+ All Categories
Home > Documents > Optimization of anodic layer properties on aluminium in mixed oxalic/sulphuric acid bath using...

Optimization of anodic layer properties on aluminium in mixed oxalic/sulphuric acid bath using...

Date post: 12-Nov-2023
Category:
Upload: independent
View: 0 times
Download: 0 times
Share this document with a friend
10
Available online at www.sciencedirect.com Materials Chemistry and Physics 108 (2008) 296–305 Optimization of mechanical and chemical properties of sulphuric anodized aluminium using statistical experimental methods W. Bensalah a , K. Elleuch b , M. Feki a , M. Wery c , M.P. Gigandet d , H.F. Ayedi a,a Unit´ e de recherche de Chimie Industrielle et Mat´ eriaux (URCIM), ENIS, B.P.W. Sfax, Tunisie b Laboratoire des Syst` emes Electrom´ ecaniques (LASEM), ENIS, B.P.W. Sfax, Tunisie c IUT Mesures Physiques d’Orsay, Plateau du Moulon, 91400 Orsay, France d LCMI-Corrosion et Traitements de surface 16, Route de Gray, 25030 Besan¸ con Cedex, France Received 21 September 2006; received in revised form 4 September 2007; accepted 29 September 2007 Abstract We described a three-step strategy to achieve simultaneous optimization of mechanical and chemical properties of an anodic aluminium oxide layer elaborated in a sulphuric acid solution. In the first two steps, a Doehlert design was carried out and then the canonical analysis has been conducted to study the four fitted models of the responses, namely: dissolution rate, Vickers microhardness, weight loss after abrasion and deflection at failure of the anodic oxide layer. Canonical analysis showed that the experimental conditions where the optima are found for each individual response are just opposite, so it is required to look for a certain compromise, which was achieved using the desirability function, in the last step. The morphology and the composition of “optimum” layer was examined by scanning electron microscopy (SEM), atomic force microscopy (AFM) and glow-discharge optical emission spectroscopy (GDOES). © 2007 Elsevier B.V. All rights reserved. Keywords: Sulphuric anodization; Experimental design and canonical analysis; Microhardness and abrasion 1. Introduction The use of aluminium alloys in industry has increased in the last few decades due to its exceptional properties (low-specific mass, good thermal and electrical conductivities, good corrosion resistance, etc.). Aluminium and its alloys are naturally protected by an oxide film when they are exposed to air. However, this natural oxide is thin and heterogeneous and it does not offer sufficient pro- tection against aggressive environments. Moreover, the friction behaviour of aluminium alloys is mediocre. In order to enhance the surface properties and mechanical strength of aluminium, anodizing has been used [1–9]. This treatment, which is an elec- trochemical process, consists on converting aluminium into its oxide by appropriate selection of the electrolyte and the anodiz- ing conditions, such as current density, voltage, temperature, etc. Corresponding author. Tel.: +216 74 274 088; fax: +216 74 275 595. E-mail address: [email protected] (H.F. Ayedi). Performances of sulphuric acid bath and properties of anodic oxide have been intensively developed during the past decades [10–16]. In general, the investigations on aluminium or dilute aluminium alloys focused on the characteristics of the formed oxide layers (e.g. thickness, density, hardness, corro- sion resistance, morphology, etc.). These previous works have investigated the influence of each anodizing parameter one at a time while keeping the others constant. This traditional step-by- step approach for optimization purposes involves a large number of independent runs and does not take into account the possi- ble interactions between factors. To avoid these disadvantages, an experimental design is the most efficient means to reach conclusions with a minimum of trials. The use of multivari- ate experimental design techniques is becoming increasingly widespread in several research fields. Multivariate designs, which allow the simultaneous optimisation of several control variables of a process, are faster to implement and more cost- effective than traditional univariate approaches [17–21]. In the present work, the Doehlert experimental design [22,23] was used to establish the effect of the input factors: bath temperature, anodic current density and sulphuric acid concentration, and their interactions on some properties of the aluminium oxide 0254-0584/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.matchemphys.2007.09.043
Transcript

A

lcar

K

1

lmr

fiitbtatoie

0d

Available online at www.sciencedirect.com

Materials Chemistry and Physics 108 (2008) 296–305

Optimization of mechanical and chemical properties of sulphuricanodized aluminium using statistical experimental methods

W. Bensalah a, K. Elleuch b, M. Feki a, M. Wery c, M.P. Gigandet d, H.F. Ayedi a,∗a Unite de recherche de Chimie Industrielle et Materiaux (URCIM), ENIS, B.P.W. Sfax, Tunisie

b Laboratoire des Systemes Electromecaniques (LASEM), ENIS, B.P.W. Sfax, Tunisiec IUT Mesures Physiques d’Orsay, Plateau du Moulon, 91400 Orsay, France

d LCMI-Corrosion et Traitements de surface 16, Route de Gray, 25030 Besancon Cedex, France

Received 21 September 2006; received in revised form 4 September 2007; accepted 29 September 2007

bstract

We described a three-step strategy to achieve simultaneous optimization of mechanical and chemical properties of an anodic aluminium oxideayer elaborated in a sulphuric acid solution. In the first two steps, a Doehlert design was carried out and then the canonical analysis has beenonducted to study the four fitted models of the responses, namely: dissolution rate, Vickers microhardness, weight loss after abrasion and deflectiont failure of the anodic oxide layer. Canonical analysis showed that the experimental conditions where the optima are found for each individual

esponse are just opposite, so it is required to look for a certain compromise, which was achieved using the desirability function, in the last step.

