DATA-DRIVEN MODELING AND DEEP LEARNING FOR FLUID-STRUCTURE INTERACTION

Abstract

Fluid-structure interaction (FSI) is ubiquitous in mechanical engineering applications. Due to the complexity of these systems, the wake-body synchronization is not well understood and real-time prediction of flow dynamics is almost impossible. This study aims to explain wake-body synchronization mechanism and efficiently predict FSI dynamics using deep learning. A self-sustaining process is introduced to compare the flow structure formation behind stationary and oscillating bluff bodies. Further, the FSI feature breakdown revealed that the shear layer feeds vorticity to the near wake and the vortices, sustaining vortex shedding and bluff body motion. The deep learning techniques are developed to predict flow dynamics accurately and efficiently. Further, analogies between deep learning algorithms and well-established numerical/ analytical techniques are presented. Finally, a hybrid of model reduction and deep learning is developed to obtain time histories of flow fields. These new tools enable FSI dynamics to be integrated into design procedures and real-time operations.