Sunteți pe pagina 1din 12

Advances in Water Resources 62 (2013) 215–226

Contents lists available at SciVerse ScienceDirect

Advances in Water Resources


journal homepage: www.elsevier.com/locate/advwatres

Flow and axial dispersion in a sinusoidal-walled tube: Effects of inertial


and unsteady flows
Marshall C. Richmond a, William A. Perkins a, Timothy D. Scheibe a,⇑, Adam Lambert b, Brian D. Wood b
a
Hydrology Group, Pacific Northwest National Laboratory, Richland, WA, United States
b
School of Chemical, Biological and Environmental Engineering, Oregon State University, Corvallis, OR, United States

a r t i c l e i n f o a b s t r a c t

Article history: In this work, we consider a sinusoidal-walled tube (a three-dimensional tube with sinusoidally-varying
Available online 5 July 2013 diameter) as a simplified conceptualization of flow in porous media. Direct numerical simulation using
computational fluid dynamics (CFD) methods was used to compute velocity fields by solving the
Keywords: Navier–Stokes equations, and also to numerically solve the volume averaging closure problem, for a range
Dispersion of Reynolds numbers (Re) spanning the low-Re to inertial flow regimes, including one simulation at
Volume averaging Re ¼ 449 for which unsteady flow was observed. The longitudinal dispersion observed for the flow was
Skewness
computed using a random walk particle tracking method, and this was compared to the longitudinal dis-
CFD
Pore-scale
persion predicted from a volume-averaged macroscopic mass balance using the method of volume aver-
aging; the results of the two methods were consistent. Our results are compared to experimental
measurements of dispersion in porous media and to previous theoretical results for both the low-Re,
Stokes flow regime and for values of Re representing the steady inertial regime. In the steady inertial
regime, a power-law increase in the effective longitudinal dispersion (DL ) with Re was found, and this
is consistent with previous results. This rapid rate of increase is caused by trapping of solute in expan-
sions due to flow separation (eddies). One unsteady (but non-turbulent) flow case (Re ¼ 449) was also
examined. For this case, the rate of increase of DL with Re was smaller than that observed at lower Re.
Velocity fluctuations in this regime lead to increased rates of solute mass transfer between the core flow
and separated flow regions, thus diminishing the amount of tailing caused by solute trapping in eddies
and thereby reducing longitudinal dispersion. The observed tailing was further explored through analysis
of concentration skewness (third moment) and its assymptotic convergence to conventional advection–
dispersion behavior (skewness = 0). The method of volume averaging was applied to develop a skewness
model, and demonstrated that the skewness decreases as a function of inverse square root of time. Our
particle tracking simulation results were shown to conform to this theoretical result in most of the cases
considered.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction fects of velocity variations at unresolved scales and dilution of sol-


utes associated with molecular diffusion. In geologic systems,
Dispersion in porous media is an important process in many macrodispersion is associated with spatially heterogeneous aquifer
problems in both natural and engineered systems. Examples in- properties (e.g., permeability and porosity) which cause large-scale
clude subsurface contaminant transport and remediation, chemical variations in advective velocity. Microdispersion is associated with
transport and reaction in packed bed reactors, and drinking water molecular diffusion coupled with pore-scale velocity variations
filtration systems. Dispersion as a mass transport process in porous including variations caused by variable pore sizes and geometries
media flows has been extensively studied, and a number of excel- as well as variations within a single pore caused by wall friction.
lent reviews of theoretical developments have been published (e.g., Dybbs and Edwards [7] defined four distinct regimes of flow in
[1–4]). A brief summary of the history of dispersion studies is given porous media as a function of the Reynolds number Re: (1) creep-
by Wood [5], and a compilation of available experimental data is ing flow (or Stokes flow), Re < 5  10, in which flow is dominated
provided by Delgado [6]. Dispersion is a macroscopic phenomenon by viscous forces; (2) steady inertial (nonlinear) flow,
that incorporates both spreading of solutes that arises from the ef- 5  10 < Re < 200þ, in which inertial forces become significant
relative to viscous forces; (3) unsteady laminar flow,
200þ < Re < 350þ, in which flow is no longer steady; and (4) cha-
⇑ Corresponding author. otic flow, Re > 350þ, in which turbulence is fully developed. In the
E-mail address: tim.scheibe@pnnl.gov (T.D. Scheibe).

0309-1708/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.advwatres.2013.06.014
216 M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226

work reported here, we employed a range of Re that spans these that formed in the expansions trapped solute particles causing long
four regimes (Re = 1, 15, 100, and 449). (For additional discussion tailing in breakthrough curves, and leading to behavior that could
and the specific definition of Re used here, see [5]). Although nat- be described in terms of a mobile-immobile model with either a
ural porous media flows are commonly assumed to fall into the single-rate or multiple-rate mass transfer model defining exchange
low-Re (Stokes flow) regime because of low velocities, inertial of solute between the core flow and the eddies. DeGroot and Stra-
and unsteady flow effects on dispersion may be important for some atman [27] evaluated closure approximations associated with vol-
problems including certain contaminant remediation processes [8] ume-averaging by direct numerical simulation in periodic unit cell
and contaminant transport near wells and in the hyporheic zone geometry at Re up to 100. They found that the dispersion coeffi-
[9–11]. An important example of large and dynamic fluxes in the cient was a function of Pe alone at low Re, but at Re > 50 the dis-
hyporheic zone exists at the U.S. DOE Hanford 300A Site along persion coefficient also depended on Re. Bouquain et al. [28]
the Columbia River in southeastern Washington State. At this site, performed two-dimensional simulations of flow in a sinusoidal-
where investigations of the fate and transport of a uranium plume walled tube with Re up to 100. They identified two competing ef-
are ongoing, large and rapid fluctuations in river stage (due to up- fects of inertial flow on axial dispersion: (1) trapping of particles
stream dam operations) combine with highly permeable sedi- in side eddies, which increased the time to achieve assymptotic
ments to create a highly dynamic flow field [12,13]. Model behavior, and (2) decreased distance between streamlines in the
estimates given in [12] (their Fig. 7) suggest that pore velocities central jet region, which increased mixing and thereby decreased
are as large as several tens of meters per day, indicating that there the time to assymptotic conditions. Our work is similar in concept
could be local regions with signficant inertial effects on flow. In to that of Wood [5], who developed a macroscopic description of
this paper, we examine the effects of inertial and unsteady flows dispersion based on volume-averaging of velocity fluctuations in
on pore-scale microdispersion using a symmetric sinusoidal- a periodic unit cell with numerical closure, and used their model
walled tube as a model system. The tube has a sinusoidally-varying to evaluate inertial effects on dispersion (Re up to 250), but not into
tube diameter, and can be considered an idealized conceptual the unsteady regime. We extend that work by considering a differ-
model of flow along a streamline in a porous medium in which ent model geometry (a 3D sinusoidal-walled tube rather than a
constrictions represent pore throats and expansions represent pore periodic unit cell), by evaluating effects of unsteady flow condi-
bodies. tions (Re  500), and by cross-validating the numerical volume
Hoagland and Prudhomme [14] numerically investigated the ef- averaging solution against a random walk particle tracking
fects of converging–diverging sinusoidal tube structure on disper- method.
sion, building on the concepts of the classic work of Taylor [15] and
Aris [16] for a straight tube, but considered only two-dimensional 2. Wavy tube geometry and computational mesh
(2D) Stokes flow (no inertial effects). Wang and Vanka [17] and
Stone and Vanka [18] studied flow and heat transfer in a 2D sinu- Following Hoagland and Prudhomme [14], the shape of the
soidal-walled channel at Re up to 700, and observed that the tran- sinusoidal wall (Fig. 1) is defined as a sinusoid in terms of the aver-
sition to unsteady flow (in their case around Re = 160) led to age radius:
increased rates of heat transfer because of increased mixing of   
the core and near-wall fluids associated with Kelvin–Helmholtz
2p x
RðxÞ ¼ Ro 1 þ e sin ð1Þ
type instabilities of the shear layers. Souto and Moyne [19] per- L
formed numerical experiments on two-dimensional periodic por- where Ro is the average radius, L is the wavelength, and e is the
ous media at Re up to 100. They observed an increased influence dimensionless amplitude.
of Reynolds number on longitudinal dispersion for Re > 20 and The sinusoidal tube dimensions correspond to the Ralph [29]
also noted that this influence was less pronounced for disordered experiments, where RðxÞ varies between 0.001 m and 0.003 m
unit cells than for ordered unit cells. Cao and Kitanidis [20] studied (mean tube radius R0 = 0.002 m) and L ¼ 0:01 m. The dimension-
dispersion and dilution in a 2D sinusoidal-walled channel in the less amplitude of the wave is e ¼ 0:5 (similar to the Hoagland
Stokes regime, and compared derived macroscopic solutions with and Prudhomme [14] case of e ¼ 0:4).
results of direct numerical simulation. Bijeljic and colleagues The Reynolds and Peclet numbers are defined as
[21,22] used a pore-network modeling method to study dispersion
in porous media, and identified inertial effects as a possible expla-
nation for changes in effective dispersion at high Re. Cardenas [23]
simulated flow in single pores with varying geometries and for
Re < 2. He observed non-Fickian transport (power-law residence
time distributions) for the larger values of Re and attributed it to
eddy formation in the pores and slow mass transfer between ed-
dies (dead zones) and the rest of the flow field (active flow). Bolster
et al. [24] used an approximate analytical solution for flow in a 2D
wavy-walled channel to study the impact of molecular diffusion
and geometry on effective dispersion using both the method of lo-
cal moments and a random walk particle tracking method. They
found that periodic fluctuations of the channel aperture could lead
to both increases and decreases of the effective dispersion. How-
ever, they also considered only low-Re (Stokes) flows. Cardenas
et al. [25] simulated flow in a rough assymetric fracture and ob-
served formation of a large eddy that trapped particles and caused
tailing and persistent non-Fickian behavior over the full range of Re
considered (up to Re ¼ 100). Lambert [26] performed numerical
simulations of flow and dispersion in a 3D sinusoidal-walled tube
for Re up to 14.5, and found that inertial effects led to flow separa-
tion in expansions that strongly impacted dispersion. The eddies Fig. 1. Geometry of the sinusoidal wavy tube.
M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226 217