The morphology and the composition of “optimum” layer was examined by scanning electron microscopy (SEM), atomic force microscopyAFM) and glow-discharge optical emission spectroscopy (GDOES). 2007 Elsevier B.V. All rights reserved.

; Micr

adofsitsobacaw

eywords: Sulphuric anodization; Experimental design and canonical analysis

. Introduction

The use of aluminium alloys in industry has increased in theast few decades due to its exceptional properties (low-specific

ass, good thermal and electrical conductivities, good corrosionesistance, etc.).

Aluminium and its alloys are naturally protected by an oxidelm when they are exposed to air. However, this natural oxide

s thin and heterogeneous and it does not offer sufficient pro-ection against aggressive environments. Moreover, the frictionehaviour of aluminium alloys is mediocre. In order to enhancehe surface properties and mechanical strength of aluminium,nodizing has been used [1–9]. This treatment, which is an elec-rochemical process, consists on converting aluminium into itsxide by appropriate selection of the electrolyte and the anodiz-

ng conditions, such as current density, voltage, temperature,tc.

∗ Corresponding author. Tel.: +216 74 274 088; fax: +216 74 275 595.E-mail address: [email protected] (H.F. Ayedi).

wveptat

254-0584/$ – see front matter © 2007 Elsevier B.V. All rights reserved.oi:10.1016/j.matchemphys.2007.09.043

ohardness and abrasion

Performances of sulphuric acid bath and properties ofnodic oxide have been intensively developed during the pastecades [10–16]. In general, the investigations on aluminiumr dilute aluminium alloys focused on the characteristics of theormed oxide layers (e.g. thickness, density, hardness, corro-ion resistance, morphology, etc.). These previous works havenvestigated the influence of each anodizing parameter one at aime while keeping the others constant. This traditional step-by-tep approach for optimization purposes involves a large numberf independent runs and does not take into account the possi-le interactions between factors. To avoid these disadvantages,n experimental design is the most efficient means to reachonclusions with a minimum of trials. The use of multivari-te experimental design techniques is becoming increasinglyidespread in several research fields. Multivariate designs,hich allow the simultaneous optimisation of several controlariables of a process, are faster to implement and more cost-ffective than traditional univariate approaches [17–21]. In the

resent work, the Doehlert experimental design [22,23] was usedo establish the effect of the input factors: bath temperature,nodic current density and sulphuric acid concentration, andheir interactions on some properties of the aluminium oxide

istry and Physics 108 (2008) 296–305 297

ltVtthabowfs

2

2

sT

aa(fsAvt

s

2

o

2

Tt

2

mof

TC

E

SMCTZFPMA

2

s

bcwm0

2

w1i2i(

2

with the ASTM B 680-80 specifications: 35 mL L 85% H3PO4 + 20 g LCrO3 at 38 ◦C. Fig. 3 shows a general evolution of the weight loss versusimmersion time. As can be seen, the weight loss increased linearly with theimmersion time up to a critical value. After that, no increase of the weight losswas observed, indicating that the oxide layer was completely dissolved. The

W. Bensalah et al. / Materials Chem

ayer. In addition, the canonical analysis has been employedo study the response fitted models namely: dissolution rate,ickers microhardness, weight loss after abrasion and deflec-

ion at failure of the anodic oxide layer. In order to find outhe best experimental conditions, which lead to maximize theardness and deflection and to minimize the dissolution ratend the abrasion, an optimization of the process under study haseen achieved using the desirability function [24,25]. Finally, thebtained aluminium oxide layer under the optimum conditionsas examined by scanning electron microscopy (SEM), atomic

orce microscopy (AFM) and glow-discharge optical emissionpectroscopy (GDOES).

. Experimental

.1. Materials and procedures

Parrallelipedic coupons 100 mm × 25 mm × 3 mm of Al were used as theubstrate for anodic conversion treatment. The composition of Al was given inable 1.

The substrates were mechanically ground to P1000 grade paper. Prior tonodizing, the specimens were treated as follows: (a) chemical polishing in15:85 (v/v) mixture of concentrated HNO3 and H3PO4 at 85 ◦C for 2 min,

b) cold water rinsing, (c) etching in 1 M NaOH solution at room temperatureor 1 min, (d) cold water rinsing, (e) chemical pickling in 30% (v/v) HNO3

olution at room temperature for 30 s and (f) deionised water rinsing and drying.fterwards, the specimens (exposed area 85 mm × 25 mm) were anodized inigorously stirred sulphuric acid solution maintained within ±0.1 ◦C of the setemperature for 90 min then washed in deionised water and dried.

