CAREER: Learning-Enabled Multiphysics Computational Infrastructure for Failure Simulation in Geohazard-Structure Interaction

Natural hazards such as landslides, debris flows, and other rapid flow-like geohazards remain challenging to predict because they involve complex interactions among soil, water, engineered structures, and failure processes that evolve across multiple spatial and temporal scales. Current computational tools relying on empirical assumptions struggle to capture the multiscale and multiphysics nature of dynamic failure at the interface between geological materials and engineered systems. This CAREER project addresses these limitations by pioneering a new computational infrastructure that integrates physics-based models with artificial intelligence and machine learning (AI/ML). The project advances the fundamental understanding and predictive modeling of failure mechanisms in civil and geomaterials and delivers an open-source platform for hazard forecasting, adaptive monitoring, and resilient design, contributing to national priorities in disaster resilience and infrastructure sustainability. The educational effort includes incorporating data-driven methods into civil and geo-engineering undergraduate curricula, developing new course materials and hands-on computational modules supporting undergraduate research, K–12 STEM programs, and community workshops, thereby equipping students and practitioners from academia, industry, and public agencies with advanced skills in computational modeling and AI for infrastructure and hazard applications. The project constructs a Learning-Enabled Multiphysics Computational Infrastructure (LEMCI) that integrates high-fidelity numerical solvers and multiscale operator learning within a unified differentiable programming framework. Through three cyberinfrastructure innovations, LEMCI enables scalable, learning-in-the-loop simulation capabilities for uncovering the mechanics of material coupling and failure propagation from geohazards to built infrastructure under extreme dynamic loading, while supporting inverse modeling, data assimilation, and resilient infrastructure design. First, it develops a differentiable, Graphical Processing Unit (GPU)-accelerated meshfree simulation framework, based on a hybrid Lagrangian–Eulerian formulation, for modeling nonlinear deformation, damage evolution, and coupled soil–fluid–structure interactions. Second, it creates multiscale neural operator models to represent path-dependent inelastic material behavior and thermo-hydro-mechanical coupling, linking macroscopic failure responses to underlying microstructural characteristics in heterogeneous geological and infrastructure materials. Third, it establishes an end-to-end, ML-integrated physics-constrained optimization framework for simulation-driven inverse modeling, enabling the discovery of hidden parameters and damage mechanisms from multisource observational data. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2543706 | Program: 01003031DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT,01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Qizhi He | Institution: University of Minnesota-Twin Cities, MINNEAPOLIS, MN | Award Amount: $358,268 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2543706 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2543706.html