Investigations and optimization for hard milling ...

Rev. Téc. Ing. Univ. Zulia. Vol. 39, Nº 1, 41 - 54, 2016

Investigations and optimization for hard milling process parameters using hybrid method of RSM and NSGA-II *1

A.Tamilarasan, 2K.Marimuthu , 3A.Renugambal

*1

Assistant Professor, Department of Mechanical Engineering, Sri Chandrasekharendra Saraswathi Viswa Maha Vidyalaya University, Kanchipuram-631561,Tamilnadu,India. 2 Associate professor, Department of Mechanical Engineering, Coimbatore Institute of Technology, Civil Aerodrome post, Coimbatore-641 014, Tamilnadu, India. 3 Teaching Fellow, Department of Mathematics, University College of Engineering, Kanchipuram, Tamilnadu, India. *

Corresponding Author: E-Mail: *[email protected]

Abstract The present work investigates the effect and optimization of process parameters on the cutting temperature, tool wear and metal removal rate during hard milling of 100MnCrW4 (AISI O1) tool steel using (TiN/TiAlN) coated carbide tools. The central composite rotatable design is utilized to plan the experiments. The empirical models were developed and analysis of variance tests were used for the investigation of significant parameters and adequacy of models. Scars, adhered materials and coating peel off were observed on the rake face of the tool due to the existence tool flank wear through the SEM and EDX analysis. In order to seek optimal parameters, a Non-dominated Sorting Genetic Algorithm (NSGA-II) has been adopted and a set of Pareto-optimal solution set was obtained. Further, a set of confirmation experiments were conducted and the adopted optimization method was proved to be feasible. Also, the results would be more useful to guide the actual hard milling process parameters for predicting the responses. Keywords Hard milling, Central composite rotatable design, Cutting temperature, Tool wear, Metal removal rate, NSGA-II. 1. INTRODUCTION Hard milling is a viable green machining technique, which is more alternative to replace the grinding process at the semi-finish stage and also replacing the EDM process (Ding T et al., 2010 and Davim JP, 2011). The technology offers for significant savings in cost, good surface finish with large metal removal rate as compared with EDM (Gopalsamy BM et al., 2010). A successful implementation of hard milling describes that to boost the productivity with improved surface quality. On the other hand, it generates more fluctuated cutting forces and temperature as compared with conventional machining process. This causes severe tool wear in the cutting tool, then reduces the tool life and deteriorates machining quality. The tool wear is associated with cutting temperature and metal removal rate, being three important responses in hard milling to designate the required surface finish. Therefore, the influence of hard milling process parameters is still questionable to predict the quantitatively the technological performance of machining operations at economic level. Ding T et al. (2010) studied the effect of cutting speed, feed rate and depth of cut on the cutting forces, and surface roughness in hard milling of AISI H13 (50±1 HRC) steel with coated (TiN/TiAlN) carbide tool. Siller HR et al. (2009) demonstrated the special tool geometric features of coated carbide tools in face milling of hardened D3 tool steel (60HRC) components of dies and molds. The surface roughness was revealed in the range between 0.1- 0.3 µm with an acceptable level of tool life. Zhang S et al. (2012) carried out experimental investigations on hard milling of AISI H13 (50±1HRC) tool steel to determine the effects of surface texture, cutting parameters, phase transformation and in-depth residual stress distributions on the surface. It has been concluded that the surface texture of machined surface highly correlated induced residual surface stress was observed. The feed per tooth and radial depth of cut were more effect and axial depth of cut has least effect on the surface residual stresses. Iqbal A et al. (2007) estimated the flank wear, cutting forces and length of cut in hard milling of D2 tool steel using coated solid carbide end mills. The tool wear was predicted through developed fuzzy expert system. Caliskan H et al. (2012) evaluated the surface roughness and cutting forces through with deposited nano-layer AlTiN/TiN, commercial TiN/TiAlN and multilayer nano-composite TiAlSiN/TiSiN/TiAlN coated tools in hard milling of AISI O2 (∼61 HRC) cold work tool steel. A factorial design was used to investigate the effects of coating, cutting speed, feed rate and axial depth cut. It has been observed that the type of coating alone is not significant effect on both surface roughness and cutting forces. Besides, AlTiN/TiN coated tools, which yield a minimum surface roughness was observed. Cui X et al. (2012) studied the influence of cutting speed in the range between 200 to 1200 m/min in face milling of AISI H13 steel using CBN tools. The cutting speed and degree of chip segmentation were initially decreased and further increased with increase of cutting speed. The same Cui X et al. (2012) found notable tool wear mechanisms in high- and ultra-high-speed face milling of AISI H13 hardened steel through experiments and validated through 3D finite element simulations. Likewise, a 41


