Fault feature extracting by wavelet transform for control system fault detection and diagnosis

The paper deals with the problem of fault feature extraction from the residual in the model-based control system fault detection and diagnosis, based on the fact that the wavelet transform of a signal will maximise the modulus at its singular point in the transform domain, and the fault error has positive singularity exponent while the noise has negative singularity exponent at the corresponding singular points; the fault error and noise mixed in the residual can be separated from each other by multi-scale wavelet transform, and the modulus maximum can be taken as the fault feature, so that the fault feature becomes clearer and more recognizable and a correct decision as to whether the system fault will take place or not can be correctly made in the transform domain. This makes it easy to detect and diagnose faults in the control system