Research On The Fault Recognition Method Of Rolling Bearing Based On The Theory Of Optimal Margin Evolution

The operation of a mechanical system is closely related to the operating conditions of rolling bearings.Rolling bearings are an important supporting component in rotating machinery.Early failure mode recognition is the key to reliability analysis of mechanical systems.Among the existing model recognition methods,the mathematical theory of Support Vector Machine(SVM)is more complete,and its application in the field of fault diagnosis is also more mature.Therefore,this article conducts in-depth research on the mathematical theory of SVM,learns to learn from the idea of maximum margin of SVM,and introduces a 0-1 percentage loss function with weight optimization on this basis to improve the sensitivity of the traditional 0-1 loss function to unbalanced data.Shortcomings,and use the adaptive covariance matrix evolution algorithm(Covariance Matrix Adaptation Evolution Strategy,CMAES)to find the best weight matrix,proposed a modified 0-1 loss function based on the optimal margin evolution theory identification algorithm,and The algorithm is applied to the diagnosis of rolling bearing fault components.The research work completed in this article includes:In terms of feature extraction,in view of the poor anti-aliasing and translational properties of the traditional Discrete Wavelet Transform(DWT)when processing noisy vibration signals,the Dual Tree Complex Wavelet Packet Transform(Dual Tree Complex)is adopted.Wavelet Packet Transform,DTCWPT)to replace it,using noisy simulation signal to verify that DTCWPT has good anti-aliasing and translatability.Aiming at the imbalance sensitivity problem reflected by the traditional 0-1 loss function when processing imbalanced data,the form of statistical misclassification loss rate is used to modify the form of counting the number of misclassification instances,because the loss rate can only represent the overall average recognition of the model The accuracy rate cannot represent the recognition effect of the model on each category of groups.Therefore,a 0-1 loss function with weight optimization is proposed.Based on the loss rate,the ratio of the model to the recognition accuracy of various samples is used to express the weight coefficient.This balances the recognition accuracy of the recognition model for each target group to be recognized.Aiming at the general step size and invalid mutation problems that may occur in the process of evolution strategy(Evolution Strategy,ES)generating mutant offspring through normal distribution,the CMAES evolution optimization strategy is introduced,and the sampling method of Gaussian distribution and the iterative update of evolution path are introduced.The strategy controls the mutation degree and direction of its offspring.The CMAES evolution optimization strategy is used to optimize the weight matrix of the optimal margin evolution theory model,which not only accelerates the speed of model convergence,but also improves the optimization of parameters effect.DTCWPT is used in the data preprocessing scenario of bearing failure,and the characteristic coefficients obtained by DTCWPT decomposition are input into the optimal margin evolution theory model and the supervised learning model for training and identification.The effect verification based on the optimal margin evolution theory model was realized in two databases: In the Western Reserve University database,the original inner ring damage,outer ring damage,rolling element damage and health signal data were input into different models for fault diagnosis and identification,the recognition effects are compared from both horizontal and vertical aspects,and then the characteristic-processed single-diameter fault data and composite-diameter fault data are also compared by corresponding horizontal and vertical tests;at the University of Paderborn,Germany In the database,the original fault data is also input into different models to identify the fault,and the recognition effect is compared from both the horizontal and vertical aspects,and then the characteristic-processed fault data is compared with the corresponding horizontal and vertical tests;the results show that The optimal margin evolution theory model has the best fault recognition effect under the balanced and unbalanced data of different databases.