Research On Rolling Bearing Fault Diagnosis Using Hybrid Entropy And Seagull Optimization Algorithm-based Support Vector Machine

Rolling bearing is one of the core components in rotating machinery and equipment,which plays a role in connection and fixing.Its operating state directly determines the long-term stability of the equipment.Therefore,accurate and timely detection and judgment of early failures of rolling bearings can effectively ensure the long-term stable operation of the equipment,reduce the probability of safety accidents,and improve the production efficiency of the enterprise.This study has developed a novel fault diagnosis method to address the difficulties of rolling bearing fault feature extraction and fault classification.The technique mainly includes the following three steps: firstly,the refined composite multiscale sample entropy(RCMSE)and refined composite multiscale fuzzy entropy(RCMFE)are used to comprehensively mine the fault characteristic information from bearing signals.Subsequently,the t-distributed stochastic neighbor embedding(t-SNE)is used for feature filtering,obtaining the low-dimensional and easily recognizable feature set.Finally,the extracted mixed-entropy+manifold learning feature set is input to the seagull optimization algorithm-based support vector machine(SOA-SVM)for fault classification.Two sets of bearing experimental data validate the effectiveness of the proposed method.The main tasks of this paper are as follows.(1)To comprehensively explore the fault information of bearing signals,a novel mixed-entropy feature extraction method is developed by combining RCMSE and RCMFE.The effectiveness of this technique is confirmed by two sets of bearing experimental data;in addition,it is compared with existing single/mixed entropy methods,such as multiscale sample entropy,multiscale fuzzy entropy,RCMSE,RCMFE,and MSE+MFE,the superiority of this method is verified.(2)To avoid the redundant interference in the presence of RCMSE+RCMFE mixedentropy,the t-SNE is utilized for effective secondary feature extraction as a means to obtain the sensitive low-dimensional result.The validity of t-SNE is verified using two sets of bearing experimental data and compared with isometric feature mapping and local linear embedding is performed.The results show that t-SNE has the best dimensionality reduction effect and can further extract effective information from the original high-dimensional data.(3)To achieve high-precision bearing fault diagnosis,the SOA-SVM classifier is proposed to classify the extracted fault feature set,while the seagull optimization algorithm is used to optimize the SVM parameters.The superiority of this classifier is verified by a set of the simulation experiment and two sets of the bearing experimental data.Moreover,the SOA-SVM classifier is compared with the existing classifiers,such as particle swarm optimization-based support vector machine,bat optimization-based support vector machine,and grey wolf optimization-based support vector machine.Through comparison,it is found that SOA-SVM can accurately and effectively identify the type of rolling bearing failure.(4)On the basis of the above methods,a novel rolling bearing fault diagnosis model is developed,based on mixed-entropy,t-SNE,and SOA-SVM.The validity and superiority of this model are verified using the bearing experimental data and comparison experiments.