longer tool life with lowest averaged cutting force was obtained at the critical cutting speed of 1400m/min. From the literature, it is clear that there is a research gap in effect of process parameters on in the combination of Tool Wear(TW), Cutting Temperature(CT) associated with metal removal rate(MRR). In the present work, investigate the effects of process parameters on the selected responses (CT,TW and MRR) in hard milling of 100MnCrW4 (AISI O1) tool steel using TiN/TiAlN coated carbide inserts. A Central Composite Rotatable Design (CCRD) using Response Surface Methodology (RSM) is employed to develop mathematical models. In process optimization, a suitable technological guideline is required to select optimum cutting conditions. Therefore, a NSGA-II technique associated with the developed quadratic models of the responses is used to optimize the process parameters simultaneously. A set of confirmation experiments are conducted and evidenced that the required solution set is more suitable for predicting the responses. 2. METHODOLOGY 2.1 Central Composite Rotatable Design The RSM based CCRD design is used to analyze the influence of four selected process parameters, namely, the feed per tooth, axial depth of cut, radial depth of cut and cutting speed on the cutting temperature, tool wear and metal removal rate. The CCRD is selected because of its efficiency with respect to the number of runs required for fitting a second order response surface mode (Montgomery DC, 2005). The following second-order polynomial equation was fitted to the experimental data of each selected dependent variable: k

k

i 1

i 1

y  0  i X i    ii X i2    ij X i X j   i

(2.1)

j

Where, β0 represents the intercept and the parameters of βi, βii, and βij represent the regression coefficients of variables for linear, quadratic, and interaction regression terms, respectively.The  is the statistical experimental error. For statistical calculations, the variables Xi were coded as xi according to the following relationship:

xi 

2[2 X  ( X max  X min )] , X max  X min

(2.2)

Where xi is the coded value of a hard milling variable X, X is the value of any variable from Xmin to Xmax. The Xmin and Xmax are the lower and upper limits of the selected variable X. The CCRD for four independent selected process parameters at five coded levels (i.e. -2,-1, 0, +1, +2) is employed, and the total number of experiments is 30 (= 2k+ 2k + 6, where k is the number of independent variables). Sixteen factorial and eight axial experimental runs are augmented with six replication runs at the centre point to evaluate the pure error. The actual values of the process parameters with their coded terms for the developed experimental design are given in Table 1. The adequacy of the model was determined the analysis of variance (ANOVA). The statistical significance of the model and model variables was determined at the 5% probability level (p < 0.05). The significant quadratic model equations are used to build response surfaces in order to under the effect of process parameters. Table 1 Experimental design in coded and un-coded form of process parameters Process parameters (unit)

Actual symbol

Reference symbol

Feed per tooth(mm/z)

fz

Radial depth of cut(mm)

Levels -2

-1

0

+1

+2

A

0.05

0.1

0.15

0.2

0.25

ae

B

0.2

0.3

0.4

0.5

0.6

Axial depth of cut(mm)

ap

C

0.2

0.4

0.6

0.8

1.0

Cutting speed(m/min)

Vc

D

200

250

300

350

400

2.2 Non-dominated Sorting Genetic Algorithm-II A multi-objective evolutionary algorithm of NSGA-II was developed by Deb K et al., (2002). Further, Srinivas N and Deb K (1994) proposed a modified version of fast elitist multi-objective optimization algorithm. The algorithm generates a set of evenly distributed solutions using non-dominated sorting and a crowdedcomparison approach. Nowadays, NSGA-II widely used in various fields of manufacturing process (Kondayya D and Krishna AG 2010, Kumar K and Agarwal S 2012, Senthilkumar C et al., 2010) due to the effectiveness and reduced computational complexity etc. Algorithm process of NSGA-II flow chart is shown in Figure 1. The step-by-step procedure of algorithm as follows (Kumar K and Agarwal S 2012, Sheshadri A 2006): 42


1. 2. 3.

4. 5. 6. 7. 8.

Set the initial run parameters for the algorithm, viz. Population size (N), maximum number of generations (gmax), crossover probability (Pc), mutation probability (Pm; generation, g=0). Randomly create an initial population Pg of size N with a good coverage of the search space, and thereby have a diverse gene pool with potential to explore as much of the search space as possible. Evaluate the objective values and rank the population using the concept of domination. Each solution is assigned a fitness (or rank) equal to its non-domination level (1 is the best level, 2 is the next best level and so on). Perform the crowding sort procedure and include the most widely solutions by using crowding distance value. The child populations Qg is produced from the parent population Pg using binary selection, recombination and mutation operators. Then the two populations are combined together to produce R g(=Pg U Qg), which is of size 2N. After this the population Rg undergoes non-dominated sorting to achieve a global non-domination check. The new population Pg+1 are filled based on the ranking of the non-dominated fronts.

