Measurement 61 (2015) 150–161

Contents lists available at ScienceDirect

Measurement

journal homepage: www.elsevier .com/ locate /measurement

Fuzzy logic based model for predicting surface roughnessof machined Al–Si–Cu–Fe die casting alloy using differentadditives-turning

http://dx.doi.org/10.1016/j.measurement.2014.10.0030263-2241/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel.: +60 379674593; fax: +60 379675330.E-mail addresses: marani.mohsen@gmail.com (M. Marani Barzani),

e.zalnezhad@gmail.com (E. Zalnezhad), ah_sarhan@um.edu.my(A.A.D. Sarhan), fsaeed2@live.utm.my (S. Farahany), ramesh79@um.edu.my (S. Ramesh).

Mohsen Marani Barzani a,⇑, Erfan Zalnezhad a, Ahmed A.D. Sarhan a,c, Saeed Farahany b,Singh Ramesh a

a Department of Mechanical Engineering, AMMP Center, Faculty of Engineering, University of Malaya, 50603 Kuala Lumpur, Malaysiab Department of Materials, Manufacturing and Industrial Engineering, Faculty of Mechanical Engineering, Universiti Teknologi Malaysia, 81310 UTM Skudai,Malaysiac Department of Mechanical Engineering, Faculty of Engineering, Assiut University, Assiut 71516, Egypt

a r t i c l e i n f o a b s t r a c t

Article history:Received 12 April 2014Received in revised form 28 August 2014Accepted 7 October 2014Available online 1 November 2014

Keywords:Fuzzy logicAluminumTurningSurface roughnessAntimonyBismuth

This paper presents a fuzzy logic artificial intelligence technique for predicting the machin-ing performance of Al–Si–Cu–Fe die casting alloy treated with different additives includingstrontium, bismuth and antimony to improve surface roughness. The Pareto-ANOVA opti-mization method was used to obtain the optimum parameter conditions for the machiningprocess. Experiments were carried out using oblique dry CNC turning. The machiningparameters of cutting speed, feed rate and depth of cut were optimized according to surfaceroughness values. The results indicated that a cutting speed of 250 m/min, a feed rate of0.05 mm/rev, and a depth of cut of 0.15 mm were the optimum CNC dry turning conditions.The results also indicated that Sr and Sb had a negative effect on workpiece machinability.The workpiece containing Bi exhibited the lowest surface roughness value, likely due to theformation of pure Bi that acted as lubricant during turning. A confirmation experiment wasperformed to check the validity of the model developed in this paper, and the predictedsurface roughness came had an error rate of only 5.4%.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Aluminum–silicon alloy is used extensively in the auto-motive and aerospace industries [1] due to its excellentcastability, good thermal conductivity, low expansion coef-ficient, and good corrosion resistance. Al–Si castings consti-tute 90% of the total aluminum cast components produced[2,3]. Silicon appears as a flake-like morphology in hypo-eutectic Al–Si alloys and it easily facilitating fractures and

decreased fracture elongation. Using a modification melttreatment that adds modifier elements such as strontium(Sr) and sodium (Na) is common practice. Modification melttreatments changes the flake morphology to a fibrous form,resulting in increased elongation and casting ductility [4].

Most Al–Si parts require some processing prior toassembly. Understanding the machining characteristics ofa workpiece is crucial for predicting surface quality aftermachining [5]. Compared to conventional Al–Si alloyscutting techniques, dry machining is an environmentallysustainable alternative owing to the absence of cuttingfluids [6]. However, cutting speeds, feed rates, and thedepth of the cut must be adjusted to achieve the bestsurface roughness and to realize cost reductions [7–9].

Nomenclature

dXn fuzzy subsets distinctiveYn fuzzy subsets distinctiveZn fuzzy subsets distinctiveWn fuzzy subsets distinctivekXn corresponding membership functionskYn corresponding membership functionskZn corresponding membership functionskWn corresponding membership functions_ minimum operation^ maximum operationCn centre of the nthrn width of the nth

a triangular fuzzy tripletb triangular fuzzy tripletc triangular fuzzy tripletD0 the output value in numerical formei individual errorHm measured valueHp predicted valueA model accuracyN total number of data set testev cutting speed (m/min)f feed rate (mm/rev)d depth of cut (mm)

M. Marani Barzani et al. / Measurement 61 (2015) 150–161 151

A few researchers have pointed out that an increase in cut-ting speed results in higher cutting temperatures duringAl–Si alloy machining in addition to a decrease in surfaceroughness due to less built-up edge (BUE) formation [10].Free-cutting aluminum alloys developed by adding free-machining elements (FME) is the most common metallur-gical technique used to improve the machinability of thesealloys [11]. In order to accurately predict the quality of amachined surface after turning, the machinability of thematerial concerned must be known. Machinabilitydepends on the type of material, its microstructure, andmachining states. Yang et al. [12] reported an improve-ment of wear resistance after the addition of antimony(Sb) and strontium (Sr) to A357 cast alloy. Bismuth (Bi) isconsidered to be a free-machining element for aluminumalloys. Additionally, Bi has a refining effect on the siliconmorphology in Al–7%Si–0.4%Mg alloy [13,14].

The traditional method of achieving low surface rough-ness at various machining parameters entails using a trialand error approach, which is very time-consuming. Hence,a reliable systematic approach for predicting surface rough-ness at different cutting conditions is required that wouldcover all parameter and only require low number of exper-iments [15]. When exact mathematical information is notavailable, soft computing techniques are the best way toanalyze experimental results. However, these techniquesdo have a few drawbacks including approximation, partialtruth, met heuristics, uncertainty, and inaccuracy.

