Tutorial: Signal processing methods for machinery diagnostics

Abstract

Conventional machinery monitoring systems rely mainly on signal energy processing methods such as power spectrum analyses in the frequency domain and envelope analysis in the time domain. Other, less well‐known, techniques can be very useful in extracting diagnostic information from the vibration or sound produced from the machine as it operates. These methods include cepstral analysis as an order analysis or deconvolution procedure, and am and fm demodulation using the Hilbert transform to study variations in loading or speed of rotational devices. They also include various system modeling methods such as Kalman filtering and functional expansions of the transfer function. Illustrations of the use of these methods, and descriptions of situations in which each may be helpful will be presented in this tutorial.