15:45
Computational Fluid and Solid Mechanics: Computational Fluid Dynamics (Part 3)
Chair: Fred Vermolen
15:45
25 mins
|
Numerical Investigation of the Boussinesq equations through a Subgrid Artificial Viscosity Method
Medine Demir, Songül Kaya
Abstract: Natural convection is a mechanism, in which the fluid motion is generated by the density differences occuring due to temperauture gradients. It has a wide range of industrial applications as well as nature such as free air cooling fluid flows around a heat-dissipation fins, solar ponds, etc.. This paper considers the approximate solutions of natural convection fluid flows which are governed by the Boussinesq system. As is the case with all multiphysics flow problems, simulation of the Boussinesq system consisting of the incompressible Navier-Stokes equations (NSE) together with the heat transport equation can be very expensive. Thus, the efficient and accurent numerical solution of these flows is known to be the core of many applications.
In this study, we propose a mixed, conforming subgrid artificial viscosity method discussed in the study of [1] for the numerical simulation of the Boussinesq system. In this method, the stability is achieved by adding an artificial viscosity on a fine scale and then removed only on the coarse mesh scale. We present a complete unconditional stability result of the method. Moreover, the efficiency and accurancy of the proposed algorithm is demonstrated through several numerical experiments.
|
16:10
25 mins
|
Multistage Preconditioning for Adaptive Discretization of Porous Media Two-Phase Flow
Birane Kane
Abstract: We present a constrained pressure residual (CPR) two-stage preconditioner applied to a discontinuous Galerkin discretization of a two-phase flow in strongly heterogeneous porous media. We consider a fully implicit, locally conservative, higher order discretization on adaptively generated meshes. The implementation is based on the open-source PDE software framework Dune and its PETSc binding.
|
16:35
25 mins
|
Numerical Modeling of Compressible Gas Flow in Zeolite Bed and above its Surface
Ondřej Pártl, Jiří Mikyška
Abstract: We shall present mathematical and numerical models for non-isothermal compressible flow of a mixture of two gases in a heterogeneous porous medium and in the coupled atmospheric boundary layer above the surface of this porous medium, where one of the flowing components adsorbs on the porous matrix, and we shall discuss the application of these models to the simulation of the hydration of a zeolite bed ([1], [2]).
In our models, the domain in which the flow occurs is divided into the porous medium subdomain and the free flow subdomain. In each subdomain, the flow is described by corresponding balance equations for mass, momentum and energy. On the interface between the subdomains, coupling conditions are prescribed.
In both subdomains, the spacial discretization of the governing equations is carried out via the finite volume method. As for the time discretization, the stiff source terms which describe the adsorption effects are handled via the operator splitting.
In this contribution, we shall also present the results of the simulations of the hydration of a zeolite bed via humid air which comes to this bed through a pipe.
References
[1] O Pártl, J. Jiří Mikyška, Numerical Modeling of Compressible Gas Flow in Zeolite Bed and above its Surface, to appear in International Journal of Heat and Mass Transfer.
[2] O. Pártl, M. Beneš, R. Fučík, T. Illangasekare, Numerical Modeling of Non-isothermal Compositional Compressible Gas Flow in Soil and Coupled Atmospheric Boundary Layer, Communications in Computational Physics, 26 (2019), pp. 346–388.
|
17:00
25 mins
|
Stationary flow predictions using convolutional neural networks
Matthias Eichinger, Alexander Heinlein, Axel Klawonn
Abstract: Computational Fluid Dynamics (CFD) simulations are a numerical tool to model and analyze the behavior of fluid flow. However, accurate simulations are generally very costly because they require high grid resolutions. In this paper, an alternative approach for computing flow predictions using Convolutional Neural Networks (CNNs) is described; in particular, a classical CNN as well as the U-Net architecture are used. First, the networks are trained in an expensive offline phase using flow fields computed by CFD simulations. Afterwards, the evaluation of the trained neural networks is very cheap. Here, the focus is on the dependence of the stationary flow in a channel on variations of the shape and the location of an obstacle. CNNs perform very well on validation data, where the averaged error for the best networks is below 3%. In addition to that, they also generalize very well to new data, with an averaged error below 10%.
|
|
