A model-based fault diagnosis in uncertain systems

This paper deals with the fault diagnosis problem in uncertain linear systems having undermodelling, linearization errors and noise inputs. The new approach proposed in this paper uses an appropriate test variable and the difference between system parameters which are estimated by the least squares method to locate the fault. The singular value decomposion is used to decouple the correlation between the estimated system parameters and to observe the trend of parameter changes. Some simulations applied to aircraft ergines show good allocation of the fault even though the system model has significant uncertainties. The feature of the approach is to diagnose the uncertain system through simple parameter operations and not to need complex calculations in the diagnosis procedure as compared with other methods.