You are on page 1of 15

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/282238991

Numerical Modeling of Three Dimensional Heat Transfer and Fluid Flow


Through Interrupted Plates Using Unit Cell Scale

Article in Special Topics & Reviews in Porous Media An International Journal · January 2015
DOI: 10.1615/.2015012321

CITATIONS READS

5 169

4 authors, including:

Chao Zhang Terrence W. Simon


Apple Inc. University of Minnesota Twin Cities
15 PUBLICATIONS 187 CITATIONS 273 PUBLICATIONS 4,599 CITATIONS

SEE PROFILE SEE PROFILE

Perry Li
University of Minnesota Twin Cities
281 PUBLICATIONS 4,307 CITATIONS

SEE PROFILE

All content following this page was uploaded by Chao Zhang on 20 February 2018.

The user has requested enhancement of the downloaded file.


DRAFT
Official version available at: http://www.dl.begellhouse.com/journals/3d21681c18f5b5e7,09dfc75d18d3a6ca,31d1f5561536ae34.html

Special Topics & Reviews in Porous Media — An International Journal, 6 (2): 145–158 (2015)

NUMERICAL MODELING OF THREE-DIMENSIONAL


HEAT TRANSFER AND FLUID FLOW THROUGH
INTERRUPTED PLATES USING UNIT CELL SCALE

Chao Zhang,∗ Terrence W. Simon, Perry Y. Li, & James D. Van de Ven

Department of Mechanical Engineering, University of Minnesota, 111 Church St. S.E.,


Minneapolis, Minnesota 55455, USA


Address all correspondence to Chao Zhang E-mail: zhan1120@umn.edu

Original Manuscript Submitted: 10/21/2014; Final Draft Received: 2/25/2015

Interrupted-plate heat exchangers are used as regenerators for absorbing and releasing thermal energy such as in a
Compressed Air Energy Storage (CAES) system in which the exchanger absorbs energy to cool the air being compressed
or liberates energy to heat the air upon expansion. The exchanger consists of layers of thin plates in stacked arrays. In
a given layer, the plates are parallel to one another and parallel to the exchanger axis. Each successive layer is rotated to
have its plates be perpendicular to those of the layer below but still parallel to the exchanger axis. As flow passes from
one layer to the next, new thermal boundary layers develop, beneficial to effective heat transfer. The interrupted-plate
heat exchanger can also be seen as a porous medium. As such, it demonstrates strong anisotropic behavior when flow
approaches the plates in a direction other than axially. Thus, pressure drops and heat transfer coefficients are dependent
upon the attack angle. Mathematical models for anisotropic pressure drop and heat transfer behavior are proposed based
on numerical calculations on a Representative Elementary Volume (REV), the unit cell model of the interrupted-plate
medium. The anisotropic pressure drop is modeled by the traditionally used Darcy and inertial terms, with the addition
of another term representing mixing effects. Heat transfer between the fluid and the plates is formulated in terms of
Nusselt number vs. Reynolds number and approach angle of the mean flow. These models are used when solving, on the
scale of the heat exchanger application, the volume-averaged Navier–Stokes equations that treat the exchanger region as
a continuum. The analysis of the heat exchanger is used for design and optimization of the medium.

KEY WORDS: porous media, numerical simulation, heat exchanger, convection, anisotropic heat transfer,
interrupted plate

1. INTRODUCTION fore, as flow passes through the plate layers, new thermal
boundary layers are created, which is beneficial to heat
The present study presents modeling of three-dimensio- transfer. One application of such medium is Compressed
nal, anisotropic heat transfer and flow resistance prop- Air Energy Storage, where it is used in the chamber of
erties of an interrupted-plate medium that is made for a liquid-piston compressor to absorb heat from air being
the purpose of thermal regeneration. The design idea compressed (Zhang et al., 2013) and liberate that energy
of the present interrupted-plate heat exchanger (heat- when the air is being expanded. Due to its shape, the
absorbing porous medium) originated from a microfabri- interrupted-plate medium is an effective heat exchanger
cated segmented-involute-foil regenerator used for a Stir- but it represents a three-dimensional, highly anisotropic
ling engine (Tew et al., 2007; Sun et al., 2009). The struc- porous medium, requiring detailed characterization of its
ture features layers of thin plates that are stacked per- fluid mechanics properties. The present study resolves the
pendicularly to each other, as shown in Fig. 1. There- anisotropic pressure drop and heat transfer properties.

Draft
2151–4798/15/$35.00 °
c 2015 by Begell House, Inc. 145
146 Draft Zhang et al.

NOMENCLATURE

*
aV surface area per volume (/m) Sm momentum source term (Pa/m)
bf inertial coefficient (/m) T temperature (K)
bf 1 anisotropic inertial coefficient t thickness (m)
for x direction (/m) u velocity
bf 2 anisotropic inertial coefficient V volume
for y direction (/m)
bf 3 anisotropic inertial coefficient Greek Symbols
for z direction (/m) α angle between mean velocity
c conduction effect (–) vector and x axis (–)
cp Specific heat J/(KgK) β angle between mean velocity
d 1 , d 2 , f1 , f 2 coefficients for mixing vector and y axis (–)
on heat transfer (–) γ angle between mean velocity
b inertial coefficient (/m) vector and z axis (–)
¯b̄ anisotropic inertial matrix (/m) ² porosity (–)
b half distance between plates θ angle between mean velocity vector
H pressure resistance term due to mixing and xOy plain (–)
effects (–) κ turbulence kinetic energy
hV volumetric heat transfer coefficient (m2 /s2 )
W/(m3 K) ρ density (kg/m3 )
K permeability (m2 ) ϕ angle between the projection
k thermal conductivity [W/(mK)] of mean velocity
L representative pores size (m) vector on the xOy plane and x axis (–)
` plate length (m) χ a flow variable (velocity, temperature
m Reynolds number exponent characterizing or pressure) (m/s, K, or Pa)
anisotropic mixing on pressure drop (–)
Nu Nusselt number Subscript
n Reynolds number exponent characterizing REV representative elementary volume
anisotropic mixing on heat transfer
℘ periodic pore-scale pressure (Pa) Superscript
q 00 wall heat flux W/m2 f fluid region
ReL Reynolds number based on a s solid region
representative pore size * dimensionless variable

