Transfer learning-enhanced transformer for virtual process sensing in resistance spot welding

Resistance Spot Welding (RSW) is a crucial manufacturing process, especially in automotive manufacturing. However, quality assurance in RSW remains a challenge, due to the lack of low-cost, non-invasive process sensing techniques. This paper investigates Transformer for generating virtual electrode force and displacement signals from physically measurable DR signals. Furthermore, to enhance the generalizability, the transformer has been enhanced by transfer learning to adapt to dynamic welding scenarios, by pre-training the transformer model with data from diverse welding conditions and fine-tuning using minimal experimental data from novel conditions. Experimental results validate the effectiveness of virtual sensing technology for RSW process monitoring.