Int J Adv Manuf Technol (2011) 57:10431051

DOI 10.1007/s00170-011-3343-1

ORIGINAL ARTICLE

Prediction and analysis of surface roughness characteristics

of a non-ferrous material using ANN in CNC turning
Chinnasamy Natarajan & S. Muthu & P. Karuppuswamy

Received: 5 October 2010 / Accepted: 11 April 2011 / Published online: 26 April 2011
# Springer-Verlag London Limited 2011

Abstract Surface roughness, an indicator of surface quality

is one of the most-specified customer requirements in a
machining process. For efficient use of machine tools,
optimum cutting parameters (speed, feed, and depth of cut)
are required. So it is necessary to find a suitable
optimization method which can find optimum values of
cutting parameters for minimizing surface roughness. The
turning process parameter optimization is highly constrained and non-linear. In this work, machining process
has been carried out on brass C26000 material in dry
cutting condition in a CNC turning machine and surface
roughness has been measured using surface roughness
tester. To predict the surface roughness, an artificial neural
network (ANN) model has been designed through feedforward back-propagation network using Matlab (2009a)
software for the data obtained. Comparison of the experimental data and ANN results show that there is no
significant difference and ANN has been used confidently.
The results obtained conclude that ANN is reliable and
accurate for predicting the values. The actual Ra value has
been obtained as 1.1999 m and the corresponding
predicted surface roughness value is 1.1859 m, which
implies greater accuracy.

C. Natarajan (*) : P. Karuppuswamy

Department of Mechanical Engineering,
Sri Ramakrishna Engineering College,
Coimbatore, Tamil Nadu, India 641 022
e-mail: cnat6666@gmail.com
S. Muthu
Department of Mechanical Engineering,
Sri Eshwar College of Engineering,
Coimbatore, Tamil Nadu, India 641 202

Keywords CNC turning process . Non-ferrous material .

Surface roughness . Artificial neural network (ANN) .
Optimization

1 Introduction
Nowadays, due to the increasing demand of higher
precision components for its functional aspect, surface
roughness of a machined part plays an important role in
the modern manufacturing process. Turning is a machining
operation, which is carried out on lathe. The quality of the
surface plays a very important role in the performance of
turning as a good quality turned surface significantly
improves fatigue strength, corrosion resistance, or creep
life. Surface roughness also affects several functional
attributes of parts, such as, contact causing surface friction,
wearing, light reflection, heat transmission, ability of
distributing and holding a lubricant, load bearing capacity,
coating, or resisting fatigue. Therefore, the desired finish
surface is usually specified and the appropriate processes
are selected to reach the required quality [1]. To achieve the
desired surface finish, a good predictive model is required
for stable machining. Generally, these models have a
complex relationship between surface roughness and
operational parameters, work materials, and chip breaker
types. Artificial neural networks (ANNs) are information
processing systems, and since their inception, they have
been used in several areas of engineering applications. In
experimental studies, some of the operating conditions of a
system can be investigated. For this type of experimental
work, experts and special equipment are needed. It also
requires too much time and high cost.
ANNs have been trained to solve non-linear and complex
problems that are not exactly modeled mathematically. ANNs

1044

eliminate the limitations of the classical approaches by

extracting the desired information using the input data.
Applying ANN to a system needs sufficient input and output
data instead of a mathematical equation. Furthermore, it can
continuously re-train for new data during the operation, thus it
can adapt to changes in the system. ANNs can also be used to
deal with problems with incomplete and imprecise data.
In this work, artificial neural network model have been
developed through feed-forward back-propagation network
method to predict the surface roughness of brass C26000
metal in the computer numerical control (CNC) turning
process.

