Fusing Data- and Model-driven Methods for RUL Prediction in Smart Manufacturing Systems

Remaining useful life (RUL) is of significance in prognostic and health management (PHM) of machinery equipment. The state-of-the-art RUL prediction methods, including data-driven, model-based, and hybrid approaches, suffer from constraints like incomplete/imprecise physical models, uncertainties in degradation processes, and measurement data noise. To address these constraints, a novel hybrid model-and data-driven RUL prediction approach is proposed in this paper, which leverage the strengths of data-based and model-driven approaches. The proposed method integrates an extended Kalman filter to estimate stochastic degradation model parameters and a multi-head attention transformer to extract information from sensor data. A regression token is introduced to seamlessly fuse the deep learning model and the stochastic filtering method. Numerical experiments conducted with real-world rolling bearing degradation datasets underscore the superiority of the proposed approach compared to competing methodologies.