Vous êtes sur la page 1sur 7

Expert Systems with Applications 38 (2011) 6822–6828

Contents lists available at ScienceDirect

Expert Systems with Applications


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

Taguchi-fuzzy multi output optimization (MOO) in high speed CNC turning


of AISI P-20 tool steel
Anil Gupta a,⇑, Hari Singh b, Aman Aggarwal a
a
Mechanical and Automation Engineering Department, Maharaja Agrasen Institute of Technology, Rohini, Delhi 110086, India
b
Department of Mechanical Engineering, National Institute of Technology, Kurukshetra 136119, India

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

Keywords: This paper presents the application of Taguchi method with logical fuzzy reasoning for multiple output
High speed CNC turning optimization of high speed CNC turning of AISI P-20 tool steel using TiN coated tungsten carbide coatings.
Taguchi-fuzzy hybrid approach The machining parameters (cutting speed, feed rate, depth of cut, nose radius and cutting environment)
Comprehensive output measure (COM) are optimized with considerations of the multiple performance measures (surface roughness, tool life,
Multi output optimization (MOO)
cutting force and power consumption). Taguchi’s concepts of orthogonal arrays, signal to noise (S/N)
AISI P-20 tool steel
ratio, ANOVA have been fuzzified to optimize the high speed CNC turning process parameters through
a single comprehensive output measure (COM). The result analysis shows that cutting speed of 160
m/min, nose radius of 0.8 mm, feed of 0.1 mm/rev, depth of cut of 0.2 mm and the cryogenic environment
are the most favorable cutting parameters for high speed CNC turning of AISI P-20 tool steel.
Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction Taguchi’s ideology has been built upon W.E. Deming’s observa-
tion that 85% of poor quality is attributable to the manufacturing
Today’s fast changing manufacturing environment requires the process and only 15% to the worker (Park, 1996). Taguchi developed
application of optimization techniques in metal cutting processes the robust manufacturing systems, insensitive to daily and seasonal
to effectively respond to severe competitiveness and to meet the variations of environment, machine and wear. His approach
increasing demand of customizable quality product (low cost, high optimizes the performance characteristics through the settings of
quality, easily deliverable) in the market Taguchi’s method is one process parameters and reduces the sensitivity of the system
of the most effective systems of off-line quality control where performance to sources of variation. Consequently the Taguchi
the quality is in-built at the product design stage instead of con- method (Taguchi, 1990) has become a powerful tool in the design
trolling it at the manufacturing stage or through the inspection of experiment methods (Roy, 1990). Aggarwal, Singh, Kumar, and
of final products. In true quality sense, a customer usually consid- Singh (2008a, 2008b) comparatively used response surface meth-
ers several correlated quality characteristics of a product. Accord- odology and Taguchi method to conclude that the low temperature
ingly, variability of a product’s response has to be reduced and cutting conditions significantly reduce the power consumption as
mean needs to be brought close to the target. compared to dry and wet cutting conditions.
Metal cutting is one of the important and widely used manufac- Most real world engineering processes comprise of multiple
turing processes in engineering industries. The metal cutting responses. Only the simplest systems can be modeled with single
studies focus on the features of tools, work material composition response optimization and bulk of the processes are too complex
and mechanical properties, and above all the machine parameter to be categorized into individual responses. However analyzing
settings that influence the process efficiency and output quality and solving a multiple performance characteristics is a challenging
characteristics (or responses). A significant improvement in process research problem as against the most published Taguchi applica-
efficiency can be obtained by process parameter optimization that tions in the area of optimization of a single performance character-
identifies and determines the regions of critical process control istic which is time consuming and exorbitant in cost. Statistical
factors leading to desired outputs or responses with acceptable design of experiments (DOE) refers to the effective process of plan-
variations ensuring a lower cost of manufacturing (Montgomery, ning the experiment in such a manner so that the appropriate data
1997). can be analyzed by statistical methods, resulting in valid and
objective conclusions (Elsayed & Chen, 1993).
Zadeh (1965) initiated the theory of fuzzy logics to deal with
⇑ Corresponding author. Mob.: +91 9868876407. uncertain and vague information. His definition of performance
E-mail address: anil_gupta10@hotmail.com (A. Gupta).
characteristics such as lower-the-better (LB), higher-the-better

