International Journal of Thermal Sciences 137 (2019) 64–74

Contents lists available at ScienceDirect

International Journal of Thermal Sciences

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

Heat transfer enhancement in panel type radiators using delta-wing vortex T

generators
Luciano Garelli∗, Gustavo Ríos Rodriguez, Jonathan J. Dorella, Mario A. Storti
Centro de Investigación de Métodos Computacionales, CIMEC (UNL - CONICET), 3000, Santa Fe, Argentina

ARTICLE INFO ABSTRACT

Keywords: In this paper, a three-dimensional numerical simulation is used to analyze the performance of delta-wing vortex
Heat transfer enhancement generators for enhancing the heat exchange in panel type radiators, which are widely used in electric power
Vortex generators transformers. The study is focused on natural convection and buoyancy-driven flows, which are common
Panel type radiators working conditions for this type of heat exchanger. First, the performance of a single delta wing between parallel
Streamwise vortices
vertical plates is analyzed to establish the best combination of characteristic parameters to obtain the highest
thermal enhancement factor. It is found that separating the vortex generator from the surface of the panel has
positive effects in this sense. Then, with the selected configuration, a set of delta-wing arrays is placed on the
surface of the heat exchanger, and the resulting thermo-fluid dynamic is analyzed. The total heat flux and local/
global heat exchange coefficients are reported. Using these passive devices, the overall heat transfer improves by
12%.

1. Introduction
Vortex generators (VGs) have been widely studied in terms of
aerodynamics, as well as heat and mass transfer. In aeronautics [1,2],
VGs are commonly used as aerodynamic devices to delay flow separa-
tion and improve the handling and stall characteristics of aircraft at low
speeds. Additionally, with aerodynamic ends, they are used in racing
cars to enhance the features of the air flow that arrives at ailerons and
diffusers with the objective of improving the downforce. Regarding
heat exchange enhancement, since 1969, with the work of Johnson [3],
extensive research has been performed to understand the improvement
in heat exchange and pressure losses produced by VGs. In these studies,
the influence of several parameters have been considered, such as the
angle of attack ( AoA ), chord length (c), aspect ratio ( ) and Reynolds
number (Re), among others. However, most of the heat exchange en-
hancement results are for prescribed inlet velocities or fixed Re num-
bers. In the work of Tiggelbeck [4], a comparison of several types of
VGs (delta and rectangular wings, delta and rectangular winglet pairs,
etc.) is presented for 2000 < Re< 9000 (based on the wing chord
length). The wing-type VGs are noted to produce stable and strong
vortices, the heat exchange coefficient can increase considerably over
an area 100 times greater than the vortex generator area, and winglet-
type VGs produce a higher heat transfer, but with a higher friction
coefficient, than wing-type VGs. On the other hand, Fiebig [5] found
that delta wing VGs are the most effective for heat transfer enhance-
ment per unit area when the results are normalized using the relation
between the heated plate and the vortex generator areas. For the Rey-
nolds number range , the experimental work of Fiebig [6] shows that
the performance difference between delta wings and delta winglets is
negligible. In the review by Jacobi [7], it is stated that the performance
of a heat exchanger is, in general, limited by the heat transfer coeffi-
cient of air and that the temperature distribution depends on the ve-
locity field. Additionally, a physical description of the phenomena be-
hind the heat exchange enhancement and pressure losses in terms of the
flow field and wall friction is provided. In the subsequent work of
Gentry [8], new experimental data are obtained for delta-wing VGs for
Re = {600; 800; 1000} , 10 AoA 60 [deg] and 0.8 2 , showing
50% to 60% improvement in heat transfer. In a more recent work,
Joardar [9] carried out a numerical study of heat transfer enhancement
using an array of delta winglets in a heat exchanger with different ar-
rangements, obtaining an overall heat transfer enhancement of 32%
and a similar pressure loss penalty. Numerical simulations were carried
out in Refs. [10–15] to analyze heat transfer enhancement in rectan-
gular and circular channels with different VGs. In Ref. [16], the authors
investigated the performance of planar and curved winglets with and
without punched holes, in laminar and turbulent flow regions. In the
context of applying VGs to heat exchangers, the publications [17,19,20]
investigated the global and local effects of VGs by means of numerical

∗
Corresponding author.
E-mail address: lgarelli@cimec.unl.edu.ar (L. Garelli).

https://doi.org/10.1016/j.ijthermalsci.2018.10.037
Received 27 February 2018; Received in revised form 4 September 2018; Accepted 27 October 2018
1290-0729/ © 2018 Elsevier Masson SAS. All rights reserved.
L. Garelli et al. International Journal of Thermal Sciences 137 (2019) 64–74

Nomenclature c wing chord, [m]

d trailing edge separation from heated wall, [m]
P pressure loss, [Pa ] Gr Grashof number
T temperature difference, [K ] L channel length, [m]
Ap heated plate area, [m2 ] Pr Prandtl number
Cp specific heat at constant pressure, [J(kgK) 1] Ra Rayleigh number
Dh hydraulic diameter, [m] W channel width, [m]
g gravity acceleration, [ms 2 ]
Hc height of the oil channels, [m] Abbreviations
hg global heat transfer coefficient, [Wm 2K 1]
hl local heat exchange coefficient, [Wm 2K 1] AoA wing Angle of Attack, [deg]
Hp spacing between panels, [m] CFD Computational Fluid Dynamics
hwall size of the mesh at the wall FVM Finite Volume Method
H channel height, [m] LIC Line Integral Convolution
h heat exchange coefficient, [Wm 2K 1] RANS Reynolds-Averaged Navier-Stokes
La arrays separation, [m] SST Shear-Stress Transport
Lp length of the panel, [m] TEF Thermal Enhancement Factor
Nu Nusselt number VG Vortex Generator
qw wall heat flux, [Wm 2 ]
q ext extrapolated total heat flux, [W] Greek letters
Re Reynolds number
T+ dimensionless temperature α thermal diffusivity, [m 2s ]
Tw wall temperature, [K ] β air thermal expansion coefficient, [K 1]
T air temperature, [K] wing aspect ratio
uin inlet air velocity, [ms−1] λ thermal conductivity, [W(mK) 1]
Wc width of the oil channels, [m] μ air dynamic viscosity, [Pa s]
Wp width of the panel, [m] u* friction velocity
y+ dimensionless wall distance ν kinematic viscosity, [m 2s ]
f friction factor ρ air density, [kg m 3]
A cross-sectional area of the channel, [m2 ] w wall shear stress
b wing span, [m]

