Applied Thermal Engineering 51 (2013) 852e863

Contents lists available at SciVerse ScienceDirect

Applied Thermal Engineering

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

Numerical modelling of the natural ventilation of underground

transformer substations
Juan Carlos Ramos a, *, Maximiliano Beiza a, Jon Gastelurrutia a, Alejandro Rivas a,
Raúl Antón a, Gorka S. Larraona a, Iván de Miguel b
a
TECNUN (University of Navarra), Department of Mechanical Engineering, Thermal and Fluids Engineering Division,
Paseo de Manuel de Lardizábal 13, 20018 San Sebastián, Spain
b
Ormazabal Distribución Secundaria, Barrio Basauntz 2, 48140 Igorre, Spain

h i g h l i g h t s

< A model of natural ventilation of underground transformer substations is developed.

< Simulations serve to analyse air flow patterns and temperature distributions.
< There is a component of reverse flow in the outflow grilles.
< There is a mass of static warm air at the top part of the switchboards zone.
< Correlations for air mass flow rate and heat transfer coefficients are proposed.

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

Article history: Ventilation by natural convection of two underground transformer substations has been numerically
Received 28 May 2012 modelled. The model has been verified in terms of discretization errors and it has been validated with the
Accepted 7 October 2012 experimental results of eight temperature rise tests carried out under different conditions of ventilation
Available online 2 November 2012
and transformer power losses. The results of the simulations serve to analyse the air flow pattern and the
air temperature distributions inside the substation. A correlation for the air mass flow rate as a function
Keywords:
of the ventilation conditions (discharge coefficient and area of the grilles) and the heat dissipated by the
Thermal modelling
transformer has been fitted. The heat transfer coefficients on the surfaces of the transformer and the
Transformer substation
Natural ventilation
walls of the enclosure can also be obtained from the simulations of the model. All this information will be
CFD used in a future paper to develop a zonal thermal model of the ventilation of the substations that can be
employed as a design and optimisation tool.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction
Underground transformer substations are used for electrical
power distribution in public networks and private installation load-
centres. These buildings are usually made of prefabricated concrete
and have a personnel access point and some ventilation grilles.
Inside of the enclosure there are one or two distribution trans-
formers with their Low Voltage (LV) boards, Medium Voltage (MV)
cubicles, and interconnecting and auxiliary devices. In the trans-
former and the LV boards the generation of heat is due to power
losses occurring in the conversion of the distributed electrical
energy from medium voltage to low voltage for domestic and
industrial applications. This heat must be removed by the natural
* Corresponding author. Tel.: þ34 943 21 98 77; fax: þ34 943 31 14 42.
E-mail address: jcramos@tecnun.es (J.C. Ramos).
convection flow of air that enters and leaves the substation through
the ventilation grilles and by the radiation exchanges with the walls
of the substation.
International Standards [1,2] state that the criterion for good
performance in a transformer substation is given by the maximum
temperature reached by the top-oil of the transformer. This
temperature must be limited in order to extend the operating life of
the transformer. As experimental tests must be run on a real
substation, obtaining a temperature that is over the limit would
invalidate the built substation, requiring a new design and new
casts. In order to avoid this slow and expensive design procedure, it
would be very useful to have a mathematical model of ventilation
in the substation and to perform a simulation of this model to
determine the temperatures in the design stage prior to the
experimental tests.
One of the first models of ventilation in transformer substations
that can be found in the literature is the one by Menheere [3]. This

