge from the impeller shears the inc ip ient gas in jected jus t ahead of i t , thereby creat ing an extremely f ine bubble size with high bubble surface area-to-volume r a t i o , for best gas- l iqu id react ion k i n e t i c s . Hence demagging can proceed very e f f i c i e n t l y . Figure 6 depicts a typ ica l demagging curve for a 150,000 l b . melt (75,000 l b . heel) with continuous charging of high magnesium bearing scrap.

FIG. 5a - E lec t r ic Drive M-30 Gas In jec t ion Pump (Metaul l ics Systems)


FIG. 6 - Demagging a 150,000 Lb. Secondary Aluminum Melt

Most m i l l or producer scrap is usually remelted to the same chemical spec i f ica t ion and hence demagging is not general ly employed t h e r e i n . Likewise, UBC recycl ing and the growth of dedicated f a c i l i t i e s committed to regeneration of UBC can body and l i d stock a l loys usually do not

FIG. 5b - Typical Secondary Smelter Pump I n s t a l l a t i o n

ACTUAL MAGNESIUM CONTENT USING PUMP

-CALCULATED MAGNESIUM CONTENT

GAS INJECTION PUMP

Z M

AG

NE

SIU

M


require demagging. Often normal magnesium loss during melting and the use of make up metal can bring the f i n a l magnesium content from the mixed a l loy (can and l i d ) into speci f icat ion for new can body stock. Indeed, economics v i r t u a l l y demand that beverage can a l loys be reproduced from the same high magnesium a l l o y s , with only a l loy "sweetener" added in small amounts to maintain preferred composition. However, the increasing amount of UBC material ava i lab le o v e r a l l , and paucity of other wrought scrap forms ava i l ab le , may mean that secondary recyclers u t i l i z i n g even a portion of UBC for t h e i r remelt stock may need to demag to produce other a l loys , especia l ly foundry ingot.

A lka l i Removal

Both mi l l producer scrap as well as consumer scrap may contain meta l l i c impuri t ies other than magnesium. The a l k a l i metals l i t h i u m , sodium, potassium, and calcium may enter the scrap stream as a resu l t of or ig ina l primary production, or l a t e r casthouse processing. Primary producers often do de l ibe ra te ly re f ine l i t h i u m , sodium and calcium from most melts, using chlor ine in jec t ion treatment of pot l ine vessels and/or i n - l i n e degassers, but i t must be remembered that th is is r e l a t i v e l y recent emphasis. Var ie t i es of consumer scrap entering the remelt arena a f t e r 20-40 years of service were not o r i g i n a l l y ref ined in the same manner and hence such contamination is possible.

The gas in jec t ion pump is sui table for continuous r e f i n i n g a l k a l i metal impuri t ies from scrap metal during the remelting process. FIG. 7 represents the continuous re f in ing of l i th ium from a mixture of sow, mi l l scrap and customer scrap in a 100,000 l b . reverberatory furnace charge as the charge is melted down(5).

I n - l i n e molten metal treatment may also be employed to reduce a l k a l i content, p a r t i c u l a r l y for m i l l recyclers with such equipment already in place. Several i n - l i n e metal treatment systems such as SNIF, Alcoa 622, Alpur, MINT, DMC are rout ine ly used for t h i s purpose, s p e c i f i c a l l y using chlorine as the r e f i n i n g gas.

25 LITHIUM CONTENT, PPM

20

15

10

6h

0 ' — 0:00 0:10 0:20 0:30 0:40 0:50

TIME, MINUTES 1:00 1:10 1:20 1:30

6 8 C F H , 6 % C H L O R I N E / 0 6 % N I T R O Q E N

FIG. 7 - Typical Refining of Lithium with the Gas In jec t ion Pump During Total Furnacing Cycle

100,000 LB CHARGE

1000 SERIES ALLOY


Degassing

Secondary smelting of consumer scrap, remelt ing of good q u a l i t y producer scrap, and melt recycl ing of UBC usually require ext ra measures of melt c leanl iness to be taken. Non-metall ic inclusions or hydrogen gas pose probable problems in the end product. Hence degassing and f i l t r a t i o n are usual ly employed, of ten in various i n - l i n e systems posit ioned between furnace and casting p i t . Other degassing measures are often used in the furnace as w e l l . Hydrogen may be removed from the melt in the furnace using hexachloroethane t a b l e t s , f l ux tubes, or with the gas i n j e c t i o n pump. The l a t t e r provides degassing i n - s i t u during melt ing and avoids the need for opening doors or a separate degassing period at the end of the heat . The bubble-shearing action of the ro ta t ing impel ler permits e f f e c t i v e degassing throughout the heat with typ ica l 5% ch lor ine , 95% nitrogen mixtures. The chlorine addit ion helps wet non-metal l ic inclusions and de l i ve r them to the surface, so providing addi t ional melt c leaning. A period of bath quiescence or holding time is desirable to allow a major i ty of inclusions in the bath to separate by sedimentation (heavier p a r t i c l e s ) or f l o t a t i o n ( l i g h t e r p a r t i c l e s ) , two "natura l" f i l t e r i n g phenomena.

In-furnace degassing with the gas in jec t ion pump can resu l t in f i n a l gas leve ls near 0.15cc H2/100g, thereby f a c i l i t a t i n g or even e l iminat ing the use of a separate i n - l i n e degasser(5) . Performance of i n - l i n e degassing and f i l t r a t i o n , i f pract iced, are usually enhanced by the prel iminary cleaning treatment in the furnace.

I n - l i n e degassing of secondary or recycled metal often demands greater concentration of react ive gases, i . e . ch lor ine , and longer residence t imes, i f possible, to achieve greater hydrogen reduction and inclusion f l o t a t i o n than is the case with primary metal . Pr ior to t reatment , melt hydrogen leve ls of secondary meta l , scrap being remelted, can often be nearly 0.45-0.50cc H2/100g A l , about twice that of clean scrap or primary metal . Nevertheless, the f u l l range of recycled aluminum al loys can be produced to v i r t u a l l y the same cleanl iness leve l as primary metal with degassing and f i l t r a t i o n .

Large scale batch degassing using a rotor system is also possible . FIG. 8 portrays a double rotor degassing system, the Dual STAR™. This patented system incorporates two rather simple rotors and accomplishes excel lent mixing k inet ics and react ion e f f i c i e n c i e s at moderate ro tor speeds. The system is capable of t rea t ing vessels such as crucibles or ladles from 5000 to 15,000 l b . Ei ther iner t or ine r t plus react ive gases may be employed. Recent degassing resul ts on recycled 6000 ser ies a l loy are shown in FIG. 9 . I t is expected that th is technology w i l l also f ind some usefulness in demagging, a l k a l i removal, and metal cleaning ( to be discussed) where batch treatment is desired.

Fluxes and Flux In jec t ion

The use of f lux ing sa l ts is commonplace in most aluminum recycl ing operations as a melting a i d . So-cal led wet f luxes , based on the binary system sodium chloride/potassium chlor ide with a few percent of f luor ide sa l t ( c r y o l i t e ) , are used with the charge to agglomerate impur i t i es , absorb oxides, and provide some surface protect ion for l i g h t gauge scrap as i t melts . Such f luxes are commonly used in melting and recycl ing of both consumer and producer scrap, and can improve recoveries by several percent, depending on speci f ic operating circumstances.


SYSTEM

FIG. 8 - Dual STAR Rotor Degassing System (Metaul l ics Systems)

HYDROGEN CONTENT, CC /100g Al

1 0 , 0 0 0 LB LADLE, 6 0 0 0 SERIES ALLOY

FIG. 9 - Degassing Performance in 10,000 Lb. Ladle Using Dual STAR

TABLET TABLET TABLET STAR STAR STAR

TREATMENT

• • BEFORE FLUXING WM AFTER TREATMENT


Salts are also used with UBC, but the reactions and f lux ing e f f i c i e n c y are complicated by the presence of magnesium. Spinel formation (MgAl204) occurs at the flux-atmosphere i n t e r f a c e , increasing density and v i s c o s i t y , and perhaps a l t e r i n g the f l u x ' s wett ing c h a r a c t e r i s t i c s ( l l ) . Metal recovery is thus lessened as the f lux entraps l i q u i d metal and as the tendency for oxide formation wi th in the f lux is increased. Physical ly such f lux appears black, rather l i k e metal - lean or t reated dross in non-magnesium a l l o y s , thereby fool ing the furnace operator. When melt ing UBC and where a f lux is required, sparing usage and a higher f luor ide ( c r y o l i t e ) f l ux content are h e l p f u l .

Flux in jec t ion is a r e l a t i v e l y new technology which can be employed in aluminum recycl ing for molten metal treatment a f t e r melt ing is completed. Fluxing sa l ts may be used for degassing, grain r e f i n i n g , or metal c leaning. In the l a t t e r , halogen containing compounds aid in wett ing out non-metal l ic oxide inclusions from the melt .

Flux in jec t ion may be coupled with rotor dispersion to achieve optimal performance especia l ly for large scale batch treatment; i . e . large furnaces or l a d l e s . FIG. 10 portrays a f lux in jec t ion system, the AMCOR INJECTA, which has been used in conjunction with the aforementioned STAR rotor to provide a f u l l y integrated metal treatment system. The combined equipment provides both small and large scale ( to 15,000 l b . ) metal treatment c a p a b i l i t i e s for degassing, demagging, a l k a l i removal, inclusion f l o t a t i o n , dross reduct ion, and p o t e n t i a l l y even a l l o y i n g .

FIG. 10 - Flux In jec t ion System (AMCOR)


Combining f lux in jec t ion with rotor dispersion provides f u l l e f f i c i e n t usage of chemical f lux ing compounds, and thorough dispersion and mixing for best react ion k i n e t i c s , especia l ly important for any large vessel treatment.

F i l t r a t i o n

Despite using other metal treatment technologies, pos i t ive f i l t r a t i o n is s t i l l usually qui te necessray to clean up recycled metal . F i l t r a t i o n of recycled a l loy melts is accomplished in the same fashion as primary melting - through furnace f lux ing to achieve f l o t a t i o n and sedimentation of p a r t i c u l a t e , and i n - l i n e metal treatment. Posi t ive f i l t r a t i o n may be performed with bed f i l t e r s , bonded p a r t i c l e f i l t e r systems, or ceramic foam p la tes . In any instance, the " d i r t loading" of secondary, recycled metal , is often substant ia l ly greater than primary metal . Hence f i l t e r l i f e may be shorter , and the need for greater f i l t e r i n g capacity evident . FIG. 11 represents an example of how higher concentration of inclusions present in recycled metal versus primary metal could resu l t in more rapid cake buildup on the surface of a ceramic p la te f i l t e r , hence less f i l t r a t e (metal passed) as a function of t ime(12 ) . Here the spec i f ic a l loy used, 7075, while not often .encounted in recyc l ing , nevertheless exhib i ts s imi lar inclusion morphologies, spinel st r ingers and oxide dispersoids to those that would normally be encountered in recycl ing UBC mixed 3000 and 5000 series a l l o y , for instance. The "contaminated" curve is thusly representat ive of the potent ia l d i r t iness involved in recycled metal .