simulations to obtain the thermal enhancement factor (TEF) as defined
in Ref. [17], which considers the increase in the heat exchange coeffi-
cient penalized by the increase in the pressure loss, using a prescribed
inlet velocity or Re number.
In this numerical study, we focus on the usage of VGs in natural
convection and buoyancy-driven flows, which are characteristic of
panel type heat exchangers, with the objective of increasing heat
transfer from the point of view of the air side. Panel type heat ex-
changers are widely used in electric power transformers [21–23],
wherein the performance is limited by the heat transfer coefficient on
the air side. Based on the reviewed literature, a delta-wing type vortex
generator is considered in this study because it provides the best TEF
values in rectangular channels. This is geometrically equivalent to the
channel between two radiator panels and results in the highest local
enhancement ([7]). In the work of Min [24], a first step is given in this
direction. A numerical investigation of rectangular winglets attached to
a radiator panel is performed, obtaining approximately 25% of increase
in the heat exchange coefficient (h ). In this numerical simulation,
symmetric boundary conditions are used in the air channels, by which
air is funneled and inlet/outlet flows do not appear at the sides. This
boundary condition setup is correct for the central air channel (i.e., the
air channel along the center of the panel) due to the mid-plane sym-
metry of the radiator. However, when the full width panel is analyzed,
ambient air enters the panel from the bottom and the sides, as explained
later in Section 4.
The work is organized as follows: the physical model is described in
the first section. Then, a mesh convergence analysis is carried out for
the highest air velocity to establish an appropriate mesh size near the
wall to accurately capture the velocity and thermal gradients. Then, the
performance of a single delta-wing VG is analyzed to establish the op-
timum AoA for typical inlet air velocities ranging from 0.30 uin 1.05
[m/s]. These values are obtained from experimental measurements and
numerical simulations presented in Ref. [23]. Having analyzed the
impact of AoA for different air velocities, an additional parameter, the
clearance (i.e., the separation between the delta wing and the panel
surface), is considered. The separation allows an air passage below the
VG that improves the heat exchange coefficient behind the delta wing.
Finally, delta wings are placed in an array configuration in the air
channel of a radiator panel with a trapezoidal geometry, such as that
described in Refs. [1,23], where the air flow is the result of natural
convection (i.e., buoyancy-driven flow). The temperature distribution
imposed on the radiator panel is obtained from Ref. [23], which is more
realistic than setting a constant value on the entire panel surface.
2. Problem statement
The use of vortex generators, particularly delta-wing type gen-
erators, introduces a secondary flow consisting of two counter-rotating
vortices that propagate streamwise and induce air flow between them
and toward the heated surface. This is denoted as inflow in Fig. 1. The
inflow increases the local heat exchange coefficient and reduces the
thickness of the fluid dynamic and thermal boundary layers. In the
region outside these vortices, the situation is reversed since the hot air
coming from the surface toward the bulk of the flow stream, indicated
as outflow in Fig. 1, reduces the local heat exchange coefficient and
increases the thickness of the fluid dynamic and thermal boundary
layers, as mentioned in Ref. [7]. Additionally, there is a strong inter-
action between the vortices, which shifts both of them toward the he-
ated surface until a stable position is reached.
The location of the vortex, combined with its intensity, is funda-
mental in the heat exchange enhancement, as mentioned in Ref. [8]. If
the vortex is located far from the boundary layer, the temperature
difference across the vortex vanishes, and an advective contribution is
not made. On the other hand, if the vortex is located deep in the

65
L. Garelli et al. International Journal of Thermal Sciences 137 (2019) 64–74

Fig. 1. Schematic view - Characteristic parameters of the problem and flow features.

boundary layer, the size of the vortex is limited by the near-wall sur- continuity equation is given by
face, and its intensity is not enough to produce a strong inflow/outflow.
In between these two extreme scenarios, the optimum location of the ( ui) = 0,
xi (2)
vortex is near the edge of the boundary layer, but the interaction be-
tween the two counter-rotating vortices makes maintaining that loca- and the momentum equation can be written as
tion difficult.
In this numerical study, the local Nu and global Nu numbers will be p ui uj 2 uk
( ui uj) = + µ + ij + gi ( 0)
reported for each configuration analyzed, as well as the TEF and loca- xi xj xi xj xi 3 xk
tion of the vortex center within the boundary layer for several down-
stream positions. After selecting the delta wing position that gives the ( u i u j ),
xi (3)
best performance, delta wing arrays are placed in the air channel be-
tween two radiator panels to obtain a global TEF for natural convection where Einstein notation is used. In Eq. (3), u i u j is the Reynolds stress
and buoyancy-driven flows. tensor that appears as result of the averaging procedure, which has to
be modeled to close the system of governing equations, g is the gravity
acceleration and ij is the Kronecker delta. Finally, the energy equation
3. Numerical model
can be expressed as follows:

To begin the numerical study, a delta-wing VG is placed between T

( ui C p T) =
two parallel plates separated by a distance H= 40 [mm]. xi xi xi (4)
(which is a typical distance between radiator panels in power
transformers [1,23]), of width W= 40 [mm] and length L= 127 [mm]. Density is assumed to be constant in Eq. (3), that is, = 0 , if an
The wing chord of the VG is c= 12.7 [mm], the wing span b= 6.35 [mm], imposed inlet velocity is used as boundary condition to analyze the
and the aspect ratio = 2b/ c= 1. The trailing edge is placed at performance of the delta wing. However, when studying the radiator
x= 2 c= 25.4 [mm] from the inlet. Three different clearance values (i.e., panel the air flows because of buoyancy forces. Therefore, a Boussinesq
the separation distance from the trailing edge to the heated surface) are model is used to take into account the driving force due to a tem-
considered, d= {0; 3; 5} [mm]. Additionally, three angles of attack are perature-dependent density. Consequently, the density variation is ap-
considered for each clearance value, namely AoA = {30; 40; 50} [deg]. proximated in Eq. (3) as
The Re number is calculated using the hydraulic diameter Dh as char- ( (5)
0 )g i 0 Tg i,
acteristic length and uin as reference velocity.
where β is the thermal expansion coefficient of the air
0 uin D h ( = 0.0033 [1/K]) and T is the temperature difference which drives
Re = .
µ (1) the buoyancy force. In this work, the maximum temperature difference
between the heated wall and the ambient is (Tw T0) 35 [K], which is
For the case of parallel plates Dh = 2 H= 80 [mm]. If a reference
a typical value for electric transformer radiators. Other non-dimen-
value T0 = 303 [K]
sional numbers important for the analysis are the Grashof (Gr ), Prandtl
is taken for the temperature, then the reference air density and
(Pr ) and Rayleigh (Ra ),
dynamic viscosity are 0 = 1.17 [kg/m3] and µ = 1.86 10 5 [Pa s], re-
spectively. The Re numbers for the air velocities considered are g TL3
Gr = ,
1519 Re 5316. As reported by White [25], fully turbulent flow is 2 (6)
ensured at Re 4000 , and the flow becomes unstable at Re 2000 .
µC p
Additionally, in the work of Hanks [26], several experimental research Pr = = ,
studies were reviewed, and a critical Re number range of (7)
2360 Re 2960 was defined. Finally, in the work of Oneissi et al.
g TL3
[17], a turbulence model was used when Re 2800 . Ra = = Gr Pr,
µ (8)
In this study, the air flow is assumed to be stationary and three
dimensional, and the fluid is Newtonian. The governing equations are where the kinematic viscosity is = µ/ 0 = 1.59 10 5 [m2/s], thermal
the Reynolds-Averaged Navier-Stokes (RANS) equations; thus, the conductivity is = 0.0262 [W/mK], specific heat is C p = 1007 [J/kg K]

66
L. Garelli et al. International Journal of Thermal Sciences 137 (2019) 64–74

and thermal diffusivity is = / C p = 2.22 10 5 [m2/s]. The character- 4. Results and discussion
istic length of the panel is L p = 1.524 [m]. With these physical properties
defined, the following values are computed for the non-dimensional 4.1. Mesh convergence analysis
numbers: Gr = 2.74 105 , Pr = 0.71 and Ra = 1.94 105. As mentioned in
Ref. [25], if 105 < Ra< 109 , then natural convection laminar flow oc- In this section, a mesh convergence analysis is carried out to es-
curs, with the transition to turbulent flow occurring within the range of tablish the appropriate mesh size near the wall to accurately capture the
10 8 < Ra< 1010 for parallel plates. velocity and thermal gradients. The flow domain consists of the air
Eqs. (2)–(4) are discretized in space using a co-located finite volume channel between parallel plates without VGs. The local (Eq. (16)) and
method (FVM) [27,28] implemented in the open source computational global (Eq. (17)) Nu numbers are compared with the Stephan correla-
fluid dynamic (CFD) code Code_Saturne [29,30]. A 3D segregated solver tions [36,37], established for a flow between parallel plates. These
is used with a SIMPLEC (semi-implicit method for pressure-linked correlations are used as reference for the air channel,
equations consistent) algorithm for coupling between velocity and
0.024x * 1.14 (0.0179Pr 0.17x * 0.64 0.14)
pressure [31]. A second-order linear upwind (SOLU) method [32–34] is Nux = 7.55 + ,
considered for the convective flux. For turbulence modeling, shear- (1 + 0.0358Pr 0.17x * 0.64)2 (16)
stress transport (SST) is used [18], where two additional equa- 0. 024x * 1.14
tions have to be solved, one for the turbulent kinetic energy κ and one ¯ = 7. 55 +
Nu ,
1 + 0. 0358Pr 0.17x * 0.64 (17)
for the specific dissipation ω. This turbulence model has been used by
where x* =
x/Dh
Oneissi et al. [17], showing good agreement with similar flows. A de- is defined.
Re Pr
tailed description of the equation discretization and implementation To evaluate the mesh size near the wall, the dimensionless wall
can be found in the Code_Saturne: Theory guide [35]. The boundary distance y+ is defined as
conditions adopted are described in the following sections. u* y w
The numerical simulations allow the analysis of global and local y+ ; u* ,
(18)
quantities in order to evaluate the vortex generator performance, as
described in Refs. [17,20]. where is the friction velocity and w is the wall shear stress.
u* T+ is the
The local heat transfer coefficient hl is defined as follows: dimensionless temperature, defined as follows:
T Tw Tb|x qw
qw T+ = ; T* .
hl = = n
, T* C p u* (19)
T (Tw Tb |x ) (9)
To accurately compute the pressure losses, mainly produced by
where viscous effects (wall shear stress), the cell center nearest the wall has to
be in the viscous sublayer (y+ < 5 ). For thermal heat flux, the ratio
A
uTdA A
uTdA 1 between the fluid boundary layer thickness (δ) and the thermal
Tb |x = = = uTdA,
m A
udA uA
¯ A
(10) boundary layer thickness ( T ) is / T Pr1/3 [39,40]. As stated in the
work of Arpaci [39], considering Pr 1, the heat diffusion is expected
and the local Nusselt (Nu ) number is to be dominated by molecular diffusion (conduction layer) in the near-
wall thermal layer and by the turbulent contribution in the core. Con-
Dh hl Dh qw
Nu = = , sequently, to accurately calculate the thermal flux in the near wall
(Tw Tb |x ) (11)
layer, y+ < 13.2 must hold, as stated by Arpaci [39] and Stolpa [38];
with qw being the thermal flux and A the cross-sectional area of the thus, in this region, T+ = Pr y+.
channel. For mesh convergence analysis, an air velocity uin = 1.05 [m/s] is set
Finally, an average or global heat exchange coefficient hg is ob- at the inlet boundary. This velocity magnitude corresponds to the lar-
tained for the panel surface, gest Reynolds number mentioned in section 3, i.e., Re = 5316. A re-
ference pressure value is set at the outlet boundary, the no-slip condi-
1 tion is imposed for the lower surface, and symmetry conditions are
hg = hl dAp,
Ap assumed at the top and lateral surfaces. The inlet temperature is
Ap (12)
Tin = 303 [K], and the temperature of the heated wall is set to
and the average Nu number is computed as follows: Tw = 343 [K].
The problem is solved using three meshes (M1; M2; M3), which
Dh hg have different refinement levels. These meshes are hexa-dominant to
Nu =
(13) avoid mesh quality issues, such as cell non-orthogonality, cell skewness
and cell offset [35], and are generated with SnappyHexMesh [41]. The
In problems where flow is driven by buoyancy forces, the inclusion mesh generation process starts from a hexahedral base mesh, known as
of the VG produces a pressure loss, which decreases the flow velocity. level 0 mesh, which sets the initial cell size. Then, the refinement levels
This can reduce the heat transfer coefficient; hence, it is important to are defined by surface regions or volume regions. Each refinement level
evaluate the pressure loss P and friction factor f, involves to split the cell (hexahedron) of the previous level in eight
PDh hexahedra, splitting in halves the edges of the cell. The three meshes
f= have the same refinement levels, namely six for the bottom wall and
( Lu¯ 2)/2 (14)
seven for the VG. The difference among the meshes is that the size the
Using the Nu number and the friction factor, the TEF can be com- elements for the level 0 mesh of M2 is twice the size of M3. Also the size
puted as of the elements for the level 0 mesh of M1 is twice the size of M2. This
leads to the following cell sizes at the wall and near region:
Nu f
1/3
hwall|M3 0.5hwall|M2 0.33hwall|M1. A more detailed description of the
TEF = ,
Nu 0 f 0 (15) mesh generation process can be found in Ref. [41].
In Table 1, the number of cells for each mesh, the size of the mesh at
where Nu 0 and f0 represent the average Nu number and the friction the wall (hwall ), the maximum and average y+ and T+, the total heat flux
factor for the case without the VG, which is taken as reference. (qw ) and the total heat flux error with respect to that obtained using