Fig. 2. A picture of the coarse mesh (28 k cells, left) and medium resolution mesh (150 k cells, right).

qU o Do 3. Governing equations and solution methods


Re ¼ ð2Þ
l
3.1. Hydrodynamics
U o Do
Pe ¼ ð3Þ The hydrodynamics (velocity and pressure fields) in the sinusoi-
Dm
dal tube are computed using the TETHYS (Transient Energy Transport
HYdrodynamics Simulator) code developed at Pacific Northwest
where the length scale is the mean tube diameter, Do ¼ 2Ro , the
National Laboratory (PNNL). TETHYS is a parallel code that uses
velocity scale is the bulk velocity based on the mean tube radius,
the finite volume method [34] to solve the governing equations
U o ¼ Q =pR2o ; Q is the volume flux of fluid, and Dm is the coefficient
of mass, momentum, and scalar conservation on an unstructured
of molecular diffusion. The kinematic viscosity (m ¼ l=q) is that of
mesh. The starting point for the finite volume method is the gov-
the glycerol solution used in the Ralph [29] experiments
erning equations in integral form:
(5:33  106 m2/s). The Reynolds number was specified for each
Z Z
simulation by prescribing a pressure drop between the inlet and @
outlet faces so as to achieve the desired mean flow velocity; the
qdV þ qv b  nbr dA ¼ 0 ð4Þ
@t Vb Abr
Peclet number was then specified by varying the molecular diffu-
sion coefficient as necessary. Z Z
@
Using mesh generation software [30], three nearly-orthogonal qui dV þ qui vb  nbr dS
@t Vb Abr
meshes, for a single wavelength were created. Three mesh resolu- Z Z
tions were used that consisted of 28,000; 150,000; and 450,000 ¼ ðsij ij  pii Þ  nbr dS þ fi dV ð5Þ
hexahedral cells (Fig. 2) for use in mesh independence tests. Simu- Abr Vb
lations in the steady regime (Re 6 100) used a single wavelength as
Z Z
the computational domain. For the unsteady case (Re ¼ 449) we @
used a domain consisting of three tube wavelengths to allow the
q/dV þ q/vb  nbr dA
@t Vb Abr
three-dimensional, unsteady flow structures to evolve with mini- Z Z
mal interference from the periodic boundary conditions at the do- ¼ Cr/  nbr dA þ q/ dV ð6Þ
Abr Vb
main ends. This domain length (three wavelengths) was selected
based on qualitative judgement and consistency with previously- where V b is volume; Abr is the surface area; v b is the fluid velocity
reported works that utilized from one to four unit cells [31–33] vector with components ui ; nbr is the outward-directed unit nor-
in order to provide confidence in the results while maintaining mal; / is a generic scalar quantity; fi is a body force; q/ is a source
computational feasibility. The same domain configurations (one term; C is diffusivity of /; and q is the fluid density.
wavelength for steady flows; three wavelengths for unsteady For an incompressible Newtonian fluid the viscous stress tensor
flows) were utilized for both (1) direct numerical simulation of is given by:
hydrodynamics with particle tracking and (2) solution of the meth-  
od of volume-averaging (MVA) equations as described below. A @ui @uj
sij ¼ l þ
summary of the simulation conditions for each of the four values @xj @xi
of Re considered is provided in Table 1.
where l is the fluid dynamic viscosity.
The finite volume method [34] is used to discretize the govern-
ing Eqs. (4)–(6) on a computational mesh consisting of arbitrarily
Table 1 shaped cells (e.g., hexahedral, tetrahedral). A co-located storage
Summary of simulated conditions for the four values of Reynolds number (Re) scheme is used where all computed variables are defined at the cen-
considered. U 0 is the average (bulk) velocity (m/s) and T F is the flow-through time for
troid of each mesh cell. A fully implicit solution scheme is used
a single unit cell (seconds).
where temporal terms are discretized using a three-time-level
Re No. of unit cells Flow regime U 0 (m/s) T F (s) 2nd order accurate scheme [34]. Convective fluxes are computed
1 One Steady 1.33E03 7.51 using deferred correction [34] with the choice of several high-reso-
15 One Steady 2.00E02 0.501 lution schemes such as central differencing (CDS), second-order up-
100 One Steady 1.33E01 0.0751
wind, and MUSCL. These are implemented using the normalized
449 Three Unsteady 5.98E01 0.0167
variable and space formulation (NVSF) of Darwish and Moukalled
218 M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226