One method of modeling fluid flow and heat transfer packed beds (Saito and de Lemos, 2009), double-pipe
in engineering applications of porous media is by solving heat exchangers (Du et al., 2010), solar receivers (Xu
the volume-averaged transport equations. In the compu- et al., 2011), or metal-foam-filled liquid piston compres-
tational domain, the porous medium region is treated as sors (Zhang et al., 2013). In this approach, flow variables
a continuum (Slattery, 1969; Vafai and Tien, 1981). As are volume-averaged based on a Representative Elemen-
such, when applied to simulating flow in a porous zone, tary Volume (REV), which is a repeating, unit cell of the
closure models are used in the Navier–Stokes equations porous medium. It is defined as
to model the interactions between the flow and the in-
Z
ternal walls of the porous medium. Such approach has 1
been applied to metal-foam-filled pipes (Lu et al., 2006), hχi = χdV (1)
VREV VREV

Special Topics & Reviews in Porous Media — An International Journal


Numerical Modeling of Three-Dimensional Heat Transfer 147

ing flows through wire screens (Hsu, 1999). Thus, the


model is given by
*
µh ui ¯ ¯ H
* ¯*¯ * * 1/2 *
Sm = − −ρb ¯h ui¯ h ui− 3/4 (ρµ|h ui|) h ui (2)
Kf K

The interfacial heat transfer term is modeled using di-


mensionless correlations based on Nusselt, Reynolds,
and Prandtl numbers. A number of such correlations for
isotropic open cell porous media and packed beds of
spheres have been developed (Kuwahara et al., 2001;
(a) Global view Kamiuto and Yee, 2005; Nakayama et al., 2009).
In anisotropic porous media, the viscous effect is mod-
eled with a permeability tensor (Dullien, 1979). The iner-
tial effects are modeled by an inertial tensor. For an or-
thotropic medium consisting of repeating square rods, an
inertial tensor was modeled (Nakayama et al., 2002) as
 
bf 1 bbf 1 cos α cos β bbf 3 cos α cos γ
¯b̄ =bb cos α cos β bf 2 bbf 2 cos β cos γ (3)
f1
bbf 3 cos α cos γ bbf 2 cos β cos γ bf 3

Direction-dependent pressure drop and heat transfer cor-


relations are usually obtained through numerical simula-
tions on the REV models of the anisotropic porous me-
dia (Nakayama et al., 2004; Alshare et al., 2010). How-
(b) Unit cell ever, most of the porous media investigated in the liter-
ature are two-dimensional and quasi-three-dimensional,
FIG. 1: Interrupted-plate medium. and have distinctively different shapes from the present
interrupted-plate medium. In the present study, closure
models for the momentum source term and the interfa-
cial heat transfer coefficients for the three-dimensional,
The variable χ represents local, pore-scale velocity,
anisotropic interrupted-plate media are developed.
temperature, or pressure. Volume-averaging the Navier–
Stokes equations yields the transport equations of flow
in a porous medium, which require closing the follow- 2. UNIT CELL MODEL
*
ing terms: a momentum
¡ sink term,¢ S m , an interfacial heatThe anisotropic pressure drop and heat transfer of the
transfer term, hV hTs is − hT if , and a thermal disper- interrupted-plate medium are determined using the micro-
*
f
sion term, −ρ∇ · cp h ũ T̃ i . The present study resolves scopic (pore-scale) results calculated in REV simulations
* ¡ ¢ using different Reynolds numbers and various flow an-
anisotropic models for terms S m and hV hTs is − hT if ,
gles. This section introduces the procedure of these REV
which are dominant terms for the present interrupted-
simulations.
plate heat exchanger application. The thermal dispersion
term can be ignored in all settings but those of very high-
speed flows. 2.1 Periodic Flow and Governing Equations
In isotropic porous media, the momentum sink term The boundaries of the REV simulation domain must sat-
represents resistance of the porous medium to the flow. isfy periodic flow conditions. As such, the flow that exits
It has been modeled using a viscous and an inertial term the REV from an outlet boundary must enter its paring
* 1/2 *
(Hsu, 2000). Another term proportional to −(ρµ| u|) u inlet boundary.
was added to capture flow transitional effects, based on The flow region, as shown by the transparent region
curve-fitting of experimental data of steady and oscillat- of the REV in Fig. 1(b), is the computational domain.

Volume 6, Number 2, 2015


148 Zhang et al.

Three Reynolds numbers are investigated in the simula- Angles αp , βp , and γp are varied in different simula-
tion runs: 1, 181, and 8309. These Reynolds numbers rep- tion runs. The pore-scale pressure gradient, (∂℘)/(∂xi ),
resent, respectively, Darcy, Darcy plus Forchheimer, and and the magnitude of the global-scale pressure drop
post-Forchheimer (turbulent) flow regimes. The Reynolds |(∂hpif )/(∂xi )| are calculated. Geometrical representa-
number is based on the mean flow velocity and a charac- tions of these angles are given in Fig. 2(a). Any direction
teristic pore size, in a 3D space can be defined by using either the (α, β, γ)
system, or a (ϕ, θ) system, as shown in Fig. 2(b).
*
ρ|h ui|L Periodic boundary conditions are applied on each pair
ReL = (4)
µ of the inlet and outlet boundaries of the REV domain,

where L is defined as 2L3 = VREV,f . Laminar flow Γ|x=−l = Γ|x=l , Γ|y=−(b+ 2t ) = Γ|y=b+ 2t ,
transport equations are solved when Reynolds number is Γ|z=−(b+ 2t ) = Γ|z=b+ 2t , Γ = ui , ℘
1 or 181; the RANS (Reynolds-Averaged Navier–Stokes)
equations are solved when Reynolds number is 8309, On the interfacial boundary between the fluid and solid,
given in the following (with the overbars being omitted): the no-slip velocity and uniform-wall-heat-flux boundary
conditions are used. The periodic thermal boundary con-
∂ui dition is defined using interfacial wall heat flux,
=0 (5)
∂xi