67
L. Garelli et al. International Journal of Thermal Sciences 137 (2019) 64–74

Table 1 Richardson extrapolation [44] are reported. Using a Richardson extra-

Mesh convergence results. polation as proposed in Ref. [44], the extrapolated total heat flux (q ext )
– Cells hwall y+ max T+ max qw ε (%) can be computed with the results from meshes M2 and M3 as,
y+ T+
(q w3 q w2) 2.58 2.65
M1 89451 8.53E-04 5.41 3.16 3.86 2.25 2.85 12.9 q ext q w3 + = 2.58 + = 2.524 [W]
( )
4.26 E 04 2
h wall2
M2 645304 4.26E-04 4.27 1.45 3.05 1.03 2.65 4.99 1 1
M3 2283254 2.84E-04 3.09 0.95 2.21 0.68 2.58 2.21
h wall3 2.84 E 04 (20)

Additionally, for the data presented in Table 1, the local Nux and the
global Nu are plotted in Figs. 2 and 3. These figures show the con-
vergence of the Nusselt number with mesh refinement. The results
calculated with mesh M1 show large discrepancies near the inlet and
outlet regions. The results obtained using mesh M2 fit much better with
the reference solution. Finally, when computed with mesh M3, the re-
sults have good correlation when compared with the reference solution
obtained from Eqs. (16) and (17).
The criterion (local sizes) used to create mesh M3 is then used to
generate the discretization of the fluid domain when the VG is included
in the air channel (see Fig. 4).

4.2. Delta-wing vortex generator performance analysis

The performance of a delta-wing VG is analyzed in this section for

Fig. 2. Local Nusselt number distribution.
Fig. 3. Global Nusselt number distribution.
different AoA = {30; 40; 50} [deg], inlet air velocities uin = {0,3; 0.45;
0.6; 0.75; 0.9; 1.05} [m/s] and separation distances from the heated
wall d = {0; 3;5} [mm]. The objective of this section is to determine the
best configuration in terms of the TEF to achieve a balance between
thermal enhancement and pressure loss. Additionally, the total heat
flux increase is analyzed. In Fig. 4, a mid-plane slice of the mesh is
shown. This figure corresponds to AoA = 30 [deg] and d = 3 [mm],
where the region near the delta wing and heated wall is refined ac-
cording to the results obtained in the mesh convergence analysis.
Longitudinal vortices will develop in this refined region, where good
spatial resolution is required.
As mentioned earlier, the reference values are obtained from a si-
mulation of the flow between parallel plates shown in Sec.(4.1). The
results of this parametric analysis are shown in Fig. 5 and Fig. 6. Some
partial conclusions can be drawn from these figures. In all cases without
a clearance between the heated wall and the delta wing, the perfor-
mance is worse than if a separation d>0 is used; therefore, the clearance
has a positive contribution in terms of the TEF and total heat flux. In
Fig. 6, the results calculated without a clearance are significantly dif-
ferent from the others (i.e., when a clearance is used), and the TEF and
relative heat flux continuously increase as the inlet velocity increases.
In terms of the total heat flux, the best performance occurs for the
AoA = 40 [deg] and d = 5 [mm] case. However, when these results are
weighted with pressure loss, the best performance is attained for
AoA = 30 [deg] and d = 3 [mm], as shown in Fig. 5. Using this

Fig. 4. Mid-plane mesh.

