Synthetic turbulent inflow generator using machine learning

Kai Fukami, Yusuke Nabae, Ken Kawai, Koji Fukagata

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

We propose a methodology for generating time-dependent turbulent inflow data with the aid of machine learning (ML), which has the possibility to replace conventional driver simulations or synthetic turbulent inflow generators. As for the ML model, we use an autoencoder-type convolutional neural network with a multilayer perceptron. For the test case, we study a fully developed turbulent channel flow at the friction Reynolds number of Reτ=180 for easiness of assessment. The ML models are trained using a time series of instantaneous velocity fields in a single cross section obtained by direct numerical simulation (DNS) so as to output the cross-sectional velocity field at a specified future time instant. From the a priori test in which the output from the trained ML model are recycled to the input, the spatiotemporal evolution of cross-sectional structure is found to be reasonably well reproduced by the proposed method. The turbulence statistics obtained in the a priori test are also, in general, in reasonable agreement with the DNS data, although some deviation in the flow rate was found. It is also found that the present machine-learned inflow generator is free from the spurious periodicity, unlike the conventional driver DNS in a periodic domain. As an a posteriori test, we perform DNS of inflow-outflow turbulent channel flow with the trained ML model used as a machine-learned turbulent inflow generator (MLTG) at the inlet. It is shown that the present MLTG can maintain the turbulent channel flow for a long time period sufficient to accumulate turbulent statistics, with much lower computational cost than the corresponding driver simulation. It is also demonstrated that we can obtain accurate turbulent statistics by properly correcting the deviation in the flow rate.

Original languageEnglish
Article number064603
JournalPhysical Review Fluids
Volume4
Issue number6
DOIs
Publication statusPublished - 2019 Jun 4

Fingerprint

Direct numerical simulation
Learning systems
Machine Learning
Generator
Turbulent Channel Flow
Channel flow
Driver
Statistics
Flow Rate
Velocity Field
Flow rate
Deviation
Multilayer neural networks
Output
Accumulate
Time series
Perceptron
Reynolds number
Turbulence
Instant

ASJC Scopus subject areas

  • Computational Mechanics
  • Modelling and Simulation
  • Fluid Flow and Transfer Processes

Cite this

Synthetic turbulent inflow generator using machine learning. / Fukami, Kai; Nabae, Yusuke; Kawai, Ken; Fukagata, Koji.

In: Physical Review Fluids, Vol. 4, No. 6, 064603, 04.06.2019.

Research output: Contribution to journalArticle

Fukami, Kai ; Nabae, Yusuke ; Kawai, Ken ; Fukagata, Koji. / Synthetic turbulent inflow generator using machine learning. In: Physical Review Fluids. 2019 ; Vol. 4, No. 6.
@article{efa0578adb804ed5ab653f34e7e658f0,
title = "Synthetic turbulent inflow generator using machine learning",
abstract = "We propose a methodology for generating time-dependent turbulent inflow data with the aid of machine learning (ML), which has the possibility to replace conventional driver simulations or synthetic turbulent inflow generators. As for the ML model, we use an autoencoder-type convolutional neural network with a multilayer perceptron. For the test case, we study a fully developed turbulent channel flow at the friction Reynolds number of Reτ=180 for easiness of assessment. The ML models are trained using a time series of instantaneous velocity fields in a single cross section obtained by direct numerical simulation (DNS) so as to output the cross-sectional velocity field at a specified future time instant. From the a priori test in which the output from the trained ML model are recycled to the input, the spatiotemporal evolution of cross-sectional structure is found to be reasonably well reproduced by the proposed method. The turbulence statistics obtained in the a priori test are also, in general, in reasonable agreement with the DNS data, although some deviation in the flow rate was found. It is also found that the present machine-learned inflow generator is free from the spurious periodicity, unlike the conventional driver DNS in a periodic domain. As an a posteriori test, we perform DNS of inflow-outflow turbulent channel flow with the trained ML model used as a machine-learned turbulent inflow generator (MLTG) at the inlet. It is shown that the present MLTG can maintain the turbulent channel flow for a long time period sufficient to accumulate turbulent statistics, with much lower computational cost than the corresponding driver simulation. It is also demonstrated that we can obtain accurate turbulent statistics by properly correcting the deviation in the flow rate.",
author = "Kai Fukami and Yusuke Nabae and Ken Kawai and Koji Fukagata",
year = "2019",
month = "6",
day = "4",
doi = "10.1103/PhysRevFluids.4.064603",
language = "English",
volume = "4",
journal = "Physical Review Fluids",
issn = "2469-990X",
publisher = "American Physical Society",
number = "6",

