Cross-Domain Electrochemical Impedance Spectroscopy Reconstruction for PEMFC Aging Characterization: A Time-Frequency PINN Framework with ECM Priors

Proton exchange membrane fuel cells (PEMFCs) hold strong prospects due to their high efficiency and low emissions; however, in engineering operation, electrochemical impedance spectroscopy (EIS) is typically discretely acquired and data-scarce, and a reliable mapping from time-domain operational data to frequency-domain EIS is still lacking, which hinders characterization and tracking of the continuous spectral evolution with aging. To address this gap, we propose a time–frequency joint learning framework, termed Physics ECM Informed Neural Network (PEINN), which integrates an equivalent circuit model (ECM) to learn a cross-domain mapping from operational time-series representations to impedance spectra, enabling frequency-domain regression and reconstruction of EIS and comparative analysis of spectral evolution across aging stages. Using only 40% of the data for training, PEINN achieves R 2 >0.99, demonstrating strong potential for health-state assessment. Cross-dataset validation on additional PEMFC datasets further shows stable predictive performance, confirming good generalization and robustness. Moreover, within the proposed time–frequency aging characterization framework, we combine the frequency-domain evolution of the predicted EIS with peak-feature verification in the distribution of relaxation times (DRT) time-constant domain, enabling decoupled identification and interpretation of dominant processes such as charge-transfer deterioration and aggravated mass-transport limitation, thereby providing an explainable basis for health assessment and degradation inference.