Research On Fault Diagnosis Of Rolling Bearing Based On EMD And Entropy Characteristics

With the continuous innovation of the industrialization process,industrial machinery systems are more complex and precise than any previous stage,and are developing in the direction of more efficient and intelligent production.As the basic component of most mechanical equipment in the manufacturing industry,rolling bearings are indispensable in industrial process assembly due to their precision and reliability design.As a part of the whole machine equipment,the health state of rolling bearings is directly or indirectly related to the performance of the whole machine equipment.In severe cases,it may even cause production and economic losses.Therefore,complete fault monitoring and diagnosis technology is the premise to ensure the safe operation of equipment and to ensure the efficient operation of industrial production.Since the measured vibration data of the bearings are mostly non-linear and non-stationary,and the phenomenon of external interference is more obvious.This paper applies the signal processing method to the vibration signals for stabilization analysis,the characteristic information of the signal is evaluated by the property of entropy,and the particle swarm optimization probabilistic neural network(PSO-PNN)classification model is constructed to complete the state recognition of the bearing,so as to improve the classification accuracy of the diagnosis stage.In addition,considering the influence of more dimensionality of feature data,the feature dimension reduction method is used to achieve dimension reduction.Based on the above research description,this paper proposes two types of discrimination models for rolling bearing faults.The main work content is summarized as follows:(1)Aiming at the nonlinear modulation characteristics of bearing vibration data,a fault diagnosis model based on ensemble empirical mode decomposition(EEMD)entropy features combined with manifold learning is established.The vibration data are decomposed and processed by EEMD,and the entropy characteristics of the corresponding four processing states are calculated according to the entropy theory for the obtained effective modal components,so as to form the fault characteristic data for discrimination.Then,the t-distributed stochastic neighbor embedding(t-SNE)is used for data dimension reduction and visualization,and the PSO-PNN classification model is used to identify the fault state.(2)Aiming at the lack of local feature analysis of fuzzy entropy,a fault diagnosis model based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)combined with fuzzy measure entropy is established.The vibration signals are decomposed and processed by CEEMDAN to obtain multiple modal components.The fuzzy measure entropy is used to calculate the corresponding entropy characteristics of the effective modal components from local and global perspectives,and the characteristic data for discrimination is constructed.Then,the PSO-PNN model is used as the diagnostic classification part to identify the fault state.Finally,the fault vibration data of Western Reserve University and the vibration data collected by the research group on the bearing life test platform are used to carry out simulation experiments and tests for the two proposed diagnosis models.In addition,several control methods are used to perform comparative analysis,which proves the validity of the proposed diagnostic model.