J384-214-9-1141

8/6/2019 J384-214-9-1141

1/9

Characterization of acoustic emission signals from

fatigue fracture

Z Shi1, J Jarzynski1, S Bair1, S Hurlebaus2** and L J Jacobs2*1 The George W. Woodru School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, Georgia,

USA2School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA

Abstract: This paper discusses a comprehensive study that is developing a quantitative under-

standing of the acoustic emission (AE) signals that emanate from fatigue cracks. Two critical com-

ponents of this study are the development of a transfer function that quanties and removes

geometric eects from a measured AE waveform and an experimental program that monitors andidenties AE signals that occur during the fatigue of cylindrical stainless steel specimens under

torsion. Typical waveforms are collected during torsional fatigue and correlated with fracture

mechanisms from dierent stages of testing. Three stages of fatigue are identied by AE waveform

characterization and conrmed by microscopic replica observation. The other portion of this study

demonstrates the eectiveness of using laser ultrasonic techniques to develop transfer functions to

quantify and remove geometric eects from measured acoustic emission waveforms.

Keywords: acoustic emission, transfer function, torsion fatigue

NOTATION

N fatigue cycle number

shear strain

1 INTRODUCTION

The detection of progressive and catastrophic failure

events is important for the increased safety and relia-

bility of engineering structures. These failures can be

identied and prevented using non-destructive testing

techniques for detecting and locating internal aws.

Acoustic emission (AE) testing is used today to evaluateand monitor structural components. This technique

oers a distinct advantage over more conventional non-

destructive testing techniques because it allows for the

real-time monitoring of in-service structures. To be

eective, AE testing must be able to identify the location

of an arbitrary defect (such as a fatigue crack) and to

characterize the severity of that defect. The diculty in

the application of AE techniques to locate and detect

these damage events from a source within a body is in

interpreting and categorizing the large quantity of dataand in rejecting the spurious information. Of funda-

mental importance for the advancement of the current

state of AE technology is the isolation and identication

of the signal from a fatigue crack.

This work characterizes the AE signals from fatigue

cracks by integrating methods from wave propagation

in elastic solids and experimental mechanics with

quantitative non-destructive evaluation techniques. This

paper presents results that provide the necessary rst

step towards accurate assessment of the integrity of

existing engineering structures.

A measured AE signal depends upon three distinct

factors:

(a) its source,

(b) component geometry and material properties and

(c) receiving sensor and instrumentation.

The source of an AE waveform determines its initial

shape, amplitude and frequency content [1, 2]; these

waveforms are transient and broad band in nature. The

component geometry and material properties inuence

the wave propagation from the emission source to the

receiver [3]. Finally, the detection sensor and instru-

mentation can bias the signal by ltering certain fre-

quencies [4, 5]. The contribution of each of these threeelements must be understood in order to identify and

characterize measured AE waveforms eectively.

The MS was received on 25 February 1999 and was accepted afterrevision for publication on 14 July 1999.

* Corresponding author: School of Civil and Environmental Engineering,Georgia Institute of Technology, Atlanta, GA 30332-0355, USA.** Present address: Institute A of Mechanics, University of Stuttgart,Stuttgart, Germany.

1141

C03199 IMechE 2000 Proc Instn Mech Engrs Vol 214 Part C

8/6/2019 J384-214-9-1141

2/9

The component geometry changes the original AE

signal as it propagates from its source to the receiving

sensor. These changes, which are caused by reection,

transmission and mode conversion at interfaces and

boundaries, have a signicant impact on the AE wave

that is measured by the receiving sensor. The geometryeects (which do not contain any useful information for

AE testing) tend to swamp and mask the initial signal,

making it dicult to identify the original source of the

acoustic emission uniquely.

This paper reports on a comprehensive study that is

developing a quantitative understanding of the AE sig-

nals that emanate from fatigue cracks. Two critical

components of this study are:

(a) the development of a transfer function that quan-

ties and removes these geometric eects;

(b) an experimental program that monitors and identi-

es AE signals that occur during the fatigue ofcylindrical stainless steel specimens under torsion.

These results advance the current state of AE and play a

critical role in quantitatively characterizing AE signals.

