1007/s12541-011-0050-7
The aim of this research is to develop an integrated study of surface roughness to model and optimize the cutting parameters when end milling of 6061 aluminum alloy with HSS and carbide tools under dry and wet conditions. A multiple regression analysis using analysis of variance is conducted to determine the performance of experimental measurements and to show the effect of cutting parameters on the surface roughness. The second-order mathematical models in terms of machining parameters have been developed for each of these conditions on the basis of experimental results. Genetic algorithm (GA) supported with the regression equation is utilized to determine the best combinations of cutting parameters providing roughness to the lower surface through optimization process. The value obtained from GA is compared with that of experimental value and found reliable. It is observed from the results that the developed study can be applied to other machining processes operating under different machining conditions.
Manuscript received: July 10, 2010 / Accepted: January 26, 2011
NOMENCLATURE
N = spindle speed d = depth of cut f = feed rate Ra = surface roughness
1. Introduction
Surface roughness is an important measure of product quality since it greatly influences the performance of mechanical parts as well as production cost. Surface roughness has an impact on the mechanical properties like fatigue behavior, corrosion resistance, etc. and functional attributes like friction, wear, light reflection, heat transmission and electrical conductivity, etc. There have been many research developments in modeling surface roughness and optimization of the controlling parameters to obtain a surface finish of desired level since only proper selection of cutting parameters can produce a better surface finish. In the manufacturing industries, various machining processes are adapted for removing the material from the workpiece for a better product. Out of these, end milling process is one of the most vital and common metal cutting KSPE and Springer 2011
operations used for machining parts because of its ability to remove materials faster with a reasonably good surface quality. In recent times, computer numerically controlled (CNC) machine tools have been implemented to utilize full automation in milling since they provide greater improvements in productivity, increase the quality of the machined parts and require less operator input. A brief review of literature on surface roughness modeling in milling is presented here. Surface roughness and dimensional accuracy have been important factors in predicting the machining performances of any machining operation.1 Kline et al.2 investigated the effect of vibration, deflection and chatter of the tool-workpiece system on roughness in end milling. Alauddin et al.3 developed the mathematical model of surface roughness for the end milling of 190 BHN steel considering only the centre line average (CLA) roughness parameter (Ra) in terms of cutting speed, feed rate and depth of cut using response surface method (RSM). Fuht and Wu4 studied using RSM the influence of tool geometries (nose radius and flank width) and cutting parameters (cutting speed, feed rate, depth of cut) on surface roughness in end milling of Al alloy. Chun and Ko5 studied machining error caused by tool deflection in the internal boring process using RSM. Chen6,7 and his co-workers studied the effect of spindle speed, feed rate and depth of cut on Ra
in surface roughness recognition and a neural fuzzy system to predict the roughness. Mansour and Abdalla8 studied the roughness (Ra) in end milling of EN 32 steel in terms of machining parameters using RSM. Ertekin et al.9 considered three different materials, viz., 6061 Al, 7075 Al and ANSI 4140 steel for roughness (Ra) study in CNC milling. Benardos and Vosniakos10 used Taguchi design to consider prediction of Ra in CNC face milling of Al alloy. Dweiri et al.11 considered neuro-fuzzy approach for surface roughness (Ra) modeling in CNC down milling of Alumic-79. Ghani et al.12 considered Taguchi method for optimization of surface roughness in end milling of hardened steel in terms of cutting parameters. Brezocnik et al.13 used genetic programming for prediction of Ra in CNC end milling of 6061 Al in terms of machining parameters as well as vibrations. Wang and Chang14 investigated surface roughness in slot end milling of Al. Oktem15,16 and co-workers analyzed the optimum cutting condition leading to a minimum roughness (Ra) in end milling by combining RSM with neural network and genetic algorithm for Al and plastic mold parts. Wang et al.17 investigated the influence of micro-end-milling cutting conditions on roughness of a brass surface using RSM. Reddy and Rao18 developed a mathematical model for surface roughness considering the cutting parameters and tool geometry during end milling of medium carbon steel using RSM. Recently, Ozcelik and Bayramoglu19 have modeled Ra in high speed flat end milling of steel including total tool operating time along with other machining variables such as spindle speed, feed rate and depth of cut. Ryu et al.20 incorporated the effect of cutting edge angle on roughness and texture generation on end milled steel surfaces. They have used RMS deviation, skewness and kurtosis for evaluating the
generated surface texture characteristics. Bagci and Aykut21 used the Taguchi optimization method for low surface roughness value (Ra) in terms of cutting parameters in CNC face milling of Cobalt based alloy. More recently, Chang and Lu22 have presented the optimization of cutting parameters for side milling of medium carbon steel with multiple roughness characteristics, viz., feeding direction roughness, axial direction roughness and waviness using grey relational Taguchi approach. In the present study, an effective mathematical model using RSM integrated with genetic algorithm (GA) optimization technique was developed to determine the combinations of machining parameters leading to the lower surface roughness when end milling of AA 6061 aluminum alloy. The roughness models are developed using RSM for four different conditions using HSS and carbide tools with and without using coolant. Second order models have been developed by using 33 factorial design of experiment. Finally an attempt has been made to obtain optimum machining conditions with respect to each of the roughness values. The lower surface roughness obtained from GA was verified with that of measured from experiments for the feasibility of optimization process.
