Key points are not available for this paper at this time.
Digital Twin (DT) technology has been explored for developing condition monitoring (CM) solutions, especially under data availability constraints. DT develops virtual replicas of physical systems to accurately represent their behaviour and generate synthetic data. A critical challenge in DT implementation is maintaining coherence between virtual and physical systems. This first requires a high-fidelity virtual model that can accurately simulate various health states. However, developing such models is complex, and even after development, the inherent simplifications and modelling assumptions lead to discrepancies between the virtual model’s response and that of the physical system. Therefore, physical-to-virtual communication is necessary to tune model parameters, enabling the virtual model to better align with the actual behaviour of the physical system. The most popular method in this category involves modal analysis, which is impractical to perform in real-world scenarios. This study proposes a sequential Bayesian optimisation approach to iteratively tune the virtual model parameters, ensuring close alignment with the actual system dynamics. Its demonstration has been shown for a ball screw feed drive system. A high-fidelity lumped parameter model serves as the virtual counterpart, representing the nut’s motion in three degrees of freedom. Once calibrated, the virtual model is used to generate synthetic signals representing different system health states. These signals are then used to train machine learning (ML) models for real-time health estimation. The proposed DT framework through this physical-to-virtual communication shows motivating results in accurately predicting the ball screw health state, specifically in computationally limited scenarios.
Gupta et al. (Fri,) studied this question.