Anda di halaman 1dari 10

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817

journal homepage: www.elsevier.com/locate/jmatprotec

Grey relational analysis coupled with principal component


analysis for optimization design of the cutting parameters
in high-speed end milling

H.S. Lu a,∗ , C.K. Chang b , N.C. Hwang a , C.T. Chung c


a Department of Mechanical and Computer-Aided Engineering, National Formosa University, 64, Wun-Hun Road,
Huwei, Yunlin 632, Taiwan
b Institute Engineering Science and Technology, National Kaohsiung First University of Science and Technology, 2, Juoyue Road, Nantz

District, Kaohsiung 811, Taiwan


c Department of Vehicle Engineering, National Formosa University, 64, Wun-hun Road, Huwei, Yunlin 632, Taiwan

a r t i c l e i n f o a b s t r a c t

Article history: This paper investigates optimization design of the cutting parameters for rough cutting pro-
Received 28 December 2007 cesses in high-speed end milling on SKD61 tool steel. The major characteristics indexes for
Received in revised form performance selected to evaluate the processes are tool life and metal removal rate, and
5 August 2008 the corresponding cutting parameters are milling type, spindle speed, feed per tooth, radial
Accepted 30 August 2008 depth of cut, and axial depth of cut. In this study, the process is intrinsically with mul-
tiple performance indexes so that grey relational analysis that uses grey relational grade
as performance index is specially adopted to determine the optimal combination of cutting
Keywords: parameters. Moreover, the principal component analysis is applied to evaluate the weighting
High-speed end milling values corresponding to various performance characteristics so that their relative impor-
Grey relational analysis tance can be properly and objectively described. The results of confirmation experiments
Principal component analysis reveal that grey relational analysis coupled with principal component analysis can effec-
Optimization tively acquire the optimal combination of cutting parameters. Hence, this confirms that the
proposed approach in this study can be an useful tool to improve the cutting performance
of rough cutting processes in high-speed end milling process.
© 2008 Elsevier B.V. All rights reserved.

1. Introduction rough cutting processes in high-speed end milling for SKD61


tool steels (40HRC).
The development of technologies for cutting and machining The main features of high-speed end milling are high spin-
tools has been advanced into a stage of considerably high cut- dle rotational speed, low depth of cut and high feed rate. This
ting speeds and feed rates. Nowadays, the terms “high speed can favorably enhance metal removal rates (Destmobes, 1984),
machining” (HSM) and “high speed cutting” (HSC) are widely and, thereby, shorten production time and lower manufactur-
used especially in the field of the high-speed end milling ing cost. Furthermore, the relatively larger portion (at least
(Dewes and Aspinwall, 1997). Followed by this trend, this paper 85–90%) of the heat generated during HSM is carried away by
is focused on the optimal design of cutting parameters of the chips when compared with conventional machining. This


Corresponding author. Tel.: +886 5 6315324; fax: +886 5 6315310.
E-mail address: Lubs@sunws.nfu.edu.tw (H.S. Lu).
0924-0136/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmatprotec.2008.08.030
j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817 3809