2. Methodology
In this work, mathematical models have been developed using experimental results with the help of response surface methodology. The purpose of developing mathematical models relating the machining responses and their factors is to facilitate the optimization of the machining process. The mathematical model has been used as an objective function and the optimization was carried out with the help of genetic algorithm. The flow chart of this procedure is briefly illustrated in Fig. 1.
Y = (N, f , d ) +
(1)
where Y is the machining response, is the response function and N, f and d are milling variables and is the error which is normally distributed about the observed response Y with zero mean. The relationship between surface roughness and other independent variables can be represented as follows:
c
R =C N f d
b a a
(2)
where C is a constant and a, b and c are exponents. To facilitate the determination of constants and exponents, the mathematical model will have to be linearized by performing a logarithmic transformation as follows:
a
ln R = ln C + a ln N + b ln f + c ln d
(3)
The constants and exponents C, a, b and c can be determined by the method of least squares. The first order linear model, developed from the above functional relationship using least squares method, can be represented as follows:
3 3
starts by defining a chromosome (solution) as an array of variable values to be optimized. If the chromosome has n variables (an ndimensional optimization problem) given by x1, x2, . . . . . .xn, a chromosome is written as an array with 1 n elements so that Chromosome = [x1, x2, . . . . . .xn] (6)
Y =Y = b x +b x +b x +b x
2 2 1 1 0 0 1
(4)
where Y2 is the estimated response based on the second order equation. The parameters b1, b2, b3, b12, b13, b23, b11, b22, b33 are to be estimated by the method of least squares.
3 33 2
+b x x +b x x +b x +b x +b x
2 22 2 1 11 2 3 2 32 3 1 31
2 1 21
Y =Y = b x +b x +b x +b x +b x x
3 3 2 2 1 1 0 0 2
(5)
where xinorm = Normalized value of variable, 0 xinorm 1 Now to begin the GA, a population of sp chromosomes is defined by a matrix with each row in the matrix being 1 n array of continuous values.
i l
i mron
i l
i u
where Y1 is the estimated response based on the first-order equation, Y is the measured surface roughness on a logarithmic scale, x0 =1, x1, x2 and x3 are logarithmic transformations of cutting speed, feed rate and depth of cut respectively, is the experimental error and b values are the estimates of corresponding parameters. The general second order polynomial response is as given below:
All variables are normalized to have values between 0 and 1, the range of a uniform random number generator. Then the values of a variable are un-normalized in the fitness function. If the range of values of an ith variable is between xil and xiu, then the unnormalized value is given by: x = (x x
i
+x
(7)
computing offspring solutions y(1) and y(2) from two parent solutions x(1), x(2) are as follows: (1) Create a random number u between 0 and 1. (2) Find a parameter using a polynomial probability distribution such that the area under the probability curve from 0 to is equal to u. Then the parameter is given by: ( u ) , = 1 2 u
1+ p 1
if u 1/ (8) , otherwise
1+ p 1
and xl, xu = lower & upper bounds of variable under consideration. It is assumed that x(1) < x(2). This procedure allows a zero probability of creating any offspring solution outside the prescribed range [xl, xu ]. A small value of p allows solutions for away from parents to be created as offspring solutions and a large value restricts only near parent solutions to be created, as offspring solutions. In all simulation results, p = 2 has been used. (3) The offspring solutions are then calculated as follows: y(1) = 0.5[(x(1)+ x(2)) | x(2) x(1)|], y
(2)
= 0.5[(x + x ) + | x x |]
Where q = Distribution index for mutation (non negative). = min [(x xl), (xu x)]/ (xu xl) This ensures that no solution would be created out side the range (xl, xu). (1) Calculate the mutated child as follows: y = x + max. (12)
where max = maximum perturbance allowed in the parent solution = xu xl. In all simulation results, q = 100 has been used.
