Equivalent Circuit Model Analysis for Data-Driven Oriented Diagnosis of High-Level CO in HT-PEMFC with EIS

Different equivalent circuit models (ECMs) of Electrochemical impedance spectroscopy (EIS) were analyzed in terms of parameter identification as features for online data-driven diagnosis of CO in the high temperature proton exchanged membrane fuel cell (HT-PEMFC). Parameter identification was performed and analyzed for feature extraction in machine learning model training. The EIS data were tested under 0, 0.75 and 1.5% CO and 5-100A load current on a 10-cell short fuel cell stack. The three levels of CO can be successfully identified via artificial neural network (ANN) and support vector machine (SVM). Anode reaction(1000-100Hz) and diffusion(100-5Hz) influenced by CO were suggested as two factors for the interpretability of the selected ECM. On the other hand, the simple ECM with fewer electrical components should be selected provided it can meet the diagnosis requirement by machine learning methods. This work contributes to the selection of ECM and the interpretation of machine learning methods for online diagnosis on HT-PEMFC with EIS.