A Digital Twin of Life-Cycle Bearing Based on a Defect-Guided Conditional Diffusion Model for Predicting Remaining Life under Limited Degradation Data

Accurate prediction of the remaining useful life (RUL) for rolling bearings was essential to ensure the operational reliability and safety of complex industrial equipment. However, the lack of high-quality life-cycle data in industry impeded the development of high-precision RUL models. Digital twin (DT) technology offered the reconstruction of physical degradation life-cycle data by integrating model and data virtual-physical mapping, thus addressing the issue of data insufficiency. This study proposed an innovative DT method for the life cycle, which was based on a defect guided conditional diffusion model integrated with virtual-physical mapping. First, a two-degree-of-freedom dynamic model of the bearing was constructed to serve as the physical digital representation and served as prior knowledge guiding the diffusion model. Subsequently, a regression model was constructed to explore the mapping relationship of defect degradation between measured vibration signals and dynamic data, and pseudo-labels were generated for the measured signals. A defect-conditioned generation module was incorporated into the diffusion process, with pseudolabels informing the synthesis of full-life-cycle twin signals aligned with the defect evolution pattern. The proposed method enabled physically interpretable and controllable synthesis of vibration signals and preserved both statistical characteristics and degradation trends observed in real data. Qualitative comparisons in both the time and frequency domains confirmed that the generated signals retained critical dynamic features, such as impulsive components, amplitude modulation, and fault-related spectral structures. The proposed method demonstrated significant potential to enhance DT construction, augment limited data sets, and support intelligent condition monitoring of rotating machinery.