Hierarchical fault diagnostic method for a polymer electrolyte fuel cell system

This study proposes a hierarchical method for on-line fault detection and diagnosis (FDD) of a stack and balance of plants (BoPs) in a polymer electrolyte fuel cell (PEFC) system. Because the fuel cell system consists of various subsystems with different characteristics, we have developed a multi-stage structure with subsystem-level FDD. In the first stage, faults were diagnosed at the subsystem level. In the next step, component-level faults were identified in the corresponding subsystem. The model-based approach in this study is composed of process estimation, residual generation, and FDD. Supervised machine learning methods were applied to train models for regression and fault classification. Residuals, the difference between analytic redundancies and measured results, were employed as fault indicators, i.e., residuals were used to detect faults and to generate fault patterns. Analytic redundancies were calculated using regression models. Several abrupt and performance degradation faults were considered. Because long-term performance degradations were difficult to introduce in the experimental system, the proposed method was evaluated using test data obtained by artificially decreasing the performance or sensor readings for a short period of time. This study focuses primarily on subsystem-level FDD and demonstrates one scenario of second level FDD. The experimental results verified the accuracy of the model-based approach and demonstrated that the proposed multi-stage hierarchical method effectively diagnosed faults in a PEFC system.