Estimation of Undercuts in Mild Steel Weldment using Artificial Neural Network

Abstract

Weld undercut is a common welding defect that occurs when the weld metal penetrates too deeply into the base metal, creating a groove or undercut along the base metal adjacent to the weld. To improve and predict this weld defect with respect to the weld current, voltage and gas flow rate factors, artificial neural network (ANN) was employed. 100 welded specimens of mild steel, measuring 60mm x 40mm x10mm were prepared and measured using the V-WAC gauge. The results were employed to train ANN. The research produced an R2 of 93% in comparison to the experimental result on a fitted line plot using regression analysis, while correlation analysis obtained in the training and validation exercise from ANN were all above 80%. Result of the study have shown that ANN is a robust predictive tool in welding which could help reduce trial and error in welding processes.