翼型
解算器
有限元法
人工神经网络
计算科学
航空航天工程
计算机科学
物理
应用数学
数学优化
数学
结构工程
工程类
人工智能
作者
Wee-Beng Tay,Tian-Chun Liao-Yang,Hongrui Luo,Kai-Peng Chen,Shuo Chen,Jun-Won Yang,Simon See,Boo Cheong Khoo
出处
期刊:Journal of Aircraft
[American Institute of Aeronautics and Astronautics]
日期:2025-08-31
卷期号:: 1-9
被引量:1
摘要
A framework has been designed to predict the flow fields and perform shape optimization of two-element airfoils using Nvidia Modulus, an open-source Physics Informed Neural Network (PINN) solver. Modulus integrates physical law constraints into the loss terms during training of its neural network so that its prediction honors the underlying physical principles. There is no need for any external data, which is important because generating data requires valuable time and resources. With a single to a few neural network trainings, Modulus can predict flow fields and force coefficients for a range of geometric variables including angles of attack, airfoil thickness, and also over a range of fluid variables such as viscosity. In comparison to traditional computational fluid dynamics solvers such as OpenFOAM, Modulus is much faster once the networks are trained. The range splitting method can also be used to further improve prediction accuracy. For flow past two-element airfoils with Reynolds number ranging from 50 to 300, prediction results from PINN is in good agreement with those of OpenFOAM. Moreover, by formulating lift and drag coefficient prediction into an objective function, one can then couple it into an optimization routine to perform airfoil shape optimization to improve the airfoil’s performance.
科研通智能强力驱动
Strongly Powered by AbleSci AI