Abstract:Particulate Stokesian flows describe the hydrodynamics of rigid or deformable particles in Stokes flows. Due to highly nonlinear fluid-structure interaction dynamics, moving interfaces, and multiple scales, numerical simulations of such flows are challenging and expensive. In this Letter, we propose a generic machine-learning-augmented reduced model for these flows. Our model replaces expensive parts of a numerical scheme with multilayer perceptrons. Given the physical parameters of the particle, our model generalizes to arbitrary geometries and boundary conditions without the need to retrain the regression function. It is 10 times faster than a state-of-the-art numerical scheme having the same number of degrees of freedom and can reproduce several features of the flow quite accurately. We illustrate the performance of our model on integral equation formulation of vesicle suspensions in two dimensions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the high computational cost encountered in the particle hydrodynamics simulation in Stokesian suspensions. Specifically, due to the highly nonlinear fluid - structure interaction dynamics, moving interfaces, and multi - scale characteristics, the numerical simulation of these fluids is very challenging and costly. To solve this problem, the author proposes a general machine - learning - enhanced reduced - model (MLARM) to accelerate the simulation of Stokesian suspensions. ### Main Objectives 1. **Reduce Computational Cost**: By using a multi - layer perceptron (MLP) to replace the expensive parts in the numerical scheme, the simulation speed is increased by more than 10 times while maintaining an accuracy similar to that of high - fidelity simulations. 2. **Improve Universality**: The proposed model can be generalized to arbitrary geometries and boundary conditions without retraining the regression function. 3. **Accurately Reproduce Fluid Characteristics**: The model can accurately reproduce multiple characteristics of the fluid in low - resolution simulations. ### Method Overview 1. **Application of Multi - layer Perceptron (MLP)**: - Use MLP to approximate multiple spatial nonlinear operators in the numerical scheme. - Select these operators based on their stiffness characteristics, computational cost, and generalization ability to unseen data. 2. **High - Fidelity Simulation**: - Use high - fidelity (i.e., highly refined in space and time) simulations to train MLP. - High - fidelity simulations are carried out in "burst mode", that is, single - particle simulations for one time step. 3. **Combination with Low - Fidelity Simulation**: - Combine the trained MLP with low - fidelity simulations to achieve multi - particle simulations over a long time range, even under restricted geometric conditions. ### Specific Applications - **Single - Particle Dynamics**: Train MLP to approximate the relaxation process of a single particle in the absence of an external flow field. - **Influence of External Flow Field**: Use Fourier series to decompose the external flow field and train MLP to approximate the influence of the flow field on the particle boundary. - **Multi - Particle System**: For the case of multiple particles in a confined flow, only the calculation method of the external flow field needs to be adjusted, and the other parts remain unchanged. ### Experimental Verification - **Benchmark Tests**: Verify the performance of MLARM through multiple benchmark tests, including the equilibrium shape in parabolic flow, cross - streamline migration in curved streamlines, dilute and dense suspensions in Taylor - Couette flow, etc. - **Accuracy and Efficiency**: The results show that MLARM is superior to traditional low - resolution simulations and simulations with the same computational cost in terms of both computational time and accuracy. ### Conclusion The author proposes a machine - learning - enhanced reduced - model (MLARM) that combines high - fidelity simulations and low - fidelity numerical schemes, which successfully reduces the computational cost of Stokesian suspension simulations while maintaining high accuracy. This method performs well under a variety of flow conditions and has broad application prospects.

Machine learning acceleration of simulations of Stokesian suspensions

Parallel contact-aware simulations of deformable particles in 3D Stokes flow

A scalable computational platform for particulate Stokes suspensions

Stacked Generative Machine Learning Models for Fast Approximations of Steady-State Navier-Stokes Equations

Machine learning accelerated computational fluid dynamics

Coupling Machine Learning Local Predictions with a Computational Fluid Dynamics Solver to Accelerate Transient Buoyant Plume Simulations

Multiscale Stochastic Simulation and Modeling

Fast Stokesian dynamics

An integral equation method for the simulation of doubly-periodic suspensions of rigid bodies in a shearing viscous flow

A Multifidelity Machine Learning Based Semi-Lagrangian Finite Volume Scheme for Linear Transport Equations and the Nonlinear Vlasov–Poisson System

Scientific Machine Learning for Modeling and Simulating Complex Fluids

Fast Simulation of Particulate Suspensions Enabled by Graph Neural Network

A numerical method for suspensions of articulated bodies in viscous flows

High-volume fraction simulations of two-dimensional vesicle suspensions

Discovering Artificial Viscosity Models for Discontinuous Galerkin Approximation of Conservation Laws using Physics-Informed Machine Learning

Parallel O(N) Stokes' solver towards scalable Brownian dynamics of hydrodynamically interacting objects in general geometries

A turbulent eddy-viscosity surrogate modeling framework for Reynolds-averaged Navier-Stokes simulations

Fast and accurate solvers for simulating Janus particle suspensions in Stokes flow

Simulation of Fluid–particles Flows: Heavy Particles, Flowing Regime, and Asymptotic-Preserving Schemes

Accelerating a hybrid continuum-atomistic fluidic model with on-the-fly machine learning

A simulation platform for slender, semiflexible, and inextensible fibers with Brownian hydrodynamics and steric repulsion