INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN FLUIDS, VOL.

25, 659±677 (1997)

PERIODIC FLOW AND HEAT TRANSFER USING

UNSTRUCTURED MESHES

J. Y. MURTHY* AND S. MATHUR

Fluent Inc., 10 Cavendish Court, Centerra Resource Park, Lebanon, NH 03766, U.S.A.

SUMMARY
A numerical scheme has been developed for computing ¯uid ¯ow and heat transfer in periodically repeating
geometries. Unstructured solution-adaptive meshes are used in a cell-centred ®nite volume formulation. The
SIMPLE algorithm is used for pressure±velocity coupling. For periodic ¯ows the static pressure is decomposed
into a periodic component and one that varies linearly in the streamwise direction. The latter is computed from
the imposition of overall mass balance at the periodic boundary. A subiteration between the periodic pressure
correction equation and the correction to the linear component is used. For heat transfer a formulation using the
physical rather than the scaled temperature is employed. The scheme is applied to both laminar and turbulent
computations of periodic ¯ow and heat transfer in a variety of heat exchanger geometries; comparison with
published computations and experimental data is found to be satisfactory. # 1997 John Wiley & Sons, Ltd.
Int. J. Numer. Meth. Fluids, 25: 659±677 (1997)
No. of Figures: 14. No. of Tables: 4. No. of References: 30.

KEY WORDS: periodic ¯ow; periodic heat transfer; unstructured mesh; ®nite volume; pressure-based; heat exchangers

1. INTRODUCTION
Periodic ¯ow and heat transfer occur widely in a variety of industrial applications. In compact heat
exchangers, for example, it is common to augment heat transfer by using arrays of ®ns.1 In
evaporators, elaborate beading and embossment patterns are frequently used to increase surface area
and to promote mixing.2 In condenser tubes, periodically interrupted ®ns are sometimes used to
destroy the integrity of the refrigerant ®lm and promote heat transfer. In other areas such as
electronics cooling, electronic components are frequently modelled as in®nite arrays of heat
generating blocks to facilitate analysis.3 Periodically arranged turbulators are used in turbine blade
cooling passages to enhance heat transfer.4 In many applications of industrial interest the geometries
involved are quite complex and the periodically repeating domains are not simple blocks.
Computational ¯uid dynamics (CFD) has been used extensively to analyse periodic ¯ow and heat
transfer. Patankar et al.5 generalized the concept of fully developed ¯ow to ducts whose geometry
varied periodically in the streamwise direction. The analysis included periodicity with speci®ed
temperature and speci®ed ¯ux boundary conditions. For the speci®ed temperature case, an iterative

* Correspondence to: J. Y. Murthy, Fluent Inc., 10 Cavendish Court, Centerra Resource Park, Lebanon, NH 03766, U.S.A.

CCC 0271±2091/97/060659±19 $17.50 Received 12 June 1996

# 1997 John Wiley & Sons, Ltd. Revised 22 October 1996
660 J. Y. MURTHY AND S. MATHUR

procedure for computing the dimensionless streamwise bulk temperature gradient was devised.
Kelkar and Patankar6 used a similar discretization to solve an antiperiodic problem for ¯ow and heat
transfer in a periodic plate channel. Unlike Patankar et al.,5, they used actual rather than scaled
temperature as the dependent variable and greatly simpli®ed the procedure for calculating the
streamwise bulk temperature gradient. Sathyamurthy and Karki7 devised an iterative scheme for the
computation of the streamwise pressure gradient; this was used to compute laminar ¯ow over heated
blocks in an electronics cooling application in Reference 3. Most of these studies used a staggered
mesh arrangement with a control volume discretization. The SIMPLE algorithm8 was used for
pressure velocity coupling.
Most published ®nite volume schemes in this area employ either regular3,5,6 or body-®tted9,10
meshes with a single-block mesh topology. Although a single- or multiblock mesh is adequate for the
simple geometries considered above, it is inconvenient to use when realistic heat exchanger
geometries such as those described in Reference 1 are to be considered. Finite element schemes that
provide greater mesh ¯exibility have also been published (see e.g. Reference 11).
In the last few years, unstructured mesh methods have been developed which greatly ease mesh
restrictions and which permit solution adaption. Unstructured mesh schemes for incompressible ¯ows
were developed during the 1980s using the control volume ®nite element method (CVFEM).12 This
method attempted to marry the geometric ¯exibility of ®nite element schemes with the conservation
property of ®nite volume schemes. More recently, researchers have attempted to extend the
incompressible ¯ow methodology of Reference 8 to unstructured meshes.13±16 Here a conservative
®nite volume formulation is applied to arbitrary polyhedra and the SIMPLE family of algorithms is
used. Though these methods continue to evolve, preliminary computations have been shown to match
standard benchmarks well.
In this paper we use the ®nite volume methodology presented in Reference 16 in developing a
solution procedure for periodic ¯ow and heat transfer. The procedures in References 5 and 6 are
generalized for unstructured solution-adaptive meshes. The method is validated against the published
literature and shown to perform satisfactorily.

