What does this research mean for the field?

Integrating a hybrid GNN-U-Net neural network to predict forcing field changes and initialize solver iterations accelerates particle-resolved simulations of dense suspensions by over a factor of two without altering governing equations or convergence criteria. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The study aims to resolve computational bottlenecks in particle-resolved simulations of dense suspensions by using a machine learning model to predict and initialize solver-internal forcing fields.

September 8, 2026Computers & FluidsOpen Access

Accelerating particle-resolved simulations of dense suspensions with hybrid GNN–U-Net neural networks

View Full Paper

Ask AI

Bookmark

Authors

MOMichel Orsi GBGianluca Boccardo DMDaniele Marchisio

Discussion

Member takes

Overview

Computational study demonstrates a hybrid GNN–U-Net reduces solver iterations in dense suspensions, indicating learned initializers accelerate physics simulations without altering governing equations.

Key Points

The study aims to resolve computational bottlenecks in particle-resolved simulations of dense suspensions by using a machine learning model to predict and initialize solver-internal forcing fields.
Developed a hybrid graph neural network (GNN) and U-Net architecture to predict time-step changes in the coupling forcing field within an OpenFOAM-based fictitious-domain solver.
Trained the model on a single simulation and evaluated its performance on unseen time intervals and six diverse test cases varying volume fraction, friction coefficient, particle interactions, domain size, and flow type.
Preserved the underlying governing equations, solver iteration loop, and convergence criteria without replacing the physics-based solver.
The learned initializer reduced mean forcing error by approximately 95% compared to naive persistence on unseen simulation intervals.
Integration into the solver produced an iteration reduction factor of 2.54 and an overall speedup of 2.13x (including inference overhead), while shifting time-averaged shear stress by only -1.65%.
Iteration speedups between 1.48x and 2.78x generalized across six distinct test cases not encountered during training.

Cite This Study

Orsi et al. (2026) studied this question.

synapsesocial.com/papers/6a9fd74958e84d0ff5b45bc8 https://doi.org/10.1016/j.compfluid.2026.107276

View Full Paper

Ask AI

Bookmark

September 8, 2026Computers & FluidsOpen Access

Accelerating particle-resolved simulations of dense suspensions with hybrid GNN–U-Net neural networks

View Full Paper

Ask AI

Bookmark

Authors

MOMichel Orsi GBGianluca Boccardo DMDaniele Marchisio

Discussion

Member takes

Overview

Key Points

The study aims to resolve computational bottlenecks in particle-resolved simulations of dense suspensions by using a machine learning model to predict and initialize solver-internal forcing fields.
Developed a hybrid graph neural network (GNN) and U-Net architecture to predict time-step changes in the coupling forcing field within an OpenFOAM-based fictitious-domain solver.
Trained the model on a single simulation and evaluated its performance on unseen time intervals and six diverse test cases varying volume fraction, friction coefficient, particle interactions, domain size, and flow type.
Preserved the underlying governing equations, solver iteration loop, and convergence criteria without replacing the physics-based solver.
The learned initializer reduced mean forcing error by approximately 95% compared to naive persistence on unseen simulation intervals.
Integration into the solver produced an iteration reduction factor of 2.54 and an overall speedup of 2.13x (including inference overhead), while shifting time-averaged shear stress by only -1.65%.
Iteration speedups between 1.48x and 2.78x generalized across six distinct test cases not encountered during training.

Cite This Study

Orsi et al. (2026) studied this question.

synapsesocial.com/papers/6a9fd74958e84d0ff5b45bc8 https://doi.org/10.1016/j.compfluid.2026.107276

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark