Performance-based Wind Engineering through Stochastic Simulation and High-fidelity Computational Modeling

Abstract

With the burgeoning growth of tall building construction around the globe and heightened public expectations for safe and sustainable urban habitats in the face of extreme natural hazards, solutions for performance-oriented building systems are in great need. Conventional design for wind effects requires the satisfaction of a series of component-level elastic acceptance criteria under strength-level wind demands stipulated in building codes and standards. By ensuring building response is at or below the first significant yield point at a predefined load intensity, satisfactory performance is deemed to be achieved for significantly higher load intensities for which severe damage or even catastrophic collapse may occur. That is, not only are structural systems designed with no knowledge of their wind-induced inelastic behavior, but also how satisfactory the post-yield performance is (e.g., the margin of safety or reliability against structural collapse) remains unknown. The evolving paradigm of performance-based wind engineering (PBWE) seeks to achieve better structural performance (through reliable prediction and control of nonlinear behavior) and improved economy while treating uncertainty through reliability. The principal objective of this dissertation is to develop computational tools within reliability frameworks to efficiently assess the nonlinear behavior of uncertain dynamic structures subject to wind loads. Practical and efficient reliability approaches based on stratified sampling were developed that not only tackle optimal sample allocation for satisfying user-defined coefficient of variation (COV) targets in high-dimensional settings, but also extend applicability to a wider range of natural hazard risk assessment problems involving explicit hazard modeling (e.g., stochastic ground motion modeling or hurricane hazard modeling). This generalization is enabled by incorporating Markov Chain Monte Carlo methods for which COV expressions are rigorously derived of the failure probability estimators. Subsequently, a reliability-based framework was proposed for the characterization of the nonlinear behavior by integrating a high-fidelity fiber-based structural modeling approach capable of capturing progressive yielding, buckling, low-cycle fatigue (LCF), and variable damping, which are crucial for collapse simulation, with an appropriate stochastic wind load model. Through illustration on a 45-story archetype steel building, discussions on the types of observed collapse mechanisms, the difference in along-wind and across-wind response, and finally, reliabilities against first yield, component failure, and system collapse were presented. Presently, PBWE lacks a validated model for the stochastic simulation of non-stationary (NS) wind pressures that can be calibrated to stationary wind tunnel data. Such a model is essential for studying load-path-dependent phenomena of inelasticity and LCF in wind-induced responses subject to NS hurricane winds. This led to the experimental validation of the NS wind pressure simulation model, with its performance confirmed through analyses of spectrogram-based errors. Ultimately, a comparative study of the nonlinear behavior of the 45-story steel building, including collapse and associated reliabilities, was conducted to highlight the differences between NS and equivalent-stationary wind load applications on the peak/residual drifts, initiation of yielding, LCF-induced damage, and reliabilities against several system/component limit states. This dissertation advances the field of PBWE on multiple fronts with cutting-edge frameworks and insights from case studies, to accelerate the shift towards truly performance-based wind design.