The used cathodes were also aluminium sheets, with larger size than thepecimens. Sulphuric, nitric and phosphoric acids are analytical grade chemicals.

.2. Testing methods

In order to characterize the anodic oxide layer the following tests were carriedut.

.2.1. Thickness measurementThe thickness of the anodic oxide layer was measured using ELCOME-

ER 355 Top Thickness Gauge equipped with eddy current probe. The averagehickness of 30 measuring points evenly distributed on both sides was taken.

.2.2. Microhardness measurement

The Vickers microhardness of the anodic film was determined using a Vickers

icrohardness tester DELTALAB HVS-1000, a 200 g load and a loading timef 15 s. The results represented the average of approximately 20 measurementsor each sample.

able 1hemical composition of the used aluminium (wt%)

lements Wt%

i 0.11n <0.005u <0.005i 0.014n 0.009e 0.37b 0.006g <0.005l Balance

F

Fig. 1. Schematic diagram of used pin-on-disc machine.

.2.3. Abrasion testIn order to characterize the tribological properties of the anodic oxide, abra-

ion resistance was carried out on a pin-on-disc machine (Fig. 1).Anodized samples with dimensions of 20 mm × 20 mm × 3 mm were

rought into contact with 320 grit SiC paper, fixed on a rotating disc with aonstant speed of 20 rpm. The applied normal load was 5 N and the test durationas 1 min. Abrasion tests were carried out in water lubrication. For each speci-en a new abrasive paper was used. An analytical balance with an accuracy of

.1 mg was used to measure the weight of the samples before and after each test.

.2.4. Flexure testMeasurements of deflection at failure of the anodic oxide films on aluminium

ere made by performing three-points flexure test on parrallelipedic samples00 mm × 25 mm × 3 mm at room temperature. A universal machine (Lloydnstruments LR 50KN) was used for this purpose. Loading speed was fixed atmm min−1 and the calibrated distance was 50 mm. Load-deflection response

s then recorded using NEXYGEN software program. The deflection at failureDf) is then deduced as shown in Fig. 2.

.2.5. Acid dissolution testingAcid dissolution tests of the anodized samples were achieved in accordance

−1 −1

ig. 2. General load-deflection curve of an anodized aluminium specimen.

298 W. Bensalah et al. / Materials Chemistry and Physics 108 (2008) 296–305

Fd

su

2

ta5

u

2

d4i3t

2

2

ec

Table 2Study domain

Variables Number of levels Centre Uj(0) Variation step �Uj

U1 (◦C) 5 14 11UU

Taifhtt

•••

fi

mfie

wce

w

TD

N

1111111

ig. 3. General weight loss vs. immersion time of an anodized specimen duringissolution test in the mixed acid solution.

lope of the linear part represents the value of the dissolution rate (per surfacenit).

.2.6. Surface characterizationAFM, performed using model Digital instrument-Nanoscope probe II (con-

act mode), was used to examine and to determine the roughness of thenodized surfaces. Surface topography was recorded over scanned areas of0 �m × 50 �m.

The morphology of the oxide layer was studied from the topside of the layersing a scanning electron microscope (Jeol JSM-6400F).

.2.7. Glow-discharge optical emission spectroscopy (GDOES)The distribution of species in the anodic oxide layer was determined by

epth profiling using a Jobin Yvon GD Profiler instrument equipped with amm diameter anode and operating at pressure of 800 Pa and power 600 W

n an argon atmosphere. The relevant wavelengths (nm) were as follows: Al,96.15; O, 130.22; S, 181.73 and C, 156.14. The sputtering layer was 6 �mhick.

.3. Methodology of experimental design

.3.1. Doehlert experimental designThe Doehlert experimental design [22] was performed to establish the

ffect of the bath temperature, anodic current density and sulphuric acid con-entration, and their interactions on some properties of the aluminium oxide.

Ioitt

able 3oehlert experimental design in coded variables and the obtained responses Yi

umber of run X1 X2 X3

1 1.0000 0.0000 0.00002 −1.0000 0.0000 0.00003 0.5000 0.8660 0.00004 −0.5000 −0.8660 0.00005 0.5000 −0.8660 0.00006 −0.5000 0.8660 0.00007 0.5000 0.2887 0.81658 −0.5000 −0.2887 −0.81659 0.5000 −0.2887 −0.81650 0.0000 0.5774 −0.81651 −0.5000 0.2887 0.81652 0.0000 −0.5774 0.81653 0.0000 0.0000 0.00004 0.0000 0.0000 0.00005 0.0000 0.0000 0.00006 0.0000 0.0000 0.0000

2 (A dm−2) 7 2 1

3 (g L−1) 3 160 40

he choice of Doehlert design is justified by a number of advantages suchs: (i) its spherical experimental domain with an uniformity in space fill-ng, (ii) its ability to explore the whole of the domain and (iii) its potentialor sequentially where the experiments can be re-used when the boundariesave not been well chosen at first. It is important to point out that an impor-ant property of Doehlert design regards the number of levels that each factorakes.

The factors Uj selected were:

U1: the anodizing temperature (◦C),U2: the current density (A dm−2),U3: the sulphuric acid concentration (g L−1).

