Intelligent Fault Diagnosis Of Gearbox Based On Particle Filter And Wavelet Neural Network

Gearbox is used to change the speed and transmit the power as an important component of mechanical equipments. The gears and bearings of the gearbox have the fatigue wear under the long-term operation with heavy load and the gearbox is easy to have failure among mechanical system. The fault diagnosis and condition monitoring of gearbox is of vital importance for machanical system to run properly. Fault diagnosis of gearbox includes the acquisition of the feature extraction and condition classification. Fault feature extraction is to obtain the fault feature of vibration signals by the signal processing technology. Fault condition classification is to recognize the type of failure components of gearbox system.Wavelet and wavelet packet transform are widely applied in signal denoising. The filter method is difficult to make full use of signal characteristics when the nonlinear vibration singal is corrupted by the large amount of non-gauss noises. It is easy for the filter to delete the useful feature signal by setting threshold through wavelet transform method. Particle filter, a signal processing technology, also perform well in filter denoising especially when it is used to process the non-gaussian and nonlinear systems. However, particle filter algorithm requires a large number of sample particles to approximately estimates the state of system while excessive sample particles bring about accumulative effects of forecasting error and result in the divergent systems. Any single adoption of signal processing technology mentioned above leads to the indistinct filtering effects and the missed useful signals.This paper applied the particle filter technique to preprocess nonlinear and low SNR(Signal Noise Ratio) vibration signal of gearbox. The experiment obtained a small amount of noise signal and filter signal. Then a filtered signal was eventually turned into an ideal denoised signal after a wavelet transform and secondary denoising by setting soft threshold. The SNR of the final filter signal is used to show that the gear vibration signal processed by combining two signal processing technology not only retained a lot of useful signal but also filtered out a great deal of noise, which decreased denoising errors and increased the success rates of neural networks fault classification in post-processing.Wavelet neural network is a signal analysis method based on wavelet transform theory. It’s time-frequency localization properties and multi-resolution feature has remarkable advantages for feature extraction of gear vibration signal information. The experiment extracted the energy eigenvalues of filtering signal through wavelet transform, the energy eigenvalues can well reflect the characteristics of gear vibration signals, The experiment extracted the energy eigenvalues of filtering signal through wavelet transform, this energy eigenvalues can well reflect the characteristics of gear vibration signals. Then the wavelet neural network model is established and the classification of fault feature is completed. The results after classifier training indicated the method of combining particle filter with wavelet denoising and extract the energy eigenvalues through wavelet transform, which completed the fault classification of neural networks and achieved the research purpose.