· µ
∂ui ∂(ui uj ) ∂p ∂ ∂ui ∂uj
ρ +ρ =− + µ +
∂t ∂xj ∂xi ∂xj ∂xj ∂xi
¶¸ ³ ´
2 ∂uk ∂
− δij + −ρu0i u0j (6)
3 ∂xk ∂xj

∂ ³ ui ui ´ ∂ h ³ ui ui ´i
ρ cp T + +ρ uj cp T + +p
∂t 2 ∂xj 2
· µ ¶
∂ ∂T ∂ui ∂uj 2 ∂uk
= k + ui µ + − δij
∂xj ∂xj ∂xj ∂xi 3 ∂xk
¸
cp µt ∂T
+ (7)
Prt xj (a) α, β and γ
Solutions are obtained when the flow reaches a steady
state in the REV simulation. The Boussinesq hypothesis
is used to relate the Reynolds stresses to the mean velocity
gradients,

∂ui ∂uj 2 ∂uk 2


−ρu0i u0j = µt ( + − δij ) − ρκδij (8)
∂xj ∂xi 3 ∂xk 3

Different flow directions in different REV simulations are


set by varying the pressure gradient direction. The pres-
sure gradient is decoupled into a macroscopic pressure
gradient, which is due to the periodic interruption of the
porous medium, and a microscopic (pore-scale) pressure
gradient, which is due to the local flow field variation in-
side the REV: (b) ϕ and θ
¯ ¯
∂p ¯ ∂hpif ¯
¯ (cos αp , cos βp , cos γp ) + ∂℘
FIG. 2: Schematic of two sets of angles used to charac-
=¯ ¯ (9)
∂xi ∂xi ¯ ∂xi terize the direction of a vector.

Special Topics & Reviews in Porous Media — An International Journal


Numerical Modeling of Three-Dimensional Heat Transfer 149

T |x=l − T |x=−l q 00 aV 2.3 Choice of Turbulence Model


= cos α (10)
2l * f
ρcp |h ui | The Transition SST model is used for solving the tur-
bulent viscosity and turbulent kinetic energy in Eq. (7)
T |y=b+ 2t − T |y=−(b+ 2t ) q 00 aV
= cos β (11) for 8309 Reynolds number flows. The model couples the
2b + t * f
ρcp |h ui | SST k − ω model with two other transport equations, one
for the intermittency, and one for the momentum thick-
T |z=b+ 2t − T |z=−(b+ 2t ) q 00 aV
= cos γ (12) ness Reynolds number, which is used as a transition onset
2b + t * f
ρcp |h ui | criterion (Menter et al., 2004). This choice of the turbu-
lence model is based on another study comparing vari-
where cos α, cos β, cos γ are projections of the unit mean ous RANS simulations to Large Eddy Simulation (LES)
velocity vector onto the x, y, and z axes, respectively. when the mean flow is along the x-axis direction. For
Length features of the medium are ReL = 1777, the Transition SST model computes a Nus-
selt number that is 2.7% different from that computed by
` = 7.5 mm, 2b = 2.75 mm, t = 0.55 mm,
LES. For ReL = 41,667, the Transition SST model cal-
² = 0.8333, aV = 643.1/m, L = 4.083 mm. culates a Nusselt number that is 1.0% different from that
by WMLES (Wall-Modeled LES). Other closure models
Twenty different mean flow directions are used for each
compared are the two-equation k − ² and k − ω models,
of the 181 and 8309 Reynolds numbers.
which give greater errors.

2.2 Numerical Procedure


3. FLOW RESISTANCE
The transport equations are solved by the finite volume
method using the commercial computational fluid dynam- In this section, flow fields and anisotropic flow resistance
ics (CFD) software ANSYS Fluent. The first-order im- of the interrupted-plate medium are discussed in specially
plicit formulation is used for transient discretization; the selected scenarios, to reveal the effects of: (1) the mean
second-order upwind scheme is used for discretizing ad- velocity on the pressure drop when the velocity is aligned
vective terms; central differencing is used for diffusive with the pressure gradient, (2) y-direction velocity com-
terms. The SIMPLE algorithm (Patankar and Spalding, ponent on z-direction pressure drop, and (3) x-direction
1972) is used for pressure–velocity coupling. The conver- velocity component on y-direction pressure drop, as well
gence criterion for residuals of all equations is set to 10−9 . as y-direction velocity component on x-direction pres-
The computational domain is discretized into rectangular sure drop. Then, the pressure drop models will be tested
hexahedral cells. The cell size gradually decreases as the against cases with randomly selected mean flow attack an-
walls are approached. The total numbers of grid cells are gles.
1,469,139, for simulation runs with a Reynolds number
of 8309, and 707,840, for simulation runs with smaller 3.1 Mean Flow Aligned with the Plate
Reynolds numbers. The maximum value of the dimen-
sionless wall distance (y + ) of the first cell adjacent to the When the mean flow direction is aligned with the plate,
wall is between 2 and 6 for simulations of 8309 Reynolds such as the Ox or Oy coordinate (see Fig. 2), the pressure
number. gradient points to the same direction as the mean flow.
Grid independence is verified by refining the mesh Velocity streamline plots are shown in Figs. 3 and 4 for
from 707,840 to 1,916,480 cells for a case with a cases with the mean flow along the Ox and Oy direc-
Reynolds number of 181, and refining the mesh from tions, at a Reynolds number of 8309. Because the thick-
1,469,139 to 2,584,833 cells for a case with a Reynolds ness of the plate is relatively small, when the mean flow
number of 8309. For the 181 Reynolds number case, the is in the Ox direction, local flow in the REV is also in the
calculated Nusselt numbers and normalized pressure drop same direction as the mean flow; only very small regions
values using the original mesh and the refined mesh are, of stagnation and separation are seen near the leading and
respectively, 22.3 and 22.2, and –0.2092 and –0.2084. For trailing edges. When the mean flow is in the Oy direction,
the 8309 Reynolds number case, they are, respectively, it is perpendicular to a set of vertical plates, as shown in
180.6 and 182.4, and –0.0956 and –0.0969. The chosen Fig. 4. In this case, flow regions interrupted by the plates
grid sizes are sufficient. are mostly stagnation regions.