2.2.4 GA Parameters
Selecting GA parameters like population size (sp), crossover
1+ q 1+ q 1
( )
()
(1)
(2)
(2)
(1)
if u 0.5, , otherwise,
=1+
+ p (
=2
min x( ) x , x x ( x
)(
probability (pc), mutation probability (pm), and number of generations (ng) is very difficult due to many possible variations in the algorithm and objective function. A real parameter GA relies on random number generators for creating the population, crossover and mutation. A different random number seed produces different results. As far as population size is considered, traditionally large number of population of solutions has been used to thoroughly explore complicated objective surfaces. The number of generations is something like termination criteria, which indicates how many times the trials (iterations) are to be made. Crossover probability is used to find the probable number of solutions (sp, pc) to be crossed over to produce an equal number of offspring solutions. In order to preserve some good solutions selected during reproduction operator, ((1-pc). sp) number of solutions are simply copied to the new population. This process helps in exploiting promising regions of objective space by combining information from promising solutions. Mutation probability is used in finding the number of solutions to be mutated (pm x spn). This process helps in exploring different areas of the objective space by randomly introducing changes, or mutations, in some of the variable.
3. Experimental details
3.1 Design of experiment
(9) (10) The design of experiments technique is an important tool, which permits us to carry out the modeling and analysis of the influence of process variables on the response variable. The response variable is an unknown function of the process variables, which are known as design factors. There are a large number of parameters that can be considered for machining of a particular material in end milling. In the present study most widely used machining parameters such as spindle speed, feed rate and depth of cut are considered as design factors. The range of values of each factor was set at three different levels as shown in Table 1. A full factorial design is used to design factors so that all the interactions between the response variable and process variables can be investigated.
(11)
surface roughness tester (Mitutoyo Surftest SJ 201) at a 0.8 mm cutoff value. An average of three measurements was used as a response value.
HSS tool, w ithout coolant 6 5.5 5 Ra m 4.5 4 3.5 3 0 10 20 Test num ber
HSS tool, w ith coolant 6 5.5 5 Ra m 4.5 4 3.5 3 0 10 20 Test num ber
Carbide tool, w ithout coolant 5 4.5 4 3.5 Ra m 3 2.5 2 1.5 1 0.5 0 0 10 20 Test num ber 30 Experimental Predicted
Experimental Predicted
30
Experimental Predicted
30
Table 2 Chemical composition of 6061 aluminum alloy (wt. %) Al Cr Cu Fe Mg Mn Si Ti Zn 95.8- 0.04- 0.15- Max. 0.8- Max. 0.4- Max. Max. 98.6 0.35 0.4 0.7 1.2 0.15 0.8 0.15 0.25
HSS w ithout coolant Carbide w ithout coolant 6 5 HSS w ith coolant Carbide w ith coolant
5 4.5 4 3.5
Ra m
Experimental Predicted
20
30
Fig. 2 The experimental values of surface roughness under varying machining conditions
Fig. 3 The variation of predicted Ra and experimental Ra under varying tool and coolant conditions
speed in HSS tools, but these are wider in carbide tools. It was also observed that the variation of surface roughness with the change of speed in carbide cutting tools is more, with that of HSS tools. Using these experimental results, empirical equations have been obtained to estimate surface roughness with the significant
parameters considered for experimentation i.e. cutting speed, feed rate and depth of cut. The second-order response equations have been fitted using Minitab software for the response variable Ra.29 The response equations were generated separately for the machining of material using HSS and coated carbide cutting tools
6 5 R a m 4 3 2 1 0 100 200
d=1.5m m
6 5 Ra m
d=1m m
d=0.5m m 6 5 Ra m 4 3 2 1
4 3 2 1
500
0 100
200
400
500
0 100
200
400
500
d=1mm
6 5 Ra m
5 4 Ra m 3 2 1 0 100
d=0.5mm
Ra m
200
400
500
200
400
500
5 4 Ra m 3 2 1 0 100 200
d=1m m
d=0.5m m
500
200
500
5 4 R a m 3 2 1 0 100
d=1.5mm
200
500
d=1mm
3.5 3 2.5 Ra m 2
d=0.5m m
Ra m
1.5 1 0.5
200
400
500
0 100
200
500
(d) Carbide tool with coolant N 500 rmp N 1000 rpm N 1500 rpm
Fig. 4 Variation of surface roughness with feed rate at the selected levels of depth of cut
The model would be optimized using a real coded GA. In this context an effort has been made to optimize the process variables that produce the best possible surface quality within the assumed variable bounds. There will be a total of four optimization problems to be solved. Each optimization problem consists of a minimization function defined by one of the second order equations given by (13), (14), (15), (16) and the following variable bounds. 500 rpm N 1500 rpm 200 mm/min f 400 mm/min 0.5 mm d 1.5 mm xil xi xiu where xil and xiu are the upper and lower bounds of process variables xi and x1, x2, x3 are logarithmic transformation of spindle speed, feed rate and nose radius. The real parameter GA was coded in C-language and a parametric study of GA is carried out in solving the problems. The study was carried out by varying different GA parameters, viz., pc, pm, sp, and ng. By this study, the best set of GA parameters which give the minimum value of objective function can be found. The same set of GA parameters is used in solving the corresponding optimization problem.
