TY - JOUR
T1 - Super-resolution reconstruction of turbulent flows with machine learning
AU - Fukami, Kai
AU - Fukagata, Koji
AU - Taira, Kunihiko
N1 - Funding Information:
K. Fukami and K. Fukagata are grateful for support from the Japan Society for the Promotion of Science (KAKENHI grant number 18H03758). K. Taira acknowledges support from the US Army Research Office (grant number W911NF-17-1-0118) and US Air Force Office of Scientific Research (grant number FA9550-16-1-0650) and thanks L. Mathelin, S. L. Brunton and J. N. Kutz for stimulating discussions
PY - 2019/7/10
Y1 - 2019/7/10
N2 - We use machine learning to perform super-resolution analysis of grossly under-resolved turbulent flow field data to reconstruct the high-resolution flow field. Two machine learning models are developed, namely, the convolutional neural network (CNN) and the hybrid downsampled skip-connection/multi-scale (DSC/MS) models. These machine learning models are applied to a two-dimensional cylinder wake as a preliminary test and show remarkable ability to reconstruct laminar flow from low-resolution flow field data. We further assess the performance of these models for two-dimensional homogeneous turbulence. The CNN and DSC/MS models are found to reconstruct turbulent flows from extremely coarse flow field images with remarkable accuracy. For the turbulent flow problem, the machine-leaning-based super-resolution analysis can greatly enhance the spatial resolution with as little as 50 training snapshot data, holding great potential to reveal subgrid-scale physics of complex turbulent flows. With the growing availability of flow field data from high-fidelity simulations and experiments, the present approach motivates the development of effective super-resolution models for a variety of fluid flows.
AB - We use machine learning to perform super-resolution analysis of grossly under-resolved turbulent flow field data to reconstruct the high-resolution flow field. Two machine learning models are developed, namely, the convolutional neural network (CNN) and the hybrid downsampled skip-connection/multi-scale (DSC/MS) models. These machine learning models are applied to a two-dimensional cylinder wake as a preliminary test and show remarkable ability to reconstruct laminar flow from low-resolution flow field data. We further assess the performance of these models for two-dimensional homogeneous turbulence. The CNN and DSC/MS models are found to reconstruct turbulent flows from extremely coarse flow field images with remarkable accuracy. For the turbulent flow problem, the machine-leaning-based super-resolution analysis can greatly enhance the spatial resolution with as little as 50 training snapshot data, holding great potential to reveal subgrid-scale physics of complex turbulent flows. With the growing availability of flow field data from high-fidelity simulations and experiments, the present approach motivates the development of effective super-resolution models for a variety of fluid flows.
KW - Computational methods
KW - Homogeneous turbulence
KW - Wakes
UR - http://www.scopus.com/inward/record.url?scp=85065398025&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85065398025&partnerID=8YFLogxK
U2 - 10.1017/jfm.2019.238
DO - 10.1017/jfm.2019.238
M3 - Article
AN - SCOPUS:85065398025
VL - 870
SP - 106
EP - 120
JO - Journal of Fluid Mechanics
JF - Journal of Fluid Mechanics
SN - 0022-1120
ER -