Optimising a Machine Learning Model for Reynolds Averaged Turbulence Modelling of Internal Flows

Anthony Man; Mohammad Jadidi; Amir Keshmiri; Hujun Yin; Yasser Mahmoudi Larimi

Optimising a Machine Learning Model for Reynolds Averaged Turbulence Modelling of Internal Flows

Anthony Man, Mohammad Jadidi, Amir Keshmiri, Hujun Yin, Yasser Mahmoudi Larimi

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

142 Downloads (Pure)

Abstract

Various machine learning models in recent years have demonstrated capability in improving the accuracy of Reynoldsaveraged Navier-Stokes (RANS) simulations. One such exampleis the tensor-basis neural network (TBNN), which has been shown to introduce improvements that would be challenging or impossible for RANS approaches to model, such as anisotropic normal Reynolds stresses. This paper reports on a hyperparameter tuning exercise, to find optimal parameter settings for TBNNs predicting Reynolds stress anisotropy in channel flow, and these are also used in TBNNs deployed on internal flow over a forward-backward facing step. The accuracy of ensemble TBNN predictions is investigated, and it is shown that further improvements to Reynolds stress anisotropy predictions can be made when an ensemble size of 25 or more is used.

Original language	English
Title of host publication	16TH INTERNATIONAL CONFERENCE ON HEAT TRANSFER, FLUID MECHANICS AND THERMODYNAMICS AND EDITORIAL BOARD OF APPLIED THERMAL ENGINEERING
Number of pages	6
Publication status	Accepted/In press - 13 May 2022

Access to Document

HEFAT_2022_manuscript_final_submission2

Cite this

@inproceedings{8d022073eb524629a107024083043718,

title = "Optimising a Machine Learning Model for Reynolds Averaged Turbulence Modelling of Internal Flows",

abstract = "Various machine learning models in recent years have demonstrated capability in improving the accuracy of Reynoldsaveraged Navier-Stokes (RANS) simulations. One such exampleis the tensor-basis neural network (TBNN), which has been shown to introduce improvements that would be challenging or impossible for RANS approaches to model, such as anisotropic normal Reynolds stresses. This paper reports on a hyperparameter tuning exercise, to find optimal parameter settings for TBNNs predicting Reynolds stress anisotropy in channel flow, and these are also used in TBNNs deployed on internal flow over a forward-backward facing step. The accuracy of ensemble TBNN predictions is investigated, and it is shown that further improvements to Reynolds stress anisotropy predictions can be made when an ensemble size of 25 or more is used.",

author = "Anthony Man and Mohammad Jadidi and Amir Keshmiri and Hujun Yin and {Mahmoudi Larimi}, Yasser",

year = "2022",

month = may,

day = "13",

language = "English",

booktitle = "16TH INTERNATIONAL CONFERENCE ON HEAT TRANSFER, FLUID MECHANICS AND THERMODYNAMICS AND EDITORIAL BOARD OF APPLIED THERMAL ENGINEERING",

}

Optimising a Machine Learning Model for Reynolds Averaged Turbulence Modelling of Internal Flows. / Man, Anthony; Jadidi, Mohammad; Keshmiri, Amir et al.
16TH INTERNATIONAL CONFERENCE ON HEAT TRANSFER, FLUID MECHANICS AND THERMODYNAMICS AND EDITORIAL BOARD OF APPLIED THERMAL ENGINEERING. 2022.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Optimising a Machine Learning Model for Reynolds Averaged Turbulence Modelling of Internal Flows

AU - Man, Anthony

AU - Jadidi, Mohammad

AU - Keshmiri, Amir

AU - Yin, Hujun

AU - Mahmoudi Larimi, Yasser

PY - 2022/5/13

Y1 - 2022/5/13

N2 - Various machine learning models in recent years have demonstrated capability in improving the accuracy of Reynoldsaveraged Navier-Stokes (RANS) simulations. One such exampleis the tensor-basis neural network (TBNN), which has been shown to introduce improvements that would be challenging or impossible for RANS approaches to model, such as anisotropic normal Reynolds stresses. This paper reports on a hyperparameter tuning exercise, to find optimal parameter settings for TBNNs predicting Reynolds stress anisotropy in channel flow, and these are also used in TBNNs deployed on internal flow over a forward-backward facing step. The accuracy of ensemble TBNN predictions is investigated, and it is shown that further improvements to Reynolds stress anisotropy predictions can be made when an ensemble size of 25 or more is used.

AB - Various machine learning models in recent years have demonstrated capability in improving the accuracy of Reynoldsaveraged Navier-Stokes (RANS) simulations. One such exampleis the tensor-basis neural network (TBNN), which has been shown to introduce improvements that would be challenging or impossible for RANS approaches to model, such as anisotropic normal Reynolds stresses. This paper reports on a hyperparameter tuning exercise, to find optimal parameter settings for TBNNs predicting Reynolds stress anisotropy in channel flow, and these are also used in TBNNs deployed on internal flow over a forward-backward facing step. The accuracy of ensemble TBNN predictions is investigated, and it is shown that further improvements to Reynolds stress anisotropy predictions can be made when an ensemble size of 25 or more is used.

M3 - Conference contribution

BT - 16TH INTERNATIONAL CONFERENCE ON HEAT TRANSFER, FLUID MECHANICS AND THERMODYNAMICS AND EDITORIAL BOARD OF APPLIED THERMAL ENGINEERING

ER -

Optimising a Machine Learning Model for Reynolds Averaged Turbulence Modelling of Internal Flows

Abstract

Access to Document

Fingerprint

Cite this