Multi-Responses optimization of dry milling of SKD61 for low cutting power and surface roughness

SCIENCE TECHNOLOGY 
Số 49.2018 ● Tạp chí KHOA HỌC & CÔNG NGHỆ 85
MULTI-RESPONSES OPTIMIZATION OF DRY MILLING OF SKD61 
FOR LOW CUTTING POWER AND SURFACE ROUGHNESS 
TỐI ƯU HÓA ĐA MỤC TIÊU QUÁ TRÌNH PHAY KHÔ THÉP SKD61 
ĐỂ GIẢM CÔNG SUẤT CẮT VÀ ĐỘ NHÁM BỀ MẶT 
Nguyễn Trung Thành1,*, Nguyễn Tuấn Nhật2 
ABSTRACT 
Optimized process parameters play a significant role in improving the 
energy efficiency and machined part quality. This paper systematically 
investigates the nonlinear relationships between machining parameters and 
responses, including cutting power Pc and surface roughness Ra of the dry milling 
(DM) using the response surface model (RSM). Three process parameters 
considered include the spindle speed S, depth of cut ap, and feed rate fz. A set of 
physical experiments was carried out with SKD61 steel on a CNC milling machine 
using the wiper insert. The target of the current complex optimization is to find 
the low cutting power and surface roughness. Finally, an evolutionary algorithm 
entitled non-dominated sorting genetic algorithm II (NSGA-II) was used to 
generate a set of feasible optimal solutions and determine the best machining 
conditions. The results show that an appropriate trade-off solution can be drawn 
with regard to the low cutting power and surface roughness. Furthermore, the 
integration of RSM model and NSGA-II can be considered as a powerful approach 
for modeling and optimizing dry milling processes. 
Keywords: Cutting power, surface roughness, dry milling, modeling, 
optimization. 
TÓM TẮT 
Thông số công nghệ tối ưu đóng vai trò quan trọng trong nâng cao hiệu suất 
năng lượng và chất lượng sản phẩm. Nghiên cứu này xây dựng mối liên hệ phi 
tuyến giữa các thông số công nghệ với công suất cắt Pc và độ nhám bề mặt Ra quá 
trình phay khô thông qua phương pháp đáp ứng bề mặt. Thông số công nghệ 
được xem xét bao gồm tốc độ trục chính S, chiều sâu cắt ap, và lượng tiến dao fz. 
Các thí nghiệm được thực hiện với thép SKD61 trên máy phay CNC. Mục tiêu của 
quá trình tối ưu hóa các thông số công nghệ là giảm công suất gia công và độ 
nhám bề mặt. Thuật toán di truyền đa mục tiêu được sử dụng để tìm giá trị tối ưu 
của các thông số công nghệ. Kết quả nghiên cứu đã đưa ra một giải pháp phù hợp 
để giảm công suất cắt và độ nhám bề mặt. Bên cạnh đó, sự kết hợp giữa phương 
pháp bề mặt đáp ứng và thuật toán di truyền đa mục tiêu có thể coi là một 
phương pháp hiệu quả để mô hình và tối ưu hóa quá trình phay khô. 
Từ khóa: Công suất cắt, độ nhám bề mặt, phay khô, mô hình hóa, tối ưu hóa. 
1Học viện Kỹ thuật Quân sự 
2Công ty TNHH MTV Cơ khí 25, Bộ Quốc phòng 
*Email:trungthanhk21@mta.edu.vn 
Ngày nhận bài: 10/01/2018 
Ngày nhận bài sửa sau phản biện: 20/02/2018 
Ngày chấp nhận đăng: 25/12/2018 
1. INTRODUCTION 
The industrial sector accounts for about 39% of the total 
energy use and manufacturing dominates the industrial 
energy consumption [1]. Machining is a common 
manufacturing process of production in workshops and 
mechanic factories. Additionally, the energy efficiency of 
machining process is less than 30% [2]. The energy 
efficiency of a case study described by Gutowski is only 14.8 
% [3]. As a result, it has great potential for energy savings in 
machining processes. Energy price is tending to increase 
without acceptable reasons. Furthermore, improving the 
energy efficiency of machining processes is an effective 
approach to reduce the environmental effects, natural 
resource saving, and achieving sustainable development. 
Therefore, reducing energy consumed in machining 
operations is a significant contribution to improving the 
energy efficiency in manufacturing. 
