Synchronous Averaging of Time-Frequency Distribution With Application to Machine Condition Monitoring

Abstract

A novel signal analysis technique capable of detecting periodic bursts of exponentially decaying oscillation from noisy discrete-time observations is presented. The time-frequency distribution of the signal is viewed as a series of time signals corresponding to different frequency variables, and each time signal is processed with the technique of synchronous average scanning over a frequency region of interest. The energies of the outputs corresponding to different frequency bands and different average reference frequencies are recorded in a 2D array and depicted as a gray level image. Detection of the signal of interest leads itself to the identification of the spot in the image. The period of the impulse train and the frequency of the oscillation can be further estimated from the location of the spot. Examples that diagnose machine faults under complicated conditions are given to confirm the validity of the approach.