CAT400 ProjectProposal Azeem Edited3

CAT400 Undergraduate Major Project Proposal

CAT 400 Project Proposal:

Automatic Laser Welding Defect Detection And

Classification Using Image Segmentation Algorithm

[Intelligent System/SC151616]

[Muhammad Azeem Bin Abdul Halim], [Ahmad Sufril Azlan Mohamed, Dr.]

[[email protected]], [[email protected]]

School of Computer Sciences, Universiti Sains Malaysia

11800 USM, Penang, Malaysia

Abstract

This paper describees a detection of commons defects in laser welding of structural

aluminium alloy. To overcome these problems, a technique has been introduced to

detect defects automatically and effectively using the image segmentation technique.

Although, this technique has been well developed, it does suffer from several

disadvantages of radiographic images taken are of poor quality, as well as the

microscopic size of the defects together with poor orientation relatively to the size and

thickness of the evaluated parts. Using image segmentation algorithm, allows the

defects to be automatically inspected and measured within the welded surface such as

cracks, porosity and foreign inclusions, which may be weakening the welded parts.

The motivation of this project is to build a system that compliments the process of

identifying the faults from the welding process by using the existing image

segmentation algorithm such as otsu’s method, k-mean clustering, watershed

segmentation and texture filters. These algorithms will then be tested and optimized to

automatically identify and classify the types of faults. The output of the developed

system will produce a measured analysis which can then be used to describe

accordingly to the conventional testing such as tensile and force test towards the

welded part of the aluminium alloy. The benefits of this project will be describe the

faults and defects of the welded aluminium for construction and manufacturing.

Keywords: Image Segmentation, weld defect detection, laser welding, aluminium

alloy

PROPOSAL-1.0 1


1. Project Background

Welding is the most important process in manufacturing industry particularly using

the steel alloy. The use of Laser welding for aluminium is gaining its popularity due

to its cheaper construction material, and with the use of laser benefits in terms of

higher temperature of heat applied and capable of welding at greater precision.

Despite that, due to the high temperature and various welding situations, the process

using the aluminium allow is naturally complicated and nonlinear, making it prone to

a number of defects, such as porosity, undercuts, surface holes, and solidification

cracking which are often found in laser welds [1]. In order to solve this problem,

several methods will be used to detect defects based on image segmentation.

Image processing is a method to process an image into digital form. So, in

order to get some useful information from it, some operations must be performed on

it. Because of fast growing technologies nowadays, this application is very useful in

various aspects of a business and manufacturing industry. There are several purpose

of image processing such as image sharpening and restoration that will create a better

image and visualization to observe the objects that are not visible, image recognition

that will distinguish the objects in an image, and so on. Meanwhile, the purpose of

image segmentation is to divide the image into regions with different characteristics

and extract its features into the related goals [12]. There are many different ways to

perform image segmentation such as thresholding methods, color-based segmentation,

transform methods, and texture methods.

The objectives of this project are laid out as below:

To identify the appropriate existing algorithms of image segmentation

using the microscopic images of welded area for fault detection.

To compare several existing algorithms of image segmentation applied in

welding fault detection and select the one with optimum result.

To identify faults and defects within welding process of aluminium alloy

and classify it.

PROPOSAL-1.0 2


To optimize the chosen algorithm and to build a complete image

segmentation system to measure and classify the fault found in the welding

process.

To provide visual analysis of the measured faults and defects to

compliment other testing operations such as the hardness test, tensile test

and elevated temperature test.

2. Problem Statement

Welding the structure of aluminium alloy may contain more defects compare to other

types of steel alloy. This is because aluminium alloy composes of different material

composition compared to other steel alloys, and usually purposely hardened to match

the strength of steel alloy, which can lead to its natural weakening states at the welded

areas. Furthermore, cracks and porosities are some of the problems in laser welding

usually found in most welded materials [11].

Cracks happen when applying stress and tension to the welded area while

porosities happen when there are gasses trapped within the solidify welded area thus

weakening the strength of the welded structure. The process of identifying these faults

require many steps and train eyes which is time consuming and very hard to be done.

