A similarity-based prognostics approach for full cells state of health

This paper details an improved method in the 2014 PHM Data Challenge which was organized by the IEEE Reliability Society. The task is to estimate the State of Health (SoH) of a proton exchange membrane fuel cell (PEMFC) system using performance data. The estimation of SoH is a key technique to improve fuel cell system's life span and reliability. This method bases on the similarity relationship in characteristic curves with different operating time. The curves include static and dynamic information which is reflected from polarization curve records and Electrochemical Impedance Spectroscopy (EIS) measurements in this data challenge. All the information contains fuel cell system's ageing degree which needs to be extracted. A back propagation neural network and polynomial models were involved in the approach to approximate the given curves and make prognostic. The complete estimation method consisted of several steps. Firstly, we processed and classified the given data for preparation. Second, we extracted the features from curves using single modeling strategy. Then we utilized the model to predict the SoH at different instants. Finally, statistic methods were adopted for fusing network and empiric model to obtain the ensemble model for final result. The performances of the proposed method were scored by the data challenge means. The improved method gained a more accurate result comparing with using single model and had a good rank in all competition results. The result shows utilizing similarity in characteristic curves is an effective method to estimate fuel cell system's SoH.