[35] modified for unstructured meshes [36]. All hydrodynamic simu- where i is x; y, or z. Each component is solved in TETHYS using the fi-
lations reported here use CDS for the convective fluxes. Diffusive nite volume method outlined in Section 3.1 with the convective
fluxes are discretized using the methods of [34] for unstructured terms computed using the MUSCL scheme.
meshes that reduce to CDS for meshes composed of hexahedral cells. The solution of b is used to compute the diagonal components
Coupling between the pressure and velocity fields is done using of the dispersivity tensor in Eq. 7 as
the iterative SIMPLE algorithm [37] modified for co-located vari- ( Z ) Z
1 1  
ables on unstructured meshes using Rhie–Chow interpolation [34]. Dii ¼ Dm 1 þ ni bi dA  ui  hui ib bi dV ð12Þ
Discretizing the governing equations and implementing the Vb Ab r Vb Vb
SIMPLE algorithm results in systems of linear algebraic equations
that are solved using iterative methods available in PETSc [38].
TETHYS is written in C++ and uses MPI [39] and the Global Array
3.3. Axial dispersion coefficient by particle tracking
toolkit [40,41] for its parallel implementation.
On solid walls the velocity is set to zero (no slip). Periodic We also compute the axial dispersion coefficient using a ran-
boundary conditions are assigned in the streamwise direction on dom walk particle tracking (PT) method [47,48,24]. The asymptotic
the ends of the tube. A constant pressure drop (between two ends dispersion coefficient is computed from the rate of change of the
of the sinusoidal tube) is specified to drive the flow. For each simu- variance of the particle positions:
lation, the pressure drop is chosen to produce the desired mass flux.
1 dr2 1 dX N
DL ðt Þ ¼ ¼ ðxi  xi Þ2 ð13Þ
2 dt 2N dt i¼1
3.2. Axial dispersion coefficient by volume averaging
where DL is component of the dispersivity tensor in the x-direction,
The axial dispersion coefficient is calculated using the method r2 ðtÞ is the variance of particle position, xi ðtÞ is the position of par-
of volume averaging (MVA) presented by Carbonell and Whitaker ticle i, and xi ðtÞ is the average position of all particles in the cloud.
[42], Whitaker [43] and Wood [5]. A key modification in this work, The stochastic particle trajectories are determined through the
following [44,45], is the addition of a time derivative term to the solution of the Langevin equation which corresponds to the differ-
MVA equations to allow for its application to unsteady flows. ential form of the scalar transport Eq. (6) when cast into the form
The total dispersion tensor is the sum of the effective diffusivity, of the Fokker–Planck equation [49]. The particle trajectories are ad-
Deff , and hydrodynamic dispersion, Db : vanced in time using the discrete form of the Langevin equation
D ¼ Deff þ Db solved using the Euler method:
( Z ) Z  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 1 
¼ Dm I þ nbr bdA  v b  hvb ib bdV ð7Þ xðt þ DtÞ ¼ xðtÞ þ uðxðtÞÞDt þ 2Dm Dt n ð14Þ
V b Abr V b Vb
where n is a random number with a zero mean and unit variance, u
where Sc is the wall boundary area, Dm is the molecular diffusivity, b is the x-velocity component, and Dt is the time step. Similar equa-
is a vector field, and hv ib is the average velocity vector computed as tions are used to compute the particle trajectories in the y and z
Z directions. The velocity field at the particle locations is estimated
1
hv b ib ¼ v b dV using linear interpolation in space and time. A reflection boundary
Vb Vb
condition is used at the tube walls and periodic boundaries are used
The vector field b is determined from the solution of at the tube ends. A set of virtual coordinates is used for each particle
  to track its absolute displacement through multiple cycles of the
@b 
þ r  v b b  Dm r2 b ¼  v b  hv b ib ð8Þ periodic sinusoidal tube solutions.
@t In all cases described here, at total of 5000 particles were used
in the calculations. We also ran one case (Pe = 100, Re = 100) using
B:C:1  n  Dm rb ¼ nDm ; at Abr ð9Þ
10000 particles as a test, and observed that the evolution of parti-
cle location variance was nearly identical to the same case with
B:C:2 bðxÞ ¼ bðx þ li Þ; at Abe ð10Þ
5000 particles. Additional confidence that the number of particles
In this work, we have followed the development illustrated by used was sufficient is provided by the close match between the
[46] to develop a local but transient closure problem. Our primary particle tracking results and the MVA results (described in Sec-
purpose in this formulation was to allow a computationally stable tion 5.3). Particles were initially placed in a random distribution
method for computing the time-asymptotic dispersion tensor com- across the plane of the tube at its maximum diameter. This reduced
ponents. Although the assumptions required for attaining the the amount of simulation time required to compute the asymptotic
time–space locality may not be met at the highest Reynolds num- dispersion coefficient which is half the slope of the variance of the
bers (where the flow is transient), we used this form of the closure particle position determined using a linear fit (see Fig. 8).
problem nonetheless to predict values of the dispersion tensor dur-
ing transient flows. We emphasize, however, that these predictions 4. Validation of simulation methods
are still essentially time asymptotic in the sense that the influence
of the initial conditions of the system do not contribute to the tran- Validation tests are summarized here to verify the accuracy of
sient behavior. Instead, the transience in the dispersion tensor is the numerical schemes used to solve the governing equations de-
caused entirely by the time-fluctuating velocity field. This is dis- scribed in Section 3.
cussed in additional detail in Section 5.3.
Eq. 8 is cast into the form of Eq. 6 for each component, i, of the b 4.1. Hydrodynamics
vector:
Z Z
@ Validation tests of the hydrodynamics computations in TETHYS
qbi dV þ qv b  nbr bi dA were performed for several canonical laminar flow cases: flow be-
@t Vb Ab r
Z Z   tween parallel plates, flow in a tube with circular cross section, lid
¼ qDm nbr  rbi dA  q ui  hui ib dV ð11Þ driven cavity flow, and flow past a circular cylinder. In the cases of
Ab r Vb parallel plate and tube flow, the computed velocity profiles and
M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226 219

Table 2
Comparison of Mean Drag Coefficient (C D ), Length of Wake Bubble (Lw =d) and Strouhal Number (St) for flow past a two-dimensional cylinder (of diameter d) compared to previous
experiments and computations [51–57]

Re 20 40 80 300
CD Lw =d CD Lw =d CD St CD St

Tritton [51] 2.22 – 1.48 – 1.29 – – –


Weiselsberger [52] 2.05 – 1.70 – 1.45 – 1.22 –
Dennis and Chang [53] 2.05 0.94 1.52 2.35 – – – –
Fornberg [54] 2.00 0.91 1.50 2.24 – – – –
Mittal and Balachandar [55] – – – – – – 1.37 0.21
Williamson [56] – – – – – 0.15 – 0.20
Ye et al. [57] 2.03 0.92 1.52 2.27 1.37 0.15 1.38 0.21
Present work 2.09 0.91 1.55 2.17 1.38 0.15 1.37 0.21