Many researchers utilized artificial intelligent (AI)methods because these methods are capable of predictingand modeling phenomena. Artificial neural network(ANN) adaptive network-based fuzzy inference system(ANFIS) and fuzzy logic methods are the most popularexamples of AI methods [16–19]. Fuzzy logic which is amethod for identifying systems, is widely used in machinemonitoring and diagnostics [20]. Nandi et al. [21] used afuzzy logic (FL) model to predict surface roughness andcutting power for drilling Aluminum AA1050. Lo [22]concluded that a fuzzy logic models can accurately predictsurface roughness. Ramesh et al. [23] used fuzzy logic toimprove productivity by the controlling forces encoun-tered during turning processes.

Fuzzy logic was introduced by Zadeh [24] and it suc-cessfully uses the fuzzy set theory. Fuzzy logic acts as anextension of set theory by using the characteristic functionreplacement of a set through a membership function withvalues ranging from 0 to 1. Fuzzy modeling is used whensubjective knowledge and expert suggestions are impor-tant for defining objective functions and decision variables[24]. Compared to other AI methods, developing fuzzylogic methods are easier and they do not require a largeinvestment in software and hardware resources. For turn-ing process, results can be obtained by conducting only afew experiments. Out of all the AI approaches, fuzzy logicmethods are appropriate for predicting parameters suchas surface roughness using a limited amount of trainingdata. The purpose of this study was to predict surfaceroughness following the machining of Al–11.3Si–2Cu–0.4Fe die casting alloy that was treated with differentadditives including Sr, Bi, and Sb and by using an artificialintelligence technique (fuzzy logic). The Pareto-ANOVAoptimization method was used to obtain optimum param-eter conditions during the machining process.

2. Experimental procedure

2.1. Workpieces material

The chemical composition of Al–Si–Cu–Fe die cast alloyis shown in Table 1. Initially, the material was cut intosmall pieces, dried, and melted in a 2 kg SiC crucible usingan electric resistance furnace. Then a weighted bismuth,antimony and strontium in the form of pure metallic shots(99.99 wt.%), pure metallic granules (99.99 wt.%, and anAl–10Sr rod master alloy, respectively, were introducedseparately into the fully molten alloy. The levels of Bi, Sb,and Sr were selected based on the author’s previous studyin which the optimum concentration for each additive wasdetermined using a combination of computer aided coolingcurve thermal analysis (CA-CCTA) and microscopic inspec-tion [25]. The author found that the optimal concentrationfor modifying or refining the eutectic Al–Si phase with Biand Sb, and Sr were 1 wt%, 0.5 wt% and 0.04 wt%, respec-tively [25]. After adding the additive elements, the molten

Table 1Chemical compositions of the fabricated workpieces (wt.%).

Element Si Cu Zn Fe Mn Mg Ni Cr Bi Sb Sr Al

Base alloy 11.3 1.99 0.82 0.35 0.33 0.27 0.06 0.036 – – – Bal.Bi-containing 11.2 1.65 0.82 0.41 0.35 0.28 0.04 0.032 0.85 – – Bal.Sb-containing 11.3 1.82 0.80 0.43 0.31 0.25 0.06 0.030 – 0.42 – Bal.

Table 2Factors and levels used in the experiments.

Factors Level 1 Level 2 Level 3 Level 4

(A) Cutting speed(m/min)

70 130 250 –

(B) Feed rate(mm/rev)

0.05 0.1 0.15 –

(C) workpiece Basealloy

ContainingBi

ContainingSr

ContainingSb

152 M. Marani Barzani et al. / Measurement 61 (2015) 150–161

metal was kept for 15 min for complete melt homogeniza-tion. Prior to casting, the alloys were stirred and the sur-face was skimmed to remove dross and other impurities.The molten alloy was then poured at a temperature of730 �C (±5 �C) into cylindrical permanent molds to fabri-cate workpieces.

2.2. Cutting tool and tool geometry

A Kennametalphysical vapor deposition (PVD) insert(ISO catalogue number VBGT110302F) with a TiN-coated,a radius of 0.2 mm, a relief angle of 5� and a rake angleof c = 0�, and Grade KU10 was mounted on a holder desig-nated by SVJBL-1616H1. The workpieces were machinedusing different cutting speeds (70, 130, and 250 m/min)and feed rates (0.05, 0.1, and 0.15 mm/rev) with a constantcutting depth of 0.5 mm. All machining conditions wereselected based on the tool maker’s recommendations. Eachexperiment was repeated two times and a new cuttinginsert was used for each set of conditions to ensure theaccuracy of the surface roughness. The experimentalscheme is illustrated in Fig. 1.

2.3. Experiment details

A number of factors and levels used in the experimentsare shown in Table 2. The parameters used in this studyincluded feed rate, cutting speed, and workpiece parame-ters. Additionally, three levels were used for the feed rateand cutting speed and a fourth level was used for the work-piece (base alloy, Bi, Sr, Sb). Machining an Al–Si–Cu–Fe diecast alloy was completed using a CNC turning machine(ALPHA 1350S) with an 8.3 kW power drive and a

Fig. 1. Experimen

6000 rpm maximum spindle speed. The surface roughnessvalues were measured immediately after the turningprocess at five different locations on the circumference ofthe workpiece using a surface roughness tester(Mitutoyo-Formtracer CS 5000) with an accuracy rate of±0.01 lm. The average surface roughness was calculatedand used in an analysis of the workpiece morphology aftermachining. A Field Emission Scanning Electron Microscopy(FESEM) Supra-35VP, Carl Zeiss and an Atom Force Micros-copy (AFM) model SPM-9500J2 with the contact tip canti-lever were also used to record surface characterization.Images were captured from a 5 � 5 lm2 scanning area.Each measurement was repeated five times and the aver-age surface roughness value was reported.