can desirably bring out low temperature of workpieces, low qualities. Lin and Lin (2002) combined orthogonal array and
cutting forces and minimal distortion for workpieces (Chaplin grey relational analysis to optimize the electrical discharge
et al., 1981), the capability of machining thin walled selections machining process with multiple performance characteris-
and the use of simple fixture (Harding, 1993). On the con- tics. The results showed that the problems of multiple quality
trary, some drawbacks have also been identified such as high characteristics can be effectively solved. Narender Singh et
tool wear as well as requiring expensive materials for tools al. (2004) adopted Grey-Relational Analysis to investigate the
(Kramer, 1984), balanced tooling, high precision tool-holder EDM parameters on machining Al–10% SiCp composites. Kao
tapers (Lewis, 1992), expensive spindles with low spindle life and Hocheng (2003) selected surface roughness and passiva-
(Society of British Aerospace Companies Ltd, 1992), higher cost tion strength as performance indexes to optimize the process
for machine tools, and sophisticated control systems (Smith, parameters on electrochemical polishing of stainless steel.
1992). However, when calculating the values of grey relational grade,
HSM was initially applied in the aerospace industry for most of the researchers determine the weighting values of
machining workpieces of light alloys, notably aluminum. In various performance characteristics based on their own sub-
recent years, this technology has begun to be utilized in jective estimation. This approach cannot objectively reveal
mould and die industries for the production of components, the relative importance for various performance characteris-
including those manufactured from hardened tool steels. It is tics. The research of Jean and Peng et al. (Jean and Tsai, 2004;
foreseeable that HSM will be a widely used machining process Peng and Kirk, 1999) about the optimization problem with
in manufacturing. Hence, it is very important to determine multiple performance characteristics using grey relational
the optimal cutting parameters for improving cutting perfor- analysis presents a remedy by calculating the correspond-
mance of HSM. ing weighting values using fuzzy logics. In addition, this
High-speed end milling can be classified as rough cutting paper specially introduces a desirable solution for the above
and finishing cutting based on their types of processes. The problem—grey relational analysis for optimizing combination
major issue of rough cutting is to quickly remove cut metal of cutting parameters and principal component analysis for
residuals so that both processing time and production cost can determining the corresponding weighting values of various
be reduced. Generally, two important indexes to evaluate the performance characteristics.
performance of rough cutting process are tool life and metal Principal Component Analysis (PCA) was proposed by
removal rate. Therefore, this kind of processes is regarded as a Pearson in 1901, and evolved as a statistical tool by Hotelling
multiple-performance problem. Basically, metal removal rate in 1993. Its main advantage is significantly alleviating loading
and tool life are correlated with the cutting parameters such and complexion of information by simplifying several cor-
as milling type, spindle speed, feed per tooth, radial depth of related variables into fewer uncorrelated and independent
cut, and axial depth of cut. Usually, the desired cutting param- principal components, and preserving as much original infor-
eters are determined based on experience or handbook values. mation as possible using linear combination. In recent times,
However, this does not ensure whether a truly or nearly opti- PCA has gradually become an analytical tool for the optimiza-
mal result of cutting performance for that particular machine tion of a system with multiple performance characteristics
and environment can be obtained or not. Therefore, it is highly (Fung and Kang, 2005; Liao, 2006; Tong et al., 2005).
expected for industry to search for a simple and effective The context is organized in the following manner: The
experiment method that can be used to quickly optimize analysis method and the experimental design are described
the cutting parameters. In this paper, the proposed method first. Then, the optimization of the high-speed end milling
is grey relational analysis associated with principal compo- process based on the grey relational analysis coupled with
nent analysis to study the optimization problem of the cutting principal component analysis is presented in detail. Finally,
parameters in high-speed end milling on SKD61 tool steel. the paper is concluded with a summary of this study.
The grey system theory proposed by Deng in 1982 has
been proven to be useful for dealing with the problems
with poor, insufficient, and uncertain information. The grey 2. Analysis method
relational analysis based on this theory can further be effec-
tively adopted for solving the complicated interrelationship 2.1. Signal-to-noise ratio
among the designated performance characteristics. Through
this analysis, a grey relational grade is favorably defined as an Taguchi method is one of the simple and effective solutions
indicator of the multiple performance characteristics for eval- for parameter design and experimental planning (Ross, 1988).
uation. In recent years, grey relational analysis has become In this method, signal-to-noise (S/N) ratio is used to repre-
a powerful tool to analyze the processes with multiple per- sent a performance characteristic and the largest value of S/N
formance characteristics. Pan et al. (2007) combined Taguchi ratio is required. There are three types of S/N ratio—the lower-
method with Grey-Relational Analysis to investigate the opti- the-better, the higher-the-better, and the nominal-the-better.
mal design of cutting parameters for Nd:YAG laser welding The S/N ratio with a lower-the-better characteristic that can
titanium alloy plates. Lin and Lin (2006) applied Grey theory to be expressed as:
evaluate the tribological performance of the a-C:H(N) coating
films prepared by differing the nitrogen content and the film ⎛ ⎞
thickness. Tarng et al. (2002) utilized the grey-based Taguchi 
n
ij = −10 log ⎝ n1 y2ij ⎠ (1)
methods to optimize the process parameters of the submerged
j=1
arc welding (SAW) in hardfacing, considering multiple weld
3810 j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817

The S/N ratio with a higher-the-better characteristic can be comparability 18 sequences xi∗ (k), where i = 1, 2, . . ., m and k = 1,
expressed as: 2, . . ., n with m = 18 and n = 3 in this study.
⎛ ⎞ The grey relational grade is a weighting-sum of the grey

n relational coefficients. It is defined as follows:
ij = −10 log ⎝ n1 1 ⎠ (2)
y2
j=1
ij

n
 (xo∗ , xi∗ ) = ˇk (xo∗ (k), xi∗ (k)) (6)
k=1
The S/N ratio with a nominal-the-better characteristic can
be expressed as:
where ˇk represents the weighting value of the kth per-
⎛ ⎞ n
formance characteristic, and ˇ = 1 In this study, the
k=1 k

n
ij = −10 log ⎝ ns y2ij ⎠
1 corresponding weighting values are obtained from the prin-
(3)
cipal component analysis.
j=1
Here, the grey relational grade  (xo∗ , xi∗ ) represents the
level of correlation between the reference sequence and the
where yij is the ith experiment at the jth test, n is the total
comparability sequence. If the two sequences are identically
number of the tests, and s is the standard deviation.
coincidence, then the value of grey relational grade is equal
to 1. The grey relational grade also indicates the degree of
2.2. Grey relational analysis
influence that the comparability sequences could exert over
the reference sequence. Therefore, if a particular comparabil-
2.2.1. Data preprocessing
ity sequence is more important than the other comparability
In the grey relational analysis, when the range of the sequence
sequences to the reference sequence, then the grey rela-
is large or the standard value is enormous, the function of
tional grade for that comparability sequence will be higher
factors is neglected. However, if the factors goals and direc-
than other grey relational grades. Grey analysis is actually a
tions are different, the grey relational might produce incorrect
measurement of absolute value of data difference between
results. Therefore, one has to preprocess the data which are
sequences, and it could be used to measure approximation
related to a group of sequences, which is called “grey relational
correlation between sequences.
generation” (Deng, 1989).
A linear date preprocessing method for the S/N ratio can
be expressed as: 2.3. Principal component analysis