2 Literature survey
Since turning is the primary operation in most of the
production processes in the industry, surface finish of
turned components has greater influence on the quality of
the product. Surface finish in turning has been found to be
influenced in varying amounts by a number of factors such
as feed rate, work material characteristics, work hardness,
unstable built-up edge, cutting speed, depth of cut, cutting
time, tool nose radius. According to these parameters, a
detailed literature survey is carried out as follows. Adeel H.
Suhail et al. [2] focused on an experimental study to
optimize the cutting parameters using workpiece surface
temperature and surface roughness by employing Taguchi
techniques. Al-Ahmari [3] developed empirical models for
tool life, surface roughness, and cutting force for turning
operations. Two important data mining techniques used
were response surface methodology and neural networks.
Avisekh Banerjee et al. [4] conducted a study of feasibility
of on-line monitoring of surface roughness in turning
operations using a developed opto-electrical transducer.
Regression and neural network (NN) models were
exploited to predict surface roughness and compared to
actual and on-line measurements. Azouzi et al. [5] proposed
an on-line prediction of surface finish and dimensional
deviation in turning using neural network-based sensor
fusion. Bajic et al. [6] focused on modeling of machined
surface roughness and optimization of cutting parameters in
face milling and examined the influence of cutting
parameters on surface roughness in face milling. David et
al. [7] described an approach to predict surface roughness
in a high-speed end-milling process and used artificial
neural networks and statistical tools to develop different
surface roughness predictors. Franic Cus [8] used binary
coded genetic algorithm (BCGA) for the optimization of
cutting parameters. This genetic algorithm optimizes the
cutting conditions having an influence on production cost,
time, and quality of the final product. Groover [9] depicted
the impact of three factors, namely, the feed, nose radius,

Int J Adv Manuf Technol (2011) 57:10431051

and cutting-edge angles, on surface roughness. Hossain et

al. [10] developed an artificial neural network algorithm for
predicting the surface roughness in end milling of Inconel
718 alloy. Kadirgama et al. [11] optimized the surface
roughness in milling of mold aluminum alloys (AA6061-T6)
based on response surface method (RSM) and radian basis
function network (RBFN) and concluded that RBFN predict
surface roughness more accurately compared to RSM. Huang
et al. [12] predicted in-process surface roughness through
multiple regression model in turning operation via accelerometer. Mahdavinejad and Sharifi Bidgoli [13] optimized the
machining parameters to predict the surface roughness with
the combination of adaptive neural fuzzy intelligent system
in dry surface condition. Lou et al. [1] examined multiple
regression model for surface finish prediction in end-milling
operations and concluded that through experimentation, the
system proved capable of predicting the surface roughness
(Ra) with about 90% accuracy. Muammer et al. [14]
addressed regression analysis and neural network-based
models used for the prediction of surface roughness and
compared for various cutting conditions in turning. Oktem
[15] utilized response surface methodology to create an
efficient analytical model for surface roughness in terms of
cutting parameters: feed, cutting speed, axial depth of
cut, radial depth of cut, and machining tolerance. Paulo
et al. [16] predicted surface roughness model using artificial
neural network to investigate the effects of cutting conditions
during turning of free machining steel, 9SMnPb28k(DIN).
Their experiments were planned as per L27 orthogonal
array with three levels defined for each of the factors in
order to develop the knowledge base for ANN training
using error back-propagation training algorithm (EBPTA).
Parappagoudar et al. [16] tackled both the forward as well
as reverse mappings in green sand mold system by using a
back-propagation neural network (BPNN) and a geneticneural network (GA-NN) also showed that GA-NN
outperforms than the BPNN and that both the NN
approaches are able to carry out the reverse mapping
effectively. Parappagoudar et al. [17] used design of
experiments (DOE) technique with response surface
methodology to develop both linear and non-linear
models, to establish the inputoutput relationships in
green sand mold system and concluded that the developed
models provided minimum percentage of deviation in
prediction. Saad and Abbas [18] predicted the surface
roughness (Ra) using multiple regression model with
average percentage deviation of 7.8%, or 92.2%, accuracy
from training data, and from testing data set that was not
included in the multiple regression analysis with average
percentage deviation of 11.95%, or accuracy of 88%, for
4-Flute end mill.
Sakir Tasdemir et al. [19] worked on the prediction of
surface roughness using artificial neural network in lathe

