Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595

www.springerlink.com/content/1738(494x
DOI 10.1007/s12206(012(0851(3

Determining the optimum process parameter for grinding operations

using robust process†
Süleyman Ne1eli1, Đlhan Asiltürk2,* and Levent Çelik2
1
Vocational High School, University of Selçuk, 42075, Konya, Turkey
2
Faculty of Technology, University of Selçuk, 42075, Konya Turkey

(Manuscript Received August 11, 2011; Revised May 21, 2012; Accepted May 31, 2012)

((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((

Abstract

We applied combined response surface methodology (RSM) and Taguchi methodology (TM) to determine optimum parameters for
minimum surface roughness ( Ra ) and vibration (Vb) in external cylindrical grinding. First, an experiment was conducted in a CNC cy(
lindrical grinding machine. The TM using L27 orthogonal array was applied to the design of the experiment. The three input parameters
were workpiece revolution, feed rate and depth of cut; the outputs were vibrations and surface roughness. Second, to minimize wheel
vibration and surface roughness, two optimized models were developed using computer(aided single(objective optimization. The ex(
perimental and statistical results revealed that the most significant grinding parameter for surface roughness and vibration is workpiece
revolution followed by the depth of cut. The predicted values and measured values were fairly close, which indicates ( RRa2
= 94.99 and
RVb = 92.73) that the developed models can be effectively used to predict surface roughness and vibration in the grinding. The estab(
2

lished model for determination of optimal operating conditions shows that a hybrid approach can lead to success of a robust process.
Keywords: CNC grinding; RSM; Taguchi design; Surface roughness; Vibration monitoring
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&

parameters in process planning. The selection of grinding

1. Introduction
parameters is traditionally executed by process planners either
In machining operations, the quality of surface finish is an on the basis of their practical knowledge gained by experience
important factor in evaluating the productivity of machined on the shop floor or with the help of data handbooks [1, 2].
parts. Grinding processes are used to smooth or improve sur( This process, however, may be time(consuming and cannot
face finish quality. Cylindrical grinding is a complex, mate( satisfy any economic criteria; also it cannot determine the
rial(removal process with a great number of influencing fac( exact optimum parameters because of restricted experiments
tors, which are non(linear, interdependent and difficult to [3]. Moreover, these solutions are not reliable or acceptable
quantify. To maximize the surface quality, the selection of for selecting the optimum cutting parameters from a produc(
grinding parameters is vital; the vibration and surface rough( tivity point of view. Efficient process set(up should be model
ness is chiefly affected by the selection of grinding parameters. based in which the required surface roughness is obtained.
So many factors influence the grinding process that a repro( Statistical design of experiments (DOE) refers to the proc(
ducible workpiece quality is rarely, if ever, achieved. Al( ess of planning the experiment so that the appropriate data can
though many efforts have been made to predict the parameters be analyzed by statistical methods, resulting in valid and ob(
of the grinding process, many difficulties remain because jective conclusions [4(7]. DOE methods such as factorial de(
abrasive processes are dynamic in nature unlike turning and sign, RSM and TM are now widely used for optimize the ma(
milling processes, as the cutting edges of the grinding wheel chining parameters in place of one(factor(at(a(time experi(
are not uniform and act differently on the workpiece at each mental approach which is time consuming and exorbitant in
grinding. To predict the parameters of the grinding process, it cost [8]. To this end, much research has been conducted to
is necessary to quantify surface roughness, which is one of the determine optimal process parameters and to apply various
most critical quality constraints for the selection of grinding optimization techniques to grinding variables of wheel speed,
workpiece speed, depth of dressing, and lead of dressing using
*
Corresponding author. Tel.: +90 3322233344, Fax.: +90 3322412179 a multi(objective function model with a weighted approach [9(
E(mail address: iasilturk@yahoo.com
†
Recommended by Associate Editor Song Min Yoo 13].
© KSME & Springer 2012 Lee at al. [14] proposed to solve the problem of optimiza(
3588 S. Ne)eli et al. / Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595
tion for the surface grinding process by using optimizing the
grinding variables such as wheel speed, workpiece speed,
depth of dressing, and lead of dressing, using a multi(objective
function model. This research provided rough(grinding and
finish(grinding results that demonstrated the applicability of
the proposed Taguchi(sliding(based differential evolution
algorithm, and the computational results showed that this pro(
posed Taguchi(sliding(based differential evolution algorithm
can obtain better results.
Krishna and Rao [15] proposed a scatter search based opti(
mization approach to optimize the grinding parameters of
wheel speed, workpiece speed, and depth of dressing and lead
of dressing using a multi(objective function model with a
weighted approach for the surface grinding process.
Habib [16] developed a comprehensive mathematical model
for correlating the interactive and higher order influences of
various Electrical discharge machining (EDM) parameters
through RSM, utilizing relevant experimental data as obtained
through experimentation.
Another important work was presented by Agarval and Rao
[17] about grinding parameter optimization. As a result of
their study, a new analytical surface roughness model was
developed on the basis of the stochastic nature of the grinding
process, governed mainly by the random geometry and the
random distribution of cutting edges on the wheel surface
having random grain protrusion heights.
Malkin [18] investigated the process monitoring and studied
various grinding aspects such as cutting mechanisms, the spe(
cific energy and the interrelationship of the parameters. His
research showed that the grinding process had very complex
cutting mechanisms and that replicability of results was diffi(
cult to obtain under the same grinding conditions.
Shaji [19] reported a study on the Taguchi method for
evaluating process parameters in surface grinding with graph(
ite as a lubricant. They analyzd the effect of the grinding pa(
rameters (wheel speed, table speed, depth of cut and the dress(
ing mode) on the surface finish and the grinding force.
Kwak [3] showed that the various grinding parameters
caused a geometric error generated during the surface grinding,
using the Taguchi method, and that the geometric error could
be predicted by means of the response surface method. Usu(
ally, the equation for predicting cutting time is unknown dur(
ing the early stages of cutting operations.
Jeang [20] studied to determine the optimal cutting parame(
ters required to minimize the cutting time while maintaining
an acceptable quality level. He formulated an objective func(
tion using CATIA software, with assistance from the statisti(
cal method and response surface methodology (RSM). Results
showed the statistical method in cooperation with the optimal
solutions found from mathematical programming can also be
used as references for the possibility of robust design im(
provements.
This paper illustrates how response surface designs and or(
thogonal array fractions can be used in a series of grinding
experiments to formulate a high performance robust product.
Taking this research a step further, the Taguchi and response
surface methodology are used to predict the wheel vibration
and the surface roughness in the external cylindrical grinding
of the hardened AISI 8620 steel and also to select the opti(
mum grinding conditions. Also, this research would benefit
from future applications, especially as introduced in a real(
world application, such as a manufacturing plant. It is also
advantageous to perform studies similar to this in academia as
class learning exercises.
2. Profile parameters of surface roughness
Surface roughness plays an important role in determining
how a real object will interact with its environment. A good(
quality workpiece surface can improve fatigue strength, corro(
sion resistance, and thermal resistance. In addition, the final
surface roughness also affects several functional attributes of
parts like friction, wearing, light reflection, heat transmission,
coating and ability of distributing, and holding a lubricant [20(
23].
The surface roughness universally recognized standard is
ISO 4287:1997. Actually, there are various simple surface
roughness amplitude parameters used in industry, such as
roughness average ( Ra ), root(mean(square ( rms ) oughness
( Rq ), and maximum peak(to(valley roughness (Ry or R max)
[24, 25]. In quantifying surface roughness the unit Pm is used
worldwide. Therefore, the parameter Ra is used for surface
roughness in this study. Ra is specified by Eq. (1).
L
1
f ( x ) dx
L ∫0
Ra = (1)
where Ra is the arithmetic average deviation from the mean
line, L is the cut(off length, f the ordinate of the profile
curve. In this study, the average surface roughness ( Ra ) is
measured within the selected cut(off value of 0.8(2.5 mm,
depending on the magnitude of the roughness.
3. Monitoring of vibration signals
Two important roles of the monitoring system are to pro(
vide information to optimize the process and to contribute to
establishing the database, which is necessary to determine the
setup parameters. The major problems in the grinding proc(
esses are chatter vibration and surface roughness deterioration
[26]. The process is highly complex and it is affected by dy(
namic stability of the workpiece. Vibration is a direct cause of
poor workpiece quality. Hence, the quality of the parts,
namely surface finish, depends largely on the stability of the
process. Here, we are going to test the ability of the vibration
signal to follow the changes of workpiece surface roughness
to achieve high quality.
 S. Ne)eli et al. / Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595 3589