friction factors matched the analytical solutions for Couette and As shown in Fig. 3, in a straight tube, both methods were in excel-
Poiseulle flow, respectively, as the mesh was refined to use at least lent agreement with the theoretical Taylor–Aris values given by
10 cells in the cross stream direction.
An additional validation test for a more complex (but still stea- DL Pe2
¼1þ ð15Þ
dy flow) case was performed for a two-dimensional lid driven cav- Dm 192
ity corresponding to the benchmark computations of Ghia et al.
where Dm is the molecular diffusivity of the fluid and the Peclet
[50]. Horizontal and vertical velocity profiles through the center-
number is computed as
line of the cavity were compared to the benchmark solutions using
different grid meshes for Re ¼ 100; 400; 1000; and 3200. Progres- Uo D
sive refinement of the mesh resulted in good agreement with Pe ¼
Dm
benchmark solutions.
Finally, flow past a circular cylinder case was simulated to val- where U o is the bulk longitudinal fluid velocity and D is the circular
idate the numerical solution scheme for cases exhibiting flow sep- tube diameter.
aration from a curved surface and unsteady vortex shedding. In addition, we compared our dispersion coefficients to those
Correct simulation of these kinds of features is important because calculated by Hoagland and Prudhomme [14] and Federspiel and
similar features are observed in the sinusoidal tube. A range of Rey- Fredberg [58] in the Stokes flow regime for a sinusoidal tube geom-
nolds numbers, spanning the steady and unsteady laminar flow re- etry similar to ours (e ¼ 0:4) and found good agreement across a
gimes (Re ¼ 20; 40; 80 and 300) were simulated. The mean drag range of Peclet numbers (Fig. 3).
coefficient (C D ) agreed well with several previous experimental
and computational studies summarized in Table 2. In the steady 5. Flow and dispersion in the wavy tube
regime (Re ¼ 20 and 40), the length of the separation bubble in
the cylinder wake (Lw ) compared well with previous work. At the In this section we present the hydrodynamic and longitudinal
higher Reynolds numbers (Re ¼ 80 and 300), the flow around the dispersion results for the specific sinusoidal tube case correspond-
cylinder became unsteady and the expected vortex street formed ing to the geometry used in the experiments by Ralph [29]. Note
downstream of the cylinder. The vortex shedding frequency, repre- that the experiments of [29] were performed with unspecified
sented by the Strouhal number (St), was in excellent agreement aqueous solutions of glycerol. In all our simulations, the fluid
with other studies as shown in Table 2. was assumed to be a 50 percent aqueous solution, by weight, of
glycerol at 20 °C. Fluid properties are: q ¼ 1126:3 kg m-3,
4.2. Dispersion l = 6.00  10-3 Pa s.
Steady simulations (Re ¼ 0:1  100) were simulated using a sin-
Axial dispersion coefficients computed using the MVA and PT gle tube wavelength. A periodic boundary condition was imposed
methods were validated for flow between parallel plates and tubes.
1000
10000 Simulated, 28k cell mesh
Straight Tube Theory Simulated, 150k cell mesh
Straight Tube VA Simulated, 450k cell mesh
Straight Tube PT Analytic, Sisavath et al. (2001)
Wavy Tube (ε = 0.4) Hoagland and Prudhomme (1985) Simulated, Ralph (1987)
1000 Wavy Tube (ε = 0.4) Federspeil and Fredberg (1988) 100
Experimental, Ralph (1987)
Wavy Tube (ε = 0.4) PT
Wavy Tube (ε = 0.4) VA

100 10
f
DL/D

10 1

1
0.1
0.1 1 10 100 1000
0.1 1 10 100 1000 Re
Pe
Fig. 4. Friction factors computed from simulations in this work compared with the
Fig. 3. Validation of dispersion calculations for a straight tube (Taylor-Aris) and for simulations and experiments of Ralph [29] and with the analytical solution of
the sinusoidal tube of Hoagland and Prudhomme [14]. Sisavath et al [59].
220 M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226

at each end, with a specified pressure drop that was adjusted to may not extend to the unsteady case (Re ¼ 449), to be conservative
achieve the desired Re. we used the fine mesh in that case. Once a mesh was selected we
The unsteady case was simulated using a domain consisting of then proceeded to hydrodynamics, MVA, and PT simulations.
three tube wavelengths. Using the data of [29], a pressure drop
(2100 Pa per wavelength) was chosen in the unsteady regime with
5.1. Friction factor
a target of Re  500. The simulation resulted in an actual value of
Re ¼ 449. The unsteady case was first preconditioned for a total
The friction factor is defined and calculated based on the pressure
of 16:7 TF where the flow through time, T F , is given by L=U o . The
loss between the two ends of the sinusoidal tube, in the form as:
simulation was then continued for an additional 13:5 TF during
which axial dispersion coefficients were computed. 1 D o 2 DP
We conducted a set of computational mesh dependency tests f ¼ ð16Þ
4 L qU 2o
(see Section 2) for hydrodynamic simulations corresponding to
Re ¼ 100 and volume averaging computations for Pe ¼ 100. The The computed friction factors are shown in Fig. 4 and these
changes in friction factor and longitudinal dispersion coefficient compared well with both experimental data and the low-Reynolds
between the medium (150 k cells) and the fine mesh (450 k cells) number analytical solution of Sisavath et al. [59]. The friction factor
were less than 0:5%. The medium mesh was therefore used for in the unsteady regime (Re ¼ 449), computed from averaging over
all other steady state cases. Because we recognize that this result 13:5 TF, is slightly higher than the experimental value.

Re = 0.1 Re = 1

Re = 10 Re = 15

Re = 25 Re = 100

Fig. 5. Streamlines for Steady-State Cases Re ¼ 0:1; 1; 10; 15; 25; 100.
M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226 221

5.2. Velocity field 100000


Re = 1 VA
Re = 1 PT
The evolution of the flow patterns that emerge as Re increases Re = 15 VA
10000 Re = 15 PT
through the steady regime are shown in Fig. 5. Flow remains at- Re = 100 VA
tached and is symmetrical at the two lowest Reynolds numbers Re = 100 PT
Re = 449 (average) VA
(0.1 and 1.0) where the effects of inertia are small. At Re ¼ 10, 1000 Re = 449 (average) PT
Re = 449 (transient) VA
the streamlines show that the flow is beginning to become asym- Re = 449 (transient) PT
metrical prior to undergoing separation which occurs when Re fur-

DL/D
100
ther increases to 15. The size of the separation zone grows and by
Re ¼ 100 a jet is formed that nearly bridges the gap between the
throats and avoids the side lobes of the sinusoidal tube. 10
At Re ¼ 449, transient flow appears which is shown in Fig. 6 at
several instances in time following the preconditioning period. The
1
isosurfaces of velocity magnitude reveal an unsteady flow field
composed of complex, asymmetrical three-dimensional structures.
Points of separation and reattachment occur at different locations 0.1
0.1 1 10 100 1000
on the tube wall. The unsteady vortices clearly promote advective
Pe
mixing across the entire tube cross section. The enhanced mixing
that occurs in the unsteady flow is further highlighted by compar- Fig. 7. Dispersion in the sinusoidal tube for both steady and unsteady cases. For the
ison to the time-averaged velocity isosurfaces in Fig. 6. The Re ¼ 449 case, ‘‘Average’’ denotes results for the time-averaged velocity field, and
time-averaged Re ¼ 449 solution is very similar in structure to ’’Transient’’ denotes results for the fully transient velocity field with unsteady
the steady-state flow at Re ¼ 100. As will be shown in the next sec- features.

tion, the enhanced mixing associated with unsteady flow reduces


trapping of particles in the side eddies and thus reduces the longi-

tudinal dispersion. Averaging out this unsteadiness (i.e., through a


turbulence closure model) will lead to over-prediction of longitudi-
nal dispersion.