2. FLOW AND TEMPERATURE PERIODICITY

~ as shown in Figure 1.
Consider a domain with periodic boundaries separated by a translation vector L,
The two periodic boundaries of the domain are denoted as periodic and periodic shadow for
~ There may
convenience. This domain represents one of a series of periodic modules translated by L.
be other periodic boundaries in the domain, but there is no net in¯ow through any of these

Figure 1. Perodic domain

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 661

boundaries. As in Reference 5, we may write the following relationships for the velocity ui and the
static pressure P at position ~r:
~ ˆ ui ~r ‡ 2L†
ui ~r† ˆ ui ~r ‡ L† ~ ˆ


; 1†

~ ˆ P ~r ‡ L†
P ~r† ÿ P ~r ‡ L† ~ ÿ P ~r ‡ 2L†
~ ˆ


: 2†
For periodic ¯ows the pressure gradient may be decomposed into two partsÐthe gradient of a
periodic component, @p=@xi , and the gradient of a linearly varying component, b^eL :
@P @p
ˆ beL;i ‡ ; 3†
@xi @xi
~
where eL;i is the i-component of the unit vector e^ L in the direction L.
Periodic heat transfer is possible for either Dirichlet or Neumann boundary conditions on
temperature. Here the scaled temperature
T ~r† ÿ Tw
y ~r† ˆ 4†
Tb ~r† ÿ Tw
is periodic. For Dirichlet boundary conditions on temperature we may write
T ~r† ÿ Tw ~ ÿ Tw
T ~r ‡ L† ~ ÿ Tw
T ~r ‡ 2L†
ˆ ˆ ˆ


: 5†
~ ÿ Tw Tb ~r ‡ 2L†
Tb ~r† ÿ Tw Tb ~r ‡ L† ~ ÿ Tw
For Neumann boundary conditions the quantity Tb ÿ Tw is constant across modules. Consequently,
we may simplify equation (5) to write
~ ÿ Tb ~r ‡ L†
T ~r† ÿ Tb ~r† ˆ T ~r ‡ L† ~ ˆ T ~r ‡ 2L†
~ ÿ Tb ~r ‡ 2L†
~ ˆ


: 6†
Here the bulk temperature Tb is de®ned as
RR
ju e jT dA
R RA i L;i ˆ 0; 7†
A jui eL;i j dA

where A is the area of cross-section.

3. GOVERNING EQUATIONS
The equations for conservation of mass, momentum and energy for the incompressible, steady ¯ow of
a Newtonian ¯uid are
@
rui † ˆ 0; 8†
@xi
!
@ @p @ @ui
rui uj † ˆ ÿ ÿ beL;i ‡ m ‡ mt † ‡ Fi ; 9†
@xj @xi @xj @xj
  
@ @ mt @T
rui cp T † ˆ k‡ ‡ Sh : 10†
@xi @xi sh @xi
Fi contains those derivatives of the stress tensor not included in the diffusion term in the momentum
equations; it may also contain the contributions of body forces consistent with the periodicity of the
¯ow. To sustain periodicity, all ¯uid properties must be independent of temperature, though they may

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)
662 J. Y. MURTHY AND S. MATHUR

have a periodic spatial variation. The energy source term Sh must also be temperature-independent
and spatially periodic; non-zero values are permissible for Neumann boundary conditions only.
Turbulence is computed using the high-Reynolds-number k ±e model17
  
@ @ m @k
rui k† ˆ m‡ t ‡ Gk ÿ re; 11†
@xi @xi sk @xi
  
@ @ m @e e e2
rui e† ˆ m‡ t ‡ C1e Gk ÿ C2e r : 12†
@xi @xi se @xi k k
The turbulence viscosity mt is computed as17
mt ˆ Cm rk 2 =e: 13†

The two-layer low-Reynolds number model described in Reference 18 is also implemented.

4. NUMERICAL METHOD
We describe brie¯y important components of the numerical method used to discretize the governing
equations. Details may be found in Reference 16. Cartesian velocities are used as the basis of the ¯ow
formulation. For heat transfer we use the temperature directly as the dependent variable, instead of
using y as done in Reference 5.

4.1. Domain discretization

The domain is discretized into arbitrary unstructured convex polyhedra. All transport variables are
stored at cell centres. The governing equations are integrated over the polyhedron. Consider the
differential equation for transport of a scalar quantitiy f:
@ @ @f
rui f† ˆ G ‡ Sf : 14†
@xi @xi @xi
Integration and discretization about the control volume C0 shown in Figure 2 yields
P P
Jf ff ˆ Df ‡ Sf Dv†0 ; 15†
f f

Figure 2. Control volume

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 663

where Jf is the mass ¯ow rate de®ned as

~ f
V~f :
Jf ˆ rf A 16†
~ f is the outward-pointing area vector at face f . Df is the diffusive ¯ux entering C0 through face f
A
and the summations are over all the faces of the control volume. Thus
~f :
Df ˆ GHf†f
A 17†