Int J Adv Manuf Technol (2011) 57:10431051

and investigated the effect of tool geometry on surface

roughness in universal lathe and carried out machining
process on AISI 1040 steel in dry cutting condition using
various insert geometry at depth of cut of 0.5 mm. Srikanth
et al. [20] proposed a Real Coded Genetic Algorithm
(RCGA) to find optimum cutting parameters and explained
various issues of RCGA and its advantages over the
existing approach of BCGA. Suresh [21] developed
optimum surface roughness predictive model using BCGA.
This GA program gives minimum and maximum values of
surface roughness and their respective optimal machining
conditions. Uros Zuperl [22] proposed a neural networkbased approach to complex optimization of cutting parameters
and described the multi-objective technique of optimization of
cutting conditions by means of the neural networks taking
into consideration the technological, economic, and organizational limitations. Yang [23] used Taguchi method for
design optimization on surface quality. An orthogonal
array, the signal-to-noise (S/N) ratio and the analysis of
variance (ANOVA) were employed to investigate the
cutting characteristics.
Optimization of machining parameters not only increases
the utility for machining economics, but also the product
quality to a great extent. The dynamic nature and
widespread usage of turning operations in practice have
raised a need for seeking a systematic approach that can
help to set-up turning operations in a timely manner and
also to achieve the desired surface roughness quality.
After a detailed literature survey it is inferred that there
are no appropriate surface recognition models for machining Brass C26000 metal in CNC turning. The experimental
works were conducted in a leading pump manufacturing
company. The seamless pipe which is being manufactured
in the pump industry made up of Brass C26000 requires
more surface finish in the inner surface area that is
considered in this work.
This work predicts the surface roughness based on ANN
technique by using back-propagation trained neural networks for Brass C26000 metal in CNC turning process. The
rest of this paper is organized as follows. Section 3
describes the problem description. Development of neural
network model, execution of experiments, architecture of
the proposed artificial neural network and artificial neural
network testing are given in Section 4. Section 5 discusses
the results. Finally, the conclusion has been summarized in
Section 6.

1045

part to be machined and to avoid trial and error

approaches to set-up turning conditions in order to achieve
the desired surface roughness. The goal of which is to
predict surface roughness (Ra) under multiple cutting
conditions determined by spindle speed, feed rate, and
depth of cut. Surface roughness would be measured directly
by surface roughness measuring instruments. Experimental
results are expected to show that parameters of spindle
speed, feed rate, and depth of cut could predict surface
roughness (Ra) under different combinations of cutting
parameters.

4 Development of neural network model

In the present work, a feed-forward back-propagation
training algorithm is employed for predicting the surface
roughness in CNC turning process. Training begins with
all weights set to random numbers. For each data record,
the predicted value is compared to the desired (actual)
value and the weights are adjusted to move the
prediction closer to the desired value. Many cycles are
made through the entire set of training data with the
weights being continually adjusted to produce more
accurate predictions.
4.1 Execution of experiments
Figure 1 shows the experimental set-up and the tests were
performed on a CNC turning center. The training data set
was developed through experiments based on L 27
Taguchi orthogonal array [16]. The orthogonal array were
assigned to cutting speed (v), feed rate (f), and depth of cut
(d), respectively, and accordingly 27 experiments were

3 Problem definition
Most of the measurement techniques have limitations to
their in-process use. The purpose of the analysis is to
develop techniques to predict the surface roughness of a

Fig. 1 Experimental set-up

1046

Int J Adv Manuf Technol (2011) 57:10431051

Table 1 Machining parameters and levels

Parameter

Cutting speed (v)

Feed rate (f)
Depth of cut (d)

Unit

rpm
mm/rev
mm

Levels
1

2,500
0.05
0.2

3,250
0.12
0.3

3,500
0.15
0.4

performed under different combinations of the factor

levels.
The brass C26000 metal specimens with dimensions
each of diameter 20 mm and length of 85 mm was clamped
onto to the turret of the machine table. Surface roughness
measurement was done in the off-line with the usage of
TIME TR100 surface roughness tester. The radius of the
stylus point is 10.02.5 micron and the traversed length is
6 mm.
The experimental set-up consists of a CNC machine,
battery unit for backup purpose, power supply, and the
whole set-up is connected to the computer interface. A
computer numeric control (CNC) program was written to
perform the turning process. According to the acceptable
ranges of cutting speed and feed rate when cutting brass
with a CNMG 120408 insert with a tool holder
PCLNR120408 and nose radius of 0.8, and then an NC
program was written to execute the cutting operations.
Three levels of each factor were selected as shown in
Table 1.
Including test runs, there were totally 36 specimens
machined in this experiment and the specimens are shown
in Fig. 2. All the specimens in this experiment were
machined without coolant. At high cutting speeds, there
will be no BUE. Furthermore, in CNC, sculptured tool
holders are used, and hence, the inserts are replaced in its
position, depending on the wear rate of the tool inserts. In
addition, the following assumptions were made: (1) The