0.000008 x + 0.220 x
2 2
(13)
(2) Machining with HSS tool with coolant Ra = 2.34 0.00275 x1 + 0.0170 x2 + 0.682 x3 + 0.000002 x1 x2 0.000387 x1 x3 0.00022 x2x3 + 0.000001 x
3 2 1 2
0.000026 x 0.193 x
2 2
(14)
(3) Machining with coated carbide tool without coolant Ra = 0.73 0.00430 x1 + 0.0196 x2 + 1.32 x3 0.000003 x1 x2 0.000820 x1 x3 + 0.00295 x2 x3 + 0.000002 x
3 2 1 2
0.000022 x 0.529 x
2 2
(15)
(a) Spindle speed and feed rate on Ra (b) Spindle speed and feed rate on Ra with (c) Spindle speed and depth of cut on Ra without coolant coolant without coolant
5.5 5.4 5.3 5.2 5.1 5 4.9 4.8 4.7 4.6 0 0.5 1 1.5 2
(d) Spindle speed and depth of cut on Ra (e) Depth of cut and feed rate on Ra without (f) Depth of cut and feed rate on Ra with with coolant coolant coolant Fig. 5 Variation of surface roughness with varying machining conditions
0002
mm 5.0 mm 1 mm 5.1
0051
5.1
mp r ,d e ep s eldnipS
m m tuc fo htp eD
0001
005
5.0
3 2
0.000013 x 0.384 x
2 2
1 2
under the conditions of with and without using coolant. The analysis of variance (ANOVA) and the F-ratio test have been performed to justify the accuracy of the fit for the mathematical models as well as the significance of the individual model coefficients. The additional experiments were conducted to test the adequacy of response variable. As the first order model has low predictability, second order surface roughness equations are considered for optimization of machining parameters. These can be given in terms of machining variables as the following: (1) Machining with HSS tool without coolant
1 2
6 5 4 3 5.5
a 5.4
5.3
4.2
9.1
R a 4.1 m
9.0
4.0
R m
0002
mp r d e ep s eldnipS
m m tuc fo htp eD
R a m
0002
0002
mm 5.0 mm 1 mm 5.1
0051
0051
mp r ,d e ep s eldnipS
mp r d e ep s eldnipS
0001
005
005
5 4R 3 5.4 5.3
5.5
1R a m
5.2
5.1
5.0
a m
Now for the optimization problem involving end milling of 6061 aluminum alloy using HSS tool without coolant initially one parameter, viz., crossover probability (pc) varied from 0.89 to 0.99 in steps of 0.005, keeping other parameters fixed to the values of pm = 0.01, sp= 20 and ng =20. Then all the objective function values are compared, and pc corresponding to minimum value of objective function value is selected as the best pc. It was found to be 0.895. The above study is then repeated for different values of mutation probability from 0.005 to 0.11 in steps of 0.005, keeping the other parameters fixed to the values of pc =0.895, sp= 20 and ng =20. Then the best pm corresponding to minimum value of objective function was found to be 0.085. By keeping the parameters pc =0.895, pm = 0.085, ng =20, the population size is varied from 20 to 220 in steps of 10. The best sp was found to be 30. Finally the number of generations is varied from 20 to 250 in steps of 10, keeping the other parameters constant at pc =0.895, pm = 0.085, sp = 30. The best ng was found to be 250. Thus the best GA parameters after the study are: pc =0.895, pm = 0.085, sp = 30, ng = 250. The convergence of GA to the minimum objective function value (Ra) for all the four optimization problems is shown in Fig. 6. The same procedure is followed for other optimization problems. The results found by GA are compared with those obtained from conformation experiments and given in Tables 3 and 4. They show fairly good agreement with each other.