For the face f between control volumes C0 and C1 in Figure 2 the diffusive ¯ux can be written as
! !
f1 ÿ f0 A~f
A~f ~f
A
A ~f
Df ˆ Gf 
A
‡ Hf ~ ÿ Hf

e^ s : 18†
ds ~ f
e^ s
A ~ f
e^ s
A

Here e^ s is the unit vector along the line joining the cell centroids. The primary gradient (the ®rst term
on the RHS of the above equation) is treated implicitly. The secondary gradient (the second term on
the RHS) is zero for orthogonal quadrilateral=hexahedral meshes and for equilateral triangular/
tetrahedral meshes. It is computed explicitly. The derivative Hf  at the face is taken to be the average
of the derivatives at the two adjacent cells. The above treatment is the equivalent of a second-order
scheme for structured meshes.
The face value ff used in computing the convective ¯ux can be taken to be the value at the upwind
cell, yielding a ®rst order scheme. To achieve second order accuracy, a linear reconstruction
procedure similar to that of Barth and Jesperson19 is used to write the face value in terms of values in
the upwind cell:
!
ff ˆ fupwind ‡ Hfrupwind
dr ; 19†
!
where dr is the vector directed from the centroid of the upwind cell to the centroid of the face. Hfr is
the reconstruction gradient of f evaluated using the divergence theorem:
a P  ~
Hfr ˆ ff Af †; 20†
Dv f

where f f is the average of the values of f at the adjacent cells and a is a factor used to ensure that the
reconstruction does not introduce local extrema. The limiter proposed by Venkatakrishnan20 is used
in the present work. The gradients used in the diffusion term in equation (18) are computed by
applying the divergence theorem with the reconstructed face values ff .
It is necessary to account for periodicity in computing cell derivatives and reconstruction gradients
and in the limiting procedures. For velocity and the periodic pressure component p this is
straightforward. For temperature, equations (5) and (6) are used to evaluate the temperature in the
cells in neighbouring modules, as explained in the subsections that follow.

4.2. Boundary conditions

In addition to cell centres, f is also stored at boundary face centroids. The boundary diffusion ¯ux
can then be linearized in the same manner as at an interior face, i.e., using equation(18). For the
boundary face in Figure 3 this yields
~f
A
fb ÿ f0 A ~f ~
A
A ~f
Df ˆ Gf ~ ÿ Hf0
e^ b f
‡ Hf0
A ; 21†
db A ~ f
e^ b ~ f
e^ b
A

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)
664 J. Y. MURTHY AND S. MATHUR

Figure 3. Boundary control volume

where fb is the value at the boundary and e^ b is the unit vector directed from the cell centroid to the
boundary face centroid.
For Dirichlet boundary conditions, equation (21) is used to express the boundary diffusion ¯ux as a
linear function of the cell and boundary values of f. It is also used to compute the boundary ¯ux of f
for postprocessing. For Neumann boundary conditions the prescribed boundary ¯ux of f is included
directly in the ¯ux balance for the boundary cell. Equation (21) is then used to compute fb . This
boundary value is used for computing cell gradients during reconstruction as described above, and for
postprocessing.
At periodic boundaries the cells C0 and C1 in Figure 1 are considered to be neighbours and share a
common periodic face. Thus appropriate neighbour values for pressure and velocity are known. The
treatment of temperature is described in a later subsection.
The discretization procedure leads to a set of linear equations relating the value of f at the cell
centre to its cell neighbours:
P
ap fp ˆ anb fnb ‡ b: 22†
nb

Here nb is the number of cell neighbours. The above equation is underrelaxed, as described in
Reference 8. As in Reference 8, ap > 0 in the above equation.

4.3. Linear solver

The number of cell neighbours in equation (22) is arbitrary for an unstructured mesh.
Consequently, familiar line-iterative solvers and their cyclic variants, e.g. the cyclic tridiagonal
matrix algorithm (TDMA),5 cannot be used. Instead, the system is solved using an algebraic multigrid
procedure21 which constructs coarse level equations by clustering a ®ne level cell with a neighbour
for which the in¯uence coef®cient is the highest. The agglomeration procedures are applied across
periodic boundaries. A variety of multigrid cycles such as the V-, W- and Brandt cycles have been
implemented. For the examples in this paper the V-cycle is used for pressure and the Brandt cycle for
all other variables. A Gauss±Seidel relaxation procedure is used at each multigrid level. The
procedure is applied across periodic boundaries.

4.4. Discretization of the momentum equation

The periodic pressure p and the Cartesian components of velocity, ui , are stored at cell centres. The
momentum equation (9) is discretized using the procedure described. The periodic pressure gradient
is computed using linear reconstruction. The linear component of the gradient, b, is either given or
assumed temporarily known and is integrated in the same manner as the term Sf . Boundary
conditions are also treated as described above.