4. Design of experiments Table 1. L27 (313 ) orthogonal array.

Control factors
Trial no.
A B C
Response surface methodology is a collection of statistical 1 1 1 1
and mathematical techniques useful for developing, improving, 2 1 1 2
and optimizing processes. The field of response surface meth( 3 1 1 3
odology consists of the experimental strategy for exploring the 4 1 2 1
space of the process or independent variables, empirical statis( 5 1 2 2
tical modeling to develop an appropriate approximating rela( 6 1 2 3
tionship between the yield and the process variables, and op( 7 1 3 1
timization methods for finding the values of the process vari( 8 1 3 2
ables that produce desirable values of the response. The most 9 1 3 3
extensive applications of response surface methodology are in 10 2 1 1
the particular situations where several input variables poten( 11 2 1 2
tially influence some performance measure or quality charac( 12 2 1 3
teristic of the process [27]. Thus performance measure or 13 2 2 1
quality characteristic is called the response. The input vari( 14 2 2 2
ables are sometimes called independent variables, and they are 15 2 2 3
subject to the control of the scientist or engineer. In most re( 16 2 3 1
sponse surface methodology problems, there is a functional 17 2 3 2
relation between responses and independent variables, and this 18 2 3 3
relation can be explained using the second order model below 19 3 1 1
[28(30]. 20 3 1 2
21 3 1 3
k k
η = β 0 + ∑ βi X i + ∑ βii X i2 + ∑∑ β ij X i X j + ε (2) 22 3 2 1
i =1 i =1 i j 23 3 2 2
24 3 2 3
where η is the estimated response (surface roughness), 25 3 3 1
β 0 is constant, β i , βii and βij represent the coefficients of 26 3 3 2
linear, quadratic, and cross(product terms, respectively. X 27 3 3 3
reveals the coded variables that correspond to the studied The factor levels are shown in each experiment. On the left side, the
grinding process parameters such as depth of cut (A), feed rate numbers indicate the experiments, and the letters at the top of the
columns indicate the factors.
(B), and work speed (workpiece revolution) (C). The relation(
ship between the surface roughness and machining parameters
is expressed as follows: (S/N) ratios (dB) to evaluate the quality characteristics [31].
In this study, the Taguchi method is employed to determine
η = β 0 + β1 ( X 1 ) + β 2 ( X 2 ) + β3 ( X 3 ) + β 4 ( X 12 ) + β5 ( X 22 ) the optimal external cylindrical grinding process parameters.
(3) The configuration of orthogonal array is based on the total
( )
+ β 6 X 32 + β 7 ( X 1 X 2 ) + β8 ( X 1 X 3 ) + β 9 ( X 2 X 3 ) . degree(of(freedom ( DoFt ) of the objective function. The most
suitable array is L27 (313 ) , which needs 27 runs and has
The function that consists of the process parameters usually 26 DoFt . To check the DoFt in the experimental design for
represents the response surface graphically as shown in Figs. 7 the three levels of the test the three main factors take 6
and 8. DoFt s (3 × 2) and the remaining DoFt s are taken by interac(
tions. Using L27 (313 ) Taguchi standard orthogonal array, the
experimental results are given in Table 1. This plan was de(
veloped for establishing the quadratic model of wheel vibra(
Taguchi design methods are robust statistical techniques tion and surface roughness using response surface analysis.
developed by Genichi Taguchi to improve the quality of
manufactured goods and to optimize the process parameters. 5. Experimental procedure of grinding
Compared to traditional experimental designs that focus on
the average characteristics obtained by several experiments,
the Taguchi method uses a special design of orthogonal array A series of experiments have been conducted to evaluate
to examine the quality characteristics through a minimal num( which grinding factors affect wheel vibration and workpiece
ber of experiments. The experimental results based on the surface roughness. Then, a literature survey and a brainstorm(
orthogonal array are then transformed into signal to noise ing session helped to identify the grinding factors and their
3590 S. Ne)eli et al. / Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595

Table 2. Control factors and their levels for grinding.

Symbol Control factor Unit Level 1 Level 2 Level 3

A Depth of cut mm 0.01 0.05 0.1
B Feed rate mm rev(1 0.2 0.4 0.6
Workpiece
C rpm 175 225 275
revolution

Table 3. Mechanical properties of AISI 8620 steel.

Tensile Tensile
Thermal
Elongation Hardness Strength. Strength.
Conductivity
(%) (HB) Ultimate Yield
W/m(K
(Mpa) (Mpa)
26 212 46.6 635 360

Table 4. Chemical composition of AISI 8620 steel. % weight.