3. Experimental results

The surface roughness of the machined samples wasmeasured with a hybrid surface contour measuringmachine. An evaluation of the machined surface finish

tal scheme.

Fig. 2. AFM image analysis software.

M. Marani Barzani et al. / Measurement 61 (2015) 150–161 153

was created using an AFM image analysis software over a5 � 5 lm2 scanning area (Fig. 2). Fig. 3 presents the FESEMimages and AFM topographical images of the machinedsurface of Al–11.3Si–2Cu–0.4Fe workpieces and for work-pieces containing Bi, Sb, and Sr at a cutting speed of250 m/min and a feed rate of 0.15 mm/rev. The imagesrevealed that the workpieces with Bi had the best surfaceroughness of about 1 lm (Fig. 3c). In the 3D images, hillsand valleys are observed and the workpiece containing Bihad the lowest hills (Fig. 3d). By contrast, the Sr-containingworkpiece had the greatest surface roughness value of3.8-lm (Fig. 3g) and the highest hills in the 3-D image(Fig. 3h). Moreover, the images show that surface tearingin Sr containing workpieces was greater than for workpiec-es composed of other additives or the base alloys.

4. Fuzzy Logic

Fuzzy logic is a continuous conversion from true to falseconditions, as opposed to the separate true–false transitionseen in binary logic. The possibilities presented in fuzzylogic provides a measure of a subset’s potential ability tobelong to another subset. Fuzzy logic has an extensivescope and range of applications compared to other statisti-cal methods. In engineering applications, fuzzy logicutilizes this continuous subset membership transition tochange wavy numeric problems into fuzzy linguisticterritories. Fuzzy logic employs conventional language todefine variables and fuzzy linguistic rules to describe rela-tionships as opposed to working with numeric variables

and mathematical functions. Fuzzy logic makes it possibleto use accrued experience and knowledge in the rules-of-thumb form that cannot be incorporated into mathemati-cal formula. The most notable use of fuzzy logic is to sim-ulate complex and non-linear systems while maintainingthe physical inferences and effects of every variable. In thisstudy, the fuzzy rule base contained a group of IF–THENdeclarations for thirty-six rules with three inputs, feed rate(A), cutting speed (B) and workpiece (C) with one output(surface roughness (D)). The general structure of fuzzyargumentation for the three inputs and one output of thefuzzy logic unit were defined as follows:

Rule 1 : if A is X1 and B is Y1 and C is Z1 then D is W1

Rule 2 : if A is X2 and B is Y2 and C is Z2 then D is W2

Rule n : if A is Xi and B is Yj and C is Zk then D is Wm

ð1Þ

Xi, Yj, Zk, and Wm are fuzzy subsets distinctive by their cor-responding membership functions, kXi, kYj, kZk, and kWm,respectively and n represents the rule’s number. Table 3portrays the value of surface roughness which measuredby surface roughness tester. Moreover, it shows a descrip-tion of all If-THEN rules. As shown in Table 3, Rule 1 is typ-ically selected as follows:

Rule 1 : if A is Low and B is Low and C is 1;then D is good:

(c)Machined surface

workpiece

workpiece

Machined surface(a)

(e)

workpiece

Machined surface

Machined surface

workpiece

(g)

Fig. 3. FESEM micrographs and atomic force microscopy topographical of machined surface for different workpiece materials (a and b) base alloy(Ra = 1.4 lm), (c and d) containing Bi (Ra = 1 lm), (e and f) containing Sb (Ra = 2.1 lm), and (g and h) containing Sr (Ra = 3.8 lm).

154 M. Marani Barzani et al. / Measurement 61 (2015) 150–161

Thirty-six fuzzy rules were established based on theexperimental conditions shown in Table 3. By adhering tothe maximum–minimum compositional process, the fuzzylogic of these rules produced fuzzy output. Assuming thatA, B, and C are the three input parameters of the fuzzy logicunit, the membership function of the fuzzy logic output isstated as [26]:

kW0ðDÞ ¼ ½kX1ðAÞ ^ kY1ðBÞ ^ kZ1ðCÞ ^ kW2ðDÞ_ . . . kX3ðAÞ ^ kY3ðBÞ ^ kZ4ðCÞ ^ kW3ðDÞ� ð2Þ

where _ is the maximum and ^ is the minimum operation.Membership functions come in different forms includ-

ing triangular, trapezoidal, Gaussian, and sigmoid. In thispresent study, both Gaussian and triangular membership

Table 3Linguistic variables used for experimental results.

Exp. Inputs Output

Feed rate (A) (mm/rev) Cutting speed (B) (m/min) Workpiece (C) Surface roughness (D) (lm)

1 0.05 Low 70 Low Base alloy 1 1.999 Good2 0.05 Low 70 Low 1 Bi 2 1.591 Excellent3 0.05 Low 70 Low 0.5 Sb 3 2.281 Good4 0.05 Low 70 Low 0.06 Sr 4 2.419 Good5 0.1 Medium 70 Low Base alloy 1 5.058 Bad6 0.1 Medium 70 Low 1 Bi 2 4.311 Average7 0.1 Medium 70 Low 0.5 Sb 3 5.251 Bad8 0.1 Medium 70 Low 0.06 Sr 4 6.106 Bad9 0.15 High 70 Low Base alloy 1 5.200 Bad

