Surrogate-based acceleration of quasi-Newton techniques for fluid-structure interaction simulations. (February 2022)

• Record Type: Journal Article
• Authors: Delaissé, Nicolas; Demeester, Toon; Fauconnier, Dieter; Degroote, Joris
• Is Part Of: Computers & structures. Volume 260(2022)
• Publication Date: 2022-02
• Subjects: Fluid-structure interaction -- Quasi-Newton method -- Surrogate model -- Partitioned algorithm; Structural engineering -- Data processing -- Periodicals; Electronic data processing -- Structures, Theory of -- Periodicals; 624.171
• DOI: 10.1016/j.compstruc.2021.106720
• Languages: English
• ISSNs: 0045-7949

Abstract

Highlights: Partitioned fluid-structure interaction is often simulated with quasi-Newton methods. A typical quasi-Newton technique uses the solvers as black boxes. Often there is knowledge about the coupled system in the form of a surrogate model. A surrogate model provides an initial guess and supplementary Jacobian. Including a surrogate model accelerates and stabilizes the coupling. Abstract: Quasi-Newton methods have proven to be an efficient way to couple partitioned solvers in fluid-structure interaction problems, as they are able to stabilize cases with high added-mass, as well as accelerate the convergence. However, these methods assume that the coupled system is a complete black-box, whereas often, its behavior is well approximated by a surrogate model. Such a model may be obtained by coarsening the system, simplifying the physics, reverting to analytical approximations or considering the system at a previous point in time. The principal idea of this work is to use an initial solution and a Jacobian provided by the surrogate model, to expedite the convergence even further. This article presents a new framework for the inclusion of surrogate models in quasi-Newton methods and positions several existing methods with respect to it.