Consequently, proper metal f i l t r a t i o n of recycled a l loy often requires "engineering" a f i l t e r appl icat ion to proper f i l t e r size (surface a rea ) , permeabi l i ty (porosi ty) and inclusion loading capaci ty , to accommodate a spec i f ic casting flow ra te and to meet a spec i f ic product requirement in f i l t r a t i o n e f f i c i ency or molten metal c leanl iness . When the " d i r t loading' is high, permeabi l i ty of the f i l t e r may need to be increased (greater porosity) to avoid excessive head buildup, although f i l t r a t i o n e f f i c i ency may thereby be reduced. Degree of f i l t r a t i o n is therefore a balancing act between several competing fac to rs .

Fluxing sa l ts that must often be used in the aluminum recycl ing process to f a c i l i t a t e melting and metal recovery w i l l also a f fec t f i l t r a t i o n r e s u l t s . As the f luxes create l i q u i d phases, such f l u i d i t y can wash out captured inclusions from wi th in a f i l t e r ( 1 3 ) , thereby reducing the ef fect iveness and e f f i c i ency of a ceramic f i l t e r system.

Grain Refining

Aluminum a l l o y s , whether primary or secondary, require a grain r e f i n i n g addit ion to f a c i l i t a t e fur ther processing. The p r inc ip le means to accomplish t h i s involves the addit ion of a t i tan ium a l loy add i t i ve , usually in rod form. Since scrap metal already contains some level of t i tan ium from the i n i t i a l treatment, there is a r e s i d u a l , and varying amount of t i tan ium in each scrap mel t . Add i t iona l l y , many coatings applied to aluminum, especia l ly beverage cans, contain t i tan ium pigment compounds which may be present i f the scrap is not f u l l y de-lacquered. Consequently the scrap melter should assess the residual t i tan ium content encountered and r e - a l l o y with a t i tanium grain r e f i n e r accordingly. An addi t ion ra te of 0.003% - 0.005% t i tan ium is usually acceptable to achieve adequate grain r e f i n i n g in recycled UBC(14).

408


700 r

600

ro E

500

UJ

5

O 400

UJ

300

UJ

>

200

100

ALLOY 7075

o,« EXPERIMENTAL DATA

FIT TO MODEL EQUATION

CONTAMINATED

100 200 300 400 500

TIME, t (s) 600

FIG. 11 - Comparison, F i l t r a t e Volume/Time on 7075 A l loy , Clean Vs. Contaminated Metal (t$

CONCLUSIONS

The use of various systems to e f f e c t i v e l y submerge and melt l i g h t gauge, low density aluminum scrap such as UBC has evolved to the point that sustainable production rates in excess of 20,000 l b s . / h r . are v i r t u a l l y rout ine . New vortexing systems o f f e r improved throughput and r e l i a b i l i t y .

The current ly p re fe r red , most e f f e c t i v e technology to enhance forced convectional heat t ransfer in remel t ing/recycl ing of scrap aluminum is the use of mechanical centr i fuga l pumping systems. The improved heat t rans fer g rea t ly assists the smelting of consumer scrap by the secondary aluminum industry, remelt of producer scrap by the mi l l products producers, and the recycl ing of UBC in dedicated f a c i l i t i e s .

Centr i fugal pumping systems increase the melt r a t e , reduce energy consumption, improve product iv i ty and reduce thermal heads in furnace r e f r a c t o r y , a iding in longer furnace l i f e . Most important ly , cent r i fuga l pumps enhance melt metal lurgical qua l i t y by homogenizing bath temperatures, overcoming temperature s t r a t i f i c a t i o n , and creat ing greater uniformity in a l loy composition.

In the remelt ing of a homogeneous scrap source, such as with processed UBC, pumping performance can assist in achieving sustained melt rates greater than 20,000 l b s . / h r . Pump longevity is extended s i g n i f i c a n t l y


with homogeneous UBC feed metal and with proper on- l ine maintenance, such that continuous usage averaging 35 days or more is possible without necessitat ing pump replacement or repa i rs .

Metals re f in ing technology for demagging and a l k a l i metal removal, degassing, f i l t r a t i o n , and grain re f in ing have evolved to render the recycl ing of aluminum scrap metal and al loys as meta l lu rg ica l l y as sat is factory as primary metal q u a l i t y . In p a r t i c u l a r , appl icat ion of the gas in jec t ion pump, i n - l i n e or batch rotor degassing, f lux i n j e c t i o n , and integrated f lux i n j e c t i o n / r o t o r dispersion systems o f f e r the aluminum recycler a f u l l range of metal treatment technology choices.

References

1 . D. V. Neff " E f f i c i e n t Melting of Low Density Scrap", Proceedings.. Recycle and Secondary Recovery of Metals, TMS, 1985, P. 51

2. R. E. Sanders, J . R. McBride - "Recycling and Fabricat ion of Used Beverage Cans (UBC's)", Proceedings, Recycle and Secondary Recovery of Metals, AIME, 1985, P. 407

"Submergence of Light Scrap Using a Linear Induction Motor", Proceedings. Recycler and Secondary Recovery of Metals, AIME, 1985, P. 121

"Molten Metal Pumping Systems - Current Applications and Benef i ts" , Light Metals. AIME, 1987, P. 805

"Recent Development in Molten Metal Pumping Applications in Aluminum Recycling", Proceedings, Second Internat ional Symposium Recycling of Metal and Engineered Mate r ia ls , TMS, 1990, P. 105

6. P. J . Bamji, F. W. Pierson - "Electromagnetic C i rcu la t ion of Molten Aluminum", Journal of Metals, AIME, 1985, P. 44

7. M. A. Th ibau l t , F. Tremblay, J . C Pomerlau - "Molten Metal S t i r r i n g ; The Alcan Jet S t i r r e r , Light Metals, TMS, 1991, P. 1005

8 . T. Cunard "Application of Graphite Pump to Direct Charged Aluminum Melting Furnace", Proceedings, Energy Conservation Workshop X, Aluminum Associat ion, 1987, P. 159

9. J . Wojciechowski, W. F. Fundine - "A S t a t e - o f - t h e - A r t UBC Recycling F a c i l i t y " , IMSAMET's Idaho Plant" , Proceedings, Second Internat ional Symposium Recycling of Metals and Engineered Mater ia ls , TMS, 1990, P. 215

10. D, V. Neff "The Use Of Gas In jec t ion Pumps in Secondary Aluminum Metal Ref in ing", Proceedings, Recycle and Secondary Recovery of Metals, AIME, 1985, P. 73

11 . T. J . Johnston, R. D. Peterson - "The Role of Magnesium in Fluxing UBC", Proceedings, Recycled Secondary Recovery of Metals, AIME, 1985, P. 417

3. P. J . Bamji

4. D. V. Neff

5. D. V. Neff


12. C E. Eckert, R. Mutharasan, D. Apel ian, R. E. M i l l e r - "An Experimental Technique for Determining Speci f ic Cake Resistance Values in the Cake Mode F i l t r a t i o n of Aluminum Al loys" , Light Metals. AIME, 1985, P. 1225

13. J . A. Eady, D. M. Smith, J . F. Grandfield - " F i l t r a t i o n of Aluminum Mel ts" , Proceedings. Aluminum Technology "86, I n s t i t u t e of Metals, London. 1986, P. 93

14. R. B. Seese, Personal Communication with author; KB A l loys , March 28, 1991

413

A qualitative assessment of sludge formation in magnesium diecasting furnaces

A. Thorvaldsen, N. Fantetti, C A . Aliravci Institute of Magnesium Technology, Inc., Ste-Foy, Quebec, Canada

ABSTRACT

Sludge generation and melting practices for magnesium alloys have a direct impact on quality and cost of diecast products. This work describes current molten metal handling practices in industry and discusses the sludge generation process.

The average composition of the sludge was found to be:

Oxides 29 ± 14% Intermetallic particles 0.8 ± 0,8% Alloyed metal Balance

Sources of oxides include: surface skin from ingots and liquid metal oxidation in the crucibles. Oxidation in the crucible is suspected to be the most important component. In order to reduce the oxide level, all agitation of the melt surface must be eliminated, the furnaces should be closed and, the protective gas system should be optimized.

KEYWORDS

Sludge, oxides, intermetallics, diecasting, magnesium, melting, dross.

INTRODUCTION

The objective of this study was to gain insight into the generation of sludge and dross in magnesium diecasting operations; four magnesium diecasting foundries were visited and sludge samples were collected and analyzed. To ensure confidentiality, samples are identified without any reference to their origin.

Magnesium diecasting alloys are found to have higher melt losses than their competitors, zinc and aluminum. The underlying cause of this higher melt loss is a higher tendency to oxidize. It is important to understand how oxides are either introduced or generated in the melting crucible and what is the structure of the oxide containing waste. Sludge settled to the bottom of the crucibles or dross floating on the melt surface have a considerable amount of metallic magnesium trapped in their structure. The challenge is therefore not only to reduce the oxidation of the molten metal, but also to change the structure of the sludge and dross in order to reduce metal loss.

This study focusses mainly on the melting practices in diecasting plants.


The objectives of the study were to:

1) describe common melt handling practices in magnesium diecasting operations;

2) develop reliable analytical methods for sludge and dross characterization;

3) characterize sludge with respect to chemical elements and phases present.

COMMON MELT HANDLING PRACTICE

The following report is based on interviews with key technical personnel in four magnesium diecasting plants. A questionnaire aimed at defining the following issues was used:

• Metal suppliers / consumption; • Melting practice / capacity; • Pot cleaning method / frequency; • Shut down - weekends / nights; • Furnace gas protection system / consumption; • Scrap recycling / disposal; • Metal transfer / metering; • Casting technology (machines, casting temperatures etc.).

Meta» Suppliers

The diecasting companies visited have annual consumptions of magnesium alloys ranging from a few hundred to several thousand metric tonnes. They are currently supplied by three magnesium producers in North America. No secondary metal is being purchased for their operations.

The main alloy used by all the companies is AZ91D. Uses of AM60 and AS41 were also reported.

No apparent difference in the sludge formation tendency, in the melts prepared with the ingots acquired from different primary producers, was reported. From the producers own chemical analyses, there seems to be slight differences from one to another in the manganese level in the primary metal. High manganese levels may cause increased precipitation of intermetallics that settle to the bottom and thus increase the amount of sludge formed.

Mating Practice - Capacity

The companies visited have crucible capacities ranging from 200 kg to 1200 kg. Eight out of twenty-five furnaces are gas fired and the rest are electrically heated, crucible furnaces. Several of the gas fired furnaces have burners that impinge direcdy on the side walls of the crucibles. We may speculate that this may increase the natural convection in the molten metal and consequently induce mixing and alter the structure of the sludge.