1359-4311/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.applthermaleng.2012.10.032
 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863
Nomenclature
Latin letters
Agrilles surface area of the ventilation grilles [m2]
cp specific heat [J/kg K]
Cd,grilles discharge coefficient of the ventilation grilles
[m/s Pa0.5]
g gravity acceleration [m/s2]
GCI Grid Convergence Index [%]
k turbulent kinetic energy [m2/s2]
m_ air air mass flow rate [kg/s]
Nuz Nusselt number [e]
p pressure [Pa]
P transformer power losses [W]
qconv,transf transformer convection heat losses [W]
is a very simplified model that uses one equation for the heat
transferred to the ventilation air and another equation to the heat
dissipated through the walls of the substation. The inputs of the
model are the power dissipated by the transformer, the dimensions
of the transformer and the substation, the thermal conductivity of
the walls, and the heights, surface areas and resistance coefficients
of the inlet and outlet ventilation grilles. The outputs of the model
are estimations of the outlet air and of the transformer’s mean
temperature rises over the ambient temperature. Although the
model takes into account the main parameters involved in the
ventilation performance of the substations (the transformer power
and the surface area and the resistance coefficient of the ventilation
grilles) the results can be used only as a “rule of thumb” in the
design stage of a transformer substation because the model does
not give the enclosure class of the substation required by IEC
Standard [2].
Another possible approach to the thermal modelling of trans-
former substations is the transient equivalent thermal circuit
model developed by Radakovic and Maksimovic in [4]. This model
is based on a small number of characteristic temperatures inside
the transformer substation, and it relies on some parameters whose
values have to be determined through experimentation for each
new design. The same authors in [5] present an improvement of the
model by including solar irradiation and wind velocity.
Iskender and Mamizadeh in [6] use the same methodology as
the dynamic thermal circuit model. They improve the previous
models by taking into consideration the variation over time of the
thermal resistances and capacitances of the top-oil, of the venti-
lation air and of the different components of the substation
enclosure.
A common characteristic of these dynamic network models is
that they can be used to simulate the load capability and ageing of
a specific and experimentally checked system using discrete
temperature measurements, but they are not able to analyse and
optimise the performance of a transformer substation in the design
stage in order to determine the enclosure class.
Thus, in order to deal with design and optimisation objectives,
the use of other types of mathematical models with a more
exhaustive treatment of the physical phenomena that is taking
place in the transformer substation is required.
Hence, there is the type of model based on the description, by
means of differential equations, of the mass and heat transfer
phenomena taking place in a flow domain under the restriction of
some conditions imposed at its boundaries. To solve these differ-
ential equations there are different numerical techniques that can
be employed.
853
Raz Rayleigh number [e]
Sh source term in energy equation [W/m3]
T temperature [K or  C]
ui velocity components [m/s]
xi Cartesian coordinates [m]
yþ dimensionless distance to the nearest wall [e]
Greek letters
dij Kronecker delta [e]
DT temperature rise over ambient temperature [K]
ε dissipation rate of k [m2/s3]
l thermal molecular conductivity [W/m K]
lT turbulent thermal conductivity [W/m K]
m dynamic molecular viscosity [kg/m s]
mT turbulent eddy viscosity [kg/m s]
r density [kg/m3]
As far as the authors are aware, there are only two attempts in
the specialized literature. Loucaides et al. [7] use the Finite Element
Method (FEM) to solve the energy and the NaviereStokes equations
in a flow domain corresponding to the air inside a transformer
substation. They assume that the flow is under a laminar regime
and the buoyancy forces are modelled by means of the Boussinesq
hypothesis. As main boundary conditions, they impose the ambient
temperature on the external surface of the walls, the ventilation
grilles are geometrically modelled with their effective area, and the
transformer losses are imposed as a uniform heat flux on the
surface of the transformer. The model is used to analyse the influ-
ence of the aperture of the ventilation grills, the transformer load
and the ambient temperature in the air temperature distribution
inside the substation.
The other work addressing the issue of modelling the ventila-
tion in substations by means of differential equations is the one by
Ramos et al. in [8], where they develop a differential model of the
ventilation (air circulation and heat transfer) of a half-buried
transformer substation solved by means of the Finite Volume
Method (FVM). Taking a more realistic approach than the previous
authors, they assumed the flow is turbulent and used the Standard
keε model to model it. The walls of the enclosure are modelled as
solids with one-dimensional heat conduction and external
convection. The ventilation grilles are modelled as boundaries
where the air flow suffers a pressure loss, and the corresponding
loss coefficients are calculated numerically. A vertical temperature
distribution is imposed on the transformer casing and the oil-filled
hollow fins are modelled as solids by means of an equivalent
thermal conductivity. The model allows the enclosure class of the
substation to be determined and for the analysis of the air velocity
and temperature distributions.
On the one hand, the principal handicap of these differential
models is the high computational cost of performing a single