0957-4174/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2010.12.057
A. Gupta et al. / Expert Systems with Applications 38 (2011) 6822–6828 6823

(HB), and nominal-the-best (NB) contains a certain degree of transformed into a signal-to-noise (S/N) ratio. This S/N ratio which
uncertainty and vagueness. Therefore, optimization of the compli- measures the deviation of the performance characteristics from the
cated multiple performance characteristics with fuzzy logic has desired values has been categorized into 3 types of performance
been considered in this study by transforming it into the optimiza- characteristics – the lower-the-better, the higher-the-better, and
tion of a single comprehensive output measure (COM) for a high the nominal-the-best. Regardless of the category of the perfor-
speed CNC turning problem. mance characteristic, a larger S/N ratio corresponds to better per-
In high speed CNC turning of a tool steel, machining parameter formance characteristic. Therefore, the optimal level of the
selection for achieving optimal performance like a good surface process parameters is the level with the highest S/N ratio. Further-
finish, close tolerance-dimensional accuracy, high strength is more, a statistical analysis of variance (ANOVA) is performed to
important and essential. The determination of desired machining identify the process parameters that are statistically significant.
parameters based on experience or handbook values does not en- The optimal combination of the process parameters can then be
sure that the selected machining parameters result in optimal or predicted based on the above analysis. Finally, a confirmation
near optimal machining performance for that particular CNC lathe experiment is conducted to verify the optimal process parameters
and environment. obtained from the parameter design.
Most of the optimization techniques do have certain con- Basically, the Taguchi method is designed to handle the optimi-
straints, assumptions and limitations for implementation in real- zation of a single performance characteristic. The usual recommen-
life cutting process problems. The optimum setting of parameters dation for the optimization of a process with multiple performance
for various responses is usually a conflicting task and the weight- characteristics is a matter of engineering judgment (Phadke, 1989),
age of responses is a matter of researcher’s judgment. Still the as it involves achieving a trade off between several multiple con-
researchers continually strive to find still better optimized cutting flicting problems and variables. For example: the category of each
conditions in order to economize the machining problem at hand. performance characteristic may not be same; the engineering unit
Lin, Wang, Yan, and Tarng (2000) optimized the electrode wear for describing each performance characteristic may be different;
rate and material removal rate of an electrical discharge machine and the importance of each performance characteristic may vary.
by finding the optimum machine parameter setting of work-piece As a result, the application of the Taguchi method in a process with
polarity, pulse on time, duty factor, voltage, current and dielectric multiple performance characteristics cannot be straightforward. In
fluid using Taguchi method with fuzzy logics. Aggarwal et al. this paper, the use of fuzzy logic to deal with the optimization of a
(2008a, 2008b) optimized the machining parameters of CNC process with multiple performance characteristics is reported.
turned parts using Principal Component Analysis with Taguchi First, several fuzzy rules are derived based on the performance
method and inferred that more than two principal components requirement of the process. The loss function corresponding to
with eigen value more than one are necessary to explain the vari- each performance characteristic is fuzzified and then a single
ability in the experimental data. Paiva Anderson, Ferreira, and COM is obtained through fuzzy reasoning on the fuzzy rules. The
Baletrassi (2007) used hybrid approach by combining RSM with COM can be used to optimize the process based on the Taguchi
PCA to optimize multiple correlated responses within the experi- approach.
mental region in turning of AISI 52100 hardened steel by consider-
ing 3 inputs (cutting speed, feed, depth of cut) to optimize several 3. The high speed CNC turning process parameters
outputs (tool life, processing cost, cutting time, surface roughness,
metal removal rate). Similarly Nian, Yang, and Tarng (1999) opti- 3.1. Process parameters
mized the turning process on lathe by considering several perfor-
mance characteristics. The process parameters selected for the present work are
In the present study, the Taguchi method with fuzzy logic is shown in Table 1.
used as an efficient approach to determine the optimal machining
parameters for CNC turned parts for optimization of tool life, cut-
3.2. Work material
ting force, power consumption and surface roughness. The experi-
mental design used orthogonal array L27 for the five controllable
The work material selected for the study was AISI P20 tool steel
factors that is, cutting speed, feed, depth of cut, nose radius and
which is used extensively for making injection moulds and occa-
cutting environment, each at three levels to find the optimum
sionally compression moulds. The steel is supplied in hardness
combination of factors and levels in high speed CNC turning of AISI
range of about 32–36 Rc. The chemical composition of this material
P-20 tool steel. The single-response optimization was conducted
is: 0.4% C, 1.5% Mn, 1.9% Cr, 1.0% Ni, and 0.2% Mo. The material is
using Taguchi method. For a multi-response case, fuzzy logic unit
machined in pre-hardened condition and subsequently no heat
(FLU) was employed to transform the four correlated responses
treatment is required to be carried out.
to a single response called comprehensive output measure
(COM). Finally, the analysis of variance (ANOVA) was used to find
out the most influential high speed CNC turning parameter for 3.3. Tool material
problems of single and multiple responses.
The cutting tool selected for machining AISI P20 steel was TiN
coated tungsten carbide inserts of Kennametal make. The tungsten
2. Optimization of multiple performance characteristics with
fuzzy logic
Table 1
Machining parameters and their levels.
Originally developed by Fisher (1925) the classical experimen-
Factors/process parameters Level 1 Level 2 Level 3
tal design methods (Montgomery, 1997) are complex in applica-
tion due to increase in the number of experiments to be S Cutting speed (m/min) 120 160 200
F Feed rate (mm/rev) 0.10 0.12 0.14
performed with the increase in the number of process parameters.
D Depth of cut (mm) 0.20 0.35 0.50
To solve this, the Taguchi’s method uses a special design of orthog- N Nose radius (mm) 0.40 0.80 1.20
onal array (OA) to study the entire parameter space with only a
E Environment Dry Wet Cryo
small number of experiments where experimental results are
6824 A. Gupta et al. / Expert Systems with Applications 38 (2011) 6822–6828

