OPTIMAL MQL AND CUTTING CONDITIONS DETERMINATION FOR DESIRED SURFACE ROUGHNESS IN TURNING OF BRASS USING GENETIC ALGORITHMS

The evolving concept of minimum quantity of lubrication (MQL) in machining is considered as one of the solutions to reduce the amount of lubricant to address the environmental, economical and ecological issues. This paper investigates the influence of cutting speed, feed rate and different amount of MQL on machining performance during turning of brass using K10 cemented carbide tool. The experiments have been planned as per Taguchi's orthogonal array and the second order surface roughness model in terms of machining parameters was developed using response surface methodology (RSM). The parametric analysis has been carried out to analyze the interaction effects of process parameters on surface roughness. The optimization is then carried out with genetic algorithms (GA) using surface roughness model for the selection of optimal MQL and cutting conditions. The GA program gives the minimum values of surface roughness and the corresponding optimal machining parameters.

Keywords: brass; genetic algorithms; Minimum Quantity of Lubricant (MQL); Response Surface Methodology (RSM); surface roughness; turning

Cutting fluids are generally used in machining to improve the characteristics of tribological processes. The cutting fluids do not only increase the tool life but also provide a better surface finish on the machined surfaces. However, in modern production, there has been an increasing attention on carefully selecting proper cutting fluids from the viewpoint of cost, ecology and environmental issues (Sreejith and Ngoi, [27]). In the recent past, a lot of efforts have been made to minimize or even completely avoid the use of cutting fluids (Sokovic and Mijanovic, [26]).

The dry cutting is preferred in the field of environmental friendly manufacturing (Cantero et al., [2]). But, they are less effective when higher machining efficiency, better surface finish and severe cutting conditions are required. The minimum quantity of lubricant (MQL) concept in machining is an alternative to completely dry or flood lubricating system for reducing the amount of lubricant (Diniz et al., [8]). The MQL refers to the use of cutting fluids in a small quantity and controlling flow rate in the range between 50 and 500 ml/hr. The minimum quantity lubricants have the advantages of better lubricating capability and advanced thermal stability over the conventional cutting fluids.

Rahman et al. ([23]) carried out an experimental study on end milling of ASSAB 718HH steel workpieces under MQL and flood coolant systems. The comparative effectiveness was investigated in terms of cutting force, tool wear, surface roughness and chip shape using carbide end mills and concluded that MQL may be considered as an economical and environmentally compatible lubrication technique. Kishawy et al. ([16]) reported the application of different coolant strategies to high speed milling of aluminium alloy (A356). The effects of flood coolant, dry cutting and MQL techniques on tool wear, surface roughness and cutting forces have been studied and concluded that the MQL could be an alternative to flood cooling. Lopez de Lacalle et al. ([17]) carried out experimental and numerical investigations on high speed milling of wrought aluminium alloys to study the effect of spray cutting fluids. The efficiency of MQL was assessed through computational fluid dynamics (CFD) simulation and experimental evidences.

Braga et al. ([1]) reported that the holes obtained during drilling of aluminium-silicon alloys with 7% silicon (SAE 323) using uncoated and diamond coated K10 carbide tools with MQL technique presented either similar or better quality than those obtained with flood lubricant system. Hanyu et al. ([11]) presented an investigative study on dry and semi-dry drilling of 12% silicon aluminium alloy (JISADC12) with finely crystallized smooth surface of diamond coatings. The study revealed that smooth surface diamond coating could lead to longer durability of drills as compared to conventional rough surface diamond coating in both dry and semi-dry cutting conditions.

The experimental studies on drilling of aluminium alloys (AA1050) with K10 carbide tools under dry, MQL and flood lubricated conditions were carried out by Davim et al. ([3]). The effects of cutting speed and feed rate on cutting power, specific cutting pressure and surface roughness were analyzed under different lubrication conditions. The results clearly showed that, by using MQL, it is possible to obtain a performance quality similar to flood cooling with a proper selection of machining parameters.

