A holistic approach for an intelligent laser beam welding architecture using machine learning for the welding of metallic bipolar plates for polymer electrolyte membrane fuel cells

Laser beam welding is the state-of-the-art technology for joining micro-formed metal foils in the manufacturing of bipolar plates for proton exchange membrane fuel cells. However, the process is limited in the achievable welding speed since humping and undercut effects can occur at high feed rates. These effects significantly reduce the weld seam quality, causing scrap or subsequent failure during operation. As a result, higher manufacturing costs arise and additional quality assurance is needed. In this work, welding experiments, including a photodiode-based sensor system, were conducted on AISI 316L metal foils to evaluate the capability of this sensor for inline and online quality assurance. Based on the results, an intelligent laser beam welding architecture is proposed, representing a holistic approach for a multi-sensor-based and self-improving quality assurance system. The theoretical architecture combines a novel laser beam welding concept with different optical and acoustic sensors for determining the current weld state. It considers sensor data fusion for relevant information on the process behavior via dedicated algorithms applying deep neural networks. The approach is an idea of a predictive weld state determination for a precise and real-time capable weld seam quality assurance.