International Journal of Advance Research and Innovation

Table: 1. Neural Networks Multilayer Perceptron [16] 
Sr. No 
s 
f 
d 
Ra 
Pred. Ra 
1 
1700 
0.1 
0.2 
0.82 
0.83 
2 
1700 
0.1 
0.3 
0.94 
0.86 
3 
1700 
0.1 
0.4 
0.96 
0.87 
4 
1700 
0.13 
0.2 
1.12 
0.99 
5 
1700 
0.13 
0.3 
1.06 
1.09 
6 
1700 
0.13 
0.4 
1.1 
1.1 
7 
1700 
0.15 
0.2 
1.44 
1.38 
8 
1700 
0.15 
0.3 
1.54 
1.44 
9 
1700 
0.15 
0.4 
1.5 
1.45 
10 
1900 
0.1 
0.2 
0.86 
0.84 
11 
1900 
0.1 
0.3 
0.92 
0.88 
12 
1900 
0.1 
0.4 
0.76 
0.88 
13 
1900 
0.13 
0.2 
1.04 
1.01 
14 
1900 
0.13 
0.3 
1.2 
1.12 
15 
1900 
0.13 
0.4 
1.1 
1.13 
16 
1900 
0.15 
0.2 
1.44 
1.4 
17 
1900 
0.15 
0.3 
1.6 
1.46 
18 
1900 
0.15 
0.4 
1.5 
1.46 
19 
2100 
0.1 
0.2 
0.88 
0.84 
20 
2100 
0.1 
0.3 
0.78 
0.88 
21 
2100 
0.1 
0.4 
1.16 
0.89 
22 
2100 
0.13 
0.2 
1.08 
1.03 
23 
2100 
0.13 
0.3 
1.14 
1.14 
24 
2100 
0.13 
0.4 
1.26 
1.15 
25 
2100 
0.15 
0.2 
0.58 
1.41 
26 
2100 
0.15 
0.3 
1.42 
1.46 
27 
2100 
0.15 
0.4 
1.86 
1.47 
Ten independent runs having different initial random 
weights were performed to achieve a good solution. Data 
sets were from experiments conducted on a CNC turning 
machine. After each turning operation, the surface 
roughness (Ra) was measured with Surface Roughness 
Tester Mitotoyo (SJ-301). The measurements were taken 
three times for each work piece. A National Instruments 
portable E Series NI DAQCard-6036E with maximum 
acquisition rate of 200,000 samples per second and 16 
channels, data acquisition card was used to transmit the data 
to PC. A software called as ilhan_daq_v01 was developed 
using Matlab 6.5 program. The constants and cutting 
parameters were entered to the interface. The outputs were 
measured as 80 samples/sec, and their average values were 
recorded as one datum. Consequently, tests were performed 
with 27 experimental runs. They concluded that ANN is a 
powerful tool in predicting surface roughness.
Ruturaj Kulkarni et al, [18], carried out tests on AISI 
4140 steel. 12 speed Jones and Lamson Lathe model was 
used for turning operation. The specimen with a diameter of 
60mm, 500mm length and hardened 35 HRC is used. The 
operating parameters that contribute to turning process are 
cutting speed, Feed rate, Depth of cut, Vibrations, tool wear, 
tool life, surface finish and cutting forces. These readings 
were used to train and validate the Neural Network. They 
found ANN to be very useful with simulations tasks which 
have complex and explicit relation between control factors 
and result of process. They created Neural Network using 
feed forward back propagation technique for simulation of 
the process using the Matlab Neural network toolbox. With 
assurance of accuracy of the predictive capabilities of the 
neural network, it was then used for optimization. Particle 
Swarm 
Optimization 
Algorithm, 
an 
evolutionary 
computation technique is used to find out the optimum 
values of the input parameters to achieve the minimum 
surface roughness. The objective function used here is to 
minimize the surface roughness. Limits of the operational 
variables are used as constraints for developing the code for 
optimization algorithm. Ranganath M S et al, [16] analyzed 
surface roughness of Aluminium (6061) through neural 
network model. To predict the surface roughness, neural 
network model was designed through Multilayer Perceptron 
network for the data obtained. The predicted surface 
roughness values computed from ANN, were compared 
with experimental data and the results obtained showed that 
neural network model is reliable and accurate for solving 
the cutting parameter optimization. 
Fig: 5. Network Layers [16] 
They concluded that the appropriate cutting parameters 
can be determined for a desired value of surface roughness. 
Durmus Karayel [9],
has used neural network approach for 
the prediction and control of surface roughness in a 
computer numerically controlled (CNC) lathe. A feed 
forward multi-layered neural network was developed and 

---

Volume 2, Issue 3 (2014) 676-683 
ISSN 2347 - 3258
International Journal of Advance Research and Innovation 
682 
IJARI 
the network model was trained using the scaled conjugate 
gradient algorithm (SCGA), which is a type of back-
propagation. The adaptive learning rate was used. He 
concluded that the appropriate cutting parameters can be 
determined for a desired value of surface roughness. Anna 
Zawada and Tomkiewicz [3], have estimated the surface 
roughness parameter with use of a neural network (NN). 
The optical method suggested in this paper is based on the 
vision system created to acquire an image of the machined 
surface during the cutting process. The acquired image is 
analyzed to correlate its parameters with surface parameters. 
In the application of machined surface image analysis, the 
wavelet methods were introduced. A digital image of a 
machined surface was described using the one-dimensional 
Digital Wavelet Transform with the basic wavelet as 
Coiflet. The increment of machined surface image 
parameters was applied as input for the neural network 
estimator. Five cross-sections of the image were loaded, 
from which six statistical parameters of the six levels of 
wavelet decomposition were computed. These six 
parameters were chosen via the Optimal Brain Surgeon 
Method. They found that by applying the increments of 
these parameters and of the estimated value in a given time, 
it made possible to establish the Ra estimator for the points 
in time when the surface roughness parameters were 
unknown. Ranganath M S et al, [15], have reviewed the 
works related to Artificial Neural Networks ANN, in 
predicting the surface roughness in turning process. They 
studied in papers that some of the machining variables that 
have a major impact on the surface roughness in turning 
process, such as spindle speed, feed rate and depth of cut 
were considered as inputs and surface roughness as output 
for a neural network model. They found that the predicted 
surface roughness values computed from ANN, were 
compared with experimental data and the results obtained. 
These results showed that the neural network model is 
reliable and accurate for solving the cutting parameter 
optimization.