The ingots are added directly into the melt and this causes temperature fluctuations of ± 10-20°C. When critical parts are produced, one company claims that only one third of an ingot is added which causes temperature fluctuations of ± 5 °C. This company also uses relatively small furnaces. Three of the four companies make no efforts to minimize the melt surface exposed to the charging operation. The fourth company has developed an ingot charging unit that reduces the exposure to air during ingot charging.


Three of the four companies practice ingot preheating where the ingots rest exposed on the top of the furnaces. The fourth company with a charging unit, electrically preheats the ingots to ensure that their ingot temperature is above 100 °C before charging.

In all operations, there is considerable dross formation on the molten metal surface.

Molten metal temperatures vary from 620 °C to 670 °C for the hot-chamber machines and from 650 °C to 710 °C for the cold-chamber machines. The different melt temperatures are a function of the complexity of the cast part, the casting process, and also of the technological level of metal transfer from the crucible to the die. This applies both to cold-chamber and to hot-chamber machines.

The lowest level of sludge is found where the melts are held at the lowest temperatures.

Pot Craning - Frequency

Pot cleaning of dross and sludge is normally accomplished utilizing a perforated ladle allowing some liquid metal to drain back into the crucible. Little attention seems to be given to the settling of the oxides remaining suspended in the melt after the cleaning procedure.

The crucible cleaning frequency is summarized in Table I.

Table I. Frequency of Crucible Cleaning

Company Dross removal Sludge removal

A once a day once a day B once a day once a week C 5 times a day 5 times a day D 3 times a day 3 times a day

A low frequency of sludge removal could have a beneficial effect by increasing the packing density of the sludge and lowering the proportion of entrapped metal.

Based on the data provided by the diecasters, the amount of sludge created in the crucibles varies from less than 2 % to more than 4 % of the annual consumption of metal. The economical importance of sludge generation means that a 10001 operation will generate 20 to 40 tonnes of sludge each year. Based on the best prices for sludge obtained by the interviewed companies and test results of the metallic content of the sludge, a conservative estimate of the loss for each kg of sludge generated is 1 $. This gives a total loss of 20 - 40 k$ per year for each 10001 of metal consumed.

Typically 10 - 20 kg of sludge and dross are removed from each furnace every day depending on the following factors:

• melting rate; • melt temperature; • ingot charging system; • metering system (hand ladling, pumps, siphons, hot-chamber etc.); • gas protection system including gas composition and hood design.


Weekends / Nights - Shut Down

Standard melt holding practice for diecasting is to lower the melt temperature during the weekends between the solidus (468 °C) and the liquidus (596 °C). Typical holding temperatures are given in Table n . Based on the Mg-Al phase diagram, the portion of the melt not solidified at these temperatures (%liquid) is also indicated as well as the concentration of aluminum in this portion (1).

Table II. Melt Temperatures During the Weekends

Company Melt temperature % liquid. %A1 in liq

A below solidus B 540 °C - 2 5 % - 2 0 % C 595 °C ~ 100% - 9 % D 500 - 550 °C 8 - 35 % 2 5 - 1 8 %

Gas Protection • Consumption

All the companies use a variation of SF6 / air / C 0 2 gas mixtures for liquid metal oxidation protection (none use flux). The only use of flux is as a fire extinguisher when the pots are cleaned and when magnesium is machined.

The gas mixture is usually blended at a central supply station. In some plants the quantity of gas protection at each furnace may be adjusted manually by the operators.

Companies using air and SF6 use air/SF6 volume ratios ranging from 400/1 to 4/3. When C 0 2 and SF6 are used, the COJS¥6 volume ratio is 250/1. The important differences in SF6 concentrations are compensated by the gas mixture flow rate into the furnace. Consumption of SF6 per furnace are given in Table III.

Table III. Consumption of SF6

Company SF6 (grams per hour per furnace)

A 2 0 - 2 5 B 25 C 2 5 - 3 0 D 2 5 - 3 0

Metal Transfer - Metering

None of the companies surveyed practice any form of molten metal transfer from furnace to furnace. The ingots are melted directly in the furnace used for casting.


Cold-chamber machines are supplied with metal either by hand ladling or a metering device. A lower level of sludge was found when metering devices are used.

S?ludg? Analyses

Elemental analyses of sludge were performed by Atomic Absorption and Inductively Coupled Plasma/Atomic Emission Spectroscopy at the Centre de Recherches Minerales laboratory in Sainte-Foy, Quebec. Figure 1 shows a typical sludge sample collected from a diecasting melting furnace. Material for analysis was obtained by drilling randomly through the sludge sample. The filings were fully dissolved in HC1 50%. The resulting solution was filtered and the filter ashed in a platinum crucible. The ash was fused with lithium metaborate and the fusion products dissolved in HC1 50% which was added to the first solution. The final volume was 50 ml for a 0.5 g sludge sample. For samples 1 through 5 , Al, Zn, Mn, Fe and Ni were determined by atomic absorption spectrometry. For the other samples, these elements were analyzed by ICP/AES. Beryllium was analyzed by ICP/AES in all samples.

The oxide content was determined by oxidizing a 0.5 g sample and determining the weight increase. A magnesium sample cannot be oxidized directly in a furnace without fume losses. Therefore, it was necessary to use a different approach. The samples were placed in a porcelain crucible.Four ml of H N 03 (20%) was carefully added, drop by drop. The acid addition was then repeated with stronger H N 03 (50%) until effervescence markedly decreased, and finally, concentrated HN03 was added. When the sample was completely dissolved, the solution was dried by introducing the solution crucible into a box furnace at 600 °C. The sample was taken out of the furnace when nitrous evaporation had ceased, i.e. when all nitrates were decomposed into oxides. The drying step was conducted very carefully in order to avoid possible losses. To complete the oxidation, the crucible was introduced into a tubular furnace at 1100°C under an oxygen atmosphere for one hour. Finally, the crucible was cooled down to room temperature in a dessicator and the weight was determined.

The results of the chemical analyses and the oxide determination are given in Table IV. The elemental analysis is the composition of the metallic part of the sludge normalized to 100 %.

RESULTS

Figure 1 - Sludge sample.


Table IV. Chemical Analyses of Sludge Samples

Sample no Mg Al Zn Mn Fe Ni Be % % (%) (%) (%) (%) ppm ppm ppm Metallic Oxides

Plant A Sludge 1

Plant B Sludge 2 Sludge 3 Sludge 4

Plant C Sludge 5

Plant D Dross 6 Dross 7

Sludge 8

Sludge 9 Sludge 10 Sludge 11 Sludge 12

90.0 8.8 0.7

89.3 9.5 0.7 89.5 9.5 0.7 88.1 11.0 0.6

83.8 11.2 0.7

89.6 9.1 0.9 88.5 10.5 0.7

86.0 11.4 0.5

85.5 12.9 0.6 90.9 8.2 0.6 90.1 8.9 0.6 90.6 8.5 0.6

0.4 275 36

0.3 1200 36 0.2 200 28 0.3 476 36

1.1 32000 89

0.3 316 11 0.2 67 4

2.0 1242 1

1.0 522 6 0.2 623 29 0.2 634 32 0.2 444 29

61 79 21

39 71 29 96 79 21

176 69 31

102 40 60

41 24 76 333 63 37

284 92 8

326 76 24 64 59 41 31 76 24 78 70 30

Figure 2 - Oxides and oxide films in sludge.

50 |im


DISCUSSION

Three main components of sludge have been identified in this study: oxides, intermetallic particles and entrapped metal. The average sludge composition is:

Component weight%

oxides 29 ± 1 4 intermetallic particles 0 .8±0.8 entrapped metal 70 ± 1 4

The amount of intermetallic particles is based on the manganese-analyses. The concentration of manganese in the entrapped metal is supposed to be in the order of 0.2%. The rest is combined with aluminum to form intermetallic particles (AlxMny) (2). Apart from the high level of entrapped metal, the analyses show that oxides predominate the sludge composition. It should be noted that the sludge is heterogeneous and sludge samples collected from the same furnace differ substantially in compo-sition. For this reason, the composition above must be regarded as average only with wide variations.

Oxides may originate from the primary metal as internal inclusions or, surface skin. The contribution from the internal oxides is assumed to be small. The amount of oxides due to the surface skin depends on the surface/volume ratio of the ingots and the thickness of the skin. Ingots from three main producers of magnesium diecasting alloys in North America have surface/volume ratio in the range of 1.4 -1.7 (dm1). If all the oxides in the sludge generated originated from the ingots only, the skin thickness should have been in the range of 0.1 to 0.2 mm. This is far thicker than that normally observed in metallographic samples. Figure 3 shows the cross-section at the surface of a magnesium ingot confirming that surface skin is very thin (<0.01 mm). We can therefore conclude that the surface skin is not the dominating source of oxides in the sludge. Oxide films found in the sludge, as shown in Figure 2, are interpreted as surface skin oxides originating from the ingots. The thickness of these films are in the same order of magnitude as the surface skin (0.001 - 0.01 mm.).

Figure 3 - Cross section of the surface of a magnesium ingot.


A glance at the melt-surface in any magnesium diecasting operation reveals the other source of oxides since heavy dross can be seen floating on top of the surface. Under the microscope, the dross-oxides appear thick and mushy as can be seen in Figure 2. Three factors may greatly influence the formation of dross: the level of agitation of the melt surface, the melt temperature and the gas-protection system.

Agitation of the melt surface occurs every time an ingot is dropped into the melt. Depending on the height and the manner of dropping, various levels of surface disturbance may occur. From a purely agitational point of view, a skilled operator that slides the ingot carefully into the melt is achieving the best of all the present charging systems. On the other hand, the necessity to open the hood door disturbs the crucible atmosphere. This has an opposite and strongly negative effect on dross formation.

Manual discharging with hand ladles is the other major source of agitation of the melt. A ladle collecting metal for the cold-chamber machine creates a considerable amount of disturbance which enhances oxidation. The level of dross formation depends heavily on the operator. With adequate training an operator may fill the ladle and transfer the metal to the shot-sleeve without visible burning. In any case, the surface is agitated and some oxidation will occur. The hood door has to be opened leading to the contamination of the crucible atmosphere. Automatic transfer systems may to a great extent eliminate the agitation and the need for door openings. Even then, valves in the melt may disturb the melt surface and the system may introduce air at the point where the transfer tube goes through the hood.

In hot-chamber systems there is no ladling. Molten magnesium is introduced into the gooseneck below the melt surface. Ideally, there should be no disturbance of the molten metal. However, on each injection cycle the piston shaft normally goes unprotected through the melt surface. Thus, every stroke of the piston emits waves throughout the crucible and these waves have a similar effect on dross formation in hot-chamber diecasting as manual discharging in cold-chamber diecasting.

The oxidation rate of molten magnesium is directly influenced by the melt temperature. At temperatures near the liquidus, the molten metal is relatively unreactive. At higher temperatures, the melt ignites more readily if the surface is disturbed. The influence of the temperature on the oxidation rate under various gas atmospheres and different levels of agitation has not been investigated.

