Study On Mechanical Fault Feature Extraction Method Based On EMD And Stochastic Resonance

With the development of modern production equipment toward automated and intelligent, condition monitoring and fault diagnosis technology of machanical equipment has become a hot research point. Time frequency analysis method can effectively realize the nonlinear, non-stationary signal analysis and feature extraction, becoming a hot spot of research in modern signal processing methods. In this paper, on the basis of previous work, aiming at the respective shortcomings and limitations of EMD and stochastic resonance, mechanical fault feature extraction method based on improved EMD and stochastic resonance is proposed. A further study is conducted on the end effect of EMD and the application of stochastic resonance in multi-frequency weak signal feature extraction. Improved EMD algorithm and stochastic resonance theory are proposed with certain reference significance practical value for mechanical fault feature.For the end effect of EMD algorithm, a novel integrated method that combines waveform feature matching extension and cosine window function is proposed. First of all, waveform feature matching extension is used to achieve a smooth transition at the junction of the original signal and its extension and avoid the instantaneous frequency jump at the boundary. Secondly, aiming at existing extension error in the extension method, the signal is processed by cosine window function. Thus the error is controlled at both ends. This ensures the correct decomposition of effective data, and raises the decomposition accuracy to realize the EMD algorithm improvement. Simulation results and misalignment fault diagnosis examples show that the improved method can inhibit end effect effectively in rotating machinery fault diagnosis.Aiming at the detection problem of the multi-frequency signal under noise background, a novel method based on Re-scaling frequency-shifted band-pass stochastic resonance (RFBSR) is proposed in this paper. In this method, different frequency bands of the signal are processed by re-scaling sub-sampling compression, so as to make each frequency band meet the conditions of stochastic resonance. Further the weak signal frequency components are enhanced through stochastic resonance system. Before the enhanced components of the signal are synthesized, they are processed through band-pass filter only leaving the enhanced sections of the signal, to achieve the detection of multi-frequency weak signals. The simulation and actual signal results show that the proposed method, a simple way to detect multi-frequency weak signals, can effectively improve the signal-to-noise ratio and has a good prospect of engineering application.Under colored noise background, a novel weak signal method based on stochastic resonance tuning by multi-scale noise is proposed. Firstly, noisy signal is processed by orthogonal wavelet transform to decompose the signal into multi-scale ingredients. According to the orthogonal wavelet transform coefficients characteristics of1/f distribution, multi-scale noise is constructed so as to make the frequency-band containing the driving frequency be enhanced through stochastic resonance system. Thus multi-frequency weak signal is detected. The method is effective to detect multi-frequency weak signal under colored noise background. Simulation and experiment signal analysis results show that the proposed method is simple for multi-frequency weak signal detection, and has good prospects for engineering applications.A novel method based on EMD after de-noising by adaptive re-scaling frequency-shifted band-pass stochastic resonance is proposed. Adaptive Re-scaling frequency-shifted band-pass stochastic resonance (RFBSR) is used to enhance different frequency scales adaptively and improve signal-to-noise ratio. Combined with stochastic resonance, the accuracy of EMD algorithm is improved. The simulation results show that the proposed method, can enhance the signal amplitude, reduce the false component and improve the accuracy of the EMD algorithm, effectively detect multi-frequency weak signal submerged by noise and extract feature of mechanical fault.