Parameter Optimization of ASSAB XW 42 Tool Steel on End Milling Process with MQCL Using Taguchi-WPCA

— Determination of a combination of process variables that are not appropriate in the end milling process will result in high surface roughness and can reduce the metal removal rate. Therefore, it is necessary to adjust the end milling process variables with the appropriate minimum quantity cooling lubrication. This study aims to obtain a combination of end milling process variables on ASSAB XW-42 material using the Taguchi-WPCA method to minimize arithmetic roughness (Ra), quadratic roughness average (Rq), and average roughness from peak to valley (Rz), and maximize metal removal rate (MRR) simultaneously. The cooling fluid method used is the minimum quantity of cooling lubrication (MQCL). The end milling process variable that is varied is the cutting fluid (soluble oil and vegetable oil), spindle speed (178 rpm, 310 rpm, and 570 rpm), feed rate (33.5 mm / minute, 59.4 mm / minute and 111.9 mm / minute) and the cutting depth (0.125 mm, 0.25 mm and 0.5 mm). The cutting tool used in this study is solid carbide end mill having four cutting edges with a diameter of 10 mm. The experimental design of the L 18 orthogonal array was used in this study. The results showed that the optimal roughness of the workpiece surface and metal removal rate (MRR) was given by vegetable oil cutting fluid, 570 rpm of spindle speed, 33.5 mm / minute of feed rate, and 0.25 mm of the cutting depth. The Cutting fluid, spindle speed, and feed rate have a significant effect on the response variables observed simultaneously.


I. INTRODUCTION
Cutting tools in the milling process that are widely used in the industry of manufacturing, such as the automotive industry, aircraft, and plastic molds are end mill. The cutting edge of the end mill cutting tool is on the tip of the face and the spiral side. The selection of variable end milling processes such as feed rate, spindle speed, cutting depth and tool type and the coolant must be precise to obtain a low workpiece surface roughness value and high metal removal rate. The function of the coolant in the machining process is to bring down the friction coefficient, bring down the heat of the cutting tool and clean the chip from the material surface. Also, the use of coolant can increase the quality of the workpiece surface [1]. Besides being useful during the metal cutting process, coolant causes health problems for operators and the environment [2]. The coolant in the machining process consists of a mixture of water and oil, containing irritant and allergic ingredients such as surfactants, alkanol amines and preservatives [3]. Exposure to excessive metal cutting fluid can cause skin irritation [3]- [5]. The method of providing cutting fluid using an environmentally friendly coolant is the focus of current research [6], [7]. One method of providing environmentally friendly coolant is to use Minimum Quantity Cooling Lubrication [8]- [10].
Surface roughness is used to check the end quality of the work surface resulting from the process of machining [11]. At the same time, the metal removal rate (MRR) is used to check the productivity. The greater the MRR, the higher the productivity. However, the characteristics of quality and productivity in the process of machining are different. The roughness of the surface has the characteristics that the smaller is better and the MRR the higher the best. Therefore, determining the combination of variable milling processes such as cutting speed, feeding and proper cutting depth to get optimal results is very important to do besides the use of environmentally friendly coolant. This step is done so that when carrying out the machining process does not use the trial and error process in determining the process variables. Taguchi is an optimization method for a single response that effectively controls the quality of the product. Whereas for multiple responses optimization, it is used a combination of the Taguchi method with weighted principal components analysis (WPCA), fuzzy logic, grey relational analysis (GRA), and genetic algorithm (GA). Das et al. carried out the optimization of the Al 7075 / SiCp MMC material in the turning process by using the Taguchi method combined with WPCA [12]. Panda et al. optimized the turning process variables to optimize the response of surface roughness using the Taguchi method combined with the WPCA method [13]. While Nayak et al. did the optimization parameters of the abrasive jet machining process using a combination of the Taguchi method with WPCA [14].