As currently used in experimental design, natural variables Uj were trans-ormed into coded variables Xj [17–22]. For the Doehlert design construction,ts centre and its variation step (Table 2) defined the study domain.

Four responses were studied: Y1: dissolution rate (g m−2 min−1), Y2: Vickersicrohardness (Hv), Y3: weight loss by abrasion (mg) and Y4: deflection at

ailure of the anodic oxide (mm). A full quadratic model with 10 coefficients,ncluding interaction terms, was assumed to describe the relationship betweenach response Yi and experimental factors Xj:

Yi = b0 + b1X1 + b2X2 + b3X3 + b11X21 + b22X

22 + b33X

23

+b12X1X2 + b13X1X3 + b23X2X3 + e

here b0 is the constant of the model, bj the first degree coefficients, bjk theross-products coefficients, bjj the quadratic coefficients and e is the randomxperimental error.

Doehlert design requires an experiment number according to N = k2 + k + N0,here k is the number of the factors and N0 the number of centre points.

n our case, the N value was fixed at 4 so with three factors the total

0

f points of Doehlert matrix was 16. All experiments were performedn randomized order to transform potential systematic errors of uncon-rolled factors into random errors and then to be able to apply statisticalests.

Y1 (g m−2 min−1) Y2 (Hv) Y3 (mg) Y4 (mm)

5.7 128 50.5 5.34.6 448 16.3 8.24.2 153 35.2 7.23.1 270 22.0 5.34.1 142 31.2 4.74.6 411 20.0 7.64.1 480 36.5 7.33.2 469 14.2 5.54.1 250 29.3 5.23.3 512 22.1 7.03.6 320 16.6 7.13.0 491 26.9 5.34.1 511 23.7 6.54.3 467 28.2 6.74.2 454 25.2 6.13.7 481 24.3 6.0

istry

ommpav

2

mttbacmc

Y

wc

ranip

2

rsc

aBssfi(ntraioho

t

3

TA

S

Y

Y

Y

Y

Tc

W. Bensalah et al. / Materials Chem

Replicates at the central level of the variables were carried out inrder to estimate the pure error variance. This ensures independent esti-ates of the model parameters. Using the F-test, a statistical test of theodel fit was made by comparing the variance due to lack of fit to the

ure error variance. The fitted model is considered adequate if the vari-nce due to the lack of fit is not significantly different from the pure errorariance.

.3.2. Canonical analysisThe response functions may be analyzed by canonical analysis [17–20], a

ethod of rewriting a fitted second-degree equation in a form nicely describeshe nature of the stationary point and the nature of the system around the sta-ionary point (in which it can be more readily understood). This is accomplishedy a rotation of axes that remove all cross-product terms. If desired, this may beccompanied by a change of origin to remove first-order terms. Using this newoordinate system, the second-order model equations are simplified and its geo-etrical nature becomes apparent. The equation developed by the transformation

alled the canonical form of the model is:

= Ys +j=3∑j=1

λjW2j

here Wj (j = 1, 2, 3) denotes the transformed independent variables or theanonical variables.

The λj is the eigen values which will describe the curvature of the

esponse. The constant Ys is the calculated response value at the station-ry point. The algebraic signs of the eigen values provide an idea about theature of its stationary point. If the values are all negative, it is a maximum;f all positive, it is a minimum, and if the signs are mixed it is a saddleoint.

3

r

able 4NOVA table for the responses Y1, Y2, Y3 and Y4

ources of variation Sum of squares Degrees of freedom

ˆ1

Regression 6.6969 9Residual 0.3125 6Lack of fit 0.1050 3Pure error 0.2075 3Total variation 7.0094 15

ˆ2

Regression 2.79804105 9Residual 1.75901104 6Lack of fit 1.57953104 3Pure error 1.79475103 3Total variation 2.97394105 15

ˆ3

Regression 1.20271103 9Residual 3.13275101 6Lack of fit 1.93575101 3Pure error 1.19700101 3Total variation 1.23404103 15

ˆ4

Regression 13.8377 9Residual 2.1398 6Lack of fit 1.8123 3Pure error 0.3275 3Total variation 15.9775 15

he significance in the table is the probability (between 0 and 1) of obtaining a rationventional manner: ***<0.001, **<0.01 and *<0.05 [27].

and Physics 108 (2008) 296–305 299

.3.3. Desirability functionAs one might expect, the optimum values (maxima or minima) for the

esponses do not occur at the same operating conditions but rather at widelyeparated points in the studied domain. This means that one must look for aompromise between conflicting criteria.

In this study, the desirability function approach, as proposed by Derringernd Suich [24], was employed to optimize, simultaneously, the four responses.riefly, the i measured responses, i = 1, 2, 3 and 4, are transformed to a dimen-

ionless desirability scale, di, defined as partial desirability function whosecale ranges between di = 0, for a completely undesired response, and di = 1or fully desired response above which further improvements would have nomportance. It is worth noting that each individual desirability function, di

i = 1–4) is a continuous function chosen from among a family of linear or expo-ential functions. Once the function di is defined for each of the responses,hey are combined into an overall desirability function, D, that weighs theesponses together, with one single criterion. A calculation algorithm is thenpplied to the D function in order to determine the set of factors that max-mizes it as close as possible to 100%. The values of D computed from thebserved responses allow locating a near-optimum region. The value of D isighest at conditions where combination of the different criteria is globallyptimal.