5.3. Axial dispersion

Axial dispersion coefficients (DL ) for the steady and unsteady


hydrodynamics cases (Re ¼ 1; 15; 100; 449) were computed using
both the MVA and PT methods. The results for a range of Pe num-
bers are compared in Fig. 7 and show very good agreement be-
tween the two methods. Fig. 8 shows the variation over time for
DL computed from the transient MVA solution and shows close
agreement with the time average of DL computed from the vari-
ance of the particle cloud using PT.
As Re increases through the steady regime axial dispersion coef-
ficients increase for Pe > 1 because of the reduced mass exchange
with the side lobes of the sinusoidal tube (formation of eddies that
trap solute in the expansions). This slow exchange between mobile
(central flow) and immobile (eddies in side lobes) regions causes
long tailing of the solute plume and correspondingly large disper-
sion. As the flow becomes unsteady, mass exchange with the side
lobes is increased by transient tongues of flow that sweep out
the expansion zones (see supplemental information for an anima-
tion that illustrates this phenomenon clearly). This causes a corre-
sponding decrease in the slope of DL at large Pe that is evident for
the largest Re (unsteady flow). As expected from the velocity fields
shown in Fig. 6, DL computed from the time-averaged velocity field
is larger than that in the actual transient flow and similar in mag-
nitude to the steady values in the inertial flow range (Re  100).
Snapshots of the particle positions at two Pe (100 and 1000) are
shown in Figs. 9 and 10 at different times. These particle plots illus-
trate the decrease in particle retention as the physical mechanism
for the reduction in dispersion coefficient for the transient case as
compared to the time-averaged case.

5.4. Skewness

In the development of the MVA mass balance, there is a skew-


Fig. 6. Velocity magnitude isosurfaces for several times in the transient Re ¼ 449
simulation (upper panel) and the time averaged of the unsteady solution (lower
ness term that is usually neglected (see [43, p. 130]). The evidence
panel). The blue, green, and yellow surfaces represent u=U o = 1.2, 2.3, and 3.5, of long tailing in the particle distributions shown in Figs. 9 and 10
respectively. suggested that this assumption may not be justified for the steady
222 M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226

3.5e-03 270
Re = 449, Pe = 1.0e+02 Re=449, Pe=1.0E+02
Simulated Transient VA
3.0e-03 Transient VA Time Average
Linear fit, DL = 5.08e-03 m2/s 260
Transient PT
2.5e-03
250
Particle σ 2, m2

2.0e-03
240

DL/D
1.5e-03
230
1.0e-03

220
5.0e-04

0.0e+00 210

-5.0e-04 200
0 5 10 15 0 1 2 3 4 5 6 7 8 9 10
t/TF t/TF

Fig. 8. Particle cloud variance time history (left) and unsteady volume averaging compared to particle tracking (right) in the unsteady Re ¼ 449 case.

inertial flow cases. Significant skewness would indicate incom-


pleteness of the macroscopic advection–dispersion equation, since
it assumes transport is fully described by the first two moments of
concentration (or equivalently, particle displacements).
To explore this finding further, we have developed a new MVA
relationship that includes higher-order moments (Appendices A
and B). This analysis indicates that skewness should decrease as
the inverse square root of time. Although our numerical experi-
ments were not designed to evaluate higher-order moments, we
were able to compute estimates of skewness from the particle
tracking results and compare them to the predicted relationship.
Figs. 11–13 show the computed skewness as a function of time
(using PT) for three different Re and four different Pe. Also shown Fig. 9. Comparison of simulated particle positions for Re ¼ 449 averaged (above)
are fitted curves of the form and (transient) with Pe ¼ 100 (T = 0.0167 s).

c ¼ ðA=t1=2 Þ þ B ð17Þ

where c is particle skewness, t is time, and A and B are fitted coef-


ficients (Table 3).
In most of the cases considered, the observed skewness as a
function of time is well fitted by a functional of the prescribed
form, as indicated in the figures. The exception to this is the cases
with low Re (=1) and high Pe (>10). For these cases, the skewness
does not assyptotically approach zero,suggesting that non-Fickian
behavior is very persistent. However, these cases may also be sus-
ceptible to numerical error, as low diffusion rates and slow veloc-
ities very close to the tube walls may lead to artificial trapping of
particles that depends on grid resolution. Our grid refinement tests
were based on convergence of second moments; it is likely that
accurate simulation of higher moments (particularly for low Re/
high Pe) may require additional grid resolution near the tube walls.
Fig. 10. Comparison of simulated particle positions for Re ¼ 449 averaged (above)
Therefore, the results for those cases should be interpreted cau-
and transient (below) with Pe ¼ 1000 (T =0.0167 s).
tiously and need additional study. Overall, however, the match be-
tween the observed and theoretical form of skewness vs. time is
surprisingly good. Even some cases that do approach zero assymp- To our knowledge, this is the first analysis of higher-order mo-
totically (e.g., Fig. 13) do so slowly. Cardenas [60] observed conver- ments using MVA, and provides new insights into early-time non-
gence to Fickian residence time distributions after transport Fickian transport behavior and long-term assymptotic convergence
distances of about 10 grains in simulations of flow through 3D cu- behavior. Because the skewness analysis was not the central focus
bic packed spheres, whereas our system exhibits non-zero skew- of this work, we have not yet taken the additional step of numer-
ness at times greater than several tens of flow-through times in ically solving the closure equations (Appendix B), which would re-
some cases. These differences may be attributable to the differ- quire additional code development. The next step in this analysis
ences in simulated geometry, differences in simulated Re, or differ- would be to perform these computations, thus allowing direct esti-
ences in the way Fickian behavior was characterized (residence mation of the skewness parameters rather than fitting to the par-
time distribution versus particle position skewness). ticle observations.
M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226 223

1.6 3.0
Re = 1.0 Re = 100
1.4 Fitted Fitted
Pe = 1.0E+00 2.5 Pe = 1.0E+00
Pe = 1.0E+01 Pe = 1.0E+01
1.2 Pe = 1.0E+02 Pe = 1.0E+02
Particle Skewness, γ

Particle Skewness, γ
Pe = 1.0E+03 Pe = 1.0E+03
1.0 2.0

0.8
1.5
0.6

0.4 1.0

0.2
0.5
0.0

-0.2 0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0 20 40 60 80 100 120 140
t/TF, s t/TF, s

Fig. 11. Plot of skewness coefficients as a function of time, computed from Fig. 13. Plot of skewness coefficients as a function of time, computed from
simulated particle distributions, for Re = 1 and four different values of Pe (red simulated particle distributions, for Re = 100 and four different values of Pe (red
dashed curves). Also shown (grey solid curves) are fitted curves of the form dashed curves). Also shown (grey solid curves) are fitted curves of the form
prescribed by the MVA relationship derived in Appendix A (Eq. 17). Note that in this prescribed by the MVA relationship derived in Appendix A (Eq. 17). (For interpre-
case a single curve was fit to the cases with Pe = 10, 100 and 1000 since the fitted tation of the references to colour in this figure caption, the reader is referred to the
portions of those curves were nearly identical. (For interpretation of the references web version of this article.)
to colour in this figure caption, the reader is referred to the web version of this
article.)

