Genetic Stochastic Resonance: A New Fault Diagnosis Method to Detect Weak Signals in Mechanical Systems

Abstract

Mechanical faults, such as defects or wear of bearings and gears, are often characterized by the presence of periodically weak signals, which are contaminated by strong background noises and difficult to differentiate using traditional methods. In this paper, we propose a novel genetic stochastic resonance scheme for feature extraction of the above signals. Firstly, the original data are input into a stochastic resonance system with double-well potential. Then, the signal-noise-ratio (SNR) of the output is selected as the fitness function of genetic algorithm and the optimal structure parameters of the bi-stable system correspond to the maximum of SNR. With the optimized parameters, weak periodical components are sufficiently amplified. Meanwhile, the proposed method is combined with frequency-shifted and re-scaling stochastic resonance (FRSR) to extract the high-frequency components, which can be hardly achieved with traditional stochastic resonance. Finally, the present algorithm is applied to simulation data as well as vibration acceleration signals measured on defective rolling bearings with outer race fault. Results demonstrate that the weak periodical features, either in low or high frequency band, can be well extracted even if there are limited sample points.