In this work, we used NEMROD W software [26] for data calculation andreatment.

. Results and discussion

.1. Response models and validation

Table 3 displays the experimental matrix and the measuredesponses. Fitted to 16 responses values, the second-order mod-

Mean square F-ratio Significance

0.7441 14.2861 0.211**0.05210.0350 0.5061 70.50.0692

3.10894104 10.6046 0.474**2.93168103

5.26510103 8.8008 5.45.98250102

1.33634102 25.5943 0.041***5.221266.45251 1.6172 35.13.99000

1.5375 4.3112 4.46*0.35660.6041 5.5338 9.70.1092

o of mean squares greater than F. The significance level is represented in the

3 emistr

e

− 0.

.9X22

22 −− 0.

Ass0fi

cieteosc

3

oicf

TfT

IY

sspcotft

s

Y

Tcsvbcdt

3

a

l

Y

Ts

3r

Y

TiOb

3r

00 W. Bensalah et al. / Materials Ch

ls are represented by the following equations:

Y1 = 4.08 + 0.52X1 + 0.42X2 + 0.02X3 + 1.07X21 − 0.46X2

2

Y2 = 478.2 − 135.6X1 + 41.8X2 + 12.2X3 − 190.3X21 − 248

Y3 = 25.35 + 15.98X1 + 0.43X2 + 2.94X3 + 8.05X21 − 0.35X

Y4 = 6.32 − 0.86X1 + 1.55X2 + 0.41X3 + 0.42X21 − 0.31X2

2

nalysis of variance (ANOVA) is performed on each responseeparately (Table 4). As can be seen the regression sum ofquares is statistically significant (their p-value are less than.05). On the other hand, none of the four models has a lack oft (p-value >0.05).

For the responses Y1, Y2, Y3 and Y4, the multiple correlationoefficient R2 are respectively 0.96, 0.94, 0.98 and 0.88. Thisndicates that the each regression model correlates well with thexperimental data. It should be noted that the multiple correla-ion coefficient is the proportion of the variability in the responsexplained by the deliberate variation of the factors in the coursef experiments. From overall results, we can conclude that eachecond-order model is adequate to describe each response andan be used as a prediction equation in the studied domain.

.2. Canonical analysis

The purpose of the following paragraph is to examine theverall shape of the response and the nature (maximum, min-mum or saddle) and uniqueness of the stationary point. Theanonical form of the four fitted models is represented by theollowing equations:

Y1 = 4.17 + 1.18W21 − 0.50W2

2 − 1.00W23

Y2 = 488.3 + 113.4W21 − 249.6W2

2 − 280.0W23

Y3 = 5.85 + 8.42W21 + 0.22W2

2 − 4.49W23

Y4 = 8.26 + 0.46W21 − 0.17W2

2 − 0.33W23

he mixed or the different eigen value signs indicate that theour responses at the stationary points are shaped like a saddle.he coordinates of the stationary point are

S1

⎛⎜⎝

−0.047

+0.530

+0.096

⎞⎟⎠ ; S2

⎛⎜⎝

−0.126

−0.026

+0.332

⎞⎟⎠ ; S3

⎛⎜⎝

−2.027

+6.387

−3.192

⎞⎟⎠ ;

S4

⎛⎜⎝

+0.391

+2.465

+0.926

⎞⎟⎠ .

t can easily seen that the stationary points corresponding toˆ1 and Y2 are inside the study domain whereas those corre-ponding to Y3 and Y4 are outside the study domain. Underuch circumstances, the response surface around the stationaryoint does not represent any real phenomenon. Accordingly, we

hoose to conduct the canonical analysis taking into accountnly a unique rotation of the co-ordinate system (withoutranslation), which removes the cross-product terms bjkXjXkrom the model while keeping the initial origin at the cen-re point. Zj was used to denote the axes of such rotated

Y

Fot

y and Physics 108 (2008) 296–305

94X23 − 0.81X1X2 + 0.04X1X3 + 0.31X2X3

+ 22.9X23 − 75.1X1X2 + 258.6X1X3 − 192.9X2X3

3.55X23 + 3.46X1X2 + 1.71X1X3 − 3.46X2X3

17X23 + 0.12X1X2 + 0.27X1X3 − 0.08X2X3

ystem. This yielded a canonical model of the form:

= Ys +j=4∑j=1

bjZj +j=4∑j=1

λjZ2j

he λj will describe the curvature of the response while the linearoefficient bj will describe the slope of the ridge in the corre-ponding direction. The constant Ys is the calculated responsealue at the stationary point. The interpretation would be easiery analysing each response along every Zj-axis separately. Theanonical analysis is only detailed for the first response i.e. theissolution rate Y1. For the three responses Y2, Y3 and Y4, onlyhe final results are given.

