A wavelet-packet approach for analytical redundancy in frequency subbands

This paper presents an analytical redundancy approach for fault detection in frequency subbands. The proposed method employs an optimized wavelet packet decomposition tree to band-limit the signals under consideration. Dynamic consistency checks can be carried out within each frequency subband by using models identified from training data. A fault detection example involving a simulated servomechanism is presented for illustration. The wavelet packet method is compared with a similar technique employing a standard wavelet decomposition tree, as well as with a classic fault- detection scheme based on a Luenberger state-observer. The results indicate that the proposed method has better sensitivity to the fault effect.