Fig.1 NSGA-II flow chart

9. Since the combined population is twice the size of the population size N, all the fronts are not allowed to be used. Therefore a crowding distance sorting is performed in descending order and the population is filled. Thus, for this new population Pg+1, the whole process is repeated. 10. Update the number of generations, g=g+1. 11. Repeat steps 3 to 10 until a stopping criterion is met. 3. MATERIAL, MACHINE TOOL AND MEASUREMENT The cutting experiments have been carried out a high precision CNC vertical machining centre (MAZAKNEXUS 510C-II) as shown in Figure 2. The workpiece material is hardened and tempered to hardness 50±1 HRC. The selected steels are more suitable for a wide variety of making moulds, dies and master tools and precision gauges (ASM Handbook 1991). The TiN+TiAlN (Recommended by Taegu Tec) coated WC inserts (TT9080) is used in the experiments. The coated inserts are a superior wear resistance, lubricate, and a thermal barrier, while the substrate provides mechanical strength and fracture resistance (Krar SF and Gill A 2003, Noordin MY et al., 2007). The insert was a screw clamped to tool holder (Type 2S-TE90AP 220-W16-09) with a diameter of 20 mm. In order to prevent tool vibration, the tool run-out and tool holder projection length are kept at 0.02 mm and 50 mm respectively. Each experiment is conducted in three times and random fashion followed as per the experimental table to minimize the effects of unexplained variability in the observed responses. Cutting test is started with a fresh, cutting tool, and the machining process was stopped at length of 300 mm. A cutting zone with measurement of each response is depicted in Figure 3. A non-contact (fluke type 8839) pyrometer was used to capture the tool-chip interface temperature (Cydas U 2010). The measuring range of a parameter is -50 to 1000°C with an accuracy of ±2°C. 43


Fig.2 MAZAK VMC

Fig.3 Hard milling cutting zone and measurement of responses Each cutting temperature is calculated for a maximum of fifteen peak points in each run and three replicates are considered. The tool flank wear (ISO 8688-2, 1989) is obtained by the average value of three points measured on the cutting edge within an axial depth of cut through an optical microscope with high resolution. Further, the tools are examined using Scanning Electron Microscopy (SEM) (JSM-6510LV, Japan) an energy-dispersive Xray spectroscopy (EDX) in order to observe wear mechanisms in the tool. The material removal rate is calculated by weight loss method (Kondayya D and Krishna AG 2012). For each experiment the machining time is noted. The weight difference of the specimens before and after machining is calculated using a Sartorius precision digital balance with 0.001 gram accuracy. The experimental results are presented in Table 2. 4. RESULT AND DISCUSSIONS 4.1 Fitted Regression Models The measured responses are analyzed using statistical design-expert software for obtaining the mathematical models to best fit at 95% of confidence intervals. Then, the backward elimination method is applied to eliminate

44


the insignificant model terms. The final quadratic models from CT, TW and MRR in terms of actual values of process parameters as follows: CT (°C) =-1207.1727+1996.746429×A+55.7226×B+529.527×C+6.51711×D-849.125×AC+2.2335×BD0.897375×CD-3289.571429×A2-744.2678571×B2-0.00739×D2 (4.1) TW (mm) =+0.06604+0.1275×A-0.635833×B-0.21645×C+0.0010575×D-0.175×AC0.0001625×CD+2.3583×A2+0.8395×B2+0.266145833×C2-0.000001241×D2

(4.2)

MRR (g/min) =-26.331+25.3575× A+1. 5279×B-5.3404×C + 0.14648× D + 41.675 ×AC +15.19375×BC77.725×A2-0.000213475×D2 (4.3) The terms associated with the positive sign co-efficient indicates a synergistic effect and negative sign coefficient suggested antagonistic effect. In order to ensure statistical validation of the developed models, normality of residual data, presence of outliers, pattern of error variance and amount of residuals in prediction are checked. The Figure 4 depicts the each response of the normal probability plot of the residuals. It is noticed that no sign of the violation since each point in the plot follows a straight line pattern implying that the residuals are normally distributed. The result supported that the experimental points are reasonably aligned with predicted or fitted points, suggesting the normality of the data. 4.2 Analysis of variance The relative importance of the hard milling process parameters with respect to the CT, TW and MRR is investigated more accurately using ANOVA. The developed mathematical models are tested at 0.05 level of significance. The ANOVA Table 3 summarizes the sum of squares of residuals and regressions together with the corresponding degrees of freedom, F-value and ANOVA coefficients (i.e. coefficients of multiple determination R2, Adjusted R2 and Predicted R2 statistic). From the Table 3, it is noticed that the CT, TW and MRR model Fvalues are 180.45, 755.877 and 84.11 respectively, and thus the model value of each response is significant. There is only a 0.01% chance that a “model F value” this large could occur due to noise. The lack of fit F-value of 4.48 for CT, of 0.593 for TW and of 0.83 for MRR implies the lack of fit is not significant relative to the pure error. There is a 5.37% of CT, 79.68% of TW and 64.49 % of MRR chance that a lack of fit F-value this large could occur due to noise. The lack-of-fit term is not significant means that the model was more significant for the each response. The Pred.R2 and is in reasonable agreement with the Adj.R2 for each response. Therefore, the developed models can be used to predict the respective responses within the limits of the selected experimental domain. Design-Expert® Software CT

Design-Expert® Software Normal Plot of Residuals TW

Color points by value of CT: 719.09

Color points by value of TW: 0.243

99

Normal Plot of Residuals 99

0.093

Normal % Probability

95 90 80 70 50 30 20 10

95 90 80 70 50 30 20 10 5

5

1

1

-1 .4 8

-2 .0 9

-1 .0 5

Design-Expert® Software Int ernally MRR

-0 .0 1

1 .0 3

-0 .6 5

0 .1 8

1 .0 1

1 .8 5

2 .0 7

Int ernally St udent ized Residuals St udent ized Residuals Normal Plot of Residuals

Color points by value of MRR: 9.64

99

1.088



244.02

95 90 80 70 50 30 20 10 5 1

-1 .6 5

-0 .8 0

0 .0 5

0 .9 0

1 .7 5

Int ernally St udent ized Residuals

Fig.4 Normal probability plots 45


Table 2 CCRD Design matrix with Responses Process Parameters Exp.no

Responses

A

B

C

D

CT

TW

MRR

mm/z

mm

mm

m/min

°C

mm

g/min

1

1

-1

-1

-1

407.01

0.183

2.282

2

0

0

2

0

591.67

0.199

8.424

3

1

1

-1

1

693.78

0.213

6.175

4

-1

-1

-1

1

575.26

0.135

2.696

5

0

0

0

0

574.94

0.143

4.664

6

0

0

0

0

559.81

0.142

4.930

7

1

-1

-1

1

650.84

0.209

3.885

8

0

0

-2

0

491.55

0.175

2.178

9

1

1

1

-1

460.71

0.201

7.981

10

1

1

1

1

705.47

0.219

9.640

11

-1

1

-1

1

609.02

0.139

3.237

12

-1

-1

-1

-1

334.09

0.107

1.196

13

1

-1

1

1

673.63

0.210

7.770

14

0

0

0

-2

244.02

0.112

1.088

15

1

1

-1

-1

378.68

0.191

4.055

16

0

0

0

0

567.64

0.148

5.865

17

-1

1

1

1

696.61

0.153

6.475

18

0

0

0

0

552.28

0.145

5.554

19

-1

1

-1

-1

349.57

0.113

2.027

20

0

-2

0

0

505.2

0.169

2.674

21

2

0

0

0

575.46

0.243

7.348

22

-1

1

1

-1

438.82

0.129

4.355

23

0

0

0

0

560.73

0.148

5.038

24

1

-1

1

-1

441.52

0.195

4.866

25

-2

0

0

0

469.66

0.093

1.696

26

0

0

0

2

719.09

0.152

5.241

27

0

0

0

0

568.37

0.142

5.839

28

-1

-1

1

-1

419.04

0.125

2.433

29

0

2

0

0

546.17

0.187

7.581

30

-1

-1

1

1

600.35

0.144

3.085

46

Rev. Téc. Ing. Univ. Zulia. Vol. 39, Nº 1, 41 - 54, 2016 Des ign-Expert® Software CT 719.09 244.02 X1 = B: Radial depth of cut X2 = D: Cutting s peed Actual Factors A: Feed per tooth = 0.15 C: Axial depth of cut = 0.60

720 600

CT

480 360 240

350

0.50 325

0.45 300

0.40

D: Cutting speed

275

0.35B:

Radial depth of cut

250 0.30

Fig.5 Response surface plot for CT

Table 3 ANOVA Results Sum of squares

DF

CT(°C) Model Total Residual Lack of fit Pure error

408300 412600 4298.96 3981.33 7514.5965

TW(mm) Model Total Residual Lack of fit Pure error MRR(g/min) Model Total Residual Lack of fit Pure error

Source

Mean square

F

Prob.>F

R2

Adj-R2

Pre.R2

10 29 19 14 5

40828.21

180.45

< 0.0001

0.9895

0.9840

0.9728

226.26 284.38 1502.919

4.48

0.0537

0.04167 0.04178 0.0001 6.5E-05 3.9E-05

10 29 19 14 5

0.00417

755.877

< 0.0001

0.9974

0.9961

0.9933

5.5E-06 4.7E-06 7.9E-06

0.593

0.7968

151.15 155.86 4.72 3.43 1.28

8 29 21 16 5

18.89

84.11

< 0.0001

0.9697

0.9582

0.9399

0.22 0.21 0.26

0.83

0.6449

47

(Radulescu&Kapoor 1994). Figure 7.19 shows that the suggested minimum

Rev. Téc.speed Ing. Univ. Vol. 39,depth Nº 1,of41 - 54, 2016 cutting temperature is noted at 250 m/min of cutting and 0.4Zulia. mm of radial cut.

