Project & Service

Project Introduction

This project aims to design an iterative solver algorithm based on neural networks to enhance the computational speed of OpenFOAM simulation cases. OpenFOAM is a widely used open-source computing platform for fluid dynamics and multiphysics simulations. However, in complex fluid dynamics problems, its traditional iterative solvers often face issues with excessively long convergence times. By introducing neural networks, the project will explore how to optimize the iterative process using deep learning algorithms.

Research Focus

1. Accelerating the computational speed of iterative algorithms using neural networks, achieving a speedup of 3 times compared to baseline algorithms.

2. The inference program for the neural network and the computational program for the iterative algorithm will be implemented based on MPI and integrated into the OpenFOAM software.

Main Outputs

A program and documentation in the form of an OpenFOAM plugin.

Specific Application Scenarios and Functions

In complex fluid dynamics simulations, traditional iterative solvers in OpenFOAM typically suffer from long computation times and low convergence efficiency when handling large-scale problems. By designing a neural network-based iterative solver algorithm, the solution process can be accelerated. Specific application scenarios include aerodynamics simulations, thermal fluid transport, and turbulence modeling. The neural network will optimize algorithm parameters or operators by learning from historical data, reducing the number of iterations needed and speeding up convergence. In engineering design, this algorithm can significantly shorten simulation times, enhance research and development efficiency, and provide technical support for rapid decision-making in fields such as aerospace and automotive manufacturing.

Collaboration Model

Joint laboratory project with Huawei Solver