The tool life and cutting force were investigated by Sun et al. ([28]) on titanium alloy based on dry cutting, flood cooling and MQL techniques. The experimental results showed that MQL machining could remarkably and reliably improve tool life and reduce cutting force due to better lubrication and cooling effect. Zeilmann and Weingaertner ([29]) performed temperature analysis during drilling of titanium alloy (Ti6Al4V) using different types of coated K10 carbide drills under MQL conditions. The results showed the potential for drilling with MQL applied internally through drills as compared to lubricant applied using an external nozzle. The use of MQL in deep hole drilling on plain carbon steel (0.45% C) using small diameter HSS twist drills has been reported by Heinemann et al. ([12]). The results of their study showed that the discontinuous supply of MQL could reduce tool life as compared to continuous supply of MQL.

Lugscheider et al. ([18]) applied the MQL technique in reaming process of grey cast iron (GG25) and aluminium alloy (AlSi12) with coated carbide tools. The significant reduction in tool wear and improvement in surface quality of the holes have been observed using MQL technique when compared to dry cutting. The investigations carried out by Kelly and Cotterell ([15]) on aluminium alloy machining revealed that the MQL technique is preferred with higher cutting speeds and feed rates. The performance of MQL technique in the grinding process was evaluated by Roberto da Silva et al. ([24]) based on the analysis of surface integrity (roughness, residual stress, microstructure and micro hardness). The results are expected to lead to technological and ecological gains in the grinding process using MQL concept.

Dhar et al. ([5]) employed the technique of MQL in turning of AISI 1040 steel and the results indicated that a mixture of air and soluble oil machining is better than conventional flood cooling system. Further study on turning of AISI 1040 steel by Dhar et al. ([7]) compared the mechanical performance of MQL over completely dry lubrication based on the experimental measurements of cutting temperature, chip reduction coefficient, cutting forces, tool wear, surface finish, and dimensional deviation. Dhar et al. ([6]) also carried out an experimental investigation on tool wear and surface roughness in turning of AISI 4340 steel workpieces at industrial speed-feed combination with uncoated carbide inserts to study the effect of MQL. The results indicated that tool wear rate and surface roughness could be significantly reduced using MQL technique mainly through the reduction in the cutting zone temperature and favorable changes in the chip–tool and work–tool interaction.

Itoigawa et al. ([13]) studied the effects and mechanisms of MQL in an intermittent turning process of aluminium silicon alloys (AlSi5) using sintered diamond and K10 carbide tools. A difference between MQL with water and the MQL without water was studied to elucidate boundary film behavior on the rake face. Davim et al. ([4]) carried out experimental investigations on machining of brass under different conditions of lubricant environments. The influences of cutting speed and feed rate on machinability were studied and concluded that the flood lubrication can be successfully replaced by MQL type of lubrication. Gaitonde et al. ([9]) applied the Taguchi method and utility concept to determine the optimal process parameters for simultaneously minimizing the surface roughness and specific cutting force in turning of brass. The tool wear and tool life were observed at relatively higher values of cutting speed by Kamata and Obikawa ([14]) on finish turning of a nickel-base super alloy (Inconel 718) with different types of coated carbide tools under MQL conditions.

As seen from the literature, the MQL technique suggests several advantages in various machining processes. But, most of the experimental studies are limited to the role and effectiveness of MQL over dry and wet machining. The surface roughness plays an important role as it influences fatigue strength, resistance to corrosion, coefficent of friction and wear rate on machined surfaces. In order to achieve good surface finish on a machined component either the cutting conditions should be carefully selected with an optimum quantity of MQL or a new tool material to be developed with lower coefficient of friction and higher heat resistance capacity.