Fig. 2 Machined specimens

cutting tools used are identical in property. (2) The hardness

of each work piece is same throughout the length of the
work piece. (3) Surface roughness values are not affected
by abnormal factors. (4) Surface roughness values measured within the measuring area are sufficient to represent
the roughness of entire work piece. (5) The effect of

Table 2 Experimental result of training patterns and actual surface

roughness values
Sample
number

Cutting
speed (v)
(rpm)

Training set
1
2
3
4
5
6
7
8
9
10
11
12
13

Feed rate
(f) (mm/
rev)

Depth of
cut (d)
(mm)

Actual
roughness
(Ra) (m)

2,500
2,500
2,500
2,500
2,500
2,500
2,500
2,500
2,500
3,250
3,250
3,250
3,250

0.05
0.05
0.05
0.12
0.12
0.12
0.15
0.15
0.15
0.05
0.05
0.05
0.12

0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4
0.2

0.7933
0.8433
0.9467
1.3267
1.1633
1.4467
0.96
1.0033
0.9733
0.81
0.9033
0.9967
0.96

14
3,250
15
3,250
16
3,250
17
3,250
18
3,250
19
3,500
20
3,500
21
3,500
22
3,500
23
3,500
24
3,500
25
3,500
26
3,500
27
3,500
Test readings
28
2,750
29
2,750
30
2,750
31
3,000

0.12
0.12
0.15
0.15
0.15
0.05
0.05
0.05
0.12
0.12
0.12
0.15
0.15
0.15

0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4

1.24
1.4867
0.9633
1.0967
1.1633
0.94
0.9533
1.11
0.8
1.23
1.09
1.01
1.06
1.0467

0.08
0.08
0.08
0.1

0.15
0.25
0.35
0.15

1.07
1.0667
1.0533
1.2833

32
33
34
35
36

0.1
0.1
0.13
0.13
0.13

0.25
0.33
0.15
0.25
0.35

1.41
1.0567
1.4433
1.1233
1.2933

3,000
3,000
3,300
3,300
3,300

Int J Adv Manuf Technol (2011) 57:10431051

1047

approach angle is not considered. (6) Vibration is negligible.

(7) Tool nose radius is constant.
After turning all the specimens, the surface roughness
(Ra) was measured by using the surface roughness tester
and shown in Table 2. The measurement was done
separately and the measured Ra values are utilized for the
purpose of training the developed neural networks. The
parameters which influence the surface roughness are taken
into consideration. A combined set of test values are also
obtained with the values of surface roughness, which can be
compared after implied with the neural network procedure.
4.2 Architecture of the proposed artificial neural network
Fig. 4 Variation of mean squared error (MSE) with number of epochs

Figure 3 shows the developed architecture of artificial

neural network and comprises one input layer with 3
neurons, one output layer with 1 neuron and one hidden
layer with 9 neurons in the layers, respectively.
The links with synaptic weights are connected between
neurons and the back-propagation training algorithm is
based on weight updates so as to minimize the sum of
squared error for K number of output neurons, given as
E

K 

1X
dk:p  ok:p
2 k1

where dk,p =desired output for the pth pattern. The weights
of the links are updated as

For hidden layer:


X
d pk wkj ; j 1; ::::J
d pj opj 1  opj

where J is the number of neurons in the hidden layer. The

training process is initialized by assigning small random
weight values to all the links. The inputoutput patterns are
presented one by one and updating the weights each time.
The mean square error (MSE) at the end of each epoch due
to all patterns is computed as
MSE

2
1 XNP XK 
d

o
kp
kp
p1
K1
NP

where n is the learning step, is the learning rate and is

the momentum constant. In Eq. 4, pj is the error term,
which is given as follows:
For output layer:



d pk dkp  okp 1  okp ; k 1; ::::K
3

where NP=number of training patterns.

