Development of online systematic condition assessment architecture for integrated PEMFC systems based on data-driven random matrix analysis

Abstract

Due to complex configuration of high power integrated PEMFC systems, the associated systematic condition assessment is still a promising challenge. In this paper, an online systematic condition assessment architecture for high-power integrated PEMFC systems is put forward based on random matrix analysis. The proposed architecture consists of two cascaded procedures, which are the streaming formulation of random characteristic matrices and random matrix analysis based systematic condition assessment, respectively. Benefited from the signal cluster transformation by the fusion of model-driven and data-driven approaches, the residuals characterizing system abnormal can be extracted to formulate the streaming random matrices. On this basis, by recursive eigenpairs' updating of random covariance matrices, high-dimensional analysis can be conducted in real-time even with random tensor augmentation-based matrix dimension expansion, and the systematic condition assessment indicators can be derived. Taking temperature anomaly awareness as an example, detailed experiment results demonstrate that, the derived indicators are more sensitive to system anomaly than traditional threshold-based condition assessment method, and the online evaluation of the operation condition of integrated PEMFC systems can be achieved more effectively under the proposed systematic condition assessment architecture. Finally, a recommended online robust systematic condition assessment procedure with the fusion of multi-indicators is demonstrated. To our best knowledge, this paper represents the attempt to put random matrix analysis into the online systematic condition assessment of high-power integrated PEMFC systems for the first time.