Research On Rolling Bearing Fault Feature Selection And Diagnosis Method Based On Support Vector Machine

Rolling bearings are widely used in mechanical equipment because of their advantages,such as low friction resistance,good interchangeability,and low price.Their good operation has a significant impact on the operation safety of mechanical equipment.This paper takes rolling bearings as the research object and researches the feature selection and fault diagnosis of rolling bearings.The main research contents of this paper are as follows:(1)In response to the problem of nonlinear vibration characteristics and more normal data than the fault data of rolling bearings,this paper adopts a support vector machine as the primary model for fault classification.The optimal kernel function is selected through comparative experimental analysis,and an SVM classification model based on the radial basis function is constructed.By testing fault data of 4 and 10 categories,it was verified that the SVM classification model based on radial basis function has the advantages of high classification accuracy and short training time compared to classification models such as BP neural network and decision tree.(2)To solve the uncertainty problem caused by the traditional fault feature selection based on artificial prior knowledge,this paper proposes a double-wrapped feature selection method based on a mixed domain.On the one hand,in terms of feature selection methods,the wrapped feature selection method is adopted,with the performance of support vector machines as the evaluation criteria for feature selection;On the other hand,in the method of feature indicator selection,first,construct the initial time-domain feature set,perform the first feature selection,and select the optimal time-domain feature subset.Secondly,construct an initial frequency domain feature set and determine similar regions based on the diagnostic classification results of the optimal time domain feature set and the optimal time domain feature subset.Conduct a second feature selection for similar regions and select the optimal frequency domain feature subset.Then,determine the frequency domain features of other regions based on the frequency domain features of similar regions.Finally,the selected time-domain features and frequency-domain features are fused.The effectiveness and accuracy of this method were verified through simulation analysis and comparison.(3)Aiming at the problem that the rolling bearing fault diagnosis classification model will be affected by the change of model parameters,an improved hybrid snake swarm optimization algorithm is proposed in this paper.Firstly,the genetic algorithm’s cross operation,mutation operation,and evolutionary reversal operation are introduced into the standard snake swarm optimization algorithm to improve the algorithm’s ability to jump out of the local optimal quickly and accelerate the convergence speed.Secondly,the classification accuracy of the classification model is taken as the objective function of the improved optimization algorithm to select the optimal penalty parameters and kernel parameters.Finally,through simulation analysis and comparison,it is verified that the classification accuracy and model running time of the proposed method are relatively optimal compared with other swarm intelligent optimization algorithms.