Residual Life Prediction of Rolling Bearings Driven by Digital Twins

To enhance the maintenance efficiency and operational stability of rolling bearings, this work establishes a methodology for bearing life prediction, employing digital twin systems to evaluate the remaining useful life of rolling bearings. A comprehensive digital twin-integrated model for the entire lifecycle of rolling bearings is constructed using the Modelica language. This model generates sufficient and reliable lifecycle twin data for the bearings. Due to the symmetrical physical structure of the bearings, the generated twin data also have symmetry. Based on this characteristic of bearings, a remaining useful life (RUL) prediction algorithm is developed using a recurrent neural network (RNN), specifically an improved gated recurrent unit (GRU) model. An optimization algorithm is employed to adjust the hyperparameters and determine the initial fault point of the bearing. A multi-feature dataset is constructed, effectively enhancing the precision and reliability of lifespan estimation. Based on existing measured data of the bearing’s entire lifecycle, the rolling bearing’s digital twin-integrated model parameters are updated. Through the parameter degradation component of the twin, the lifecycle twin data of the rolling bearing are generated. By combining twin data with actual measurement data, this method addresses the limitations of traditional approaches in situations where complete lifecycle data of bearings are scarce, providing reliable technical support for the intelligent maintenance and optimization of rolling bearings.