Stochastic-resonance based fault diagnosis for rolling element bearings subjected to low rotational speed

Abstract

Despite been introduced about more than 30 years ago, the Stochastic-Resonance (SR) theory has only been gaining considerable attention in the field of condition based maintenance (CBM) in recent years. SR is a nonlinear physical phenomenon where weak signals (i.e. signals with low signalto- noise ratio) can be enhanced by a cooperative interaction of noise and periodic excitation (stimulus) in particular nonlinear systems, e.g. “bistable” systems. In other words, SR serves as a non-linear filter that can amplify a (periodic) signal of interest heavily masked by a large background noise by adjusting both the parameters of the nonlinear bi-stable system and the intensity of artificially generated noise to be added to the measurement signal. This paper discusses an improvement of SR filtering with multi scale noise tuning recently published for multi-fault diagnosis of a rolling element bearing subjected to low rotational speed and low load. Prior to application of the SR filtering, gear-related signals are removed from a measured acceleration vibration signal (i.e. signal pre-whitening) by means of the cepstra-based discrete components removal. The pre-whitened signal is further processed by employing an optimized SR filter in which the filter parameters are pre-determined based on the faulty data. Finally, features defined as the ratio between the peaks around the corresponding bearing fault frequencies and the background noise level are extracted from the spectrum of the output signal obtained from the SR filtering. A number of experiments have been carried out on a gearbox dynamics simulator with healthy and faulty bearings in order to demonstrate and verify the effectiveness of the proposed approach for bearing fault detection under low shaft rotational speed (348 rpm) and low load (1.2 Nm). For comparison purpose, the experimental data have been analyzed both with the well known “envelope” method and the proposed framework. The experimental results show that the proposed framework outperforms the envelope method.