Robust Spectral Peaks Detection in Vibration and Acoustic Signals

Abstract

This paper brings a practical solution to the problem of spectral peak detection in non-uniform spectra. It applies a robust probabilistic approach that fits the histogram of trimmed spectral data with a truncated Gamma distribution. The estimated distribution parameters are used to derive a threshold through a hypothesis test on the presence of peaks. The proposed approach gains its robustness from the formulation of the no-peak distribution, while no knowledge is available about the amount of peaks in spectral data. The authors propose a pre-processing step to cope with a non-uniform spectrum. The proposed methodology is validated on both simulated and experimental vibration and acoustic signals.