Quality Monitoring of Hairpin Joints Using Optical Coherence Tomography and Machine Learning

Modern stators for electric traction drives rely on a technique known as hairpin winding, which necessitates laser beam welding to join rectangular copper conductors. As stators contain a vast number of hairpins, it is vital to ensure that pin joints meet both electrical and mechanical requirements to optimize stator performance. As defects from downstream processes culminate here, the laser beam welding process significantly contributes to the overall connection quality. Recently, this proliferated the adoption of Optical Coherence Tomography (OCT) for quality monitoring in laser beam welding of hairpins. Furthermore, the application of machine learning techniques in this context indicates the feasibility of automated quality monitoring. Thus, in this work, we present a machine learning-based method to monitor defects in hairpin welds. We utilize OCT-based multivariate time series data to train a classifier to assess the quality of the joints following the welding step. Our evaluation encompasses both classical machine learning as well as current deep learning methodologies to compare their effectiveness in distinguishing between high and low-quality welding.