Fault Diagnosis for LTI Systems Based on Subspace Identification Technique

This paper concerns the fault estimation for a class of systems in an innovation form. We propose a new fault diagnosis strategy based on the well-known subspace model identification technique. The underlying faults, mainly sensor bias and actuator additive bias, are parameterized in the model which is estimated in two stages. In the first stage, the contribution of the fault, similar to the residual in classical observer based fault detection system, is estimated. In the second stage, the faults are recovered from the residual. For the component faults, we use online parameter identification to monitor the fault. Recursive online monitoring algorithms are derived. One example is given to illustrate the effectiveness of the proposed method.