However, current technology enables images to be taken at microscopic level with

high resolution, and with the use of image processing technique, image segmentation

can be used to automatically extract and classify these faults and measuring them

[1,3]

Figure 2.1(a) shows an example of welding defect using the Scanning Electron

Microscope (SEM) taken at the The School of Mechanical Engineering, USM. The

welded interface has porosity and slag usually occur when the gas used is not enough

or the surface contained contaminants, and thus weakening the strength of the welded

area (see figure 2.1b).

PROPOSAL-1.0 3


Figure 2.1: (a) Shows the examples of welding defect at

microscopic level using the Scanning Electron Microscope (SEM). (b) porosity after

welding process [11].

3. Motivation

Due to fault detection found in the steel alloy several problems have been heavily

researched and developed [7-9]. The motivation of the proposed system is to

overcome the problem arises from the defect detection found in the hardened

aluminium alloy which is very new in the construction and manufacturing industries.

After the system is develop completely, it can be implement in the manufacturing

industry and helps the engineers to detect any faults found visually and classify them.

The results are then further validated with the conventional method of testing such as

tensile and strength tests. This will ease the burden of having a train eyes to manually

calculate and measure the faults and can effectively provides an automatic visual

inspection at microscopic level.

4. Proposed Solution

The Segmentation processes are the part an image is partitioned into regions. In this

paper, we presented a number of methods and show the possible improvements that

PROPOSAL-1.0 4

(b)(a)


can be achieved. By using the image segmentation algorithm technique, 2d images are

converted to grayscale [5] from the images and analyzed to automatically to detect

and isolate the defects found right after the welding process. The 2d grayscale image

will then be used to detect cracks of the outer surface and specimens are to be cut into

smaller parts which will then be captured again using SEM to detect anomalies of the

inner surface.

An effective and useful technique will be listed and selected to perform image

segmentation on the image which is where the system will based on during the

development phase and provides a comprehensive environment for data analysis, and

data visualization There are several general-purpose algorithms and technique that

will be used for image segmentation and are shown in figure 4.1 [6].

Figure 4.1: List of segmentation techniques

The Otsu’s method, is the simplest method of image segmentation. This

method is based on a threshold value to turn a gray-scale image into a binary image

(see figure 4.2)

PROPOSAL-1.0 5


Figure 4.2: image of Otsu’s technique [10].

As for K-means clustering, the algorithm partitioned and relocates instances by

moving them from one cluster to another until desired clustering structure is obtained

(see figure 4.3).

Figure 4.3: image of K-means clustering technique [10].

Watershed segmentation however [2], is a classical algorithm used for separating

different objects in an image. Starting from user-defined markers, the watershed

algorithm treats pixel values as a local topography (elevation) – see figure 4.4.

Figure 4.4: image of watershed segmentation technique [10].

Lastly, the texture filters is a set of metrics calculated in image processing designed to

quantify the perceived texture if an image. Image texture gives us information about

PROPOSAL-1.0 6


the spatial arrangement of color or intensities in an image or selected region of an

image (see figure 4.5).

Figure 4.5: image of texture filters technique [10].

The development process is listed out in figure 4.6 showing the flowchart of

each process. The algorithm starts with the reading of SEM image. The image which

displays fault in the aluminium alloy is going to be processed one at a time. After the

image is being read and stored, it is going to be converted into 2d grayscale image.

This process is important because it helps the system to analyze the image correctly

using the mentioned image segmentation techniques.

Next, the edges and lines of the image are processed and each of the features

are identified and extracted. In order to facilitate the process, a library called OpenCV

will be used during the development of the system. Before the fault detection is

calculated the image is being segmented into several segments. After this process, the

fault detection is measured and the result is being displayed to the user either by

visual data or in table forms. Overall, the GUI and the development of the algorithm

are to be programmed using Visual C++.

PROPOSAL-1.0 7


Figure 4.6: Development Process

In general, these image segmentation process contributes part of the fault

detection process typically used in a welding process. Figure 4.7 shows the steps

taken for identifying the faults and the image segmentation analysis is positioned

before the hardness test. SEM images used are of x10, x25 and x27 microns and these

are taken after the grinding and polishing process.

