Data-driven methods in natural hazards engineering

Current research projects

  • Hybrid and physics-based machine learning models for structural resiliency
  • Data-driven predictive models of structural components and materials

Past research projects

Data-driven approaches for improving site response prediction

Common site condition proxies in ground motion models cannot sufficiently represent site response. This project developed data-driven models to estimate site terms using horizontal-to-vertical spectral ratios and compared their advantages to conventional models.

Data-driven Surrogate modeling for performance-based assessment of mid-rise concrete frames at early design

The performance-based early design of buildings requires numerical models that can estimate performance using crude design parameters. This project developed and compared different machine learning techniques to estimate economic loss and embodied carbon of an inventory of mid-rise RC buildings using limited information.

INSSEPT: An open-source relational database of seismic performance estimation to aid with early design of buildings

Relational databases can serve as a flexible and powerful tool to develop next-generation of performance inventories. This study presents the metadata of a database structured around published results of 222 PBEE assessments of mid-rise buildings, as a curated resource for future machine learning applications, early design, and seismic regional assessments.