Yarn Defect Detection Using Fuzzy Logic and BPN

					      Yarn Defect Detection Using Fuzzy Logic and BPN
                                     Soundarajan
                                  Assistant Professor
                             Annai College of arts & Science

Abstract:                                      industry are working toward just-in-time
                                               delivery and a poor quality production
Textile industry is the ever green             run can be disastrous. Presently, the
business in trend. Many industries are         inspection is done manually after a
using automated defect detection system        significant amount of fabric is produced,
to identifying error rates after production    removed from the weaving machine,
over. This research implements a textile       batched into large rolls and then sent to
defect detector which uses computer            an inspection frame.
vision      methodology        with     the
combination of multi-layer neural              An would be to automatically inspect
networks to identify the classification of     fabric [1] [2][3] as it is being produced
textile defects and detect it. In This         and to alert maintenance personnel when
Paper we provide an effective method to        the machine needs attention to prevent
detect the YARN (thread) defect using          production of defects or to change
Fuzzy Logic and Neural Networks.               process parameters automatically to
Fuzzy logic used to detect the pattern         improve product quality. Reducing the
identification. After identifying the          number of defects produced by timely
pattern error rate will calculated using       maintenance or control would result in
Neural Network methods.                        obvious savings. Also if inspection is
                                               accomplished on the machine, the need
Keywords:                                      for 100% manual inspection is
    Yarn,  Fuzzy Logic, Neural                 eliminated.
Networks, BPN (Back Propagation
Network).                                      Costs to inspect fabric manually range
                                               from 1 .O to 1.5 cents/yard. The cost to
1. INTRODUCTION:                               inspect the annual production of a
                                               machine would be $1250 to $1900.
    The textile industry is continually
                                               Other tangible and intangible benefits
pressing for higher product quality and
                                               could be factored into the savings
improved productivity to meet both
                                               equation. Computer vision systems do
customer demands and to reduce the
                                               not suffer from some of the limitations
costs associated with off-quality. Higher
                                               of humans (such as exhaustion) while
production speeds make the timely
                                               offering the potential for robust defect
detection of fabric defects [6] [7] more
                                               detection with few false alarms.
important than ever. Newer weaving
technologies also tend to include larger       1.1 Neural Networks:
roll sizes and this translates into greater
                                               ANNs apply the principle of function
potential for off-quality production           approximation by example, meaning that
before inspection. Many segments of the        they learn a function by looking at
examples of this function. One of the        network. Neurons send impulses to each
simplest examples is an ANN learning         other through the connections and these
the XOR function, but it could just as       impulses make the brain work. The
easily be learning to determine the          neural network also receives impulses
language of a text, or whether there is a    from the five senses and sends out
tumor visible in an X-ray image. If an       impulses to muscles to achieve motion
ANN is to be able to learn a problem, it     or speech. The individual neuron can be
must be defined as a function with a set     seen as an input output machine which
of input and output variables supported      waits for impulses from the surrounding
by examples of how this function should      neurons and, when it has received
work. A problem like the XOR function        enough impulses, it sends out an impulse
is already defined as a function with two    to other neurons.
binary input variables and a binary
output variable, and with the examples       2. PROPOSED METHOD:
which are defined by the results of four              In this paper we are providing an
different input patterns.                    effective computer vision automated
                                             system for detecting the fabric faults.
However, there are more complicated          This method likes a proactive technique
problems which can be more difficult to      because we find error detection in yarn
define as functions. The input variables     itself. To enhance the quality of the
to the problem of finding a tumor in an      detection we add the Fuzzy Logic and
X-ray image could be the pixel values of     Neural networks. Fuzzy Logic is the
the image, but they could also be some       estimation technique which provides the
values extracted from the image. The         effective window size (frame size) in
output could then either be a binary         this paper. A neural network is iteration
value or a floating point value              process used to classify the patterns as
representing the probability of a tumor in   well as detecting the errors. In this paper
the image. In Ann’s this floating-point      we are using Back propagation network
value would normally be between 0 and        (BPN) for detecting the errors.
1, inclusive.
                                             2.1 Flow Diagram:
A function approximator like an ANN
[4] can be viewed as a black box and               Digital Image
when it comes to FANN, this is more or
less all you will need to know. However,
basic knowledge of how the human brain        Determine Window size
operates is needed to understand how            using Fuzzy Logic
ANN’s work. The human brain is a
highly complicated system which is
capable to solve very complex problems.               Resize
The brain consists of many different
elements, but one of its most important
building blocks is the neuron, of which it
                                               Pattern Classification
contains approximately 1011. These
neurons are connected by around 1015
connections, creating a huge neural
                                               Error Detection Using
                                                       BPN