A. Materials
ASSAB XW-42 tool steel was used in this experiment as a workpiece with a dimension of (80x30x30) mm, as shown in Fig. 1. The tool used was four cutting edge end mills with a diameter of 10 mm. The traditional Milling machines with a maximum spindle rotation of 2000 rpm was used in this experiment. Mitutoyo surf test and stopwatch were respectively used to measure surface roughness and cutting time. Then, the cutting time is put in equation 1 to get the MRR. MRR is the volume of material removed per unit of time or minute [15].

B. Experimental Design
Process variables used in this study are shown in Table 1. The response variables used in this experiment are average arithmetic roughness (Ra), quadratic roughness average (Rq), and average roughness from peak to valley (Rz) and metal removal rate (MRR). The method of applying coolant used is the minimum quantity of cooling lubrication (MQCL). Based on Table I, the total degrees of freedom of the response variable is 7. The degrees of freedom from the orthogonal matrix used must be larger than or equal to the total degrees of freedom of the predetermined factors and levels [16]. According to the choices available, orthogonal L18 matrices meet the requirements to be used as experimental designs, as shown in Table 2.

C. Optimization Taguchi-WPCA
The Taguchi Method seeks to achieve this goal by making products and processes insensitive to various noise factors, such as materials, manufacturing equipment, human labor, and operational conditions. However, the Taguchi technique is only used to make one response [17]. Taguchi techniques can be combined with WPCA to do multiple-responses optimization simultaneously.
The WPCA is used to remove the correlation among the responses and to change the correlated responses to an uncorrelated responses index named the major components (Principal Components) [18]. The main components that each have different variance values are independent of each other; therefore, to produce the total variance value, each variance of the main component is considered or used as a weight. The main components are accumulated first to count the Multi Responses Performance Index (MPI). Next, the value of combined quality loss (CQL) is calculated, which is defined as the deviation from the MPI value from the desired ideal value. CQL aims to reduce the MPI deviation from the ideal value [14]. The Taguchi-WPCA optimization stage can be seen in Fig. 2.

III. RESULTS AND DISCUSSION
The value of arithmetic roughness (Ra), quadratic roughness average (Rq), and average roughness from peak to valley (Rz) and metal removal rate (MRR) results from this study are shown in Table 3. Surface roughness Ra, Rq and Rz are measured by using surftest Mitutoyo SJ-210 while the metal removal rate is calculated by using equation 1. Based on Table III, the lowest surface roughness of Ra is 0.494 µm in combination experiment of number 18. While the lowest surface roughness of Rq and Rz is 0.693 and 2.534 μm, respectively in combination experiment of number 17. Whereas the most significant metal removal rate (MRR) is in combination experiment of no. 8 at 332,760 mm 3 /minute. Therefore, an optimization process is needed to get the setting of the end milling process variables that produce a low surface roughness value with a high metal removal rate (MRR).
The results of the data normalization are shown in Table  4. The results of the normalization process Lij are between 0 and 1. The maximum value of Lij is 1 and is considered an ideal condition.

B. Calculating the Pearson Correlation Coefficient (ρ)
The next step is estimating the Pearson correlation coefficient (ρ). The correlation coefficient is used to see whether there is a correlation between the observed response variables. The calculation of the Pearson Correlation is done using equation 4.
The correlation coefficient between responses is shown in Table 5. The Pearson correlation coefficient (ρ) does not equal zero, so there is a correlation among responses. The highest Pearson correlation coefficient value of 0.710 is the correlation between the surface roughness of Rq and Rz. While the lowest correlation value of -0.230 is the correlation between Rq surface roughness with Metal Removal Rate (MRR).
The Correlation value can be positive and negative numbers. If it is positive, the relationship is one-way. If it is negative, the relationship is not unidirectional.

C. Calculating Principal Components (PC)
Based on Table 5, all responses are correlated. The calculation of the principal component score (PC) is carried out to eliminate the correlation between responses. The principal component consists of eigenvalue, AP (accountability proportion), eigenvector, and CAP (cumulative accountability proportion). They all can be seen in Table 6. The eigenvector values of Y 1 , Y 2 , Y 3 and Y 4 in Table 6 are used to calculate the principal component scores of PC 1, PC 2 , PC 3 and PC 4 by using equation 5.
Where Y i is the value of the principal component score (PC), S ij is the normalized data, and β is the eigenvector value. The result of the Individual principal component (PC) values is shown in Table 7. The principal components (PC) shown in Table 7 Table 6 is used as a weight to calculate MPI.