(a)

(b)

(d)

(c) Coating delamination

(c)

Flank wear zone Trace of landed chip surface

Fig.6Figure SEM7.20 images on toolonrake faceface (a) (a) Before SEM images tool rake Beforemachining machining (b) Magnified view of rake face (c) Magnified of rake face aftermachining machining (d) viewview of of rake (b) Magnified view of rake face (c) Magnified view view of rake face after (d)Magnified Magnified rake face (e) Magnified view near worn cutting edge face and (e) Magnified view near worn-out cutting edge 4.3 Effect of process parameters on cutting temperature Figure 5 illustrates a response surface plot showing the interaction effect of radial depth of cut and cutting speed on cutting temperature with other two process parameters at the middle level. It is believed that a noticeable amount of cutting temperature increases when the increase of cutting speed with any values of radial depth of cut from 0.3 to 0.5 mm. This is due to the fact that the induced strain rate becomes high in the shear zone beginning with chip load; thus more heat energy would be generated resulting in a high cutting temperature (NG EG and Aspinwall DK 2002). Therefore the increase of cutting temperature in the cutting zone leads to softening of the tool. This reduces the hardness of the tool. Likewise, the hardness of the cutting edge decreases as the temperature increases with the speed (Gu et al., 1999). Due to the thermal softening at the cutting edge, the tool exceeds the maximum working temperature of the coated layer. Further, the increase of the temperature causes the protective layer of coating to get detached (delamination) from the tool rake face as shown in Figure 6 (d) (Ding T et al., 2010). Figures.6 (a) - (e) provide different forms of tool edge and surface before and after machining. Accordingly, Figures. 6 (a) and (b) showed the tool insert of the rake face before machining. Figures. 6 (c), (d) and (e) represent the rake face of the tool after machining. From Figure.3, it is observed that the chip flew away carried with very high temperature. In the same way as shown in Figure.6 (e), it is evident that some of the chips landed in the tool rake face because of interaction between the tool and chip with high cutting temperature. Consequently, the effect accelerates the tool wear and shortens the tool life. This fact is in agreement with the results for CBN tool with 1200m/min (Cui X et al., 2012). In the present study of experimental results for the performance of TiN/TiAlN coated carbide tool, which exhibits the same effect at 350 m/ min of cutting speed. Therefore, reducing the cutting speed may produce less cutting temperatures and as improve the machinability indices. Hence, the hard milling process parameters must be selected in such a way that the predicted temperature does not exceed the softening temperature of the coated tool. It is possible to state that the tool-chip interface temperature is directly proportional to the cutting speed followed by the radial depth of cut (Radulescu R and Kapoor SG 1994). Besides, Figure.5 shows that the suggested minimum cutting temperature is observed at 250 m/min of cutting speed and 0.4 mm of radial depth of cut.

48


4.4 Effect of process parameters on tool wear The interaction between axial depth of cut and cutting speed on tool wear as illustrated in 3D response surface plot is shown in Figure.7. It is observed that the tool wear decreases with increase of axial depth of cut up to 0.6 mm for any value of cutting speed. The main reason is the coating materials act as a thermal barrier and provide less co-efficient of friction in the tool-workpiece and tool-chip interface zones. This possibly reduces the tool wear up to 0.6 mm of axial depth of cut. When increase axial depth of cut from 0.6 to 0.8 mm, the tool wear is tremendously increased. Because of contact area increases with increase of axial depth of cut. This leads to higher magnitude of normal cutting forces and temperatures, which in turn promotes severe tool wear. When the cutting speed increased from 250 to 350 m/min, the revolutions per second increased subsequently. This shows that the contact time between the tool and workpiece is amplified at higher cutting speeds. Due to this, the friction force in the tool-workpiece interface and the contact area of the wear land are increasing, which result in a significant increase in the cutting force. It implies the heat source at the tool-workpiece contact area and consequently triggers tool wear for all axial depth of cut. Figures .8 (a)-(c) demonstrate the SEM images and EDX analyzes of cutting tool for 7th the experimental run (Table 2). The SEM image on Figure.8 (a) clearly depicts the combined effect of cutting temperature and high-frequency impacts on the tool edge with variable chip thickness, which produces small scars near the cutting edge during chip formation. Further, failure of the tool edge started with the propagation of flank wear due to fatigue loading of the cutting edge at higher speed. In addition, hard machining can facilitate the material plastification under high shear strength of workpiece hardness (50HRC), which in turn produces extremely high mechanical and thermal loads along the arc of cut. Due to the effect, plastically deformed work material adheres to the tool surface. This is the experimental evidence that highly heated red chips were produced in the cutting zone to propagate the tool flank wear. Furthermore, there are two points A (unworn location) and B (worn out location) are considered for the composition of the coated layer. As shown in Figure.8 (b) the composition of titanium, aluminum and nitride were noticed on Point A. This indicates PVD coatings of TiN/TiAlN tool insert. The EDX analysis of point B (zone in contact with hot chips) is shown in Figure.8 (c). It clearly traces the content of element O and Fe is 32.50 wt. % and 3.57 wt. %, respectively (Gopalsamy BM et al., 2010). This indicates that the influence of adhesion and oxidation occurs during the hard milling process. These findings demonstrate the influence of tool wear by the possible existence of different tool wear mechanisms. Des ign-Expert® Software TW 0.243 0.093 X1 = C: Axial depth of cut X2 = D: Cutting s peed Actual Factors A: Feed per tooth = 0.15 B: Radial depth of cut = 0.40

0.199 0.17725

TW

0.1555 0.13375 0.112

350

0.80 325

0.70 300

D: Cutting speed

0.60 0.50 C: Axial

275

depth of cut

250 0.40

Fig.7 Response surface plot for TW 49


(a) Some adhered materials

Scars Rake face

B

Zone in contact with hot chips

A A

(b)