The training process will be terminated when the
specified goal of MSE or maximum number of epochs is
achieved. The activation function for the input and the two
hidden layers is chosen as tansigmoidal function. The
activation function for the output layer is chosen as pure
linear function. The network is then simulated for the input
values and a graph is plotted between the output and target

Fig. 3 ANN architecture with a single hidden layer

Fig. 5 Correlation of the predicted and actual Ra values

wjin1 wjin hdpj opi awjin

1048

Int J Adv Manuf Technol (2011) 57:10431051

Table 3 Experimental results and predicted reading and actual roughness

Sample
number

Cutting
Speed (v)
(rpm)

Feed rate
(f) (mm/
rev)

Depth of
cut (d)
(mm)

Training set
1
2,500
2
2,500
3
2,500
4
2,500
5
2,500
6
2,500
7
2,500
8
2,500
9
2,500

0.05
0.05
0.05
0.12
0.12
0.12
0.15
0.15
0.15

10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

3,250
3,250
3,250
3,250
3,250
3,250
3,250
3,250
3,250
3,500
3,500
3,500
3,500
3,500
3,500
3,500
3,500
3,500

Test readings
28
2,750
29
2,750
30
2,750
31
3,000
32
3,000
33
3,000
34
3,300
35
3,300
36
3,300

Predicted roughness (Ra)

Actual
roughness
(m)

Difference

% Error

Trial
1

Trial
2

Trial
3

Average

0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4

0.95
1.01
1.07
1.44
1.33
1.56
1.13
1.17
1.12

0.93
0.96
1.08
1.47
1.3
1.59
1.1
1.14
1.09

0.92
0.98
1.11
1.19
1.28
1.61
1.07
1.12
1.13

0.93333333
0.98333333
1.08666667
1.36666667
1.30333333
1.58666667
1.1
1.14333333
1.11333333

0.7933
0.8433
0.9467
1.3267
1.1633
1.4467
0.96
1.0033
0.9733

0.140033333
0.140033333
0.139966667
0.039966667
0.140033333
0.139966667
0.14
0.140033333
0.140033333

17.65200218
16.60539942
14.78469068
3.012487123
12.0375942
9.674892284
14.58333333
13.95727433
14.38747902

0.05
0.05
0.05
0.12
0.12
0.12
0.15
0.15
0.15
0.05
0.05
0.05
0.12
0.12
0.12
0.15
0.15
0.15

0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4
0.2
0.3
0.4

0.94
1.01
1.14
1.13
1.36
1.62
1.1
1.24
1.33
1.06
1.11
1.23
0.93
1.35
1.24
1.14
1.18
1.19

0.95
1.05
1.15
1.08
1.38
1.63
1.08
1.26
1.29
1.08
1.07
1.25
0.93
1.35
1.24
1.14
1.18
1.19

0.96
1.07
1.12
1.09
1.4
1.63
1.13
1.21
1.29
1.1
1.1
1.27
0.93
1.39
1.21
1.15
1.21
1.18

0.95
1.04333333
1.13666667
1.1
1.38
1.62666667
1.10333333
1.23666667
1.30333333
1.08
1.09333333
1.25
0.93
1.36333333
1.23
1.14333333
1.19
1.18666667

0.81
0.9033
0.9967
0.96
1.24
1.4867
0.9633
1.0967
1.1633
0.94
0.9533
1.11
0.8
1.23
1.09
1.01
1.06
1.0467

0.14
0.140033333
0.139966667
0.14
0.14
0.139966667
0.140033333
0.139966667
0.140033333
0.14
0.140033333
0.14
0.13
0.133333333
0.14
0.133333333
0.13
0.139966667

17.28395062
15.50241706
14.0430086
14.58333333
11.29032258
9.414587117
14.53683518
12.76253001
12.0375942
14.89361702
14.6893248
12.61261261
16.25
10.8401084
12.8440367
13.20132013
12.26415094
13.3721856

