Lossy Vibration Compression through Matching Pursuit

Abstract

This paper introduces a method for lossy compression of mechanical vibrations, based on the Matching Pursuit (a general purpose time-frequency-scale signal transform) and some usual compression algorithms. The compressed vibration can be sent through industrial communication channels. The ultimate goal is to detect possible mechanical faults, while analyzing the vibration at the destination. The algorithm employs a harmonic wavelets dictionary (atoms), from which the best approximates of the initial vibration are extracted. Thus, the samples of the raw vibration are replaced by projection coefficients on the dictionary and their indexes. In other words, instead of compressing the initial vibration samples, one has to compress the most fitted structure of the dictionary atoms, while removing the final residual that cannot encode useful information. Several compression algorithms are applied on both signals (the raw vibration and the harmonic wavelet coefficients), in order to compare their performances. One proves that the compression ratio significantly decreases when using coefficients and indexes instead of vibration samples.