C Si Mn P S Cr Mo Ni Fe
0.2 0.25 0.79 0.02 0.006 0.5 0.18 0.48 97.6
Fig. 2. Experimental set up to measure the vibration of the process.

ilhan_daq_v01 was developed using Matlab 7 program. Fig. 2

Fig. 1. Workpiece used in the experiments.
levels for the experiment. Finally, we selected three parame(
ters that can be associated with wheel vibration and surface
roughness. The factors and levels identified are given in Table
2. The other factors such as workpiece material, wheel speed,
and grain size were constant.
The cutting experiments were carried out on the
STSUPERTEC G32A CNC grinding machine [32]. Cutting
tests using aluminum oxide ( Al2O3 ) abrasives with vitrified
bond in the grinding wheel were run.
The grinding parameters include: work piece speed, feder(
ate, depth of cut, fluid: water based fluid 4(7%; grinding
wheel: Karbosan trademark; NK60K6V 500 × 50 × 152, and
including 95(96% Al2O3. Mechanical properties, chemical
composition and geometry of workpiece material AISI 8620
which was used in the experiment are shown in Tables 3, 4
and Fig. 1 respectively.
An accelerometer was connected to the wheel head spindle
motor to measure vibration signals. Sensor model Kistler
8692C50 (Acceleration range ± 50 g, sensitivity 100 mV/g,
frequency response ± 5%) was connected to a Kistler (type
5134B) coupler, which provides a DC power and signal proc(
essing by adjustable gains and cut(off frequencies. Data acqui(
sition card was a National Instruments portable E Series NI
DAQCard(6036E with maximum acquisition rate of 200000
samples per second and 16 channels. A software called
shows the experimental setup used for vibration acquisition.
After the experiments, the surface roughness value meas(
ured by Mitutoyo Surftest SJ(201 stylus type surface texture
measuring instrument, and the mean roughness values meas(
ured from three different points of the machined surface were
calculated. In this paper a robust approach has been repre(
sented which uses both the Taguchi and response surface
methodology to obtain optimum combination for the grinding
parameters. For this purpose, the first step in the optimization
process is to determine the S/N ratio for all the experimental
runs by using the Taguchi methodology. The next step is to
find the objective function. The objective function is formu(
lated using the response surface methodology.
6. Experimental procedure of grinding
There are several types of quality characteristics, such as the
lower(the(better, the higher(the(better and the nominal(the(
better in the case of vibration and surface roughness that
should be a minimum. In this study, therefore, the smaller(the(
better type of the signal(to(noise (S/N) ratio has been used and
is defined as follows [33]:
1 n 
S N = −10log  ∑ yi2  (4)
 n i =1 
where n is the number of repeated experiments for each
combination of control factors, and yi is the observed re(
sponse at ith trial. The negative sign in Eq. (4) is for showing
the smaller(the(better quality characteristic. The response y
in this case is the wheel vibration and surface roughness, re(
spectively. Response tables of S/N ratio for vibration and sur(
 S. Ne)eli et al. / Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595 3591

Table 5. Experimental results for responses and their S/N ratios. Table 6. Response table for S/N ratios (Smaller(is(better) for Vb.

Control factor levels S/N S/N Level A B C

Vibration Surface
Exp. A B C ratio ratio 1 #1.971a #2.273a #1.388a
Vb roughness
No. Deep Feed Wrkp. for Vb for Ra
(mV/g) Ra (Pm) 2 (2.655 (2.376 (2.329
of cut rate rev. (dB) (dB)
3 (2.721 (2698 (3.630
1 0.01 0.2 175 1.08 0.20 (0.67 13.98
Max(min 0.749 0.424 2.241
2 0.01 0.2 225 1.16 0.17 (1.29 15.39
Rank 2 3 1
3 0.01 0.2 275 1.4 0.13 (2.92 17.72 a
Optimum level levels of the factors (the largest value in the column)
4 0.01 0.4 175 1.113 0.22 (0.93 13.15 Max(min = Difference between maximum and minimum S/N ratio
5 0.01 0.4 225 1.232 0.20 (1.81 13.98 response values
6 0.01 0.4 275 1.44 0.18 (3.17 14.89
7 0.01 0.6 175 1.137 0.27 (1.12 11.37 Table 7. Response table for S/N ratios (Smaller(is(better) for Ra.