0.08
0.08
0.08
0.1
0.1
0.1
0.13
0.13
0.13

0.15
0.25
0.35
0.15
0.25
0.33
0.15
0.25
0.35

1.21
1.2
1.2
1.44
1.55
1.21
1.57
1.24
1.44

1.19
1.21
1.18
1.43
1.55
1.19
1.58
1.27
1.42

1.23
1.21
1.2
1.4
1.55
1.19
1.6
1.28
1.44

1.21
1.2067
1.1933
1.4233
1.55
1.1967
1.5833
1.2633
1.4333

1.07
1.0667
1.0533
1.2833
1.41
1.0567
1.4433
1.1233
1.2933

(neural network output) values. The network created is trained

for the input and output values. The stopping criterion for
training was number of epochs and is given as 590.
The network is again simulated for the input values and the
target values of the experiments conducted. The input values for

0.14
0.14
0.14
0.14
0.14
0.14
0.14
0.14
0.14
Average

13.08411215
13.12458986
13.29155986
10.90937427
9.929078014
13.24879341
9.699993071
12.46327784
10.82502126
12.93591353

the test readings are then given and the network is trained. The
target value is then obtained and compared with actual output.
The predicted Ra values are compared with the actual Ra values
and the predicted Ra values obtained from the present study
show minimal in variations. The parameters taken could be

Int J Adv Manuf Technol (2011) 57:10431051

1049

Surface Plot of Ra vs Speed, Feed

confidently used for the above method for predicting the Ra

values. The behaviors of the parameters are also noted.
The predicted value of Ra is compared with the
respective measured average values of Ra and the absolute
percentage error is computed, which is given as


 Ra ;Actual  Ra ;predicted 

x100
6
% Absolute error 

Ra ;Actual
where Ra, actual is the measured value (average) and Ra,
predicted is the ANN predicted value of the response for ith
trial (Figs. 4 and 5).

1.4
Ra

1.2
1.0

3500

0.8

3000

5 Results and discussion

0. 10
Feed

The actual roughness values have been calculated for each

set of readings and the same is compared with predicted
roughness values. The behavior of roughness with various
parameters is analyzed and the influence of each parameter
over surface roughness is identified from the experiments
carried out. The percentage deviation between actual
roughness values and predicted roughness values have
been obtained and shown in Table 3 and also it is calculated
that the average percentage of error is 12.93%.
5.1 Comparison of graphical results
Figure 6 shows comparison between actual and predicted
roughness values and it is observed that there is a good
agreement between experimental and predicted Ra values.
The test values obtained from the neural network after
training the values are found to be closer to the experimental
values. The 3D plots were drawn using Minitab 5.0 software
package to identify the influence of the parameters over
surface roughness for the experimental values. Optimization
of cutting parameters can be obtained for the usage of the
same in machining area in mass production. This will reduce
the inspection of the product, which is a quality check in any
industry. The material is widely used in the pump industry

Speed

0.05
2500
0.15

Fig. 7 Interaction plot for surface roughness vs. speed, feed

where surface roughness plays an important role. The

methodology using ANN for predicting parameters are
utilized for the same actual roughness values. The test values
obtained from the neural network after training the values are
found to be closer to the experimental values.
The 3D plots were drawn to identify the influence of the
parameters over surface roughness for the experimental
values. Optimization of cutting parameters can be obtained
for the usage of the same in machining area in mass
production. This will reduce the inspection of the product,
which is a quality check in any industry. The material is
widely used in the pump industry where surface roughness
plays an important role. The methodology using ANN for
predicting parameters are utilized for the same.
Figure 7 shows the interaction plot between surface
roughness and speed with respect to feed. If the feed rate
increases gradually towards the speed, the surface rough-