10 0.15 High 70 Low 1 Bi 2 5.012 Bad11 0.15 High 70 Low 0.5 Sb 3 5.881 Bad12 0.15 High 70 Low 0.06 Sr 4 6.567 Bad13 0.05 Low 130 Medium Base alloy 1 1.502 Excellent14 0.05 Low 130 Medium 1 Bi 2 1.295 Excellent15 0.05 Low 130 Medium 0.5 Sb 3 2.167 Good16 0.05 Low 130 Medium 0.06 Sr 4 2.723 Good17 0.1 Medium 130 Medium Base alloy 1 3.555 Average18 0.1 Medium 130 Medium 1 Bi 2 3.131 Average19 0.1 Medium 130 Medium 0.5 Sb 3 3.791 Average20 0.1 Medium 130 Medium 0.06 Sr 4 4.142 Average21 0.15 High 130 Medium Base alloy 1 3.449 Average22 0.15 High 130 Medium 1 Bi 2 3.731 Average23 0.15 High 130 Medium 0.5 Sb 3 4.163 Average24 0.15 High 130 Medium 0.06 Sr 4 4.885 Average25 0.05 Low 250 High Base alloy 1 1.355 Excellent26 0.05 Low 250 High 1 Bi 2 1.034 Excellent27 0.05 Low 250 High 0.5 Sb 3 2.072 Good28 0.05 Low 250 High 0.06 Sr 4 2.665 Good29 0.1 Medium 250 High Base alloy 1 1.992 Good30 0.1 Medium 250 High 1 Bi 2 1.725 Good31 0.1 Medium 250 High 0.5 Sb 3 2.489 Good32 0.1 Medium 250 High 0.06 Sr 4 2.801 Good33 0.15 High 250 High Base alloy 1 3.028 Average34 0.15 High 250 High 1 Bi 2 2.401 Good35 0.15 High 250 High 0.5 Sb 3 4.029 Average36 0.15 High 250 High 0.06 Sr 4 4.573 Average

M. Marani Barzani et al. / Measurement 61 (2015) 150–161 155

functions were used for the input parameters (feed rate,cutting speed and different workpieces) and the outputparameter of (machining surface roughness, respectively.

The Gaussian fuzzy membership function often charac-terizes vague, linguistic terms as illustrated by:

kAnðXÞ ¼ exp�ðCn � XÞ2

2r2n

!ð3Þ

where Cn and rn are the centre and width of the nth fuzzyset An, respectively.

The triangular membership function for output wasdefined using three parameters

f ðx; a; b; cÞ ¼

0; x 6 a:x�ab�a a 6 x 6 b:c�xc�b b 6 x 6 c:

0; c 6 x:

8>>><>>>:

9>>>=>>>;

ð4Þ

In terms of the minimum and maximum, an alternatestatement for the next equation was presented:

f ðx; a; b; cÞ ¼ max minx� ab� a

;c � xc � b

� �; 0

� �ð5Þ

where a, b, and c define the triangular fuzzy triplet anddetermine the x coordinates of the three corners of theunderlying triangular membership function.

The numerical input–output values are connected bylinguistic variables, something attained by designing mem-bership functions consisting of a set of fuzzy set values.Linguistic values such as LOW, MEDIUM, and HIGH showthe input variables feed rate, cutting speed and workpieces1, 2, 3 and 4 that represent the base alloy workpiece andthe workpieces containing Bi, Sb and Sr, respectively. Theoutput numerical values (surface roughness) are likewiseclosely related by using membership functions includingEXCELLENT, GOOD, AVERAGE, and BAD. The linguistic vari-ables and experimental results that served as input andoutput parameters are summarized in Table 3.

The Gaussian membership functions for the inputparameters of feed rate and cutting speed and differentworkpieces along with the triangular membership functionfor the output parameter of surface roughness are illus-trated in Fig. 4.

A defuzzification process was carried out as part of thisstudy. Defuzzification is imperative in the theory of fuzzysets as it translates fuzzy set information into numeric

Membership func�onMembership func�on

Membership func�onMembership func�on

Cu�ng speed (mm/min)Feed rate (mm/rev)

Surface roughness (μm)Workpiece (1, 2, 3, 4)

(a) (b)

(c) (d)

Fig. 4. Membership functions for input and output parameters using triangular membership function (a) MF for feed rate (b) MF for cutting speed (c) MF forworkpiece and (d) MF for surface roughness.

156 M. Marani Barzani et al. / Measurement 61 (2015) 150–161

data. This process is accompanied by fuzzification, which iscritical in designing fuzzy systems because both of fuzzifi-cation and fuzzy system procedures deliver a series of con-nections between the fuzzy set region and real-valuedscalar region. Defuzzification with a centroid form wasselected for use in this study because it provided the centrearea of the possibility distribution of the inference outputand it is a popular defuzzification method for calculatingthe centroid of the area under a membership function [27].

D0 ¼P

DkW0ðDÞPkW0ðDÞ

ð6Þ

The non-fuzzy value D0 generates an output value innumerical form. The shape difference of the input and out-put membership functions and the selection of the type ofthe input membership functions is related to the influenceof each input on the output and the complexity of the phe-nomenon. In choosing membership functions for fuzzifica-tion, the event and type of membership functions arelargely dependent on the relevant event. Gauss shape ofthe membership function employed to describe the fuzzysets for input variables. In the output variable fuzzy set, tri-angular membership functions are used. The triangularmembership function has been selected among differenttypes of membership functions such as trapezoidal, Gauss-ian and sigmoid in order to obtain the most accurate pre-diction model. All four types of membership functionswere examined and the best result was obtained by usingtriangular membership function in terms of percentage ofthe error. The linearity of the triangular membership func-tion is another advantage of this model in comparison toGaussian and sigmoid membership functions that has asignificant role in describing the behavior of the phenom-enon. Additionally, the number of consequent parameters

(unknown parameters in output membership function) intriangular membership function is three while this valuefor trapezoidal membership function is four. Thus, theresult of using triangular membership function as the out-put membership function has better agreement withexperimental result due to the less numbers of experimen-tal test to obtain the value of consequent parameters [28].

