Effects of compressed sensing on classification of bearing faults with entropic features

The ability of automatically determining the underlying fault type in-situ for a roller element bearing is highly desired in machine condition monitoring applications nowadays. In this paper, we classify roller element fault types under a compressed sensing framework. Firstly, vibration signals of roller element bearings are acquired in the time domain and resampled with a random Bernoulli matrix to emulate the compressed sensing mechanism. Sample entropy based features are then computed for both the normalized raw vibration signals and the reconstructed compressed sensed signals. Classification performance using Support Vector Machine (SVM) shows slight per formance degradation with significant reduction of the bandwidth requirement.