A Hybrid Diagnosis for Gas Starvation Faults in Proton Exchange Membrane Fuel Cells

Fuel cell technology is recognized as a green energy alternative for the transport sector and stationary power applications, boasting high current density and zero emissions. Nonetheless, broad adoption and commercial viability of the technology is limited by its low durability and reliability. These limitations can be addressed by developing and implementing a fuel cell management system. In this paper, gas starvation faults in proton exchange membrane fuel cells (PEMFCs) have been studied. The causes, consequences, and the fault indicators (features) are identified. The sensitivities of fault indicators and how they affect the performance have been evaluated. Subsequently, a hybrid fault diagnosis method that combines residual-based fault detection with data-driven fault isolation has been proposed. Fault detection is achieved using a physics-based white-box model while fault isolation is achieved using a data-driven approach. Two supervised machine learning classifiers, namely the k-nearest neighbor (kNN) and the support vector machine (SVM) have been developed. The performances of these methods are compared in terms of their accuracy and computation time. Moreover, the effect of additional indicators has been evaluated. The results show that starvation faults on the PEMFC can be detected and isolated efficiently and correctly, thanks to the fault indicators and the hybrid nature of the diagnostic method. Also, from the presented results, it can be deduced that the kNN classifier has outperformed the SVM classifier.