As per the knowledge of the authors, no systematic work has been reported in the literature to analyze the interaction effects of MQL and cutting conditions on surface roughness. Further, no attempt has been made to estimate the optimal MQL with appropriate cutting conditions for achieving a better surface finish within the chosen constraints. Keeping this consideration in view, this paper illustrates the application of response surface methodology (RSM) to analyze the interaction effects of MQL, cutting speed and feed rate on surface roughness in turning of brass using K10 cemented carbide tool. The second order surface roughness model has been developed on the basis of experimental results. The develped mathematical model is further utilized to determine the best combination of the machining parameters using genetic algorithms (GA).

The response surface methodology (RSM) using design of experiments (DOE) has been proved to be an efficient modeling tool (Montgomery, [21]). The methodology not only reduces cost and time, but also helps to gain the required information about the main and interaction effects of the parameters with minimum number of experiments. The RSM is a combination of mathematical and statistical techniques, which are useful in building the models and analyzing the problems. The mathematical model of the response to independent parameters can be predicted by employing the multiple regression analysis with reduced number of experiments planned through DOE.

In the present investigation, the second order RSM based mathematical model (Montgomery, [21]) has been developed to study the effect of three machining parameters, namely, quantity of lubricant (Q), cutting speed (v) and feed rate (f) on surface roughness (Ra). The non-linear response surface equation is given by:

Graph

where, Y: response, i.e., Ra ; bo,...., b33: regression coefficients of polynomial equation to be determined. The values of regression coefficients are determined by (Montgomery, [21]):

Graph

where, b: matrix of parameter estimates; X: calculation matrix consisting of linear, interaction, and quadratic terms; XT: transpose of X; Y: matrix of response.

The genetic algorithms (GA) as a tool for process optimization combines the Darwinian principle of natural selection "survival of the fittest" strategy to eliminate unfit solutions and use the random information exchange with the exploitation of knowledge contained in old solutions, generating a search mechanism with surprisingly high power and speed (Goldberg, [10]). The use of GA with gene information and chromosome processing to optimize the given objective function has proved to be an efficient optimization tool.

The GA is more likely to converge to a global optimum as compared to conventional optimization techniques, since GA search from a population of points and on probabilistic rules. The conventional optimization techniques are ordinarily based on deterministic hill-climbing methods, which may find local optima. The GA can also tolerate discontinuities and noisy function evaluations.

In each cycle of genetic operation, termed as an evolving process, a subsequent generation is created from the chromosomes in the current population. This consists of manipulation of genes, where genes of parents are mixed and recombined for the production of offspring in the next generation. This evolution process consists of selection or reproduction, crossover and mutation.

The solution of a problem that GA attempts to solve is coded into a string of binary numbers, termed as chromosomes. Each chromosome contains the information of a set of process parameters. In GA, a population of chromosome is formed randomly and then fitness of each chromosome is evaluated using an objective function after the chromosome has been decoded. After the completion of evaluation, pairs of chromosomes will be selected randomly to undergo the genetic operations, namely, crossover and mutation to produce the offsprings for the evaluation of fitness. This process continues until a near optimal solution is obtained.

The planning for experimentation is necessary to develop the RSM based surface roughness model. The quantity of lubricant (Q), cutting speed (v) and feed rate (f) are selected as the process parameters. The ranges of the cutting conditions were determined from the previous investigations carried out by Gaitonde at al. ([9]). In the present work, the experiments are planned as per Taguchi's orthogonal array (Ross, [25]). Three level tests for each process parameter have been selected to study the non-linearity effect of the machining parameters. The experimental control factors and their levels are illustrated in Table 1. According to Taguchi design, L27 (313) orthogonal array (Phadke, [22]; Ross, 2006) is employed for the experimentation, which requires 27 trials to be conducted. Table 2 shows the planning for experimental design considered for the current investigation. The first column, second column and fifth column of L27 (313) array were assigned to quantity of lubricant (Q), cutting speed (v) and feed rate (f), respectively.

