Fault diagnosis method for proton exchange membrane fuel cell system based on digital twin and unsupervised domain adaptive learning

In recent years, supervised deep learning-based methods have achieved significant results in fuel cell system fault diagnosis. However, most existing deep learning-based fault prediction methods suffer from missing fault labels and data limitations due to the difficulty in obtaining fault and degradation data in real fuel cell systems. To address the above challenges, this paper proposes a fault diagnosis method based on digital twin and unsupervised domain adaptive learning. The method has two key features: First, a maximum-relevance minimum-redundancy algorithm is used to select the input signals. Then a high-order fuel cell system model is developed to obtain a large amount of digital domain fault data at low cost by simulating fault injection. Second, domain-invariant features are extracted using a domain-adaptive adversarial learning approach to reduce the distribution differences between the digital and real domains. The method successfully diagnosed nine typical faults in the fuel cell air, hydrogen, and thermal subsystems without real data fault labels. Under dynamic load conditions, the diagnostic accuracy reached 92.5 %. In addition, the method achieves a diagnostic accuracy of over 90 % under domain adversarial training using only normal real data. The experimental results show that the proposed method can achieve fuel cell system fault diagnosis without fault labels and significantly reduce the dependence on fault data.