Multimodal sensor fusion with SWIR imaging and audio for inline gas-tightness monitoring in laser-welded bipolar plates

The gas-tightness of laser-welded bipolar plates (BPPs) is a key quality feature for the functionality of fuel cells. Pressure drop tests to check gas-tightness are very time-consuming and expensive, posing a challenge for large-scale production. This study proposes a novel sensor fusion approach combining short-wavelength infrared (SWIR) imaging and airborne audio signals, analyzed using recurrent neural networks, to enable real-time gas-tightness monitoring during the manufacturing process. The experimental results show a classification accuracy of 96 %, a substantial reduction in overkill rates compared to single-sensor evaluations, and the elimination of escapes that could lead to leaks. The new approach reduces the number of gas-tightness tests required by 47 %. • Combines SWIR imaging and audio for real-time gas-tightness monitoring. • Enables scalable, inline quality control for fuel cell manufacturing. • Fusion of multimodal sensors outperforms single-sensor approaches.