2 ACOUSTIC EMISSION SIGNALS FROM

TORSION FATIGUE

This portion of the paper reports on a study of the

fatigue of a cylindrical stainless steel specimen under

torsional loading. The fatigue process is monitored with

a two-channel, AE system that digitizes the completeAE event waveform. By examining the cumulative AE

event counts versus number of fatigue cycles curve, three

`event growth' zones are observed. It is found that spe-

cic characteristics of the measured AE waveforms

(such as amplitude, energy distribution and frequency

content) change signicantly as the fatigue test pro-

gresses. In addition, these AE waveform changes are

correlated with the damage mechanisms that occur

during the dierent stages of the fatigue test. The three

distinct stages of fatigue damage observed are:

(a) the uniform growth and distribution of small,

subgrain sized cracks,(b) the growth of microcracks across the grain bound-

aries;

(c) the coalescence of microcracks into localized cracks.

These stages are conrmed by the microscopy of surface

replicas taken at regular loading cycle intervals.

3 EXPERIMENTAL PROCEDURE

The specimen used is a 304 stainless steel cylinder, with

an outer diameter of 2 in (50.8 mm), an inner diameter of

1 in (25.4 mm) and a gauge length of 2.5 in (63.50 mm).

Torsion tests are performed using an MTS fatigue-testing

machine. The loading is triangular, with a strain of=2

0:35 per cent. The loading frequency is 0.2 cycles/s.

A two-channel AE acquisition system (with a capacity

of up to four channels), the fracture wave detector from

digital waves, is used to digitize the AE signals. Two

critical features of this system are:

(a) digitization and storage of the entire AE waveform

in real time and

(b) separation of event triggering from signal acquisi-

tion.

These features are critical for waveform-based AE stu-

dies, since spurious noise due to electricalmechanical

interference (EMI) can be largely eliminated with proper

system conguration (with proper ltering and thresh-

old settings). The AE transducers used in this study are

single-chip, piezoelectric (PZT) elements with a size of

1 2 mm. These sensors were developed as part of this

research program, and their response to an impulse

input (a laser source) is shown in Fig. 1. Figure 1

compares the time and frequency domain responses of

both a single-chip PZT element (used in this study) and

a typical, commercially available AE transducer. Note

that both these sensors are relatively at in the fre-

quency range of interest: 100 kHz to 2 MHz. However,

the small size of the PZT chip sensors enables point

detection (it is considered to be a point receiver) which

avoids the averaging eect of transducers having a large

surface area. A critical factor in conducting repeatable

AE test results is mounting the sensor on the specimen

in a way that achieves both a stable triggering schemeand the reproducible capture of AE waveforms.

The research presented in this portion of the paper

determines:

(a) the statistical features of AE events acquired from

material fatigue and fracture;

(b) the characterization of typical AE waveforms at

dierent stages of fatigue;

(c) the correlation of AE signal features with damage

mechanisms observed with surface microscopy.

Since the geometry eects and transducer response are

constant during these fatigue tests, it is logical to assumethat measured changes in an AE waveform are due to

the evolution of the damage sources.

4 EXPERIMENTAL RESULTS AND DISCUSSION

Figure 2 shows the cumulative number of AE event

counts versus number of fatigue cycles in a typical tor-

sion test. In traditional AE testing, AE event counting

has been correlated with stressstrain data and then

used to characterize material fatigue behaviour. Using

this approach, turning points or transition zones are

usually associated with the changes in crack formation

or fracture. With waveform-based AE analysis, the sig-

1142 Z SHI, J JARZYNSKI, S BAIR, S HURLEBAUS AND L J JACOBS

Proc Instn Mech Engrs Vol 214 Part C C03199 IMechE 2000

8/6/2019 J384-214-9-1141

3/9

nals during, before and after these stages are analysed in

detail and used to relate any changes in waveform

features to possible evolution of fatigue damage

mechanisms. Several features are observed in Fig. 2:

there are zones of signicant AE activity, followed byplateaus of inactivity. This work relates these `acoustic'

zones of activity (and inactivity) to specic fracture

phenomena (fatigue damage) occurring in the torsion

specimen. This correlation between AE signals and

fracture phenomena is possible through the microscopic

examination of surface replicas that are taken at regular

intervals during the fatigue test. This microscopic eva-

luation is performed at the beginning and after 70 000,140 000, 210 000 and 240 000 cycles (microscopic images

at 70 000, 140 000 and 210 000 cycles are shown in Figs

3a to c). Figures 2 and 3 reveal quantitative information

Fig. 1 Response of point PZT element and typical commercial AE transducer

Fig. 2 Cumulative AE counts versus fatigue cycles

CHARACTERIZATION OF ACOUSTIC EMISSION SIGNALS FROM FATIGUE FRACTURE 1143


8/6/2019 J384-214-9-1141

4/9

Fig. 3 Surface replica (a) at N 70 000 cycles, (b) at N 140000 cycles and (c) at N 210000 cycles



8/6/2019 J384-214-9-1141

5/9

concerning the fatigue characteristics of this particular

material.

