Towards Bayesian Digital Twins: Enabling Uncertainty Propagation between Modules and across Assets

Ohio State University, The, Columbus OH

Investigators

Abstract

This project develops a statistical framework for understanding Digital Twins, a next-generation technology that integrates modeling, data collection, prediction, and decision-making in a two-way, real-time cycle for a physical, biological, or engineering system. This project explores a Digital Twin for a robotic surgical plate bending system that is integrated into a virtual surgical planning process to assist surgeons in planning and executing cranio-facial reconstructive surgery. This Digital Twin could maintain synchronized models of a human mandible after traumatic injury and a surgical fixation plate designed to stabilize it. A robotic plate bender - part of the system’s physical asset - would iteratively bend the plate, monitor changes in its shape, update the corresponding digital model, and plan the next bending operation, all in real time. This model is then coupled with the patient’s jaw geometry to ensure that the applied forces will not induce long-term weakening or failure. Each step of this complex process involves uncertainty, including errors in the plate or mandible models, noisy observations, and unmodeled variation in the robot’s performance. These uncertainties propagate throughout the Digital Twin system in nonlinear, interacting ways, posing potential risks to patient outcomes. Current Digital Twin frameworks often ignore or oversimplify these sources of uncertainty. This project addresses the critical need for robust, real-time methods to quantify, propagate, and manage uncertainty in Digital Twin systems, thereby promoting safer and more reliable medical interventions and supporting NSF’s mission to improve health and advance technological innovation. The investigators will develop a Bayesian data assimilation framework for Digital Twins, enabling structured uncertainty quantification (UQ), verification, and validation across modular components and real-time physical interactions. This project addresses key Digital Twin UQ challenges by: (1) designing modular representations of uncertainty across system components; (2) developing a Bayesian approach to dynamically update Digital Twin models with real-time sensor data; and (3) formalizing two-way digital–physical interactions to support predictive decision-making. The framework will be demonstrated using a Virtual Surgical Planning system that includes robotic plate bending for autonomous point-of-care manufacturing of patient-specific implants. This testbed provides a rigorous platform for evaluating how Bayesian UQ methodologies improve real-time adaptation, predictive reliability, and decision support in safety-critical applications. The developed methods will also be portable to Digital Twin applications in autonomous manufacturing, such as those pursued in the NSF HAMMER-ERC Engineering Research Center (Hybrid Autonomous Manufacturing Moving from Evolution to Revolution). This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.