Christos Stamopoulos has been selected to receive the Victor Peng ’81 Summer Research Project Award through Rensselaer Polytechnic Institute’s Future of Computing Institute (FOCI) for his project, “Foundation-Model Energy Landscapes: AI-Driven Bayesian Inverse Inference for Structural Dynamics and Digital Twins.”
The project addresses a growing challenge in aerospace structures, additive manufacturing, and unmanned aircraft systems: efficiently inferring latent damage, defects, and operating states from rich, multi-sensor time-series data.
Traditional Markov chain Monte Carlo (MCMC) methods can become computationally expensive when applied to high-dimensional and multimodal posterior distributions, particularly when high-fidelity finite element or computational fluid dynamics models are combined with stochastic, time-varying system dynamics. This computational burden can limit the speed and practicality of digital-twin applications.
The proposed research explores the use of Time-Series Foundation Models (TSFMs), including IBM’s Tiny Time Mixers (TTM-R2), to generate transferable, low-dimensional representations of multivariate sensor data. By using these learned representations to characterize the geometry of complex posterior distributions, the project aims to support more efficient Bayesian inverse inference and enable faster, more scalable digital-twin frameworks.
AI-Driven Foundation-Model Energy Landscapes for Structural Health Monitoring and Digital Twins