TABLE 1 Experimental Factors and Their Levels

TABLE 2 Experimental Conditions as per L27 Orthogonal Array and Average Surface Roughness (Ra)

The work material used for the turning tests is CuZn39Pb3 brass, according to DIN specification (12164:1998). The brass material has an average hardness of 66 HRB. The chemical composition of the brass workpiece is: Cu – 58.98%; Zn − 38.37%; Pb − 2.87%; Sn − 0.287%; Fe − 0.334% and Ni − 0.13%. 50 mm diameter and 170 mm length workpieces were machined throughout the investigation.

The turning tests were performed on Kingsbury (USA) MHP 50 CNC lathe of 18 kW spindle power with maximum spindle speed of 4500 rpm. K10 cemented carbide inserts of TCGX 16 T3 08- Al H10 (Sandvik, Sweden) were used with the following tool geometry: rake angle: 15° (positive); clearance angle: 7°; major edge cutting angle: 91o; cutting edge inclination angle: 0° and corner radius: 0.8 mm. 'STGCL2020K16' type tool holder was used throughout the investigation. 2 mm depth of cut was kept constant during the turning tests. The experiments were performed as per orthogonal array under different conditions of MQL (lubricated with emulsion oil, Microtrend 231L).

A Hommelwerke® (Germany) T1000 profilometer was used to measure the surface roughness on machined components with a cut off distance of 0.8 mm, in accordance with ISO/DIS 4287/1E. An average of three measurements was used as a response value (Ra) and are given in Table 2.

The multiple regression analysis has been performed to develop the RSM-based, second-order mathematical model of surface roughness by utilizing the experimental results of Table 2. The mathematical model to predict surface roughness in turning of brass using K10 cemented carbide tool is given by:

Graph

where, Q in ml/hr; v in m/min; f in mm/rev; Ra in microns.

The statistical testing of the developed quadratic surface roughness model has been performed by Fisher's statistical test for ANOVA (Montgomery, [21]) at 95% confidence interval. The ANOVA table consists of sum of squares and degrees of freedom. The sum of squares is usually performed into contributions from regression model and residual error. The mean square is the ratio of sum of squares to degrees of freedom. F-ratio is the ratio of mean square of regression model to mean square of the residual error. As per this technique, the calculated value of F-ratio of the developed model should be more than tabulated value of table (F-table, see Table 3) for the model to be adequate at 95% confidence interval. The results of ANOVA for surface roughness model is presented in Table 3 and is found to be adequate at 95% confidence level.

TABLE 3 Summary of ANOVA for RSM-Based Surface Roughness Model

The goodness of fit of the models was also tested by the coefficient of determination (R2) and adjusted coefficient of determination (). R2 provides a measure of variability in the observed response and can be explained by the process parameters along with their interactions. On the other hand, is the coefficient of determination adjusted for the number of independent variables in the regression model. The R2 and values are found to be 0.9812 and 0.971, which clearly indicate the significance of non-linear regression model.

The developed surface roughness model is then used to test the accuracy and the % prediction error is given by:

Graph

where, yi,expt: observed response corresponding to trial i; yi,pred: predicted response corresponding to trial i; n: number of trials. The average prediction error of the developed models with the experimental data of an orthogonal array was found to be 5.99%. The comparison of the predicted and experimental values as per orthogonal array is shown in Figure 1.

Graph: FIGURE 1 Experimental and RSM predicted values for surface roughness (Ra).

The analysis of parametric influences on surface roughness has been carried out through the developed RSM based model by generating 3D response surface plots (Figures 2-4). The response surface plots were generated using Minitab statistical software (Minitab Inc, [20]).

Graph: FIGURE 2 Response surface plot of quantity of lubricant (Q) and cutting speed (v) on surface roughness (Ra).

Graph: FIGURE 3 Response surface plot of quantity of lubricant (Q) and feed rate (f) on surface roughness (Ra).

Graph: FIGURE 4 Response surface plot of cutting speed (v) and feed rate (f) on surface roughness (Ra).