carbide inserts used were of ISO coding CNMG 120404, CNMG degrees of freedom associated with interaction between two pro-
120408 and CNMG 120412 and tool holder of ISO coding PCLNR cess parameters are given by the product of the degrees of freedom
1616H07. The selection of parameters of interest and their ranges for the two process parameters. In the present study, the interac-
was based on the literature reviewed and results of some prelimin- tion between the machining parameters is neglected. Therefore,
ary experiments. there are 10 degrees of freedom due to five three-level machining
parameters in the high speed CNC turning process.
3.4. Experimental set up Design of experiment (DOE) and Taguchi’s techniques have
been used to accomplish the objective of the experimental study.
The experimental work was carried out at Solid State Physics L27 orthogonal has been used for conducting the experiments and
Lab (SSPL), Delhi under Ministry of Defence on a Scaublin make is shown in Table 2.
high speed CNC turning center as shown in Fig. 1. AISI P20 tool
steel bars (65 mm diameter and 275 mm length) were used for 4.2. Signal-to-noise ratio
the experimentation. Canned cycle was used for machining and
the machining was done in absolute mode. A Cryocan was used Taguchi’s loss function estimates the deviation between exper-
for pouring liquid nitrogen in the form of jet on machining area. imental value and desired value. Usually, there are three categories
of the performance characteristics in the analysis of the signal-to-
3.5. Machining performance evaluatory factors noise ratio, i.e., the lower-the-better, the higher-the-better, and the
nominal-the-best. To obtain optimal machining performance, the
The machining performance is evaluated by four parameters minimum SR, CF, PC are desired therefore lower-the-better charac-
namely surface roughness (SR), tool life (TL), cutting force (CF) teristic is chosen.
and power consumption (PC). Three components of force viz tan- For ‘lower the better’ type of machining quality characteristic,
gential, feed and radial force were measured with a three-dimen- the S/N ratio is given by
sional turning dynamometer. The square root of sum of squares   
of the average value of these forces is equal to cutting force. Power S=N Ratio ¼ 10 Log 1=n y21 þ y22 þ    þ y2n ; ð1Þ
consumption was measured with a set of Watt meters as three
where y1 ; y2 ; . . . ; yn are the responses of the SR, CF, PC taken sepa-
phase supply was used and their readings were added after multi-
rately for a trial condition repeated n times. The S/N ratios were
plying by suitable multiplying factor to get power consumption in
computed using Eq. (1) for each of the 27 trials.
Watts. A pocket surf (Mahr Germany) was used to measure the Ra
For ‘higher the better’ type of machining quality characteristic,
value of surface roughness in microns. A fixture was specially de-
the S/N ratio is given by
signed to ensure holding of the component accurately while taking
  
measurement of surface roughness. S=N Ratio ¼ 10 Log 1=n 1=y21 þ 1=y22 þ    þ 1=y2n ; ð2Þ

