.png){kind=icon}
返回
A Predictive Model for Turbulence Evolution and Mixing Using Machine Learning
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2024.3442189.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
The high cost associated with high-fidelity computational fluid dynamics (CFD) is one of the main challenges that inhibit the design and optimisation of new fluid-flow systems. In this study, we explore the feasibility of a physics-informed deep learning approach to predict turbulence evolution and mixing without requiring a classical CFD solver. The deep learning architecture was inspired by integrating U-Net with inception modules for capturing the multi-scale nature of turbulent flows. In addition, a physics-constrained loss function was designed to enforce the mass and pressure conservation of the predicted solution. After trained, the optimised model was validated in the large eddy simulation (LES) of a forced turbulent mixing layer at two distinct Reynolds numbers ( $\mathrm {Re} =3000$ and 30000). The results demonstrate that the proposed approach achieves a promising solution accuracy and extrapolation ability with a significant reduction in computing time when compared to those obtained using a classical LES flow solver. The success in developing such a physics-informed deep learning approach not only justifies the potential of ML-based surrogate solvers for fast prototyping and design of generic fluid-flow systems but also highlights the key challenges arising from data-driven surrogate solver development for turbulence modelling.
Keyword:
Mathematical models
Computational modeling
Predictive models
Convolutional neural networks
Computer architecture
Feature extraction
Accuracy
Fluid dynamics
Machine learning
Computational fluid dynamics
data-driven approaches
machine learning
physics-informed neural networks
turbulence modeling
turbulent flows
期刊
IF:
3.6
论文数:
9.8W
被引数:
29.4W
机构
引用论文
Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations物理信息神经网络: 一种用于解决涉及非线性偏微分方程的正反问题的深度学习框架
OAAI
OAAI分享收藏Structural study of lanthanides(III) in aqueous nitrate and chloride solutions by EXAFS通过EXAFS对硝酸盐和氯化物水溶液中镧系元素 (III) 的结构研究
PREAI
分享收藏The aluminum-in-olivine thermometer for mantle peridotites — Experimental versus empirical calibration and potential applications.png)
Lithos
LithosIF0
PREAI
分享收藏
查看更多内容