B

Fig.8

(c)

(a) SEM image on tool rake face near cutting edge (b) EDX analysis on unworn location at A, and (c) EDX analysis on worn-out location at B

Des ign-Expert® Software M RR 9.64 1.088

X1 = B: Radial depth of cut X2 = C: Axial depth of cut Actual Factors A: Feed per tooth = 0.15 D: Cutting s peed = 300

8.5

MRR

6.9 5.3 3.7 2.1

0.80

0.50 0.70

0.45 0.60

0.40

C: Axial depth of cut 0.50

0.35B:

Radial depth of cut

0.40 0.30

Fig.9 Response surface plot for MRR 50


4.5 Effect of process parameters on metal removal rate Figure.9 shows the effect of significant process parameters (axial depth of cut and radial depth of cut) on metal removal rate when the feed per tooth (0.2 mm) and cutting speed (250 m/min) is constant level. It can be observed from the figure that the both axial depth of cut and radial depth of cut influence the metal removal rate i.e. the value of the metal removal rate increases with all values axial depth of cut and radial depth of cut range. The combination of 0.8 mm axial depth of cut and 0.5 mm radial depth of cut produces a higher metal removal rate of 8.4 g/min. Therefore, the higher values of axial depth of cut and radial depth of cut boost the productivity in the hard milling operation (Kondayya D and Krishna AG 2012). On the contrary, the increase of axial depth of cut and radial depth of cut induces high cutting forces, tool wear and cutting temperatures in the cutting zone. 4.6 Multi-Objective Optimization In multi-objective optimization, the mathematical models of all responses are considered as each objective function and optimized using NSGA-II approach. The present goal of the study is simultaneously to minimize both the cutting temperature (CT), tool wear (TW) and maximize metal removal rate (MRR) for increase the production efficiency. Since the objectives are conflicting in nature are observed. Therefore, a third objective function is modified to get into negative for minimization. The formulated objectives for optimization are as follows: Objective 1 = CT Objective 2 = TW Objective 3 = - (MRR) TheNSGA-II algorithm code is implemented in MATLAB V10a and the range of lower and upper bound values are set by LB= [0.05 0.2 0.2 200];UB= [0.25 0.6 1.0 400].The control parameters are adjusted to obtain the best improved solutions. Finally, the following parameters have been selected to obtain the best Pareto optimal solutions after several simulations. The parameters are : pz (population size)=100, pc( Probability of crossover) = 0.9, pm( mutation probability)=0.25. Number of generations= 1000(to yield better convergence). The formation of the Pareto front leading to the final set of solutions was depicted in three dimensional plot as illustrated Figure.10. A set of 25 solutions with corresponding cutting conditions are depicted in Table 4. The choice of important response (CT or TW or MRR or combined all) within the experimental range is very important because the selected process parameters are determined to improve production efficiency. Depending upon the requirements, the process engineer can select any of optimal solutions of cutting conditions for the Pareto optimal solution set. Because, none of the solutions were in the non-dominated set is better than any other solution in the set as observed from the Table 4. A little consideration showed that, for boosting the productivity at higher MRR is more desired one. The suitable solutions are 5, 10, 17, and 21 will be selected, while the cutting temperature and tool wear values will be compromised. While, the solutions of 13, 25 and 4, 23 were noted for to obtain minimum cutting temperature and tool wear respectively. Similarly, the other responses are compromised with consideration of one response at a time. The remaining solutions are treated as to obtain balanced higher MRR is associated with minimum cutting temperature and tool wear during the process.

Fig.10 Simulated results

51


In addition, the predicted result of 13th solution number (Table 4) is closely matched with a 1 st experimental run (Table 2). It is observed that, the slight increase of cutting temperature from 334.09°C to 335.982°C, results to decrease the MRR from 1.196 g/min to 0.576 g/min with the same tool wear of 0.107mm. This also confirms the obtained solutions are good for predicting the responses. Furthermore, two more examples were addressed for improved performance characteristics of the hard milling process. For experimental run 18 (Table 2) produces the CT, TW and MRR as 552.28°C, 0.145mm and 5.554 g/min respectively. Correspondingly, solution number 8 (Table 4) indicates that the same value of 0.145mm of TW results, the reduced CT value of 552.28°C to 543.013°C and increased MRR value of 5.557g/min to 5.837 g/min were observed. On the other hand, 1.6779%of CT decreased and 4.8483% of MRR increased after the optimization. By choosing the experimental run 12 (Table 2), which yields the CT, TW and MRR as 335.982°C, 0.107mm and 0.576 g/min respectively. It can be seen that the CT is reduced to 0.5631% and MRR is raised to 51.839% for the same TW of 0.107 mm as suggested in solution number 13 (Table 4). Finally, first three solutions of cutting conditions were selected for experimental work to verify the surface quality of the workpiece from the Table 4. The Fig.11 illustrates the measured the surface roughness of milled surfaces. Using mitutoyo (SJ 301 type) surface roughness tester, the measured Ra value 0.19µm,0.13µm, and 0.23µm are obtained corresponding to the solutions number of 1, 2 and 3 (Table 4) respectively. These results were reasonable agreement with the grinding surface finish (ASM Hand book 1978).This confirms, the hard milling process definitely may replace the grinding process, at least in the semi finishing of the component (Ding T et al., 2010). Table 4 Optimal combination of parameters No 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