}

TY - JOUR

T1 - Synthetic turbulent inflow generator using machine learning

AU - Fukami, Kai

AU - Nabae, Yusuke

AU - Kawai, Ken

AU - Fukagata, Koji

PY - 2019/6/4

Y1 - 2019/6/4

N2 - We propose a methodology for generating time-dependent turbulent inflow data with the aid of machine learning (ML), which has the possibility to replace conventional driver simulations or synthetic turbulent inflow generators. As for the ML model, we use an autoencoder-type convolutional neural network with a multilayer perceptron. For the test case, we study a fully developed turbulent channel flow at the friction Reynolds number of Reτ=180 for easiness of assessment. The ML models are trained using a time series of instantaneous velocity fields in a single cross section obtained by direct numerical simulation (DNS) so as to output the cross-sectional velocity field at a specified future time instant. From the a priori test in which the output from the trained ML model are recycled to the input, the spatiotemporal evolution of cross-sectional structure is found to be reasonably well reproduced by the proposed method. The turbulence statistics obtained in the a priori test are also, in general, in reasonable agreement with the DNS data, although some deviation in the flow rate was found. It is also found that the present machine-learned inflow generator is free from the spurious periodicity, unlike the conventional driver DNS in a periodic domain. As an a posteriori test, we perform DNS of inflow-outflow turbulent channel flow with the trained ML model used as a machine-learned turbulent inflow generator (MLTG) at the inlet. It is shown that the present MLTG can maintain the turbulent channel flow for a long time period sufficient to accumulate turbulent statistics, with much lower computational cost than the corresponding driver simulation. It is also demonstrated that we can obtain accurate turbulent statistics by properly correcting the deviation in the flow rate.

AB - We propose a methodology for generating time-dependent turbulent inflow data with the aid of machine learning (ML), which has the possibility to replace conventional driver simulations or synthetic turbulent inflow generators. As for the ML model, we use an autoencoder-type convolutional neural network with a multilayer perceptron. For the test case, we study a fully developed turbulent channel flow at the friction Reynolds number of Reτ=180 for easiness of assessment. The ML models are trained using a time series of instantaneous velocity fields in a single cross section obtained by direct numerical simulation (DNS) so as to output the cross-sectional velocity field at a specified future time instant. From the a priori test in which the output from the trained ML model are recycled to the input, the spatiotemporal evolution of cross-sectional structure is found to be reasonably well reproduced by the proposed method. The turbulence statistics obtained in the a priori test are also, in general, in reasonable agreement with the DNS data, although some deviation in the flow rate was found. It is also found that the present machine-learned inflow generator is free from the spurious periodicity, unlike the conventional driver DNS in a periodic domain. As an a posteriori test, we perform DNS of inflow-outflow turbulent channel flow with the trained ML model used as a machine-learned turbulent inflow generator (MLTG) at the inlet. It is shown that the present MLTG can maintain the turbulent channel flow for a long time period sufficient to accumulate turbulent statistics, with much lower computational cost than the corresponding driver simulation. It is also demonstrated that we can obtain accurate turbulent statistics by properly correcting the deviation in the flow rate.

UR - http://www.scopus.com/inward/record.url?scp=85068975949&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85068975949&partnerID=8YFLogxK

U2 - 10.1103/PhysRevFluids.4.064603

DO - 10.1103/PhysRevFluids.4.064603

M3 - Article

AN - SCOPUS:85068975949

VL - 4

JO - Physical Review Fluids

JF - Physical Review Fluids

SN - 2469-990X

IS - 6

M1 - 064603

ER -