International Journal of Advance Research and Innovation


Neural Network in Image Processing and Compression



Yüklə 0,71 Mb.
Pdf görüntüsü
səhifə9/13
tarix27.01.2023
ölçüsü0,71 Mb.
#81145
1   ...   5   6   7   8   9   10   11   12   13
IJARI-ME-14-09-106 (1)

Neural Network in Image Processing and Compression: 
Windows based neural network image compression and 
restoration, Real time data compression, Image compression 
using direct solution method (DSM) method based neural 
network, Application of NN to wavelet filter selection in 
multispectral image compression, Human face detection in 
visual scenes, Rotation invariant neural network-based face 
detection 
Neural Network in Control: The neural network can be act 
as NN controller and can be used in the feedback path as 
NN estimator for parameter estimation. The applications 
included are: Neural network based controller for induction 
motor. In this application the neural networks are used as 
controller to control the speed of the induction motor. 
Neural network control of the neutralization system. In this 
application the pH value is used to control the neutralization 
of the system. The pH value is controlled by controlling the 
flow rate of the base. Neural network based controller for 
broom stick balancer. In this application the position and 
angle of broom stick is used to control the balance of the 
broom stick. 
Neural Network in Pattern Recognition: Pattern 
recognition is the association of an observation to past 
experience or knowledge. It involves many stages such as 
making the measurements, pre-processing and segmenting, 
finding a suitable numerical representation for the objects 
we are interested in and finally classifying them on these 
representations. 
3. Literature Review
 
Diwakar Reddy V. et al, [8] have carried out 
machining process on Mild steel material in dry cutting 
condition in a lathe machine and surface roughness was 
measured using Surface Roughness Tester. To predict the 
surface roughness, an artificial neural network (ANN) 
model was designed through back propagation network for 
the data obtained. Comparison of the experimental data and 
ANN results showed that there is no significant difference 
and ANN was used confidently. Three cutting parameters 
speed, feed, depth of cut have been considered. The 
machining tests have been carried out by straight turning of 
medium carbon steel (mild steel) on a lathe by a standard 
HSS uncoated and carbide insert with ISO designation-
SNMG 120408 at different speed-feed and depth 
combinations. 
By 
using 
the MATLAB 
command 
„postmnmx‟ the network values have been predicted, 
regression analysis was adopted to find the coefficient of 
determination value (R
2
) for both training and testing phases 
to judge performance of each network. The multilayer feed 
forward network consisting of three inputs, 25 hidden 
neurons (tangent sigmoid neurons) and one output (network 
architecture represented as 3-25-1) was found to be the 
optimum network for the model developed in their study. 
They concluded from their results obtained that ANN is 
reliable and accurate for solving the cutting parameter 
optimization. A. V. N. L. Sharma et al, [1] studied 
prediction of surface roughness, using an Artificial Neural 
Network (ANN) model which was designed through back 
propagation network using MATLAB 7.1 software for the 
data obtained. Experimental details and specifications used 
in their study were, Machine tool : Engine Lathe, Work 
material: Mild steel, Cutting tool : High speed steel, 
Cemented carbide tipped tool. Cutting conditions: Dry 
environment Surface roughness measuring instrument 
Mitutoyo SJ-201P Traverse Speed: 1mm/sec, Measurement: 
Metric/Inch. Three cutting parameters speed, feed, depth of 
cut have been considered .They have concluded in their 
studies that the approach is suitable for fast determination of 
optimum cutting parameters during machining, where there 
is not enough time for deep analysis. U. Natarajan et al, 
[22] analyzed the feasibility of fully automated futuristic 
actories. Their research studies deals with the use of 
machine vision techniques to inspect surface roughness of a 
work piece under a variation of turning operations. The 
surface image of the work piece is first acquired using a 
digital camera and then the feature of the surface image has 
been extracted. A neural fuzzy network using a self-
organizing adaptive modeling method have been applied to 


Volume 2, Issue 3 (2014) 676-683 
ISSN 2347 - 3258
International Journal of Advance Research and Innovation 
680 
IJARI 
constructing the relationships between the feature of the 
surface mage and actual surface roughness under various 
parameters of turning operations. They have, concluded 
from their results obtained that ANN is reliable and accurate 
for solving the cutting parameter optimization. 
Sita Rama Raju K et al, [20] studied three soft 
computing techniques namely Adaptive Neuro Fuzzy 
Inference System ANFIS, Neural Networks NN and 
regression in predicting the surface roughness in turning 
process. The work piece material used was AA 6063 
aluminum alloy. Here 27 data sets were considered for 
training and 9 data sets were considered for testing .The 
predicted surface roughness values computed from ANFIS, 
NN and regression are compared with experimental data. 
Based on the their experimental results they observed that, 
surface roughness value increases as the feed and depth of 
cut increases and as the spindle speed increases the surface 
roughness value decreases. The minimum surface roughness 
value is observed at spindle speed of 150 rpm, feed of 0.05 
mm/rev and a depth of cut of 0.2 mm respectively. Zsolt 
Janos and Viharos, [25] applied ANN models to estimate 
the roughness of a given finishing operation. They have 
used acoustic emission sensor as an information source to 
improve the estimation capability of the ANN model. To 
avoid the problem of overlapping and non-invertable 
dependencies they have used a new approach for building 
the ANN model. To estimate the surface roughness 
parameters describing the energy content of the Acoustic 
Emission signals sensor were used beside the three 
machining parameters depth of cut (a), feed per revolution 
(f) and cutting speed (v).Four parameters related to different 
frequency range were used to describe the energy content. 
Tugrul Ozel et al, [21] studied the effects of tool corner 
design on the surface finish and productivity in turning of 
steel parts. Surface finishing has been investigated in finish 
turning of AISI 1045 steel using conventional and wiper 
(multi-radii) design inserts. Multiple linear regression 
models and neural network models have been developed for 
predicting surface roughness, mean force and cutting Power. 
The Levenberg-Marquardt method was used together with 
Bayesian regularization in training neural networks in order 
to obtain neural networks with good generalization 
capability. 
Fig: 3. Multilayer Feed-forward Neural Network 
Neural network based predictions of surface roughness 
were carried out and compared with a non-training 
experimental data. These results showed that neural network 
models are suitable to predict surface roughness patterns for 
a range of cutting conditions in turning with conventional 
and wiper inserts. 
Fig: 4. Architecture of Multilayer Feed-Forward Neural 
Network used for Predictions 
Yue Jiao et al, [24] used combined neural -fuzzy 
approach (fuzzy adaptive network, FAN), to model surface 
roughness in turning operations. The FAN network has both 
the learning ability of neural network and linguistic 
representation of complex, not well-understood, vague 
phenomenon. A model representing the influences of 
machining parameters on surface roughness have been 
established and verified by the use of the results of pilot 
experiments. Ilhan Asiltürk and Mehmet Çunkas [10] used 
artificial neural networks (ANN) and multiple regression 
approaches to model the surface roughness of AISI 1040 
steel. Full factorial experimental design is implemented to 
investigate the effect of the cutting parameters (i.e. cutting 
speed, feed rate, and depth of cut) on the surface roughness. 
In order to predict the surface roughness, the second-order 
regression equation can be expressed as: R

Yüklə 0,71 Mb.

Dostları ilə paylaş:
1   ...   5   6   7   8   9   10   11   12   13




Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©azkurs.org 2024
rəhbərliyinə müraciət

gir | qeydiyyatdan keç
    Ana səhifə


yükləyin