RSW-AI-three-layered – Methodology for Parameterizing 3-Sheet Joints in Resistance Welding Using Artificial Intelligence

First, the welding process for a three-sheet joint is investigated both experimentally and numerically. The numerical simulations expand the data space by introducing additional process variations and, at the same time, provide insights into the welding process that would not be possible through experimentation alone.

Based on this process data, an initial machine learning (ML) model is developed that predicts the formation of the weld bead (the load-bearing weld joint). The algorithm predicts the lower current limit at which the required weld spot diameter is achieved. A second model predicts the upper current limit at which spatter begins to occur. It takes into account thermoelectric processes and physical boundary conditions.

Four requirements are central to model development:

• Robustness: The model responds to real-world process influences in the production environment without overinterpreting them.
• Generalizability: It can be used on different production lines and with various materials.
• Data Security: Since the model processes real production data, it is protected against external attacks.
• Local Operation: The model runs locally, so no data leaves the company.

Finally, all the findings are integrated into a prototype component and demonstrated.