Advances in high-performance computing (HPC) have empowered us to perform large-scale simulations for millions of variables in coupled fluid-structure systems involving complex geometries and multiphysics effects. These high-fidelity simulations via nonlinear partial differential equations have been providing invaluable physical insight for the development of new design and devices in aerospace and marine/offshore engineering. Despite efficient numerical methods and powerful supercomputers, the state-of-the-art computational fluid dynamics (CFD) and coupled fluid-structure simulations are somewhat inefficient hence less attractive with regard to the design optimization, parameter space exploration and the development of control and monitoring strategies for aerospace and offshore structures. In this talk, I will present some of our recent developments to integrate and to complement the HPC-based high-fidelity computations with the emerging field of data science and machine learning. The primary focus of this seminar is: (i) to discuss efficient reduced-order models for the physical modeling of dynamical fluid-structure systems, and (ii) to explore the integration of projection-based model reduction with deep neural networks for dynamical predictions. A series of canonical academic test cases will be covered to elucidate the integration of standard CFD with model reduction and deep learning techniques for the stability analysis and prediction of unsteady fluid flow and fluid-structure interaction. Some efforts on the iterative optimization and feedback active control will be demonstrated with the learned dynamical models. The proposed hybrid high-fidelity CFD with data-driven computing framework is precisely aligned with the current aerospace industry needs on structural life prediction, real-time flow control, diagnosis and monitoring via physics-based digital twin. Finally, I will provide some review of challenging problems that keep intense interest in this important topic from the industry standpoint.