(o) (o) Pearson and Hotelling initially developed PCA to explain the
xi (k) − min xi (k)
xi∗ (k) = (4) structure of variance-convariance by way of the linear com-
(o) (o)
max xi (k) − min xi (k) binations of each quality characteristic. The procedure is
described as follows (Fung and Kang, 2005):
(o)
where xi∗ (k) is the sequence after the data processing; xi (k)
is the original sequence of S/N ratio, where i = 1, 2, . . ., m and
(o) (o) (o) 1. The original multiple quality characteristic array
k = 1, 2, . . ., n; max xi (k) is the largest value of xi (k); min xi (k)
(o)
is the smallest value of xi (k). xi (j), i = 1, 2, . . . , m; j = 1, 2, . . . , n

2.2.2. Grey relational coefficient and grey relational grade


After data processing is carried out, a grey relational coeffi- ⎡ ⎤
x1 (1) x1 (2) ··· ··· x1 (n)
cient can be calculated with the preprocessed sequences. The
⎢ x2 (1) x2 (2) ··· ··· x2 (n) ⎥
grey relational coefficient is defined as follows: X=⎢ . .. ⎥ (7)
⎣ . .. ⎦
. . ··· ··· .
min +  · max xm (1) xm (2) ··· ··· xm (n)
(xo∗ (k), xi∗ (k)) = (5)
oi (k) +  · max

where m is the number of experiment and n is the num-


where oi (k) is the deviation sequence of the reference
ber of the quality characteristic. In this paper, x is the grey
sequence xo∗ (k) and the comparability xi∗ (k), i.e. oi (k) = |xo∗ (k) −
relational coefficient of each quality characteristic. In this
xi∗ (k)| is the absolute value of the difference between xo∗ (k) and
paper, m = 18, n = 3.
xi∗ (k),
2. Correlation coefficient array
min = minminoi (k) The correlation coefficient array is evaluated as follows:
∀i ∀k
 
Cov(xi (j), xi (l))
max = maxmaxoi (k) Rjl = , j = 1, 2, . . . , n, l = 1, 2, . . . , n (8)
∀i ∀k xi (j) × xi (l)
: distinguishing coefficient,  ∈ [0,1].  is set as 0.5 in this
study. where Cov(xi (j), xi (l)):the convariance of sequences xi (j) and
The purpose of defining this coefficient is to show the rela- xi (l); xi (j):the standard deviation of sequence xi (j); xi (l):the
tional degree between the reference sequence xo∗ (k) and the standard deviation of sequence xi (l).
j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817 3811

3. Determining the eigenvalues and eigenvectors indexes coupled with metal removal rate (MRR) are defined
The eigenvalues and eigenvectors are determined from the respectively as
correlation coefficient array,
average width of flank wear of end cutting edge
TWR1 = (11)
(R − k Im )Vik = 0 (9) cutting time

n TWR2
where k eigenvalues,  = n, k = 1, 2, . . . , n; Vik =
k=1 k
[ak1 ak2 . . . akn ]T : eigenvectors corresponding to the eigen- average width of flank wear of peripheral cutting edge
=
value k . cutting time
4. Principal components (12)
The uncorrelated principal component is formulated as:
total metal removal volumes

n MRR = (13)
Ymk = xm (i) · Vik (10) cutting time

i=1
The cutting performance of rough cutting processes in
where Ym1 is called the first principal component, Ym2 is end milling operations is closely correlated with the cutting
called the second principal component and so on. parameters such as milling type, spindle speed, feed per tooth,
axial depth of cut, and radial depth of cut, that are regarded as
The principal components are aligned in descending order controllable factors in the study. The selected cutting param-
with respect to variance, and therefore the first principal com- eters and their assigned values of levels are listed in Table 1.
ponent Ym1 accounts for most variance in the data. The selection of the orthogonal array is concerned with the
total degree of freedom of the cutting parameters. The degree
of freedom is defined as the number of comparisons among
3. Experimental apparatus the process parameters required to optimize the parameters.
In the present study, there are 4 cutting parameters with 3
The experiments were carried out on a Papars B8 HSM center levels and 1 cutting parameter with 2 levels. By neglecting the
with maximum spindle speed of 22,000 rpm, maximum feed interaction among the cutting parameters, the total degree of
rate of 30 m/min, and spindle power of 20 hp. freedom is 9.
The test workpieces are made of SKD tool steel and Once the degree of freedom is known, the next step is
200 mm × 80 mm × 80 mm in size. The hardness of the work- to select an appropriate orthogonal array. The degree of
piece material was measured to be 40HRC. TopGreen freedom for the orthogonal array should be greater than or
AAE-SS4RD0880R050 tools were utilized in the experiments. at least equal to that of the process parameters. Thereby, a
The cutting tools are made of tungsten carbide and coated L18 orthogonal array is considered. The experimental layout
with TiAlN under a major specification, such as diameter of is shown in Table 2.
8 mm, 4 flutes, helix angle of 30◦ , and corner radius of 0.5 mm.
4.2. Experimental results

4. Experimental design and results The experiments were carried out on a Papers B8 HSM cen-
ter with air blow. Based on the experimental layout depicted
4.1. Experimental design in Table 2, the experiments are performed in order and each
specific experiment is repeated three times. The procedures
For rough cutting processes in high-speed end milling opera- are described as follows:
tions, tool life and metal removal rate are selected as indexes
to evaluate cutting performance. In practice, the longer tool (1) Tools are sintered with its holder using a high-frequency
life is, the smaller tool wear rate is. Therefore, the former is sintering machine. The overhang length of the tools is
favorably replaced by the latter for convenience of calculation. selected at a fixed value of 38 mm.
Furthermore, tool wear rate can be divided into two indexes: (2) An dynamic-balance meter is used to measure the offset
one is flank wear rate of end cutting edge (TWR1) and the other of tools. The tool dynamic under test is required to be less
is flank wear rate of peripheral cutting edge (TWR2). These two than 0.02 G.