68
L. Garelli et al.
Fig. 5. TEF vs. air inlet velocity.
Fig. 6. Relative heat flux enhancement (qwall/q 0) vs. air inlet velocity.
configuration the total heat flux increases from 6% for the lowest ve-
locity to 22% for the highest one with respect to the reference values
(q 0 ) obtained from the simulation of the flow between parallel plates
Fig. 7. Nu computed without (left) and with a VG (center). Relative (Nu/Nu 0 ) (right).
69
International Journal of Thermal Sciences 137 (2019) 64–74
without the VG.
Several previous works [7,17,20] have explained how VGs operate,
and in this manuscript, a brief description was given in section §2. In
Fig. 7, Nu is shown for a flat plate without (left) and with (center) a VG.
A centerline ruler with a graduation of one chord (c= 12.7 mm ) is
placed in the central figure. In the zone between 1c and 3c behind the
VG, Nu increases significantly due to the intense inflow toward the
panel surface. Downstream of this region, the vortices reach a stable
position, their intensity decays, and the thermal boundary layer thick-
ness increases. In the outflow region, Nu is reduced in comparison with
the reference flat plate values (Nu 0 ), reaching half the magnitude in
some places, which is apparent in Fig. 7 (right). Outside of the inflow/
outflow region, Nu/Nu 0 1. Finally, in the outlet region, the local
Nusselt is Nux 21, which is similar to the computed value in the mesh
convergence analysis shown in Fig. 2.
The downstream evolution of the vortices can be observed in Fig. 8,
where the temperature and velocity fields are shown. Several slices
along the channel length are shown to visualize the vortices. To easily
find the vortex center, the secondary flow is visualized using Line In-
tegral Convolution (LIC) [42] implemented in ParaView [43]. The
thermal and fluid boundary layers are reduced in the central (inflow)
region, and the thickness increases downstream. Additionally, the
center of the vortex is placed near the edge of the thermal boundary
layer in this configuration, which is a desired location, as mentioned by
Gentry in Ref. [8], to obtain higher efficiency in the air mixing process.
The separation d of the delta wing from the heated wall has two
positive effects. The first effect is an increase of the heat transfer in the
region very close to the delta wing, thus avoiding strong recirculation
behind it. The second effect is a decrease of the shear stress near the
vortex generator, which is evident in Fig. 9 when both solutions are
compared. Additionally, in the inflow region, the shear stress is 2 or 3
times larger than the reference value.
Having determined the optimal configuration of the delta wing by
means of a parametric analysis, the next step is to use a delta wing array
in a panel type radiator to compute the thermal enhancement for a
buoyancy-driven flow.
L. Garelli et al.
Fig. 8. Temperature and velocity fields.
Fig. 9. Relative shear stress factor w / w0 . Upper d = 3 [mm]; lower d = 0 [mm].
Fig. 10. Computational domain with delta wings.
4.3. Panel type radiator with delta-wing arrays
Next, a set of VG arrays are attached to a radiator panel to de-
termine if they enhance the thermal exchange. The flow is driven by
buoyancy, unlike the previous case where the air velocity was imposed
at the inlet boundary. The panel geometry is equivalent to that shown
by Min [24], with some modified dimensions in order to be more
consistent with the size of the radiators used in power transformers. The
total length of the panel is L p = 1524 [mm], the width is Wp = 450 [mm]
and the spacing between panels is Hp = 45 [mm]. The panel has six
trapezoidal oil channels [24], each one of width Wc = 75 [mm] and
height Hc = 4.5 [mm]. Three delta wings are placed widthwise on the
outer surface of each channel, with 18 VGs used in each array. The
arrays are separated by La = 10 c= 127 [mm], lengthwise with respect
to the panel. This distance was selected based on the results of the
70
International Journal of Thermal Sciences 137 (2019) 64–74
previous section, where it was observed that the vortex influence on the
surface of the panel is 8c . In Fig. 10, the computational domain is
shown (drawing not to scale). In this figure, the characteristic dimen-
sions and boundary conditions are denoted. To reduce the computa-
tional cost of the simulation, only a quarter of the total domain is si-
mulated using two symmetrical planes (Sym Z and Sym X), at the top of
the domain a pressure outlet boundary conditions is set. At the bottom
and right side a free inlet/outlet condition ([35]) is set in order to allow
the air flow to enter and exit from the computational domain. Ten
delta-wing arrays are placed along the panel, with AoA = 30 [deg] and
d = 3 [mm].
The physical properties of air are listed in Sec.(4.1), with the same
air inlet temperature Tin = 303 [K]. On the other hand, a temperature
distribution is imposed on the surface of the heated wall. This dis-
tribution is obtained from a conjugate heat transfer analysis for a ra-
diator panel solved by Ríos in Ref. [23] and experimentally validated.
Note that the use of this variable temperature distribution allows for a
more realistic prediction of the flow rate. If a constant temperature
distribution was used, then the buoyancy forces would be larger, and
the flow rate would be overestimated. The oil temperature difference
between the inlet and outlet depends on the heat exchange between the
oil and the ambient. This temperature difference generates a driving
force that causes the oil to circulate inside the radiator cooling chan-
nels. To simplify the setting of the temperature distribution, an analy-
tical function Tw = T(x, y) is proposed. This function is linear along the
x-axis (longitudinal) and quadratic along the transverse direction (y-
axis).
Tx = 10.74 x+ 338.6, 0.127 x 1.397 [m],
Tw = 147.5 y 2 + Tx , 0.225 y 0 [m], (21)
where Twmax = 340 [K] (air outlet region) and Twmin = 323.6 [K] (air inlet
region).
The fluid domain is discretized with the same criteria used to gen-