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 665

4.5. Discretization of continuity equation

The continuity equation(8) is integrated over the control volume in Figure 2. A scheme similar to
that of Rhie and Chow22 is used to interpolate pressure to the cell face and to avoid pressure
checkerboarding. We de®ne the face mass ¯ow rate as

rf Dv p1 ÿ p0 A
~
A ~f
~ f
V~f ÿ
Jf ˆ rf A  f
~es † f
ÿ Hp : 23†
a p ds ~ f
e^ s
A

Here V~f is the average value of velocity at the face and a p is the average of the cell coef®cients ap .
The second term is a third-order added dissipation which prevents checkerboarding. The linear
pressure gradient b appears in V~f as the momentum-averaged value of b^eL
A ~ f =jA
~ f j†.
The discrete continuity equation is written as
P
Jf ˆ 0: 24†
f

We use a procedure similar to the SIMPLE algorithm8 to solve for pressure and velocity. If the
momentum equation (9) is solved with guessed values of the pressure gradients @p*=@xi and b*, the
resulting mass ¯ow rate Jf* on face f does not satisfy the discrete continuity equation (24). We
propose a correction Jf0 , i.e.
Jf ˆ Jf* ‡ Jf0 ; 25†

such that the corrected ¯ow rate Jf satis®es equation (24). Further, we postulate that
Jf0 ˆ ÿd1;f p01 ÿ p00 † ÿ d2;f b0 ; 26†

where

rf Dv ~f
A
A ~f
d1;f ˆ ; 27†
a p ds ~ f
e^ s
A

rf Dv
d2;f ˆ ~ f
e^ L :
A 28†
a p

Substituting equations (25) and (26) into equation (24) yields the pressure correction equation
P P
ap p0p ˆ anb p0nb ‡ b0 d2;f ‡ b: 29†
nb f
P P
Here, as in Reference 8, anb ˆ d1;f and ap ˆ nb anb . The term b is the net mass in¯ow f Jf*. We
see that the correction to the linear pressure gradient, b0 , appears in the correction equation for the
periodic pressure p.

4.6. Computation of b
Periodic ¯ows may be computed either for a speci®ed streamwise pressure gradient b or for a
speci®ed mass ¯ow rate M . If b is given, it is used as a known quantity in equation (9), and b0 is set
equal to zero in equations (26) and (29). However, for many problems of practical interest, M is given
and b must be computed.

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)
666 J. Y. MURTHY AND S. MATHUR

In order to ®nd b, we enforce overall mass balance. If the momentum equation (9) is solved with a
guessed value b*, the resulting mass ¯ow rate through the periodic boundary,
P
M* ˆ Jf*; 30†
pf

does not satisfy overall mass balance. We require that the sum of all corrected face ¯ow rates through
the periodic boundary be equal to the desired ¯ow rate M:
P
Jf ˆ M: 31†
pf

Here, pf denotes the set of periodic faces. Using (25) and (26), we may write
P
M ÿ M* ‡ d1;f p01 ÿ p00 †
pf
b0 ˆ ÿ P : 32†
d2;f
pf

p00 p01
The corrections and are associated with the cells that lie on either side of the periodic boundary.
Equations (29) and (32) are solved alternately until convergence.
Once the corrections p0 and b0 are obtained, the face ¯ow rate Jf* is corrected using equation (25).
Cell pressure and b are corrected using
b ˆ b* ‡ ab b0 ; 33†

pp ˆ pp* ‡ ap p0p : 34†

For the co-located scheme used here, it is useful to correct cell velocities u*i computed from the
momentum equation
P
Af;i p0f † ‡ Dvb0 eL;i
f
ui ˆ u*i ÿ : 35†
aip
Here, aip is the centre coef®cient of the momentum equation in the direction i and Af ;i is the i-
component of the face area vector A ~f .
The procedure for the computation of b described above is different from those used in the
literature.3,5 In Reference 5 it is recognized that the Reynolds number is the single governing
parameter in the problem; for every b there exists a unique Reynolds number (and vice versa).
Consequently, a physical property such as viscosity is iteratively changed to impose the correct
Reynolds number; the resulting b, suitably non-dimensionalized, is the desired dimensionless
streamwise pressure gradient. Kelkar et al.3 describe an iterative scheme for b whereby the velocities
ui are scaled by a factor g ˆ M=M * every iteration to match the imposed ¯ow rate; the factor b is also
scaled by ga where a is an underrelaxation factor, typically chosen to be less than unity. The
procedure adopted here is similar to that used in parabolized Navier±Stokes solvers to compute the
streamwise pressure gradient.23 It is similar to the procedure described by Kelkar24 for single-block
structured meshes. Here the parameter b is the streamwise pressure gradient, assumed to be aligned
with one of the mesh directions, x, of the x; Z† system, and M* in equation (31) is an average over all
x ˆ constant grid lines.

4.7. Computation of temperature ®eld

The energy equation (10) is discretized using the procedures described previously. Temperature is
used as the dependent variable. For cells with periodic or periodic shadow faces, equation (22)

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 667

requires temperature in adjacent modules. We describe below the procedure used to obtain neighbour
temperature.

