A NEW, NON-LINEAR, ADAPTIVE, BLIND SOURCE SEPARATION APPROACH TO GEAR TOOTH FAILURE DETECTION AND ANALYSIS

Abstract

Past condition monitoring techniques of gearboxes have utilised many different approaches such as time-series averaging, amplitude and phase demodulation, time–frequency distribution and wavelet analysis. Only recently have statistical approaches taken a hold in gear tooth failure detection. Non-linear adaptive algorithms for independent component analysis (ICA) have been shown to separate unknown, statistically independent sources that have been mixed in dynamic systems. This paper proposes the application of an information maximisation based blind source separation algorithm (a type of ICA) to gear vibration measurements. It is shown that the individual gear and pinion vibrations cannot be separated using the blind separation algorithm, but the learning curve of the updated parameter can be used to detect impulsive and random changes in the data. It is shown that the algorithm is capable of tracking the higher-order statistics of the meshing signature using a single measure. This results in a detection scheme that is shown to find localised damage in a single tooth failure, two adjacent teeth failures, two non-adjacent teeth failures and multiple adjacent teeth failures. This method does not need a priori information about the loading, speed or type of gear measured.