An imbalance multi-faults data transfer learning diagnosis method based on finite element simulation optimization model of rolling bearing

Abstract

In practical engineering applications, rotating machinery is often in a healthy state, with more healthy samples and fewer multi-faults samples, which leads to be misdiagnosed. A finite element simulation model (FEM) is proposed to expand imbalanced multi-faults sample data. However, there is a significant difference in vibration response between the FEM simulation data and the measured data collected on the rotating machinery. Therefore, a method based on vibration response eigenvalue model correction has been adopted, aiming to narrow the gap between the simulation data and the measured data. In order to solve the problem of low accuracy of sample imbalance fault diagnosis under different working conditions, a transfer learning (TL) fault diagnosis method based on finite element simulation optimization model of rolling bearings to supplement imbalanced multi-fault data is proposed. Firstly, the FEM of healthy bearings is established, and the optimal radial clearance of bearings is obtained by modifying the vibration response eigenvalue model. Then, corresponding simulated fault vibration responses were obtained by setting composite faults on the optimal model. Finally, supplementing the unbalanced fault data with simulation data to create a mixed balanced dataset as samples, the TL model is trained to accomplish intelligent fault diagnosis for multi-faults. The experimental results show that using mixed data as training samples to train neural network models can improve the accuracy of transfer learning fault diagnosis by over 10%. Expanding fault samples by modifying the bearing clearance of the FEM is an effective method to address sample imbalance.