                                                      Results
2.2 Algorithm:                               FL incorporates a simple, rule-based IF
                                             X AND Y THEN Z approach to a
Step1: Source Image fetched from             solving control problem rather than
camera.                                      attempting to model a system
Step2: Window size optimization using        mathematically. The FL model is
fuzzy Logic                                  empirically-based, relying on an
Step3: Resize or Zooming the image.          operator's experience rather than their
Step4: Pattern classification using          technical understanding of the system.
Perceptron1 (linear separable).              For example, rather than dealing with
Step5: Error detection and optimization      temperature control in terms such as "SP
using Back propagation neural network.       =500F", "T <1000F", or "210C <TEMP
Step6: Identified Output Result              <220C", terms like "IF (process is too
                                             cool) AND (process is getting colder)
                                             THEN (add heat to the process)" or "IF
3. DIGITAL IMAGE:                            (process is too hot) AND (process is
                                             heating rapidly) THEN (cool the process
 At the first step we have to catch the      quickly)" are used. These terms are
sample image of yarns what we are            imprecise and yet very descriptive of
going to use for production. Using           what must actually happen. Consider
digital camera we can fetch the digital      what you do in the shower if the
image of yarns. Ever material having         temperature is too cold: you will make
different type of yarns and single           the water comfortable very quickly with
material may contain the multiple type       little trouble. FL is capable of mimicking
of yarns. Every yarn having own color,       this type of behavior but at very high
contrast, thin, style, twist, pattern etc.   rate.
after fetch the digital image we have to
determine the window size of sample          In this paper FL used to determine the
image to match the existing samples.         frame size. Frame size is the unit of
                                             picture which is repeatedly present in the
4. FUZZY LOGIC:                              yarn pattern. Cosine transformation used
                                             to calculate the distance between the two
In this context, FL is a problem-solving     directions of point.
control system methodology that lends
itself to implementation in systems          While defining the frame size we have to
ranging from simple, small, embedded         consider     about       the     intensity
micro-controllers to large, networked,       transformation of repeated unit. For the
multi-channel PC or workstation-based        processing capability the original sample
data acquisition and control systems. It     will divided into many sub images,
can be implemented in hardware,              among the sub images which pattern
software, or a combination of both. FL       having more probability that will take as
provides a simple way to arrive at a         frame size. Frame size will chose based
definite conclusion based upon vague,        on the following algorithm.
ambiguous, imprecise, noisy, or missing
input information. FL's approach to                Input sample (sub images)
control problems mimics how a person               Scaling factors
would make decisions, only much faster.
                                                   Fuzzification
      Defuzzication                          technique is fast convergence. Seven
      Plant                                  cotton fiber properties measured by HVI
      Frame size                             and yarn count were taken as input
                                              parameters and the output was breaking
Sub images taken as inputs. In that we        elongation.
apply the scaling factors to determine the
frame size. Actual processing happens in            The increase in MSE values in
fuzzification phase. The output of            comparison with the optimized ANN
fuzzification always in membership form       model values was considered as the
from that we should predict the               indicator of importance of eliminated
optimized image in plant phase. Finally       output. Cotton yarn hairiness from the
we will get the optimized frame size.         fiber characteristics and process
                                              parameters are using a feed forward back
5.  BACK               PROPAGATION            propagation       network.       Reverse
NETWORK:                                      engineering approach using ANN for the
                                              prediction of fiber properties form the
       ANN and the multiple linear            yarn .Yarn tenacity, breaking elongation,
regression models for predicting the          unevenness and hairiness index were the
tensile properties of cotton-covered          inputs and with the use of feed forward
nylon core yarns. The input parameters        back propagation network, cotton SCI
were the count of core part, count of         and micronaire were predicted the bulk
sheath part, twist factor of core-spun        and yarn instability values well
yarn and presentation of core part.           compared to yarn tenacity.
Breaking      strength     and    breaking
elongation of yarn were the outputs. As
the input parameters are four, the four
cross     validation     technique     was
employed. Over fitting or memorization
was prevented by using the weight decay
technique. The models were assessed by
verifying mean square error and
correlation coefficient. The result
showed that ANN models gives better
prediction than multiple linear regression
models.
      Number of fiber parameters, yarn              Fig 1: BPN Network Model
count, twist, and processing parameters
were taken as input and number of fibers      BPN having three layers are input layer,
in yarn cross section, yarn unevenness,       hidden layer, output layer. According to
hairiness, tenacity, and ends down were       our research we took four parameters of
considered as spinning performance            yarn like 1) count of core part, 2) count
outputs. Perceptron was built and trained     of sheath part, 3) twist factor of core-
to classify the patterns based on the fiber   spun yarn and 4) presentation of core
of yarns and type. The convergence of         part. Hidden layer may vary from 15 to
this method of training is back               20 according to the pattern. The output
propagation algorithm. The use of such        layer produce the two outputs half of the
input layer. The out put vary from 00 to
11. We classified into 00, 01, 10, and 00
like no error, minor error, small error
and major error.
Original values compared with BPN
values using Standardized Euclidean
distance. The error rate up to .1 called no
error, up to .4 called minor errors, up to
.7 called small errors, beyond those
major errors.

