Non-stationary feature extraction by the stochastic response of coupled oscillators and its application in bearing fault diagnosis under variable speed condition

Abstract

Non-stationary feature information is common in engineering applications, and its signal features are irregular compared to stationary signals. Because of its strong volatility, the traditional signal analysis method is no longer applicable. In some environments, strong background noise makes it difficult to extract feature information. Adaptive cascade stochastic resonance is an effective method to enhance the stationary signal. The weak characteristic signal is enhanced step by step, but it is difficult to extract non-stationary information under strong noise background. Compared with the classical bistable system, the piecewise linear system can overcome the disadvantage of output saturation and has high output signal-to-noise ratio. Therefore, an adaptive cascaded stochastic resonance method is proposed to extract and enhance the non-stationary weak feature information. Firstly, the simulated non-stationary signal of a faulty bearing is preprocessed. The non-stationary feature information is transformed into the stationary feature information by computed order analysis method. Combined with maximum correlation kurtosis deconvolution filtering, the periodic feature of the characteristic signal is highlighted. Then, the adaptive stochastic resonance in a piecewise linear system and variational mode decomposition are applied to enhance characteristic signal and reduce noise interference. The cascaded mechanism is used to filter the interference signal and enhance the characteristic information step by step. Finally, the effectiveness of the method is verified by experimental signals, which can significantly improve the output characteristic amplitude and signal-to-noise ratio.