Energy saving technologies for cutting process can be 
divided into two solutions. The first solution mainly focuses 
on machine design and improvement as well as new 
cutting technologies used. The second solution pays 
attention to investigate the relationship among cutting 
conditions and energy consumption and leads to the 
development of energy consumption models and optimal 
parameters in terms of energy savings. Design 
methodologies [4] and the intelligent control [5] were 
proposed to improve the energy efficiency of cutting 
process. Additionally, devices consumed less energy also 
were used to improve the energy efficiency [6]. Apparently, 
the first branch based on hardware technologies is too 
costly to renew or replace existing devices. Improving the 
energy efficiency should be made firstly in existing 
machines and the second solution is an intelligent choice. 
The optimizing cutting process is less expensive and has 
better social sustainability compared to making drastic 
changes due to the low investment needed and user 
acceptance [7]. Consequently, optimal cutting conditions 
selection plays an important role in reducing energy 
consumption in cutting process. 
To meet the challenge of reducing energy 
consumption, a multi-objective optimization of the dry 
 CÔNG NGHỆ 
 Tạp chí KHOA HỌC & CÔNG NGHỆ ● Số 49.2018 86
KHOA HỌC
milling has considered in this paper. The material, namely 
SKD61 was chosen as the workpiece due to wide 
applications in molding, automotive, aerospace, and 
marine industrial. Moreover, the practical analysis indicated 
that machining parameters has complicated effects on the 
machining responses, such as cutting energy and surface 
roughness. Therefore, an effective approach for modeling 
dry cutting and optimizing process parameters is still 
urgent demand. This paper is expected as a significant 
contribution to exhibit the impacts of machining factors on 
the cutting power and surface roughness as well as help 
the DM operators select the appropriate conditions. 
2. MATERIALS AND METHODS 
The systematic research procedure for experimental 
conductions and parameter optimization is depicted in Fig. 
1. The Box-Behnken method was applied instead of the full-
factorial in order to decrease the number of experiments 
and guarantee the predicting accuracy [8, 9]. Three 
machining parameters, including the spindle speed S, 
depth of cut ap, and feed rate fz with their levels were 
exhibited in Table 1. The parameter ranges were identified 
through machine tool characteristics as well as 
recommendations of cutting tool manufacturers and 
verified then using cutting trials. The output models 
considered of PC and Ra were developed with the aid of 
experimental data and RSM [10, 11]. A non-dominated 
sorting genetic algorithm II (NSGA-II) was used to solve the 
complicated problem with two objectives. In the NSGA-II, 
each objective parameter is treated separately. Standard 
genetic operation of mutation and crossover are performed 
on the designs. The selection process is based on two main 
mechanisms, including non-dominated and crowding 
distance sorting. By the end of the optimization run a 
Pareto set is constructed where each design has the best 
combination of objective values and improving one 
objective is impossible without sacrificing one or more of 
the other objectives. 
 Table 1. Machining parameters and their values 
Symbol Parameters level-1 level 0 level +1 
S Spindle speed (revolution/min) 2000 3000 4000 
ap Depth of cut (mm) 0.2 0.6 1.0 
fZ Feed per tooth (mm/tooth) 0.04 0.10 0.16 
Fig. 1. Optimizing procedure for cutting power and surface roughness 
(a) Tool holder specifications 
(b) Wiper insert dimensions 
(c) CNC machine and workpiece 
(d) Control unit and PC 
(e) Surface roughness measurement 
Fig. 2. Experimental facilities 
SCIENCE TECHNOLOGY 
Số 49.2018 ● Tạp chí KHOA HỌC & CÔNG NGHỆ 87
The dimensions of the rectangular SKD61 plate used 
were 350 mm×150 mm×25 mm in the experiments. The 
tool holder namely EPO07R012M12.0-02 mounting two 
wiper inserts AOMT 070204PDPR-MJ of Tungaloy 
Corporation was used to perform machining runs. The total 
length, effective length, and effective diameter of the tool 
holder are 68 mm, 18 mm, and 12 mm, respectively. The 
detail data of the tool holder and cutting insert can be 
found in Figs. 2a and 2b, respectively. A new insert was 
adopted for each machining experiment to eliminate any 
possible interference during the cutting process. 