4
5. Conclusions
The approach presented in this paper is to develop mathematical models, based on experimental results for obtaining a surface roughness using the response surface methodology. The predicted surface roughness from the model is compared to the values measured experimentally. The feed rate is a dominant parameter and the surface roughness increases rapidly with the increase in feed rate and decreases with increase in spindle speed, where as the effect of depth of cut is not regular. This technique can produce accurate relationship between machining parameters and surface roughness. Surface roughness decreases with the use of carbide tool as compared to HSS tool. It was also observed that surface roughness decreases with the use of coolant. GA has been used to estimate the optimum machining conditions to produce the best possible surface quality within the permissible bounds. Optimum machining parameter combinations for response variable is also tested through conformation experiments that show fairly good agreement with values obtained from GA. This study can be extended to other machine tools by using more cutting parameters, tool geometries and different cutting tools and workpiece materials.
REFERENCES
1. Mittal, A. and Metha, M., Surface finish prediction models for fine turning, International Journal of Production Research, Vol. 26, No. 12, pp. 1861-1876, 1988. 2. Kline, W. A., Devor, R. A. and Shareef, I. A., The prediction of surface accuracy in end milling, Journal of Engineering for Industry, Transactions of ASME, Vol. 101, No. 3, pp. 272-278, 1982. 3. Alauddin, M., Baradie, M. A. and Hashmi, M. S. J., Computeraided analysis of a surface roughness model for end milling, Journal of Materials Processing Technology, Vol. 55, No. 2, pp.
Fig. 6 Convergence graphs of GA for the optimization problems Table 3 Optimization results using HSS tool Condition Without coolant With coolant GA Parameters pc =0.895, pm =0.085 sp =30, ng = 250 pc =0.945, pm =0.05 sp =140, ng = 250 N (rpm) 1487.5 1465.24
Table 4 Optimization results using coated carbide tool Condition Without coolant With coolant GA Parameters pc =0.89, pm =0.095 sp =20, ng = 240 pc =0.88, pm =0.1 sp =150, ng = 250 N (rpm) 1500 1476.1 Cutting parameters f (mm/min) d (mm) 200 200 1.5 1.5 Objective function Ra (m) Genetic algorithm Experimental 0.75 0.67 0.77 0.68
052
002
. oN n oi ta r e n eG
051
001
05
O 1 b j e c 5.1 t i v e f u n c
Objective function Ra (m) Genetic algorithm Experimental 3.61 3.07 3.67 3.05
123-127, 1995. 4. Fuh, K. H. and Wu, C. F., A proposed statistical model for surface quality prediction in end-milling of A1 alloy, International Journal of Machine Tools and Manufacture, Vol. 35, No. 8, pp. 1187-1200, 1995. 5. Chun, S.-H. and Ko, T. J., Study on the Response Surface Model of Machining Error in Internal Lathe Boring, Int. J. Precis. Eng. Manuf., Vol. 12, No. 2, pp. 177-182, 2011. 6. Lou, M. S., Chen, J. C. and Li, C. M., Surface roughness prediction technique for CNC end milling, Journal of Industrial Technology, Vol. 15, No. 1, pp. 1-6, 1998. 7. Lou, M. S. and Chen, J. C., In-process surface roughness recognition (ISRR) system in end-milling operations, Int. J. Adv. Manuf. Technol., Vol. 15, No. 3, pp. 200-209, 1999. 8. Mansour, A. and Abdalla, H., Surface roughness model for end milling: a semi-free cutting carbon casehardening steel (EN32) in dry condition, Journal of Materials Processing Technology, Vol. 124, No. 1-2, pp. 183-191, 2002. 9. Ertekin, Y. M., Kwon, Y. and Tseng, T. L., Identification of common sensory features for the control of CNC milling operations under varying cutting conditions, International Journal of Machine Tools and Manufacture, Vol. 43, No. 9, pp. 897-904, 2003. 