4. Optimization of high speed CNC turning parameters where y1 ; y2 ; . . . ; yn are the responses of TL for a trial condition re-
peated n times. The S/N ratios were computed using Eq. (2) for each
In this section, the use of the Taguchi method with fuzzy logic to of the twenty seven trials.
determine the machining parameters with optimal machining per-
formance in the high speed CNC turning process is illustrated. Table 2
Experimental layout using an L27 orthogonal array.
4.1. Orthogonal array experiment S. No. Machining parameter and their levels
Speed Feed Depth Nose Environment
To select an appropriate orthogonal array for the experiments, of cut radius
the total degrees of freedom need to be computed. The degrees S (m/min) F (mm/rev) D (mm) N (mm) E
of freedom are defined as the number of comparisons between 1 120 0.10 0.20 0.40 Dry
process parameters that need to be made to determine which level 2 120 0.10 0.35 0.80 Wet
is better and specifically how much better it is. For example, a two- 3 120 0.10 0.50 1.20 Cryo
level process parameter counts for one degree of freedom. The 4 120 0.12 0.20 0.80 Wet
5 120 0.12 0.35 1.20 Cryo
6 120 0.12 0.50 0.40 Dry
7 120 0.14 0.20 1.20 Cryo
8 120 0.14 0.35 0.40 Dry
9 120 0.14 0.50 0.80 Wet
10 160 0.10 0.20 1.20 Wet
11 160 0.10 0.35 0.40 Cryo
12 160 0.10 0.50 0.80 Dry
13 160 0.12 0.20 0.40 Cryo
14 160 0.12 0.35 0.80 Dry
15 160 0.12 0.50 1.20 Wet
16 160 0.14 0.20 0.80 Dry
17 160 0.14 0.35 1.20 Wet
18 160 0.14 0.50 0.40 Cryo
19 200 0.10 0.20 0.80 Cryo
20 200 0.10 0.35 1.20 Dry
21 200 0.10 0.50 0.40 Wet
22 200 0.12 0.20 1.20 Dry
23 200 0.12 0.35 0.40 Wet
24 200 0.12 0.50 0.80 Cryo
25 200 0.14 0.20 0.40 Wet
26 200 0.14 0.35 0.80 Cryo
27 200 0.14 0.50 1.20 Dry
Fig. 1. Schematic diagram of the high speed CNC turning process.
A. Gupta et al. / Expert Systems with Applications 38 (2011) 6822–6828 6825

Table 3 shows the experimental results for surface roughness, the smaller-the-better methodology of S/N ratio was employed
tool life, cutting force and power consumption and their S/N ratio for the aforesaid responses. On the other hand for tool life higher
based on the experimental parameter combinations (Table 2). The the better methodology of S/N ratio is employed, as higher value
table also shows that the engineering units for all the four mea- of the same is desired in metal machining. The S/N ratios of four
sured outputs are different. To consider these four different perfor- responses of the 27 experimental runs are listed in Table 3 along
mance characteristics in the Taguchi method, the S/N ratios with their experimentally measured values.
corresponding to the surface roughness, tool life, cutting force The response table of S/N ratios of tool life was calculated, as
and power consumption are processed by the fuzzy logic unit. shown in Table 4. The S/N ratio of factors S–F–D–N–E is maximum
at S1, F1, D1, N2 and E3, respectively. As a result, the factor/level
4.3. Single characteristic optimization combination S1F1D1N2E3 was recommended. The results of ANOVA
for the S/N ratios of the tool life are also shown in Table 4. It can be
When a single-response problem is considered, Taguchi method seen that the contribution of factor E to the tool life was the largest
can be employed to obtain the optimal level/factor combination of (69.27%) followed by factor S (24.36%). Thus, cutting environment
high speed CNC turning process. In the past a lot of researchers was most important factor followed by cutting speed as far as tool
studied the effect of various machining parameters in optimizing life is concerned.
the machined surface integrity using Taguchi’s method (Beaucb- The same analysis procedure was applied to optimize the high
amp Thomas, Youssef, & Masounave, 1997; Choudhury & EI-Bara- speed CNC turning conditions for the surface roughness, power
die, 1997; Lin, Lee, & Wu, 2001). The S/N is used to represent consumption and cutting force also. The levels that gave the largest
quality characteristic and the largest S/N ratio is demanded. In average response were S3F2D2N3E3 for the surface roughness,
addition, the contribution of each factor is evaluated using the S1F1D1N2E3 for the power consumption, and S2F1D1N2E3 for the cut-
ANOVA. ting force. A combined analysis of all the four ANOVAs showed that
A smaller value of power consumption, surface roughness and the optimal factor/level combination, or the most important factor,
cutting force is normally required in metal machining. Therefore, for one quality was usually different from that for another quality.

