Session: 08-12-01: Optimization, Uncertainty and Probability

Paper Number: 165524

Surrogate Modeling of Fluid-Structure Interactions in Vertical Towers Under Downburst Wind Loads Using Artificial Neural Networks: Preliminary Results 

The growing occurrence and severity of extreme wind events, such as tornadoes and downbursts, have made it increasingly important to assess their effects on vertical tower structures. These events pose significant challenges due to the complexity of accurately modeling the wind field, the loads and the resulting fluid-structure interaction. Traditional methods for evaluating structural response under non-stationary wind conditions typically rely on Monte Carlo-based simulations. Although these approaches offer reliable probabilistic assessments, their high computational demands limit their applicability to small-scale studies or unrealistic applications.

This study proposes the use of Artificial Neural Network (ANN) surrogate modeling to predict fluid-structure interaction effects in uncertain, non-stationary wind environments through fragility analysis. The study analyzes the response of vertical tower structures, subjected to random downburst wind loads. By significantly reducing the computational burden, the methodology facilitates a more efficient response analysis while maintaining accuracy. The primary objective is to lower the computational time required for high-fidelity “fragility analysis” and to provide reliable estimations of structural fragility functions with a reduced number of support points.

The surrogate model is trained and calibrated using data, extracted from a series of random dynamic simulations that incorporate uncertainties in both the downburst wind field and the structural response. The wind field model consists of two primary components: a deterministic mean wind field, characterized by a distinctive "nose-shaped" velocity profile - fundamentally different from conventional stationary wind profiles - and two, non-stationary turbulence components, primarily in the horizontal direction, modeled as amplitude-modulated extensions of stationary turbulence. Furthermore, the model captures the temporal and spatial evolution of the downburst’s mean wind field.

To validate the proposed simulation approach, the method is first applied to the structural model and response of a “point-like” (plate) system, i.e., a monopole tower, represented by two generalized degrees of freedom. Second, the approach is applied to the study of a horizontal-axis wind turbine tower structure; the dynamic model describes in a generalized form the coupled blade/tower assembly, i.e., by two degrees of freedom; considering the motion in both horizontal directions (in and out of the rotor plane), the effective number of degrees of freedom becomes four. Third, the model of the CAARC benchmark tall building is analyzed using a two-mode generalized dynamic representation.

The accuracy of the ANN model is evaluated by comparing ANN-based fragility predictions with those obtained from high-fidelity Monte Carlo simulations. As part of Performance-Based Wind Engineering (PBWE), this study demonstrates the potential of ANNs to enhance the risk assessment of various structural systems under the effect of non-stationary thunderstorm downburst loads. The proposed formulation improves computational efficiency and offers a practical tool for life-cycle cost analysis in wind engineering.

This material is based upon work supported in part by the National Science Foundation (NSF) of the United States of America, Award CMMI-2330150 (sub-award from Iowa State University). The financial support of Sapienza University of Rome is also acknowledged.

Presenting Author: Matteo Polucci Northeastern University

Presenting Author Biography: I am PhD student enrolled in a dual program between Northeastern University and Sapienza University of Rome. I am working on wind engineering with a focus on the extreme wind events.

Authors:

Matteo Polucci Northeastern University
Luca Caracoglia Northeastern University
Francesco Petrini Sapienza Università di Roma