8 0.01 0.6 225 1.334 0.24 (2.50 12.40 Level A B C

9 0.01 0.6 275 1.468 0.19 (3.33 14.42 1 14.15a 13.80a 11.91
10 0.05 0.2 175 1.16 0.24 (1.29 12.40 2 12.85 13.26 13.03
11 0.05 0.2 225 1.453 0.22 (3.25 13.15 3 12.50 12.43 14.55a
12 0.05 0.2 275 1.52 0.20 (3.64 13.98 Max(min 1.65 1.37 2.64
13 0.05 0.4 175 1.2 0.25 (1.58 12.04 Rank 2 3 1
14 0.05 0.4 225 1.32 0.23 (2.41 12.77 For details see Table 6.
15 0.05 0.4 275 1.539 0.19 (3.74 14.42
16 0.05 0.6 175 1.216 0.27 (1.70 11.37
17 0.05 0.6 225 1.326 0.24 (2.45 12.40
18 0.05 0.6 275 1.555 0.22 (3.83 13.15
19 0.1 0.2 175 1.208 0.29 (1.64 10.75
20 0.1 0.2 225 1.249 0.24 (1.93 12.40
Fig. 3. Main effects plot for SN ratios for wheel vibration (Vb).
21 0.1 0.2 275 1.555 0.19 (3.83 14.42
22 0.1 0.4 175 1.216 0.27 (1.70 11.37
23 0.1 0.4 225 1.263 0.23 (2.03 12.77
24 0.1 0.4 275 1.587 0.20 (4.01 13.98
25 0.1 0.6 175 1.24 0.29 (1.87 10.75
26 0.1 0.6 225 1.461 0.25 (3.29 12.04
27 0.1 0.6 275 1.618 0.20 (4.18 13.98 Fig. 4. Main effects plot for SN ratios for surface roughness (Ra).
The assigned values of the factors are given in the table (the letters
correspond to the letters in Table 1).
the control factors and how it is changed when settings of each
control factor are changed from one level to another.
face roughness are given in Table 5 along with their experi( The influence of each control factor can be more clearly
mentally measured values. presented with response graphs (see Figs. 3 and 4, respec(
The S/N ratio for each level of factors is computed based on tively). These figures reveal the level to be chosen for the
the S/N analysis using Eq. (4). A smaller value of wheel vibra( ideal cutting parameters (the level with the highest point on
tion and surface roughness is normally required in metal ma( the graphs), as well as the relative effect each parameter has
chining. Therefore, the smaller(the(better methodology of S/N on the S/N ratio (the general slope of the line). As seen in the
ratio was employed for the aforesaid responses. Regardless of S/N ratio effects graphs (Figs. 3 and 4), the slope of the line
the category of the performance characteristics, the largest which connects between the levels can clearly show the power
S/N ratio corresponds to a better performance. Therefore, the of influence of each control factor. Especially, the workpiece
optimal level of the process parameters is the level with the revolution and depth of cut are shown to have a strong effect
greatest S/N ratio. on wheel vibration and surface roughness and their S/N ratios.
The influence of each control factor on the wheel vibration The feed rate had a smaller effect, as evidenced by the shallow
and the surface roughness has been analyzed with a signal(to( slope of the lines.
noise ratio response table. Response tables of S/N ratio for An analysis of variance (ANOVA) summary table is most
wheel vibration and surface roughness are shown in Tables 6 widely used to summarize the test of the regression model, test
and 7, respectively. They show the S/N ratio at each level of of the significance factors on response and their lack(of(fit test.
3592 S. Ne)eli et al. / Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595

Table 8. Analysis of variance for Vb.

Adj Adj PC
Symbol DF F(value P(value R2
SS MS (%)
A 2 3.10 1.55 14.61 0.000 10.72
B 2 0.88 0.44 4.16 0.031 3.04
92.7
C 2 22.80 11.40 107.49 0.000 78.89
%
Error 20 2.12 0.11 7.33
Total 26 28.90
DF: degrees of freedom; SS: sum of squared deviations; F(value: F(test
value (at least 95% confidence); P(value: probability level (less than
0.05); P: percentage contribution.

Table 9. Analysis of variance for Ra.

Adj Adj PC
Symbol DF F(value P(value R2
SS MS (%)
A 2 13.56 6.78 14.29 0.000 21.42
B 2 8.54 4.27 9.00 0.002 13.49
85.0
C 2 31.68 15.84 33.38 0.000 50.06
%
Error 20 9.49 0.48 14.99
Total 26 63.28
For details see Table 8. Fig. 5. Relationship between observed and predicted responses values.