Table 3
Experimental results for output variables and their S/N ratio.

Run No. SR TL CF PC
lm S/N Min S/N N S/N W S/N
1 1.41 2.985 29.00 38.489 171.30 40.731 1066 58.759
2 0.71 2.894 34.00 40.175 147.50 43.379 1560 63.862
3 0.6 4.485 54.67 44.297 111.74 44.677 866 60.562
4 0.47 6.554 34.67 40.341 120.30 41.611 1493 63.484
5 0.19 14.256 51.66 43.809 180.60 45.135 987 59.885
6 1.18 1.414 27.00 38.174 236.20 47.468 1187 61.488
7 0.67 3.522 50.00 43.523 157.70 43.959 960 59.650
8 1.16 1.264 24.66 37.391 214.40 46.627 1134 61.088
9 0.92 0.724 28.33 38.590 286.90 49.157 1813 65.170
10 0.18 14.895 27.66 38.387 116.37 41.320 1586 64.010
11 0.45 6.871 47.66 43.107 133.33 42.627 1013 60.117
12 0.43 7.195 21.66 36.260 191.23 45.633 1240 61.871
13 0.58 4.731 45.66 42.728 125.40 41.966 893 59.022
14 0.72 2.885 20.33 35.723 149.43 43.491 1253 61.962
15 0.31 10.170 25.66 37.731 212.46 46.547 1773 65.041
16 0.66 3.609 20.00 35.570 162.93 44.242 1107 60.882
17 0.64 3.831 22.33 36.523 190.23 45.588 1533 63.713
18 0.75 2.460 41.33 41.869 177.76 45.000 1373 62.756
19 0.16 16.097 40.00 41.585 106.23 38.208 1053 60.562
20 0.23 12.862 15.67 33.446 208.50 46.383 1373 62.756
21 0.67 3.434 21.67 36.260 209.80 46.437 2094 66.417
22 0.4 7.953 14.67 32.874 200.20 46.032 1286 62.190
23 0.5 6.016 20.33 35.709 178.80 45.048 1866 65.422
24 0.18 14.886 37.66 41.062 168.70 44.546 1613 64.155
25 0.64 3.831 18.00 34.657 162.00 44.196 1573 63.937
26 0.31 10.170 34.33 40.258 162.50 44.217 1453 63.248
27 0.48 6.374 16.66 30.111 276.16 48.827 1667 64.438

Table 4
The response table and ANOVA of S/N ratios of tool life.

Variable Average level DOF SS MS F Contribution (%)


1 2 3
S 40.53 38.66 36.22 2 84.062 42.031 29.168 24.36
F 39.11 38.68 37.61 2 10.612 5.306 3.682 3.07
D 38.68 38.46 38.26 2 0.646 0.323 0.224 0.19
N 38.71 38.84 37.86 2 4.990 2.495 1.731 1.45
E 35.34 37.60 42.47 2 239.068 119.534 82.952 69.27
Error 4 5.762 1.441 1.67
Total 14 345.140 24.653 100.00
6826 A. Gupta et al. / Expert Systems with Applications 38 (2011) 6822–6828