Table 1 – Cutting parameters and their levels


Symbol Cutting parameter Level 1 Level 2 Level 3

A Milling type Down milling* Up milling


V Spindle speed (rpm) 12,000* 16,000 20,000
F Feed per tooth (mm/t) 0.04 0.07* 0.10
Da Axial depth of cut (mm) 0.60* 0.80 1.00
Dr Radial depth of cut (mm) 0.60* 0.80 1.00


Initial cutting conditions.
3812 j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817

Table 2 – Experimental layout


No. A V (rpm) F (mm/t) Da (mm) Dr (mm)

1 Down milling 12,000 0.04 0.60 0.60


2 Down milling 12,000 0.07 0.80 0.80
3 Down milling 12,000 0.10 1.00 1.00
4 Down milling 16,000 0.04 0.80 1.00
5 Down milling 16,000 0.07 1.00 0.60
6 Down milling 16,000 0.10 0.60 0.80
7 Down milling 20,000 0.04 1.00 0.80
8 Down milling 20,000 0.07 0.60 1.00
9 Down milling 20,000 0.10 0.80 0.60
10 Up milling 12,000 0.04 0.80 0.80
11 Up milling 12,000 0.07 1.00 1.00
12 Up milling 12,000 0.10 0.60 0.60
13 Up milling 16,000 0.04 0.60 1.00
14 Up milling 16,000 0.07 0.80 0.60
15 Up milling 16,000 0.10 1.00 0.80
16 Up milling 20,000 0.04 1.00 0.60
17 Up milling 20,000 0.07 0.60 0.80
18 Up milling 20,000 0.10 0.80 1.00

(3) The workpieces of SKD61 tool steel are mounted on the speed end milling operation is described step by step as
Papars B8 HSM center and continuously cut for a maxi- follow:
mum length of 160 m.
(1) Convert the experimental data into S/N values.
(2) Normalize the S/N ratio.
As the experiments are finished, the cutting time is (3) Calculate the corresponding grey relational coefficients.
counted and the flank wear of end cutting edge as well as that (4) Calculate the grey relational grade using principal compo-
of peripheral cutting edge are measured by placing the tool nent analysis.
at a microscopic of tool head with magnification of 100. The (5) Perform statistical analysis of variance (ANOVA).
features of the flank wear land on the end-mill are shown in (6) Select the optimal levels of cutting parameters.
Fig. 1. The average flank wear of end cutting edge w1 is cal- (7) Conduct confirmation experiments.
culated by averaging A, B, C, and D on the end cutting edges.
The average flank wear of peripheral cutting edge w2 is cal- 5.1. Optimal combination of the cutting parameters
culated by averaging E, F, G, and H on the peripheral cutting
edge. Finally, the values of flank wear rate and metal removal The performance characteristics obtained from the experi-
rate are calculated based on Eqs. (11)–(13) the results are listed mental results are initially converted into S/N ratio to search
in Table 3. for a desirable result with the best performance and the
smallest variance. Generally, the longer tool life, the higher
metal removal rate and the better cutting performance. Thus,
5. Analysis and discussion metal removal rate is of higher-the-better type. As for tool
wear rate, it can be clearly recognized as one of lower-
The algorithm of grey relational analysis coupled with prin- the-better type. The experimental results listed in Table 3
cipal analysis to determine the optimal combinations of are substituted into Eqs. (1) and (2) to calculate the S/N
the cutting parameters for rough cutting process in high- ratios of flank wear rate and metal removal rate shown in

Fig. 1 – Features of the flank wear land on the end-mill. Average width of flank wear of end cutting edge
w1 = (A + B + C + D)/4, average width of flank wear of peripheral cutting edge w2 = (E + F + G + H)/4
j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817 3813

Table 3 – Experimental results


No. Cutting time Flank wear rate (mm/min) × 10−4 Metal removal
(min) rate (mm3 /s)
End cutting edge Peripheral cutting edge