In this study, the surface roughness predicted by fuzzylogic in relation to changes in the parameters is shown inFig. 5. The surface roughness predicted by fuzzy logic forchanging feed rates and cutting speeds is shown in Fig. 5aand the results show that surface roughness decreased withincreasing cutting speeds from 70 to 250 m/min. Fig. 5 fur-ther demonstrates that the best surface roughness wasobtained using low feed rates (0.05 mm/rev) and high cut-ting speeds (250 m/min). Fig. 5b illustrates the surfaceroughness predicted using fuzzy logic for changing feedrates and workpiece materials. Fig. 5b also shows that sur-face roughness increased when feed rate increased from0.05 to 0.15 mm/rev. At a lower feed rates, however, Work-piece 2 containing Bi had superior surface roughness com-pared to other workpieces. Fig. 5c shows surface roughnesspredicted using fuzzy logic for changes in workpieces andcutting speeds. In this figure, surface roughness clearlydecreased at the highest cutting speeds for all workpieces.The best surface roughness was obtained by the workpiececontaining Bi (workpiece 2).

4.1. Investigating fuzzy model accuracy and error

To investigate the accuracy of the fuzzy model and errorrate, twelve new experimental tests were carried out afterthe construction of fuzzy rules and the proposed fuzzymodel helped predict the surface roughness for the same

Fig. 5. The predicted surface roughness by fuzzy logic in relation to parameters change (a) cutting speed and feed rate (b) workpiece and feed rate and (c)workpiece and cutting speed.

M. Marani Barzani et al. / Measurement 61 (2015) 150–161 157

conditions (Table 4). Errors were calculated by measuringthe gap between the main measured value and predictedvalue. The measured values can be calculated usingEq. (7), where ei represents an individual error, Hm is themeasured value, and Hp is the predicted value. Thepercentage for individual errors was obtained by dividingthe absolute difference of the predicted by the measure value.

ei ¼jHm � Hpj

Hm

� �� 100% ð7Þ

Table 4Error and accuracy of the fuzzy logic model prediction.

No. ofexperiment

Parameters (inputs) Sur

Merou

Feed rate (A)(mm/rev)

Cutting speed (B)(m/min)

Work piece(C)

Av

1 0.07 90 Bas alloy 2.32 0.07 90 Bi 1.83 0.07 90 Sb 2.44 0.07 90 Sr 2.85 0.13 180 Bas alloy 3.76 0.13 180 Bi 3.87 0.13 180 Sb 3.58 0.13 180 Sr 4.19 0.18 230 Bas alloy 2.7

10 0.18 230 Bi 2.611 0.18 230 Sb 4.012 0.18 230 Sr 4.8

Accuracy is calculated by finding the proximity of thepredicted value to the measured value. In Eq. (8), A is theaccuracy of the model and N is the total number of datasets tested. The accuracy of the model is the average indi-vidual accuracy.

A ¼ 1N

XN

i¼1

1� jHm � HpjHm

� �� 100% ð8Þ

face roughness result (output) (lm)

asured surfaceghness

Predicted surfaceroughness (Fuzzy)

Error%

Accuracy%

erage (D)

2.4 4.35 95.657 2 6.95 93.041 2.5 3.73 96.264 3 5.63 94.366 3.9 3.72 96.281 3.9 2.36 97.630 3.8 8.57 91.425 4.4 6.02 93.978 3 7.91 92.081 2.8 7.28 92.725 4.2 3.70 96.298 5.1 4.51 95.49

Accuracy ofmodel = 94.60%

Fig. 6. Comparison of fuzzy logic model prediction with the measuredresults of surface roughness.

Table 5The values of the calculated (S/N) ratio.

Exp. Calculated S/N ratio

1 6.0162562 4.0334043 7.1625064 7.6727175 14.079586 12.691567 14.404848 15.715149 14.3200710 14.0002211 15.3890212 16.3473413 3.53339914 2.24539515 6.71717816 8.70095317 11.0167918 9.91366119 11.5750820 12.344221 10.7538622 11.436523 12.3881324 13.7772925 2.63878626 0.29041127 6.32779528 8.51394429 5.98578730 4.73578231 7.92049832 8.94626233 9.62311734 7.60784335 12.1039536 13.20402

Table 6S/N response values for surface roughness.

158 M. Marani Barzani et al. / Measurement 61 (2015) 150–161

The error percentage for the dataset result was calcu-lated and the accuracy of the fuzzy logic model was deter-mined. The experimental conditions, surface roughnessresults, and fuzzy model predicted values are shown inTable 4. The average percentage of error for fuzzy modelprediction was 5.4%. The low error level indicates that sur-face roughness results predicted by the fuzzy logic modelwere very close to the actual experimental values. Table 4also shows that the accuracy of the fuzzy model was94.60%. The accuracy percentage shows that the proposedmodel can be successfully used to predict the surfaceroughness of machining Al–11.3Si–2Cu–0.4Fe die castingalloys.

Fig. 6 compares the predictions from the fuzzy logicmodel with the actual measured surface roughness aftermachining the workpieces. Fig. 6 shows that the proposedmodel was able to accurately predict the surface roughnessof machined workpieces. The similarities between surfaceroughness values means that the fuzzy logic model is apromising solution for predicting surface roughness valuesfor a specific range of parameters.

