Spatiotemporal parallel physics-informed neural networks: A framework to solve inverse problems in fluid mechanics

doi:10.1063/5.0155087

CORC > 力学研究所 > 中国科学院力学研究所 > 流固耦合系统力学重点实验室(2012-)

	Spatiotemporal parallel physics-informed neural networks: A framework to solve inverse problems in fluid mechanics
	Xu SF(许盛峰); Yan C(闫畅); Zhang, Guangtao ; Sun ZX(孙振旭); Huang RF(黄仁芳); Ju SJ(鞠胜军); Guo DL(郭迪龙); Yang GW(杨国伟)
刊名	PHYSICS OF FLUIDS
	2023-06-01
卷号	35 期号:6 页码:65141
ISSN号	1070-6631
DOI	10.1063/5.0155087
英文摘要	Physics-informed neural networks (PINNs) are widely used to solve forward and inverse problems in fluid mechanics. However, the current PINNs framework faces notable challenges when presented with problems that involve large spatiotemporal domains or high Reynolds numbers, leading to hyper-parameter tuning difficulties and excessively long training times. To overcome these issues and enhance PINNs' efficacy in solving inverse problems, this paper proposes a spatiotemporal parallel physics-informed neural networks (STPINNs) framework that can be deployed simultaneously to multi-central processing units. The STPINNs framework is specially designed for the inverse problems of fluid mechanics by utilizing an overlapping domain decomposition strategy and incorporating Reynolds-averaged Navier-Stokes equations, with eddy viscosity in the output layer of neural networks. The performance of the proposed STPINNs is evaluated on three turbulent cases: the wake flow of a two-dimensional cylinder, homogeneous isotropic decaying turbulence, and the average wake flow of a three-dimensional cylinder. All three turbulent flow cases are successfully reconstructed with sparse observations. The quantitative results along with strong and weak scaling analyses demonstrate that STPINNs can accurately and efficiently solve turbulent flows with comparatively high Reynolds numbers.
分类号	一类/力学重要期刊
WOS研究方向	Mechanics ; Physics
语种	英语
WOS记录号	WOS:001021259300005
资助机构	National Key Research and Development Project [2022YFB2603400] ; International Partnership Program of Chinese Academy of Sciences [025GJHZ2022118FN] ; China National Railway Group Science and Technology Program [K2023J047]
其他责任者	Sun, ZX (corresponding author), Chinese Acad Sci, Inst Mech, Key Lab Mech Fluid Solid Coupling Syst, Beijing 100190, Peoples R China.
内容类型	期刊论文
源URL	[http://dspace.imech.ac.cn/handle/311007/92588]
专题	力学研究所_流固耦合系统力学重点实验室(2012-)
作者单位	1.{Xu, Shengfeng, Yan, Chang, Sun, Zhenxu, Huang, Renfang, Ju, Shengjun, Guo, Dilong, Yang, Guowei} Chinese Acad Sci, Inst Mech, Key Lab Mech Fluid Solid Coupling Syst, Beijing 100190, Peoples R China 2.{Xu, Shengfeng, Guo, Dilong, Yang, Guowei} Univ Chinese Acad Sci, Sch Engn Sci, Beijing 100049, Peoples R China 3.{Yan, Chang} Univ Chinese Acad Sci, Sch Future Technol, Beijing 100049, Peoples R China 4.{Zhang, Guangtao} SandGold AI Res, Guangzhou 510642, Peoples R China 5.{Zhang, Guangtao} Univ Macau, Fac Sci & Technol, Dept Math, Macau 519000, Peoples R China
推荐引用方式 GB/T 7714	Xu SF,Yan C,Zhang, Guangtao,et al. Spatiotemporal parallel physics-informed neural networks: A framework to solve inverse problems in fluid mechanics[J]. PHYSICS OF FLUIDS,2023,35(6):65141.
APA	许盛峰.,闫畅.,Zhang, Guangtao.,孙振旭.,黄仁芳.,...&杨国伟.(2023).Spatiotemporal parallel physics-informed neural networks: A framework to solve inverse problems in fluid mechanics.PHYSICS OF FLUIDS,35(6),65141.
MLA	许盛峰,et al."Spatiotemporal parallel physics-informed neural networks: A framework to solve inverse problems in fluid mechanics".PHYSICS OF FLUIDS 35.6(2023):65141.