Fault detection in a 3-DOF laboratory helicopter using wavelet packets and subspace identification in frequency subbands

This paper proposes a wavelet-packet approach for fault detection using subspace identification. The wavelet-packet decomposition tree is used to define the frequency subbands at which the plant models are created. The best decomposition tree is chosen by using a dynamic programming algorithm that achieves a trade-off between accuracy and parsimony. A recently published framework is adopted to allow for the identification of subband models with a subspace approach. The proposed technique is validated with experimental data acquired with a Quanser™ 3-DOF laboratory helicopter and presents better results than those of a standard time-domain fault-detection technique.