Firstly, the evolution of fatigue damage is not a

continuous process, especially as evidenced by the

intermittent nature of the occurrence of AE waveforms.

There are stages when AE activity (and the associatedfatigue damage) accumulated continuously, and there

are stages during which the AE activity is eectively

dormant. About 7000, 4000 and 6000 events are cap-

tured during cycles 30 00046 000, 77 00089 000 and

118 000158 000 respectively. The slope of the AE count

curve indicates the severity of the AE activity. Almost

no (or at least limited) AE activity occurs at the begin-

ning of the test and two other regions (between cycles

46 000 and 77 000 and cycles 89 000 and 118 000); the

latter two regions are between two vicinities of high AE

activity. Note that, during the period of AE dormancy,

fatigue damage is accumulating to produce enough

energy to proceed to the next level of fatigue damage. By

referring to Fig. 2, the progressive material damage

consists of several stages: crack incubation, gradual

initiation/formation and eventually signicant crack

propagation leading to fracture.

Secondly, there are signicant changes in the ampli-

tude, frequency content and other signal characteristics

of the AE waveforms during the fatigue experiment.

These changes occur over a transition period, followed

by a constant increase in AE activity (throughout which

the waveforms keep the same pattern). Typical AE sig-

nals taken from the rst two stages are shown in Figs 4

and 5. The AE events observed during the rst stage (seeFig. 4) show low amplitude as well as frequency contents

below 300 kHz, which is similar to EMI noise. The sig-

nal in Fig. 5 is a typical burst-type AE event, which also

represents the majority of the waveforms observed

during cycles 77 00089 000. The same phenomenon (a

signicant change in AE waveforms) is also observed

when comparing AE waveforms captured during stage 3

with that of stage 2 (after a dormant period from cycles

89 000118 000).Thirdly, only one burst-type signal, with approxi-

mately the same amplitude and frequency content, is

observed during the entire stage 2, cycles 77 00089000.

A variety of AE signals, both burst and continuous type,

with dierent amplitudes and frequency distributions,

are observed during stage 3 (cycles 118 000158 000).

Note that the pattern of AE signals is an evolutionary

process, with more activity and diversity of signals

occurring in the later stages of fatigue. Since the geo-

metry eects and transducer response are constant

throughout the entire fatigue test, relative changes in AE

waveform attributes represent changes in the sources of

these AE signals: fatigue damage. As a result, the fol-

lowing conclusions are possible: there is basically one

type of crack initiation during stage 2, as revealed by the

single-type AE event, while multiple crack formation

mechanisms contribute to the variety of AE phenomena

observed in stage 3.

Three stages of fatigue are recognized from the AE

waveforms, count distribution and surface replicates:

the uniform growth and distribution of small subgrain-

sized cracks; the growth of these microcracks across the

grain boundaries; and coalescence of these microcracks

into localized cracks. Specic features of each stage

follow:

Stage 1. Uniform growth of subgrain-sized cracks. No

damage activity occurs at the beginning of the test

Fig. 4 Typical AE signal and its frequency spectrum, stage 1



8/6/2019 J384-214-9-1141

6/9

(cycles 030 000), but this is then followed by a sharp

increase in AE activity. These AE events show low

signal amplitude and frequencies below 300 kHz.

These AE events are generated from a damage source

emitting AE waves that have similar amplitudes tothe ambient noise level present in the system. The

damage activity during this initial stage is basically

uniform growth and distribution of small subgrain-

sized cracks. The microscopic image of a surface

replica taken at N 70 000 cycles (shown in Fig. 3a)

shows a uniform distribution of the subgrain-sized

fractures that developed during this stage.

Stage 2. Growth of microcracks across the grain

boundaries. The AE events measured at this stage are

burst type and have a frequency content that extends

from 150 to 600 kHz. It is observed that only one

single AE signal occurs during stage 2, suggesting aunique AE source. The microscopic image taken at

N 140 000 cycles (see Fig. 3b) clearly shows mul-

tiple cracks in the 100 mm length range. Dislocation

motion across the boundaries is the major con-

tributor of AE signals in this stage.

