Diagnosis of flooding and membrane dry fault of proton exchange membrane fuel cell based on clustered physical-informed neural network

Proton exchange membrane fuel cells (PEMFCs) are highly efficient clean energy technologies, but their reliability in automotive applications is severely hindered by internal water management faults. Accurate, real-time fault diagnosis remains challenging due to complex internal nonlinearities and the scarcity of labeled fault data in real-world operations. To bridge this gap, this study proposes a novel Clustered Physics-Informed Neural Network (C-PINN) for the unsupervised diagnosis of PEMFC flooding and membrane drying faults. By integrating the Gaussian hybrid model with the dual-domain PINN architecture, the dual-domain physical constraints are embedded in the PINN and the physical residuals are calculated in real time. Flood and dry faults are accurately diagnosed through cascade diagnostic strategies. The results show that the detection accuracy is 99.44%, and the Area Under the ROC Curve (AUC) value is more than 0.90 under sensor drift, providing a physically explainable and scalable solution for fuel cell health management.