Volume 6, Number 2, 2015


150 Zhang et al.

K1 = 4.79 × 10−7 m2 , K2 = 2.59 × 10−7 m2 ,


b1 = 24.69/m, b2 = 23.85/m

The model (the curve) and the CFD results (the points)
are compared in Fig. 5, which shows good agreement.
The dimensionless pressure drop decreases with increas-
ing Reynolds number, because of non-dimensionalization
with the inertial effects.

3.2 Flow Parallel to the yOz Plane


Under this condition, the mean flow may impinge onto
both plates in the REV. As a result, flows in different re-
FIG. 3: Velocity and wall temperature distributions gions in the REV follow distinctively different directions,
(ReL = 8309, θ = 0, ϕ = 0). which allows the local flow to run parallel to the nearest
plate to avoid direct impinging on it. As shown in Figs. 6
and 7, in different locations inside the REV, the flows are
in different directions, yet approximately parallel to the
nearest plates.
Impingement of flow onto the plates results in flow
separation and mixing effects. The pressure drop will

FIG. 4: Velocity and wall temperature distributions


(ReL = 8309, θ = 0, ϕ = 90◦ ).

The pressure drop in these flow situations is caused (a)


by viscous and by inertial effects in, respectively, the Ox
and Oy directions. It is modeled using a Darcian and an
inertial term,
µ ∗¶
∂p 1 L2

=− − Lb1 (13)
∂x Ox ReL K1
µ ∗¶
∂p 1 L2

=− − Lb2 (14)
∂y Oy ReL K2
where the dimensionless pressure drop is based on
L
∇p∗ = ¯ ¯2 ∇p (15) (b)
¯*¯
ρ ¯h ui¯
FIG. 5: Pressure drop vs. Reynolds number when the
Using calculated pressure drop values from the CFD runs, mean flow is along the symmetric lines of the plate
the permeability and inertial coefficients in the Ox and (curves: model; points: CFD results); (a) mean flow in
Oy directions are found: Ox direction, (b) mean flow in Oy direction.

Special Topics & Reviews in Porous Media — An International Journal


Numerical Modeling of Three-Dimensional Heat Transfer 151

state. In the present study, varying the mean flow direction


from one that is aligned with the plate to one that has an
attack angle on the plate transitions the flow from a rela-
tively stable state to an unstable state with increasing mix-
ing effects. Thus, the amount of pressure drop caused by
this mixing effect is modeled based on a variation of the
term proposed in (Hsu, 1999): ρH23 Rem L hu2 ihu3 i, or
23

m23 * 2 ∗
ρH23 ReL cos β|h ui| hu3 i . In sum, the pressure drop
in the Oy direction is
µ ∗¶
∂p 1 L2 ∗
= − hu2 i −[Lb2 Lb23 cos β cos γ]
∂y ∗ yOz ReL K2
· ∗ ¸
hu2 i m23 ∗
FIG. 6: Velocity and wall temperature distributions × ∗ − H23 ReL cos βhu3 i (16)
hu 3 i
(ReL = 8309, θ = 45◦ , ϕ = 90◦ ).
Due to the identical periodic features of the shape along
the Oy and Oz directions, the pressure drop in the Oz
direction must be modeled as
µ ∗¶
∂p 1 L2 ∗ £ ¤
= − hu3 i − Lb23 cos β cos γ Lb2
∂z ∗ yOz ReL K2
· ∗ ¸
hu2 i ∗
× ∗ − H23 Rem L cosγhu2 i
23
(17)
hu3 i

Good agreement is found between the model and the


directional pressure drop values calculated from 10 CFD
runs with 5 different mean flow angles and 2 Reynolds
numbers, as shown in Fig. 8. A directional component
of the pressure drop increases as the velocity component
along the same direction increases. This is the result of an
FIG. 7: Velocity and wall temperature distributions increasing amount of flow impinging onto the plate. The
(ReL = 8309, θ = 64◦ , ϕ = 90◦ ). directional inertial coefficient and the mixing coefficients
are
be characterized using an inertial and an additional b23 = −11.92/m, H23 = 3.605 × 10−2 ,
mixing term. The pressure drop caused by inertia is
m23 = 0.1050.
direction-dependent; the pressure drop in the Oy direc-
tion results from the inertial effect in the Oy direc-
tion, as well as that in the Oz direction. According 3.3 Flow Parallel to the xOy Plane
to Eq. (3), the inertial effects in the Oy and Oz di- When the mean flow direction is parallel to the xOy
rections result in a pressure drop in the Oy direction, plane, the local flow may impinge on the plate that is per-
which should be modeled in a more general form, as pendicular to the xOy plane. As shown in Figs. 9 and 10,
* *
−ρb2 |h ui|2 u∗3 i and −ρb23 cos β cos γ|h ui|2 hu∗3 i, respec- when the mean flow has an attack angle to the vertical
tively. Unsteady, mixing effects are found along with plates but parallel to the horizontal plane, the local flows
flow separation after flow impinges on a plate. A mix- in different regions in the REV follow different directions.
ing term must be used in the modeling of the pressure The flow in the gap space between the two vertical plates
drop to account for this effect. In (Hsu, 1999), the term, in Fig. 8 has an angle ϕ greater than that of the mean flow;
1/2 *
(−H/K 3/4 )(ρµ|h ui|) h ui, or −ρHRe−0.5
* * *
L |h ui|h ui, the flows alongside the two vertical plates are nearly par-
was used to capture mixing effects of an oscillatory flow allel to the plates, due to the large flow resistance. Flow
as it transitions from a relatively stable state to an unstable separation zones form due to impingement effects.