.2.1. Study of the dissolution rate (response Y1)The variable transformations:

X1 = 0.971Z1 + 0.230Z2 − 0.072Z3

X2 = −0.241Z1 + 0.915Z2 − 0.323Z3

X3 = −0.008Z2 + 0.331Z2 + 0.943Z3

nd

Z1 = 0.971X1 − 0.241X2 − 0.008X3

Z2 = 0.230X1 + 0.915X2 + 0.331X3

Z3 = −0.072X1 − 0.323X2 + 0.943X3

ead to the following canonical form of the model:

ˆ1 = 4.07 + 0.41Z1 + 0.51Z2 − 0.15Z3 + 1.18Z21

−0.50Z22 − 1.00Z2

3

hese data allow us to determine the features of the responseurface in each direction of the study domain (Fig. 4).

.2.1.1. Analysis along the OZ1 direction. The equation of Y1esponse is reduced to:

ˆ1 = 4.07 + 0.41Z1 + 1.18Z21

he corresponding curve is represented in Fig. 4(1). The min-mization of Y1 can be obtained for Z1 = −0.1. Because theZ1-axis is almost parallel to OX1-axis, Y1 should be optimizedy choosing X1 = −0.1.

.2.1.2. Analysis along the OZ2 direction:. The equation of Y1esponse is reduced to:

ˆ1 = 4.07 + 0.51Z2 − 0.50Z22

ig. 4(2) shows that the minimization of Y1 requires a low levelf Z2. According to the equations of the variable transformations,his can be achieved by choosing the low level (−1) for X2.

W. Bensalah et al. / Materials Chemistry and Physics 108 (2008) 296–305 301

3r

Y

FbOb

3

a

t

Y

Tis

•••

3

a

T

Y

The curvatures of the Y3 response along the OZj are given inFig. 6. As mentioned above, we can deduce that the response Y3should be minimized by:

Fig. 4. Curvature of Y1 response vs. Zj.

.2.1.3. Analysis along the OZ3 direction:. The equation of Y1esponse is reduced to

ˆ1 = 4.07 − 0.15Z3 − 1.00Z33

rom Fig. 4(3), one can see that that the minimization of Y1 cane obtained either at a high or a low level of Z3. Because theZ3-axis is almost parallel to OX3-axis, Y1 should be optimizedy choosing either a high or a low level of X3.

.2.2. Study of the Vickers microhardness (response Y2)Using the variable transformations:

X1 = 0.405Z1 + 0.909Z2 − 0.099Z3

X2 = −0.274Z1 + 0.224Z2 + 0.935Z3

X3 = 0.872Z1 − 0.352Z2 + 0.340Z3

nd

Z1 = 0.405X1 − 0.274X2 + 0.872X3

Z2 = 0.909X1 + 0.224X2 − 0.352X3

Z3 = −0.099X1 + 0.935X2 + 0.340X3

he canonical form of the model is represented by:

ˆ2 = 478.3 − 55.7Z1 − 118.2Z2 + 56.6Z3

+113.4Z21 − 249.6Z2

2 − 280.0Z23

he corresponding curves along the three axes are representedn Fig. 5. Following the same analysis as above, the response Y2hould be maximized by:

choosing a level for X1 between 0 and −0.1,choosing a level for X2 between 0 and 0.1,either increasing or decreasing X3: (+1) or (−1).

Fig. 5. Curvature of Y2 response vs. Zj.

.2.3. Study of the weight loss by abrasion (response Y3)By the transformations:

X1 = 0.982Z1 − 0.148Z2 − 0.120Z3

X2 = 0.185Z1 + 0.886Z2 + 0.426Z3

X3 = 0.043Z1 − 0.440Z2 + 0.897Z3

nd

Z1 = 0.982X1 + 0.185X2 + 0.043X3

Z2 = −0.148X1 + 0.886X2 − 0.440X3

Z3 = −0.120X1 + 0.426X2 + 0.897X3

he second-order model of Y3 becomes:

ˆ3 = 25.35 + 15.89Z1 − 3.27Z2 + 0.90Z3 + 8.42Z21

+0.22Z22 − 4.49Z2

3

Fig. 6. Curvature of Y3 response vs. Zj.

3 emistry and Physics 108 (2008) 296–305

•••

3(

a

t

Y

TFm

t

t

oombr

3

02 W. Bensalah et al. / Materials Ch

decreasing X1 (−1),increasing X2 (+1),either increasing or decreasing X3: (+1) or (−1).

.2.4. Study of the deflection at failure of the anodic oxideresponse Y4)

Taking into account the transformations:

X1 = 0.977Z1 − 0.169Z2 − 0.130Z3

X2 = 0.063Z1 − 0.357Z2 + 0.932Z3

X3 = 0.204Z1 + 0.919Z2 + 0.338Z3

nd

Z1 = 0.977X1 + 0.063X2 + 0.204X3

Z2 = −0.169X1 − 0.357X2 + 0.919X3

Z3 = −0.130X1 + 0.932X2 + 0.338X3

he expression of the model of Y4 is:

ˆ4 = 6.33 − 0.66Z1 − 0.03Z2 + 1.70Z3

+0.46Z21 − 0.17Z2

2 − 0.33Z23

he curvatures of the Y4 response along the OZj are given inig. 7.As above, we can deduce that the response Y4 should beaximized by fixing X1 at (−1), X2 at (+1) and X3 at (0).

