High-fidelity hazard and infrastructure simulations (e.g., CFD/FSI, nonlinear finite element analysis,
probabilistic fragility modeling) are essential for realism but often too computationally expensive for
regional-scale risk analysis, uncertainty quantification, and time-sensitive decision support.
This research thrust develops physics-informed machine learning surrogates that preserve governing
laws and engineering constraints while enabling rapid prediction of loads, responses, damage states, and
reliability metrics across large infrastructure portfolios.

Core Research Focus

• Physics-informed and constraint-aware surrogate modeling for hazard–structure interaction
• Fast emulators of high-fidelity simulations (CFD/FSI, nonlinear FE, progressive damage models)
• Data-efficient learning under limited observations and heterogeneous datasets
• Uncertainty quantification (UQ) and reliability analysis using surrogate-enabled workflows
• Fragility surface and vulnerability function generation at scale

What Has Been Done

• Established machine learning foundations to accelerate engineering prediction tasks while maintaining accuracy
• Developed workflows that link high-fidelity simulation datasets to surrogate-based prediction of key response quantities
• Advanced optimization strategies that reduce training/search cost and improve surrogate deployment efficiency

What We Are Doing Now

• Building surrogate models to accelerate hazard-to-load pipelines (e.g., wind/wave fields and pressure distributions)
• Developing fast predictors for nonlinear structural response and progressive damage outcomes
• Embedding physical constraints (conservation, monotonicity, stability, failure bounds) into learning architectures
• Integrating surrogate-enabled UQ into fragility and reliability assessment workflows

Strategic Plan

1. Surrogate-ready simulation libraries: curate validated datasets from CFD/FSI and FE models to enable reusable training pipelines.
2. Physics-consistent learning: enforce governing laws and engineering constraints to improve generalization across geometries and hazards.
3. Uncertainty-aware prediction: provide calibrated confidence estimates for risk and reliability decision-making.
4. Regional-scale deployment: integrate surrogates into lifeline and community digital twin workflows for fast scenario evaluation.
5. Open, reusable toolchains: develop modular, documented software components for research reproducibility and adoption.

How This Connects

This thrust is DM2L’s computational acceleration and uncertainty engine. It enables scalable
deployment of high-fidelity mechanics and hazard models to lifeline systems (power and transportation),
interdependency analysis, and community resilience applications—without sacrificing physical credibility.

Figure (TBA)