Table 2. Experimental results 
No. S (rpm) ap (mm) fz (mm/tooth) Pc (kW) Ra (µm) 
1 4000 1.0 0.10 1.0695 0.94 
2 3000 0.6 0.10 0.7243 0.71 
3 3000 0.6 0.10 0.7319 0.73 
4 3000 0.6 0.10 0.7206 0.73 
5 3000 0.2 0.16 0.6196 1.13 
6 4000 0.6 0.04 0.5811 0.51 
7 3000 1.0 0.04 0.5848 0.93 
8 3000 0.6 0.10 0.7300 0.73 
9 3000 0.6 0.10 0.7187 0.73 
10 3000 1.0 0.16 0.9801 1.51 
11 2000 0.2 0.10 0.4752 0.73 
12 2000 1.0 0.10 0.6897 1.13 
13 2000 0.6 0.04 0.4554 0.7 
14 4000 0.2 0.10 0.6116 0.52 
15 2000 0.6 0.16 0.6724 1.29 
16 3000 0.2 0.04 0.4055 0.53 
17 4000 0.6 0.16 0.9947 0.99 
The experiments were performed dry condition along 
the direction of the width of the specimen. The machining 
tests were performed on a SPINNER milling machine having 
spindle speed of 20.000 RPM and spindle power of 22 kW 
(Fig. 2c). The cutting forces were measured using the quartz 
three-component dynamometer KISTLER 9257B with 
control unit 5233A. These amplified signals are the 
acquired by the personal computer through the acquisition 
card. DynoWare software was used to process these signals 
and expresses the three force components (Fig. 2d). The 
cutting power was calculated using the following equation: 
2 2 2
x y z cc c
c
F F F VF VP
60000 60000
 (1) 
where Pc is the cutting power (kW). Vc is the cutting 
speed (m/min). Fx, Fy, and Fz are the cutting forces in x, y, 
and z direction (N), respectively. 
The surface roughness values were measured by a 
tester Mitutoyo SJ-301. The average response values were 
observed from repeated three times at different positions 
(Fig. 2e). 
3. EXPERIMENTAL RESULTS 
In this paper, the significance of the models proposed 
and factors considered are evaluated using an analysis of 
variance (ANOVA). The confidence level of 95% was used 
and the factors with p-values less than 0.05 are considered 
as significant. The experimental results of the dry milling 
are given in Table 2. ANOVA results of the objective 
functions are presented in Table 3 and 4 respectively. 
As shown in Table 3, the R2 value of 0.9945 revealed that 
cutting power model was highly adequate to represent the 
experimental data. Additionally, the F-value of 141.79 
indicated that the second quadratic model is significant. As 
a result, the S, ap, fz, Sap, Sfz, apfz and fz^2 are significant terms. 
The percentage contribution of 35.57% revealed that fz is 
the most effective factor with regard to the single term. The 
percentages of S and ap are 21.52% and 33.99%, 
respectively. The insignificant terms (S^2, ap^2) were 
eliminated in the design space in order to save the 
computational costs and time. 
The ANOVA results of the surface roughness model are 
presented in Table 4. The R2 value of 0.9980 indicated that 
proposed model was significantly adequate to represent 
the experimental data. The surface roughness model is 
significant due to the p-value of less than 0.0001. For this 
model, the single terms (S, ap, fz), quadratic terms (S2, ap^2, 
fz^2), and the interaction term (Sfz) were considered as the 
significant terms. The interaction terms (Sap, apfz) were 
found to be insignificant model terms. Especially, fz is the 
most effective parameter due to the highest contribution 
(50.17%). The percentages of S and ap are 25.28% and 
8.28%, respectively. Additional, the percentages of fz^2, ap^2, 
and S2
were 10.08%, 5.74%, and 0.20%, respectively. 
Table 3. ANOVA results for cutting power 
Source Sum of Squares 
Mean 
Square 
F-value 
p-value 
 Remark 
Contri. 
(%) 
Model 0.540732 0.060081 141.7965 < 0.0001 Significant 
S 0.116215 0.116215 274.2754 < 0.0001 Significant 21.51 
ap 0.18368 0.18368 433.4976 < 0.0001 Significant 33.99 
fz 0.192207 0.192207 453.6237 < 0.0001 Significant 35.57 
Sap 0.014812 0.014812 34.95849 0.0006 Significant 2.74 
Sfz 0.009658 0.009658 22.79388 0.0020 Significant 1.79 
apfz 0.008215 0.008215 19.38913 0.0031 Significant 1.52 
S^2 0.000231 0.000231 0.544303 0.4846 Insignificant 0.04 
ap^2 0.001851 0.001851 4.368016 0.0750 Insignificant 0.34 
fz^2 0.013483 0.013483 31.822 0.0008 Significant 2.50 
R2 = 0.9945 
Table 4. ANOVA results for surface roughness 
Source Sum of Squares 
Mean 
Square 
F-value 
p-value 
 Remark 
Contri. 