Volume 6, Number 2, 2015


152 Zhang et al.

(a)

FIG. 10: Velocity and wall temperature distributions


(ReL = 8309, θ = 0, ϕ = 18◦ ).

to follow the Oz direction. As a result, the impingement


effects onto the vertical plates in the situation of Figs. 9
and 10 are much stronger than those in the situation of
(b) Figs. 6 and 7. Therefore, when the mean flow is parallel
to the xOy plane, stronger separation zones behind those
FIG. 8: Pressure drop for different mean flow direc- plates are created, which induces larger pressure drop in
tions with different Reynolds numbers, all parallel to the the Ox direction. As can be seen, the flow separation and
yOz plane (curves: model; points: CFD results); (a) y- mixing effect are strongly affected by mean flow angle
directional pressure drop, (b) z-directional pressure drop. of attack. The pressure drop must be modeled to capture
these effects.
The directional pressure drop is modeled using a Dar-
cian term, an inertial term, and a mixing term representing
flow separation due to impingement:
µ ∗¶
∂p 1 L2 ∗
= − hu1 i −[Lb1 Lb12 cos α cos β]
∂x∗ xOy ReL K1
· ∗ ¸
hu1 i ∗
× ∗ − H12 RemL
12
cos αhu2 i (18)
hu2 i

µ ¶
∂p∗ 1 L2 ∗ £ ¤
=− hu2 i − Lb21 cos α cos β Lb2
∂y ∗ xOy ReL K2
· ∗ ¸
hu1 i ∗
× ∗ − H21 Rem
L
21
cos βhu1 i (19)
hu2 i
FIG. 9: Velocity and wall temperature distributions
(ReL = 8309, θ = 0, ϕ = 5.4◦ ). Results from 12 CFD runs with different mean flow an-
gles and two different Reynolds numbers are used to find
the inertial and mixing coefficients in the model.
Comparing Figs. 9 and 10 to Figs. 6 and 7, one sees
that if the mean flow is parallel to the yOz plane (Figs. 6 b12 = −63.12/m, b21 = 221.6/m, H12 = 0.6252,
and 7), the local fluid follows the Oy direction in some re-
gions of the REV while in other regions it follows the Oz H21 = 2.271, m12 = −0.04198, m21 = 0.02699
direction. However, if the mean flow is parallel to the xOy The model and the CFD runs are in good agreement as
plane (Figs. 9 and 10), it is not possible for the local fluid shown in Fig. 11. The maximum values for the directional

Special Topics & Reviews in Porous Media — An International Journal


Numerical Modeling of Three-Dimensional Heat Transfer 153

(a) (b)
FIG. 11: Pressure drops with different mean flow directions with different Reynolds numbers, all parallel to the xOy
plane (curves: model; points: CFD results); (a) x-directional pressure drop, (b) y-directional pressure drop.

components of the pressure drop, ∂p∗ /∂x∗ and ∂p∗ /∂y ∗ , neither ϕ nor θ is or 90◦ . The flow fields for such cases
are reached when the mean flow has an angle of attack are shown in Figs. 12 and 13. Even though the mean flow
with respect to the plates. is in a certain direction, the local flow features do not fol-
In the previous discussions, specially chosen CFD runs low this direction, and they run in very different directions
have resolved the pressure drop characteristics of the in- to reduce impingement on the walls, leading to complex
terrupted plate medium with respect to different flow an- flow separation and mixing pattern. This feature is bene-
gles and the directional pressure drop components are ficial for heat transfer, as the amount of mixing that leads
shown to be functions of directional velocity components. to heat transfer is determined by the pore-scale mixing
The full model is given by activities.
 2  The model [Eq. (20)] and pressure drop values cal-
L culated from all the simulation cases are compared in
 K1 0 0  
∗ Fig. 14. The following features of the anisotropic pressure
1   hu1 i
∗ ∗  L2  ∗  drop of the interrupted-plate medium can be observed.
∇ p =−  0 0  hu 2 i
ReL  K2  ∗ For a very small θ, as ϕ increases from 0 to 90◦ , (1)
 hu i
L2  3
the x-directional pressure drop increases slightly when
0 0
K2 ◦
  ϕ reaches around 30 , due to an increase in viscous ef-
Lb1 Lb12 cos α cos β Lb12 cos α cos γ fect in the x direction, and it then decreases to 0 as ϕ
− Lb21 cos α cos β Lb2 Lb23 cos β cos γ
Lb21 cos α cos γ Lb23 cos β cos γ Lb2
 ∗ 
hu1 i

×  hu2 i 

hu3 i
 
0 H12 RemL
12
cos α H12 Rem L
12
cos α
− H21 ReL 21 cos β
m
0 H23 ReL 23 cos β
m
m21 m23
H21 ReL cos γ H23 ReL cos γ 0
 ∗ 
hu1 i

×  hu2 i  (20)

hu3 i

3.4 Mean Flow Impinging on Both Plates


Twenty additional CFD runs are made for cases in which
the mean flow is in a direction that is not parallel to ei- FIG. 12: Velocity and wall temperature distributions
ther of the plates or the x, y, z coordinates. In this case, (ReL = 8309, θ = 33◦ , ϕ = 37◦ ).