Machining parameters S/N response

Level 1 Level 2 Level 3 Level 4

Feed rate (A) 5.321 10.777 12.579 –Cutting speed (B) 11.820 9.533 7.325 –Workpiece (C) 8.663 7.440 10.443 11.691

5. Result analysis using Pareto ANOVA: An alternativeanalysis

This section presents the surface roughness analysis ofmachining (turning) parameters produced using Pareto-ANOVA. Pareto-type analysis facilitates the investigationof the significance of factors and their interactions, whileallowing for optimal factor levels to be attained [29]. Par-eto ANOVA analysis is an excellent tool for determiningthe contribution of each input parameter and their interac-tions with the output parameters. It is a simplified ANOVAanalysis method that does not require an ANOVA table.Moreover, the Pareto-ANOVA analysis method not onlyrequires minimal knowledge of the ANOVA technique,but is also appropriate for industrial and engineering appli-cations [30]. Table 5 shows the calculated S/N ratios forsurface roughness. Pareto-ANOVA was used to create astandard orthogonal array to accommodate this require-ment. S/N ratio was used as a measurable value insteadof standard deviation because of the mean decreases. Inother words, the standard deviation could not be mini-mized so that the target could meet requirements.

Table 7 presents the Pareto-ANOVA analysis for surfaceroughness of a machined workpiece. The sum of squares ofthe difference between the levels of A, Sa and Sb are asfollows:

Sa;b ¼ ðA1 � A2Þ2 þ ðA1 � A3Þ2 þ ðA2 � A3Þ2 ð9Þ

The sum of the squares of the differences for Sc can becalculated as shown below:

Sc ¼ ðA1 � A2Þ2 þ ðA1 � A3Þ2 þ ðA1 � A4Þ2

þ ðA2 � A3Þ2 þ ðA2 � A4Þ2 þ ðA3 � A4Þ2 ð10Þ

The contribution ratio for each parameter is the per-centage of the sum of the squares of the differences foreach parameter compared to the total sum of the squaresof differences. The reason for using Pareto-ANOVA in this

Table 7Pareto-ANOVA analysis for surface roughness of specimens after machining process.

Factor and interaction Feed rate (A) Cutting speed (B) Workpiece (C)

Summation at the level of input parameter1 5.321 11.820 8.6632 10.777 9.533 7.4403 12.579 7.325 10.4434 – – 11.691

Total of summation at factor level 28.678 28.678 38.237Summation of squares of differences (S) Sa = 85.70 Sb = 30.304 Sc = 42.43Total of summation of squares of differences St = Sa + Sb + Sc 158.434Contribution ratio (%) 54.092 19.127 26.781Pareto diagram

0

10

20

30

40

50

60

70

80

90

100

A C B

26.78119.127

54.092

Cumulative contribution ratio 54.092 80.873 100Optimum combination A1 C2 B3Remarks on optimum condition The significant factors are chosen from the left-hand side in the above Pareto diagram

which cumulatively contribute UP TO 90%Overall optimum conditions for all factors A1B3C2

Fig. 7. Average value of response graph for surface roughness.

M. Marani Barzani et al. / Measurement 61 (2015) 150–161 159

study was to analyze the machining parameters that leadto reduced surface roughness, which subsequentlyrevealed that the best surface finish was provided by anAl alloy.

Table 7 presents the Pareto-ANOVA analysis of the sur-face roughness for specimens after machining. Table 7shows that the Pareto diagram, feed rate (A1 = 0.05 mm/rev) had the greatest effect on surface roughness duringmachining process and cutting speed (B3 = 250 m/min)had the least effect on surface roughness. The Contributionratio (%) for machining parameters included feed rate (A1),cutting speed (B3), and workpiece (C2, containing Bi) are54.092, 19.127, and 26.781, respectively. The overall opti-mum conditions for achieving the best surface roughnesswas A1B3C2.

The Pareto-ANOVA analysis (Table 7) indicates that feedrate (54.092) was the parameter that had the most signif-icant effect on surface roughness. The influence of variousparameters on machining with respect to parameter levelsis demonstrated as a response table for surface roughness(Table 6). This table shows the average for each responsecharacteristic for every level of machining factors of Al–Si–Cu–Fe die cast alloy. These values are plotted in Fig. 7to create a response (surface roughness) graph for thethree parameters, where it is evident that surface rough-ness increased when the feed rate was increased from0.05 to 0.15 mm/rev. Fig. 7 also indicates that surfaceroughness improved by adding Bi. By contrast, other addi-tives such as 0.05 Sb and 0.06 Sr drastically increased sur-face roughness. In addition, surface roughness wasenhanced by increasing the cutting speed from 70 to

250 m/min. This analysis was not based on error terms;thus some sources of uncertainty were omitted. For thisreason, the effect of the parameters on machining was fur-ther analyzed using ANOVA. Based on the Pareto-ANOVApresented in Table 7, the response table and the responsegraph, minimum surface roughness can be achievedusing a low feed rate (0.05 mm/rev), high cutting speed(250 m/min) for Bi-containing workpieces (workpiece 2).Table 7 emphasized that the optimal levels illustrated bythese results are close to those based on the parametersand their predicted levels. A study of Fig. 7 suggests thatthe most favorable combination for reduced surfaceroughness is A1B3C2.

160 M. Marani Barzani et al. / Measurement 61 (2015) 150–161

6. Discussion

The dry machining of Al–Si cast alloy has been used bymanufacturing industries, especially the automobileindustry, because significant reductions in mass can beattainable compared to traditional cast iron and steelmaterials. Workpieces with different compositions andmicrostructures require different machining parameters.The machinability of Al–11.3Si–2Cu–0.4 Fe die castingalloy with Bi-, Sb- and Sr-additives was predicted and opti-mized in terms of cutting speed and feed rate using a fuzzylogic method and Pareto-ANOVA analysis. Fuzzy logic is animportant tool and is used in many engineering applica-tions. In the present work, fuzzy logic was used to predictsurface roughness for the machining of Al–11.3Si–2Cu–0.4Fe die casting alloy. In addition, Pareto-ANOVA was used tooptimize the best combination of machining parametersfor enhanced surface roughness and to find the mostfavorable combination for reduced surface roughness. Theresults were evaluated using field emission scanningelectron microscopy and atom force microscopy.

