Research on full-life rolling bearing virtual-real fusion defect prediction with digital twin

Abstract The digital twin model of full-life rolling bearings is of great significance for the prediction of their working condition. The existing digital twin model mainly simulates the two degrees of freedom of the outer ring of the bearing, ignoring the realism of the displacement of the inner ring and the external excitation, and does not consider the time series of the bearing vibration signals in the process of the interaction between the real and the imaginary. In this paper, a six-degree-of-freedom full-life rolling bearing digital twin model is proposed. This model establishes a six-degree-of-freedom dynamics model through the unit resonator, generates simulation data closer to the real working conditions, and combines the improved CycleGAN and DTW techniques to realize the mapping and time series regularization of the virtual simulation to the physical measured data. Then the CNNN-LSTM-Attention network model is trained using multi-feature inputs from the simulation data, and thus the complete digital twin model is established. Finally, by using the XJTU-SY bearing dataset, the digital twin model was verified to be effective in the simulation data and improved the local defect extension prediction accuracy.