fz 0.13479 0.11123 0.14076 0.10124 0.19797 0.11992 0.11459 0.14627 0.14895 0.19852 0.15522 0.14289 0.10041 0.13850 0.10200 0.12242 0.19999 0.17394 0.19347 0.10848 0.19772 0.10443 0.10010 0.18744 0.10041

ae 0.42011 0.41387 0.41053 0.35305 0.49724 0.42906 0.41147 0.43936 0.43696 0.46065 0.44943 0.43162 0.30073 0.42467 0.41904 0.41366 0.49978 0.4727 0.44141 0.39837 0.5000 0.30552 0.37782 0.44109 0.30073

ap 0.60922 0.56602 0.59474 0.50517 0.79208 0.5948 0.57822 0.66503 0.68368 0.77465 0.68716 0.63654 0.40527 0.62486 0.54729 0.59109 0.79778 0.72909 0.76711 0.55144 0.79998 0.42698 0.51689 0.76152 0.40527

Vc 281.614066 261.58703 288.221199 250.471418 341.924932 267.326856 264.35519 289.299626 303.904985 336.128113 298.386793 292.144425 250.617557 286.438017 254.934924 271.882727 343.822624 319.911934 337.978615 260.085218 343.821701 255.303293 250.609444 328.888594 250.617557

CT 506.6409 424.1113 524.9164 364.9237 690.236 453.8308 437.4465 543.0126 583.2536 674.7424 573.6914 544.1726 335.9822 524.6004 391.881 467.8161 694.2746 636.0650 674.8046 413.8378 694.3826 359.8122 368.6801 656.6379 335.9822

TW 0.13113 0.1094 0.13574 0.09933 0.21563 0.11805 0.11232 0.14535 0.15136 0.20754 0.15642 0.14128 0.10675 0.13580 0.10228 0.11964 0.21883 0.18141 0.19996 0.10642 0.21703 0.10882 0.09821 0.19301 0.10675

MRR -4.69404 -3.02927 -4.83204 -1.41782 -9.8178 -3.82271 -3.27451 -5.83723 -6.31985 -9.15384 -6.58684 -5.50310 -0.57578 -5.08212 -2.44326 -3.8589 -9.96833 -8.09026 -8.69435 -2.65577 -9.93169 -0.97803 -1.67001 -8.43427 -0.57578

Fig.11 Measured surface roughness for a Pareto-optimal solution set number one

52


4.7 Confirmation Experiments Table 5 Confirmation Experiments and Results Experiment fz

ae

ap

Vc

0.1012

0.353

0.5052

0.1552

0.4494

0.2000 0.1004

Predicted

Relative error(%)

CT

TW

MRR

CT

TW

MRR

CT

TW

MRR

250.47

363.46

0.096

1.387

364.924

0.0993

1.418

0.401

3.351

2.173

0.6872

298.39

571.43

0.149

6.823

573.691

0.1564

6.587

0.394

4.743

-3.59

0.4998

0.7978

343.82

699.05

0.228

9.734

694.275

0.2188

9.968

-0.69

-4.19

2.351

0.3007

0.4053

250.62

332.34

0.109

0.563

335.982

0.1067

0.576

1.084

-2.11

2.22

In order to validate the optimization results, a set of confirmation experiments are carried out. From a Table 4, the solution numbers of 4, 11, 17 and 25 are randomly preferred. The experiments are continued in the highly configured same CNC machining centre in thrice times to ensure the most reliable results. The average values are taken to compare with the Pareto optimal solutions. Table 5 illustrates the results of confirmatory experiments with Pareto optimal solutions. The relative error percentage between the experimental results and predicted values is less than ±5%. This confirms the reproducibility of the results. 5 CONCLUSIONS The experimental observations of the cutting temperature, tool wear associated with the metal removal rate of the hard milled part of tool steel using coated (TiN/TiAlN) carbide inserts were studied and optimized. The outcomes of the investigation could be summarized as follows: 1. A central composite rotatable design effectively used for the experiment. 2. The predicted values match the experimental values reasonably well with R2 of 0.9895 for CT, R2 of 0.9974 for TW, and R2 of 0.9697 for MRR. 3. The cutting temperature is more influenced by cutting speed and feed per tooth and a coating peel off over the tool rake face observed when cutting at higher cutting speed. 4. Due to tribological improvement of (TiN/TiAlN) coating acts as a thermal barrier and lubricity layer up to 0.6mm of axial depth of cut to prevent tool wear. Further, increase of cutting temperature can promote the tool wear rapidly. 5. The examination of worn out inserts through SEM reveals that scars and adhered material on the rake face of the tool. 6. The metal removal rate monotonically increases with all axial depths of cut and radial depths of cut. 7. Using NSGA-II, and a Pareto-optimal solution set is obtained. The solution set helps the process engineer to select the cutting conditions as per the requirement of the product. 8. The first three sets of Pareto solutions give 0.19µm, 0.13µm, and 0.23µm of surface roughness value, which confirms the hard milling definitely to replace the grinding process at least in semi-finish stage.