PROPOSAL-1.0 8


Figure 4.7: Welding Process

PROPOSAL-1.0 9


5. Benefits / Impact / Significance of Project

The main significant of the project is to better analyze the fault found in aluminum

welded materials. This will benefit the manufacturing industry and simultaneously

saves time and costs of using trained eyes by manual inspection. In addition, the

algorithm can represent a visual form of inspection and the data analyzed can be used

into something meaningful to the engineers to compliment their method of measuring

the aluminium alloy’s strength and durability

6. Uniqueness of Proposed Solution

The use of hardened aluminium alloy for welding is new for the construction and

manufacturing domain. The algorithm used for developing this system is unique as it

helps to solve the manufacturing problems of using hardened aluminimum alloy

materials and automatically detect and measure the faults and defects after the

welding process. This will enhance the overall simplicity and effectiveness of the

system.

7. Expected Outcomes

The expected outcomes of this project is a measureable analysis inspected visually in

terms of the quality of the welded surface. In order to better visualize the data, a graph

is to be presented accordingly based on the result obtained. This process will help the

engineers to better understand the outcome of every welded surface of every

conditions and will give further insight and evidence of how and why the welded

areas can fail.

PROPOSAL-1.0 10


8. Status of the Project

The proposed solution is an enhancement over existing algorithm and it is going to be

developed from scratch based on the requirements set by the welded process from the

engineer at School of Mechanical Engineering, Universiti Sains Malaysia (USM) and

most of the data to be analyzed are gathered and captured from there. The overall

progress of the system is laid out in the Gantt chart form in appendix A. The chart

also listed the validation process which will be validated based on the success rate of

fault detection, which will be done by the experts at the School of Mechanical

Engineering, Universiti Sains Malaysia (USM).

9. References

[1] R. Nejatpour, A.A. Sadabad, “Automated Weld Defects Detection Using Image

Processing and CAD Methods”, ASME International Mechanical Engineering

Congress and Exposition, Oct 8, 2008

[2] Anju Bala, “An Improved Watershed Image Segmentation Technique using

MATLAB”, International Journal of Scientific & Engineering Research Volume 3,

Issue 6, June-2012

[3] Jiannan Shen, “Application of Image Segmentation In Inspection Of Welding –

Practical research in MATLAB”, University Of Boras, 2012

[4] Paul R. Hill, C. Nishan Canagarah, and David R.Bull, “Image Segmentation Using

a Texture Gradient Based Watershed Transform”, IEEE TRANSACTIONS ON

IMAGE PROCESSING, VOL. 12, NO. 12, DECEMBER 2003

[5] K. Parvati, B. S. Prakasa Rao, and M. Mariya Das, “Image Segmentation Using

Gray-Scale Morphology and Marker-Controlled Watershed Transformation”, Discrete

Dynamics in Nature and Society, Volume 2008, Article ID 384346, 8 pages

[6] Krishna kant Singh, Akansha Singh, “A Study Of Image Segmentation Algorithms

For Different Types Of Images”, IJCSI International Journal of Computer Science

Issues, Vol. 7, Issue 5, September 2010

PROPOSAL-1.0 11


[7] D. Narsimhachary, “Effect of Laser Welding Parameters on 6061 Aluminium

Alloy”, Department Of metallurgical And Materials Engineering, National Institute

Of Technology Rourkela, Odisha-769008. 2014

[8] Remi Cogranne, (2014) “Statistical detection of defects in radiographic images

using an adaptive parametric model”, Signal Processing, Volume 96, Part B, Pages

173–189

[9] Saba Madani, Mortaza Azizi "Detection of Weld Defects in Radiography Films

Using Image Processing". 2015

[10] MathWorks. 2015. Image Segmentation. [Online]. [Accessed 2 Oct 2015].

From: http://www.mathworks.com/discovery/image-segmentation.html

[11] WELDER’S Visual Inspection, HANDBOOK, May 2013, pp. 32-33.

[12] Li, C.-T, and Chiao, R. (2003) “Multiresolution Genetic Clustering Algorithm for

Texture Segmentation”, Image and Vision Computing, 21 (11), pp. 955-966.

PROPOSAL-1.0 12

http://www.mathworks.com/discovery/image-segmentation.html


10.Appendix A

Figure A: Project timeline (Gantt Chart)

PROPOSAL-1.0 1

CAT400 ProjectProposal Azeem Edited3

Documents

Transcript of CAT400 ProjectProposal Azeem Edited3