Automating the Analysis of Resistance Spot Welding Requirements in Automotive Manufacturing Using Machine Learning

In automotive manufacturing, the quality of resistance spot welding is significant to structural safety of Body-in-White assemblies. The traditional rule based methods for joining requirements verification are usually manual and time-consuming. This thesis investigates how artificial intelligence can enhance the prediction and analysis of weld quality by comparing traditional methods with data-driven validation methods. The dataset derived using a 3D analysis tool was processed using feature engineering techniques including correlation analysis, minimum redundancy maximum relevance, principal component analysis, and t-distributed stochastic neighbor embedding. Multiple machine learning models were developed and evaluated, including eXtreme Gradient Boosting, support vector machine, and feedforward artificial neural network. Ensemble models, particularly eXtreme Gradient Boosting combined with support vector machine, outperformed individual models with 99% recall on defects and 90% on acceptable welds. These findings demonstrate the potential of artificial intelligence to automate and improve the resistance spot welding verification process, ultimately increasing production efficiency and product safety in the automotive industry.