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

13Citations
Citations of this article
18Readers
Mendeley users who have this article in their library.
Get full text

Abstract

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.

References Powered by Scopus

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

8549Citations
N/AReaders
Get full text

Physics-informed machine learning

3536Citations
N/AReaders
Get full text

Validity of the single processor approach to achieving large scale computing capabilities

2999Citations
N/AReaders
Get full text

Cited by Powered by Scopus

A comprehensive review of advances in physics-informed neural networks and their applications in complex fluid dynamics

7Citations
N/AReaders
Get full text

Physics-informed neural networks for incompressible flows with moving boundaries

6Citations
N/AReaders
Get full text

Transfer learning through physics-informed neural networks for bubble growth in superheated liquid domains

5Citations
N/AReaders
Get full text

Register to see more suggestions

Mendeley helps you to discover research relevant for your work.

Already have an account?

Cite

CITATION STYLE

APA

Xu, S., Yan, C., Zhang, G., Sun, Z., Huang, R., Ju, S., … Yang, G. (2023). Spatiotemporal parallel physics-informed neural networks: A framework to solve inverse problems in fluid mechanics. Physics of Fluids, 35(6). https://doi.org/10.1063/5.0155087

Readers' Seniority

Tooltip

PhD / Post grad / Masters / Doc 4

67%

Professor / Associate Prof. 1

17%

Lecturer / Post doc 1

17%

Readers' Discipline

Tooltip

Engineering 4

67%

Physics and Astronomy 1

17%

Computer Science 1

17%

Save time finding and organizing research with Mendeley

Sign up for free