Research On Experimental And Simulation Data-Driven Fault Severity Assessment Of Rolling Element Bearings

In modern machineries,rolling element bearings are one of the most extensively used basic components,which are the key to reliability and stability of equipment.Status monitoring and fault diagnosis to rolling element bearings can not only avoid the occurrence of serious accidents effectively,but also formulate maintenance strategies effectively.At present,fault diagnosis in rolling element bearings has entered the era of big data.Various intelligent diagnosis approaches have been verified in open source data-driven.While in practical engineering,the actual fault data is hard to acquire,resulting in the lack of corresponding fault data for fault labels,which is a "small sample" problem.In order to solve this problem,experimental and simulation data-driven fault diagnosis in rolling element bearing is proposed.That is,the data obtained from measured experimental signal and the simulation signal obtained from the dynamic model are used to assess the fault severity of rolling element bearings by the intelligent fault diagnosis method based on convolutional neural network.Aiming at deep-groove ball bearing,a fault model is established by studying the fault morphology of the outer race surface of rolling element bearing and its geometric relationship based on the Hertzian contact theory.A four DOF vibration model is established for the fault bearing model.By analyzing the contact force and contact damping force between the rolling element and the race surface,dynamic differential equations are obtained.The vibration response simulation signal with different fault size can be acquired by solving differential equations.The simulation signal is compared with experimental signal in time domain,frequency domain and time frequency domain,which is extremely similar with experimental signal.Based on time-frequency analysis and deep learning,SPWVD-CNN intelligent fault diagnosis approach is proposed.Firstly,obtain SPWVD time-frequency images of bearing signal and use them as the input of CNN.After setting the fault label,the time frequency images of each fault degree are identified by CNN.Thus,the fault severity assessment of rolling element bearing is completed.Simulation data are used to provide effective and diverse fault samples for fault diagnosis in rolling element bearings,which is an effective supplement to the fault data set.Based on the SPWVD-CNN,the cross-assessment of fault severity is carried out in the data sets where the training set is simulation data and the test set is experimental data,and the mixed-assessment of fault severity is carried out in the data sets where the experimental data and simulation data are completely and partly mixed in the training set.The result of cross-assessment and mixed-assessment verify the effectiveness of the experimental and simulation data-driven to solve the "small sample" problem in fault diagnosis and provide a brand new and effective solution to the fault diagnosis in practical engineering.