The consumption of SF6 per furnace appears to be consistent amongst the four diecasters interviewed. Typical levels of SF6 were found to be approximately 25-30 grams per furnace per hour. This is comparable with consumption previously reported by Busk and Jackson (3), although they related the SF6 level to the production rate (SF6 per lb) and therefore found a great variance in the SF6-usage from diecaster to diecaster. The present study shows that the consumption of SF6 is relatively unaffected by the production rate on the diecasting machine.

The optimum conditions of SF6 consumption are not known to the industry. A too low level of gas-protection will increase the oxidation rate and enhance the melt loss. On the other hand, a high level will increase the cost of gas protection to an unacceptable level. The minimum cost level is were the loss due to oxidation is balanced with the cost of SF6.

An important factor affecting the gas protection is the design and operation of the furnace door; charging and discharging. Fresh, cold air flows into the furnace chamber while protective SF6-containing gas leaks out each time the door is activated. This occurs at the same time as the surface is disturbed by the ingot charged to the crucible or the ladle is removing molten metal.


Intermetallic particles may originate from the ingots or they may be precipitated in the diecasting crucible. These particles play an important role as precipitation sites for Fe, and thereby lowering the level of dissolved Fe and increasing the overall corrosion resistance of the diecastings. Since these particles represent approximately 1 % of the sludge content, there is little gain by reducing the amount of intermetallics.

CONCLUSIONS

This study reveals that:

• the amount of sludge removed from the melting crucibles vary between 2 and 4% of total metal input for the surveyed diecasters;

• the three major constituents in sludge are in the following proportion:

oxides: 29% ± 14% intermetallics: 0.8 ± 0.8% entrapped metal: 70 ± 14%;

• the oxide content is mainly due to process parameters rather than to the oxide level in the ingots. Ingot oxides represent less than 10% of total oxides in the sludge.

If the oxidation of molten metal in the diecasting crucibles can be drastically reduced and the manganese content of the primary metal is kept at a low level, a reduction in the sludge formation by a factor of 10 may be obtainable. This would require completely covered furnaces, careful control of the charging and discharging operations, new shapes and sizes of the primary metal ingot and, optimized furnace atmosphere.

REFERENCES

1. A. A. Nayeb-Hashemi and LB. Clark, "Phase Diagrams of Binary Magnesium Alloys", ASM International, Ohio, 1988.

2. C. J. Simensen, B.C. Oberlander, J. Svalestuen and A. Thorvaldsen, Z. Metallkunde, 79, p. 696, 1988.

3. R.S. Busk and R.B. Jackson, Proc. 37

th World Conference on Magnesium, International

Magnesium Association, 1980.

423

The effects of V and Be microadditions on age-hardening behaviour of Al-Li alloy 8090

A. Luo, W.V. Youdelis Department of Mechanical Engineering, University of Windsor, Windsor, Ontario, Canada

ABSTRACT

The effects of microadditions of V and Be on the age-hardening behaviour and microstructure of Al-Li alloy 8090 are investigated. Small additions of V and Be significantly increase the peak hardness levels of the alloy on aging. An optical, SEM/EDS, and X-ray diffraction investigation shows a higher S (Al2CuMg) and 8(AlLi) precipitate density for alloys containing V and Be, indicating an enhanced nucleation rate for their precursor phases, S \ 5' and GPB zones.

KEYWORDS

Al-Li alloy, Age-Hardening, Microalloying Al Alloy

INTRODUCTION

Al-Li alloys offer the promise of substantial weight savings in aerospace application by virtue of their reduced density and increased elastic modulus compared with conventional aluminum alloys (1). The increased strength of Al-Li alloys is attributed to precipitation of 5'(Al3Li); however, 5' precipitation also lowers ductility and toughness by strain localization and PFZ (precipitate free zone) formation. In recent years, research on Al-Li alloys has concentrated on improving ductility and toughness in two alloy systems: the ternary Al-Li-Cu alloys (2090 type) in which 5' precipitation may be supplemented by the formation of T^A^CuLi) and/or 0' (CuAl2); and the quaternary Al-Li-Cu-Mg alloys (8090 type) in which the precipitation of S' (Al2CuMg) can occur (2).

The presence of a second precipitating phase can alter the deformation mode (3-6). In the case of 8090 alloy containing S' phase, extensive cross-slip occurs to give irregular slip lines, indicating S' phase is not sheared by glissile dislocations. Tl phase was found less effective in dispersing slip than S' (3). Also, precipitation of S' does not result in PFZ's, along either low- or high-angle grain boundaries (7). Both Tx and S' nucleate heterogeneously on dislocations and other defect sites, and hence the practice of prior-aging deformation (3-5% stretch) to ensure widespread nucleation in the matrix (8). The prior-aging deformation step is not feasible for some alloy products, and so the enhancement of nucleation by other means must be explored. Pickens and co-workers (12,13) have shown that relatively small amounts of Ag and Mg are extremely effective in stimulating Tx


precipitation in a Al-Li-Cu alloy, and led to the patent of Weldalite™ 049. It has also be shown that small addition of Ag (9), Si (10), and Be (11) in Al-Cu-Mg alloys increase the nucleation rate and refinement of the S' precipitate, with an attendant increase in the peak hardnesses, for the age alloys. The above suggests a similar study in search of suitable microalloying elements to stimulate S' phase precipitation in 8090 alloy.

The object of the present research is to study the effects on the age-hardening behaviour and microstructure of the Al-Li alloy 8090 by microalloying with V and Be.

EXPERIMENTAL

Materials

The as-cast Al-Li-Cu-Mg-Zr alloy 8090 (base alloy) was provided by Alcan International (Kingston R & D Centre). A1-5V

1 and Al-5.23Be master alloys were used in

the preparation of the base alloys containing 0.30V, 0.60V, 0.15Be, and 0.30V 4- O.lOBe (designated as 30V, 60V, 15Be, and 30V10Be respectively). The alloys were prepared in graphite crucibles by induction melting under argon to prevent the possible loss of lithium. The melts were heated to well above the liquidus temperature (~800°C), maintained for about 10 min. to ensure complete homogenization and then poured into graphite molds 25 mm (dia.) x 70 mm (length) at room temperature. The nominal compositions of the alloys in this study are given in Table I.

Table I - Nominal Composition of Alloys

Alloy Composition (wt%)

Li Cu Mg Zr V Be Al

8090 2.47 1.24 0.77 0.10 Bal. 30V 2.33 1.17 0.73 0.09 0.30 — Bal. 60V 2.21 1.11 0.69 0.09 0.60 — Bal. 30V10Be 2.29 1.15 0.71 0.09 0.30 0.10 Bal. 15Be 2.40 1.20 0.75 0.10 - 0.15 Bal.

Heat Treatment

The as-cast ingots were annealed for 40h at 590°C for homogenization. For the age-hardening study, cylindrical sections ~5mm thick were cut from the central region of the ingots. Several solution treatment temperatures (560°C, 590°C and 620°C) and times (1.5h and 4h) were investigated to determine solution treatment for optimum aging response for the alloys. The 620°C solution treatment temperature was attempted with the aim of dissolving the maximum amount of V and Be in solid solution. The solution treatment was followed by a quench in iced brine. The samples were then aged at four temperatures: room temperature (22°C), 190°C, 240°C and 385°C. The solution and aging treatments were performed under ambient conditions, resulting in some surface oxidation. The surface oxide layer, estimated at less than 0.5mm, was removed by surface grinding.

JA11 concentrations in wt.%.


Hardness Measurement and Microstructure Analysis

Microhardness measurements (HV0.05) were used to monitor the age-hardening process by periodically interrupting the aging treatments. At least five readings were randomly taken for each hardness determination to obtain a mean with a typical uncertainty of ±5%. Specimens for optical and SEM microscopy were prepared in the conventional manner, starting with 1 /xm and then 0.05 /xm alumina, and finishing with colloidal silica suspension for the polishing media. The etchant used was dilute Keller's etch. Debye-Scherrer X-ray diffraction analyses were performed on the bulk specimens, using CuKa radiation to identify the major phases.


Age-Hardening Behaviour of 8090 Alloy

The effect of solution temperature and time on room temperature aging for 8090 alloy is shown in Figure 1. The maximum hardness is obtained for a solution temperature of 590°C. Hardness is also slightly increased by increasing the solution time from 1.5h to 4h at 590°C. When the solution temperature was increased to 620°C, the peak hardness decreased. To obtain maximum peak hardnesses, the solution treatment used for all of the alloys investigated was a 4h anneal at 590°C.

en

u

K

1 0 0

9 0

8 0

7 0

6 0

5 0

4 0

3 0

R o o m T e m p e r a t u r e

A 5 6 0 ° C . 1 . 5 h

x 5 9 0 ° C . 1 . 5 h

* 5 9 0 ° C . 4 h

o 6 2 0 ° C . 1 . 5 h I t I 1 > I T I I I I I \ ~ \ "\ I I I » '

1 0 0 2 0 0

i i i I i i i i i i i 3 0 0 4 0 0

I i i i i i i i i i I 5 0 0

A g e T i m e , h . o u r

Figure 1 - Age-hardening curves for 8090 alloy for various solution temperature and times.

In order to determine the effect of a natural (room temperature) age on preceding higher-temperature age, a duplex aging treatment, consisting of a 70h natural aging followed by aging at 190°C, was conducted. Figure 2 shows the effect of 70h natural aging on the 190°C aging response. Duplex aging slightly accelerates the age-hardening and also gives a slightly higher peak hardness and stability (resistance to overage) for the alloy.

An acceleration in age-hardening of Al-Li alloy by a duplex aging treatment has been reported by other investigators (14-16). According to Flower et al (14), homogeneous nucleation of S' phase is dependent on the density of free vacancies, which is a function of


solution temperature. The vacancies are strongly bound to Li and as such are not active for nucleation. However, during low temperature aging, precipitate particles grow and incorporate lithium, releasing the bound vacancies for S' nucleation, which is evidenced by dislocation loop and helix formation.

1 4 0 3

2 0 -=1 1 1—r i i r i 11 1 1—i i i i i 11 1 1—« i i i r i | 1 • i i

1 1 0 1 0 0 1 0 0 0

A g e T i m e , m i n . Figure 2 - Effect of a 70h natural aging on the age-hardening curve at 190°C for 8090 alloy.

