Fault detection for LPV systems using model parameters that can be estimated via linear least squares

Artikel i vetenskaplig tidskrift, 2014

This paper presents a fault detection approach for discrete-time affine linear parameter varying systems with additive faults. A finite horizon input-output linear parameter varying model is used to obtain a linear in the model parameter regression residual form. The bias in the residual term vanishes because of quadratic stability of an underlying observer. The new methodology avoids projecting the residual onto a parity space, which in real time requires at least quadratic computational complexity. When neglecting the bias, the fault detection is carried out by an χ2 hypothesis test. Finally, the algorithm uses model parameters that can be identified prior to the on-line fault detection with linear least squares. A realtime experiment is carried out to demonstrate the viability of the proposed method.