Multi-Resolution Digital Twins Under Uncertainty With Applications to Multiscale Systems

Many physical systems of engineering interest exhibit behavior governed by interactions across various spatial and temporal scales. Accurately capturing these multiscale phenomena in a digital twin requires models of sufficiently high-resolution, yet such models are often too computationally expensive for real-time state estimation, uncertainty quantification, and decision-making. This work addresses this challenge by introducing a multi-resolution digital twin framework that explicitly represents and manages uncertainty while dynamically alternating between low- and high-resolution model representations. The architecture leverages a computationally inexpensive single-scale model that is used for frequent data assimilation and rapid updates, whereas a detailed multiscale finite element model that is invoked selectively to refine predictions when elevated uncertainty or critical events warrant additional resolution. This selective activation enables the digital twin to preserve predictive accuracy without incurring the prohibitive cost of running the high-resolution model continuously. As a proof-of-concept, the proposed framework is demonstrated through a multiscale finite element example for static analysis of an aircraft structural component, where both model resolutions describe the same physical asset but encode different levels of mechanistic detail.