What problem does this paper attempt to address?

This paper discusses how to combine machine learning (ML) with computational fluid dynamics (CFD) to improve the efficiency of technology and natural system simulations. The authors point out that implementing CFD+ML algorithms in large-scale problems poses challenges, mainly involving data exchange, synchronization, and computation on heterogeneous hardware. To address this, they propose an efficient and scalable solution using the open-source software OpenFOAM and SmartSim. SmartSim provides an Orchestrator that simplifies the programming of CFD+ML algorithms and enables scalable data exchange between ML and CFD clients. The paper demonstrates how to effectively couple different parts of OpenFOAM, such as preprocessing applications, function objects, and grid motion solvers, using SmartSim. Additionally, an OpenFOAM submodule and examples are provided as starting points for real-world applications. The paper focuses on three questions: 1. How to embed ML into OpenFOAM as part of the simulation? 2. Which computational architecture is effective in utilizing CPU and GPU resources for CPU-based codes like OpenFOAM? 3. What is the basic workflow design pattern for building complex CFD+ML applications? To overcome these challenges, the paper suggests a loosely coupled assumption, where scientific simulations and ML algorithms maintain loose coupling on the software and computational levels, leveraging a centralized, computation-enabled database for data sharing. The paper provides three examples to illustrate this loosely coupled approach, including using Bayesian optimization to adjust turbulence model parameters, streaming CFD data from simulations for calculating reduced bases, and approximating grid point displacements using online training and inference. The paper's structure consists of an introduction, architecture, integration of OpenFOAM and SmartSim, implementation and results of use cases. Using SmartSim and the SmartRedis library, the paper demonstrates how to implement these loosely coupled CFD+ML workflows in the OpenFOAM environment, highlighting their adaptability and scalability to high-performance computing.

Combining Machine Learning with Computational Fluid Dynamics using OpenFOAM and SmartSim

Multiscale Fluid Mechanics and Modeling

In Situ Framework for Coupling Simulation and Machine Learning with Application to CFD

Recent Advances on Machine Learning for Computational Fluid Dynamics: A Survey

Machine learning accelerated computational fluid dynamics

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

Enhancing computational fluid dynamics with machine learning

Emerging Trends in Machine Learning for Computational Fluid Dynamics

Computational simulation of mass transfer in membranes using hybrid machine learning models and computational fluid dynamics

Demystifying the Data Need of ML-surrogates for CFD Simulations

Machine Learning for Fluid Mechanics

An Exploration of Flow Control Using Machine Learning and Computational Fluid Dynamics

Machine Learning Based Fluid-Transportation Monitoring and Controlling

Challenges and Opportunities for Machine Learning in Fluid Mechanics

Improving CFD simulations by local machine-learned correction

Towards a new paradigm in intelligence-driven computational fluid dynamics simulations

Integration of machine learning and CFD for modeling mass transfer in water treatment using membrane separation process

Coarse-Grid Computational Fluid Dynamic (CG-CFD) Error Prediction using Machine Learning

NeurIPS 2024 ML4CFD Competition: Harnessing Machine Learning for Computational Fluid Dynamics in Airfoil Design

High-resolution fluid–particle interactions: a machine learning approach