The response surface models for wheel vibration and surface
roughness were tested with the ANOVA analysis under 0.05
level of significance. The quality of the model fit was evalu(
ated by the coefficient of determination ( R 2 ). The R 2 value
is always between 0 and 1, and its magnitude indicates the
aptness of the model. For a good statistical model, the R 2
value should be close to 1.0.
Table 8 shows the results of ANOVA for wheel vibration.
Workpiece revolution is the significant cutting factor for af(
fecting wheel vibration. The changes of the depth of cut, and
feed rate in the ranges given in Table 8 have insignificant
effects on wheel vibration. Therefore, based on the S/N and
ANOVA analyses, the optimal cutting parameters for wheel
vibration is A1B1C1 i.e., A1 = 0.01 mm, B1 = 0.2 mm rev −1
and C3 = 175 rpm .
Table 9 shows the results of ANOVA for surface roughness.
Depth of cut, feed rate and workpiece revolution are the sig(
nificant cutting factors for affecting surface roughness. How(
ever, the contribution order of the cutting parameters for sur(
face roughness is workpiece revolution, then depth of cut, and
then feed rate. The optimal cutting parameters for surface
roughness is A1B1C3 i.e., A1 = 0.01 mm, B1 = 0.2 mm rev −1
and C3 = 275 rpm .
As mentioned in the section (4(1) of this study, it is desir(
able to be able to predict external grinding wheel vibration and
surface roughness according to the grinding conditions. One
means of doing this is to represent grinding outcomes mathe(
matically as a function of the applicable grinding parameters.
In this study, the response surface method was used to achieve
the aim. The quadratic response surface model depicting the
wheel vibration and surface roughness can be expressed as a
function of grinding factors such as depth of cut (A), feed rate
(B) and workpiece revolution (C). The models have been de(
veloped in a form of Eq. (3) reduced final equation in terms of
coded factors.
Vb = 1.23 + 3.02 A − 0.25B − 0.003C + 0.31AB
(5)
+ 0.004 AC − 25.88 A2 + 0.44 B 2
Ra = 0.21 + 2.34 A + 0.064 B − 1.63 AB
(6)
− 0.003 AC − 5.56 A2 + 0.15B 2
The upper models can be used to predict wheel vibration
and surface roughness at the particular design points. The
differences between measured and predicted responses are
illustrated in Fig 5. These figures indicate that the quadratic
models are capable of representing the system under the given
experimental domain.
In order to better understand the interaction effect of vari(
ables on wheel vibration and surface roughness, three(
dimensional (3D) plots for the measured responses were cre(
ated based on the model equations (Eqs. (5) and (6)). Since
each model had three variables, one variable was held con(
stant at the centre level for each plot; therefore, a total of 6
response surface plots were produced for the responses (Fig. 6
and 7).
Fig. 6 gives the 3D surface graphs for the wheel vibration.
Wheel vibration increases with increase in depth of cut, feed
rate and workpiece revolution. Wheel vibration improved with
 S. Ne)eli et al. / Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595
Fig. 6. Response surface plots showing the effect of two variables on
Vb (the other variable is held at center level). A(Deep of cut; B(Feed
rate; C(Workpiece revolution.
a decrease in all factors. Hence, a minimum amount of depth
of cut, feed rate and workpiece revolution equivalent to level 1
is required for minimum wheel vibration.
The 3D surface graphs for the surface roughness are shown
in Fig. 7. It is clear that surface roughness increases with in(
crease in depth of cut and feed rate. From Fig. 7(b) and c, if
depth of cut and feed rate are increased and workpiece revolu(
tion is decreased the surface roughness is increased. Therefore,
the ideal combination of control factors for lowest surface
roughness should consist of minimum depth of cut, minimum
feed rate and maximum workpiece revolution.
3593
Fig. 7. Response surface plots showing the effect of two variables on
Ra (the other variable is held at center level). A(Deep of cut; B(Feed
rate; C(Workpiece revolution.
7. Non#linear response optimization
Response optimization is used to help identify the combina(
tion of input variable settings that jointly optimize a single
response or a set of responses. The desirability function ap(
proach is widely used in industry for the optimization of the
multiple response process. The objective is to maximize the
desirability function. The weight, wt, can be assigned to a goal
to adjust the shape of the desirability function, shown in Fig. 8.
The value of one creates a linear ramp function between the
low value, the goal and the high value. Increased weight
moves the result towards the goal or its decrease creates the
opposite effect [2].
The minimization goal importance was set to a high value
3594 S. Ne)eli et al. / Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595

Table 10. Response optimization for wheel vibration factor components.

Global solution
Parameter Weight Goal Lower Target Upper Pre. Ra Desirability
A B C
Vb (mV/g) 0.1 Minimum 0.01 0.249 175 1.08 1.08 1.618 1.089 0.9983
Vb (mV/g) 1 Minimum 0.01 0.249 175 1.08 1.08 1.618 1.089 0.9831
Vb (mV/g) 10 Minimum 0.01 0.249 175 1.08 1.08 1.618 1.089 0.8440

Table 11. Response optimization for surface roughness factor components.

Global solution
Parameter Weight Goal Lower Target Upper Pre. Ra Desirability
A B C
Ra (Pm) 0.1 Minimum 0.01 0.2 275 0.13 0.13 0.29 0.145 0.9904
Ra (Pm) 1 Minimum 0.01 0.2 275 0.13 0.13 0.29 0.145 0.9089
Ra (Pm) 10 Minimum 0.01 0.2 275 0.13 0.13 0.29 0.145 0.3847

Weight feed rate and 275 rpm workpiece revolution with a predicted
1
0.1 Ra of 0.145 Pm.
1 Taguchi and response surface robust analysis are very suit(
d
able to evaluate the wheel vibration and surface roughness
10
0 problem as described in this paper.
Target Upper This robust optimization can increase production rate and
decrease grinding costs.
Fig. 8. Desirable curves to minimize responses.