Volume 6, Number 2, 2015


154 Zhang et al.

4. INTERFACIAL HEAT TRANSFER


Results of the REV calculations are analyzed for inter-
facial heat transfer between the fluid and the wall. The
wall temperature distributions for different mean flow di-
rections are also shown in Figs. 3, 4, 6, 7, 9, 10, 12, and
13. Because the wall has a uniform negative heat flux, the
fluid is heating the wall and, thus, locations on the wall
with high temperature values indicate higher local heat
transfer coefficients. Observing these figures, one sees
that wall heat transfer effect is strong next to the regions
where the flow is active. The weakest local heat trans-
fer effect on the wall is found next to the stagnant fluid
regions in Fig. 4. Even though the mean flow direction
FIG. 13: Velocity and wall temperature distributions is fixed, local flows in different regions inside the REV
(ReL = 8309, θ = 31◦ , ϕ = 44◦ ). follow different directions, leading to complex mixing ef-
fects that enhance heat transfer, such as in the case shown
by Figs. 12 and 13.

increases to 90 , due to a decrease in the x-directional A correlation for anisotropic heat transfer is developed
inertial effect, and (2) the y-directional pressure drop in- based on dimensionless numbers. The Nusselt number is

creases as ϕ increases from 0 to around 45 , due to in- a function of Reynolds number as well as angles that char-
creasing flow separation caused by flow impinging on the acterize the mean flow direction. These angles relate the
plate, and then decreases to smaller values, due to an in- pore-scale mixing effect to the macroscopic heat transfer.
creasing size of the flow stagnation region. For large θ A volumetric heat transfer coefficient is defined as
values, the effects of changing ϕ on the pressure drop q 00 aV
values in the Ox and Oy directions are small because the hV = · Z ¸ (21)
1 f
local fluid in the REV is allowed to go in the Oz direction T dA − hT i
A wall
to avoid direct impingement onto the plate, which avoids
flow separation or stagnation. For very small values of ϕ, The Nusselt number is defined accordingly,
as θ increases from 0 to 90◦ , the z-directional pressure
drop first increases and then decreases, because impinge- hV L2
Nu = (22)
ment of flow onto the plates first causes separation and κ
then stagnation. For a large value of ϕ, the effect of θ on
Studies in (Nakayama et al., 2002, 2004) used the follow-
the z-directional pressure drop is small because local fluid
ing expression to model anisotropic heat transfer behav-
in the REV can avoid directly impinging upon the plates
ior:
while maintaining the mean flow direction.
1/nc
In the development of the model, an anisotropic mix- Nu = (cn1 c cos2 α + cn2 c cos2 β + cn3 c cos2 γ)
ing term [the last term on the right-hand side of Eq. (20)] 1/nd
was introduced. This term is special to the 3D porous + (dn1 d sin2 γ + dn2 d cos2 γ) Re0.6 Pr1/3 (23)
medium in the present study, because the plates in the
1/n
porous medium cause strong separation and mixing when The term (cn1 c cos2 α + cn2 c cos2 β + cn3 c cos2 γ) c repre-
flow impinges on them, which is different from the flow sents conduction effects in the situation of a very small
through an open-cell or a packed bed. The root-mean- Reynolds number. Other porous media heat transfer cor-
square (rms) values of the model with inclusion of only relations have neglected the conduction effect for con-
the Darcian and inertial terms, and of the model with in- venience of normalizing the Nusselt number on Prandtl
clusion of all three terms are calculated. The rms values number (Kamiuto and Yee, 2005). Given this, the con-
are normalized on the differences between the maximum duction term is approximately modeled to be proportional
and minimum values of the samples. Figure 15 shows that to Pr1/3 and is independent of the mean flow direction.
inclusion of the mixing term gives predicted pressure drop The second term on the right-hand side of Eq. (23) de-
values closer to the CFD results. pends on Reynolds number and mean flow angle, and

Special Topics & Reviews in Porous Media — An International Journal


Numerical Modeling of Three-Dimensional Heat Transfer 155

(a) (b)

(c) (d)

(e) (f)
FIG. 14: Comparison of anisotropic pressure drop between model (curved surfaces) and CFD solutions (points).
Two locations are labeled with brackets to give comparisons between the CFD “x” (first entry in the bracket) and the
corresponding model results (second entry in the bracket). The locations being labeled are (ϕ = 55◦ , θ = 31◦ ) and
(ϕ = 75◦ , θ = 65◦ ) for ReL = 181; (ϕ = 44◦ , θ = 31◦ ) and (ϕ = 63◦ , θ = 42◦ ) for ReL = 8309. (a) x-direction
pressure drop, ReL = 181, (b) x-direction pressure drop, ReL = 8309, (c) y-direction pressure drop, ReL = 181,
(d) y-direction pressure drop, ReL = 8309, (e) z-direction pressure drop, ReL = 181, (f) z-direction pressure drop,
ReL = 8309.

represents convective heat transfer due to fluid inertial two-dimensional and quasi-three-dimensional porous me-
and mixing effects. Furthermore, this term is based only dia investigated. Given this, a more generalized form for
on the angle γ because it is the dominant angle for the this term is adopted in the present study by including

Volume 6, Number 2, 2015


156 Zhang et al.

(a)

FIG. 15: RMS values of the model with inclusion of only


the Darcian and inertial terms, and of the model with in-
clusion of all three terms.

all independent angles using trigonometric functions in


a quadratic form. Also, considering the fact that angles
ϕ and θ have similar effects on volumetric heat transfer,
the proposed heat transfer model for the interrupted plate
medium is (b)

