- Home
- Documents
*A MIXED FINITE ELEMENT APPROXIMATION OF .A MIXED FINITE ELEMENT APPROXIMATION OF ... this setting,*

prev

next

out of 31

View

215Download

0

Embed Size (px)

A MIXED FINITE ELEMENT APPROXIMATION OF

STOKES-BRINKMAN AND NS-BRINKMAN EQUATION FOR

NON-DARCIAN FLOWS

INGRAM, R.N

Abstract.

We propose a finite element discretization of the Brinkman equation for modeling non-Darcianfluid flow by allowing the Brinkman viscosity and permeability K 0 in solid obstacles, andK in fluid domain. In this context, the Brinkman parameters are generally highly discontinuous.Furthermore, we consider non-generic constraints: non-homogeneous Dirichlet boundary conditionsu| = 6= 0 and non-solenoidal velocity u = g 6= 0 (to model sources/sinks). Couplingbetween these two conditions makes even existence of solutions subtle. We establish well-posednessof the continuous and discrete problem, a priori stability estimates, and convergence as and K 0 in solid obstacles, as K in fluid region, and as the mesh width h 0. Fornon-solenoidal Brinkman flows, we include a small data condition to ensure existence of solutions(idea applies directly to the steady Navier-Stokes equations). In addition, we propose a pseudo-skew-symmetrization of the discrete convective term

R

u v w required for analysis of discrete

non-solenoidal Brinkman problem.

1. Introduction. This report considers the approximation of high velocity flowthrough complex geometries involving pores. Motivating examples include the flowthrough closely placed turbines on a windfarm [12], [24] and the high velocity flowof helium gas through a packed bed of (tennis-ball sized, uranium fuel) spheres ina pebble bed nuclear reactor [28], [27]. In both applications, the fluid velocities aretoo large to model accurately with Darcys equation and the pore geometry is toocomplex to approximate by the Navier-Stokes equations (NSE) in the pore regionwith no-slip boundary conditions on the solid obstacles. Therefore, appropriate forthis setting, we propose a finite element method for the Brinkman model, beginningwith the equilibrium case.

Problem 1.1. (NS-Brinkman model) For incompressible, viscous fluid flow in, find velocity u and pressure p satisfying

2 (D(u)) + u u + p + K1u = f, in

u = g, in u = , in

Here, Rd is an open domain for d = 2 or 3 consisting of both the pores andsolid obstacles, D

(u)

= 0.5(u +(u

)t) is the deformation tensor, f represents

body forces, g represents sources and/or sinks in , K is the permeability tensor, is the kinematical viscosity, and is the Brinkman viscosity. In particular, we fix0 <

f - The Fluid Domain

= f

K = kf

SampleDomain = f p s

s

= s

K = ks

K = kp

= p

p

p

s

p

p

f

fp = f p

f

s

s

fs = f s

p

p

s

s

ps = p s

Fig. 1.1. Sample subdomains with parameters , k, black regions not part of indicated. (top-left) Problem domain fps = , (top-right) Fluid-Porous domain fp, (bottom-left) Fluid-Solid domain fs, (bottom-right) Porous-Solid domain ps

For model parameters and K of order O(1), the numerical analysis of theBrinkman model fits within the framework for the abstract error analysis of the NSE,e.g. [11], [32]. However, the targeted applications of the Brinkman model are oftenhighly non-generic flows involving

complex geometries, i.e. dense swarm of porous and solid obstacles highly discontinuous parameters and K non-homogeneous boundary conditions, i.e. u| = 6= 0 general divergence conditions on velocity, i.e. u = g 6= 0

Thus, we consider herein the numerical analysis associated with the asymptotic limitsand rates of convergence as the discretization parameter h tend to 0. The last twoconditions u| 6= 0 and u 6= 0 in are necessary for many natural and industrialflows in porous media.

We derive a weak formulation of Stokes and NS-Brinkman models in Section 2.Note that Hopf proved in [15] that solutions to the steady NSE exist for generalboundary data under certain restrictions on for the case u = 0. In Section 2.2,we note coupling between u| = and u = g 6= 0 preventing a general existenceresult for nonzero boundary conditions and nonzero divergence. Our analysis is basedon the construction of an extension operator u of boundary data satisfying theconstraint u = g. We show that for g L2 () and H1/2 () satisfying

g 0, or g has compact support in ,

g = 0, and g small enough, or

g has compact support in , and g and is small enough2

there exists a solution(u, p

) H1 () L2 () to the NS-Brinkman Problem 1.1.