In such a case, an engineering judgement that refers to past expe-  Rule 1: if x1 is A1 and x2 is B1 and x3 is C1 and x4 is D1 then y is E1
rience is the only real guarantee of correct decision-making in the else
high speed CNC turning process.  Rule 2: if x1 is A2 and x2 is B2 and x3 is C2 and x4 is D2 then y is E2
else
 ...
4.4. Fuzzy logic implementation and results for multiple responses in
 Rule n: if x1 is An and x2 is Bn and x3 is Cn and x4 is Dn then y is En.
high speed CNC turning

Ai, Bi, Ci, Di and Ei are fuzzy subsets defined by the corresponding
A fuzzy logic unit (FLU) comprises a fuzzifier, membership func-
membership functions, i.e., lAi, lBi, lCi, lDi and lEi. In this paper,
tions, a fuzzy rule base, an inference system, and a defuzzifier.
two fuzzy subsets (low and high) are assigned to the four inputs
First, the fuzzifier uses membership functions to fuzzify the S/N ra-
(Fig. 2. for SR). Five fuzzy subsets are assigned to the only output
tios. Next, the inference system performs a fuzzy reasoning on fuz-
COM (Fig. 3). Various degree of membership to the fuzzy sets is cal-
zy rules to generate a fuzzy value. Finally, the defuzzifier converts
culated based on the values of x1, x2, x3, x4 and y. Sixteen fuzzy
the fuzzy value into a COM. In the present work, fuzzy reasoning is
rules (Table 5) are derived directly based on the fact that larger
based on the four-input–one-output fuzzy logic unit. The fuzzy
is the S/N ratio, the better is the performance characteristic. By tak-
rule base consists of a group of if–then control rules with the four
ing the max–min compositional operation (Zimmermann, 1985),
inputs, x1 and x2, and one output y, i.e.,
the fuzzy reasoning of these rules yields a fuzzy output. Supposing
that x1, x2, x3 and x4 are the four input values of the fuzzy logic unit,
the membership function of the output of fuzzy reasoning can be
expressed as

lC 0 ðyÞ ¼ ðlA1 ðx1 Þ ^ lB1 ðx2 Þ ^ lC 1 ðx3 Þ ^ lD1 ðx4 ÞlE1 ðyÞÞ
_    ðlAn ðx1 Þ ^ lBn ðx2 Þ ^ lC n ðx3 Þ ^ lDn ðx4 ÞlC n ðyÞÞ; ð3Þ

where ^ is the minimum operation and _ is the maximum


operation.
Finally, a defuzzification method, called the center-of-gravity
method (Zimmermann, 1985), is adopted here to transform the
fuzzy inference output lC0 into a non-fuzzy value y0, i.e.,
Fig. 2. Membership functions for the SR. P
ylco ðyÞ
yo ¼ P : ð4Þ
lco ðyÞ
In this paper, the non-fuzzy value y0 is called COM. Based on the
above discussion, the larger is the COM, the better is the perfor-
mance characteristic. Table 6 shows the MATLAB 7 generated fuzzy
logic unit results for the COM using the experimental combinations
of Table 2.

Table 6
Results for the COM.

Run No. Comprehensive output


measure (COM)
1 0.592
Fig. 3. Membership functions for the COM. 2 0.472
3 0.575
4 0.553
5 0.748
Table 5 6 0.395
Fuzzy rule table. 7 0.630
8 0.417
S/N for COM 9 0.307
SR TL CF PC 10 0.588
11 0.644
Low Low Low Low Lowest 12 0.486
High Low 13 0.683
High Low Low 14 0.472
High Medium 15 0.431
High Low Low Low 16 0.478
High Medium 17 0.438
High Low Medium 18 0.481
High High 19 0.792
High Low Low Low Low 20 0.442
High Medium 21 0.372
High Low Medium 22 0.436
High High 23 0.428
High Low Low Medium 24 0.568
High High 25 0.432
High Low High 26 0.547
High Highest 27 0.279
A. Gupta et al. / Expert Systems with Applications 38 (2011) 6822–6828 6827

