Feasibility of Physics Constrained Digital Twins for Prescriptive Space Domain Resiliency

Our nation’s space capabilities anchor infrastructure in national defense, commerce, agriculture, and disaster relief efforts. Space domain awareness (SDA) integrates space surveillance, environmental monitoring, status of U.S. and cooperative satellite systems and an understanding of U.S. and multinational space readiness. Access to space is no longer limited to the most technologically advanced entities: a modern-era truth that has added some layers of complexity to joint space operations, especially with the operational domain expanding into the cislunar regime. For U.S. to maintain its vanguard status in this burgeoning space and safely sustain the cislunar domain as a gateway to the moon and beyond, we must engage in translational efforts that will formulate and solve grand challenges in asset sustainment and mission optimization through data analysis, uncertainty quantification, threat prognostics and mitigation. Our SBIR Phase I project is focused on discovery, leading to a technology feasibility study for optimized space asset sustainment. PointPro’s technology is a suite of digital twin solutions that includes i.) deep learning tools to translate data and physics constraints into dynamic models, and ii.) closed-loop, adaptive predictive uncertainty quantification tools that help forecast uncertainty through complex dynamic systems with a prescribed level of accuracy (ensuring trust). Prescriptive analytics demands timely, accurate forecasts of a system’s path to failure. PointPro’s uncertainty quantification tools integrate with physics-based, data-driven and hybrid evolutionary models to deliver forecasts within user defined accuracy. Our software leverages the time-tested robustness and versatility of Monte Carlo simulations in a closed-loop architecture, making it possible to deliver guaranteed accuracy in the forecast of system specific quantities of interest. When prediction errors are detected to exceed user-defined bounds, a sequence of optimization problems are solved, that in effect answer a long-standing question in predictive simulation: to meet a required level of accuracy, how big should the simulation size be? Consequently, a decision-quality, certifiably actionable forecast is generated in a single-shot, in minimal time. The platform integrates equally with physics-based evolutionary models and with data-driven (ML) models; and offers the unique ability to improve such models. In benchmark studies conducted so far, up to 3x productivity gain has been demonstrated. Moreover, gains improve with increasing system complexity. This project will seek feedback from subject matter experts in the Space Force with special emphasis on two key areas for space domain resiliency: i.) trustworthy conjunction assessment, and ii.) satellite sub-system level dynamics learning and failure prediction. Feedback related to specific users, applications, and necessary additions to the software will serve as a workplan for further work.