Stage 3. Coalescence of microcracks into localized

cracks. The variety of AE events that occur during

this stage suggests that multiple damage mechanisms

are responsible for AE signal generation. Possible

mechanisms include dislocation, crack growth and

propagation, void formulation and fracture. The

continuous-type AE signals measured in this stage

(which contain much higher energy) are caused by

highly localized crack growth. The frequency content

of these AE events extends from 100 kHz to over

3 MHz. The microscopic image taken at N 210 000

cycles (shown in Fig. 3c) clearly shows a single crack

over 900mm in length.

5 DEVELOPMENT OF TRANSFER FUNCTIONS

The second portion of this research develops a transfer

function that quanties and removes geometric eects

from experimentally measured AE waveforms. This

transfer function is developed using a repeatable, broad-

band AE source, a pulsed laser, and a broad-band, high-

delity sensor, a laser interferometer [6]. The steps in the

development of the transfer function are as follows:

1. An AE signal is generated and detected in the

specimen whose geometry is being characterized.2. This experiment (same source) is repeated in a

`geometry-less specimen', i.e. one that contains no

geometric features.

3. The `geometry-less' signal is divided by the specimen

signal (in the frequency domain) to form the transfer

function. This transfer function can operate on an

AE signal, measured in the same specimen, but

caused by a dierent source.

The optical techniques used both to generate and detect

the acoustic emission signals are critical to the success of

this research. In order for the proposed development to

be viable, the source must be a repeatable, point source

that is broad band enough to represent a typical acoustic

Fig. 5 Typical AE signal and its frequency spectrum, stage 2



8/6/2019 J384-214-9-1141

7/9

emission event; the ablation source (created by a pulsed

laser) used in this research ts these criteria. The laser

interferometer used in this work is capable of making

localized (point), broad-band, absolute measurements

with very high delity; as a result, the receiving sensor

will not add any frequency bias to the calculated transferfunction. An additional advantage of this non-contact,

optical detector is that the measurement technique does

not interfere with, or aect, the process being mon-

itored.

This work investigates a specimen with a `simple'

geometry, a 90 corner. These measurements are made

with the source and receiver on the same surface (side).

As a result, the corresponding `geometry-less specimen'

is a half-space. Signals measured in the half-space only

contain the eect of the source; they are not inuenced

by any specimen traits. It is important to note that the

dominant acoustic features in both these applications

are Rayleigh surface waves; this would not be the case

for thin plate-like specimens, or if the source and

receiver were on dierent surfaces (e.g. a buried source).

A dierent `geometry-less specimen' would be needed

for these applications. Once the respective transfer

functions are developed, removing their geometric

eects and isolating the source signal from two new

sources tests their robustness: a double laser pulse and apinducer.

6 EXPERIMENTAL RESULTS

The specimen examined has a single, uncomplicated

geometric feature, a 90 corner. The source and receiver

(interferometer) are placed on the same side of the

specimen, in front of the 90 corner, as shown in Fig. 6.

The laser source is used to create incident Rayleigh,

longitudinal and shear waves. These three waves pro-

pagate directly to the receiver and afterwards theyinteract with the corner. A part of the incident wave is

transmitted at the corner and a part is reected. The

reected wave travels back to the receiver.

Figure 7 shows the specimen and the half-space

waveforms, in the time domain, that are used to make

the transfer function for the 90 corner. These signals

are obtained in the ablation regime (without oil) and

represent the average 40 waves. Note that the only dif-

ference between the specimen and the half-space signals

is the reection from the 90 corner seen in the specimen

signal. Each of the signals in Fig. 7 is transformed into

the frequency domain by taking the FFT of the entire

waveform as shown in Fig. 8. The half-space signal in

the frequency domain is point-by-point divided by the

specimen signal in the frequency domain to get the

transfer function.

