Weld Defect Categorization from Welding Current using Principle Component Analysis

Abstract

Real time welding quality control still remains a challenging task due to the dynamic characteristic of welding. Welding current of gas metal arc welding possess valuable information that can be analyzed for weld quality assessment purposes. On-line monitoring of motor current can be provided information about the welding. In this study, current signals obtained during welding in the short-circuit metal transfer mode were used for real-time categorization of deliberately induced weld defects and good welds. A hall-effect current sensor was employed on the ground wiring of the welding machine to acquire the welding current signals during the welding process. Vector reduction of the current signals in time domain was achieved by principle component analysis. The reduced vector was then classified by various classification techniques such as support vector machines, decision trees and nearest neighbor to categorize the arc weld defects or pass it as a good weld. The proposed technique has proved to be successful with accurate classification of the welding categories using all three classifiers. The classification technique is fast enough so it can be used for real time weld quality control as all the signal processing is carried out in the time domain.

On welding defect detection and causalities between welding signals

“…Huang [1] proposed a Support Vector Machine (SVM) model based on the multi-scale entropy of the current and voltage signals that detects three types of defects. Arabaci & Laving [2] carried out a study to categorize welding defects from raw current signals using principal component analysis. Pernambuco [3] used Artificial Neural Network for the classification of sound signals to detect the absence of shielding gas.…”