Uncertainty quantification and guidance on the use of wind tunnel-informed stochastic wind load models for application in performance-based wind engineering

Abstract

Probabilistic performance-based wind engineering (PBWE) is a state-of-the-art framework that allows for direct uncertainty propagation through the hazard, aerodynamic, structural response, damage, and loss models in assessing the performance of building systems subject to wind loads. To enable the application of PBWE, the aerodynamic model needs to be capable of simulating multi-variate stochastic processes representing the dynamic wind loads acting on the system. For efficient simulation, models based on spectral proper orthogonal decomposition (POD) are of interest because such an approach allows subprocesses to be generated independently while a few spectral modes are needed to adequately represent the fluctuating component of the wind load, hence significantly reducing the computational cost. To ensure that the aerodynamic phenomena observed in the wind tunnel are captured and used to inform the aerodynamic model, a data-driven POD-based stochastic wind model has been recently proposed where wind tunnel data is used to calibrate the spectral modes and eigenvalues of the target load processes. While the proposed model can serve as a general wind simulation tool for any wind direction and geometry, errors associated with such a model need to be quantified and treated properly in applications of PBWE. To investigate the errors, an extensive experimental study on a rectangular building model was conducted at the University of Florida NHERI boundary layer wind tunnel for varying wind directions and experimental settings. In particular, tests were performed for a total duration, at model scale, of 15 minutes and repeated five times in increments of 10 degrees. The datasets were divided into two groups; one was used to define the target spectra while the other was used as a testing set to evaluate the errors. Three types of errors were analyzed including the errors associated with using standard wind tunnel data (e.g., a single 30-second record), numerical scheme, and truncation of spectral modes. Results have revealed insights into typical model errors and control techniques that help facilitate the development of practical guidance on the use of the proposed wind tunnel-informed stochastic load model in a PBWE setting.