The validity of this transfer function is demonstrated

by removing the geometry eects (in this simple

Fig. 6 Location of source and receiver for the 90 corner test,

together with the incident and re ected Rayleigh

waves

Fig. 7 Specimen and half-space signal in the time domain



8/6/2019 J384-214-9-1141

8/9

specimen, the only geometry eect is the reection from

the 90 corner) from signals created with a dierent

source: the same laser creates a double-pulse signal; it

produces two distinct pulses (separated by approximately

15ms) at higher energy levels. Figure 9 compares the sig-

nals created by a double pulse and a single pulse (both

in the ablation regime) in the time domain. Figure 10

compares the reproduced half-space (double pulse in the

specimen, operated on by the transfer function) with the

actual half-space (double pulse in the half-space); thespecimen geometry is completely removed with the

transfer function. However, there is a small dierence

between the two signals in Fig. 10; the reproduced half-

space signal contains more noise than the actual half-

space signal.

7 CONCLUSION

This paper discusses a comprehensive study to develop a

quantitative understanding of the AE signals that ema-

nate from a fatigue crack. Two critical components of

this study are the development of a transfer function

that quanties and removes these geometric eects and

an experimental program that monitors and identies

AE signals that occur during the fatigue of cylindrical

stainless steel specimens under torsion.

The fatigue of a stainless steel cylindrical specimen

under torsional loading is studied using AE techniques.

Typical waveforms are collected and correlated with

fracture mechanisms from dierent stages of fatigue.

Three stages of fatigue are identied by AE waveform

Fig. 8 Specimen and half-space signal in the frequency domain

Fig. 9 Specimen signal with double pulse and single pulse



8/6/2019 J384-214-9-1141

9/9

characterization and conrmed by microscopic replica

observation. These stages are (a) the uniform growth

and distribution of small subgrain-sized cracks, (b) the

growth of these microcracks across the grain boundaries

and (c) the coalescence of these microcracks into loca-

lized cracks. More torsional fatigue tests under dierent

strain levels are being conducted to conrm these con-

clusions and to further the understanding of material

characterization with acoustic emission techniques.The second portion of this work demonstrates the

eectiveness of using laser ultrasonic techniques to

develop transfer functions to quantify and remove geo-

metric eects from measured acoustic emission wave-

forms. The eectiveness of these transfer functions is

directly dependent upon the broad-band, repeatable,

non-contact, high-delity point measurements that are

possible with the laser-based techniques used in this

work. While very sophisticated experimental techniques

are used in this study, the accompanying signal proces-

sing tools are somewhat naive; the thrust of this eort is

in making accurate experimental measurements of wave

propagation in nite specimens.Current work is developing transfer functions to

model and remove the eect of the sensor response of

the point PZT chips. These transfer functions, combined

with transfer functions that remove the geometric eect

of a fatigue specimen (being developed with the proce-

dure described in this paper), are being used to isolate

the `pure' AE signal from fatigue damage. In addition,

more advanced signal processing techniques such as

short-time Fourier transforms (SFT), wavelets and

neural networks are being used to isolate and char-

acterize these AE signals.

ACKNOWLEDGEMENTS

This work was partially supported by the Oce of

Naval Research under the M-URI Program `Integrated

Diagnostics' (Contract N00014-95-1-0539). Special

thanks are extended to Ms Valerie Bennett of the

Materials Research Group at Georgia Tech for running

the fatigue tests and taking the microscope images.

REFERENCES

1 Jacobs, L. J., Scott, W. R., Granata, D. M. and Ryan, M. J.

Experimental and analytical characterization of acoustic

emission signals. J. Nondestructive Evaluation, 1991, 10(2),

6370.

2 Ohtsu, M. and Ono, K. The generalized theory and source

representations of acoustic emission. J. Acoust. Emission,

1986, 5, 124133.

3 Hurlebaus, S., Jacobs, L. J. and Jarzynski, J. Laser

techniques to characterize the eect of geometry on acousticemission signals. Nondestructive Test. and Evaluation, 1998,

4, 2137.

4 Jacobs, L. J. and Woolsey, C. A. Transfer functions for

acoustic emission transducers using laser ultrasonics. J.

Acoust. Soc. Am., 1993, 94, 35063508.

5 Moss, B. C. and Scruby, C. B. Investigation of ultrasonic

transducers using optical techniques. Ultrasonics, 1988, 26,

179188.

6 Bruttomesso, D. A., Jacobs, L. J. and Fiedler, C. Experi-

mental and numerical investigation of the interaction of

Rayleigh surface waves with corners. J. Nondestructive

Evaluation, 1997, 16(1), 2130.

Fig. 10 Reproduced and actual half-space signal, double pulse



Documents

J384-214-9-1141