Figure 2 exhibits the interaction effect of quantity of lubricant (Q) and cutting speed (v) on surface roughness (Ra) with feed rate (f) held at 0.1 mm/rev. At high values of Q, the sensitivity of Ra with variations in v is large as compared to low values of Q. The minimum Ra occurs at high value of Q with low cutting speed. The reason might be, the thinner chips produced at lower cutting speed are pushed by high MQL due to capillary effect, which enables it come closer to hot tool-chip zone to remove the heat more effectively and hence surface roughness decreases.

Figure 3 depicts the interaction effect due to Q and f on Ra with v held at 200 m/min. It can be seen from Figure 3 that, the value of Ra sharply increases with f irrespective of Q. This may be due to the fact that, at higher feed rate more amount of heat is generated due to high MRR and hence increase in temperature at tool-chip interface. Due to rise in temperature at tool-chip interface, the tool wear increases and hence surface roughness increases. It is also seen from figure that, variations in Ra are minimal with the variations in Q at all values of f.

Figure 4 shows the variation of v and f on Ra with Q at 100 ml/hr. The Ra increases with f irrespective of v and the variations in Ra with the variations in f are less at all values of v. This is due to the fact that, more amount of material has to be removed per revolution, which in turn requires more amount of energy leading to further increase in cutting forces and temperature. Due to the combined effect of increase in temperature and cutting forces, tool wear is very high, which ultimately leads to increase in surface roughness.

From the preceding discussion, it is evident that there is a definite scope for finding the optimal values of MQL and cutting conditions for selecting the best possible surface finish within the chosen constraints of the machining parameters in turning of brass. This requires an efficient optimization tool and hence GA was selected, which employs the quadratic surface roughness model.

The optimization problem of turning process is stated as minimizing the surface roughness subject to a set of constraints. In the present investigation, the constrained optimization problem using GA is given by:

Find the optimal values of Q, v and f

• To minimize the surface roughness using the model given here:

•

Graph

• Subject to the constraints:
• 50 ≤Q ≤ 200 ml/hr; 100 ≤v ≤ 400 m/min; 0.05 ≤f ≤ 0.15 mm/rev; xil ≤ xi ≤ xiu
• where, xil and xiu are the upper and lower bounds of machining parameter xi.

In GA, the genes are searching parameters, which are represented with finite length of binary codes, 0 and 1. The chromosomes are the strings of defining genes. Thus, the chromosome for the GA optimization in the present investigation consists of 3 genes corresponding to three searching parameters Q, v and f. Each gene is represented by 20 bits of binary codes and hence a chromosome is of length 60 bits.

The optimization process was initialized with 50 chromosomes randomly generated as initial chromosome population. The genes of each chromosome are decoded as:

Graph

for Z = Q, v and f.

where, D: decimal equivalent of binary; Zmax: maximum limit for Z; Zmin: minimum limit for Z; nb: number of binary bits for genes = 20.

The decoded values of genes, Q, v and f were employed with RSM based model of Equation (3) to predict the surface roughness. Using the present generation of 50 chromosomes, the fitness evaluation is given by:

Graph

where, fit is the fitness function and Ra is the objective function. Thus, the minimization of Ra requires the maximization of fit.

The selection is the operator carrying old chromosomes into new population, implements Darwin's survival of the fittest strategy. In the "expected number control method of selection" (Goldberg, [10]), the parent selection is done with a probability in direct proportion to their fitness values. This is carried out based on relative average fitness of each individual in the population of N chromosomes, given as:

Graph

The number of representatives (Ni) of ith chromosome in the first scan is given by:

Graph

where, q indicates the integer part. The residue of the relative average fitness (fitr,i – Ni) is carried as fitness of ith chromosome in the next scan. Whenever a chromosome is selected, its copy is made. The selection process is continued until 50 chromosome representatives are selected.