erate mesh “M3” introduced in Table 1 to ensure similar y+ and T+
values. For the reference results (i.e., the domain without delta wings),
a mesh of 7,086,000 cells is generated. On the other hand, for the do-
main with VGs, a mesh of 12,313,280 cells is generated. The increased
mesh size is due to the refined regions around the delta wings, as shown
in Fig. 11.
The simulations are carried out using 20 nodes of the computer
cluster “Seshat” belonging to the “Centro de Investigación de Métodos
Computacionales” (CIMEC), equipped with an Intel® Xeon® CPU E5-
2640 v2 @2.00 GHz server and 69 Intel® Xeon® CPU E5-1620 v2 @
3.7 GHz computing nodes with 16 GB RAM Micron® DDR3 1600 MHz,
interconnected with an Infiniband® network.
Before analyzing the effects of the vortex generators on the panel, a
simple verification is carried out. A constant temperature Tw = 340 [K]
L. Garelli et al.
Fig. 11. Domain discretization (near delta-wing region detail).
Fig. 12. Nux along the panel centerline.
Fig. 13. Nu along the panel centerline.
is imposed on the heated wall, and the resulting Nux and Nu at the
panel centerline are compared with the analytical values obtained from
Eqs. (16) and (17). These equations assume a constant velocity or Re
number, but in the panel simulation, there is a change in air velocity
along the longitudinal axis due to the buoyancy effect and air entering
from the sides. Therefore, the resulting Nux and Nu are compared with
the results for upper (u= 0.65 [m/s]) and lower (u= 0.2 [m/s]) velocities.
As shown in Fig. 12 and Fig. 13, the results of the simulation in the
entrance region (x<0.1 [m]) are close to the lower velocity results.
There is a transition region within 0.1<x<0.6 [m], and for ×>0.6 [m],
71
International Journal of Thermal Sciences 137 (2019) 64–74
the simulation is close to the higher velocity solution. This verification
gives confidence in the numerical results computed when using VGs
because the Nu value and, therefore, the heat exchange coefficient are
correctly computed with the proposed numerical model.
The next step is to analyze the effect of the delta-wing arrays placed
on the panel. To establish if the heat exchange is enhanced, local and
global characteristics are reported. In Fig. 14, the imposed temperature
distribution (a) and the heat exchange coefficient (b) are shown.
Figs. 14 to 16 were generated using ParaView [43] as postprocessing
tool. For the panel without VGs, the air enters the region between the
radiator fins from the bottom of the panel and from the sides at the
given reference temperature; thus, the highest heat exchange coeffi-
cient is obtained near the inlet. At a point downstream, the air tem-
perature increases, and the heat transfer coefficient decreases gradu-
ally. A similar effect is observed when VGs are used, but now the delta-
wing arrays promote the air mixing, locally increasing the heat ex-
change. VGs placed near the centerline of the panel result in a larger
increase than VGs placed near the sides. This is due to a local higher
velocity produced by the buoyancy and the chimney effect. Ad-
ditionally, this effect causes the air entering through the sides resulting
in a side-slip angle for delta wings placed near the boundary. From
Fig. 14 (c), the positive effect of VGs on the heat exchange coefficient is
observed. Without VGs, the corresponding average value is
hg = 4.66 [W/m2K], and with VGs it is hg = 5.4 [W/m2K]. Therefore,
16% increase is reported.
For the simulation without VGs, y+max = 1.86, and the average is
y = 0.62 . When VGs are considered, since the near region is refined,
+
the respective values are y+max = 2.15 and y+ = 0.61.
The velocity magnitude and air temperature fields are depicted in
Figs. 15 (a) and (b), respectively. Both fields are shown on a slice of the
solution domain at z = 6 [mm] from the panel surface (i.e., a cutting
plane parallel to the panel that cuts the VGs at half their height),
without (left) and with (right) vortex generators. The maximum velo-
city magnitude and temperature values are attained at the outlet.
However, the addition of VGs reduces the maximum values for both
variables. For example, the maximum velocity magnitude computed
with VGs is umax 0.8 [m/s], and without VGs, it is umax 0.9 [m/s].
In Table 2, a summary of global results calculated from both si-
mulations is given. The average outlet air temperature (Tout ), outlet air
velocity (u out ), heat transfer coefficient (hg ) and total heat flux (qw ) are
reported. The heat flux increases by 12% when VGs are used.
Finally, several slices normal to the x-axis at coordinates x = {-0.12;
0.0; 0.3; 0.6; 0.9; 1.2} are made to visualize the vortex generation. In
Fig. 16, the streamlines are colored with the temperature field, and the
most intense vortices are located near the centerline of the panel where
the velocity is higher. Additionally, the thermal boundary layer is
thicker in this region than in the region near the sides, where ambient
air enters the air channel.
5. Conclusion
In this article, three-dimensional numerical simulations are carried
out with the objective of analyzing the performance of delta-wing
vortex generators in enhancing the heat exchange coefficient in panel
type radiators. The study is focused on natural convection and buoy-
ancy-driven flows, which are the working conditions for this type of
heat exchangers.
A mesh convergence analysis was performed to establish the cell
sizes required near the surface of the radiator panels and the vortex
generator to accurately solve the fluid and thermal boundary layer. It is
important to properly capture these boundary layers because heat ex-
change and pressure losses are mainly affected by the temperature
gradient and shear stress in this region. After defining a reference mesh
(cell size distribution), a parametric study was carried out for three
variables, namely, AoA , uin and d, to determine a configuration that
L. Garelli et al. International Journal of Thermal Sciences 137 (2019) 64–74