FIG. 16: Comparison of Nusselt numbers between model


Nu ¡ ¢ n
= c + d1 cos2 ϕ + cos θ2 ReLd1 (curved surfaces) and CFD solutions (points). Two loca-
Pr1/3 ¡ ¢ n tions are labeled with brackets to give comparisons be-
+ f1 sin2 ϕ + sin2 θ + d2 (cos ϕ + cos θ ReLd2 tween the CFD “x” (first entry in the bracket) and the cor-
n responding model results (second entry in the bracket).
+ f2 (sin ϕ + sin θ)ReLf2 (24)
The locations being labeled are: (ϕ = 55◦ , θ = 31◦ ) and
Using results from 40 REV simulations, we calculate the (ϕ = 75◦ , θ = 65◦ ) for ReL = 181; (ϕ = 44◦ , θ = 31◦ )
coefficients in Eq. (24): and (ϕ = 63◦ , θ = 42◦ ) for ReL = 8309, (a) ReL = 181,
c = 7.012, d1 = −2.766, f1 = −0.530, (b) ReL = 8309.
d2 = 2.685, f2 = 0.444,
nd1 = 0.387, nf1 = 0.637, nd2 = 0.476,
40◦ is that an increasing amount of flow impinging on the
nf2 = 0.682 plates leads to increasingly complex pore-scale fluid mo-
The model is compared with CFD solutions in Fig. 16. tion. Representative cases are shown by Figs. 9 and 10.
Two locations that are in the middle of the ϕ − θ domain The reason for the decrease in heat transfer when ϕ in-
are labeled by the CFD and modeled values. In general, creases from 40◦ to around 90◦ is that impingement of the
the two are in good agreement with one another. flow at a more directed angle onto the plate leads to more
The directional heat transfer of the interrupted plate fluid being stagnated. Representative cases are shown by
medium has the following characteristics: (1) For a given Figs. 6 and 7, where the flow impinges on the plates at
θ (especially for one between 0 and 65◦ ), heat transfer in- a directed angle and causes large stagnation zones, thus
creases as ϕ increases from 0 to around 40◦ , and then de- less heat transfer. (2) Due to identical geometric period-
creases as ϕ continues to reach 90◦ . The reason for the in- icity in the Oy and Oz directions, for a given ϕ (espe-
crease in heat transfer when ϕ increases from 0 to around cially between 0 and 65◦ ) heat transfer also increases as

Special Topics & Reviews in Porous Media — An International Journal


Numerical Modeling of Three-Dimensional Heat Transfer 157

θ increases from 0 to around 40◦ and then decreases as θ flow and its projection on the xOy plane, and ϕ, the an-
continues to reach 90◦ . (3) Because of the aforementioned gle between the projected vector on the xOy plane and the
aspects, the highest heat transfer effects occur when the Ox axis. The directional pressure drop is largest when ei-
mean flow approaches in a direction with both ϕ and θ ther θ or ϕ is around 45◦ while the other is 0. The reason
being about 33◦ . The reason for this is that with this direc- for this is that the largest flow separation effect is real-
tion, the pore-scale fluid is most agitated, forming com- ized when the mean flow approaches from this direction.
plex fluid movement in many different directions inside If either θ or ϕ is smaller than 45◦ while the other is 0,
the REV, even though the mean flow is pointing in one the flow separation effect is smaller; if either θ or ϕ is
specific direction. This situation is illustrated by the flow greater than 45◦ while the other is 0, stagnation zones in
fields in Figs. 12 and 13. Heat transfer is enhanced by the the REV play important roles in suppressing fluid move-
agitated pore-scale fluid. ment and, thus, reducing pressure drop. If either θ or ϕ
The proposed models are further tested against a CFD is around 45◦ while the other is greater than 0, the pore-
simulation for a Reynolds number that was not used for scale fluid attempts to run parallel to the plate to avoid
developing the models. A flow through the REV in the x direct impingement on the plate, leading to reduced flow
direction with a Reynolds number of 4969 is simulated separation effects. Heat transfer is greatest when both θ
using the Transition SST model, and a mesh count of and ϕ are around 33◦ , again due to agitation on the pore
1,323,532 cells, with a maximum y+ value of 1.88 for scale. For a given Reynolds number, the largest heat trans-
the first layer of cells next to the wall. The Nu/Pr1/3 val- fer and the largest pressure drop do not occur for the same
ues and the dimensionless pressure drop values resulting mean flow angle of attack. This indicates that heat trans-
from the CFD simulation and the model are, respectively, fer can be maximized with reduced cost in flow resistance
166.9 and 167.1, and –0.1217 and –0.1078. The model is by carefully positioning the interrupted plate medium at a
relatively good. particular angle.
Next, numerical heat transfer values are compared be- The models developed in the present study are nec-
tween the present interrupted-plate heat exchanger and essary for conducting numerical simulations of porous
heat exchangers studied in other references. For com- media for various applications by solving the volume-
parison purposes, converting dimensionless numbers to averaged Navier–Stokes equations based on the porous-
be based on the hydraulic diameter, which is twice the continuum approach.
plate separation distance, we obtain that a Reynolds num-
ber of 249 and the Nuhydraulic /Pr1/3 value of 14.2 for
ACKNOWLEDGMENTS
the present interrupted-plate medium for an x-directional
flow. In the study of the microfabricated involute-foil re- This work is supported by the National Science Foun-
generator (Sun et al., 2009), the Nuhydraulic /Pr1/3 value dation under grant NSF-EFRI 1038294, and University
of the regenerator is 16.4. Overall, the heat transfer val- of Minnesota, Institute for Renewable Energy and En-
ues computed in the present study are comparable to those vironment (IREE) under grant: RS-0027-11. The authors
published for these other heat exchangers. would like to thank also the Minnesota Super-Computing
Institute for the computational resources used in this
5. CONCLUSIONS work.