simulation and the long time period that is required to obtain
results. On the other hand, they provide results that cannot be
obtained either experimentally or with other types of models,
namely, the air flow pattern and temperature distributions, the air
mass flow rate and the heat transfer coefficients on the transformer
surfaces.
Nevertheless, all this information obtained in the simulation of
the differential model can be used to develop an intermediate level
model, an approach known as zonal modelling [9,10], which
requires fewer computational resources and less simulation time,
so as to allow its implementation in a software tool oriented toward
designing and optimizing the thermal performance of transformer
substations.
854 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863
This methodology has been successfully applied to the model-
ling of the cooling of distribution transformers by Gastelurrutia
et al. In [11] they developed a detailed numerical model of the
movement by natural convection of the oil inside distribution
transformers and, based on the oil flow and temperature patterns
obtained, they produced in [12] an algebraic zonal thermal model
that can be used for design and optimization purposes.
In this paper a mathematical model based on the numerical
resolution of the differential equations describing the air flow
movement and the heat transfer inside underground trans-
former substations is presented. The numerical model is
described, attending to its main characteristics: flow domain,
governing equations, treatment of turbulence, boundary condi-
tions, discretization and resolution, convergence and grid inde-
pendency. The model is validated with the experimental results
of eight temperature rise tests. The simulations of the model
provide a description of the main features of the natural
convection air movement and the heat transfer inside trans-
former substations. Correlations for the air mass flow rate and
for the heat transfer coefficients on the surfaces of the trans-
former and of the walls of the enclosure have been fitted. All this
information will be used to build a zonal thermal model of the
ventilation of transformer substations that will demand fewer
computational resources and that will be easier to implement in
design-oriented software, following the methodology described
in Refs. [11,12].
2. Description of underground transformer substations
Fig. 1 shows internal and external views of the modelled
transformer substations identified henceforth as TS01 (left) and
TS02 (right). TS01 has horizontal ventilation because the grilles are
at ground level, and TS02 has vertical ventilation because the vents
protrude from the ground. Each of them includes two distribution
transformers (1); their respective Low Voltage boards (2); the
Medium Voltage switchgears (3); the protection metallic plates (4),
which separate the transformer zone from the switchboards zone;
the concrete walls of the enclosure (5); the prefabricated concrete
vents (6), the internal inflow (7) and outflow (8) windows and
louver grilles in the substation with horizontal ventilation; and the
inflow (9) and outflow (10) horizontal (left image) or vertical (right
image) external ventilation grilles.
Fig. 1. Configuration of the two modelled transformer substations with their main components.
The main external dimensions of the transformer substations are
6.56 m, 2.46 m and 2.79 m in length, width and height, respectively.
The input voltage during the standard performance of the trans-
formers is 25 kV at 50 Hz, and it handles 1000 kVA. The nominal
power dissipated in the form of heat by the transformers is 12,200 W.
During the operation of the substation the LV also has power losses.
The ventilation of the substations occurs as follows. The internal
power losses of the core and the windings of the transformer are
transferred to the external surfaces, casing and hollow fins by
means of the natural convection flow of the internal oil. From the
external surfaces the heat is dissipated in two ways: by the venti-
lation air and by radiation to the walls of the enclosure, and from
the walls to the surrounding soil by conduction. The ventilation air
flows through the substation by natural convection. The air enters
the substation through the inflow grilles (9 in Fig. 1) located over
the LV and MV switchboard zone of the substation. From here the
air passes to the transformer zones through a free pass below the
separating metallic plates. Finally, the air leaves the substation
through the outflow ventilation grilles (10 in Fig. 1) located on top
of the transformers.
Substation TS01 has three types of ventilation grilles. Two of
them can be seen in the left-hand image of Fig. 1, where the most
external are the grating grilles situated at ground level at the top of
the concrete vents (9 and 10). Over the ventilation windows at the
bottom of the concrete vents, on the walls of the enclosure, there
are two types of grilles: the louvered ones (7 and 8) and the
expanded metal grilles (not shown in Fig. 1). The function of the
latter is to prevent small animals from entering the substation.
Substation TS02 (Fig. 1, right) has only two types of ventilation
grilles: the louvered vents protruding from the ceiling of the
substation (9 and 10), and the expanded metal grilles that in this
case are situated in the internal part of the louvers of the protruding
vents (9 and 10).
3. Experimental temperature rise tests
One of the objectives of the model that has been developed is to
provide the enclosure class of the substation obtained during
a temperature rise test [1,2]. Thus, in order to validate the model it
is necessary to have the results of the temperature rise or heating
test carried out in the two substations. But another objective of the
model is to analyse the air circulation and the air temperature
 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863 855