Dirichlet boundary condition. Here the (same) temperature Tw is speci®ed on all walls in the
domain. For this case the procedure adopted is similar to that in Reference 6. Each iteration of the
calculation procedure we compute the factor a de®ned as
Tb ÿ Tw †pf
aˆ : 36†
Tb ÿ Tw †psf
Here pf denotes the set of periodic faces in Figure1; psf denotes the periodic shadow faces. The bulk
temperature may be considered given at the in¯ow periodic boundary, say pf; thus Tb †pf is known.
The bulk temperature at the out¯ow boundary, Tb †psf , may be found from equation (7) using existing
values of T x; y†. Thus a is nominally known and the temperatures T0 and T2 are related by
 
T0 1
T 2 ˆ ‡ 1 ‡ Tw : 37†
a a
Similarly, T1 and T3 are related by
T3 ˆ aT1 ‡ 1 ÿ a†Tw : 38†
The relationships (37) and (38) are used in an implicit manner and modify the coef®cients anb and
ap in equation (22). The Dirichlet problem is non-linear, since the eigenvalue a is not known a priori
and must be found iteratively as part of the solution.

Neumann boundary condition. Consider the case where the heat ¯ux q00 x; y† enters the domain at
the walls; a heat source Sh x; y† may also be present. We may compute the change in bulk temperature
across the domain, DTb ˆ Tb †psf ÿ Tb †pf by performing a heat balance for the module:
P 00 P
qf Af ‡ DvSh
wf c
DTb ˆ P : 39†
cp Jf
pf

The heat ¯ux is summed over all wall faces (wf) in the domain. The heat source is summed over all
cells (c) in the domain. The relationship between the temperatures T0 and T2 may be written as
T2 ˆ T0 ‡ DTb : 40†
Similarly, the temperatures T1 and T3 are related by
T3 ˆ T1 ÿ DTb : 41†
These relations are incorporated in an implicit manner in equation (22), and modify the term b.
Unlike the case of Dirichlet boundary conditions, the speci®ed ¯ux problem is a linear one.
Furthermore, the formulation given above admits T and T ‡ C as solutions, where C is an arbitrary
constant. In practice the temperature level is determined by the speci®ed inlet bulk temperature. To
ensure the stability of the multigrid linear solver to round-off, we discard one discrete equation in the
algebraic system (22) and replace it with a ®xed temperature. After the temperature solution is
obtained, all temperatures are scaled using equation (4) to the desired inlet bulk temperature.

4.8. Overall solution procedure

The solution procedure for computing periodic ¯ow is summarized below.

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)
668 J. Y. MURTHY AND S. MATHUR

1. Solve the momentum equations with current values of pressure and the streamwise gradient, p*
and b*.
2. Compute the face mass ¯ow rates Jf*
3. Iterate between the p0 -equation (29) and the b-correction equation (32) until p0 and b0 stabilize.
4. Correct the face ¯ow rate Jf* to obtain Jf . This will satisfy both local and global continuity.
Correct p* and b*, as well as the cell-centred velocity u*.i
5. Check for convergence. If unconverged, repeat steps 1±4.
Once a converged ¯ow solution is obtained, the temperature may be computed. The solution
procedure for computing the temperature ®eld is summarized below.
1. For the Dirichlet case, compute the eigenvalue a. For the Neumann case, compute the bulk
temperature difference DTb ; the latter need be computed only once for the entire calculation.
2. Use a (or DTb ) to compute the modi®ed coef®cients and source terms in equation (22).
3. Solve the temperature equation.
4. For Neumann boundary conditons, repeat steps 1±3 until convergence if secondary gradients
exist; else the solution is complete. For Dirichlet boundary conditions, repeat steps 1±3 until
convergence.
5. For Neumann boundary conditons, scale the temperature to obtain the desired bulk temperature.

4.9. Grid adaption

The solution algorithm described in previous subsections can handle both the traditional h-
re®nement25 as well as grids with `hanging nodes' resulting from non-conformal grid adaption. In the
latter case, cells marked for adaption are subdivided by introducing midpoint nodes at edges and=or
cell and face centres. Since the discretization scheme can be used for arbitrary polyhedra, the non-
conformal interfaces created by hanging node adaption require no special treatment.

4.10. Non-periodic meshes

Most published work for periodic ¯ow has required that the underlying mesh be periodic. One
advantage of admitting arbitrary polyhedra is that non-periodic meshes can be handled easily.
Consider the non-periodic mesh shown in Figure 4. The periodic boundary is non-conformal since the
faces of cells on either side are not aligned. However, if the non-conformal faces are split into aligned
subfaces, the cells on either side of the boundary can be considered to be n-sided polyhedra (a±b±c±
d±e in Figure 4, for example). The procedures outlined above can then be used without modi®cation.
Thus it is not necessary to have access to mesh generators capable of generating periodic meshes.