Surface Plot of Ra vs Feed, Speed

1.4

Ra

1.2
1.0

0.15

0.8

0.10
2500
3000

S peed

Fig. 6 Comparison of actual and predicted roughness values

0.05
3500

Fig. 8 Interaction plot for surface roughness vs. feed, speed

Feed

1050

Int J Adv Manuf Technol (2011) 57:10431051

Surface Plot of Ra vs DOC, Speed

Surface Plot of Ra vs DOC, Feed

1.4
Ra

1.4

1.2

Ra

1.0

1.0

0.4

0.8

0.3
3000

0.4

0.8

DOC

2500
Speed

1.2

0.3

DOC

0.05

0. 2

0. 10

3500

0. 2
0.15

Feed

Fig. 9 Interaction plot for surface roughness vs. DOC, speed

Fig. 11 Interaction plot for surface roughness vs. DOC, feed

ness value decreases proportionately. The surface roughness

will be improved by increasing the feed for higher speeds.
Figure 8 shows the interaction plot between surface
roughness and f with respect to speed. If the Speed is
increased with respect to the feed, the surface roughness
value increases and decreases. The surface roughness will
be improved for higher speeds by increasing the feed rate.
Figure 9 shows the interaction plot between surface
roughness and depth of cut with respect to speed. Depth of
cut (DOC) influences more on surface roughness and if
DOC is increased, the surface roughness values will also
increase for variable speeds. The roughness values are
decreased for minimum DOC values with respect to speed
Figure 10 shows the interaction plot between surface
roughness and speed with respect to DOC. If the speed is
increased gradually, the surface roughness value increases with
respect to the DOC. The surface roughness will be improved

by decreasing the speed for depth of cut values. The cutting

parameters play an important role in obtaining the surface
roughness of a machined part. These variables are independent
and hence it is analyzed with a methodology using ANN to
obtain a model which will be useful for the industries.
Figure 11 shows the interaction plot between surface
roughness and DOC with respect to feed. If the feed rate is
increased gradually towards the DOC, the surface roughness value decreases .The surface roughness will be
improved by increasing the feed for DOC values.
The surface roughness value is higher when the feed rate
is not increased for higher depth of cut values. The DOC
plays an important role as a parameter in CNC turning.
Hence, the parameters are analyzed for obtaining improved
surface roughness characteristics. Optimization of the same
may be done with the help of ANN.

Surface Plot of Ra vs Feed, DOC

Surface Plot of Ra vs Speed, DOC

1.4

1.4
Ra

1.2

Ra

1.0

3500

0.8

3000

1.2
1.0

0.15

0.8

0.10

Speed

0.2

0.2
0.3
DOC

2500
0.4

Fig. 10 Interaction plot for surface roughness vs. DOC, speed

0.3
DOC

0.05
0.4

Fig. 12 Interaction plot for surface roughness vs. DOC, speed

Feed

Int J Adv Manuf Technol (2011) 57:10431051

Figure 12 shows the interaction plot between surface

roughness and feed with respect to DOC. If the feed rate is
increased gradually towards the DOC, the surface roughness
value decreases.

6 Conclusions
In this work, 36 specimens which are made up of the brass
C26000 material have been machined in a CNC turning
machine and then a TIME TR 100 surface roughness tester was
used to measure the roughness average (Ra) values of all the
specimens. The model was developed to predict the surface
roughness of Brass C26000 through Artificial Neural Networks technique by utilizing feed-forward back-propagation
method. This type of model has been evaluated by means of
the percentage deviation between the predicted Ra values and
the actual Ra values. The important conclusions drawn from
the present research are summarized as follows:
1. The surface roughness could be effectively predicted by
using spindle speed, feed rate, and depth of cut as the
input variables.
2. Considering the individual parameters, feed rate has
been found to be the most influencing parameter,
followed by spindle speed and depth of cut.
3. Model (including interaction terms), considering the
interaction between the individual parameters, could
achieve an accuracy of 87.07%.
4. The average actual roughness Ra value has been
obtained as 1.1999 m and the corresponding predicted
surface roughness value is 1.1859 m.
5. As the spindle speed increases, for lower feed rates, the
surface roughness decreases, for higher feed rates, the
surface roughness changes considerably.
6. The depth of cut influences the surface roughness
considerably for a given feed rate.
7. The increase in feed rate causes the surface roughness
to increase and then decrease. For lower depth of cut,
as the feed rate increases, surface roughness decreases
and then increases.