E. Calculate CQL (Combined Quality Los)
After calculating the MPI value, the next step is to perform a CQL calculation. CQL value calculation is done by calculating the absolute difference between MPI values in ideal conditions with MPI values from response data. The results of calculating the combined Quality Loss (CQL) are shown in Table 9.

F. Calculating the value of the signal to noise (S/ N) ratio
The ratio of S/N is calculated based on the characteristics of the CQL value, which is smaller the better, using equation (7). Calculation of the ratio of S/N is done to minimize the value of the estimated loss from the CQL. The results of the calculation of the ratio of S/N are shown in Table 10.

G. Selecting the Optimal End Milling Process Parameter
The following step is to determine the average of the ratio of S/N for each level and group them as in Table 11. The plot for the average ratio of S/N in Table 11 is shown in Fig.  3. The end milling is processing variable level combinations that produce the optimum response based on Fig. 3 are Cutting Fluid (CF) level 2 namely vegetable oil, level 3 of spindle speed (N) of 570 rpm, level 1 of feed rate (Vf) of 33.5 mm / min and level 2 of the cutting depth (A) of 0.25 mm.

H. Confirmation Experiment
Confirmation experiments are carried out to validate the results that have been obtained [19]. Confirmation experiment is conducted by comparing the results of the combination of optimization with the initial combination. The initial combination and optimum combination can be seen in Table 12, and the result of the confirmation experiment can be seen in Tables 12.  Table 13 shows that the surface roughness value of Ra has decreased by 40.1%, the surface roughness value of Rq has decreased by 20.8%, the Rz surface roughness value has decreased by 5.4%, and the metal removal rate (MRR) has reduced by 43.5%. The amount of the contribution and the significant influence of process variables on the response variables studied can be determined by using ANOVA. In this study, ANOVA is carried out on the value of the ratio of signal to noise (S/N) of CQL, which represents all responses simultaneously. The results of ANOVA calculation of the ratio of S/N of CQL are shown in Table XIV. Based on Table XIV, the calculated F value for cutting fluid (CF) and the spindle speed (N) successively are 10,382 and 7,399, and the feed rate (Vf) is 7,511, higher than the F table which is 3,370. This value shows that the cutting fluid (CF), spindle speed (N), and feed rate (Vf) process variables have a significant effect on the response variables observed simultaneously. The variable feed rate gave the most significant contribution in decreasing the total variance by 23,709%, spindle speed by 23,301%, cooling fluid by 17,080%, and cutting depth by 4,961%. The process can be explained that the level of surface quality of the workpiece will increase with increasing cutting speed and will decrease with increasing feed rate [20]. Giving the right coolant on the end milling process can reduce heat during the process and improve the surface quality of the workpiece.

IV. CONCLUSION
This study aims to obtain a combination of end milling process variables on ASSAB XW-42 material using the Taguchi-WPCA method to minimize arithmetic roughness (Ra), quadratic roughness average (Rq), and average roughness from peak to valley (Rz) and maximize metal removal rate (MRR) simultaneously. The cooling fluid method used is the minimum quantity of cooling lubrication (MQCL). Based on the results of the study, the following conclusions can be drawn as follow: The coolant (CF), spindle speed (N) and feed rate (V f ) had a significant effect on the variable arithmetic roughness (Ra), quadratic roughness average (Rq), and average roughness from peak to valley (Rz) and metal removal rate (MRR) which were observed simultaneously. The variable feed rate gave the biggest contribution in decreasing the total variance by 23,709%, spindle speed by 23,301%, cooling fluid by 17,080%, and cutting depth by 4,961%. The end milling process variables are set as follows to get optimal surface roughness response (Ra, Rq and Rz) and metal removal rate (MRR), Vegetable oil coolant type, spindle speed of 570 rpm, a Feed rate of 33.5 mm / minute and radial cutting depth of 0.25 mm.