The considerable variation in peak hardness with solution temperature requires comment. Flower et al (14) reported a slight increase in the hardness when the solution temperature increased from 540°C to 580°C, and attributed this to an increased vacancy concentration after quenching. The lower hardness values for the room temperature aging curve for the alloy solution treated at 620°C is believed due to the solution temperature exceeding the solidus temperature for the alloy. The solidus temperature of alloy 8090 may be estimated by using the experimental temperature-composition formula provided by Dorward (18), i.e.,

T {solidus) - 730-24.5 (%Li) -33.2 (%Cu) -49.0 (%Mg) +10 (%Mg)2

- 730-24.5 x 2.47 33.2 x 1.24-49.0 x 0.77+lOx (0.77)

2

- 596.5 (°C) (1)

Effect of V and Be on Age-Hardening of 8090 Allov

The age-hardening curves for the alloys in Table I are shown in Figure 3(a) to (c). The room temperature age-hardening results for alloy 8090 reported by Welpmann et al (17) show good agreement with the present investigation. The V- and Be-containing alloys (30V, 60V, 30V10Be and 15Be) attain hardness maximums up to 24 points above those for the base alloy. The highest peak hardnesses are obtained for the 190°C age, with 15Be at —HV142 and followed by 60V at -HV132 . The addition of both V and Be together decreased the beneficial effects of the individual components (V or Be) in the alloy. The time for maximum hardness is significantly decreased when aging at 240°C, but at the cost of a significant decrease in peak hardness.

a l 9 0 ° C A g e

a D u p l e x A g e

CO on

a u

a

U

O

>


R o o m T e m p e r a t u r e

Age Time, hour 1 90°C

100 Age Time, m i n .

100

i — i — i i i 1 1 11 i — i — i i i 11 n —

10 100 1000 Age Time, m in . Figure 3 - Age hardening curves for alloys aged at various temperatures.

427

OS cr <L G U

«d a M o

>

a 8 090 A 3 OV * 3 0V10BO 0 6 O V x 1 5 B e

on on

d

« 00 a

o 8090 A 3 OV * 3 0 V 1 OBo 0 60V x 1 5 B e

o 8 09 O ^ 3 OV 4c 3OV10BG 0 6 OV x 1 5Bo

00 09

( -1

>


Microstructure

The as-cast microstructures of the alloys are shown in Figure 4. Extensive microsegregation of alloying components occurs in the 8090 base alloy during solidification (Figure 4(a)), and in V-containing alloys a relatively coarse V-rich precipitate is evident (Figure 4(b)). The precipitate, identified as A l nV by both X-ray diffraction and EDS analysis, is present in all the V-containing alloys (30V, 60V and 30V10Be). The finer precipitates, evident in the microstructure of Figure 4(b), are assumed to be 5(AlLi), S(Al2CuMg) and T-type phases, but only T2(Al6CuLi3) could be identified by X-ray diffraction.

(b) 60V

Figure 4 - As-cast microstructure for 8090 and 60V alloys.

An optical and SEM study was carried out to determine the effect of V and Be on the distribution and density of the precipitating phases after aging. To obtain optically resolvable precipitates requires severely overaging the alloys, which was accomplished by a 30 min. age at 385°C. The precipitates in the optical microstructures shown in Figure 5(a)-(c) are identified as S and 5 by their morphology and Cu content (EDS analysis), with the lath-like S particles containing a high Cu content. The round-like particles are d phase and show no

(a) 8090


(a) 8090

(b) 60V

(c) 15Be

ure 5 - Microstructures of alloys aged for 30 min. at 385°C, with lath-like S particles and round-like 5 particles.


Cu content. T2-type phase, identified by X-ray diffraction, is found at grain boundaries (Figure 6). Crooks and Starke (5) made a similar observation and reported the diffraction pattern of the phase. It is noted that no PFZ's are present along grain boundaries, even around the T2-type phase, in contrast to precipitation in Al-Li-Cu alloys, which shows extensive PFZ formation (7).

The morphologies of S and 5 phases are more clearly shown in the SEM micrographs of Figure 7(a)-(c). Both the optical and SEM micrographs show a higher density of precipitate particles in V- and Be-containing alloys, with the highest precipitate density the 15Be alloy. The higher S and precipitate densities in V- and Be-containing alloys suggest an enhanced nucleation rate for the precursors, i.e., S' and 6' phases, and GPB zones.

Figure 6 - T2-type phase at grain boundaries for alloy 8090 (aged for 30 min. at 385°C).

(a) Optical (b) SEM


(a) 8090

(b) 60V

(c) 15Be

Figure 7 - Microstructures of alloys aged for 30 min. at 385 °C, with lath-like S particles and round-like 5 particles.

431


CONCLUSIONS

1. Small additions of V and Be increase the age-hardening rate and peak hardness for the 8090 alloy.

2. The increase in density of S and 5 precipitates resulting from the additions of V and Be to the 8090 base alloy suggests an enhanced nucleation rate for the transformation.

3. A duplex aging treatment enhances the precipitation hardening response of the 8090 alloy.

4. To obtain the optimum age-hardening response, the solution temperature should approach but not exceed the solidus temperature, estimated at 596°C.

5. The addition of both V and Be together decreases the beneficial effects of the individual components (V or Be) in 8090 alloy.

ACKNOWLEDGEMENTS

The authors wish to acknowledge the Natural Sciences and Engineering Research Council of Canada for financial support (OGP 836) and Alcan R & D Centre, Kingston, Ontario, Canada, for providing the 8090 alloy and Al-V and Al-Li master alloys.

REFERENCES

1. E.J. Lavernia and N.J. Grant, "Aluminum-Lithium Alloys," J. Mater. Sci.. Vol. 23, 1987, 1521.

2. P.J. Gregson and S.J. Harris, "Foreword to Book II: Physical Metallurgy of Al-Li Alloys," Aluminum-Lithium III. Proc. 3rd Int. Al-Li Conf.. Inst, of Met., London, 1986, 327.

3. P.J. Gregson and H.M. Flower, "Microstructural Control of Throughness in Aluminium-Lithium Alloys," Acta Met.. Vol. 33, 1985, 527.

4. W.S. Miller, M.P. Thomas, D.J. Lloyd and D.K. Creger, "Deformation and Fracture in Al-Li Base Alloys," Aluminum-Lithium HI. 584 (see ref. 2).

5. R.E. Crook and E.A. Starke, Jr., "The Microstructure and Tensile Properties of a Splat-Quenched Al-Cu-Li-Mg-Zr Alloy," Metall. Trans. A. Vol. 15A, 1984, 1367.

6. R.E. Crooks, E.A. Kenik and E.A. Starke, Jr., "HVEM In-Situ Deformation of Al-Li-X Alloys," Scr. Met.. Vol. 17, 1983, 643.

7. T.H. Sanders, Jr. and E.A. Starke, Jr., "The Physical Metallurgy of Aluminium-Lithium Alloys — A Review," Aluminum-Lithium V. Proc. 5th Int. Al-Li Conf.. MCE, 1989, 1.

8. D.J. Lloyd, private communication, Alcan R & D Centre, Kingston, Ontario, June 1989.

9. I.J. Polmear, "The Effects of Small Additions of Silver on Aging in Some Aluminium Alloys," Trans. Met. Soc. AIME. Vol. 230, 1964, 1331.


10. R.N. Wilson, D.M. Moore and J.E. Forsyth, "Effects of 0.25% Silicon on Precipitation Processes in an Aluminium-2.5%Copper-1.2%Magnesium Alloy," L Inst. Met.. Vol. 95, 1967, 177.

11. W. Fang and W.V. Youdelis, "Effect of Be on S Phase Precipitation in Pseudobinary Al(a)-Al2CuMg Alloy," Proc. Light Met. Symp.. Met. Sci. CIM, Hamilton, 1990, 279.

12. J.R. Pickens, F.H. Heubaum, TJ. Langan and L.S. Kramer, "Al-4.5-6.3Cu-l.3Li-0.4Ag-0.4Mg-0.14Zr Alloy Weldalite 049," Aluminum-Lithium V. 1397 (see ref. 7).

13. T.J. I^ngan and J.R. Pickens, "Identification of Strengthening Phases in Al-Cu-Li-Alloy Weldalite 049," Aluminum-Lithium V. 691 (see ref. 7).

14. H.M. Flower, P.J. Gregson, C.N.J. Tite and A. Mukhopadhyay, "The Effect of Composition and Heat Treatment upon Microstructure/Property Relationships in Al-Li-Cu-Mg-Zr Alloy," Aluminium Alloys. Their Physical and Mechanical Properties. Eng. Mater. Advisory Services, 1986, 734.

15. X. Xia and J.W. Martin, "The Effects of Stretch and Heat Treatment on Microstructure and Mechanical Properties of an Al-Li-Cu-Mg-Zr Alloy," Mater. Sci. Eng.. Vol. A128, 1990, 113.

16. V. Radmilovie, G. Thomas, G.J. Shiflet and E.A. Starke, Jr., "On the Nucleation and Growth of Al2Cu Mg (S') in Al-Li-Cu-Mg and Al-Cu-Mg Alloys," Scr. Metall.. Vol. 23, 1989, 1141.

17. K. Welpmann, M. Peters and T.H. Sanders, Jr., "Age-Hardening Behaviour of DTD XXXA," Aluminum-Lithium III. 524 (see ref. 2).

18. R.C. Dorward, "Solidus and Solvus Isotherms for Quaternary Al-Li-Cu-Mg Alloys," Metall. Trans. A. Vol. 19A, 1988, 1631.

http://Al-4.5-6.3Cu-l.3Li-

435

Permissible gas content in aluminum alloy melts relative to the cooling rate of the casting

S. Kitaoka NIKKEI Techno-Research Co. Ltd., Shizuoka-ken, Japan

K. Nishina Nippon Light Metal Co., Ltd., Shizuoka-ken, Japan

A B S T R A C T

Pure aluminum and several aluminum alloy melts with various gas content were cast into permanent molds which show different cooling rate. Porosity measurement was carried out on the cast ings derived to understand the relationship between gas content in melt and amount of porosity formed in the casting relative to cooling rate which was measured as solidification t ime. Porosity was found to appear when the gas content in melt exceeds an permissible gas content which is dependent on chemical composition of alloy and cooling rate. Relationship between solidification time and permissible gas content obtained from density measurement was found to be different from that of dye penetrant test.

KEYWORDS

aluminum alloy, pure Al, AA-6063, JIS-AC1A, JIS-AC2B, JIS-AC7A, JIS-AC9A, hydrogen gas, degassing, cooling rate, solidification time, porosity


INTRODUCTION

Porosity is one of the most serious problems to be solved to provide aluminum alloy castings with sound quality and high productivity. In the production of aluminum alloy castings such as automobile wheels, porosity is usually reduced to a low level so as to satisfy mechanical property requirements, especially, on elongation and ductility. However, some of the castings, such as cylinder heads of automobile engines, which have to be pressure-tight, with very complicated internal shape made by using a lot of sand cores, sometimes require porosity to a low extent in order to compensate a part of shrinkage of alloy during solidification. These examples lead to the necessity of controlling a porosity level properly according to castings to be produced.

Degassing is a major process in molten aluminum treatment, and also essential to eliminate hydrogen gas from aluminum melt. As for the relationship between gas content and amount of porosity, it is well known that there exists threshold gas content, below which porosity will not appear and above which porosity increases linearly as the gas content increases(l). Relation between gas content and porosity varies according mainly to alloys used and solidification time and/or cooling rate of the castings. Although it is very important to know the exact data on the threshold gas content, i.e. permissible gas content, which will permit the production of sound castings, there are still very little data published. In the present invest igat ion, several al loys were selected and subjected to tests to obtain the relationship among gas content, solidification time and porosity.

