Fault Diagnosis Technology Of Gearbox Based On Particle Filter Method

Gearbox is widely used in various fields of mechanical transmission, therefore gearboxfault condition monitoring and fault diagnosis is of great practical significance. Gearbox stateis reflected by measuring the gearbox vibration acceleration signal, however the actualgearbox vibration signals are often non-stationary, and the fault information is oftensubmerged in the strong background noise, so it needs to carry out noise reduction when thegearbox’s fault diagnosis.The particle filter with its nonlinear, non-gaussian signal processing ability, has beenwidely applied in the signal noise reduction processing of gearbox faultdiagnosis.Unscented particle filter because of its importance density function produced by theUKF set larger overlapping part with the true state of the probability density function,so itestimates higher accuracy and more accurate.The paper proposes a smoothing method of the acquisition vibration signal, which notonly can make the such time series be accord with the requirement of modeling smooth, butalso can achieve the goal of signal noise reduction at the beginning.First, use EMD todisintegrate the vibration signal, then use the wavelet threshold de-noising to preprocess thehigh frequency IMF component, then choose the high frequency IMF after wavelet thresholdde-noising and the part of low frequency IMF to reconstruction.In the paper the AR model is established, and on the basis of the statistical characteristicsof time series autocorrelation function and partial correlation function the AR model isverified. The criterions of FPE, AIC and BIC are used to identify AR model order number,at the same time using three methods can make it more reasonable to determine the ordernumber of the AR mode. The least squares method is used to estimate the parameters of the AR model and determine the corresponding parameters of the AR model.Finally we change AR model established into the internal space system model, anddetermine the corresponding parameters of the model, and use the UPF noise reduction toprocess the experimental data.Combined with the above theoretical analysis, using the unscented particle filtertechnology to reduce noise of gearbox vibration acceleration signal, to contrast of thetime-domain curve and signal characteristic parameter before and after filtering, you can findthat the signal after filtering effect is ideal.In this paper extract gearbox signal characteristicvalue before and after filtering respectively, then input the characteristic value into thesupport vector machine (SVM) respectively for gearbox fault diagnosis, the results shows thatthe accuracy of the test sample after unscented particle filter is higher and the UPF has goodfiltering effect.