Abstract: A methodology for the design of a general comb filter is proposed, and it is demonstrated that the traditional time domain averaging is just a special case of the general comb filtering. On this basis, a novel signal analysis technique capable of extracting the periodic waveform with inexactly known period from noisy discrete-time observations is presented. Firstly the region that the signal fundamental frequency may lie in is finely discretized, resulting in an array of candidates of the fundamental frequency estimate. Then the contaminated signal is independently cleaned by an array of comb filters, each with a candidate of the fundamental frequency estimate as the reference frequency. Finally the obtained series of filtered signals are depicted as a gray image, and the interested waveforms are picked up for further analysis. The validity of the approach is confirmed by analyzing the synthetic signal and the vibration signal produced by multiple point defects in a rolling element bearing.

Key words: Periodic waveform, Array, Comb filter, Fault diagnosis, Signal processing