(%) 
Model 1.258329 0.139814 392.2645 < 0.0001 Significant 
S 0.103513 0.103513 290.4158 < 0.0001 Significant 8.28 
 CÔNG NGHỆ 
 Tạp chí KHOA HỌC & CÔNG NGHỆ ● Số 49.2018 88
KHOA HỌC
ap 0.316013 0.316013 886.6082 < 0.0001 Significant 25.28 
fz 0.6272 0.6272 1759.679 < 0.0001 Significant 50.17 
Sap 0 0 0 1.0000 Insignificant 0.00 
Sfz 0.003025 0.003025 8.486974 0.0226 Significant 0.24 
apfz 2.5E-05 2.5E-05 0.07014 0.7988 Insignificant 0.00 
S^2 0.002527 0.002527 7.090813 0.0323 Significant 0.20 
ap^2 0.071706 0.071706 201.18 < 0.0001 Significant 5.74 
fz^2 0.126017 0.126017 353.5543 < 0.0001 Significant 10.08 
R2 = 0.9980 
The response models (cutting power, surface 
roughness) were developed in terms of input parameters 
using response surface methodology. From the 
experimental data, the coefficients of the regression 
equations are calculated. The regression coefficients of 
insignificant terms were eliminated based on ANOVA 
results. Consequently, the regression response surface 
models showing the cutting power (Pc) and surface 
roughness (Ra) are expressed as follows: 
Pc = 0.37294-0.000097S-0.10917ap+2.13734fz 
 +0.000152Sap+0.000819Sfz+1.88832apfz 
 -0.13104ap
2
-15.71919fz
2
(2) 
Ra= 0.71039+0.000079S-0.47146ap-3.50694fz 
 -0.000000025S
2
+0.81563ap
2
+48.05556fz
2
(3) 
(a) For cutting power 
(b) For surface roughness 
Fig. 3. Pareto chart 
To confirm the analyzed results, the Pareto charts of all 
terms were generated based on the F-values. The aim of 
the Pareto charts is to rank in descending order the effects 
of the machining parameters and their interactions on the 
technological outputs. The Pareto charts of Pc and Ra were 
shown in Fig. 3a and 3b, respectively. It can be stated that 
the Pareto charts are similar to the ANOVA results. 
The effects of process parameters on the responses 
were investigated using the contour plots. Figs. 4a and 4b 
showed that an increase of the spindle speed, depth of cut, 
and feed rate results in a higher cutting power. This 
phenomenon can be explained as follows. Increasing ap or 
fz increased the material removal volume in the same unit 
of time, thus resulting in a higher cutting force or power 
consumed. An improved spindle speed causes an increased 
cutting speed and a higher cutting power is observed. 
Fig. 4c and 4d exhibited that the surface roughness was 
also decreased with an increment of S. A reduction of 
cutting force can be observed at the higher spindle speed, 
resulting in a smoother surface. An increased cutting force 
or cutting power caused by a higher depth cut or feed rate 
results in a coarser surface roughness. 
(a) Pc versus S and ap 
(b) Pc versus fz and ap 
SCIENCE TECHNOLOGY 
Số 49.2018 ● Tạp chí KHOA HỌC & CÔNG NGHỆ 89
(c) Ra versus S and ap 
(d) Ra versus fz and ap 
Fig. 4. Interaction plots for machining responses 
4. OPTIMIZATION RESULTS 
As a result, the inputs, including S, ap, and fz have 
complicated effects on the technological parameters, 
including cutting power and surface roughness. The 
optimizing issue can be described as follows: 
Find X = [S, ap, fz] 
Minimize cutting power Pc and surface roughness Ra 
Constraints: 
2000 ≤ S ≤ 4000 (revolution/min), 0.2 ≤ ap ≤ 1.0 (mm), 
0.04 ≤ fz ≤ 0.16 (mm/tooth). 
After building the statistical regression equations 
showing the relationship between process parameters and 
machining responses, these equations are used to find 
optimal parameters. The optimal parameters of the multi-
objective optimization are selected from the Pareto front. 
