Predictive Digital Twins at Scale for Space Systems
Abstract
A digital twin is an evolving virtual model that mirrors an individual physical asset throughout its lifecycle. This project will develop the mathematical and computational framework to create predictive digital twins for space systems. The mathematical approach uses the formalisms of probabilistic graphical models, which are used widely in the robotics community. Mathematically the physical-twin-digital-twin system is represented as a set of coupled dynamical systems, evolving over time and interacting via observed data and control inputs. Computationally, the physical-twin-digital-twin system is implemented as a dynamic Bayesian network built on an integrated combination of physics-based models and data-driven learning. Challenges of creating digital twins for space systems include that observational data is limited, indirect and sparse; the digital twin must be deployed in a resource-constrained environment, where computing power and communication are limited; and the systems may be operating in an adversarial environment. Thus, a critical aspect of the proposed approach is to embed rigorous uncertainty quantification in all elements of the digital twin representation and its use. The approach will be demonstrated on the testbed problem of tracking unknown space objects and detecting their intent to actively maneuver to intercept a target. The digital twin will issue predictions of the space object’s future trajectory and intentions. All predictions will include quantified uncertainties. Most existing approaches to monitoring satellites and predicting their intentions concentrate on the estimate of the state (orbital characteristics) and controls (translations maneuvers) from available observations. In contrast, the proposed predictive digital twin methodology offers the mathematical tools to incorporate the space object’s intentions as a reward in the Bayesian network.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502210419
Entities
People
- Karen Willcox
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Texas at Austin