Research Of Multi-fault Rotating Machinery Diagnosis Method Based On Alpha Stable Mixture Distribution Model

As one of the most important type of mechanical equipment, rotatingmachinery plays an important role in human's life and production. Gears andbearings the main components of rotating machinery are the most vulnerable partsto failure, and there are quite a few cases that the faults are concurrent, in whichcase the single fault diagnosis method loses its effective. So the research ofmulti-fault diagnosis method is also an important part of the rotating machineryfault diagnosis. This thesis will be focused on the new parametric probabilitydistribution model, the mixture of alpha stable distribution model, for faultsignals and propose new fault diagnosis methods.The machinery multi-fault condition is discussed in this thesis by using themixture of alpha stable distribution model. Bayesian inference and Markov chainMonte Carlo are using for the estimation of model parameters and prove theaccuracy of the method based on the synthetic data.Two kinds of the alpha stable mixture model are proposed. First themulti-fault signals are modeled by alpha stable mixture model after preprocessing.Second the non-stationary characteristics of the multi-fault signals are extractedby wavelet coefficients clustering and Shannon entropy concept, and thenmodeled by alpha stable mixture model. Comparing with the Gaussian mixturemodel, the model established in this thesis is more superior.The model parameters are regarded as the feature vector and the input ofSOM network to cluster. Combining with the specific experimental equipment tenkinds of working condition signals including nine multi-fault signals and onenormal signal are collected and used to prove the accuracy and superiority of ourmethod.