A Comparison of Traditional Fourier and Maximum Entropy Spectral Methods for Vibration Analysis

Abstract

The structural vibrations monitored on a process plant often contain burst phenomena where several discrete frequencies are present. Spectral analysis of these vibration signals by Fourier transform methods is generally considered incapable of resolving discrete frequencies close together in a short time slice. The power spectral density estimates computed by traditional Fourier transform methods are compared with those computed by more recent maximum entropy spectral analysis (MESA) methods for both data from a test system where the outputs of two tuned filters, excited by pseudo-random binary noise, are summed, and the vibration signals from an accelerometer mounted on a die in an extrusion process. The frequency resolution characteristics of the two methods are compared, and the problems associated with the choice of an optimum order autoregressive prediction error filter for the MESA method are discussed.