Impedance Field-Based Fault Diagnosis for Fuel Cells

Electrochemical impedance spectroscopy (EIS), serving as an invasive yet insightful diagnostic method, unveils a trove of status information and has been utilized in data-driven state classification. However, the depth of research into the impedance evolution trend is hindered by limited experimental data. While machine learning (ML) has been employed to forecast state, most research in this area has neglected the field perspective, thereby underutilizing the rich informational potential for state recognition inherent in three-dimensional (3D) projections. This paper introduces an impedance field paradigm, leveraging extensive lithium-ion battery (LIB) EIS data for model training. By fabricating a mirrored impedance field and projecting it onto a plane, we generate impedance density images for the purpose of state classification. Based on the correlation between specific states and their impedance evolution patterns throughout cell lifecycle, we developed a ML model capable of anticipating impedance field fluctuations, which was applied to the fault diagnosis of proton exchange membrane fuel cells (PEMFCs). This method not only reconstructs the lifespan impedance field from single impedance curve but also employs image recognition to diagnose faults such as flooding, drying and oxygen starvation. The method has achieved 98.5% accuracy in PEMFCs fault diagnosis.