The Research Of Gearbox Fault Diagnosis And State Classification Based On Improved Particle Filter

Gearbox is commonly used in rotating machinery as a transmission, it often occursfailure because of its long-term load operation and suffers from the influence of fatiguewear frequently. Therefore, the fault diagnosis and status classification of gearbox has thevital significance for the normal operation of the equipment. However, the vibration signalsof the gearbox covered by all kinds of noise cannot be obtained directly in the process ofactual operation, whose real signal need to be extracted through the signal processingtechnology. So, de-noising pretreatment process of vibration signals is critical before faultdiagnosing and classifying of the gearbox.Particle filter, as a kind of signal processing method has a good effect in terms ofsignal filtering noise reduction, its outstanding advantages are shown especially inaddressing the problem of nonlinear non-gaussian system. The particle filter algorithm is akind of filtering method to achieve the optimal bayesian estimation based on monte carlosimulation. Its basic principle is to make the sample particle simulate the system model andspread forward in chronological order, through which to get the corresponding system statesamples of each moment and obtaining the approximate posterior probability densityfunction of the true state system. This algorithm can be applied to the nonlinear systemrepresented by any state space model.Due to the traditional particle filter algorithm exist the problems of particleimpoverishment as well as the prediction error accumulation effect lead to system becomedivergent, which has a certain influence on handling the signal containing noise, the radialbasis function network(RBFN) algorithm will be introduced into the particle filteringmethod in this article for training and optimizing the process of particle filter in thesampling. This new method updates the state of each particle and improves the precision of the prior probability density distribution. It not only removes the estimation error causedby the process noise but also to avoid the lack of particles, and making the particle filterperformance be improved.For the fault diagnosis of gearbox the probability model is established aim at thenonlinear characteristics of vibration signals in this article, with which to optimize therelated parameters in the particle filter algorithm.The optimized parameters will be appliedin the improved particle filter algorithm. Using this new algorithm to preprocess thegearbox vibration signal acquired from the experiment and extracting the real signal, whichis preparing for the high precision time domain analysis. Then, extracting the featureparameters of the signal by using time domain analysis after pretreatment. And putting thefeature parameters into the wavelet neural network(WNN) for fault state classification. Thetraining result shows the effectiveness of the improved algorithm, which is achieving thegoal of this research.