Rolling bearings remaining useful life estimation using digital twin and physics-informed methods with uncertainty quantification

This paper presents an innovative framework for monitoring the degradation process of rolling bearings, predicting their remaining service life and quantifying the associated uncertainties. By integrating digital twin technology with artificial intelligence, the framework significantly enhances the accuracy and reliability of remaining service life predictions for rolling bearings. First, an advanced dynamic model of the rolling bearing is developed, forming the foundation for the digital twin to accurately simulate the bearing's operating conditions. Next, a dynamic update method for the digital twin model is proposed, based on a backpropagation neural network, which ensuring real-time adaptation to changes in actual bearing conditions. Building on this, an innovative uncertainty-quantifying physics-informed neural network is introduced, which leverages a dropout technique. This network takes the output of the digital twin model as input to predict the remaining service life of the bearing while quantifying the uncertainty. The performance of the proposed framework is evaluated through experiments, demonstrating its ability to accurately reflect the bearing's operational state and predict its remaining service life with quantified uncertainty. The experimental results show that the mean absolute error and root mean square error of this framework are 0.071 and 0.084, respectively, which are both significantly better than the 0.100 and 0.118 observed for other methods, fully validating the superiority of this method.