Contribution to on-line diagnosis, fault classification and prognosis for PEMFC ; Contribution au diagnostic en ligne, à la classification des défauts et au pronostic des piles à combustible PEMFC

The proton exchange membrane fuel cell (PEMFC) is a promising energy source that offers several advantages such as no pollution, high efficiency, and low operating temperature. However, durability and reliability remain barriers to its large-scale commercialization. The development of tools for on-line diagnosis, fault classification and prognosis of PEMFC is a very important research topic to lift these barriers. The main objective of this thesis is to contribute to the development of these tools. Thus, we have proposed three diagnostic algorithms and one prognostic algorithm to address these challenges. First, a voltage fluctuation-based diagnostic method is proposed, and the faults resulting from different operating temperatures, stoichiometry and relative humidity are studied. The voltage fluctuation model is extracted by the autoregressive model (AR model), and the coefficients of the model are directly applied as diagnostic features. Four fault classification algorithms are proposed, applied, and compared under both single-fault and multiple-fault conditions. In the second stage, two electrochemical impedance spectroscopy (EIS)-based diagnostic methods are proposed and validated in real time, which can distinguish between flooding, drying-out and mass transport faults. The first method is based on an equivalent circuit model (ECM), in which the parameters of electrical elements can be identified and applied as diagnostic features. In addition, the adaptive neuro-fuzzy inference system (ANFIS) is proposed to perform the diagnosis, and the whole diagnosis process is implemented and validated in real time on a digital signal processor (DSP) system. The third proposed diagnostic method is based on zero-phase impedance and turning phase of EIS characterization. Experiments have shown that the two proposed features can represent the health status of the PEMFC in a practical way; therefore, they can be applied as features to perform fast and efficient diagnosis. The K-nearest neighbours (KNN) algorithm is applied ...