Super-resolution reconstruction of turbulent flow fields at various Reynoldsnumbers based on generative adversarial networks

This study presents a deep learning-based framework to recover high-resolution turbulentvelocity fields from extremely low-resolution data at various Reynolds numbers byutilizing the concept of generative adversarial networks. A multiscale enhancedsuper-resolution generative adversarial network is a...

Full description

Saved in:
Bibliographic Details
Published inPhysics of fluids (1994) Vol. 34; no. 1
Main Authors Yousif, Mustafa Z, Yu Linqi, Hee-Chang, Lim
Format Journal Article
LanguageEnglish
Published Melville American Institute of Physics 01.01.2022
Subjects
Direct numerical simulation
Fluid dynamics
Fluid flow
Generative adversarial networks
High resolution
Mathematical models
Physics
Reynolds number
Turbulent flow
Velocity
Velocity distribution
Online AccessGet full text

Cover

More Information
Summary:This study presents a deep learning-based framework to recover high-resolution turbulentvelocity fields from extremely low-resolution data at various Reynolds numbers byutilizing the concept of generative adversarial networks. A multiscale enhancedsuper-resolution generative adversarial network is applied as a model to reconstruct thehigh-resolution velocity fields, and direct numerical simulation data of turbulent channelflow with large longitudinal ribs at various Reynolds numbers are used to evaluate theperformance of the model. The model is found to have the capacity to accuratelyreconstruct the high-resolution velocity fields from data at two different down-samplingfactors in terms of the instantaneous velocity fields, two-point correlations, andturbulence statistics. The results further reveal that the model is able to reconstructhigh-resolution velocity fields at Reynolds numbers that fall within the range of thetraining Reynolds numbers.
ISSN:1070-6631
1089-7666