A Software Library for Digital Twins Based on Probabilistic Graphical Models
Please login to view abstract download link
The Digital Twin (DT) framework constitutes a focal point in the automation of complex systems by integrating simulation, parameter estimation and decision-making within a unified loop. DTs have been successfully applied to structural health monitoring, autonomous vehicles, personalized medicine, and more. Probabilistic Graphical Models (PGMs) represent dependencies between random variables through graphs and provide an elegant and efficient framework for inference and uncertainty quantification. In [1], PGMs are proposed to abstract and formalize the bidirectional observation-action interaction in physical-digital coupled systems. This approach has shown promising results in structural health monitoring of civil structures [2] and aircraft mission planning [3]. However, actual implementations of DTs remain highly customized, preventing practitioners from easily extending, evaluating, or integrating them with recent methodological advances. To streamline the adoption of PGM-based DTs, we present a new open-source Python library for digital twins, built on top of established packages such as pgmpy, gymnasium, and stable-baselines3. Thanks to its modular design, the library allows users to easily extend built-in models and integrate custom implementations for physical modeling, parameter estimation, and decision-making. We demonstrate the capabilities of the library by implementing a DT for a structural health monitoring application.