Eventos Anais de eventos
ENCIT 2020
18th Brazilian Congress of Thermal Sciences and Engineering
MACHINE LEARNING PREDICTIONS OF THE TURBULENT FLOW IN THE SQUARE-DUCT EMPLOYING A TRANSPORT EQUATION FOR THE REYNOLDS STRESS TENSOR
Submission Author:
Matheus de Souza Santos Macedo , RJ
Co-Authors:
Matheus de Souza Santos Macedo, Roney Thompson, Matheus Cruz
Presenter: Matheus de Souza Santos Macedo
doi://10.26678/ABCM.ENCIT2020.CIT20-0401
Abstract
The use of Machine Learning (ML) techniques to correct RANS simulations has been recently explored by a number of works. In the present paper a novel approach for these corrections is introduced. For the first time, the ML target was not a turbulent quantity to be directly injected into the mean-momentum balance. A transport equation for the Reynolds Stress Tensor R, fueled by a ML predicted source-term, is presented and employed to correct the turbulent flow in a square-duct. DNS data was used to train Neural Networks (NN) which were employed to predict the source-term. Predictions from the NN were then injected into the RANS environment through a data-driven turbulence model, which coupled the proposed Reynolds Stress' transport equation with the mean-momentum equations and a pressure-velocity correlation. The data-driven model consistently corrected the baseline RANS velocity field and turbulent stresses of the square-duct flow.
Keywords
Turbulence, machine learning, RSTE, Square Duct, OpenFOAM
DOWNLOAD PDF VIEW PRESENTATION