E X P E R I M E N T A L

99.8% pure aluminum and typical commercial alloys including AA-6063, JIS (Japanese standard) -AC1A (Al-4.5Cu), JIS-AC2B (Al-6Si-3Cu), JIS-AC7A (Al-5Mg), JIS-AC9A (Al-23Si- lCu-lMg-lNi) were selected for tests. 10 kg of each metal was melted in a # 4 0 crucible in a furnace heated by electrical resistance. Melting and casting temperature of the alloys other than AC9A was 993K. AC9A, hypereutectic Al-Si alloy which contains 23%Si, was melted and cast at 1073K. Molten metal treatment such as degassing or gas enrichment was carried out to obtain five levels of gas content in the range between 0.1 and 0.8 ml H2 (STP)/100gAl.

Metals were cast into three types of mold. The first one was a mold with a cavity of 50 mm x 50 mm x 120 mm high, placed on the 50 mm thick cast iron


plate, and surrounded by ceramic fiber insulator to obtain unidirect ional solidification. Three small cubic samples for density measurement were derived at 20 mm, 50 mm, 80 mm from the bottom of each casting (designation for each sample: U l , U2 , U3). The second one was a cast iron mold which is usually used to cast 5 kg foundry alloy ingot. One small cubic sample for density measurement was cut from the center of the block from each casting (designation: IN). The third one was a cup type mold made of cast iron with a cavity 40 mm dia. x 80 mm high. One small cylindrical sample was derived at the bottom of each casting. Three molds were prepared and mold temperatures were controlled at room temperature, 473K and 673K, respectively (designation: CI, C2, C3).

Cooling curves of the castings were recorded at the point where density measurement samples were cut off. Solidification time was obtained as the time necessary for cooling until the temperature reached to 773K after pouring the melt. Porosity was also observed on the cross section of each casting after dye penetrant test. The dissolved gas content in each melt was determined for the sample cast into copper made Ransley mold using subfusion method. Density measurement was carried out using Archmedes method, where samples for the measurement of theoretical values were derived from the permanent mold castings which were cooled rapidly in a cavity of 30 mm x 30 mm x 10 mm with a feeder.

R E S U L T S

Pores on the cross sections of the three types of castings were examined after dye penetrant test. Results of the tests on the unidirectionally solidified AC1A and AC2B are shown in Figures 1 and 2, as typical examples. In the case of AC1A, when the gas content was low (0.15 ml H2/IOO g Al), only a small amount of pores was found in the upper part of the casting. Amount of pores increased gradually as the gas content increased. Size of pores increased as the distance from the bottom increased. In the case of AC2B, overall tendency as to the relationship among gas content, amount of pores and pore size with regard to location in the casting was not changed. However, the amount of pores and size of pores seemed to be much more than AC1A.

Pores on the cross section of cylindrical castings of AC1A and AC2B are shown in Figures 3 and 4. Porosity levels of both alloys were very low as compared to those of unidirectionally solidified castings. Pores of both alloys increased as the gas content increased and also the mold temperature increased. The pore size in the castings of AC2B was larger than that of AC1A, as was observed in the case of unidirectionally solidified castings.

0.1

5 0

.34

0.3

8 0

.42

0.5

4

Ga

s c

on

ten

t,

ml/

10

0g

Al

Fig

ur

e 1

- P

or

osi

ty

in t

he

cr

oss

se

cti

on

of

un

idir

ec

tio

na

lly

soli

dif

ied

AC

1A


XTRACTION, REFINING AND FABRICATION OF LIGHT METALS 439

0.1

6 0

.23

0.3

1 0

.44

0.5

7

Ga

s c

on

ten

t,

ml/

10

0g

Al

Fig

ur

e 2

- P

or

osi

ty

in

the

cr

oss

se

cti

on

of

un

idir

ec

tio

na

lly

soli

dif

ied

AC

2B

4 4 0 EXTRACTION, REFINING AND FABRICATION OF LIGHT METALS

G a s c o n t e n t ,

m l / 1 0 0 g Al

2 9 3 4 7 3 6 7 3

M o l d t e m p e r a t u r e , K

F i g u r e 3 - P o r o s i t y in t h e c r o s s s e c t i o n o f A C 1 A c y l i n d e r



m l / 1 0 0 g A!

2 9 3 4 7 3 6 7 3

M o l d t e m p e r a t u r e , K

F i g u r e 4 - P o r o s i t y in t h e c r o s s s e c t i o n o f A C 2 B c y l i n d e r

441


The relationship between gas content of the melt and density of the sample for all castings on which density measurement was carried out, is summarized in Figures 5 to 10. All of the metal and alloys tested showed the same tendency as to the relationship between gas content of the melt and porosity of the sample. A set of seven lines were obtained for each of the metal and alloys. Each line represents different cooling conditions in the castings (Ul-~3, IN, C1--3), every line has a threshold value below which porosity does not exist and over which porosity increases linearly as the gas content increases. Both of the threshold value and the slope of the line depend on the alloys tested and the cooling speed of the sample. Increase in cooling speed leads to higher threshold gas content and gentler slope. Porosity level of 99.8% pure Al was lower as compared to the alloys tested. Among the alloys tested AC7A and AC9A showed relatively higher porosity level.

Permissible gas content obtained from density measurement data as threshold gas content with regard to the solidification time for AC1A and AC2B is summarized in Figures 11 and 12. Permissible gas content obtained from dye penetrant test is also summarized in the same figures. Both of the permissible gas content were decreased as the solidification time increased. It was found that the permissible gas content obtained from density measurement was lower than that from dye penetrant test in the case of AC1A, on the other hand the adverse relation between the two was obtained in the case of AC2B.

Permissible gas content obtained from density measurement and dye penetrant test of all of the metal and the alloys is summarized in Figures 13 and 14, respectively. AC9A, hypereutectic Al-Si alloy, showed very high level of permissible gas content. AC7A, Al-Mg alloy, showed a steep drop of the line when the solidification t ime exceeded 400 sec. Almost all l ines except AC9A, permissible gas content fell in the range between about 0.1 and 0.3 ml H2/IOO g Al.

DISCUSSION

Utilization of solidification analysis enables us to predict solidification time which is necessary to determine the target of hydrogen gas level in melt. If gas content in melt is lower than permissible gas content, the casting will be sound without porosity, otherwise there will be a certain amount of pores, which results in deterioration of mechanical properties. For this reason, control of gas content is very important. It has been difficult to introduce such technology into daily

Ga

s c

on

te

nt

ml/

10

0 g

Al

Ga

s c

on

ten

t,

ml/

10

0 g

Al

Fig

ur

e 5

- R

ela

tio

ns

hip

b

et

we

en

ga

s c

on

ten

t a

nd

po

ro

sity

in

F

igu

re

6-

Re

lati

on

sh

ip

be

tw

ee

n g

as

co

nt

en

t a

nd

p

or

os

ity

^

va

rio

us

ca

stin

gs

(99

.8 %

Al)

in

va

rio

us

ca

sti

ng

s (6

06

3)

g

A

U1

—-

U2

a

U3

o

IN

• C

1 a-—.

C2

• C

3

1 1

A y |

A

U2

a

U3

o IN

• C

1 a-

-—

C2

• C

3

Po

ro

sity

,

ml/

10

0g

Al

1.0

-

Po

ro

sity

,

ml/

10

0g

Al


Ga

s c

on

ten

t,

ml/

10

0g

Al

Fig

ur

e 7

- R

ela

tio

ns

hip

b

et

we

n g

as

co

nte

nt

an

d

pr

osi

ty

in v

ar

iou

s c

ast

ing

s (A

C1

A)

Ga

s c

on

ten

t,

ml/

10

0g

Al

Fig

ur

e 8

- R

ela

tio

ns

hip

b

et

we

n g

as

co

nt

en

t a

nd

p

ro

sity

in

va

rio

us

ca

sti

ng

s (A

C2

B)

Po

ro

sity

,

ml/

10

0g

Al

Po

ro

sity

,

ml/

10

0q

Al

A

U1

a-

—

U2

a U

3

o

IN

• ci

C

2

• C

3

i i

A

U1

—

—

U2

a U

3 o

IN

• ci

C

2 •

C3

i i


0 0

.1

0.2

0

.3

0.4

0.5

0

.6

0.7

0

.2

0.3

0

.4

0.5

0

.6

0.7

0.8

0.9

Ga

s c

on

ten

t,

ml/

10

0 g

G

as

co

nte

nt,

m

l/1

00

gA

I

Fig

ur

e 9

- R

ela

tio

nsh

ip

be

tw

ee

n g

as

co

nte

nt

an

d F

igu

re

10 -

R

ela

tio

ns

hip

b

et

we

en

ga

s c

on

te

nt

an

d p

ro

sity

in

va

rio

us

ca

stin

gs

(AC

7A

) p

ro

sity

in

va

rio

us

ca

sti

ng

s (A

C9

A)

Po

ro

sity

,

ml/

10

0g

Al

Po

ro

sity

,

ml/

10

0g

Al

1.0

—

a

U1

a-

—

U2

a U

3

o

IN

•

C1

c C

2

• C

3

i i

) i

A

U1

^ y

2

a U

3

o

|N

• C

1

a

C2

• C

3

: 1

1


4 4 6 EXTRACTION, REFINING AND FABRICATION OF LIGHT METALS

0 . 6

0 . 4


m l / 1 0 0 g Al

0 . 2

0

1 i 1 1 i

f

f

•

r o m d y e p e n e r o m d e n s i t y n

' t r a n t t e s t l e a s u r e m e n t -

•

0

• • • • • • -

•![ • I

C

1

•

5 «

I t

c

DE 1

>

i i

-11 1 1 1 1 I I I I • I

0 2 0 0 4 0 0 6 0 0 8 0 0 1 0 0 0

S o l i d i f i c a t i o n t i m e , s e c

F i g u r e 11 - P e r m i s s i b l e g a s c o n t e n t o f A C 1 A in r e l a t i o n t o s o l i d i f i c a t i o n t i m e

0 . 6

0 . 4


m l / 1 0 0 g Al

I i i i

-f r o m d y e p e n f r o m d e n s i t y

e t r a n t t e s t m e a s u r e m e n t

_ o

a '—-

-

D 0

i

D

•

• •

i ' I 1 1 1 I I I I I I I

0 2 0 0 4 0 0 6 0 0 8 0 0 1 0 0 0

S o l i d i f i c a t i o n t i m e , s e c

F i g u r e 1 2 - P e r m i s s i b l e g a s c o n t e n t o f A C 2 B in r e l a t i o n t o s o l i d i f i c a t i o n t i m e

EXTRACTION, REFINING AND FABRICATION OF LIGHT METALS 4 4 7

0 . 6

S o l i d i f i c a t i o n t i m e , s e c ,

F i g u r e 1 3 - P e r m i s s i b l e g a s c o n t e n t o b t a i n e d f r o m d e n s i t y m e a s u r e m e n t in r e l a t i o n t o s o l i d i f i c a t i o n t i m e

F i g u r e 1 4 - P e r m i s s i b l e g a s c o n t e n t o b t a i n e d f r o m d y e p e n e t r a n t t e s t in r e l a t i o n t o s o l i d i f i c a t i o n t i m e

0 . 4 —

G a s -c o n t e n t ,

m l / 1 0 0 g A l

V


m l / 1 0 0 g Al

S o l i d i f i c a t i o n t i m e , s e c ,

I 1 1 1 1 1 1 1 1 1 1


operation in foundries, however it has become easier than ever by the recent development of degassing system and hydrogen measurement system.