6. RESULT ANALYSIS:
                                                        Fig: 3 Minor errors
    Matlab 2008 image processing
toolbox and built in functionalities assist
to implementation of this research. With
that many of Fuzzy logic tools and ANN
tools assist to accomplish this task. The
following images describe the classified
errors by BPN.




                                                        Fig: 4 Small errors




             Fig: 2 No errors




                                                        Fig: 5 Major errors

                                              7. CONCLUSION:

                                                    So far the major research
                                              happening in finished product only. In
                                              this paper we provided an effective
method to avoid the yarn errors in terms     Automated Fabric Defect Detection
of computer automation using fuzzy           System” 2007.
logic and neural networks. Fuzzy logic       [7] Chrpova E., “Surface Quality
gives an affective optimization for          Control Based on Image Processing
determine the frame size. Determine the      Methods.” Fibres and Textiles Vol. 10
frame size is the tedious process but        (2), Bratislava, Slovakia, 2003.
fuzzy logic is the one of the best method    [8] Sadrei A. H., “Fault Detection on
to determine. BPN is one of the best         Simple Circular Knitted Fabrics Using
neural models to calculate the error         Wavelet”, M. Sc. Project, Amirkabir
rates, which predict the accurate error      University Technology, 2001.
deviation compare with the original
samples. Its our belief this research will
help to textile industry much and more
to produce the good materials.

8. REFERENCES:

[1] M. Bennamoun, A. Bodnarvova,
Automatic visual inspection and flaw
detection in textile materials: past,
present and future, Proceedings of the
IEEE Conference SMC, 1998, pp. 4340–
4343.
[2] A. Kumar, G.K.H. Pang, Defect
detection in textured materials using
Gabor filters, IEEE Trans. Ind. Appl. 38
(2) (2002) 425–440.
[3] A.L. Amet, A. Ertuzun, A. Ercil,
Texture defect detection using subband
domain co-occurrence matrices, IEEE
Southwest Symposium on Image
Analysis and Interpretation, April 1998,
pp. 205–210.
[4] A. Hyvarinen, E. Oja, “Independent
component analysis: algorithms and
applications”, Neural Networks, 2000,
pp. 411-430.
[5] O. G. Sezer, A. Ertuzun and A. Ercil
"Independent Component Analysis for
Texture Defect Detection", Pattern
Recognition and Image Analysis, vol. 14
no. 2, 2004, pp. 303-307.
[6] Ahmed Ridwanul Islam, Farjana
Zebin Eishita, Jesmine Ara Bubly,
“Implementation of a Real Time

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:112
posted:5/9/2011
language:English
pages:6