Bearing Failure Based On Treelets And Deep Forest Network Research On Feature Fusion And Classification

Rolling bearings are essential basic components in rotating machinery,and have a wide range of applications in important national economic fields such as equipment manufacturing,energy and petrochemicals,national defense security,and aerospace.However,under extreme conditions such as high load,variable operating conditions and continuous operation,different degrees of wear on rolling bearings can cause performance degradation or even failures,thereby inducing the failure of the entire system,so the safe and stable operation of rolling bearings is related to the entire mechanical system.Reliable and safe operation.This paper takes rolling bearings as the research object,uses the collected vibration signal data to extract the characteristic parameters,and studies the bearing fault identification.The specific contents are as follows:(1)In view of the inadequacy of EMD algorithm and EEMD algorithm,choose to use CEEMDAN algorithm to process the signal to reduce the problem of IMF component redundancy in EEMD algorithm.In view of the running time problem still existing in the CEEMDAN algorithm,the algorithm is optimized based on the EMD’s stopping judgment criteria,which accelerates the speed of the algorithm and greatly reduces the running time of the algorithm.Finally,the simulation signal analysis of MATLAB and the decomposition speed of the actual bearing experimental data verification method greatly reduce the running time of the algorithm.(2)Aiming at the problem of dimensional disasters easily caused by fault characteristics,a method of dimensionality reduction of rolling bearing characteristics based on Treelets is proposed.The eigenvalue decomposition of traditional PCA and other dimensionality reduction methods has certain limitations and is only applicable in the case of non-Gaussian distribution,and the principal component obtained by the PCA method may not be the optimal solution.For this,the Treelets algorithm is used.In order to solve the problem of binary dimensionality reduction of Treelets algorithm,it is improved to obtain multidimensional dimensionality reduction suitable for specific situations.Finally,the vibration signal data of the rolling bearing is used for comparison.Compared with the traditional PCA dimensionality reduction method,the advantages of the Treelets algorithm over other algorithms are verified.(3)In view of the problem that classification and identification are prone to overfitting problems and the actual amount of fault data is small,this paper studies the deep forest algorithm and applies it to rolling bearing fault recognition.For the problem that traditional classification algorithms are prone to overfitting,the deep forest algorithm is introduced to avoid overfitting through the multiple decision tree structure of the random forest algorithm.And optimize the cascade forest process in the deep forest algorithm,reduce the feature vector dimension of the input multi-granular scanning process,and improve the running speed of the algorithm.Compared with the traditional SVM algorithm and other classification and recognition algorithms,the reliability of the proposed method is verified.Based on the fault identification,this paper carries out research and proposes corresponding improvement methods from the three aspects of feature extraction,feature dimensionality reduction and fault identification,and verifies by using bearing vibration signal data.The experimental results show that the proposed sub-method can be used to Rolling bearing fault identification has high accuracy.