10. Bernados, P. G. and Vosniakos, G. C., Prediction of surface roughness in CNC face milling using neural networks and Taguchis design of experiments, Robotic and Computer Integrated Manufacturing, Vol. 18, No. 5-6, pp. 343-354, 2002. 11. Dweiri, F., Al-Jarrah, M. and Al-Wedyan, H., Fuzzy surface roughness modeling of CNC down milling of Alumic-79, Journal of Materials Processing Technology, Vol. 133, No. 3, pp. 266-275, 2003. 12. Ghani, J. A., Choudhury, I. A. and Hassan, H. H., Application of Taguchi method in the optimization of end milling parameters, Journal of Materials Processing Technology, Vol. 145, No. 1, pp. 84-92, 2004. 13. Brezocnik, M., Kovacic, M. and Ficko, M., Prediction of surface roughness with genetic programming, Journal of Materials Processing Technology, Vol. 157-158, pp. 28-36, 2004. 14. Wang, M. Y. and Chang, H. Y., Experimental study of surface roughness in slot end milling AL2014-T6, International Journal of Machine Tools and Manufacture, Vol. 44, No. 1, pp. 51-57, 2004. 15. Oktem, H., Erzurumlu, T. and Kurtaran, H., Application of response surface methodology in the optimization of cutting conditions for surface roughness, Journal of Materials Processing Technology, Vol. 170, No. 1-2, pp. 11-16, 2005. 16. Oktem, H., Erzurumlu, T. and Erzincanli, F., Prediction of minimum surface roughness in end milling mold parts using
neural network and genetic algorithm, Materials and Design, Vol. 27, No. 9, pp. 735-744, 2006. 17. Wang, W., Kweon, S. H. and Yang, S. H., A study on roughness of the micro-end-milled surface produced by a miniatured machine, Journal of Materials Processing Technology, Vol. 162-163, pp. 702-708, 2005. 18. Reddy, N. S. K. and Rao, P. V., Selection of optimum tool geometry and cutting conditions using surface roughness prediction model for end milling, Int. J. Adv. Manuf. Technol., Vol. 26, No. 11-12, pp. 1202-1210, 2005. 19. Ozcelik, B. and Bayramoglu, M., The statistical modeling of surface roughness in high speed flat end milling, International Journal of Machine Tools and Manufacture, Vol. 46, No. 12-13, pp. 1395-1402, 2006. 20. Ryu, S. H., Choi, D. K. and Chu, C. N., Roughness and texture generation on end milled surfaces, International Journal of Machine Tools and Manufacture, Vol. 46, No. 3-4, pp. 404-412, 2006. 21. Bagci, E. and Aykut, S., A study of Taguchi optimization method for identifying optimum surface roughness in CNC face milling of cobalt-based alloy (satellite 6), Int. J. Adv. Manuf. Technol., Vol. 29, No. 9-10, pp. 940-947, 2006. 22. Chang, C. K. and Lu, H. S., Design optimization of cutting parameters for side milling operations with multiple performance characteristics, Int. J. Adv. Manuf. Technol., Vol. 32, No. 1-2, pp. 18-26, 2007. 23. Oktem, H., An integrated study of surface roughness for modeling and optimization of cutting parameters during end milling operation, Int. J. Adv. Manuf. Technol., Vol. 43, No. 910, pp. 852-861, 2009. 24. Michalewicz, Z., Genetic Algorithms - Data StructuresEvolution Programs, Springer, p. 17, 1999. 25. Deb, K., An Efficient Constraint handling method for Genetic Algorithms, Computer Methods in Applied Mechanics and Engineering, Vol. 186, No. 2-4, pp. 311-338, 2000. 26. Goldberg, D. E. and Deb, K., A comparative analysis of selection schemes used in genetic algorithms, in: Foundations of Genetic Algorithms - I, Morgan Kaufmann Publishers Inc., pp. 69-93, 1991. 27. Deb, K. and Agarwal, R. B., Simulated binary crossover for continuous search space, Complex Systems, Vol. 9, No. 2, pp. 115-148, 1995. 28. Deb, K. and Goyal, M., A combined genetic adaptive search (Gene AS) for engineering design, Computer Science and Informatics, Vol. 26, pp. 30-45, 1996. 29. Minitab Inc., Meet MINITAB Release, Ver. 14, 2003.