| Summary: | An automatic defect classification system was developed in this research using
simulated image database and polar radius signature and neural network classifier to
identify different types of defect in radiographic images of welds. Programs were
developed by using Languange C to obtain polar radius signature and roughness
parameters from simulated images for subsequent classification. The image processes
involved are blob analysis, binarization, edge detection, etc to extract polar radius signature.
Several roughness parameters such as Ra, Rq, Rz, Rp and Rv were then extracted from the
polar radius signature from the simulated images. Neural network was employed to train
the simulated data. A total of 4 defect types were studied and the classification was carried
out using several different combinations of roughness parameters and types of weld defect.
The highest accuracy of 81.25% was achieved in classifying crack and incomplete
penetration by using five parameters. Therefore, roughness parameters which are extracted
from polar radius signature have potential in weld defect classification.
|
|---|