Acknowledgment
of five, while the factors’ importance was equal and set to
values of three. The minimum seeking starts randomly from a This study was supported by the Scientific Research Pro(
hundred starting points in the design space. The numerical jects Coordinators (BAP) of Selçuk University and TUBITAK.
non(linear optimization algorithm produces a system of con( This support is greatly appreciated.
straints that can be solved as an unconstrained problem via a
penalty function approach in a downhill simplex multidimen( Nomenclature########################################################################
sional pattern search [34] (Tables 10 and 11).
Ra : Average surface roughness
Ry : Maximum peak(to(valley roughness
8. Conclusions : Root(mean(square ( rms ) roughness
Rq
This paper applies the taguchi and response surface meth( Vb : Vibration Voltage levels
odology from the point of view of external grinding. The con( β0 : Constant
cept of S/N ratio from Taguchi methodology was used to R2 : The coefficient of determination
measure the variance of wheel vibration and surface rough( DOE : Statistical design of experiments
ness. The response surface methodology was used to formu( S/N : The signal(to(noise ratio
late an objective function and non(linear optimization to ob( DoFt : Total degree(of(freedom of the objective function
tain the optimum condition for the process. By the experimen(
tal and statistical results, the conclusions were as follows.
References
The statistical analysis by using Taguchi and response sur(
face methodology has demonstrated that the depth of cut and [1] A. G. Krishna, Optimization of surface grinding operations
workpiece revolution are the dominant parameters among the using a differential evolution approach, J. Mater. Process.
three controllable factors that influence the wheel vibration Technol, 183 (2007) 202(209.
and surface roughness in the external grinding operation dur( [2] P. Krajnik, J. Kopac and A. Sluga, Design of grinding fac(
ing AISI 8620 steel machining. tors based on response surface methodology, J. Mater. Proc&
The optimum conditions to minimize wheel vibration were ess. Technol, 162 (163) (2005) 629(636.
determined to be 0.01 mm depth of cut, 0.249 mm rev(1 feed [3] J. S. Kwak, Application of Taguchi and response surface
rate and 175 rpm workpiece revolution with a predicted Vb of methodologies for geometric error in surface grinding proc(
1.089 mV/g. For surface roughness, the optimum condition ess, Int. J. Mach. Tools & Manuf, 45 (3) (2005) 327(334.
was determined to be 0.01 mm depth of cut, 0.2 mm rev(1 [4] D. C. Montgomery, Design and Analysis of Experiments,
 S. Ne)eli et al. / Journal of Mechanical Science and Technology 26 (11) (2012) 3587~3595
(4th ed.) Wiley, New York, 1997.
[5] H. W. Lee and W.T. Kwon, Determination of the minute
range for RSM to select the optimum cutting conditions dur(
ing turning on CNC lathe, 24 (8) (2010) 1637(1645.
[6] S. H. Baek, S. H. Hong, Seok(Swoo Cho, Deuk(Yul Jang
and Won(Sik Joo, Optimization of process parameters for
recycling of mill scale using Taguchi experimental design,
24 (10) (2010) 2127(2134.
[7] J. H. Jung and W. T. Kwon, Optimization of EDM process
for multiple performance characteristics using Taguchi
method and Grey relational analysis, 24 (5) (2010) 1083(1090.
[8] A. Aggarwal, H. Singh, P. Kumar and M. Singh, Optimizing
power consumption for CNC turned parts using response
surface methodology and Taguchi's technique(A compara(
tive analysis, Journal of Materials Processing Technology,
200 (1(3) (2008) 373(384.
[9] X. Lin and H. Li, Enhanced Pareto particle swarm approach
for multi(objective optimization of surface grinding process,
Proceedings of the second International Symp. Intelligent In&
formation Technology Application, Shanghai, China, (2008)
618(623.
[10] J. Kopac, M. Bahor and M. Sokovic, Optimal machining
parameters for achieving the desired surface roughness in
fine turning of cold pre(formed steel workpieces, Int. J.
Mach. Tools & Manuf, 42 (6) (2002) 707(716.
[11] M. Tomas and Y. Beauchamp, Statistical investigation of
modal parameters of cutting tools in dry turning, Int. J. Mach.