Fig. 14. a) Temperature distribution, b) heat exchange coeff. of panel, and c) heat exchange coeff. of panel with VGs.

Fig. 15. a) Velocity magnitude, and b) temperature.

Fig. 16. Streamlines colored with the temperature field.

results in the best TEF. The best performance was obtained with
AoA = 30 [deg] and d= 3 [mm], with an increase in heat transfer ran-
ging from 6% to 22%, depending on the inlet velocity, and a maximum
TEF 10%. Other authors report higher values in heat transfer coeffi-
cient enhancement (e.g., Oneissi et al. [17], based on the research of
Tiggelbeck et al. [4], reported a maximum Nu/Nu 0 1.56 for
Re = 4600 ), but is important to mention that these values depend on the
ratio between the area influenced by the vortex generator and a re-
ference area. In these references, the ratio between the vortex generator
span b and the channel width W is b/ W= 2/5, where 40% of the

72
L. Garelli et al.
Table 2
Global result comparison between simulations.
– Tin [K] Tout [K] u out [ms 1] hg [Wm 2K] qw [W]
Without VGs 303 314 0.48 4.66 32.6
With VGs 303 316 0.42 5.41 36.4
heated surface is influenced by the vortex generator. In this work, the
span b= 6.35 [mm] and W= 40 [mm]; hence, the ratio of b/ W= 0.158,
and 16% is influenced by the vortex generator. If the channel width is
reduced to W= 15.8 [mm], then 40% of the heated surface is influenced,
with Nu/Nu 0 1.6 for Re 5300 , which is in agreement with the re-
sults presented by Oneissi et al. [17] and Tiggelbeck et al. [4].
When delta-wing vortex generators are placed in an array config-
uration on a plate-type radiator and the fluid is driven by buoyancy
forces, the pressure loss increases, and the average outlet velocity de-
creases. However, heat transfer is improved, and the average heat
transfer coefficient increases 15%, as shown in Table 2. As observed in
Fig. 14, VGs placed in the center of the panel produce higher local
exchange coefficients (hl ) due to higher local air velocity. In the case of
delta wings placed near the lateral side of the air channel, the lower air
velocity and side-slip velocity produced by air entering through the
sides reduce the efficiency of the device. Anyway, under this condition
a 12% increase in heat transfer is obtained, which is a significant im-
provement.
From the technological implementation point of view, in order to
carry out future experimental validations, the delta wings can be made
by means of 3D printing or sheet metal forming. Also, it has to be
analyzed how to improve the performance of the vortex generators
placed near the lateral of the panel. This could be achieved by better
aligning the vortex generators with the air flow or by channeling the air
flow between the radiator panels. In this manner, the heat exchange
coefficient in this region could be increased.
Acknowledgments
This work has received financial support from Consejo Nacional de
Investigaciones Científicas y Técnicas (CONICET, Argentina, grant PIP
11220150100588CO), Universidad Nacional del Litoral (UNL,
Argentina, grant CAI + D-2011-01-00012-LI, CAI + D-
50420150100112LI), Agencia Nacional de Promoción Científica y
Tecnológica (ANPCyT, Argentina, grants PICT-2016-0708, PICT-2014-
2660 and PICT-2015-2904), Agencia Santafesina de Ciencia, Tecnología
e Innovación (ASACTEI, Argentina, grant 00010-18-2014), Ministerio
de Defensa (AR, PIDDEF-4/2014), and CYTED-516RT0512.
This work was performed with Free Software Foundation/GNU-
Project resources such as GNU–Linux OS, GNU–GFortran, GNU–Octave,
GNU–Git, GNU-GCC, and GNU–GIMP, as well as other open source
resources such as Salome, ParaView, Code_Saturne, and LATEX .
Additionally, the authors made use of the computer cluster “Seshat”,
which is part of the computer center of CIMEC. This computer center is
integrated into the Sistema Nacional de Computación de Alto
Desempeño (SNCAD), which is a joint project of Ministerio de Ciencia,
Tecnología e Innovación Productiva (MINCyT) and Consejo
Interinstitucional de Ciencia y Tecnología (CICyT).
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.ijthermalsci.2018.10.037.
References
[1] J.C. Lin, Review of research on low-profile vortex generators to control boundary-
layer separation, Prog. Aero. Sci. 38 (45) (2002) 389–420, https://doi.org/10.
International Journal of Thermal Sciences 137 (2019) 64–74
1016/S0376-0421(02)00010-6.
[2] M. Brüderlin, M. Zimmer, N. Hosters, M. Behr, Numerical simulation of vortex
generators on a winglet control surface, Aero. Sci. Technol. (71) (2017) 651–660,
qw /qref https://doi.org/10.1016/j.ast.2017.10.018.
[3] T.R. Johnson, P.N. Joubert, The influence of vortex generators on the drag and heat
1 transfer from a circular cylinder normal to an airstream, ASME. J. Heat Transfer. 91
(1) (1969) 91–99, https://doi.org/10.1115/1.3580126.
1.12
[4] S. Tiggelbeck, N.K. Mitra, M. Fiebig, Comparison of wing-type vortex generators for
heat transfer enhancement in channel flows, ASME. J. Heat Transfer 116 (4) (1994)
880–885, https://doi.org/10.1115/1.2911462.
[5] M. Fiebig, P. Kallweit, N.K. Mitra, S. Tiggelbeck, Heat transfer enhancement and
drag by longitudinal vortex generators in channel flow, Exp. Therm. Fluid Sci. 4
(1991) 103–114.
[6] M. Fiebig, H. Guntermann, N.K. Mitra, Numerical analysis of heat transfer and flow
loss in a parallel plate heat exchanger element with longitudinal vortex generators
as fins, ASME J. Heat Transfer 117 (1995) 1064–1067.
[7] A.M. Jacobi, R.K. Shah, Heat transfer surface enhancement through the use of
longitudinal vortices, a review of recent progress, Exp. Therm. Fluid Sci. 11 (1995)
295–309, https://doi.org/10.1016/0894-1777(95)00066-U.
[8] M.C. Gentry, A.M. Jacobi, Heat transfer enhancement by delta-wing-generated tip
vortices in flat-plate and developing channel flows, ASME. J. Heat Transfer 124 (6)
(2002) 1158–1168, https://doi.org/10.1115/1.1513578.
[9] A.A. Joardar, A.M. Jacobi, A numerical study of flow and heat transfer enhance-
ment using an array of delta-winglet vortex generators in a fin-and-tube heat ex-
changer, ASME. J. Heat Transfer 129 (9) (2006) 1156–1167, https://doi.org/10.
1115/1.2740308.
[10] A. Sohankar, Heat transfer augmentation in a rectangular channel with a vee-
shaped vortex generator, Int. J. Heat Fluid Flow 28 (2) (2007) 306–317, https://
doi.org/10.1016/j.ijheatfluidflow.2006.03.002.
[11] J.M. Wu, W.Q. Tao, Numerical study on laminar convection heat transfer in a
rectangular channel with longitudinal vortex generator. Part A: verification of field
