Hostname: page-component-7b7567c8c-fvjsw Total loading time: 0 Render date: 2026-11-16T09:12:45.278Z Has data issue: false hasContentIssue false

High-resolution fluid–particle interactions: a machine learning approach

Published online by Cambridge University Press:  15 March 2022

Tsimur Davydzenka
Affiliation:
College of Engineering and Applied Science, University of Wyoming, Laramie, WY 82071, USA
Pejman Tahmasebi*
Affiliation:
College of Engineering and Applied Science, University of Wyoming, Laramie, WY 82071, USA
*
† Email address for correspondence: ptahmase@uwyo.edu

Abstract

Modelling of fluid–particle interactions is a major area of research in many fields of science and engineering. There are several techniques that allow modelling of such interactions, among which the coupling of computational fluid dynamics (CFD) and the discrete element method (DEM) is one of the most convenient solutions due to the balance between accuracy and computational costs. However, the accuracy of this method is largely dependent upon mesh size, where obtaining realistic results always comes with the necessity of using a small mesh and thereby increasing computational intensity. To compensate for the inaccuracies of using a large mesh in such modelling, and still take advantage of rapid computations, we extended the classical modelling by combining it with a machine learning model. We have conducted seven simulations where the first one is a numerical model with a fine mesh (i.e. ground truth) with a very high computational time and accuracy, the next three models are constructed on coarse meshes with considerably less accuracy and computational burden and the last three models are assisted by machine learning, where we can obtain large improvements in terms of observing fine-scale features yet based on a coarse mesh. The results of this study show that there is a great opportunity in machine learning towards improving classical fluid–particle modelling approaches by producing highly accurate models for large-scale systems in a reasonable time.

Information

Type
JFM Papers
Copyright
© The Author(s), 2022. Published by Cambridge University Press
Figure 0

Figure 1.Comparison between (a) coarse-scale and (b) fine-scale modelling for the same time step.

Figure 1

Figure 2.Numerical model set-up and the initial position of granular particles of the training dataset.

Figure 2

Figure 3.Sensitivity analysis of the effect of the number of layers and neurons on prediction accuracy, using the ELU activation function. The numbers in the matrix indicate ${R^2}$.

Figure 3

Figure 4.Our proposed ML architecture for estimating the drag force in the X and Z directions.

Figure 4

Table 1.Summary of sensitivity analysis for different parameters in our proposed ML network.

Figure 5

Table 2.Specifications used for training.

Figure 6

Figure 5.Runoff, velocity and drag force evolution for ground truth fine-scale (4 cm mesh) CFD-DEM, coarse-scale (8 cm mesh) ML-assisted modelling using the ELU activation function, coarse-scale (8 cm mesh) ML-assisted modelling using the ReLU activation function and coarse-scale (8 cm mesh) ML-assisted modelling using the Softplus activation function.

Figure 7

Figure 6.Quantitative comparison between the performances of several activation functions on the prediction capability of the proposed ML model.

Figure 8

Figure 7.Numerical model set-up and the initial position of the utilized granular system for testing the proposed ML model.

Figure 9

Table 3.Model parameters used in the granular failure simulation.

Figure 10

Figure 8.Parity plots between predicted and exact values of ${\boldsymbol{F}_{d,x}}$ and ${\boldsymbol{F}_{d,z}}$; (a) fine scale (4 cm mesh), (b) coarse scale (8 cm mesh), (c) coarse scale (16 cm mesh), and (d) coarse scale (32 cm mesh).

Figure 11

Figure 9.Runoff and drag force evolution for (a) ground truth fine-scale (4 cm mesh) CFD-DEM, (b) three coarse-scale CFD-DEM (8, 16, 32 cm meshes) and (c) coarse-scale (8, 16, 32 cm mesh) ML-assisted modelling.

Figure 12

Figure 10.Runoff and particle velocity evolution for (a) ground truth fine-scale (4 cm mesh) CFD-DEM, (b) three coarse-scale CFD-DEM (8, 16, 32 cm meshes) and (c) coarse-scale (8, 16, 32 cm meshes) ML-assisted modelling.

Figure 13

Figure 11.Fluid velocity evolution for (a) ground truth fine-scale (4 cm mesh) CFD-DEM, (b) three coarse-scale CFD-DEM (8, 16, 32 cm meshes) and (c) coarse-scale (8, 16, 32 cm meshes) ML-assisted modelling.

Figure 14

Figure 12.Comparison of average drag force evolution using fine (4 cm) mesh CFD-DEM; coarse (8 cm) mesh CFD-DEM using three most common models for calculating drag force, i.e. Koch–Hill, Di Felice, Gidaspow; and ML-assisted coarse (8 cm) CFD-DEM modelling.

Figure 15

Figure 13.Uncertainty quantification of the effect of different initializations on the proposed ML model for predicting (a) drag force, (b) particle velocity, (c) collision force and (d) kinetic energy evolution of the granular system. The pink colour indicates fluctuations of the ML-assisted modelling (8 cm), where the solid red curve indicates the average value based on the conducted simulations.

Figure 16

Figure 14.Runoff (a) and height (b) evolution of the granular system for fine-scale, coarse-scale and ML-assisted cases.

Figure 17

Figure 15.Evolution of average pressure of the fluid phase for fine-scale, coarse-scale and ML-assisted cases.

Figure 18

Figure 16.Evolution of velocity of the fluid phase for fine-scale, coarse-scale and ML-assisted cases; (a) average, (b) median.