Permissible gas content obtained from density measurement is generally lower than that from dye penetrant test except AC2B. This exception seems to be caused by the shrinkage behavior of the alloy. AC2B is an Al-Si-Cu alloy which has wide semisolid range(2) with relatively higher level of eutectic liquid left at the end of solidification, because of its higher content of copper.

Sharp drop of the permissible gas content line was observed only for AC7A. Permissible gas content reached to zero when the solidification time exceeded 500 sec. It also seemed to be caused by the wide semisolid range of the alloy(2).

CONCLUSION

A series of experiment were conducted to know permissible gas content of the melt of 99.8 % pure Al, AA-6063, JIS-AC1A, AC2B, AC7A and AC9A, for the production of sound castings. The following conclusions can be drawn from the present study.

(1) Permissible gas content decreases as the solidification time increases. (2) Relationship between permissible gas content and solidification time can be

expressed as a straight line in general, however, in the case of Al-Mg alloy (AC7A), the line drops abruptly downward when the solidification time exceeds 400 sec.

(3) Permissible gas content can be obtained from density measurement as well as dye penetrant test.

(4) Permissible gas content obtained from density measurement is usually lower than that obtained from dye penetrant test in general. However, AC2B was found to be exceptional.

R E F E R E N C E S

1. P.M. Thomas and J.E. Gruzleski, 'Threshold Hydrogen for Pore Formation Dur ing the Solidif ication of A luminum Alloys", METALLURGICAL TRANSACTIONS B, Vol. 9B, 1978,139-141

2. T. Isobe, M. Kubota and S. Kitaoka, "Castability of Aluminum Alloys", IMONO, Vol. 47 ,1975 , 345-355

449

The growth and properties of Al-Li single crystals

Tsing-Shien Sheu Department of Mechanical Engineering, Chung Cheng Institute of Technology, Taoyoun, Taiwan, Republic of China

Shuye-Jong Chen, Shih-Chin Chang Department of Materials Science and Engineering, National Tsing Hua University, Hsinchu, Taiwan, Republic of China

ABSTRACT

Al-Li single crystals containing up to 2.31 wt % Li was grown successfully by using graphite crucible with a spiral grain selector. The dislocation density of the as grown single crystal was

9.1 x 10

7 m -

2 preferred grain growth orientation was < 1 0 0 > for Al- 2.31 wt % Li alloys.

For Al-Li single crystals an addition of each wt % Li results in a reduction of about 2.3 % in density.


INTRODUCTION

Aluminum-Lithium alloys have been the subject of much recent study and were recognized for their low density and high stiffness which are beneficial for aerospace structural applications [ 1 - 4 ]. One major problem for using Al-Li alloys was their poor tensile ductility and fracture toughness. The low ductility of Al-Li alloys has been associated with the strain localization in the grain which leads to cracking in the intense slip bands or produce stress concentrations across grain boundaries [ 5 ] .

Chang and Huang studied the fracture and shear band formation in an Al-Cu-Li-Mg-Zr alloy [ 6 ]. It is proposed that the shear band formation is closely related to the slip of dislocation in each grain and the { 110 } < 001 > Goss texture formed in the rolled material could promote the shear band formation and fracture.

In order to get a more clear picture of the deformation and shear band formation, it is desirable to study the properties of Al-Li single crystals. In recent years, some researches has been done on Al-Li single crystals [ 7 - 8 ]. However, since lithium is a very active element, difficulties has been encountered in preparing good quality Al-Li single crystals with controlled lithium content. In this paper, a method of making graphite crucible for growth of Al-Li single crystals and the properties of the resultant crystals will be reported.

EXPERIMENTAL

Materials

Ingots of Al-Li binary alloys containing various amount of Li were prepared by melting commercially pure Al ( > 99.8 wt % ) and Al-10.5 wt % Li master alloy and casting in a vacuum induction furnace under an atmosphere of 20000 N / m

2 high purity argon. After homogenization

at 723 K in a salt bath for 12 hours, the chemical composition of each ingot was determined by Inductively Coupled Plasma Atomic Emission Spectroscopy. Then the ingots were machined into small pieces and were used as starting materials in single crystal growth.

Single Crystal Growth

Figure 1 shows the modified vertical Bridgman single growth system designed by the authors. In that system, the crucible holding the starting material for single crystal growth was sealed in a quartz tube fitted in a tubular muffle furnace.

In addition to the lost wax shell mold with either a double bulb-capillary or spiral single crystal selector, made of conventional ceramic materials such as Alumina (A12 O 3) a n (j zircon ( Z r S i O 4 ) t graphite crucible with spiral single crystal selector were designed and used in this study. As shown in Figure 2a, a graphite crucible was machined with a small hole at the bottom end. After the cavity of the graphite crucible was filled with wax, a spiral selector wax component ( Figure 2b) was welded to the crucible at the bottom hole. The conventional method of applying the ceramic coating in the lost wax mold preparation process was then applied to the assembled components. After dewaxing and mold firing, a graphite crucible with a spiral selector as shown in Figure 2c was obtained.


Fig. 1. Single crystal growth system: (1) quartz tube, (2) moving furnace, (3) shell mold and (4) driving system. The vacuum system is not shown in this picture

Fig. 2. The design of the graphite mold: (a) graphite crucible, (b) wax component of a spiral selector and (c) the finished mold

In growing Al-Li single crystal, about 80 g starting material was loaded in the cavity at the top end of the crucible. After evacuation to below 0.13 Pa , the tube was flushed with high

purity (5N9) argon and an argon pressure of ^ x *0 5 p a w as maintained in the tube throughout the single crystal growing period. The starting material was melted by heating the furnace to 983 K at a rate of about 0.15 K/s and held at that temperature for 1200 s before

solidification was accomplished by moving the furnace upward at a speed of ^ x 10 6 m / s

Characterization

The as grown rods were removed from the crucible. Cross sections of both ends of each rod were cut and metallographically polished. Keller's etchant was used to check for grain boundaries in the cross sections. Etch pits were revealed by a solution of 50 HC1 +50 H N O 3 + 10 HF +30 H 20 [ 9 ] and examined under an optical microscope or a scanning electron microscope. The dislocation density was measured by counting the etch pit density (EPD ). The back-reflction Laue method was used to determine the orientation of crystals. For each crystal, three 1.5 mm wafer


samples were cut at three equally spaced positions and several Laue pictures were taken at different positions on each wafer. The lack of any detectable difference in the obtained Laue patterns was employed as the criterion that the resultant rod is a single crystal.

The chemical composition of each rod was analyzed to study the loss of Li content. The density of crystal was obtained by measurements of weight and volume. The hardness was measured by a model HV-1 MATSUZAWA Vickers hardness tester under a load of 1 kgf applied for 10 seconds.

(a) (b)

Figure 4 - (a).Etch pits on a cross section of the rod B, (b) Etch pits on a cross section of the rod D.

(a) (b)

Figure 3 - (a) Back-reflection Laue pattern of the wafer cut from rod B, (b) Back-reflection Laue pattern of the wafer cut from rod E.



The results of this work were summarized in Table I and II. Each results listed in Table I was reproduced at least three times. The data listed in Table II were the average of three or more measurements. From Table I, it could be noticed that the spiral type grain selector was superior to the double bulb-capillary type grain selector in producing single crystals. Good single crystal of Al-Li alloys could not obtained by using alumina crucible since even for the starting material containing as low as 1 wt % Li ( Test B ), the Li loss was excessive and the crystal possessed a mosaic structure. Figure 3a shows the Laue pattern of the wafer cut from rod B. Figure 4a shows the etch

pits on specimen of that rod. The dislocation density was ^-5 x 10 m

2 g v u s n lg zircon

( Z r S i O 4) crucible with spiral grain selector (as in Test D ) , single crystal rod could be produced. However, the Li loss was still too excessive and could not be controlled. The difficulty of high Li loss was also encountered by other researchers in preparing Al-Li single crystals [ 7 , 8 ] .

When the graphite crucible with spiral grain selector was used, 15 mm diameter single crystal rod of Al-Li alloy with Li content of 2.31 wt % was produced with the starting material of Al-2.7 wt %

Li alloy. The Li loss was only 14 % and the dislocation density was ^-1

x 10

7 m"

2 which was

more than one order of magnitude less than those resulted from using ceramic molds. The Laue pattern and etch pits of a Al-2.31 wt % Li single crystal (Test E ) were shown in Figures 3b and 4b.

Table I - Results of Crystal Growth

Test Starting Material

Crucible Material

Selector Resultant rod

A B C D E

Al-1 wt%Li Al-1 wt % Li Al-2.7 wt % Li Al-2.7 wt % Li Al-2.7 wt % Li

Alumina Alumina Alumina Zircon Graphite

Double neck Spiral Spiral Spiral Spiral

Polycrystal Moasic Polycrystal Single crystal Single crystal

Table II - The Properties of Single Crystals

Test Final Li loss

Li%

Density (Mg/m3)

Density VHN

Change/Li %

Dislocation

Density (

m"

2)

B 0.35 - 65 % 2.68 - 2 . 1 % 20.8 4.5 x 10

10

D 0.47 - 83 % 2.67 - 2.4 % 19.8 2.8 x 10

9

E 2.31 - 1 4 % 2.56 - 2 . 2 % 31.6 9.1 x 10

7


From the standard Gibbs Free Energy data of oxides [ 10 ] , it is known that L i 2 O i s much more stable than A12 O 3 and Zr S i O4. Therefore, when alumina or zircon mold is used in holding

melted Al-Li alloys, the displacement reaction 6Li + A l 20 3 - » 3 L i 20 + 2Al or 4 L i + S i 0 2 —> 2 L i 2 0 + S i wJ J J t ak e p i a ce These reactions could cause the excessive loss of Li in these alloys and promote the nucleation of new grains, which is detrimental to the growth of good single crystal. In contrast, graphite is a material inert to melted Al and Al-Li alloys which makes it a good crucible material in the growth of single crystal.

It was noticed that for single crystals grown from Al-2.7 wt % Li starting material with graphite crucible, the growth orientations were generally very close to < 001 > direction. Similar result was reported in Ni based superalloy single crystal growth [ 11 ] . However, no preferred growth orientations was found in this work for Al-Li alloys with less Li content.

It was found in this study that the crystal density decreases with increasing lithium content, and an addition of each wt % lithium to these crystals results in a reduction of about 2.3 % in density. This result is quite close to but somewhat smaller than the reported value for polycrystalline Al-Li alloys [ 12 ].