5. RESULTS
5.1. Fully developed ¯ow in a triangular duct
Flow and heat transfer in triangular ducts are of considerable importance in louvred ®n heat
exchangers and extensive analyses of this geometry have been published.26,27 The equations
governing fully developed ¯ow and heat transfer are two-dimensional; the axial velocity and a scaled
temperature may be solved on the triangle to obtain the velocity and temperature ®elds. Our intent is
to validate the method developed above; consequently, we compute three-dimensional ¯ow and heat
transfer using a prismatic domain bounded by two periodic boundaries, as shown in Figure 5. The
ducts considered are isosceles; the apex angle is 2a.

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 669

Figure 4. Non-periodic mesh in periodic domain

Because of symmetry, only one half of the cross-section is considered. The domain is discretized
using prismatic cells; a typical mesh is shown in Figure 6. The axial mesh is deliberately kept only
®ve cells deep because of the fully developed nature of the ¯ow. A ratio of triangle base to length of
5
0 is used. Mesh sizes range from 648 to 3424 cells in the cross-section, depending on the apex
angle. The mesh size employed in each case is listed in Tables I and III respectively.
Computations are performed for 2a ˆ 20 ; 60 and 90 . A ¯ow rate is prescribed and the
corresponding pressure drop computed using the iterative procedure described above. Both laminar
and turbulent ¯ows are considered. For turbulence the two-layer k±e model described in Reference 18
is used. Heat transfer is computed only for laminar ¯ow. Both the speci®ed temperature and the
speci®ed heat-¯ux boundary conditions are considered. For the latter case, the heat ¯ux is assumed to
be axially and peripherally uniform.
Tables I and II present the computed fRe product and Nusselt number respectively for laminar
¯ow. NuT corresponds to the Nusselt number for the speci®ed temperature case; NuH corresponds to
that for the speci®ed heat ¯ux case. The computed results compare well with values compiled by
Shah and London27 from a variety of variational, least squares and ®nite difference solutions in the
literature. A typical plot of b versus iteration is shown in Figure 7 for 2a ˆ 60 ; b is found to reach
99% of its converged value within 100 iterations. The local velocity distribution for 2a ˆ 40 is
plotted in Figure 8 for different radial lines (corresponding to different values of the angle y shown in

Figure 5. Triangular ductÐschematic diagram

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)
670 J. Y. MURTHY AND S. MATHUR

Figure 6. Triangular ductÐprismatic mesh

Figure 5). The comparison with a point-matching solution obtained by Sparrow28 in Figure 8 is good.
For turbulent ¯ow the correlation parameter
C ˆ 4fRe0
25 42†
is tabulated for different values of the apex angle in Table III for Re ˆ 10,000. Experimental values
are interpolated from the data presented in Reference 26. Computed values compare reasonably well
with the data.

5.2. Flow over a tube bundle

Transverse ¯ow over an in-line tube bundle is computed using the methodology described above.
The geometry is shown in Figure 9. The ¯ow is laminar and is governed by the Reynolds number Re
which is de®ned as
Re ˆ rDUm =m; 43†
where Um is the average velocity in the minimum transverse gap between the tubes. Computations
are done for two values of the Reynolds number, Re ˆ 10 and 100. A single geometry corresponding
to ST =D ˆ SL =D ˆ 1
25 is used.
A triangular mesh of 3160 cells is used. Hanging node adaption is used to concentrate cells in
regions of high gradient; the resulting mesh is shown in Figure 10. The mass ¯ow rate corresponding
to the desired Reynolds number is prescribed and the pressure coef®cient, de®ned as
DP
Cp ˆ ; 44†
0
5rUm2

Table I. Triangular ductÐlaminar ¯ow friction factors

2a (deg) Cross-section grid size Computed Shah and London Difference (%)
20 3424 12
90 12
822 0
65
40 1584 13
31 13
22 0
66
60 1080 13
40 13
33 0
51
90 648 13
174 13
153 0
16

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 671
Table II. Triangular ductÐlaminar ¯ow Nusselt number

2a (deg) Computed Shah and London Difference (%)

NuT NuH NuT NuH NuT NuH

20 2
07 0
375 2
0 0
366 3
5 2
5
40 2
43 1
42 2
39 1
38 1
7 2
6
60 2,51 1,91 2
47 1
89 1
6 0
8
90 2
38 1
37 2
34 1
34 1
5 2
0

Figure 7. Triangular ductÐconvergence history for b

Figure 8. Triangular ductÐaxial velocity pro®les for 2a ˆ 40

Table III. Triangular ductÐcorrelation parameter for turbulent ¯ow

2a (deg) Cross-section Computed Experimental Difference (%)

grid size

20 3424 0.278 0.285 2.4

40 1584 0.283 0.301 6.0
60 1080 0.295 0.301* 2.0

* Schiller's data

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)
672 J. Y. MURTHY AND S. MATHUR

Figure 9. Tube bundleÐshematic diagram

Figure 10. Tube bundleÐadapated mesh

computed. Here DP is the pressure difference resulting from the axial pressure gradient b across a
periodic module.
Table IV compares computed values of Cp with those measured by Antonopoulos.29. The
agreement with experimental data is good.

