Feature extraction through chaotic metrics for weld flaw classification

Abstract

Most of the Mechanical structures formed by metals are only through the fusion of metals at high temperatures through various welding methods. The strength of the structures depends only on the perfection in welding process failing in which would lead to loss of structural stability. This ultimately results in disasters and attracts huge investment to reconstruct the structures. It is always preferred to check the quality of the weld before the final welded structure is used for its actual application. Though visual inspections could solve problems tentatively valid for low production rates, there are scenarios where visual inspection fails and needs high end methods to analyze the quality of welded joints. Several measurement techniques have evolved and help the user community. The paper aims at proposing a novel feature extraction namely, Kolmogorov-Sinai Entropy which is widely used in chaotic analysis. The classification of weld flaws are done along with the additional metrics such as kurtosis and skewness calculated from the x-ray images took from ‘GRIMA’ open database.