Skip to Main Content
This paper develops and experimentally demonstrates a new class of high-fidelity model-based fault detection and isolation filters for three-phase AC-DC power electronics systems. The structure of these filters is similar to that of a piecewise linear observer and in the absence of faults the filter residual converges to zero. On the other hand, whenever a fault occurs, by appropriately choosing the filter gain, the filter residual will exhibit certain geometric characteristics that allow the fault to be detected and, in certain cases, also isolated. Key advantages of these filters include fast detection of all possible component faults and the ability to capture slow degradation in individual components. In order to experimentally demonstrate their feasibility, the filters are implemented on an ultra-fast application-specific real-time processor. While the theoretical framework developed is general, the analysis, simulations, and experiments are focused on widely used power electronics systems implementing three-phase AC-DC converters that are used in, e.g., motor drive applications and distributed static compensators.