Fault detection in a laboratory helicopter employing a wavelet-based analytical redundancy approach

Wavelet-based techniques for fault detection usually employ one of two basic approaches, namely (a) decomposition of a measured signal containing fault-related information or (b) decomposition of a residue calculated as the difference between sensor readings and the output of a model. An alternative approach, which was recently proposed in, consists of employing the wavelet transform to identify a subband model for the normal dynamical behaviour of the system. The resulting subband model is then used to generate a residual signal. Such a fault detection approach was shown to provide good results in terms of sensitivity and false alarm rate. However, the examples presented for validation were previously restricted to simulation studies. The present work is concerned with the application of this wavelet-based fault detection technique to a more elaborate case study involving experimental data. The system at hand consists of a laboratory helicopter operating under closed-loop control in the presence of a persistent disturbance. The results indicate that the technique under consideration can successfully detect a fault of small magnitude, consisting of a 10% reduction in the pitch sensor gain. Moreover, the wavelet approach is shown to outperform a time-domain detector with similar configuration.