CONCLUSIONS

1. A technique of making graphite crucible with a single crystal selector was described. By using the graphite crucible in a modified vertical Bridgman single crystal growth system, 15 mm diameter rod of Al-Li single crystals containing up to 2.31 wt % Li were produced successfully. The

dislocation density of the as grown single crystal was ^-1

x 10

7 m -

2

2. The preferred growth orientation of Al-2.31 wt % Li single crystal was generally close to < 100 >. However, no preferred growth orientations was found in this work for Al-Li alloys with less Li content.

3. For Al-Li single crystals, the density decreases with Li content increases. An addition of each wt % lithium to the crystals results in a reduction of about 2.3 % in density

ACKNOWLEDGMENT

The authors are grateful for the support of this research by the National Science Council, Republic of China under grant NSC79-0405-E007-15.

REFERENCES

l.T.H. Sanders, Jr. and E.A. Starke, Jr, Eds., Aluminum-Lithium Alloys. The Metallurgical Society of AIME, Warrendale, PA, USA, 1981.

2.T.H. Sanders, Jr. and E.A. Starke, Jr, Eds., Aluminum-Lithium Alloys II. The Metallurgical Society of AIME, Warrendale, PA, USA, 1984.


3.C. Baker, PJ. Gregson, SJ. harris and CJ. Peel, Eds., Aluminum-Lithium Allovs III. The Institute of Metals, London, 1986.

4.G. Champier, B. Dubost, D. Miannay and L. Sabetay, Eds., 4th International Aluminum-Lithium Alloys Conference. J. De Physique, Paris, 1987.

5. E.A. Starke, Jr. T.H. Sanders, Jr. and I.G. Palmer,"New Approaches to Alloy Development in the Al-Li System", Journal of Metals. August, 1981, 24-33.

6. S.C. Chang and J.H. Huang, "The Fracture and Shear Band Formation in an Al-Cu-Li-Mg-Zr Alloy", Acta metall.. Vol. 34, No. 8,1986, 1657-1662.

7. Y. Miura, A. Matsui, M. Furukawa and M. Nemoto, "Plastic Deformation of Al-Li Single Crystals", Aluminum-Lithium Allovs III. C. Baker, PJ. Gregson, S.J. harris and C.J. Peel, Eds., The Institute of Metals, London, 1986,427-434.

8. Y. Miura, Y. Yusu, M. Furukawa and M. Nemoto,"Temperature Dependence of Yield Strength of Al-Li Single Crystals", 4th International Aluminum-Lithium Allovs Conference. G. Champier, B. Dubost, D. Miannay and L. Sabetay, Eds., J. De Physique, Paris, 1987, C3-549-C3-555.

9. S.M. Lan and S.S. Rau, "The Study of Dislocation in Al Single Crystals", Chinese Journal of Materials Science. Vol. 8, No. 3, 1976, 109-117.

10. R.A. Swalin, Thermodynamics of Solids. John Wiley & Sons, New York, 1972, 116.

11. M. Gell, D.N. Duhl, D.K. Gupta and K.D. Sheffler, "Advanced Superalloy Airfoils", Journal of Metals, July, 1987,11-15.

12. E.A. Starke, Jr., "The Application of the Fundamentals of Strengthening to the Design of New Aluminum Alloys", Strength of Metals and Alloys. R.C. Gifkins, Eds., Pergamon Press, Oxford, 1982, 1025-1044.


Authors' Index

Albright, D.L, 57 Liu, H., 83 Aliravci, C.A., 413 Loong, C.A., 315 Arcade, Ph. 231 Luo, A., 423 Avedesian, M.M., 31

Monaghan, D.J., 257 Barber, M., 329 Barr, P., 367 Neff, D.V, 393 Bellamy, G., 273 Nishina, K., 435 Browne, D.J., 257 Nussbaum, G., 19,55 Bui, R.T., 307,329,367

Pekguleryuz, M.O., 31 Champier, G., 231 Perron, J., 367 Chang, S-C, 449 Pinfold, P.M.D., 43 Charette, A., 367 Posey, W., 131 Chen, S-J., 449 Potocnik, V., 329 Cook, R., 257 Proulx, A , 367

Davis, B.R., 191 Qui, Z., 203 Dellamore, G.W., 69 Desclaux, P., 163 Ravindran, C, 245 Dimayuga, F.C., 3 Regazzoni, G.., 55 Dion, J.L, 217 Richards, N.E., 131 Dionne, S., 107 Rolseth, S., 177 Dorward, R.C., 383

Sahoo, M., 217 Edmonds, D.V., 257 Salloum, G., 315 Engh, T.A., 339 Samuel, F.M., 83,231,293 Engh, T.A., 339

Sastry, S.M.L, 99 Fantetti, N., 31,413 Sheu, T-S., 449 Fortin, G., 293 Simard, G., 367 Frayce, D., 315 Smith, R.W., 69 Frisvold, F., 329 Soboyejo, W.O., 99

Suzuki, T., 83 Gallerneault, M., 69 Gjestland, H., 57 Tabereaux, AT., 131

Thomas, P.M., 257 Harris, R., 147 Thompson, W.T., 191 Hibbins, S.G., 3 Thonstad, J., 177 Hogg, J.C., 57 Thorvaldsen, A., 413 Hunt, J.D., 257 Tikasz, L, 329

Toguri, J.M., 119 Jain, S., 367 Torres, J.H., 231 Jue, B., 245

Utigard, T.A, 163,353 Karpynczyk, J., 245 Kaya, M., 69 Warczok, A., 163 Kitaoka, S., 435 Watson, K.D., 119 Kocaefe, Y., 367 Westengen, H., 57 Krishnadev, M.R., 107

Yang, X., 257 Lacroix, M., 367 Youdelis, W.V., 423 Langlais, J., 147 Yun, M.., 257 Lederich, R.J., 99 Lefebvre, F., 19 Zhai, X., 203

457


Subject Index

Activity, 191,194,196-198 Age Hardening, 423-426, 432 Aging, 232,235 Alkali Removal, 394 Alloy Development, 270 Alloying Additions, 85,91 Alumina, 177 Alumina Feeding, 329,332,333 Alumina Solubility, 358 Aluminothermic Reduction, 147 Aluminum, 127 Aluminum Alloys, 114,219,265,

273,274,276,277,279,280 Aluminum Electrolysis, 177,329 Al-Li Alloys, 232,423,424,425,449 Al-Li-Fe Alloys, 232

Bath Density, 167 Bismuth, 147-161 Boundary Conditions, 299 Boundary Layer Flow, 169 Brine Purification, 46-48

Carbon, 127 Castfill-3D, 319 Casting, 31-36,57,273-283 Cathodic Protection, 13,14 Centrifugal Atomization, 100,232 Coke Calcination, 367,368,371 Cold Finger Technique, 171 Cold Shut, 226 Collision Efficiency, 339 Composites, 83 Computational Fluid Dynamics, 307,313 Conduction, 296 Contact Angle, 127 Control, 331,332 Corrosion, 9,10,13,31-36,40 Creep Rates, 105 Crusting, 178

Degassing, 405,407 Dehydration, 46-49 Demagging, 401,402 Die Cast, 10,11,13,31,32,34,

315-317,319,321,323,325,326 413,414,416,417,420,421

Directional Solidification, 70 Dislocation Density, 450 Dissolution, 182 Dross, 413-415, 420

Electrolysis, 47,48,50,51 EMF, 193,194 Enthalpy, 194,197 Entrapment, 70 Evaporative Foam, 245 Excess Molten Reductant, 147,160 Experimental Caster, 261 Extraction, 147-161

FACT, 157,158 Fatigue, 31,35,36,39,40 Feeding, 57 Fibres, 83,85,88,91 Filling, 57 Filtration, 405,408 Filtration Efficiency, 339 Finite Elements, 301,315-318,319,325 Flotation, 203 Flow Rate, 245 Fluidity, 221-223 Flux, 402,405,407 Flux Injection, 406,407 Forced Convection, 395,397,398,399 Fracture Toughness, 104,273,274,276,277,279,280 Free Surface, 315,316

Gibbs - Duhem, 195 Gibbs Energy, 354 Graphite Crucible, 451 Grain Refining, 408

HCI Synthesis, 47,48,51 Heat Flux, 167 Heat Transfer, 294,367-370, 395-397 Heat Transfer Coefficient, 164 Heat Treatment, 110,111 High Purity, 3,6,9 Hot Extrusion, 232,233 Hot Water Tank Anodes, 14,15,16

Impact, 33,35 Interception, 339 Interface Reaction, 88,90 Interface Tension, 358 Interface Wettability, 85,90 Intermetallics, 413,419,421

Liquidus Temperature, 167 Lithium Chloride, 191-200 Low Pressure, 57

Magnesite Dissolution, 47-49 Magnesium, 31-34,40,57,147-161,

413,414,417,419,420 Magnesium Casting, 47,48,52,53 Magnesium Chloride, 191-200 Magnesium Metal Markets, 44,52 Magnesium Production, 43-53 Mathematical Modelling, 260,307-309,

315,317,318,326,367,369 Melting, 413,414,417 Metallurgical Processes, 307,313 Metal Matrix Composites, 69,70,99,108,114 Metal-mould Interface, 299 Microalloying, 424 Microstructures, 268 Modelling, 329,330 Modulus of Ti Alloys, 104 Molten Salt, 353 Mould Filling Time, 224

459


Oxides, 413,415-421 Oxygen Affinity, 158

Particle Pushing, 69,73,74 Permeability, 245 Pidgeon Silicothermic, 3,4,5 Plasma Arc Melting, 100 Plaster, 31-38,40 Porosity, 245 Powder Metallurgy, 232 Process Capability, 51-53 Process Control, 49 Prometheus-3D, 319 Prototypes, 31,32 Pumping, 395,397,399-401

Rapid Solidification, 99,232 Reaction Mechanism, 158 Recovery of Cryolite, 203 Recycling, 393-395,400,401,403 Refractory Coating Wafer, 245 Rotary Kiln, 367,371

Sand, 57 Sand Casting, 219-221 Sand Penetration, 225 Scrap Submergence, 394,396,401

Sedimentation, 339 Segregation, 258 SiC Particulate, 69,71,73,108,109 Simulation, 332 Single Crystal, 449 Sludge, 413-415,417-419,421 Spent Potlining, 203 Sodium Chloride, 191-200 Solidification Rate, 274,275,278,279,284 Strontium, 147-161 Strontium Carbonate, 147-149,160 Surface Tension, 157

Temperature Distribution, 296 Tensile Properties, 31,33,35-37,40,102,103,112,113 Thermodynamic Properties, 191 TiB2, 127 Ti Composites, 99 Twin Roll Casting, 257

Underground Anodes, 13,14

Vacuum Casting, 217,219

Water Model, 339 Wetted Cathode, 119 Wetting, 119,122-126

Documents

Extraction, Refining, and Fabrication of Light Metals. Proceedings of the International Symposium on Extraction, Refining and Fabrication of Light Metals, Ottawa, Ontario, August 18â€“21,