The orthogonality in the experimental data makes it possible to


separate out the effect of each machining parameter at different
levels. For example, the mean of COM for the cutting speed at lev-
els 1, 2 and 3 can be calculated by averaging the multi-response
performance indexes for experiments 1–9, 10–18 and 19–27,
respectively (Table 2 and 6). The mean of COM for each level of
the other machining parameters can be computed in a similar
manner and is summarized in the COM table (Table 7). In addition,
the total mean of the COM for the 27 experiments is also calculated
and listed in Table 7. Basically, the larger is the COM, the smaller is
the variance of the performance characteristics around the desired
average COM value. However, the relative importance amongst the
machining parameters for the multiple performance characteris-
Fig. 5. The surface graph for COM for TL and SR combination.
tics still needs to be known so that the optimal combinations of
the machining parameter levels can be determined more
accurately. 4.5. Analysis of variance (ANOVA) of COM
Fig. 4 shows the fuzzified rule viewer for COM, which can accept
any value of 4 multiple output responses of SR, TL, and CF and PC. The ANOVA investigates those process parameters which signif-
The results for 27 experimental runs are shown in Table 6. Fig. 5 icantly affect the performance characteristics. This is accomplished
shows the surface view of COM for TL and SR combination. by separating the total variability of the multi-response perfor-
mance indexes, which is measured by the sum of the squared devi-
ations from the total mean of the COM, into contributions by each
Table 7
COM table. of the process parameter and the error. First, the total sum of the
squared deviations SST from the total mean of the COM lm can
Symbol Machining parameter COM value level averages Max–Min
be calculated as
1 2 3
X
n
 
S Speed 0.521 0.522 0.477 0.045 SST ¼ li  lm 2 ; ð5Þ
F Feed 0.551 0.524 0.445 0.106 i¼1
D Depth of cut 0.576 0.512 0.433 0.143
N Nose radius 0.494 0.519 0.507 0.025 where n is the number of experiments in the orthogonal array and l
E Environment 0.444 0.447 0.630 0.186 is the mean of the COM for the ith experiment.
Mean COM 0.506 dB The total sum of the squared deviations SST is decomposed into
two sources: the sum of the squared deviations SSd due to each

Fig. 4. The fuzzy rule COM graph for 4 input combinations.


6828 A. Gupta et al. / Expert Systems with Applications 38 (2011) 6822–6828

Table 8
Results of the ANOVA for comprehensive output measure (COM).

Symbol Machining parameter Degrees of freedom Sum of squares Mean sum of squares F-value Contribution (%)
S Speed 2 0.012 0.006 0.813 3.04
F Feed rate 2 0.055 0.027 3.683 13.78
D Depth of cut 2 0.093 0.047 6.267 23.45
N Nose radius 2 0.003 0.002 0.219 0.82
E Environment 2 0.204 0.102 13.748 51.43
Error 4 0.030 0.007 7.48
Total 14 0.397 0.028 100.00

