Automatic detection of defects in welding using deep learning - a systematic review

Abstract

The large number of research studies on defect detection and classification using Deep Learning implies the demand for AI in industrial automation systems. Welding is one of the crucial processes used in various industries with diverse applications, and one of the challenges faced in welding automation is the Defect identification using Non-Destructive Testing. This paper aims to provide a systematic review on the application of defect detection in Tungsten Inert Gas Welding using Computer Vision. This study analyses and compares the existing methodologies used in the earlier days and what is being used now. In addition, this paper discuss how Convolution Neural Network(CNN), a deep learning technique can solve the problems faced by traditional machine vision and how the Vision Transformers can be incorporated to get the better results.