Furthermore, we show that the continuous (Section 2.3) and discrete (Section 3.1)Stokes- and NS-Brinkman models are well-posed (with small data for the nonlinearproblem). We derive a priori estimates for u with explicit dependence on , , andK. For > 0, both the continuous and discrete velocities for both Stokes-Brinkmanand NSE-Brinkman are of order O(

/) in H1 in the solid obstacles embedded in

and O(1/) in H1 in all ; hence, for fixed > 0, u is uniformly stable with respectto with respect to 0.

For the numerical scheme, we provide a condition for interpolating non-smoothboundary data used in the analysis of the finite element discretization of the Brinkmanmodel. We also propose an innovative (explicitly pseudo-skew symmetrized, definedin Section 3, for general g) discrete form for the convective term u u . In Section3.1, we show that the proposed conforming finite element discretization provides aconvergent approximation u,h of u as h 0 uniformly with respect to the penaltyparameter . In Section 3.2, letting u be a solution to the Stokes problem in thepurely fluid domain f with no-slip boundary conditions on the solid obstacles,we show that the the discrete Brinkman velocity u,h converges to u as h, 0 suchthat

u,h u

H1() C

(

+u u,h

H1()

)

Finally, we provide numerical validations of our theory in Section 4.

1.1. Overview of Brinkman flow model. Whereas Darcys law assumes thatvelocity is proportional to the pressure gradient for a particular porous medium,Brinkman noted that, in general, the viscous effects must also be taken into accountto model flow accurately through porous media, see [6], [7]. Heuristic generalizationsof Darcys law have been considered to model non-Darcy flows in porous media (e.g.[14], [18], [26], [5]). Along with heuristic developments, theoretical justifications existfor the Brinkman model as an asymptotic approximation to the NSE, e.g. see [1], [16]and references therein. Straughan presents several of the most popular non-Darcymodels for flow in porous media in [30] (a well-cited compilation of his and otherscontributions to this theory).

The Brinkman model has been applied to approximate non-Darcian flows in avariety of contexts; e.g. it is used to model oil filtration flows [17], groundwater flows[8], forced convective flows in metal foam-filled pipes (used in the cooling of electronicequipment) [23], gas diffusion through fuel cell membranes [13], Casson fluid flow inporous media (e.g. blood flow in vessels obstructed by fatty plaques and clots) [9],and interstitial fluid flow through muscle cells [31] with good accuracy. The Brinkmanequation is also used to model turbulence in porous media in the macroscopic scales[19] (for a discussion concerning turbulence modeling at the macroscopic versus themicroscopic pore level see [25]).

Numerical analysis of a discretization of the Stokes and NS-Brinkman flow modelis limited. In [33], Xie et.al. provide an innovative numerical analysis of the Stokes-Brinkman equations with a condition that ensures stable finite element spaces forthe discrete Stokes-Brinkman equation in the limiting condition for high Reynoldsnumber. In [2], Angot provides a beautifully detailed error analysis for the continuousStokes-Brinkman fluid velocity in fluid-porous and fluid-solid domains compared toDarcy-Stokes velocities.

3

1.2. Approximating Brinkman flow in . Solving the NSE in the pores of with no-slip boundary conditions at solid interfaces or non-stationary and/or complexdomain boundaries is cumbersome at best and most often simply not feasible [2], [3].Furthermore, the coupling condition between Stokes flow domains and Darcy flowdomains (used for flow in porous media) is physically unresolved even though theBeavers-Joseph-Saffman (BJS) interface condition is widely accepted and generallyused in practice [4], [29], [21]. Furthermore, Layton et.al show in [21] that coupledStokes-Darcy flow using the BJS interface condition is well-posed, but such a conclu-sion has not been verified for the nonlinear NS-Darcy coupling. In addition, furthercomplications arise because Stokes velocity has meaning of a pointwise velocityand Darcy velocity has meaning of an averaged velocity providing an unresolvedcompatability issue between these two velocities, see e.g. [18], [26], [5].

It is exactly these shortcomings in coupling Stokes or NSE with Darcys equa-tions that are the strengths of the Brinkman flow model. To this end, we considerthe penalized Brinkman problem formulated and described in [20] with convergenceanalysis in [2]. In particular, when approximating flows in , we want u be as smallas possible inside all solid obstacles s and recover the no-slip condition on eachsolid interface s. This is attained by imposing a large Brinkman viscosity andsmall permeability k in s. In addition, in the purely fluid region f , there areno medium obstacles impeding the flow; thus, the permeability k in f should belarge. Consider a small, parameter 0 <

We write for = f , p, s, fp, fs, ps, and fps such that

fp := f pfs := f sps := p sfps :=

See Figure 1.1 for an illustration.

We assume that > 0 is constant in . Also, >