Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
Transformer-based models for predictive simulations of vortex-induced vibrations
Submission Author:
Gabriel Mario Guerra Bernadá , RJ , Brazil
Co-Authors:
Jacques Honigbaum, Fernando Rochinha, Rodolfo S. M. Freitas, Souleymane Zio, Gabriel Mario Guerra Bernadá
Presenter: Jacques Honigbaum
doi://10.26678/ABCM.COBEM2023.COB2023-1881
Abstract
The impact of vortex-induced vibrations (VIV) can assume a fundamental part in the motion of submerged objects, especially those related to offshore designed structures. Mainly due to large displacements, vortex shedding on the wake can result in continued degradation of structural performance or even catastrophic failure, thus accurate prediction of the structural response is needed. The structure nonlinear dynamics is coupled to the surrounding flow dynamics by a typical high-fidelity fluid-structure interaction (FSI) model required for the description of vortex-induced vibrations. The latter usually is based on a Navier-Stokes formulation to be solved with computational fluid dynamics (CFD) methods with a fine mesh that must be suited to structure motion. Nevertheless, considering extensive multi-query analysis like optimization, real-time response, or uncertainty quantification, the resulting model often demands a high computational cost. High-fidelity codes based on physics-based models are frequently difficult to use due to these time-consuming tasks. A decent option to beat such limits is the development of surrogate models that have become well known in some areas of research because of their aptitude in being effective alternatives for high-fidelity models. As essential tools for simplifying analysis, these kinds of models can be very useful in a wide range of industrial applications because they can make predictions at a much lower computational cost than CFD. Due to their potential to enhance the capacity of computational simulations to describe complex physical systems, data-driven machine learning (ML) models, which can combine field or experimental data with high-fidelity simulations, have gained prominence in this context. Data-driven ML predictive models that provide accurate predictions at a low cost have been the subject of several studies. Self-attention based transformer models have recently been used to model dynamic systems, which have the potential to take the place of costly computational models. Different dynamical systems can be accurately predicted by this model, which outperforms traditional approaches used in scientific machine learning literature. As a surrogate model for VIV dynamics, we present in this work a machine learning strategy based on self-attention transformers. Through numerical testing, we demonstrate that the surrogate model can accurately predict VIV dynamics. Overmore, it makes it easier to study important aspects of commonly used wake oscillator models.
Keywords
Vortex-induced vibrations Transformers, Surrogate models, Data-driven machine learning (ML) models