Table 3
Fitted parameters for skewness relationship (Eq. (17)) as a function of Re and Pe. For
2.5 the case of Re = 100, the B parameter was fixed at zero.
Re = 15.0
Fitted Re 1 15 100
2.0 Pe = 1.0E+00
Pe = 1.0E+01 A B A B A B
Pe = 1.0E+02
Pe = 1.0E+03 Pe = 1) 1.5608 0.0763 0.3745 0.0446 0.1344 0.0000
Particle Skewness, γ

1.5 Pe = 10) 5.3463 2.3026 0.1700 0.0628 0.1933 0.0000


Pe = 100) 5.3463 2.3026 0.9863 0.1536 0.6421 0.0000
Pe = 1000) 5.3463 2.3026 1.8502 0.2307 2.0612 0.0000
1.0

0.5
of Bijeljic and colleagues [21,22] who hypothesized that inertial ef-
fects led to decreases in effective dispersion at high Re. This pro-
0.0
vides a physical explanation for the change in curvature of
logðDL Þ vs logðPeÞ at high Re that has been observed in many exper-
-0.5 imental studies (see summary in Fig. 3 of Wood [5]).
0 10 20 30 40 50 60
t/TF, s We also demonstrated the application of volume averaging
(MVA) to dispersion in unsteady flows and cross-validated the re-
Fig. 12. Plot of skewness coefficients as a function of time, computed from sults using particle tracking. The dispersion coefficients estimated
simulated particle distributions, for Re = 15 and four different values of Pe (red using volume averaging and particle tracking agree closely with
dashed curves). Also shown (grey solid curves) are fitted curves of the form each other, from the creeping flow to transient flow regime. This
prescribed by the MVA relationship derived in Appendix A (Eq. 17). (For interpre-
demonstrates that MVA is applicable to unsteady flows as well as
tation of the references to colour in this figure caption, the reader is referred to the
web version of this article.) steady.
Finally, we presented a preliminary analysis of the third mo-
ment (skewness) of concentration using MVA, and demonstrated
6. Summary that the particle tracking results conform to the theoretical form
of the relationship for most of the cases evaluated.
In this work we performed computational experiments based
on a 3D sinusoidal tube as a model for porous media to examine Acknowledgments
the effects of inertial and unsteady flows on pore-scale dispersion.
The primary finding of our studies is that DL as a function of Pe in- This research was supported by the U. S. Department of Energy,
creases following a power law through the inertial region (up to Office of Science, Biological and Environmental Research (BER)
Re ¼ 250) but under unsteady flow conditions (Re approaching Division through the Subsurface Biogeochemical Research (SBR)
500) the rate of increase is diminished. Our interpretation is that Science Focus Area (SFA) at Pacific Northwest National Laboratory
the flow separation at intermediate Re leads to the formation of (PNNL).
relatively immobile regions (consistent with [26] who modeled Computations described here were performed using computa-
transport effectively with single- and multi-rate mass transfer tional facilities of the Environmental Molecular Sciences
models), causing long tailing and log-linear increase in dispersion, Laboratory (EMSL), a national scientific user facility sponsored by
whereas once flow becomes unsteady these immobile regions get DOE-BER and located at PNNL, and computational facilities of the
flushed out and the rate of increase in dispersion is lower. This con- National Energy Research Supercomputing Center, which is
clusion was also reached by early experimental studies of Weaver supported by the DOE Office of Science under Contract No.
and Ultman [61], and is consistent with the numerical experiments DE-AC02-05CH11231.
224 M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226

PNNL is operated for the DOE by Battelle Memorial Institute un- Defining the total dispersion tensor by 43, Section 3.3.3]
der Contract No. DE-AC06–76RLO 1830. Z !
1
Db ¼ DAb Iþ ~ b bib
nbr bðr1 ÞdAðr1 Þ  hv ðA:10Þ
Vb r1 2Abr ðxÞ
Appendix A. The nonlocal form and higher moments
And the total skewness tensor by
The local closure problem for dispersion has been developed by Z
DAb
Whitaker [43, Section 3]. The quasi-steady statement of the closure Sb ¼  ~ b Bib
nbr Bðr1 ÞdAðr1 Þ þ hv ðA:11Þ
Vb r1 2Abr ðxÞ
problem is given by
 then these definitions allow Eq. (A.10) to be represented by
0 ¼ r  v b ~cAb  Db  r~cAb ðA:1Þ
@hcAb ib
B:C:1  nbr  ðDb r~cAb Þ ¼ 0; x 2 Acj;M ðA:2Þ eb þ r  ðeb hv b ib hcAb ib Þ
@t
B:C:2 ~cAb ðr þ li Þ ¼ ~cAb ðrÞ; x 2 Abe;M ðA:3Þ h i h i
¼ r  eb Db  rhcAb ib  r  eb Sb : rrhcAb ib ðA:12Þ
The general solution to this problem can be stated by a convolution
integral of the form In one-dimension, when variations in eb ; Db , and Sb can be ne-
Z
glected, this balance equation takes the form (cf. [62, Section 7.7.7])
~cAb ðrÞ ¼ kðr; r1 Þ  rr1 hcAb ib jr1 dVðr1 Þ ðA:4Þ
r1 2VðxÞ @hcAb ib @hcAb ib @ 2 hcAb ib @ 3 hcAb ib
¼ v 0 þ Db 2
 Sb ðA:13Þ
This result can be combined with the unclosed volume averaged @t @x @x @x3
equation given by [43, Section 3.3.3] where, for convenience, we have defined v 0 ¼ hv b ib .
@hcAb i b A few notes about this macroscopic equation are in order.
eb þ r  ðeb hv b ib hcAb ib Þ Third-order transport equations have been studied by a number
@t" !#
Z of previous researchers (e.g., [63], [62, Section 7.7.7], [64]).
b 1
¼ r  eb DAb rhcAb i þ ~
nbr cAb dAðr1 Þ  r  hv
~ b ~cAb i Gardiner [62] points out that such equations generate solutions
V b r1 2Abr ðxÞ that are not always positive, and thus some caution is necessary
ðA:5Þ to understanding them in the context of transport of a strictly
positive scalar. A third-order equation of this form does, however,
to give a simplified nonlocal form as described by Wood and
have an interpretation in the context of quasi-probability, and the
Valdes-Parada [46]
implication is that the moments of this equation will have physical
@hcAb ib meaning even if some local values of the dependent variable do not
eb þ r  ðeb hv b ib hcAb ib Þ (because they are negative).
@th 
To provide some additional interpretation of the localized
¼r eb DAb rhcAb ib
!# transport equation given by Eq. (A.13), we assume that we are
Z Z
1 interested in macroscopically 1-dimensional transport on the
þ nbr ðr1 Þkðr1 ;r2 Þ  rr2 hcAb ib jr2 dVðr2 ÞdAðr1 Þ
Vb r1 2Abr ðxÞ r2 2VðxÞ
entire real line. Defining the spatial moment of any function,
Z Z ! f ðx; tÞ on the real line by
1 Z
r v~ b ðr1 Þkðr1 ;r2 Þ  rr hcAb ib jr1 dVðr2 ÞdVðr1 Þ 1
V np f ðn; tÞdn
2
r1 2Vb r2 2VðxÞ Mp ðtÞ ¼ ðA:14Þ
1
ðA:6Þ
Then the moments of Eq. (A.13) are readily determined by multiply-
Although nonlocal forms offer more capacity to model highly ing through by np , and integrating by parts. The results are
heterogeneous systems, they additional capability comes at a cost reasonably straightforward to derive, and the set of equations defin-
in terms of model complexity. One potential solution to this ing the first three moments are
problem is to localize the model by expanding the source terms
in the integrals. If we consider a first-order expansion, i.e., M1 ðtÞ ¼ v 0 t ðA:15Þ

rr2 hcAb ib jr2 ¼ rhcAb ib jx þ ðr2  xÞðr2  xÞ M2 ðtÞ ¼ v 20 t2 þ 2 Db t ðA:16Þ


b b
: rrhcAb i jx þ OðrrrhcAb i jx Þ ðA:7Þ M3 ðtÞ ¼ v 3 3
0t þ 6 Db v0 t 2
þ 6 Sb t ðA:17Þ
Employing this approximation allows Eq. (A.8) to be For these computations, we have assumed that the concentration
represented by field is normalized such that M 0 ðtÞ ¼ 1. It is important to note here
that each moment is independent of the moments of higher order
~cAb ðrÞ ¼ bðrÞ  rhcAb ib jx þ BðrÞ : rrhcAb ib jx ðA:8Þ
than itself. It is often helpful to express this information in terms
Substituting this result into Eq. (A.5) of the cumulants of the distribution rather than the moments. The
first three cumulants are
@hcAb ib
eb þ r  ðeb hv b ib hcAb ib Þ
@t " ! #
j1 ðtÞ ¼ v 0 t ðA:18Þ
Z
1 j2 ðtÞ ¼ 2 Db t ðA:19Þ
¼ r  eb DAb I þ nbr bðr1 ÞdAðr1 Þ  rhcAb ib
V b r1 2Abr ðxÞ
  j3 ðtÞ ¼ 6 Sb t ðA:20Þ
 r  eb hv ~ b bi  rhcAb ib
b

