Optimization of wavelet filters for parity relation-based fault detection

This paper revisits a wavelet approach for parity relation-based fault detection and proposes an improvement through the adaptation of the wavelet filters employed in the decomposition of the residue signal. In the parity space approach under consideration, the parity vector is obtained by minimizing a cost that expresses a compromise between sensitivity to faults and robustness against external disturbances. The proposed improvement consists of optimizing the wavelet filter parameters in order to further reduce the resulting cost value. An example involving the model of a two-mass-spring system is presented for illustration. The results show that the proposed filter optimization procedure results in a larger increase of the residue following the onset of a fault, without introducing additional time delays in the detection process.