5.3. Flow and heat transfer in a serrated channel

Turbulent ¯ow and heat transfer are computed for periodic ¯ow in a serrated channel. The
geometry is shown in Figure 11. Flow enters the domain at a Reynolds number of 1
2 6 104 (based
on the mid-channel height Hm and the average mid-channel velocity). A speci®ed heat ¯ux condition
is imposed on both the ¯at and serrated walls. A single geometry corresponding to Hs =Hm ˆ 1
0,
Ls =Hm ˆ 8 and L=Hm ˆ 8
075 is considered. The objective is to predict periodic ¯ow and heat
transfer and to compare predictions with measurements in Reference 30.

Table IV. Tube bundleÐcomparison of computed Cp with

experiment

Re Computed Experimental Difference (%)

10 17.77 17.96 1.06
100 1.949 1.946 0.15

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 673

Figure 11. Serrated channelÐschematic diagram

The high-Reynolds-number k±e model is employed.17. The mass ¯ow rate corresponding to the
desired Re is prescribed and the axial pressure gradient computed as a part of the solution. The
calculation domain is shown in Figure 12. It consists of quadrilateral mesh of 2400 cells with non-
conformal periodic boundaries.
The computed axial velocity U =Um is compared with experimental data from Reference 30 in
Figure 13. Figure 14 shows a comparison of the local Nusselt number Nu, with experimental data. Nu
is de®ned as

q00 Hm
Nu ˆ : 45†
k Tw ÿ Tb †

The comparison with data is reasonable and of the same order as the k±e model predictions in
Reference 30 (not shown).

Figure 12. Serrated channelÐ(a) mesh and (b) detail of periodic boundary

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)
674 J. Y. MURTHY AND S. MATHUR

Figure 13. Serrated channelÐaxial velocity comparisons: , experiment; Ð, computed

Figure 14. Serrated channelÐlocal Nusselt number distribution: , experiment; Ð, computed

6. CLOSURE
A novel conservative ®nite volume scheme is used to develop a solution procedure for periodic ¯ow
and heat transfer in complex geometries. The computational scheme employs unstructured convex
polyhedra and admits solution adaption. Previously published formulations for periodic ¯ow are
generalized to unstructured meshes. A correction procedure is used for computing the streamwise
pressure gradient subject to a mass ¯ow rate constraint. The method is applied to ¯ow and heat
transfer in a variety of heat exchanger geometries and its geometric ¯exibility demonstrated through
the use of prismatic, triangular and quadrilateral cells with hanging node adaption and non-conformal

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 675

periodic interfaces. Both laminar and turbulent ¯ow problems are computed and good agreement with
published numerical and experimental data is found.

ACKNOWLEDGEMENTS

We wish to acknowledge the contributions of N. Carter, D. Choudhury, S. E. Kim, J. Maruszewski

and G. Spragle of Fluent Inc. to various aspects of this work. The use of Fluent Inc.'s FLUENT/UNS,
PreBFC and TGrid codes is gratefully acknowledged.

APPENDIX: NOMENCLATURE
anb neighbour coef®cient
ap neighbour coef®cient
A area
A~ area vector
cp speci®c heat at constant pressure
D diffusive ¯ux
Fi body force in direction i
Gk turbulence production
J mass ¯ow rate
k turbulence kinetic energy, thermal conductivity
L~ periodic translation vector
M imposed mass ¯ow rate
p periodic component of static pressure
P static pressure
q00 heat ¯ux
r position vector
Sh heat source
Sf source of f
T temperature
Tb bulk temperature
Tw wall temperature
ui velocity component in Cartesian direction i
xi co-ordinate direction

Greek letters
a bulk temperature ratio
ap ; ab underrelaxation factors
b streamwise pressure gradient
G diffusion coef®cient
DTb bulk temperature rise across module
Dv volume of control volume
e turbulence dissipation rate
y periodic temperature
m molecular viscosity
mt turbulence viscosity
r density

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)
676 J. Y. MURTHY AND S. MATHUR

sh turbulence Prandtl number for temperature

sk turbulence Prandtl number for k
se turbulence Prandtl number for e
f transported scalar

Subscripts and superscripts

f face
nb neighbour cells
pf periodic face
psf periodic shadow face
( )* value after solving momentum equation
†0 correction

Other
P
Pf summation over all faces of control volume
Pnb summation over all cell neighbours
pf summation over all faces on periodic boundary

