Metamodelagem e machine learning em estudo de interação fluído-estrutura de uma esfera em rotação

Abstract

Computational fluid dynamics has been one of the biggest revolutions in applied physics studies. Understanding how the interaction between structures and fluids occurs is a basic field of industries such as aeronautics, maritime and oil. Within the scope of the study of flows, the analysis of a sphere immersed in a fluid becomes a canonical case. In the present work, a CFD code was used to simulate the behavior of the flow over a sphere at various rotations about its own axis. An adaptive mesh was used in combination with the immersed boundary method. The data obtained by probes along the simulation domain were used in the production of a metamodel capable of predicting the rotation of the sphere through the variation of the recorded velocities. The metamodel was constituted by the XGBoost machine learning algorithm, which was able to predict the rotation speed through the simulation data.