A Data-Driven Robust Fault Detection Method for Linear Systems with Full-Order Sensors

This paper proposes a data-driven robust fault detection scheme for linear systems with full-order sensors but unknown system internal parameters. Considering the disturbances during practical system operation, the robust fault detection index is incorporated to design the residual generator for fault detection. The key parameters of the robust index are derived from the system data based on the subspace identification, and an optimal parity vector is solved to construct the data-driven robust residual generator. The proposed data-driven method can better adapt to the fault detection task in actual systems compared with the model-based methods. Meanwhile, compared with the relevant data-driven results, the proposed method can reduce the conservatism in the robust fault detection scheme with a lower false alarm rate. The simulation results of a benchmark DC circuit system illustrate the improved performance of the proposed scheme.