Tools & Manuf, 43 (11) (2003) 1093(1106.
[12] T. R. Lin, The use of reliability in the Taguchi method for
the optimization of the polishing ceramic gauge block, Int. J.
Adv. Manuf. Technol, 22 (3(4) (2003) 237(242.
[13] T. R. Lin, Experimental design and performance analysis of
TiNcoated carbide tool in face milling stainless steel, J. Ma&
ter. Process. Technol, 127 (1) (2002) 1(7.
[14] K. M. Lee, M. R. Hsu, J. H. Chou and C. Y. Guo, Improved
differential evolution approach for optimization of surface
grinding process, Expert Syst. Appl., 38 (2011) 5680(5686.
[15] A. G. Krishna and K. M. Rao, Multi(objective optimisation
of surface grinding operations using scatter search approach,
Int. J. Adv. Manuf. Technol. 29 (2006) 475(480.
[16] S. S. Habib, Study of the parameters in electrical discharge
machining through response surface methodology approach,
Appl. Math. Model., 33 (12) (2009) 4397(4407.
[17] S. Agarwal and P. Venkateswara Rao, Modeling and pre(
diction of surface roughness in ceramic grinding, Int. J.
Mach. Tools & Manuf. 50 (2010) 1065(1076.
[18] T. W. Hwang and S. Malkin, Upper bound analysis for
specific energy in grinding of ceramics, Wear, 231 (2)
(1999) 161(171.
[19] S. Shaji and V. Radhakrishnan, Analysis of process pa(
rameters in surface grinding with graphite as lubricant based
of the Taguchi method, J. Mater. Process. Technol. 141 (1)
(2003) 51(59.
[20] A. Jeang, Robust cutting parameters optimization for pro(
duction time via computer experiment, Appl. Math. Model.,
3595
35 (3) (2011) 1354(1362.
[21] M. S. Lou, J. C. Chen and C. M. Li, Surface Roughness
Prediction Technique For CNC End(Milling, Journal of In&
dustrial Technology, 15 (1) (1999).
[22] S. Ne1eli, S. Yaldız and E. Turke1, Optimization of tool ge(
ometry parameters for turning operations based on the response
surface methodology, Measurement, 44 (2011) 580(587.
[23] I. Asilturk and M. Cunkas, Modelling and prediction of
surface roughness in turning operations using artificial neu(
ral network and multiple regression method, Expert Syst.
Appl., 38 (5) (2011) 5826(5832.
[24] http://www.predev.com/smg/parameters.htm, PDI Web(
master, Surface Metrology (2011, March 10).
[25] O. Çolak, C. Kurbanoğlu and M. C. Kayacan, Milling sur(
face roughness prediction using evolutionary programming
methods, Mater. Design, 28 (2) (2007) 657(666.
[26] I. Inasaki, Sensor fusion for monitoring and controlling
grinding processes, Int. J. Adv. Manuf. Technol, 15 (1999)
730(736.
[27] K. Dehghania, A. Nekahia and M. A. M Mirzaieb, Opti(
mizing the bake hardening behavior of Al7075 using re(
sponse surface methodology, Mater. Design, 31 (4) (2010)
1768(1775.
[28] M. C. Kathleen, Y. K. Natalia and R. Jeff, Response surface
methodology, center for computational analysis of social and
organizational systems (CASOS) Technical Report, 2004.
[29] H. Raymond, C. Douglas, Montgomery and M. Christine
Anderson(cook, process and product optimization using designed
experiments, John Wiley & Sons, Inc. Third Edition, 2009.
[30] K. M. Desai, S. A. Survase, P. S. Saudagar, S. S. Lele and
R.S. Singhal, Comparison of artificial neural network (ANN)
and response surface methodology (RSM) in fermentation
media optimization: Case study of fermentative production
of scleroglucan, Biochem. Eng. J., 41 (2008) 266(270.
[31] J. L. Rosa, A. Robin, M. B. Silva, C. A. Baldan, A. Carlos
and M. P. Peres, Electrodeposition of copper on titanium
wires: Taguchi experimental design approach, J. Mat. Proc&
essing Tech., 209 (2009) 1181(1188.
[32] L. Celik, Monitoring vibration in grinding process and
regression modelling of surface roughness, M. S. Thesis,
Selçuk University, Turkey, 2010.
[33] G. Taguchi, Introduction to quality engineering, Tokyo:
Asian Productivity Organization, 1990.
[34] I. Maros, Computational techniques of the simplex method,
Kluwer Academic Publishers, 2003.
Süleyman Ne1eli received his B.Sc.
degree in Technical Education Faculty
in Gazi University, Turkey, in 2002. He
is currently a Ph.D student in the School
of Mechanical Engineering, Selçuk
University, Turkey. His research inter(
ests include chatter vibration control,
process damping analysis and statistical
optimization.