Project Grant 2552621

The National Science Foundation Division of Civil, Mechanical, and Manufacturing Innovation awarded the Regents of the University of Michigan $304,793 on August 1, 2026, for a Project Grant under the Engineering program (CFDA 47.041) focused on data-driven agentic artificial intelligence for trustworthy design of autonomous engineered systems.

The award funds development of computational and data-enabled approaches to multidisciplinary design optimization. The project addresses the mathematical and computational challenges posed by systems involving interactions across multiple physics domains—such as floating offshore platforms, aircraft, additive manufacturing, and marine robots—where design tradeoffs across disciplines create complexity. The research will develop a novel level set-based approach using probabilistic kernel-based methods from machine learning to overcome the nested fixed-point iteration bottleneck central to multidisciplinary design. The project will leverage Bayesian decision theory to adaptively improve probabilistic models through information-theoretic acquisition policies, reducing computational costs. A large language model-based agent will translate structured engineering intent into verified coupled simulation workflows, addressing the challenge of coupling disparate computer codes available as oracles without exposing derivative information.

Work is performed at Ann Arbor, Michigan, with a period of performance from August 1, 2026, through July 31, 2029. The University of Michigan serves as the prime recipient through its Office of Research and Sponsored Projects Division.