AUTOMATIC FAULT DETECTION BY USING WAVELET METHOD

1
AUTOMATIC FAULT DETECTION BY USING WAVELET METHOD

• Soundararajan Ezekiel, Gary Greenwood, David
  Pazzaglia
• Computer Science Department
• Indiana University of Pennsylvania
• Indiana, PA, USA

2
ABSTRACT

• we propose a diagnostic model to automatically
  detect and identify faults in manufacturing
  processes by using a wavelet-based method.
• The idea behind our method is to use an image
  processing system that performs the following
  phases
• image capturing,
• image preprocessing,
• determination of region of interest,
• object segmentation,
• computations of object features and
• decision-making.
• For the above phases, we use a bank of filters,
  statistical, morphological, and wavelet
  operations.
• Developed in this paper is a method that
  automatically detects and isolates faults in
  manufacturing products by dividing our system
  into three sub modules.
• These sub modules are the sensor, computer, and
  logistical interface modules that are
  straightforward to analyze.

3
Continue

• We have focused only on the design and object
  features.
• We demonstrate our method for various product
  images and extract
• characters,
• numbers,
• and object features such as area, major/minor
  axis length, orientation, diameter, convex area,
• Euler number and centroid.
• The availability of this system may
  significantly impact the quality control process
  of the manufacturing sector.
• The underlying algorithms and system
  architecture are described, as well as the
  hardware and software aspect of the
  implementation.

4
Introduction

• Over the last two decades, the natural quality
  control process of manufacturing has undergone
  many technological advances.
• The nature of diagnosis, in general, has been
  done by visual inspection.
• Due to the complexity of the problem the demand
  on the workforce has increased tremendously.
• At the same time the product quality assurance
  has been reduced, while the cost of goods sold
  have risen.
• As a direct result of this, supply and demand
  are not in equilibrium.
• Image processing can provide tools to solve this
  problem.

5
Introduction -- Continue

• A well-designed image processing system will
  increase the product quality assurance and lower
  production costs.
• In this paper, we propose a reliable robust
  system for automatic fault detection and
  identification.
• The system is further divided into sub modules
  depending on their task.
• The required software and the sub modules are
  integrated reliably.
• The scope of our system is not only automatic
  fault detection and isolation, but it also
  encompasses data storage for further research and
  development analysis.

6
Introduction --- continue

• The data stored is composed of a number of
  elements.
• These elements include the following the image
  itself, the physical characteristics, image
  faults, and the image analysis results.
• Our results are based on thresholding functions.
  We use a threshold value predefined by the
  manufacturer of the product.
• All of these parameters can be easily modified
  by the graphical user interface (GUI).
• Modern processing plants are very complex and
  consist of a large number of parameters.
• These can be implemented in our GUI, which is
  portable and adapts easily.

7
Introduction Continue

• In this paper, we use wavelet,
• statistical,
• and morphological methods for automatic detection
  and isolations
• it is simple, effective, and it can be
  implemented in embedded systems.
• This method seems to be well suited for a wide
  variety of products.

8
Wavelet

• A wavelet is a waveform of effectively limited
  duration that has an average value of zero.
• So, wavelet analysis is done by breaking up a
  signal into shifted and scaled versions of the
  original (mother) wavelet.
• From this observation, we can define a continuous
  wavelet transform as the sum over all time of the
  signal multiplied by a scaled and shifted version
  of the wavelet function i.e.
• where scaling means stretching (or compressing)
  and position means shifting the wavelet.

9
Basics
Thresholding is the transformation of an input
image I to an output binary image BI as follows
where T is the threshold.
Morphological Operations   Morphological
operations can be used to construct spatial
filters in image enhancement . The basic
operators such as dilation, erosion, opening and
closing are defined, but many others exist. Let
and be input image and
structured element, respectively .
10
(No Transcript)
11
SYSTEM DESIGN

• A variety of methods are widely used for
  automatic detection.
• Most methods of fault detection rely on a single
  statistical parameter thresholding .
• Thresholding represents the difference between
  the calculated value and the expected value.
• For quality purpose we would like to see the
  threshold value is zero, but in practical this is
  not the case.
• Typically one parameter is measured and quality
  control is based on this parameter that may lead
  to undetected defects in other parameters. To
  avoid such problems, it is necessary to check all
  possible parameters.
• Since we are using the wavelet, morphological,
  and statistical methods, the system is able to
  provide in-depth analysis.
• Based upon this analysis, faults can be
  effectively detected.

12
Image Processing System
13
Continue

• The system is divided into three sub modules
• sensors,
• computers, and
• logistical interfaces.
• Sensor Module
•  
• This model consists of the charge coupled device
  (CCD) image sensors, lenses, driver control
  circuits or high quality cameras and illumination
  setups.

14
Continue

• Computer Module
• The computer module determines if the picture
  sent to it is an analog or digital picture. If
  the picture is analog, the frame grabber will
  convert it to a digital image. If the picture is
  digital, it will bypass the frame grabber and the
  analysis process will begin. The analysis
  process determines if the image of the product
  matches the predefined criteria.
• Logical Interface Module
• The logistic interface receives a message from
  the computer containing a number or parameters.
  These parameters include when the product will
  reach the control arm, whether the product
  matches the criteria, and what to do with the
  product. Based on the parameters the control arm
  will take action and accept or deny the product
  when the time is right

15
RESULTS
Original and enhanced image
Edges and gear segmentation
16
Various segmentations
Original image and Object perimeters
17
Original and enhanced image
Object perimeters
18
Original and enhanced washer
Object perimeters
19
barcodes with noise added
Extracted and matching characters
20
Conclusion

• Using the system described above, we have been
  able to automatically detect and identify faults
  in manufacturing processes by using wavelet,
  morphological, and thresholding operations.
• Our experimental results have demonstrated that
  our algorithm is effective for image capturing,
  image preprocessing, determination of region of
  interest, object segmentation, computations of
  object features and decision-making.
• Although the system has not been fully
  implemented, a foundation for an automatic image
  processing system for fault detection and
  isolation has been set forth. Our system can be
  applied to a variety of manufacturing processes.
• However, further experimental analysis needs to
  be carried out for different manufacturing
  processes in order to adapt to the vast range of
  manufactured products.
• More information check out website
• http//www.cosc.iup.edu/sezekiel
• Contact sezekiel_at_iup.edu