References ASM Handbook (1978). Machining, vol. 16, ASM International, USA. ASM Handbook (1991). Heat Treating, vol. 4, ASM International, USA. Calişkan H, Kurbanoğlu C, Panjan P, Kramar D. Investigation of the performance of carbide cutting tools with hard coatings in hard milling based on the response surface methodology. Int. J. Adv. Manuf. Technol. 2012; 66 (5-8) 883-893. Cui X, Zhao J, Tian X. Cutting forces, chip formation, and tool wear in high-speed face milling of AISI H13 steel with CBN tools. Int. J. Adv. Manuf. Technol. 2012;64 (9-12) 1737-1749. Cui X, Zhao J, Tian X. Tool wear in high-speed face milling of AISI H13 steel. P I Mech Eng B-J EngManuf.2012;226(10) 1684-1693. Cydas U. Machinability evaluation in hard turning of AISI 4340 steel with different cutting tools using statistical techniques.P I MechEng B-J Eng. 2010; 224(7) 1043-1055. Davim JP. Machining of Hard Materials. Springer-Verlag;2011.

53


Deb K, Pratap A, Agarwal S. A fast and elitist multi-objective genetic algorithm: NSGA-II. IEEE I Evolut Comput. 2002;6 (2) 182-197. Ding T, Zhang S, Wang Y, Zhu X. Empirical models and optimal cutting parameters for cutting forces and surface roughness in hard milling of AISI H13 steel. Int. J. Adv. Manuf. Technol. 2010; 51(1-4) 45-55. Gu J, Barber G, Tung S, Gu RJ. Tool life and wear mechanism of uncoated and coated milling inserts. Wear. 1999; 225-229 273-284. Gopalsamy BM, Mondal B, Ghosh S, Arntz K, Klocke F. Experimental investigations while hard machining of DIEVAR tool steel (50 HRC). Int. J. Adv. Manuf. Technol. 2010; 51(9-12) 853-869. ISO 8688-2, 1989. Tool Life Testing in Milling-Part 2: End-milling. International Organization for Standardization, Geneva, Switzerland. Iqbal A, He N, Dar NU, Li L. Comparison of fuzzy expert system based strategies of offline and online estimation of flank wear in hard milling process. Expert Syst Appl. 2007: 33(1) 61-66. Kondayya D, Krishna AG. An integrated evolutionary approach for modelling and optimisation of CNC end milling process, International Journal of Computer Integrated Manufacturing. 2012;25(11) 1069-1084. Kondayya D, Krishna AG. An integrated evolutionary approach for modelling and optimization of laser beam cutting process. Int. J. Adv. Manuf. Technol. 2010;65(1-4) 259-274. Krar SF,Gill A. Exploring Advanced Manufacturing Technologies. Industrial Press Inc; 2003. Kumar K, Agarwal S. Multi-objective parametric optimization on machining with wire electric discharge machining. Int. J. Adv. Manuf. Technol. 2012; 62(58) 617-633. Montgomery DC. Design and analysis of experiments. John Wiley and Sons; 2012. Ng EG, Aspinwall DK. The effect of workpiece hardness and cutting speed on the machinability of AISI H13 Hot work die steel when using PCBN Tooling. J Manuf Sci E-T ASME. 2002; 124 588-594. Noordin NG,Yang YC, Kurniawan D. The use of TiAlN coated carbide tool when finish machining hardened stainless steel. International Journal of Precision Technology. 2007; 1(1) 21-29. Radulescu, R, Kapoor SG. An analytical model for prediction of tool temperature fields during continuous and interrupted cutting. Journal of Engineering Industry. 1994; 116 (2) 135-143. Senthilkumar C, Ganesan G, Karthikeyan R. Bi-performance optimization of electrochemical machining characteristics of Al/20%SiCp composites using NSGA-II. P I MechEng B-J Eng Manuf. 2010; 224 13991407 Siller HR, Vila C, Rodríguez CA, Abellan JV. Study of face milling of hardened AISI D3 steel with a special design of carbide tools. Int.J. Adv. Manuf. Technol. 2009; 40 (1-2) 12-25. Srinivas, N, Deb K. Multi objective optimization using non-dominated sorting algorithm. Evol Comput. 1994; 2(3) 241–248. Sheshadri A. A multi objective optimization algorithm, Matlab Central, file exchange, Mathworks, http:// www.mathworks.com/mat lab central (2006, accessed 12 July 2011). Zhang S, Ding TC, Li JF. Determination of surface and in-depth residual stress distributions induced by hard milling of H13 steel. Prod. Eng. Res. Devel. 2012;6(4-5)375-383.

54