Table 5 summarizes the experimental conditions, leading to

he optimization of the four separately taken responses.The examination of all the results obtained by means of

he canonical analysis, allows us to conclude that it is notuS

Fig. 8. Evolution of elementa

Fig. 7. Curvature of Y4 response vs. Zj.

bvious how one can find experimental conditions that canptimize the four responses simultaneously. Accordingly, aulticriteria methodology that looks for a certain compromise

etween experimental conditions fulfilling our expectations isequired.

.3. Optimization

The optimization of the four responses was performed bysing the desirability functions as proposed by Derringer anduich [26]. In our case, one-sized transformation was chosen

ry desirability di vs. Yi.

W. Bensalah et al. / Materials Chemistry and Physics 108 (2008) 296–305 303

Table 5Overall results of the canonical analysis

X1 X2 X3

Y1 (g m−2 min−1) −0.1 −1 (+1) or (−1)Y2 (Hv) 0–(−0.1) (0)–(0.1) (+1) or (−1)Y3 (mg) −1 +1 (+1) or (−1)Y4 (mm) −1 +1 0

Table 6Optimal conditions

Xj Uj

The anodizing temperature (◦C) −0.039671 13.6The current density (A dm−2) 0.217253 2.2T

ft

rsfriTt

e

aYcrtYw

Fo

acflnbvc

3o

ssmooth.

Fm

he sulphuric acid concentration (g L−1) 0.978730 199

or the four responses (Fig. 8). The undesirable responses andhe fully desired responses are also reported on Fig. 8.

Taking into account all the requirements for the fouresponses, we choose to compute an overall desirability mea-ure D as a weighted geometric mean of the desirability valuesor individual parameters. Equal weights were given for the fouresponses. Therefore, the function D, over the studied domain,s calculated by using the following equation: D = (d1d2d3d4)1/4.he value of D is highest at conditions where a combination of

he different criteria is globally optimal.After calculation by NEMROD W software, the optimal

xperimental conditions are obtained (Table 6).Under these conditions, the estimated response values

re 3.3 g m−2 min−1, 464 Hv, 23.4 mg and 6.9 mm for Y1,2, Y3 and Y4, respectively. In order to validate the cal-ulated optimal conditions, an additional experiment wasun with the levels of the optimum. The experimen-

−2 −1

al responses (Y1 = 3.1 ± 0.3 g m min , Y2 = 458 ± 24 Hv,3 = 22.5 ± 2.0 mg and Y4 = 6.7 ± 0.3 mm) are in agreementith the predicted ones.

id

ig. 9. Graphical representation of the overall desirability function D: (a) X2 is ploaintaining U1 at 13.6 ◦C.

ig. 10. AFM contact mode images of the anodic film surface obtained underptimal conditions.

The resulting contour plots from modelling the overall desir-bility function can be seen in Fig. 9. It can be noted that the areaorresponding to the optimal conditions (D = 1.00) was ratherat. This would imply that the values of the four responsesear this point are stable. Hence, the maximum can be said toe insensitive to small variations in the studied experimentalariables and represents the robustness of the predicted optimalonditions.

.4. Morphology and composition of the anodic layerbtained under the optimum conditions

Fig. 10 is the three-dimensional AFM image of the anodizedample. As can be seen, the surface is relatively flat and

The oxide layer structure revealed by SEM is illustratedn Fig. 11. The top surface exhibits nano-pores uniformlyistributed, together with some spherical shaped dots, of approx-

tted against X1 maintaining U3 at 199 g L−1 and (b) X3 is plotted against X2

304 W. Bensalah et al. / Materials Chemistr

Fc

itlTd

tpi[atThiim[

Fc

4

ialonbu(ld

A

Ng

R

[

[

ig. 11. Scanning micrograph of anodized surface obtained under the optimalonditions.

mately 20–30 nm in diameter, heterogeneously distributed inhe intervening areas. In addition, the pores are roughly circu-ar in section and gradually merge along domain boundaries.he pore spacing is relatively high compared to the poreiameter.

Fig. 12 shows a depth profile of the oxide layer. The dis-ribution of Al, O and S species are revealed clearly. Theresence of sulphur specie suggests that SO4

2− anions migratenward in the thickening film as reported in the literature28–30]. Its distribution is practically uniform through thenodic oxide coating whereas in the inner layer, adjacent tohe aluminium substrate, the sulphur concentration decreases.his phenomenon has been previously mentioned [29,30] andas been explained by the greater inward mobility of O2−ons relative to SO4

2− ions under the field during anodiz-ng. The presence of Al and O is explained by the outward

igration of Al3+ and the inward migration of O2− ions

29,30].

ig. 12. The GDOES depth profile of the anodic oxide obtained under optimalonditions.