The Pareto front generated by the NSGA-II algorithm was 
exhibited in Fig. 5, in which the blue points are feasible 
solutions. The optimal solution is determined as a blue 
point with the red crossed line. The optimal values of 
design variables and objective functions were presented in 
Table 5. 
Table 5. Optimal values of process parameters and responses 
Parameters S (rpm) 
ap 
(revolution/min) 
fz 
(mm/tooth) 
Pc 
(kW) Ra (µm) 
Optimal 
values 
3996 0.2 0.04 0.4057 0.43 
Initial values 3000 0.6 0.10 0.7243 0.71 
- 44% 
- 
39.44% 
Fig. 5. Pareto front for selecting optimal values 
As compared to initial values in the Table 5, the cutting 
power Pc is decreased approximately 44% and surface 
roughness Ra is reduced around 39.44%. 
5. CONCLUSIONS 
This work addressed the process parameters 
optimization of the dry milling for low cutting power as 
well as surface roughness. A hybrid approach combining 
machining experiments, RSM model, and NSGA-II was 
proposed in order to develop predictive models and 
determine the optimal values. An ANOVA analysis was 
performed to evaluate the model adequacy and factor 
significance. The main conclusions from the research 
results of this work can be drawn as follows within 
parameter ranges: 
1. The low process parameters were commented to 
decrease the cutting power, in which depth of cut and feed 
rate have the higher contribution, compared to the spindle 
speed. 
2. The surface roughness values decrease with increased 
spindle speed and increase with higher depth of cut and 
feed rate. 
3. The optimizing issue, in which the cutting power and 
surface roughness is practical and realistic in the dry milling 
processes, compared to single objective optimization (i.e. 
Minimizing surface roughness). 
 CÔNG NGHỆ 
 Tạp chí KHOA HỌC & CÔNG NGHỆ ● Số 49.2018 90
KHOA HỌC
This work is expected as a significant contribution to 
improve the dry milling efficiency (i.e. Low energy 
consumed and surface roughness). The holistic 
optimization considering more objectives, such as material 
removal rate and tool wear will be addressed in the future 
work. 
ACKNOWLEDGMENT 
This research is funded by Vietnam National Foundation 
for Science and Technology Development (NAFOSTED) 
under grant number 107.04-2017.06 
REFERENCES 
[1]. Viet Nam Ennergy Sector, Assessment, Strategy, and road map, 
https://www.adb.org/documents/viet-nam-energy-sector-assessment-
strategy-and-road-map, 2015. 
[2]. S.H. Hu, 2012. Energy consumption characteristics of multiple-component 
of modern cnc machine tools. Chongqing University. 
[3]. Gutowski T.G., Branham M.S., Dahmus J.B., Jones A.J., Thiriez A., Sekulic 
D.P., 2009. Thermodynamic analysis of resources used in manufacturing processes. 
Environmental Science and Technology, 43 (5), pp. 1584-1590. 
[4]. ISO/TC 39/SC, Machine tools-environmental evaluation of machine 
tools-part 1: design methodology for energy-efficient machine tools, Draft 
International Standard ISO/DIS 14955-1, 2012. 
[5].  
[6]. Zhongde, S., Shaoyan, Q., Qian, L., Feng, L., 2012. Key manufacturing 
technology & equipment for energy saving and emissions reduction in mechanical 
equipment industry. International Journal of Precision Engineering and 
Manufacturing 13(7), pp. 1095-1100. 
[7]. Pusavec, F., Krajnik, P., Kopac, J., 2010. Transitioning to sustainable 
production-part I: application on machining technologies. Journal of Cleaner 
Production 18 (2), pp. 174-184. 
[8]. C.F. Jeff, R. C. Z., Y. Lan, 2009. Experiment Design & Analyses and 
Parameters Optimization. Wiley. 
[9]. H. S. Park, T. T. Nguyen, X. P. Dang, 2016. Multi-objective optimization of 
turning process of hardened material for energy efficiency. International Journal of 
Precision Engineering and Manufacturing 17 (12), pp. 1623-1631. 
[10]. H. S. Park; T. T. Nguyen, 2017. Development of a new staking process for 
an automotive part. The International Journal of Advanced Manufacturing 
Technology 89 (1-4), pp. 1053-1068. 
[11]. H. S. Park, T. T. Nguyen, 2014. Optimization of injection molding process 
for car fender in consideration of energy efficiency and product quality. Journal of 
Computational Design and Engineering Vol. 1, No. 4, 256-265.