Measured 1 Measured 2 Measured 3 Measured 1 Measured 2 Measured 3

1 83.13 6.3490 6.3540 6.3500 2.4724 2.4356 2.4536 11.52


2 47.62 9.5534 9.7662 9.6508 4.4387 4.5071 4.4730 35.84
3 33.33 14.9879 15.2551 15.1200 7.9987 8.7429 8.3698 80.00
4 56.25 7.5821 7.5645 7.5698 3.9745 4.1321 4.0503 34.13
5 35.63 11.5594 11.7356 11.6485 5.2139 5.1707 5.1918 44.80
6 25.00 14.5938 14.6862 14.6398 8.8984 9.3416 9.1408 51.20
7 31.25 10.0211 11.0989 10.5600 5.0324 4.5676 4.8090 42.67
8 17.14 18.9877 19.9789 19.5005 9.2743 9.3955 9.3450 56.00
9 19.95 23.1032 22.5108 22.8070 11.4027 11.2539 11.3503 64.00
10 83.33 6.7538 6.5428 6.6498 3.7104 3.5858 3.6519 20.48
11 47.62 9.5787 9.7409 9.6608 5.5879 5.7519 5.6708 56.00
12 22.17 18.6124 18.1940 18.4030 9.4210 9.0724 9.2477 28.80
13 56.25 8.2318 7.2704 8.2842 4.4003 4.3285 5.3025 25.60
14 35.71 11.4387 11.6921 11.5655 5.9981 6.5475 6.2808 35.84
15 25.00 16.6739 16.1261 16.3998 12.8364 12.6036 12.7199 85.33
16 33.25 15.2158 14.5356 13.1960 9.6925 8.7895 8.8834 32.00
17 21.43 19.1469 18.7031 22.0660 11.6274 11.4588 11.2116 44.80
18 6.00 57.9398 59.7268 58.8303 36.7897 36.5437 36.6657 106.67

Table 4. Usually, the larger S/N ratio, the better performance


characteristic. o1 (2) = |xo∗ (2) − x1∗ (2)| = |1.0000 − 1.0000| = 0.0000,
All the original sequences of S/N ratio in Table 4 are then
substituted into Eq. (4) to be normalized. The outcomes are o1 (3) = |xo∗ (3) − x1∗ (3)| = |1.0000 − 0.0000| = 1.0000,
shown in Table 5 and denoted as xo∗ (k) and xi∗ (k) for reference
therefore o1 = (0.0000, 0.0000, 1.0000).
sequence and comparability sequence respectively. According
The same calculating method was performed for i = 1–18,
to Deng (1982), larger values of the normalized results corre-
and the results of all oi for i = 1–18 are listed in Table 6.
sponds to better performance, and the maximum normalized
Investigating the data presented in Table 6, we can find that
results that are equal to 1 indicates the best performance.
max (k) and min (k) are as follows:
According to Table 5, the deviation sequences o1 can be
calculated as follows: max = o18 (1) = o18 (2) = o1 (3) = 1.000,

o1 (1) = |xo∗ (1) − x1∗ (1)| = |1.0000 − 1.0000| = 0.0000,

Table 5 – The sequences after data preprocessing


No. Flank wear rate Metal
Table 4 – The sequences of S/N ratio removal rate
End cutting Peripheral
No. Flank wear rate Metal edge cutting edge
removal rate
End cutting Peripheral Reference sequence 1.0000 1.0000 1.0000
edge cutting edge
Comparability sequence
1 63.94 72.20 21.23 1 1.0000 1.0000 0.0000
2 60.30 66.99 31.09 2 0.8117 0.7782 0.5101
3 56.41 61.54 38.06 3 0.6105 0.5462 0.8707
4 62.42 67.84 30.66 4 0.9214 0.8144 0.4878
5 58.67 65.69 33.03 5 0.7274 0.7229 0.6105
6 56.69 60.79 34.19 6 0.6249 0.5143 0.6705
7 59.52 66.36 32.60 7 0.7713 0.7514 0.5882
8 54.20 60.59 34.96 8 0.4961 0.5057 0.7103
9 52.84 58.91 36.12 9 0.4258 0.4342 0.7703
10 63.55 68.75 26.23 10 0.9793 0.8531 0.2587
11 60.30 64.93 34.96 11 0.8117 0.6905 0.7103
12 54.70 60.68 29.19 12 0.5220 0.5096 0.4118
13 62.00 66.56 28.16 13 0.8996 0.7599 0.3585
14 58.74 64.05 31.09 14 0.7310 0.6526 0.5101
15 55.70 57.91 38.62 15 0.5737 0.3917 0.8996
16 56.87 60.79 30.10 16 0.6342 0.5143 0.4589
17 53.97 58.84 33.03 17 0.4842 0.4312 0.6105
18 44.61 48.71 40.56 18 0.0000 0.0000 1.0000
3814 j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817