process parameter and the sum of the squared error SSe. The per- roughness, cutting force and power consumption compared
centage contribution by each of the process parameter in the total to wet (conventional coolant ILO cut 154 Indian Oil recom-
sum of the squared deviations SST can be used to evaluate the mended for CNC machine) and dry machining.
importance of the process-parameter change on the performance
characteristics. In addition, the F-test named after Fisher (1925) The paper has presented the use of fuzzy logics to the Taguchi
can also be used to determine which process parameters have a method in optimization of the high speed CNC turning with multi-
significant effect on the performance characteristic. Usually, the ple performance characteristics. A fuzzy reasoning of the multiple
change of the process parameter has a significant effect on the per- performance characteristics has been performed by the fuzzy logic
formance characteristic when the F-value is large. unit. As a result, four performance characteristics namely surface
The results of ANOVA for COM (Table 8) indicate that environ- roughness, tool life, cutting force and power consumption can be
ment is the most significant machining parameters in affecting improved. It can be concluded that the optimization methodology
the multiple performance characteristics followed by depth of developed in this study is useful in improving multiple perfor-
cut and feed rate. Based on the above discussion, the optimal mance characteristics in high speed CNC turning.
machining parameters are the cutting speed at level 2
(S = 160 m/min), feed rate at level 1 (F = 0.10 mm/rev), depth of References
cut at level 1 (D = 0.20 mm), nose radius at level 2 (N = 0.80 mm),
and environment at level 3 (E = Cryo) or S2F1D1N2E3 in short. Aggarwal, A., Singh, H., Kumar, P., & Singh, M. (2008a). Multi-characteristic
optimization of CNC turned parts using principal component analysis.
International Journal of Machining and Machinability of Materials, 3(1/2), 208–223.
5. Conclusions Aggarwal, A., Singh, H., Kumar, P., & Singh, M. (2008b). Optimizing power
consumption for CNC turned parts using response surface methodology and
Taguchi’s technique – A comparative analysis. Journal of Materials Processing
The hybrid Taguchi-fuzzy approach is used in this study to opti- Technology, 200, 373–384.
mize the high speed CNC turning conditions of AISI P-20 tool steel. Beaucbamp Thomas, M., Youssef, Y., & Masounave, Y. A. (1997). An experimental
design for surface roughness and built-up edge formation in lathe dry turning.
The cutting parameters optimization is carried out through exper-
International Journal of Quality Science, 2(3), 167–180.
iments with minimum number of trials as against full factorial de- Choudhury, I. A., & EI-Baradie, M. A. (1997). Surface roughness prediction in the
sign. The fuzzification process takes care of the vagueness in the turning of high-strength steel by factorial design of experiments. Journal of
information and produces the best suitable cutting conditions in Materials Processing Technology, 67, 55–61.
Elsayed, E. A., & Chen, A. (1993). Optimal levels of process parameters for products
the present study. The results are summarized as follows: with multiple characteristics. International Journal of Production Research, 31(5),
1117–1132.
(1) The factor/level combination S3F2D2N3E3 for surface rough- Fisher, R. A. (1925). Statistical methods for research worker. London: Oliver & Boyd.
Lin, W. S., Lee, B. Y., & Wu, C. L. (2001). Modeling the surface roughness and cutting
ness, S1F1D1N2E3 for the tool life, S1F1D1N2E3 for power con- force for turning. Journal of Materials Processing Technology, 108, 286–293.
sumption and S2F1D1N2E3 for cutting force are the Lin, J. L., Wang, K. S., Yan, B. H., & Tarng, Y. S. (2000). Optimization of the electrical
recommended optimum parameters, for high speed CNC discharge machining process based on the Taguchi method with fuzzy logics.
Journal of Materials Processing Technology, 102, 48–55.
turning when all four responses are considered Montgomery, D. C. (1997). Design and analysis of experiments (4th ed.). New York:
independently. Wi1ey.
(2) In the multi-response problem, all the four responses tool Nian, C. Y., Yang, W. H., & Tarng, Y. S. (1999). Optimization of turning operations
with multiple performance characteristics. Journal of Materials Processing
life, power consumption, cutting force and surface rough- Technology, 95, 90–96.
ness were simultaneously considered, and S2F1D1N2E3 was Paiva Anderson, P., Ferreira, J. R., & Baletrassi, P. P. (2007). A multivariate hybrid
the recommended optimum condition as per the hybrid approach applied to AISI 52100 hardened steel turning optimization. Journal of
Materials Processing Technology, 189, 26–35.
Taguchi-fuzzy approach.
Park, S. H. (1996). Robust design and analysis or quality engineering. London:
(3) It can be concluded that middle level of cutting speed Chapman and Hall.
(160 m/min) and nose radius (0.8 mm) and lower level of Phadke, M. S. (1989). Quality engineering using robust design. Englewood Cliffs, NJ:
feed (0.1 mm/rev) and depth of cut (0.2 mm) yield the opti- Prentice-Hall.
Roy, R. K. (1990). A primer on Taguchi method. New York: Van Noshtrand Reinhold
mal result. Cryogenic environment is the most favourable Int. Co. Ltd.
condition out of three cutting environments. Taguchi, G. (1990). Introduction to quality engineering. Tokyo: Asian Productivity
(4) Both single response and multi-response optimization anal- Organization.
Zadeh, L. (1965). Fuzzy sets. Information and Control, 8, 338–353.
ysis proved that cryogenic machining environment E3 is Zimmermann, H. J. (1985). Fuzzy set theory and its applications. London: Kluwer
favourable in increasing tool life and reducing surface Academic Publishers.

Vous aimerez peut-être aussi