References
1. Lou MS, Chen JC, Li M (1998) Surface roughness prediction for
CNC End milling, materials and processes quality control
manufacturing. J Ind Technol 15(1):26
2. Suhail AH, Ismail N (2010) Optimization of cutting parameters
based on surface roughness and assistance of workpiece surface
temperature in turning process. Amer J Eng Appl Sci 3(1):102108
3. Al-Ahmari AMA (2007) Predictive machinability models for a
selected hard material in turning operations. J Mater Process
Technol 1(190):305311

1051
4. Banerjee A, Bordatchev EV, Choudhury SK (2009) On-line
monitoring of surface roughness in turning operations with optoelectrical transducer. Int J Manuf Res 4(1):5773
5. Azouzi R, Guillot M (1997) On-line prediction of surface finish
and dimensional deviation in turning using neural network based
sensor fusion. Int J Mach tools Manuf 37(9):12011217
6. Bajic D, Lela B, Zivkovic D (2008) Modeling of machined
surface roughness and optimization of cutting parameters in face
milling. Metalurgija 47(45):331334
7. Villaseor D, Morales R, Menndez C, Rodrguez J, Alique R (2006)
Neural networks and statistical based models for surface roughness
prediction. International Association of Science and Technology for
Development. Proceedings of the 25th IASTED International
Conference on Modeling, Identification and Control, pp. 326331
8. Cus F, Balic J (2003) Optimization of cutting process by GA
approach. Robot Comput-Integr Manuf 19:113121
9. Groover M (1996) Fundamentals of modern manufacturing. Prentice
Hall, Upper Saddle River (now published by Wiley, New York)
10. Hossain MI, Amin AKM, Patwari AU (2008) Development of an
artificial neural network algorithm for predicting the surface
roughness in end milling of Inconel 718 alloy. Int Conf Comput
Commun Eng ICCCE 13(15):13211324
11. Kadirgama K, Noor MM (2008) Optimization of surface roughness in end milling on mould aluminum alloys using response
surface method and radian basis function network. Jordan Journal
of Mechanical and Industrial Engineering 2(4):209214
12. Huang L, Chen JC (2001) A multiple regression model to predict
in-process surface roughness in turning operation via accelerometer. J Ind Technol 17(2):18
13. Mahdavinejad RA, Sharifi Bidgoli H (2009) Optimization of
surface roughness parameters in dry turning. J Achiev Mater
Manuf Eng 37(2):571577
14. Nalbant M, Gokkaya H, Toktas I (2007) Comparison of regression
and artificial neural network models for surface roughness prediction
with the cutting parameters in CNC turning. Modelling and
Simulation in Engineering. Hindawi Publishing Corp, New York
15. Oktem H, Erzurumlu T, Kurtaran H (2005) Application of
response surface methodology in the optimization of cutting
conditions for surface roughness. J Mater Process Technol
170:1116
16. Davima JP, Gaitondeb VN, Karnik SR (2008) Investigations into the
effect of cutting conditions on surface roughness in turning of free
machining steel by ANN models. J Mater Process Technol 205:1623
17. Parappagoudar MB, Pratihar DK, Datta GL (2008) Forward and
reverse mappings in green sand mould system using neural
networks. Appl Soft Comput 8:239260
18. Saad Kareem Shather,Abbas Fadhel Ibrheem (2007) Prediction of
surface roughness in end milling with multiple regression model.
Engineering and Technology 26(3)
19. Sakir Tasdemir, Sleyman Neseli, Ismail Saritas, Sleyman Yaldiz
(2008) Prediction of surface roughness using artificial neural
network in lathe. In: Proceedings on CompSysTech '08 Proceedings of the 9th International Conference on Computer Systems
and Technologies and Workshop for PhD Students in Computing
20. Srikanth T, Kamala V (2008) A real coded genetic algorithm for
optimization of cutting parameters in turning. Int J Comput Sci
Net Secur 8(6):189193
21. Suresh PVS, Venkateswara RP, Deshmukh SG (2002) A genetic
algorithmic approach for optimization of surface roughness
prediction model. Int J Mach Tools Manuf 42:675680
22. Uros Z, Franci C (2003) Optimization of cutting conditions during
cutting by using neural networks. Robot Comput-Integr Manuf
19:189199
23. Yang WH, Tarng YS (1998) Design optimization of cutting
parameters for turning operations based on the Taguchi method. J
Mater Process Technol 84:122129