Research on Remaining Useful Life Prediction of Bearing with Digital Twin and TSMixer

To advance the prediction accuracy of remaining useful life (RUL) for rolling element bearings, this study proposes a novel hybrid prognostic framework that synergistically integrates digital twin (DT) technology with the Time-series Mixer (TSMixer) architecture. A high-fidelity virtual bearing digital prototype is constructed via multi-physics fusion modeling of DT, enabling real-time state perception and dynamic evolution simulation of the physical bearing system, which further generates abundant, high-reliability augmented operational data to compensate for the scarcity of measured full-life degradation samples. By fusing multi-domain features extracted from both historical monitoring datasets and DT-augmented simulation sequences, a comprehensive multi-source feature space covering the full degradation trajectory of bearings is established. Subsequently, the TSMixer model, which excels in parallel capturing of local temporal patterns and global long-range dependencies, is employed to learn the implicit time-varying degradation characteristics from the constructed time-series datasets, so as to achieve accurate RUL estimation of the tested bearings. Experimental validation on the publicly available benchmark dataset demonstrates that the proposed framework yields superior prognostic performance, with a prediction accuracy improvement of approximately 9.23% compared with several state-of-the-art baseline models. Furthermore, the proposed hybrid paradigm exhibits remarkable robustness and generalization capability under variable working conditions, which provides a promising technical solution for intelligent predictive maintenance and active fault prevention of high-end mechanical equipment.