REFERENCES
1. W. M. Kays and A. L. London, Compact Heat Exchangers, 3rd edn, McGraw-Hill, New York, 1984.
2. T. Ohara and T. Takahashi, `High performance evaporator development', SAE Tech. Paper 880047, 1988.
3. K. M. Kelkar, P. S. Sathyamurthy, K. C. Karki and S. V. Patankar, `Solutions of laminar ¯ow and heat transfer over an
array of heated blocks', ASME Paper 93-WA=EEP, 1993.
4. Y. M. Zhang, W. Z. Gu and J. C. Han, `Heat transfer and friction in rectangular channels with ribbed or ribbed±grooved
walls', J. Heat Transfer, 116, 58±65 (1994).
5. S. V. Patankar, C. H. Liu and E. M. Sparrow, `Fully developed ¯ow and heat transfer in ducts having streamwise-periodic
variations of cross-sectional area', J. Heat Transfer, 99, 180±186 (1977).
6. K. M. Kelkar and S. V. Patankar, `Numerical prediction of ¯ow and heat transfer in a parallel plate channel with staggered
®ns', J. Heat Transfer, 109, 25±30 (1987).
7. P. S. Sathyamurthy and K. C. Karki, Internal Rep., Innovative Research Inc., 1992.
8. S. V. Patankar, Numerical Heat Transfer and Fluid Flow, McGraw-Hill, New York, 1980.
9. R. C. Xin and W. Q. Tao, `Numerical prediction of laminar ¯ow and heat transfer in wavy channels of uniform cross-
sectional area', Numer. Heat Transfer, 14, 465±481 (1988).
10. D. Choudhury and K. C. Karki, `Calculation of fully developed ¯ow and heat transfer in streamwise-periodic dimpled
channels', J. Thermophys. Heat Transfer, 5, 81±87 (1991).
11. G. Segal, K. Vuik and K. Kassels, `On the implementation of symmetric and antisymmetric periodic boundary conditions
for incompressible ¯ow', Int. j. numer. methods ¯uids, 18, 1153±1165 (1994).
12. B. R. Baliga and S. V. Patankar, `A control-volume ®nite element method for two-dimensional ¯uid ¯ow and heat
transfer', Numer. Heat Transfer, 6, 245±261 (1983).
13. Y. Jiang and A. J. Przekwas, `Implicit, pressure-based incompressible Navier±Stokes equations solver for unstructured
meshes', AIAA Paper 94±0305, 1994.
14. I. Demirdzic and S. Muzaferija, `Numerical method for coupled ¯uid ¯ow, heat transfer and stress analysis using
unstructured moving meshes with cells of arbitrary topology', Comput. Methods Appl. Mech. Eng., 125, 235±255 (1995).
15. L. Davidson, `A pressure correction method for unstructured meshes with arbitrary control volumes', Int. j. numer. methods
¯uids, 22, 265±281 (1996).
16. S. R. Mathur and J. Y. Murthy, `A pressure based method for unstructured meshes', Numer. Heat Transfer, 31, 195±216
(1997).
17. B. E. Launder and D. B. Spalding, Lectures in Mathematical Models of Turbulence, Academic, London, 1972.
18. S. E. Kim, `Application of a two-layer zonal approach to computations of complex turbulent ¯ows', in Turbulent Flows,
FED Vol. 208, ASME, New York, 1995, pp. 75±80.
19. T. J. Barth and D. C. Jespersen, `The design and application of upwind schemes on unstructures meshes', AIAA Paper 89-
0366, 1989.
20. V. Venkatakrishnan, `On the accuracy of limiters and convergence to steady state solutions', AIAA Paper 93-0880, 1993.

INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997) # 1997 by John Wiley & Sons, Ltd.
 PERIODIC FLOW AND HEAT TRANSFER 677

21. B. R. Hutchinson and G. D. Raithby, `A multigrid method based on the additive correction strategy', Numer. Heat
Transfer, 9, 511±537 (1986).
22. C. M. Rhie and W. L. Chow, `Numerical study of the turbulent ¯ow past an airfoil with trailing edge separation', AIAA J.,
21, 1523±1532 (1983).
23. S. V. Patankar and D. B. Spalding, `A calculation procedure for heat, mass and momentum transfer in three-dimensional
parabolic ¯ows', Int. J. Heat Mass Transfer, 15, 1787±1805 (1972).
24. K. M. Kelkar, personal communication, 1992.
25. W. A. Smith and K. R. Blake, `Multigrid solution of internal ¯ows using unstructured solution-adaptive meshes', Tech.
Rep. NAS3-25785, NASA Lewis Research Center, Cleveland, OH, 1992.
26. W. M. Kays and H. C. Perkins, in Handbook of Heat Transfer, McGraw-Hill, New York, 1973, pp. 7-110±7-138.
27. R. K. Shah and A. L. London, Laminar Flow Forced Convection in Ducts, Academic, New York, 1978.
28. E. M. Sparrow, `Laminar ¯ow in isosceles triangular ducts', AlChE J., 8, 599±604 (1962).
29. K. A. Antonopoulos, `The prediction of turbulent inclined ¯ow in rod bundles', Comput. Fluids, 14, 361±378 (1986).
30. S. Obi, K. P. Kobayashi, A. M. Bethancourt, H. Yoshida, T. Asano and R. Echigo, `Experimental and computational study
of turbulent heat transfer characteristics in serrated channel ¯ow', Int. J. Heat Fluid Flow, 16, 398±404 (1995).

# 1997 by John Wiley & Sons, Ltd. INT. J. NUMER. METH. FLUIDS, VOL. 25: 659±677 (1997)