A Neural-Network Approach for the Stability of Loosely-Coupled FSI Simulations in Hemodynamics

77/2026

A Neural-Network Approach for the Stability of Loosely-Coupled FSI Simulations in Hemodynamics

Tuesday 13th October 2026

Renzi, F.; Pagani, S.; Antonietti, P.F.; Vergara, C.

The development of accurate and efficient computational methods for fluid-structure interaction (FSI) problems remains an active area of research, given the wide range of engineering and biomedical applications involving FSI. Among the available approaches, partitioned loosely coupled (LC) schemes are particularly attractive for large-scale simulations due to their modularity and reduced computational cost. However, their applicability is limited by stability issues in high added-mass regimes, where the fluid and structure densities are comparable, as occurs in hemodynamics. In our recent work, we proposed an LC algorithm based on Dirichlet-Neumann interface coupling, referred to as LC-DN-alpha scheme, which improved the stability of the coupling by introducing a stabilization parameter alpha. Although theoretical stability has been demonstrated under suitable constraints on alpha, the dependence of the optimal value of alpha on the physical and numerical characteristics of general FSI problems is not yet fully understood, making its practical selection challenging. In this work, we address this open question by proposing a data-driven strategy for automatically estimating the optimal value of alpha. Specifically, we develop a feed-forward neural network to predict alpha from the descriptors of the FSI problem. The model is trained on a dataset of baseline FSI problems in large added-mass regimes and is subsequently evaluated on more realistic configurations to assess its robustness and generalization capabilities. The numerical results show that the proposed model generalizes across the considered FSI test scenarios and provides robust estimates of alpha under moderate variations in the problem setup, including changes in geometry and loading conditions, that do not completely alter the fluid-structure interaction mechanics with respect to those represented in the training set. The methodology has been intentionally developed and validated on simplified FSI problems, providing a proof of concept for data-driven strategies to improve the efficiency of partitioned FSI simulations and supporting their extension to more complex and realistic scenarios.