A neural network surrogate model for the performance assessment of a vertical structure subjected to non-stationary, tornadic wind loads

Abstract

Despite significant advancements in computational technologies and methods, the comprehensive assessment of the performance capacities and risk of structures built in environments prone to severe natural hazards is still a daunting task under standard Monte Carlo-based simulation schemes. This issue is particularly relevant for the consideration of wind actions from loads generated by non-stationary phenomena (e.g. tornadoes) because of extreme complexities in the simulated flow field and the fluid-structure interaction. To mitigate such computational burdens, this study proposes a surrogate modeling approach that utilizes predicted fragilities from artificial neural networks (ANNs) to facilitate the performance-based assessment of a vertical structure subjected to tornadic wind loads. Calibration data for the feedforward ANNs are extracted from numerically generated responses based on a derived wind loading model that capitalizes on the developments of various analytical formulations of a tornado’s wind field. Uncertainties in the structural behavior and in the overall modeling procedure are incorporated in the process, culminating in a life-cycle cost assessment that incorporates a practical, economic value to the simulation framework. The novel application of ANNs in this study, therefore, empowers a more robust performance-based framework for the risk evaluation of structures subjected to tornado wind loads.