synergy principle, Int. J. Heat Mass Transfer 51 (5–6) (2008) 1179–1191, https://
doi.org/10.1016/j.ijheatmasstransfer.2007.03.032.
[12] Li-Ting Tian, Ya-Ling He, Yong-Gang Lei, Wen-Quan Tao, Numerical study of fluid
flow and heat transfer in a flat-plate channel with longitudinal vortex generators by
applying field synergy principle analysis, Int. Commun. Heat Mass Tran. 36 (2)
(2009) 111–120, https://doi.org/10.1016/j.icheatmasstransfer.2008.10.018.
[13] Li Li, Xiaoze Du, Yuwen Zhang, Lijun Yang, Yongping Yang, Numerical simulation
on flow and heat transfer of fin-and-tube heat exchanger with longitudinal vortex
generators, Int. J. Therm. Sci. 92 (2015) 85–96, https://doi.org/10.1016/j.
ijthermalsci.2015.01.030.
[14] Gaofeng Lu, Guobing Zhou, Numerical simulation on performances of plane and
curved winglet-Pair vortex generators in a rectangular channel and field synergy
analysis, Int. J. Therm. Sci. 109 (2016) 323–333, https://doi.org/10.1016/j.
ijthermalsci.2016.06.024.
[15] Y. Xu, M.D. Islam, N. Kharoua, Numerical study of winglets vortex generator effects
on thermal performance in a circular pipe, Int. J. Therm. Sci. 112 (2017) 304–317,
https://doi.org/10.1016/j.ijthermalsci.2016.10.015.
[16] Guobing Zhou, Zhizheng Feng, Experimental investigations of heat transfer en-
hancement by plane and curved winglet type vortex generators with punched holes,
Int. J. Therm. Sci. 78 (2014) 26–35, https://doi.org/10.1016/j.ijthermalsci.2013.
11.010.
[17] M. Oneissi, C. Habchi, S. Russeil, D. Bougeard, T. Lemenand, Novel design of delta
winglet pair vortex generator for heat transfer enhancement, Int. J. Therm. Sci. 109
(2016) 1–9, https://doi.org/10.1016/j.ijthermalsci.2016.05.025.
[18] F.R. Menter, Two-equation eddy-viscosity turbulence models for engineering ap-
plications, AIAA J. 8 (32) (1994) 1598–1605, https://doi.org/10.2514/3.12149.
[19] A. Esmaeilzadeh, N. Amanifard, H.M. Deylami, Comparison of simple and curved
trapezoidal longitudinal vortex generators for optimum flow characteristics and
heat transfer augmentation in a heat exchanger, Appl. Therm. Eng. (125) (2017)
1414–1425, https://doi.org/10.1016/j.applthermaleng.2017.07.115.
[20] Qiang Zhang, Liang-Bi Wang, Yong-Heng Zhang, The mechanism of heat transfer
enhancement using longitudinal vortex generators in a laminar channel flow with
uniform wall temperature, Int. J. Therm. Sci. 117 (2017) 26–43, https://doi.org/10.
1016/j.ijthermalsci.2017.03.003.
[21] S.B. Paramane, K. Joshi, W. Van der Veken, A. Sharma, CFD study on thermal
performance of radiators in a power transformer: effect of blowing direction and
offset of fans. Power delivery, IEEE Transactions on 6 (29) (2014) 2596–2604.
[22] S.B. Paramane, W. Van der Veken, A. Sharma, A coupled internal-external flow and
conjugate heat transfer simulations and experiments on radiators of a transformer,
Appl. Therm. Eng. 103 (2016) 961–970.
[23] G. Ríos Rodriguez, L. Garelli, M. Storti, D. Granata, M. Amadei, M. Rossetti,
Numerical and experimental thermo-fluid dynamic analysis of a power transformer
working in ONAN mode, Appl. Therm. Eng. (112) (2016) 1271–1280, https://doi.
org/10.1016/j.applthermaleng.2016.08.171.
[24] C. Min, J. Jin, X. Wang, C. Qi, Numerical investigation of natural convection heat
transfer for panel type radiator mounted with longitudinal vortex generators, Int.
Conf. Mater. Renew. Energy & Environ. Shanghai (2011) 1267–1270, https://doi.
org/10.1109/ICMREE.2011.5930567.
[25] F.M. White, Viscous Fluid Flow, McGraw-Hill Education, 2005.
[26] R.W. Hanks, H.C. Ruo, Laminar-turbulent transition in ducts of rectangular cross
section, Ind. Eng. Chem. Fundam. 4 (5) (1966) 558–561, https://doi.org/10.1021/
i160020a022.
[27] H. Versteeg, W. Malalasekra, An Introduction to Computational Fluid Dynamics: the
Finite Volume Method, Prentice Hall, 2007.
[28] F. Moukalled, L. Mangani, M. Darwish, The Finite Volume Method in
73
L. Garelli et al.
Computational Fluid Dynamics, Springer International Publishing, 2015.
[29] Code_Saturne an open source CFD software. http://code-saturne.org.
[30] F. Archambeau, N. Mchitoua, M. Sakiz, Code Saturne: a finite volume code for the
computation of turbulent incompressible flows-industrial applications, Int. J. Finite
1 (1) (2004) 1–62.
[31] D.S. Jang, R. Jetli, S. Acharya, Comparison of the PISO, SIMPLER, and SIMPLEC
algorithms for the treatment of the pressure-velocity coupling in steady flow pro-
blems, Numer. Heat Tran. 3 (10) (1986) 209–228.
[32] T.J. Barth, D. Jespersen, The design and application of upwind schemes on un-
structured meshes, Technical Report AIAA-89-0366, AIAA 27th Aerospace Sciences
Meeting. Reno, Nevada, 1989.
[33] P. Tamamidis, D.N. Assanis, Evaluation of various high-order-accuracy schemes
with and without flux limiters, Int. J. Numer. Methods Fluid. (16) (1993) 931–948,
https://doi.org/10.1002/fld.1650161006.
[34] S.E. Norris, A Parallel Navier-stokes Solver for Natural Convection and Free Surface
Flow, Ph.D. Thesis, Department of Mechanical Engineering, Sydney, Australia,
2000.
[35] Code_Saturne: Theory guide http://code-saturne.org/cms/sites/default/files/docs/
5.0/theory.pdf.
[36] Y.S. Muzychka, M.M. Yovanovich, Laminar forced convection heat transfer in the
74
International Journal of Thermal Sciences 137 (2019) 64–74
combined entry region of non-circular ducts, ASME. J. Heat Transfer. 126 (1)
(2004) 54–61, https://doi.org/10.1115/1.1643752.
[37] A. Bejan, A. Kraus, Heat Transfer Handbook vol. 1, New Jersey Willey, 2003, pp.
395–438.
[38] S. Stolpa, Turbulent Heat Transfer, Department of Aerospace and Mechanical
Engineering, University of Notre Dame, Notre Dame, USA, 2004.
[39] V.S. Arpaci, P.S. Larsen, Convection Heat Transfer, Prentice-Hall, 1984.
[40] T.L. Bergman, A.S. Lavine, F.P. Incropera, Fundamentals of Heat Mass Transfer,
seventh ed., Wiley, Hoboken (NJ), 2011.
[41] OpenFOAM Foundation, SnappyHexMesh, (2014) http://www.openfoam.org/
docs/user/snappyHexMesh.php.
[42] B. Cabral, L. Leedom, Imaging vector fields using line integral convolution,
Proceedings of the 20th Annular Conference on Computer Graphics and Interactive
Techniques, Anaheim, California, 1993, pp. 263–270.
[43] J. Ahrens, B. Geveci, C. Law, ParaView: an End-user Tool for Large Data
Visualization, Visualization Handbook, Elsevier, 2005 ISBN-13: 978-0123875822.
[44] P.J. Roache, Quantification of uncertainty in computational fluid dynamics, Annu.
Rev. Fluid Mech. 29 (1) (1997) 123–160, https://doi.org/10.1146/annurev.fluid.
29.1.123.