Abrupt Sensor Fault Diagnosis Based on Wavelet Network

The possible faults of a sensor may be classified as abrupt (sudden) faults and incipient (slowly developing) faults. This paper focuses on the abrupt faults of a sensor. Due to the limited number of scales, a single wavelet amplitude map has not enough scales to describe all details of the signal. The sampling grid in the scale direction is rather sparse, Some of the fault information will be leaked under such sparse grid. To make up for the deficiency of scalar orthogonal wavelet transform in the application of abrupt fault diagnosis, multiwavelet packets transform was introduced into the field of abrupt fault diagnosis. The distribution differences of the signal energy on decomposed multiwavelet scales of the signal before and after the fault occurring are extracted as the fault feature and used as the input of multi-dimensional wavelet network. A new model-free diagnostic method for isolating abrupt sensor faults is developed based on a proposed algorithm of multi-dimensional wavelet network constructing. The method has been proved to be quite effective in the detection of sensor abrupt fault.