Hierarchical fault diagnosis-mitigation for a high-power proton exchange membrane fuel cell with an ammonia-based hydrogen source based on a deep learning method

The hierarchical fault diagnosis-mitigation is implemented based on a deep leaning method, that is, based on the multi-scale convolutional neural networks (MCNN) model. The MCNN model is utilized for diagnosing the fatal and recoverable faults in the high-power proton exchange membrane fuel cell (PEMFC) with ammonia-based hydrogen sources (AHS) system. In the AHS-PEMFC system, the hierarchical fault diagnosis result shows that when the various faults appear in components, the average diagnostic accuracy based on the MCNN model is 99.11 %, which is composed of 99.31 % for diagnosing the normal state, 99.07 % for the fatal faults and 98.84 % for the recoverable faults. Moreover, the identification rate of fault severity exceeds 98 % for all fault scenarios, where allows the adoption of mitigation strategies based on the severity of the faults. The hierarchical fault diagnosis-mitigation ensures continuous diagnosis of the health status and accurate application of fault mitigation strategies, thereby enabling the system to preserve fault stabilization or restore normal operation. The robustness of hierarchical fault diagnosis-mitigation is confirmed under various fault scenarios in the AHSPEMFC system.