Research

This research leverages deep learning techniques and model order reduction in addressing the heavy computational demand of nonlinear response analysis of large-scale engineered structures. Projection through proper orthogonal decomposition basis is considered to reduce the dimensionality of the system. Sequence-to-sequence mapping technique: long-short-term memory is leveraged to capture the dynamics of reduced time series. Through transfer learning, this scheme is first integrated into a stratified sampling strategy to allow rapid and accurate estimation of small failure probabilities. 

Shakedown analysis has been incorporated in the Prestandard of PBWD because of its capability to highly efficiently assess the collapse safety of inelastic structures under stochastic wind load. This work develops an alternate algorithm: an adaptive fast nonlinear analysis (AFNA) scheme, which is capable of analyzing inelastic structures at shakedown and beyond, and providing a full range of response histories. To further facilitate industrial adoption, user-friendly software was developed and published recently. 

As wind-induced damage becomes increasingly prevalent in coastal areas, risk forecasting is essential for emergency preparedness and rescue activity. To this end, we developed a framework to predict the risk of hurricane-induced damage to buildings. This work integrated metamodeling, probabilistic performance-based damage assessment, hurricane forecasting, and wind field modeling. The proposed framework allows multiple-day risk forecasting for imminent hurricanes with negligible computational effort. This work can facilitate informed decisions for governments and emergency management agencies, without the need for expensive computational resources.

In this research, isolation techniques were introduced to convert the massive scuttles to tuned mass dampers (TMD). Efficient optimization design frameworks by embedding model order reduction within metaheuristic algorithms were developed to solve this high-dimensional design problem of multiple TMDs, with a full range of uncertainty included. To further facilitate current engineering use, a semi-empirical design framework was developed and validated through a shaking table test.