Table 6 – The deviation sequences Table 8 – The Eigenvalues and explained variation for
principal components
Deviation sequences oi (1) oi (2) oi (3)
Principal component Eigenvalue Explained
No. 1, i = 1 0.0000 0.0000 1.0000
variation (%)
No. 2, i = 2 0.1883 0.2218 0.4899
No. 3, i = 3 0.3895 0.4538 0.1293 First 2.634 87.79
No. 4, i = 4 0.0786 0.1856 0.5122 Second 0.315 10.49
No. 5, i = 5 0.2726 0.2771 0.3895 Third 0.051 1.72
No. 6, i = 6 0.3751 0.4857 0.3295
No. 7, i = 7 0.2287 0.2486 0.4118
No. 8, i = 8 0.5039 0.4943 0.2897
No. 9, i = 9 0.5742 0.5658 0.2297
No. 10, i = 10 0.0207 0.1469 0.7413
No. 11, i = 11 0.1883 0.3095 0.2897
In order to objectively reflect the relative importance for
No. 12, i = 12 0.4780 0.4904 0.5882 each performance characteristic in grey relational analysis,
No. 13, i = 13 0.1004 0.2401 0.6415 principal component analysis is specially introduced here
No. 14, i = 14 0.2690 0.3474 0.4889 to determine the corresponding weighting values for each
No. 15, i = 15 0.4263 0.6083 0.1004 performance characteristic. The elements of the array for
No. 16, i = 16 0.3658 0.4857 0.5411
multiple performance characteristics listed in Table 7 rep-
No. 17, i = 17 0.5158 0.5688 0.3895
resent the grey relational coefficient of each performance
No. 18, i = 18 1.0000 1.0000 0.0000
characteristic. These data are used to evaluate the corre-
lation coefficient matrix, and determine the corresponding
eigenvalues from Eq. (9), shown in Table 8. The eigenvector
min = o1 (1) = o1 (2) = o18 (3) = 0.000.
corresponding to each eigenvalue is listed in Table 9, and
According to Table 6 and Eq. (5), the grey relational coeffi- its square can represent the contribution of the correspond-
cient (xo∗ (k), x1∗ (k)) can be calculated as follows: ing performance characteristic to the principal component.
0.0000 + 0.5 × 1.0000 Table 10 shows that the contributions of flank wear rate of
(xo∗ (1), x1∗ (1)) = = 1.0000, end cutting edge, flank wear rate of peripheral cutting edge,
0.0000 + 0.5 × 1.0000
and metal removal rate are indicated as 0.3502, 0.3542, and
0.0000 + 0.5 × 1.0000 0.2956. Moreover, the variance contribution for the first prin-
(xo∗ (2), x1∗ (2)) = = 1.0000,
0.0000 + 0.5 × 1.0000 cipal component characterizing the whole original variables,
i.e. the three performance characteristics, is as high as 87.79%.
0.0000 + 0.5 × 1.0000 Hence, for this study, the squares of its corresponding eigen-
(xo∗ (3), x1∗ (3)) = = 0.3333,
1.0000 + 0.5 × 1.0000 vectors are selected as the weighting values of the related
thus (xo∗ (k), x1∗ (k)) = (1.0000, 1.0000, 0.3333), k = 1–3. performance characteristic, and the coefficients ˇ1 , ˇ2 , ˇ3 in
Similar procedure is applied for i = 1–18. Table 7 shows the Eq. (6) are thereby set as 0.3502, 0.3542, and 0.2956, respec-
grey relational coefficient for all 18 comparability sequences. tively. Based on Eq. (6) and the data listed in Table 7, the grey

Table 7 – The calculated grey relational coefficient and grey relational grade for 18 comparability sequences
No. (comparability sequences) Grey relational coefficient Grey relational grade

Flank wear rate Metal removal rate

End cutting edge Peripheral cutting edge

1 1.0000 1.0000 0.3333 0.8029


2 0.7264 0.6927 0.5051 0.6491
3 0.5621 0.5242 0.7945 0.6174
4 0.8642 0.7293 0.4940 0.7070
5 0.6472 0.6434 0.5621 0.6207
6 0.5714 0.5073 0.6028 0.5580
7 0.6862 0.6679 0.5484 0.6390
8 0.4981 0.5029 0.6332 0.5397
9 0.4655 0.4691 0.6852 0.5317
10 0.9602 0.7729 0.4028 0.7291
11 0.7264 0.6177 0.6332 0.6604
12 0.5112 0.5048 0.4595 0.4937
13 0.8328 0.6756 0.4380 0.6304
14 0.6502 0.5900 0.5051 0.5860
15 0.5398 0.4511 0.5328 0.5950
16 0.5775 0.5073 0.4803 0.5239
17 0.4922 0.4678 0.5621 0.5042
18 0.3333 0.3333 1.0000 0.5304
j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817 3815

Table 9 – The Eigenvectors for principal components


Quality characteristic Eigenvector

First principal Second principal Third principal


component component component

Flank wear rate of end cutting edge 0.5918 0.4093 −0.6945


Flank wear rate of peripheral cutting edge 0.5951 0.3592 0.7189
Metal removal rate −0.5437 0.8387 -0.0310

No. 10, No. 11, and No. 12 were the experimental runs at which
Table 10 – The contribution of each individual quality
characteristic for the fist principal component cutting parameter V was set at level 1. The associated values
of grey relational grade for V1 are those experimental runs’
Quality characteristic Contribution
grey relational grades. Therefore, their average is the average
Flank wear rate of end cutting edge 0.3502 grey relational grade for V1 :
Flank wear rate of peripheral cutting edge 0.3542
Metal removal rate 0.2956
V 1 = 0.8029 + 0.6491 + 0.6174 + 0.7291 + 0.6604 + 0.4937

= 0.6587,
relational grade  (xo∗ , x1∗ ) can be calculated as follows:

 (xo∗ , x1∗ ) = 0.3502 × 1.000 + 0.3542 × 1.000 + 0.2956 × 0.3333 V 2 = 0.7070 + 0.6207 + 0.5580 + 0.6304 + 0.5860 + 0.5950
= 0.8029 = 0.6162,