According to the predictions, surface roughnessdecreases when cutting speeds rose from 70 to 250 m/minand surface roughness increased when feed rates increasedfrom 0.05 to 0.15 mm/rev due to the domination offeed marks and the increase of distance from peak to valleyof the machined surface. The Pareto-ANOVA resultsshowed that minimum surface roughness is feasible whenlow feed rates (0.05 mm/rev) and high cutting speeds(250 m/min) are used. In addition, workpieces containingBi demonstrated the least amount of surface roughnessand best machinability in contrast to the base alloy and

Aluminium

Matrix

Plate-like Si shape

Al

F

Lamellar Si

Fig. 8. Optical microstructures at feed rate of 0.05 mm/rev and cutting speed ofworkpieces.

workpieces containing Sb and Sr. One explanation is thata decrease in BUE formation occurred, and thus at a greatercutting speed, chips could be easily removed from the cut-ting tool which lead to the best surface finish. Furthermore,Bi is a known lubricant for the machining process [8].

Optical workpiece microstructures are illustrated inFig. 8. Coarse, plate-like Si formations was observed inthe aluminum matrix of the base alloy (Fig. 8a). Coarse,plate-like Si formations are considered to be harmful forelongation and fatigue and they represent a potential sitefor cracks. Therefore, the alloy was modified by addingonly small amounts of certain elements. Fig. 8b showsthe how the Si morphology changed from a plate-like for-mation to a lamellar structure after the addition of Bi,which is supported by previous studies [27]. Fig. 8b alsoshows that the silicon particle changed in the size andshape. Moreover, adding Sb had a similar effect to addingBi, as shown in Fig. 8c. The coarse plate-like Si formationstransformed into a refined lamellar structure. However, thesurface roughness results from topographical microscopywere different for Bi and Sb-containing workpieces. Theaddition of Sr had a significant effect on Si morphologyas the coarse plate-like Si formations were convert to a finefibrous form as shown in Fig. 8d. Since the Si morphologychanged after adding Bi, Sb, and Sr, it can be deduced thatchanges in topography were related to changes in themorphology of the silicon hard phase embedded in thealuminum soft matrix. Imai and Rabinowiczg found thatlow-melting point metals, like Bi (272 �C), in their pureform, may act as lubricants. Dasch et al. [31] reported thatdry machining is significantly improved by incorporatingsmall amounts of Bi in the alloy composition.

Lamellar Si

Al

Al

ibrous Si

250 m/min (a) base alloy, (b) Bi-treated, (c) Sb-treated and (d) Sr-treated

M. Marani Barzani et al. / Measurement 61 (2015) 150–161 161

7. Conclusion

In this study, the effect of different additives (bismuth,antimony and strontium) and machining parameters onthe surface roughness of Al–11.3Si–2Cu–0.4 Fe die castingalloy was predicted using a fuzzy logic technique. Fuzzylogic and Pareto-ANOVA were used to predict and optimizethe best combination of machining parameters to enhancesurface roughness. The results showed that surface rough-ness increases with feed rates the increased from 0.05 to0.15 mm/rev, and that surface roughness improved whencutting speed increased from 70 to 250 m/min. In termsof additives, the best surface roughness was obtained forBi-containing alloys possibly because the formation of pureBi behaves as lubricant during turning. The fuzzy modelpercentages of error and accuracy were 5.4% and 94.60%,respectively, indicating that the fuzzy logic predictionmodel can accurately predict the surface roughness ofAl–11.3Si–2Cu–0.4 Fe die casting alloy. Pareto-ANOVArevealed that the optimal levels were A1B3C2. The experi-mental results agreed the optimal levels based on theparameters and their levels.

Acknowledgements

This research was funded by the high impact research(HIR) Grant number: HIR-MOHE-D000023-16001 fromthe Ministry of Higher Education, Malaysia and the Univer-sity of Malaya Postgraduate Research Grant (PPP) ProgramNo. PG016-2013A.

References

[1] D.B. Miracle, Metal matrix composites – from science totechnological significance, Compos. Sci. Technol. 65 (2005) 2526–2540.

[2] J. Grum, M. Kisin, Influence of microstructure on surface integrity inturning—part I: the influence of the size of the soft phase in amicrostructure on surface-roughness formation, Int. J. Mach. ToolsManuf. 43 (2003) 1535–1543.

[3] M. Marani Barzani, S. Farahany, N. Mohd Yusof, A. Ourdjini, Theinfluence of bismuth, antimony and strontium on microstructure,thermal and machinability of aluminium–silicon alloy, J. Mater.Manuf. Process 28 (2013) 1184–1190.

[4] X.H. Zhang, G.C. Su, C.W. Ju, W.C. Wang, W.L. Yan, Effect ofmodification treatment on the microstructure and mechanicalproperties of Al–0.35%Mg–7.0%Si cast alloy, Mater. Des. 31 (2010)4408–4413.

[5] S. Gopalakannan, T. Senthilvelan, Application of response surfacemethod on machining of Al–SiC nano-composites, Measurement 46(2013) 2705–2715.

[6] J.M. Dasch, C.C. Ang, C.A. Wong, R.A. Waldo, Y.T. Chester, The effect offree-machining elements on dry machining of B319 aluminum alloy,J. Mater. Process. Technol. 209 (2009) 4638–4644.

[7] A.J. Makadia, J.I. Nanavati, Optimisation of machining parameters forturning operations based on response surface methodology,Measurement 46 (2013) 1521–1529.

