Collaborative Research: Resource-Constrained Optimal Learning Framework for Post-Seismic Regional Building Damage Inference

This grant will support research that will contribute new knowledge related to resource allocation for post-seismic building damage assessment, promoting the progress of science and preserving the national welfare. During emergency relief operations after an earthquake or other disaster, it is critical to accurately assess the infrastructural damage across the impacted region.

Critical resources must be allocated quickly, before labor-intensive reconnaissance surveys are able to inspect each building. Thus, certain inspections should be prioritized in order to deliver an optimal damage assessment survey in time to benefit first-response relief efforts. This award supports fundamental research to develop a mathematical modeling framework to guide inspection teams through post-seismic reconnaissance missions.

This new approach will holistically identify buildings to be prioritized for inspection and design inspection schedules to efficiently visit these buildings with limited time and inspection crew members. Results from this research will expedite regional hazard damage assessment, which will improve disaster management and thereby help save human lives, ensure ethical resource allocation, and preserve the welfare of our society.

The project will prepare future civil engineers, mathematicians and statisticians with multi-disciplinary knowledge, and will broaden the participation of underrepresented groups in research which positively impact engineering education.

This research integrates concepts from statistical and optimal learning with models for routing and scheduling. These two aspects have been extensively studied separately, but never jointly. This knowledge gap poses a serious challenge to the guidance of inspection teams, which collect information in the field subject to resource constraints.

The research team will develop an integrated modeling framework which bridges optimal learning and combinatorial optimization to identify inspection routes and schedules that maximize the predictive power of machine learning models for post-seismic building damage assessment. The new methodology will be validated on real-world benchmarks, including data from the 2011 Chile and 2015 Nepal earthquakes, as well as a regional earthquake simulation testbed for the San Francisco Bay.

The results will improve crisis management, while also providing new insights into other domains (such as health and disease control) that face a tight tradeoff between data expense and information gain.

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.