LAUSR

Abstract Fault diagnosis technology has been developed to improve the reliability of fuel cell systems in pursuit of successful commercialization. In this study, a hybrid fault diagnosis method is proposed to improve diagnosable fault magnitudes and diagnostic accuracy. Six types of faults in the air supply system of a proton exchange membrane fuel cell system were therefore defined and diagnosed according to the relevant components and locations as actuator, sensor, and piping faults. The proposed method applies an artificial neural network classifier as a data-based diagnostic tool within a model-based diagnosis method that relies upon residual patterns to address the limitations of the model-based diagnosis method (insufficient accuracy for initial fault diagnosis) and the data-based diagnosis method (need for a large dataset to generate a classifier). The proposed method is shown to improve the diagnostic accuracy and decrease the diagnosable fault magnitude compared to solely model-based and data-based methods. Moreover, the proposed method enables a faster diagnosis of air supply system faults, preventing loss of system efficiency and stack degradation. By providing fast and accurate diagnoses, the proposed method is expected to help develop an effective fuel cell health management system.