Simulation-Driven Diagnostic Method via a Multibody Dynamics Model for Rotating Machinery

Paper in proceeding, 2026

Fault diagnosis is a pivotal aspect of condition monitoring in rotating machinery. Although fault diagnosis methods that use deep-learning approaches have achieved significant success, they require massive, labelled datasets from operating machines, which is often impractical and expensive for industrial applications. To overcome this challenge, simulation-driven fault diagnosis provides a scalable alternative by generating simulated data for training diagnostic models. This study presents a simulation-driven diagnostic framework for a physical test rig, where fault characteristics are learned via a multibody dynamics (MBD) model capable of incorporating more complex fault conditions in rotating machinery. The MBD model is first validated against experimental measurements from the physical test rig, demonstrating its capability to accurately replicate realistic fault signatures. The simulated dataset generated by the MBD model is then used to train a diagnostic model, which is further adapted to the physical test rig through transfer-learning methods. By capturing the realistic operating defect responses, this study demonstrates the potential of simulation-driven fault diagnosis and provides a promising foundation for future industrial integration.