Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Azzeddine Tiba, Thibault Dairay, Florian de Vuyst, Iraj Mortazavi, Juan-Pedro Berro Ramirez

TL;DR

This work tackles the computational burden of unsteady partitioned FSI under strong added-mass effects by introducing a non-intrusive data-driven predictor that couples solid and fluid reduced-order models (ROMs) with full-order models (FOMs). The predictor uses a POD-based encoder-regressor-decoder architecture for both subproblems and includes an online adaptation mechanism to incorporate high-fidelity data, yielding a physics-aware initialization for the next time step. Across three challenging FSI cases, the approach achieves faster convergence and up to 3.2x speedups compared with classical predictors, while maintaining high accuracy in the solid and stress fields. The method is non-intrusive, compatible with existing solvers, and can extrapolate to unseen time-parameter regimes, with potential extensions to replacing the solid solver entirely by a ROM in future work.

Abstract

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.

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

TL;DR

This work tackles the computational burden of unsteady partitioned FSI under strong added-mass effects by introducing a non-intrusive data-driven predictor that couples solid and fluid reduced-order models (ROMs) with full-order models (FOMs). The predictor uses a POD-based encoder-regressor-decoder architecture for both subproblems and includes an online adaptation mechanism to incorporate high-fidelity data, yielding a physics-aware initialization for the next time step. Across three challenging FSI cases, the approach achieves faster convergence and up to 3.2x speedups compared with classical predictors, while maintaining high accuracy in the solid and stress fields. The method is non-intrusive, compatible with existing solvers, and can extrapolate to unseen time-parameter regimes, with potential extensions to replacing the solid solver entirely by a ROM in future work.

Abstract

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.
Paper Structure (26 sections, 58 equations, 28 figures, 2 tables, 4 algorithms)

This paper contains 26 sections, 58 equations, 28 figures, 2 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. FOM-FOM fluid-structure interaction black-box coupling
  3. Coupling scheme
  4. Convergence acceleration
  5. Predictors
  6. Non-intrusive ROM-FOM fluid-structure interaction acceleration
  7. Solid reduced-order model
  8. Forces encoder $\mathcal{E_F(\cdot)}$
  9. Regressor $\mathcal{I_S(\cdot)}$
  10. Displacement decoder $\mathcal{D_S(\cdot)}$
  11. In a nutshell:
  12. Adaptive data-driven predictors with fluid and solid ROMs
  13. Prediction of a better initial guess: Inner coupling
  14. Fluid ROM components:
  15. Fluid ROM online update:
  16. ...and 11 more sections

Figures (28)

  • Figure 1: Brief illustrative representation of the presented approach.
  • Figure 2: Fluid and solid ROM components
  • Figure 3: Fluid and solid ROM components - Hybrid variant
  • Figure 4: Global FSI coupling scheme with the ROM-FOM coupling and data-driven adaptive predictors.
  • Figure 5: Example 1 - 1D flexible tube test case schematic explanation (from degroote_stability_2008). $\sigma_{\varphi \varphi}$ is the vessel hoop stress, $h$ is the thickness and $\Delta x$ is the length of the finite volume cell.
  • ...and 23 more figures