A Prior Knowledge-Based SimpleNet Model for Fuel Cell Bipolar Plates Defect Detection

Fuel cells play a crucial role in future energy systems, and their key components, metal bipolar plates (BPPs), are prone to various anomalies during the production process, which can seriously affect their overall efficiency and lifetime. In this paper, prior knowledge-based SimpleNet model is proposed to detect the BPPs, but the significant scale difference between different defect types makes the original architecture difficult to achieve high-precision anomaly detection. For this reason, this paper constructs anomaly feature synthesis related to the prior knowledge of defects to adapt to the detection of different defect types. The results show that the detection accuracy is significantly improved by ${2 - 1 1 \%}$ after constructing the prior knowledge. This study highlights the potential of unsupervised learning and representation for anomaly detection in metal BPPs with significant scale differences.