Digital twin information adaptive model and measured data fusion based bearing fault diagnosis

Abstract

Bearings are widely utilized in industrial applications. Their condition critically influences the stability, reliability, and accuracy of the equipment. Traditional fault diagnosis approaches typically depend on comprehensive fault data for accurate identification. However, challenges such as inadequate fault data and elevated equipment damage costs often limit the effectiveness of data-driven fault diagnosis methods in many industrial contexts. The digital twin information model can generate a virtual bearing entity, enabling the simulation of its operation and the generation of fault data under varying conditions. In this paper, a digital twin information model and measured data fusion based bearing fault diagnosis method is proposed. Initially, a virtual twin information model of the bearing is constructed using the finite element method. This model allows for the injection of fault depth and the generation of corresponding fault feature data. Subsequently, an optimization process is conducted to enhance domain adversarial networks and improve data feature mapping between real and virtual domains. Finally, a transformer feature extractor is utilized for effective fusion of multi-source features, facilitating fault diagnosis modeling. The case study results demonstrate that the enhanced features derived from fused data significantly improves diagnostic performance. The proposed method represents a substantial advancement in bearing health status evaluation and fault prediction.