Research On Bearing Fault Diagnosis Method Based On Deep Learning And Transfer Learning

With the improvement of precision and tightness of industrial equipment,any small fault in the bearing can cause the equipment to stop running or even damage,so it is very important to diagnose its health status.With the development of intelligence,various data-driven intelligent bearing diagnosis methods can achieve better diagnosis results under the data training set with rich fault types,sufficient sample number and label information.However,in practical engineering,it is difficult to obtain a large amount of historical fault data of the equipment to be diagnosed in advance.Aiming at this problem,this paper proposes a bearing fault diagnosis method based on finite element simulation and transfer learning.First,a large number of simulated fault data sets with labels are obtained through finite element simulation,and then a deep domain adaptive transfer learning fault diagnosis model is constructed to realize the accurate diagnosis of faults.Diagnostic classification of actual bearing failures.The main work of this paper is as follows:(1)Construct the finite element model of the rolling bearing in the normal state and in the presence of outer ring,inner ring and rolling element faults,and use the ANSYS/LS-DYNA module to simulate the motion process of the rolling bearing,so as to obtain a large number of simulation data with fault information.,and the reliability of the simulation model is verified from three aspects: time domain,frequency domain and time-frequency domain.The results show that the established simulation model can accurately reflect the fault characteristics of the bearing.(2)Considering the one-dimensional characteristics of bearing vibration signals,the shortcomings of two-dimensional convolutional neural network(2D CNN)applied to bearing fault diagnosis and the advantages of one-dimensional convolutional neural network(1D CNN)in processing one-dimensional signals are analyzed..A bearing fault diagnosis method based on 1D CNN is proposed,a lightweight 1D CNN is designed,and the network loss function and training algorithm are determined.The diagnostic model is trained with simulated data and tested on the actual failure dataset.The test results show that the performance of the fault diagnosis model on the actual data set is greatly reduced due to the difference of the data set,indicating that the diagnosis model trained on the simulation data cannot be directly used for the diagnosis of actual faults,and the transfer learning technology needs to be introduced.(3)In order to solve the problem of low fault diagnosis accuracy caused by the difference between simulated data and actual data,a deep domain adaptation transfer based on generative adversarial network and multi-kernel maximum mean difference(MK-MMD)is further proposed.learning method.The domain difference between the simulated data and the actual data is narrowed by the deep domain adaptation method,so that the fault diagnosis model trained on the simulated data can be applied to the fault diagnosis of physical entities.Finally,the performance of the deep domain adaptive transfer learning method is verified through the experiments of variable working conditions and variable bearings,and the rationality of the rolling bearing fault diagnosis method based on finite element simulation and transfer learning is also proved.