Data-driven CFD of turbulent bubbly flows using machine learning

Researchers : Hokyo Jung, Serin Yoon, Youngjae Kim in collaboration with SNUST, SNU & KNU

Motivation & Objectives

-For safety improvement of nuclear power plant, it is necessary to understand and estimate turbulent bubbly flows

-Previous models for bubble size has insufficient predictive capabilities

-Because of complex and nonlinear interaction between the bubble and turbulence

Background - ANN

-Advantages of artificial neural network (ANN) as a modeling method

- It can detect complex nonlinear relationships between parameters

: It is very flexible and effective for nonlinear problems

-It needs no pre-defined model, assumption and prior knowledge

: It is useful for modeling variables like 𝐷_𝑠𝑚 that has many parameters

-Process of using ANN to model 𝐷_𝑠𝑚/𝑅

Data-driven modeling

-Data-driven D_sm modeling in turbulent bubbly flows using ANN

-Modeling procedure

-Collect data from experiments or literatures

-Data-driven D_sm modeling in turbulent bubbly flows using ANN

RANS with ANN-based D_sm model

-Structure of data-driven RANS simulation using ANN-based model for 𝐷_𝑠𝑚

-Governing equations : An Eulerian-Eulerian method

-Validation results and comparison with previous studies

Errors from RANS with ANN-based model were significantly lower than previous models

Page updated

Google Sites

Report abuse