Using the same procedure, the grey relational grade of com-


parability sequence for i = 1–18 can also be obtained and shown
V 3 = 0.5580 + 0.6390 + 0.5397 + 0.5239 + 0.5042 + 0.5304
in Table 7. Thereby, the optimization design can be desirably
performed with respect to a single grey relational grade rather = 0.5448.
than complicated performance characteristics.
The response table of Taguchi method is employed here Using the same method, calculations are performed for
to calculate the average grey relational grade for each cutting each cutting parameter level and the response table is con-
parameter level. It is done by sorting the grey relational grades structed as shown in Table 11. Basically, the larger the grey
corresponding to levels of the cutting parameter in each col- relational grade, the better the corresponding multiple per-
umn of the orthogonal array, and taking an average on those formance characteristic. Accordingly, we select the level that
with the same level. For instance, in the second column in the gives the largest average response. From the response table for
orthogonal array (as shown in Table 2) the No. 1, No. 2, No. 3, the grey relational grades shown in Table 11, the best combina-

Table 11 – Response table for the grey relational grade


Symbol Cutting parameter Level 1 Level 2 Level 3 Max-Min

A Milling type 0.6295 0.5838 0.0457


V Spindle speed 0.6587 0.6162 0.5448 0.1139
F Feed per tooth 0.6721 0.5934 0.5544 0.1177
Da Axial depth of cut 0.5882 0.6222 0.6094 0.0340
Dr Radial depth of cut 0.5932 0.6124 0.6142 0.0210

Average grey relational grade = 0.6066

Table 12 – Results of the analysis of variance


Symbol Cutting parameter Degree of freedom Sum of square Mean square F value Contribution (%)

A Milling type 1 0.0081 0.0081 3.1153 6.67


V Spindle speed 2 0.0404 0.0202 7.7692 33.28
F Feed per tooth 2 0.0472 0.0236 9.0769 38.88
Da Axial depth of cut 2 0.0025 0.0013 0.5000 2.06
Dr Radial depth of cut 2 0.0022 0.0011 0.4231 1.81
Error 8 0.0209 0.0026 17.22

Total 17 0.1214

F0.05(1,8) = 5.32, F0.05(2,8) = 4.46.


3816 j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817

Table 13 – Comparison between initial level and optimal level


Best combination Tool life (min) Total metal removal volumes (mm3 ) Metal removal rate (mm3 /s)

Initial design A1 V1 F2 Da1 Dr1 79.17 95764.032 20.16


Optimal design A1 V1 F1 Da2 Dr3 100.00 153600.000 25.60
Final gain 20.83 57835.968 5.44

tion of the cutting parameters is the set with A1 (down milling),


V1 (spindle speed of 12000 rpm), F1 (feed rate of 0.04 mm/t), Da2
(axial depth of cut of 0.8 mm), and Dr3 (radial depth of cut of
1.0 mm).

5.2. Analysis of variance

A statistical software with an analytical tool of ANOVA is used


to determine which parameter significantly affects the per-
formance characteristics. The results of ANOVA for the grey
relational grades are listed in Table 12. It shows that the two
parameters V and F are found to be the major factors with the
selected multiple performance characteristics, because their
corresponding F ratio are greater than 4.46 that are the ratio
of F0.05(2,8) .
The percentage of error, error , can be used to evaluate if an Fig. 3 – Effect of cutting time on flank wear of peripheral
experiment possesses feasibility and sufficiency or not, since cutting edge.
it is related to the uncertain or uncontrollable factors. Based on
experience, it can be regarded that an experiment is acceptable
without neglecting major factors if error is less than 15%, while Green tool manufacturer. The definition of tool life, shown
unacceptable with losing some significant factors if error is in Table 13 is the total cutting time before the breakage of
greater than 50% (Su, 2002; Huang et al., 2006). The percent- cutting edge of the tool or, that is, as flank wear of cutting
age of error for contribution is 17.22% as shown in Table 12, edge reaches 0.2 mm. As shown in Table 13, tool life increases
indicating that the proposed method as well as the outcome from 79.17 min to 100.00 min; total metal removal volumes
in this study is proven to be highly acceptable. increases from 95764.032 mm3 to 153600 mm3 , metal removal
rate increases from 20.16 to 25.60 mm3 /s. Figs. 2 and 3 show
5.3. Confirmation tests the effects of cutting time on flank wear of cutting edge under
the conditions of the initial cutting parameters compared with
Once the optimal level of the cutting parameters is identi- those of the final optimal cutting parameters. Consequently,
fied, the following step is to verify the improvement of the these confirmation tests reveal that the proposed algorithm
performance characteristics using this optimal combination. for solving the optimal combinations of the cutting parame-
Table 13 compares the results of the confirmation experiments ters in this work improves both tool life and metal removal
using the optimal cutting parameters (A1 , V1 , F1 , Da2 , Dr3 ) rate.
obtained by the proposed method and with those of the ini-
tial cutting parameters (A1 , V1 , F2 , Da1 , Dr1 ) provided by Top
6. Conclusions

This paper presents an application of grey relational analysis


coupled with principal component analysis for optimizing the
cutting parameters of rough cutting process in high-speed end
milling operation for SKD61 tool steel. It should be noted that
the optimal results can only be applied for a process specified
as rough cutting in high-speed end milling. The results are
summarized as follows:

1. The principal component analysis, used to determine


the corresponding weighting values of each performance
characteristics while applying grey relational analysis to
a problem with multiple performance characteristics, is
proven to be capable of objectively reflecting the relative
importance for each performance characteristic.
Fig. 2 – Effect of cutting time on flank wear of end cutting 2. Based on analysis of variance, the major controllable
edge. factors significantly affecting the multiple performance
j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 3808–3817 3817

characteristics are milling type, spindle speed, and feed per Jean, M.D., Tsai, J.S., 2004. Intelligent designs of experiment for
tooth with a desired total contribution of 79%. multiple characteristics optimizing a small-scale aquaculture
3. The optimal combination of the cutting parameters environment. J. Technol. 19 (4), 349–357.
Kramer, B.M., 1984. On tool materials for high speed machining. J.
obtained from the proposed method is the set with A1 ,
Eng. Ind. Trans. ASME 109 (2), 87–91.
V1 , F1 , Da2 , and Dr3 . The corresponding confirmation tests Kao, P.S., Hocheng, H., 2003. Optimization of electrochemical
shows that tool life, metal removal rate, and total removal polishing of stainless steel by grey relational analysis. J. Mater.
volumes increase by 26.31%, 27%, and 60.39%, respectively. Process. Technol. 140, 255–259.
4. The proposed algorithm greatly simplifies the optimization Lewis, D.L., 1992. Factors for Successful Rotating Tool Operation
design of cutting parameters with multiple performance at Hi-Speeds. SME High Speed Machining Clinic, Oak Brook, IL,
characteristics. Thus, the solutions from this method can USA, p. 19.
Lin, J.L., Lin, J.F., 2006. Grey theory applied to evaluate the
be an useful reference for tool manufacturers and operators
tribological performance of the a-C:H(N) coating films
who are willing to search for an optimal solution of cutting prepared by differing the nitrogen content and the film
conditions. thickness. Int. J. Adv. Manuf. 27, 845–853.
Lin, J.L., Lin, C.L., 2002. The use of the orthogonal array with grey
relational analysis to optimize the electrical discharge
Acknowledgement machining process with multiple performance
characteristics. Int. J. Machine Tools Manuf. 42, 237–244.
Financial supported from the National Science Council of the Liao, H.C., 2006. Multi-response optimization using weighted
China, Taiwan under grant number NSC-94-2212-E-150-037 is principal component. Int. J. Adv. Manuf. Technol. 27, 720–725.
acknowledged with gratitude. Narender Singh, P., Raghukandan, K., Pai, B.C., 2004. Optimization
by Grey relational analysis of EDM parameters on machining
Al-10% SiCp composites. J. Mater. Process. Technol. 155–156,
references 1658–1661.
Pan, L.K., Wang, C.C., Wei, S.L., Sher, H.F., 2007. Optimizing
multiple quality characteristics via Taguchi method-based
Chaplin, J., Miller, J.A., King, R.I., 1981. Summary of recent Grey analysis. J. Mater. Process. Technol. 182, 107–116.
Lockheed research re: high-speed machining. In: Proceedings Peng, Z., Kirk, T.B., 1999. Wear particle classification in a fuzzy
of the Ninth North American Manufacturing Research grey system. Wear 225–229, 1238–1247.
Conference (NAMRC), Pennsylvania, PA, USA, pp. 311–317. Pearson, K., 1901. On lines and planes of closest fit to systems of
Deng, J., 1982. Control problems of grey systems. Syst. Contr. Lett. points in space. Philos. Mag. Series 62, 559–572.
5, 288–294. Ross, P.J., 1988. Taguchi Techniques for Quality Engineering. Mc
Deng, J.L., 1989. Introduction to Grey system. J. Grey Syst. 1 (1), Graw-Hill, New York.
1–24. Su, C.T., 2002. Quality Engineering. Chinese Society for Quality,
Dewes, R.C., Aspinwall, D.K., 1997. A review of ultra high speed Taipei, pp. 128.
milling of hardened steels. J. Mater. Process. Technol. 69, 1–17. Society of British Aerospace Companies Ltd, London, UK, 1992.
Destmobes, Y., 1984. Very high speed machining of ferrous and The repercussions of high speed machining, Study Team
solid composite materials: test campaign results. High Speed Report, pp. 45.
Machining, the Winter Annual Meeting of ASME, New Smith, G.T., 1992. Ultra-high speed machining-the problems
Orleans, LA, USA, PED, Vol. 12, pp. 263–289. associated with spindle design, tooling and machined parts
Fung, H.C., Kang, P.C., 2005. Multi-response optimization in when milling. In: Proceedings of the Second International
friction properties of PBT composites using Taguchi method Conference on Flexible Automation and Information
and principal component analysis. J. Mater. Process. Technol. Management (FAIM), Falls Church, VA, USA, pp. 947–961.
170, 602–610. Tarng, Y.S., Juang, S.C., Chang, C.H., 2002. The use of grey-based
Harding, M., 1993. Get a grip on high speed Machining. Taguchi methods to determine submerged arc welding
Metalworking Prod. 137 (6), 33–34. process parameters in hardfacing. J. Mater. Process. Technol.
Huang, C.Y., Ying, K.C., Tsai, X.L., 2006. The optimization of 128, 1–6.
stencil printing process for lead free materials. J. Technol. 21 Tong, L.I., Wang, C.H., Chen, H.C., 2005. Optimization of multiple
(3), 227–236. responses using principal component analysis and technique
Hotelling, H., 1993. Analysis of a complex of statistical variables for order preference by similarity to ideal solution. Int. J. Adv.
into principal components. J. Educ. Psychol. 24, 417–441. Manuf. 27, 407–414.

Anda mungkin juga menyukai