The present study investigates anisotropic pressure drop


and heat transfer for an interrupted plate medium. It is REFERENCES
done with numerical simulations. The results could be Alshare, A. A., Strykowski, P. J., and Simon, T. W., Modeling
used in other heat exchangers of generally the geome- of unsteady and steady fluid flow, heat transfer and disper-
try of Fig. 1. Flows through the REV of the medium sion in porous media using a unit cell scale, Int. J. Heat Mass
with different mean flow angles are simulated. Because Transfer, vol. 53, pp. 2294–2310, 2010.
the medium consists of long, flat plates, impingement Du, Y. P., Qu, Z. G., Zhao, C. Y., and Tao, W. Q., Numerical
of flows at different angles onto the plates, which leads study of conjugated heat transfer in metal foam filled double-
to substantially different pore-scale fluid activity, are re- pipe, Int. J. Heat Mass Transfer, vol. 53, pp. 4899–4907,
sponsible for the features observed for the directional ef- 2010.
fects on pressure drop and heat transfer. The mean flow Dullien, F. A. L., Porous Media: Fluid Transport and Pore
angle is characterized by θ, the angle between the mean Structure, Academic Press INC., pp. 215–219, 1979.

Volume 6, Number 2, 2015


158 View publication stats Zhang et al.

Hsu, C. T., On pressure-velocity correlation of steady and os- pp. 1787–1806, 1972.
cillating flows in regenerators made of wire screens, Trans. Saito, M. B. and S. de Lemos, M. J., Laminar heat transfer
ASME, vol. 121, pp. 52–56, 1999. in a porous channel simulated with a two-energy equations
Hsu, C. T., Dynamic Modeling of Convective Heat Transfer in model, Int. Commun. Heat Mass Transfer, vol. 36, pp. 1002–
Porous Media, in K. Vafai, ed., Handbook of Porous Media 1007, 2009.
(2nd ed.), Taylor and Francis Group, LLC, pp. 39–80, 2000.
Slattery, J. C., Single-phase flow through porous media, AIChE
Kamiuto, K. and Yee, S. S., Heat transfer correlations for open- J., vol. 15, pp. 866–872, 1969.
cellular porous materials, Int. Commun. Heat Mass Transfer,
vol. 32, pp. 947–953, 2005. Sun, L., Simon, T. W., Mantell, S.C., Ibrahim, M. B.,
Gedeon, D., and Tew, R., Thermo-fluid Experiments Support-
Kuwahara, F., Shirota, M., and Nakayama, A., A numerical
ing Microfabricated Regenerator Development for a Stirling
study of interfacial convective heat transfer coefficient in two-
Space Power Engine, AIAA-2009-4579, in Proc. of the 7th
energy equation model for convection in porous media, Int. J.
Intl. Energy Conversion Engineering Conf., 2009.
Heat Mass Transfer, vol. 44, pp. 1153–1159, 2001.
Lu, W., Zhao, C. Y., and Tassou, S. A., Thermal analysis on Tew, R., Ibrahim, M., Danila, D., Simon, T. W., Mantell, S. C.,
metal-foam filled heat exchangers. Part I: Metal-foam filled Sum, L., Gedeon, D. Kelly, K., Mclean, J., Wood, G., and
pipes, Int. J. Heat Mass Transfer, vol. 49, pp. 2751–2761, Qiu, S., A Microfabricated Involute-Foil Regenerator for Stir-
2006. ling Engines, NASA Technical Report, NASA/CR – 2007-
215006, 2007.
Menter, F. R., Langtry, R. B., Likki, S. R., Suzen, Y. B.,
Huang, P. G., and Volker, S., A correlation based transi- Vafai, K. and Tien, C. L., Boundary and inertial effects on flow
tion model using local variables. Part I: Model formulation, and heat transfer in porous media, Int. J. Heat Mass Transfer,
ASME-GT2004-53452, 2004. vol. 24, pp. 195–203, 1981.
Nakayama, A., Kuwahara, F., Umemoto, T. and Hayashi, T., Xu, C., Song, Z., Chen, L., and Zhen, Y., Numerical investiga-
Heat and fluid flow within an anisotropic porous medium, tion on porous media heat transfer in a solar tower receiver,
Transactions of ASME, vol. 124, pp. 746–753, 2002. Renewable Energy, vol. 36, pp. 1138–1144, 2011.
Nakayama, A., Kuwahara, F., and Hayashi, T., Numerical mod- Zhang, C., Shirazi, F. A., Yan, B., Terrence, W. S., Li, P. Y.,
eling for three-dimensional heat and fluid flow through a bank and Van De Ven, J., Design of an Interrupted-Plate Heat Ex-
of cylinders in yaw, J. Fluid Mech., vol. 498, pp. 139–159, changer Used in a Liquid-Piston Compression Chamber for
2004. Compressed Air Energy Storage, in Proc. of the ASME 2013
Nakayama, A., Ando, K., Yang, C., Sano, Y., Kuwahara, F., and Summer Heat Transfer Conference, 2013.
Liu, J., A study on interstitial heat transfer in consolidated Zhang, C., Wiebedink, J. H., Shirazi, A. F., Yan, B., Si-
and unconsolidated porous media, Heat Mass Transfer, vol. mon, T. W., and Li, P. Y., Numerical investigation of metal-
45, no. 11, pp. 1365–1372, 2009. foam filled liquid piston compressor using a two-energy
Patankar, S. V. and Spalding, D. B., A calculation procedure equation formulation based on experimentally validated mod-
for heat, mass and momentum transfer in three-dimensional els, in Proc. of the ASME 2013 Intl. Mechanical Engineering
parabolic flows, Int. J. Heat Mass Transfer, vol. 15, no. 10, Congress and Exposition, Houston, TX, 2013.

Special Topics & Reviews in Porous Media — An International Journal

You might also like