Advancing Welding Quality through Intelligent TIG Welding: A Hybrid Deep Learning Approach for Defect Detection and Quality Monitoring

Modern welding procedures are intricate, requiring a variety of variables and occasionally lacking a complete understanding of their underlying mechanics. Despite the adoption of intelligent welding processes in a few applications, there are still several obstacles. By combining advanced search, combinatorial optimisation, geometric reasoning techniques, and comprehensive Artificial Intelligence (AI) modelling cognitive capabilities, the proposed research aims to build intelligent welding. The three main scientific foci of the research are feature correlation to forecast process performance and facilitate corrective actions, feature extraction utilising intense signal analysis, and the use of simulated or supplied data for analysis. Previous research led to the development of an intelligent Tungsten Inert Gas (TIG) welding platform for materials made of aluminium. On the other hand, TIG welding is susceptible to fluctuations in the root gap, which affect the quality of the weld and could result in electrode contamination. Common welding errors include excessive heat-affected zone width, fusion width, bead height, and inadequate penetration. These errors directly affect the strength and load-bearing capacity of the joint while also making it more susceptible to stress and fracture propagation. The proposed AI-powered welding tool is made to overcome common weld imperfections. Therefore, the research's objective is to develop a hybrid deep learning-powered platform for TIG welding. Convolutional Neural Networks (CNNs) will be employed to extract discrete visual characteristics linked to each type of weld defect, establish correlations between these features, and give weld images to identify the types of defects or their absence. The objective of the research is to create a neural network model that can determine whether a given weld image is good or bad due to contamination, burn-through, or lack of fusion. These findings will make precise weld quality monitoring and process improvement possible.