" # Finally, the influence of the third moment is often measured by the
Z
DAb b skewness, defined by
þ r  eb nbr BdAðr1 Þ : rrhcAb i
V b r1 2Abr ðxÞ 3Sb
  j3
c¼ 3
¼ 1
ðA:21Þ
 r  eb hv ~ b Bib : rrhcAb ib ðA:9Þ j22 Db ð2Db tÞ2
M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226 225

Note that the skewness is inversely proportional to the square- S/ source term,
root of time, and goes to zero as t becomes arbitrarily large; this is Uo ¼ Q bulk velocity based on mean tube radius,
a requirement, if we assume that the concentration field assumes a pR2o
m/s
normal distribution asymptotically.
Re ¼ qUlo Do Reynolds number
Pe ¼ UDo mDo = ReSc Peclet number
Appendix B. The closure problems
L tube wave length, m
One of the advantages of localizing the transport equation is nbr unit normal vector pointing from the b
that the resulting effective parameters, Db and Sb are predictable phase toward the r phase
from a set of partial differential equations written for a representa- T ¼ ULo flow through time, s, for one tube wave
tive volume element of the porous medium. To see this, recall that length
the simplifications made in A lead to the form of the expression for DP pressure drop over length of tube, Pa
~cAb being DL longitudinal dispersion coefficient, m2/s
Sc ¼ qDlm ¼ Dmm Schmidt Number
~cAb ðrÞ ¼ bðrÞ  rhcAb ib jx þ BðrÞ : rrhcAb ib jx ðB:1Þ
r2 particle cloud variance, m2
The closure problem itself for this problem is given by [43] c particle cloud skewness, m3

rr  v b ðrÞ~cAb ðrÞ  Db  r~cAb ðrÞ ~ b ðrÞ  rhcAb ib jx
¼ v ðB:2Þ
B:C:1  nbr  Db r~cAb ðrÞ ¼ 0; x 2 Acj;M ðB:3Þ
References
B:C:2 ~cAb ðr þ li Þ ¼ ~cAb ðrÞ; x 2 Abe;M ðB:4Þ
[1] Bear J. Dynamics of fluids in porous media. New York: Dover; 1972.
Substituting Eq. (B.1) into Eq. (B.4), and using the linearity property [2] Brenner H. Macrotransport processes. Langmuir 1990;6(c12):1715–24.
[3] Mei CC, Auriault JL, Ng CO. Some applications of the homogenization theory.
of the solutions yields the following two closure problems for pre-
Adv Appl Mech 1996;32:277–348.
dicting b and B. [4] Cushman J, Bennethum L, Hu BX. A primer on upscaling tools for porous media.
Adv Water Resour 1992;25(8–12):1042–67.
Problem I [5] Wood BD. Inertial effects in dispersion in porous media. Water Resour Res
2007;43.
rr  v b ðrÞbðrÞ  Db  rbðrÞ ¼ v~ b ðrÞ ðB:5Þ [6] Delgado J. A critical review of dispersion in packed beds. Heat Mass Trans
2006;42(4):279–310.
B:C:1  nbr  Db rbðrÞ ¼ 0; x 2 Acj;M ðB:6Þ [7] Dybbs A, Edwards R. A new look at porous media mechanics – darcy to
turbulent. In: Bear J, Corapcioglu Y, editors. Fundamentals of transport
B:C:2 bðr þ li Þ ¼ bðrÞ; x 2 Abe;M ðB:7Þ phenomena in porous media. Zoetermeer, Netherlands: Martinus Nijhoff;
1984. p. 199–254.
[8] Das D, Nassehi V. Modeling of contaminant mobility in underground domains
Problem II with multiple free/porous interfaces. Water Resour Res 2003;39(3).
[9] Prinos P, Sofialidis D, Keramaris E. Turbulent flow over and within a porous
rr  v b ðrÞBðrÞ  Db  rBðrÞ ¼ 0 ðB:8Þ bed. J Hydraulic Eng 2003:129.
[10] Cardenas M, Wilson JL. The influence of ambient groundwater discharge on
B:C:1  nbr  Db rBðrÞ ¼ 0; x 2 Acj;M ðB:9Þ exchange zones induced by current–bedform interactions. J Hydrol
2006;331(1):103–9.
B:C:2 Bðr þ li Þ ¼ BðrÞ; x 2 Abe;M ðB:10Þ [11] Wagner F, Bretschko G. Interstitial flow through preferential flow paths in the
hyporheic zone of the oberer seebach, austria. Aquat Sci Res Across Boundaries
Solution of these two problems allows one to predict the diffusion 2002;64(3):307–16.
[12] Yabusaki SB, Fang Y, Waichler SR. Building conceptual models of field-scale
tensor, Db , and skewness tensor, Sb , via Eqs. A.10,A.11
uranium reactive transport in a dynamic vadose zone-aquifer-river system.
Water Resour Res 2008;44(12):W12403. http://dx.doi.org/10.1029/
2007WR006617.
Appendix C. Nomenclature [13] Ma R, Zheng C, Tonkin M, Zachara JM. Importance of considering intraborehole
flow in solute transport modeling under highly dynamic flow conditions. J
Contam Hydrol 2011;123:11–9.
[14] Hoagland DA, Prudhomme RK. Taylor–Aris dispersion arising from flow in a
b closure variable defined by Eq. (A.8), m sinusoidal tube. AIChE J 1985;31(2):236–44.
B closure variable defined by Eq. (A.8), m2 [15] Taylor G. Dispersion of soluble matter in solvent flowing slowly through a
Do mean tube diameter, m (primary length tube. Proc R Soc London Ser A Math Phys Sci 1953;219(1137):186–203.
[16] Aris R. On the dispersion of a solute in a fluid flowing through a tube. Proc R
scale) Soc London Ser A Math Phys Sci 1956;235(1200):67–77.
Dm molecular diffusivity of solute, m2/s [17] Wang G, Vanka S. Convective heat transfer in periodic wavy passages. Int J
DL ¼ Dxx ¼ i  Db  i longitudinal component of the total Heat Mass Trans 1995;38(17):3219–30.
[18] Stone K, Vanka SP. Numerical study of developing flow and heat transfer in a
hydrodynamic dispersion tensor, m2/s wavy passage. J Fluids Eng Trans ASME 1999;121(4):713–9.
Db hydrodynamic dispersion tensor, m2/s [19] Souto H, Moyne C. Dispersion in two-dimensional periodic porous media. 2.
Dispersion tensor. Phys Fluids 1997;9(8):2253–63.
f ¼ 14 DLo q2DUP2 friction factor
[20] Cao J, Kitanidis P. Pore-scale dilution of conservative solutes: an example.
o
Water Resour Res 1998;34(8):1941–9.
Ro ¼ D2o mean tube radius, m
[21] Bijeljic B, Muggeridge A, Blunt M. Pore-scale modeling of longitudinal
l fluid viscosity, Pa-s dispersion. Water Resour Res 2004;40(11).
q fluid density, m3/kg [22] Bijeljic B, Blunt M. Pore-scale modeling and continuous time random walk
analysis of dispersion in porous media. Water Resour Res 2006;42(1).
m fluid dynamic viscosity, m2/s [23] Cardenas MB. Three-dimensional vortices in single pores and their effects on
Q discharge, m3/s transport. Geophys Res Lett 2008;35(18). http://dx.doi.org/10.1029/
v velocity vector, m3/s 2008GL035343.
[24] Bolster D, Dentz M, Le Borgne T. Solute dispersion in channels with
p pressure, Pa
periodically varying apertures. Physics of Fluids 2009;21(5).
f body force, N [25] Cardenas MB, Slottke DT, Ketcham RA, Sharp Jr JM. Effects of inertia and
/ scalar directionality on flow and transport in a rough asymmetric fracture. J Geophys
Res – Solid, Earth 2009;114. http://dx.doi.org/10.1029/2009JB006336.
[26] Lambert A. Advection–diffusion in inertial flow through periodically
converging–diverging tubes. Master’s thesis, Oregon State Univeristy; 2010.
226 M.C. Richmond et al. / Advances in Water Resources 62 (2013) 215–226

