Machine-Learning Enhanced Predictors for Accelerated Convergence of Partitioned Fluid-Structure Interaction Simulations

Stable partitioned techniques for simulating unsteady fluid-structure interaction (FSI) are known to be computationally expensive when high added-mass is involved. Multiple coupling strategies have been developed to accelerate these simulations, but often use predictors in the form of simple finite-...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Computer physics communications 2025-05, Vol.310, p.109522, Article 109522
Hauptverfasser: Tiba, Azzeddine, Dairay, Thibault, De Vuyst, Florian, Mortazavi, Iraj, Ramirez, Juan Pedro Berro
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Stable partitioned techniques for simulating unsteady fluid-structure interaction (FSI) are known to be computationally expensive when high added-mass is involved. Multiple coupling strategies have been developed to accelerate these simulations, but often use predictors in the form of simple finite-difference extrapolations. In this work, we propose a non-intrusive data-driven predictor that couples reduced-order models of both the solid and fluid subproblems, providing an initial guess for the nonlinear problem of the next time step calculation. Each reduced order model is composed of a nonlinear encoder-regressor-decoder architecture and is equipped with an adaptive update strategy that adds robustness for extrapolation. In doing so, the proposed methodology leverages physics-based insights from high-fidelity solvers, thus establishing a physics-aware machine learning predictor. Using three strongly coupled FSI examples, this study demonstrates the improved convergence obtained with the new predictor and the overall computational speedup realized compared to classical approaches. •Novel predictor to accelerate convergence of fluid-structure interactions problems.•Coupled solution of solid and fluid reduced models is used as the next initial guess.•Reduced models take the form of encoder-regressor-decoder data- driven models.•Online adaptation of the reduced models is used for more robust extrapolation.•Faster convergence and speedups up to 3.2 versus classical- predictor based coupling. Highlights (for review).
ISSN:0010-4655
DOI:10.1016/j.cpc.2025.109522