Research On Rolling Bearing Fault Recognition Algorithm Based On The Maximum Expectation And Potential Energy Function

Bearing fault recognition has caused the people a great deal of attention.According to statistics, more than70%of the failures are displayed in the form ofvibration. Therefore, this paper chooses vibration signal for fault feature extraction.Because it is a very complicated nonlinear relation between feature vector and itsdiagnostic model, we choose characteristics of vibration signal from time domainand frequency domain as feature parameters to collectively reflect the features offault, and select the effective nonlinear classification algorithm for faultidentification.Two important links in fault diagnosis is fault feature selection and classifierselection. In this paper, wavelet analysis is conducted for the fault signal，and timedomain and frequency domain statistical features are calculated through denoisedsignal， from which original fault feature vector is obtained. However, whenundertaking fault recognition, different characteristics can identify fault fromdifferent aspects, but they have different sensitivity to identify faults. Somecharacteristics is sensitive an closely related to fault and other characteristics are not.So before inputting feature set to classifier, feature selection is a very importantproblem. This paper proposes distance evaluation factor to use for reducingdimension of fault data based on ideas of distance ratio, Basic idea of distanceevaluation factor evaluation criteria is: distances distance among classes and in classare calculated from one characteristic parameter separately, and ratio of twodistances is regarded as evaluation factors. Then characteristic parameters will beordered from big to small, and fusion of some parameters is regarded as a featurevector, which realizes objective choice of fault characteristic parameters.By referring to a large number of literatures, it is found that Gaussian MixtureModel can approximate density distribution of any shape smoothly, which is widelyused in pattern recognition in recent years and obtains better effect. Therefore thispaper selects Gaussian Mixture Model for fault identification based on the above ideas, and combine it with maximum expected algorithm(EM algorithm), is proposedEM algorithm based on finite Gaussian Mixture Model.After mastering the significance of potential energy function classificationalgorithm and the limitation of the multiple classification problems, this paperintroduces principle of binary tree to classification algorithm and combine it with thepotential energy function, and proposes improved potential energy functionclassification algorithm. Basic idea of the algorithm is: diagnosis types is dividedinto fault and not fault, and fault type is devised specifically again. In addition, faulttype recognition can be adjusted according to different object of fault diagnosis,calculate the probability of each fault type according to existing data of the diagnosisobject, and order from big to small, fault type of having the bigger probability can bediagnosed firstly, which can improve the efficiency of fault diagnosis effectively.Simulation experiments are undertaken in the MATLAB platform, theexperimental results show that the proposed distance evaluation factor dimensionmethod is effective and feasible; The EM algorithm based on gaussian mixturemodel can improve the ability of data classification; Choosing potential energyfunction classification of fault diagnosis provide new ideas for rolling bearing faultdiagnosis; Introducing theory of binary tree classification algorithm to the potentialenergy function solve the multiple classification problems.