[27] DeGroot CT, Straatman AG. Closure of non-equilibrium volume-averaged and comparison with ensemble averaging. Water Resour Res 2003;39(8):1210.
energy equations in high-conductivity porous media. Int J Heat Mass Trans http://dx.doi.org/10.1029/2002WR001723.
2011;54(23–24):5039–48. http://dx.doi.org/10.1016/ [46] Wood B, Valdes-Parada F. Volume averaging: local and nonlocal closures using
j.ijheatmasstransfer.2011.07.018. green’s functions. Adv Water Resour 2013;51:139–67. http://dx.doi.org/
[28] Bouquain J, Meheust Y, Bolster D, Davy P. The impact of inertial effects on 10.1016/j.advwatres.2012.06.008.
solute dispersion in a channel with periodically varying aperture. Phys Fluids [47] Maier RS, Kroll DM, Bernard RS, Howington SE, Peters JF, Davis HT.
2012;24(8). http://dx.doi.org/10.1063/1.4747458. Hydrodynamic dispersion in confined packed beds. Phys Fluids
[29] Ralph ME. Steady flow structures and pressure drops in wavy-walled tubes. J 2003;15(12):3795–815.
Fluids Eng 1987;109:255–61. [48] Fischer H, List EJ, Koh R, Imerger J, Brooks N. Mixing in inland and coastal
[30] Pointwise, Inc., Gridgen User Manual, Version 15; 2003. waters. New York, New York: Academic Press; 1979.
[31] Cherukat P, Na Y, Hanratty TJ, McLaughllin JB. Direct numerical simulation of a [49] Gardiner C. Handbook of stochastic methods: for physics, chemistry and the
fully developed turbulent flow over a wavy wall. Theor Comput Fluid Dyn natural sciences. Berlin: Springer; 2004.
1998;11(2):109–34. [50] Ghia U, Ghia KN, Shin CT. High-Re solutions for incompressible-flow using the
[32] Yue W, Lin CL, Patel VC. Large eddy simulation of turbulent open-channel flow navier stokes equations and a multigrid method. J Comput Phys
with free surface simulated by level set method. Phys Fluids 2005;17. http:// 1982;48(3):387–411.
dx.doi.org/10.1063/1.1849182. [51] Tritton DJ. Experiments on the flow past a circular cylinder at low reynolds
[33] Henn DS, Sykes RI. Large-eddy simulation of flow over wavy surfaces. J Fluid numbers. J Fluid Mech 1959;6(4):547–60.
Mech 1999;383(3):75–112. [52] Wieselsberger C. New data on the laws of fluid resistance. NACA TN 1922;84.
[34] Ferziger JH, Perić M. Computational methods for fluid dynamics. Springer- [53] Dennis SCR, Chang GZ. Numerical solutions for steady flow past a circular
Verlag; 2002. cylinder at reynolds numbers up to 100. J Fluid Mech 1970;42:471–80.
[35] Darwish MS, Moukalled FH. Normalized variable and space formulation [54] Fornberg B. A numerical study of steady viscous flow past a circular cylinder. J
methodology for high-resolution schemes. Numer Heat Trans Part B – Fundam Fluid Mech 1980;98(JUN):819–55.
1994;26(1):79–96. [55] Mittal R, Balachandar S. Effect of 3-dimensionality on the lift and drag of
[36] Darwish MS, Moukalled F. Tvd schemes for unstructured grids. Int J Heat Mass nominally 2-dimensional cylinders. Phys Fluids 1995;7(8):1841–65.
Trans 2003;46:599–611. [56] Williamson CHK. Vortex dynamics in the cylinder wake. Annu Rev Fluid Mech
[37] Patankar SV. Numerical heat transfer and fluid flow. Washington, 1996;28:477–539.
DC: Hemisphere; 1980. [57] Ye T, Mittal R, Udaykumar HS, Shyy W. An accurate cartesian grid method for
[38] Balay S, Brown J, Buschelman K, Gropp W.D, Kaushik D, Knepley M.G, et al. viscous incompressible flows with complex immersed boundaries. J Comput
PETSc Web page; 2012. Http://www.mcs.anl.gov/petsc. Phys 1999;156(2):209–40.
[39] MPI Forum, MPI: A message-passing interface standard. Version 2.2; 2009. [58] Federspiel WJ, Fredberg JJ. Axial dispersion in respiratory bronchioles and
Http://www.mpi-forum.org. alveolar ducts. J Appl Physiol 1988;64(6):2614–21.
[40] Nieplocha J, Harrison R, Littlefield R. Global arrays: a nonuniform memory [59] Sisavath S, Jing XD, Zimmerman RW. Creeping flow through a pipe of varying
access programming model for high-performance computers. J Supercomput radius. Phys Fluids 2001;13(10):2762–72.
1996;10(2):169–89. [60] Cardenas MB. Direct simulation of pore level Fickian dispersion scale for
[41] Nieplocha J, Palmer B, Tipparaju V, Krishnan M, Trease H, Aprà E. Advances, transport through dense cubic packed spheres with vortices. Geochem
applications and performance of the global arrays shared memory Geophys Geosyst 2009;10. http://dx.doi.org/10.1029/2009GC002593.
programming toolkit. Int J High Perform Comput Appl 2006;20(2):203–31. [61] Weaver D, Ultman J. Axial dispersion through tube constrictions. AIChE J
[42] Carbonell RG, Whitaker S. Dispersion in pulsed systems. 2. Theoretical 1980;26(1):9–15.
developments for passive dispersion in porous media. Chem Eng Sci [62] Gardiner C. Handbook of stochastic methods. Springer series in
1983;38(11):1795–802. synergetics. Berlin: Springer-Verlag; 1990.
[43] Whitaker S. The method of volume averaging. Kluwer Academic Publishers; [63] Orsingher E. Processes governed by signed measures connected with third-
1999. order heat-type equations. Lith Math J 1991;31(2):220–31.
[44] Edwards D, Shapira M, Brenner H. Dispersion and reaction in two-dimensional [64] Beghin L, Kozachenko Y, Orsingher E, Sakhno L. On the solution of linear odd-
model porous media. Phys Fluids A: Fluid Dyn 1993;5:837–45. order heat-type eequaitons with random initial conditions. J Stat Phys
[45] Wood BD, Cherblanc F, Quintard M, Whitaker S. Volume averaging for 2007;127(4):721–39.
determining the effective dispersion tensor: closure using periodic unit cells

S-ar putea să vă placă și