The crossover is a recombination operator where gene information is exchanged at random locations between the two parent chromosomes, which are randomly selected high fit chromosomes from the population. One point crossover is performed between pairs of randomly selected high fit parent chromosomes to produce their offsprings. The crossover location is also randomly selected. If crossover probability is pc, then the number of crossover is given by:

Graph

where N = No. of chromosomes in the population.

The mutation causes individual genetic representations to be changed according to some probabilistic rule so as to overcome destructive crossover. The mutation operation consists of complementing the bits (replacing 0 by 1 and vice-versa) at random locations. If the mutation probability is pm, then the number of mutations is given by:

Graph

where, nl = length of chromosome = 60.

The steps involved in the GA optimization in the present investigation are as follows:

• Step 1: Generate an initial chromosome population randomly.
• Step 2: Decode the genes Q, v and f of all chromosomes.
• Step 3: Evaluate the predicted values of Ra using the RSM model of Equation. (3).
• Step 4: Determine the fitness of all chromosomes and obtain the maximum fitness (fitmax).
• Step 5: If fitmax ≤ required fitness (fitrequired), then carryout following genetic operations
• a. Selection based on expected number control method,
• b. Crossover, and
• c. Mutation
• to generate new chromosome population and go to step 2.
• Else stop.

MATLAB (Math Works Corporation, [19]) was used to develop the GA code. The following GA parameters have been selected to obtain the best possible solutions:

• Maximum number of generations = 100;
• Total string length = 60;
• No. of chromosomes = 50;
• Cross over = Two points;
• Crossover probability = 0.8;
• Mutation = Two bits;
• Mutation probability = 0.003.

The levels of input process parameter were fed to GA program and the values of MQL and cutting conditions were predicted for minimum surface roughness. The minimum values of surface roughness predicted by GA program with respect to the ranges of machining parameters and the corresponding optimal MQL and cutting conditions are presented in Table 4. From the optimization results of GA, it is seen that the minimum surface roughness value varies from 0.23 microns to 0.5 microns. Further, it is also observed that the optimal conditions for achieving better surface finish can also be determined using GA. Hence, it can be concluded that with the given boundaries of surface roughness and the machining parameters, the turning of brass with K10 carbide tool can be performed with selected operating conditions of process parameters. The determination of optimal machining conditions using GA is really beneficial at the CAPP stage in the manufacturing process especially with tight tolerances and in the adaptive control machine tools.

TABLE 4 Output Values of GA for Chosen Machining Parameters and the Corresponding Optimal Values of Machining Parameters for Minimum Surface Roughness

The optimization results illustrated in the present study are one of such possible optimal solutions. Further, the GA solution also depends on various factors such as chromosome population size and on the crossover and mutation probabilities. The GA parameters employed in the present investigation were selected after several trials of GA simulation and were found to be suitable for the drilling optimization.

The application of GA optimization for minimization of surface roughness in turning of brass with K10 cemented carbide tool using RSM model is presented in this paper. The second order surface roughness model based on RSM was developed using the experimental database obtained from Taguchi's orthogonal array. Three process parameters, namely, MQL, cutting speed and feed rate were considered for the model development. The developed model was then tested through ANOVA. The best combination of the process parameters has been determined using GA. The following conclusions can be drawn from the present investigation within the ranges of the parameters selected:

• Second-order response surface model can be effectively used to predict the surface roughness on machining of brass under different conditions of MQL, cutting speed and feed rate at 95% confidence interval.
• Response surface analysis indicates that high values of quantity of lubricant, the sensitivity of surface roughness with the variations in cutting speed is large as compared to low values. The surface roughness sharply increases with feed rate irrespective of the amount of quantity of lubricant.
• The GA optimization carried out in this work gives the optimal MQL and cutting conditions for achieving a better surface quality within the chosen machining conditions.
• The GA simulation also reveals that the minimum surface roughness value varies from 0.23 microns to 0.5 microns, which is within the ranges of the process parameters selected.

The authors acknowledge Jose Silva, Mechanical Engineer, for his participation in conducting the turning tests.