[8] E.O. Ezugwu, Z.M. Wang, C.I. Okeke, Tool life and surface integritywhen machining Inconel 718 with PVD-and CVD-coated tools,Tribol. Trans. 42 (1999) 353–360.

[9] M.M. Barzani, M.Y. Noordin, A.A. Farid, S. Farahany, A. Davoudinejad,Effects of cutting condition on surface roughness when turning

untreated and Sb-treated Al–11% Si alloys using PVD coated tools,Appl. Mech. Mater. 315 (2013) 413–417.

[10] A. Gupta, H. Singh, A. Aggarwal, Taguchi-fuzzy multi outputoptimization (MOO) in high speed CNC turning of AISI P-20 toolsteel, Expert Syst. Appl. 38 (2011) 6822–6828.

[11] E. Elgallad, F. Samuel, A. Samuel, H. Doty, Machinability aspects ofnew Al–Cu alloys intended for automotive castings, J. Mater. Process.Technol. 210 (2010) 1754–1766.

[12] C.-Y. Yang, S.-L. Lee, C.-K. Lee, J.-C. Lin, Effects of Sr and Sb modifierson the sliding wear behavior of A357 alloy under varying pressureand speed conditions, Wear 261 (2006) 1348–1358.

[13] M.M. Barzani, A.A.D. Sarhan, S. Farahany, R. Singh, I. Maher,Machinability of Al-Si-Cu cast alloy containing bismuth andantimony when dry turning using coated carbide insert,Measurement (2014), http://dx.doi.org/10.1016/j.measurement.2014.10.030.

[14] M.M. Barzani, M.Y. Noordin, S. Farahany, A. Ourdjini, Effect ofmachining parameters on cutting force when turning untreated andSb-treated Al–11% Si–1.8% Cu alloys using PVD coated tools, Appl.Mech. Mater. 234 (2012) 74–77.

[15] E. Zalnezhad, A.A.D. Sarhan, M. Hamdi, A Fuzzy Logic based Model toPredict surface hardness of thin film TiN Coating on AerospaceAL70751, Int. J. Adv. Manuf. Tech. 68 (1–4) (2013) 415–423.

[16] A. Garg, K. Tai, Review of genetic programming in modeling ofmachining processes, in: Modelling, Identification & Control (ICMIC),2012 Proceedings of International Conference on, IEEE, 2012, pp.653–658.

[17] C.S.N. Azwadi, M. Zeinali, A. Safdari, A. Kazemi, Adaptive-network-based fuzzy inference system analysis to predict the temperatureand flow fields in a lid-driven cavity, Numer. Heat Transfer, Part A:Appl. 63 (2013) 906–920.

[18] A. Garg, Y. Bhalerao, K. Tai, Review of empirical modellingtechniques for modelling of turning process, Int. J. Model. Ident.Control 20 (2013) 121–129.

[19] Q. Ren, M. Balazinski, L. Baron, High-order interval type-2 Takagi-Sugeno-Kang fuzzy logic system and its application in acousticemission signal modeling in turning process, Int. J. Adv. Manuf.Technol. 63 (2012) 1057–1063.

[20] E. Zalnezhad, A.A.D. Sarhan, A Fuzzy Logic Predictive Model forMeasured Surface Roughness of Ti-TiN Coating on AL7075-T6 Alloyfor Longer Fretting Fatigue Life, Measurement 49 (2014) 256–265.

[21] A.K. Nandi, J. Paulo Davim, A study of drilling performances withminimum quantity of lubricant using fuzzy logic rules, Mechatronics19 (2009) 218–232.

[22] S.-P. Lo, An adaptive-network based fuzzy inference system forprediction of workpiece surface roughness in end milling, J. Mater.Process. Technol. 142 (2003) 665–675.

[23] S. Ramesh, L. Karunamoorthy, K. Palanikumar, Fuzzy modeling andanalysis of machining parameters in machining titanium alloy,Mater. Manuf. Processes 23 (2008) 439–447.

[24] L.A. Zadeh, Fuzzy sets, Inf. Control 8 (1965) 338–353.[25] S. Farahany, A. Ourdjini, M.H. Idris, The usage of computer-aided

cooling curve thermal analysis to optimize eutectic refiner andmodifier in Al–Si alloys, J. Therm. Anal. Calorim. 109 (2012) 105–111.

[26] H. Zimmermann, Fuzzy Set Theory and Its Applications Second,Revised Edition, Springer, 1992.

[27] J. Masounave, Y.A. Youssef, Y. Beauchamp, M. Thomas, Anexperimental design for surface roughness and built-up edgeformation in lathe dry turning, Int. J. Qual. Sci. 2 (1997) 167–180.

[28] E. Zalnezhad, A.A.D.M. Sarhan, M. Hamdi, Surface hardnessprediction of CrN thin film coating on AL7075-T6 alloy using fuzzylogic system, Int. J. Prec. Eng. Manuf. 14 (2013) 467–473.

[29] I. Maher, M.E.H. Eltaib, A.A.D. Sarhan, R.M. El-Zahry, Investigate theeffect of the machining parameters on the surface quality ofmachined Brass (60/40) in CNC End-Milling-Automotiveapplication, Int. J. Adv. Manuf. Tech., 74 (1–4), 2014, 531-537.

[30] S. Tanaydin, Robust design and analysis for quality engineering,Technometrics 40 (1998) 348–349.

[31] J.M. Dasch, C. Ang, C. Wong, R. Waldo, D. Chester, Y. Cheng, B. Powell,A. Weiner, E. Konca, The effect of free-machining elements on drymachining of B319 aluminum alloy, J. Mater. Process. Technol. 209(2009) 4638–4644.