Rapid deployment of digital twin for life prediction of rolling bearings

Abstract

Finite element modeling (FEM) is widely recognized as a relatively accurate approach for constructing digital twin (DT) models for predicting remaining useful life (RUL). However, FEM suffers from long computation times, high operational complexity, and an inability to meet the real-time requirements of DT. This study proposes a k-nearest neighbor Kriging Radial basis function Digital Twin (KKR-DT) system. Initially, the full working condition results of the roller bearing were calculated using Ansys software. Subsequently, a reduced-order (OR) model was developed following the agent model approach. KNN was used to find neighboring values near the OR points, and Kriging was employed to interpolate at the OR points, obtaining an OR model with a single working condition. Finally, using RBFs all single-working condition OR models were transformed into full-working condition OR models, thereby establishing a five-dimensional DT model and DT user interface. The stress-life (S–N) degradation curve of the material was used to predict the roller bearing RUL. The proposed stress field diagram addressed the challenge of reverse validation in interpolation models. Ultimately integrated as the KKR-DT system. Compared the full working condition average accuracy of KKR-DT was 96.6938%, with maximum and minimum average accuracies of 99.9993% and 99.9978%, respectively. Real-time dynamic operation calculation time for a single instance was achieved within 0.35 s. Remote DT testing was conducted using actual spinning frame equipment, to demonstrate the accuracy and real-time DT capabilities of the system, a solution is provided for the practical application of digital twins in dynamic operation and prediction.