Blind component separation for highly corrupted ultrasonic signals in real-time spot weld inspection

This paper introduces an efficient technique to analyze ultrasonic signals obtained in the process of nondestructive testing of resistance spot welds. The ultrasonic transducer, integrated inside a water-cooled welding electrode, sends multiple impulses into the welded metal sheets in reflection mode. The harsh environment surrounding the inspection system generates multiple sources of noise. Bubbles in the cooling water stream, temperature variations around transducer, electromagnetic noise from power electronics, transformers and servo motors corrupt the received A-scans. The method describes how to separate signal components and extract desired features from corrupted A-scans to distinguish the reflections from multiple interfaces of spot weld structure. As welding progresses, the technique is capable of monitoring the dynamics of the key parameters of spot weld growth and the presence of the liquid nugget within welded sheets. Several experiments were carried out using the welding system and robots available at the Institute for Diagnostic Imaging Research as well as the data received from leading auto makers in the North America and Europe.