High-precision and efficiency diagnosis for polymer electrolyte membrane fuel cell based on physical mechanism and deep learning

Abstract

As a nonlinear and dynamic system, the polymer electrolyte membrane fuel cell (PEMFC) system requires a comprehensive failure prediction and health management system to ensure its safety and reliability. In this study, a data-driven PEMFC health diagnosis framework is proposed, coupling the fault embedding model, sensor pre-selection method and deep learning diagnosis model. Firstly, a physical-based mechanism fault embedding model of PEMFC is developed to collect the data on various health states. This model can be utilized to determine the effects of different faults on cell performance and assist in the pre-selection of sensors. Then, considering the effect of fault pattern on decline, a sensor pre-selection method based on the analytical model is proposed to filter the insensitive variable from the sensor set. The diagnosis accuracy and computational time could be improved 3.7% and 40% with the help of pre-selection approach, respectively. Finally, the data collected by the optimal sensor set is utilized to develop the fault diagnosis model based on 1D-convolutional neural network (CNN). The results show that the proposed health diagnosis framework has better diagnosis performance compared with other popular diagnosis models and is conducive to online diagnosis, with 99.2% accuracy, higher computational efficiency, faster convergence speed and smaller training error. It is demonstrated that faster convergence speed and smaller training error are reflected in the proposed health diagnosis framework, which can significantly reduce computational costs.