Optimization of wavelet multiresolution analysis of shock signals. Application to the signals generated by defective rolling bearings

The aim of this paper is to propose the wavelet analysis as an effective tool allowing to improve the sensitivity of scalar indicators (Kurtosis and crest factor) for the identification of mechanical faults inducing impulsive forces. Indeed, these indicators are very sensitive to the variations in temporal signal due to the periodic shocks caused by the defect. Nevertheless, their reliability is immediately limited by the presence of high level of random noise associated to a small gravity of the defect. Wavelet Multiresolution Analysis (WMRA) is proposed as a solution to this problem. Adapted for such objective, several of its parameters were chosen, even optimized. Initially, the proposed method is applied to simulated signal. For the experimental validation, several series of experiments were realized on ball and cylindrical roller bearings, on which various defects were caused. Measurements were taken with various configurations, 210 signals were measured at different sampling rates and rotation speeds on a laboratory test rig, the industrial application is carried out on a turbo alternator.