A Data-driven Approach to Actuator and Sensor Fault Detection, Isolation and Estimation in Discrete-Time Linear Systems
For control engineers, this work offers a practical data-driven fault diagnosis approach that avoids complex system identification steps, though it is an incremental improvement over existing methods.
This paper presents a data-driven method for fault detection, isolation, and estimation in discrete-time linear systems using only input-output measurements, without requiring system identification of the observability matrix. The proposed filters provide asymptotically unbiased estimates and allow error compensation, with simulations demonstrating improved performance over existing methods.
In this work, we propose explicit state-space based fault detection, isolation and estimation filters that are data-driven and are directly identified and constructed from only the system input-output (I/O) measurements and through estimating the system Markov parameters. The proposed methodology does not involve a reduction step and does not require identification of the system extended observability matrix or its left null space. The performance of our proposed filters is directly connected to and linearly dependent on the errors in the Markov parameters identification process. The estimation filters operate with a subset of the system I/O data that is selected by the designer. It is shown that the proposed filters provide asymptotically unbiased estimates by invoking low order filters as long as the selected subsystem has a stable inverse. We have derived the estimation error dynamics in terms of the Markov parameters identification errors and have shown that they can be directly synthesized from the healthy system I/O data. Consequently, the estimation errors can be effectively compensated for. Finally, we have provided several illustrative case study simulations that demonstrate and confirm the merits of our proposed schemes as compared to methodologies that are available in the literature.