[

[

[

[

[

[

[

[

[

[

[[

[

y and Physics 108 (2008) 296–305

. Conclusion

The achievement of a Doehlert design followed by canon-cal analysis and optimization by using desirability functionllows us to determine the best experimental conditions, whichead to a compromise between studied properties of the anodicxide layer on pure aluminium namely: Vickers microhard-ess, abrasion resistance, oxide film dissolution rate and flexureehaviour. Characterizations of the anodized sample obtainednder optimal conditions by SEM, AFM and GDOES reveal:i) the nano-porous structure of top surface of the anodic oxideayer (ii) and the “homogeneous” distribution of Al, O, S in theepth of the oxide layer.

cknowledgment

The authors would like to thank Mr. Ayadi Hajji (Ecoleationale d’Ingenieurs de Sfax-Tunisie) for his valuable lin-uistic assistance.

eferences

[1] Y.H. Choo, O.F. Devereux, J. Electrochem. Soc. 123 (12) (1976) 1868.[2] M. Maejima, K. Saruwatari, K. Isawa, Met. Finish. 10 (1998) 36.[3] M.B. Spoelstra, A.J. Bosch, D.H. Van der Weijde, J.H.W. de Wit, Mater.

Corros. 51 (2000) 155.[4] J. Rasmussen, Met. Finish. 9 (2001) 46.[5] K. Elleuch, S. Fouvry, Ph. Kapsa, Thin Solid Films 426 (2003) 271.[6] V. Moutarlier, S. Pelletier, F. Lallemand, M.P. Gigandet, Z. Mekhalif, Appl.

Surf. Sci. 173 (2003) 87.[7] X. Li, X. Nie, L. Wang, D.O. Northwood, Surf. Coat. Technol. 200 (2005)

1994.[8] S. Mezlini, K. Elleuch, S. Fouvry, Ph. Kapsa, Surf. Coat. Technol. 200

(2006) 2852.[9] G.E. Thompson, Thin Solid Films 297 (1997) 192.10] A.W. Brace, The Technology of Anodized Aluminum, Robert Droper,

Teddington, 1968, pp. 1–11.11] R.A. Wodehouse, in: A.K. Grahan (Ed.), Electroplating Engineering Hand-

book, 3rd ed., Nostrand Reinhold Company, 1971, p. 456.12] G.E. Thompson, G.C. Wood, in: J.C. Scully (Ed.), Anodic Films on Alu-

minum, Academic Press, London, 1983, p. 250.13] S. Wernick, R. Pinner, P. Sheasby, The Surface Treatment of Aluminum

and its Alloys, 5th ed., ASM International, Metals Park, Ohio, 1987, pp.444, 801–806.

14] M.B. Spoelstra, E.P.M. Van Westing, J.H.W. de Wit, Mater. Corros. 52(2001) 661.

15] L.E.F. Apachitei, F.D. Tichelaar, G.E. Thompson, H. Terryn, P. Skeldon, J.Duszczyk, L. Katgerman, Surf. Coat. Technol. 49 (2004) 3169.

16] J.P. O’Sullivan, G.C. Wood, Proc. Roy. Soc. Lond. A 317 (1970)511.

17] D. Mathieu, R. Phan-Tan-Luu, Plans d’Experiences: Application al’entreprise, Technip, Paris, 1995.

18] D.C. Montgomrey, Design and Analysis of Experiments, 3rd ed., Wiley,New York, 1991.

19] J. Goupy, Plans d’Experiences pour Surfaces de Reponses, Dunod, Paris,1999.

20] A.I. Khuri, J.A. Cornell, Response Surfaces: Design and Analyses, M.Dekker, New York, 1996.

21] A.L. Gareth, D. Mathieu, R. Phan-Tan-Luu, Pharmaceutical Experimental

Design, M. Dekker, New York, 1999.

22] D.H. Doehlert, Appl. Stat. 19 (1970) 231.23] L. Chalumeau, M. Wery, H.F. Ayedi, M.M. Chabouni, C. Leclere, J. Appl.

Electrochem. 34 (2004) 1177.24] G. Derringer, R. Suich, J. Qual. Technol. 12 (1980) 214.

istry

[[

[

W. Bensalah et al. / Materials Chem

25] E.C. Harrington Jr., Ind. Qual. Control 21 (1965) 494.26] D. Mathieu et, J. Noney et, R. Phan-Tan-Luu, NEMROD-W Software,

LPRAI, Marseille, 2002.27] R.A. Fisher, F. Yels, Statical Tables for Biologic and Agricultural and

Medical Research, Olivier and Boyd Ed, London, 1948.

[

[

[

and Physics 108 (2008) 296–305 305

28] G.C. Wood, P. Skeldon, G.E. Thompson, K. Shimizu, J. Electrochem. Soc.143 (1996) 74.

29] K. Shimizu, H. Habazaki, P. Skeldon, G.E. Thompson, G.C. Wood, Elec-trochim. Acta 4 (2000) 1805.

30] G.E. Thompson, G.C. Wood, Nature 290 (1981) 230.


Recommended