distribution in the substation. So other experimental tests that Table 2

differ from the standard heating test have also been performed Grid independency analysis of the results for the simulation of test TS01-E02.

with the aim of better understanding the flow and thermal Mass flow Transformer Air temperature
boundary conditions to be imposed on the model and in order to rate [kg/s] power losses rise [ C]
validate it again. [W]

Table 1 shows the list of the experimental tests performed in Variable values f1 0.3184 12,356 36
f2 0.3152 12,283 36.6
both substations, varying the ventilation condition and the power
f3 0.3093 12,597 37.6
dissipated by the transformers. Test E00 corresponds to the heating Absolute differences a21 0.0032 73 0.6
test of the transformer outside the substation [1]. This test is a32 0.0059 314 1.0
necessary to calculate the enclosure class of the substation [2]. Apparent order p 4.843 20.879 11.072
In tests E01 and E05, all the ventilation grilles of the substations Extrapolated values fExt,21 0.3337 12,414 34.9
fExt,32 0.3279 12,208 35.9
were removed. In test TS01-E04, four of the six inflow windows Approximate relative EAprox,21 [%] 1.0 0.6 1.7
(7 in Fig. 1) and two of the four outflow windows of each trans- errors EAprox,32 [%] 1.9 2.6 2.7
former (8 in Fig. 1) were closed by means of expanded polystyrene Extrapolated relative EExt,21 [%] 4.6 0.5 3.2
plates. In test TS02-E08 the standard protruding louvered vents error
Grid convergence GCI21 [%] 6.0 1.5 3.8
consisting of eight louvers where replaced by shorter vents con-
index
taining only four louvers.
The measurement devices employed during the tests were: 34
T-type thermocouples with an accuracy of 1 K; 26 T-type ther- The most relevant results of the eight tests used to validate the
mocouples with an accuracy of 0.5 K; 15 K-type thermocouples numerical model are presented in Table 3.
with an accuracy of 1.5 K; 3 one-directional hot wire anemome-
ters with an accuracy of 0.015 m/s; 2 omni-directional hot wire 4. Mathematical model
anemometers with a reading accuracy of 3%; and an infrared
thermographic camera. The location of the thermocouples was the The natural convection of air inside the described transformer
following: 38 were on the external surface of one of the trans- substations can be studied by solving the governing differential
formers: lid, casing, fins and base (Fig. 2a); 5 registered the air equations restricted to the boundary conditions inside the flow
temperature over the lid and between the fins of the transformer; 4 domain.
recorded the external ambient temperature; 6 were on the external
and internal surfaces of the walls of the enclosure (Fig. 2b); 2 4.1. Flow domain
measured the internal oil temperature at the upper part of the
transformers; and 20 were on the external surfaces of the LV and The computational flow domain consists mainly of the air inside
MV switchboards. One omni-directional and one one-directional one half of the substation, delimited by the internal surfaces of the
anemometer both were placed inside the substation, over one of walls of the enclosure, the external surfaces of the inner compo-
the transformers (Fig. 2c), the other omni-directional and one one- nents (the transformer, the LV boards and the MV cubicles) and the
directional anemometer were placed at the outlet external grilles symmetry plane. Fig. 3 represents the flow domain of substations
(Fig. 2d), and the remaining one-directional anemometer at the TS01 and TS02, with the external limits and the main components
inlet ventilation external grilles. identified. The flow domain has been extended outside the venti-
The temperatures were recorded during the whole heating test lation grilles in order to include a small portion of the ambient air
with a sample frequency of 1, 3 or 5 min, depending on the case. where the corresponding boundary conditions of pressure and
The velocities were only recorded at the end of the heating test temperature are imposed.
with a sample frequency of 1 s. An infrared camera was used to The main simplification adopted in defining the flow domain
photograph the temperature distributions on the external surfaces has been to consider only one half of the substation, assuming
of the walls of the enclosure when the steady state of the heating a geometrical symmetry with respect to a vertical plane that cuts
test was reached. The total power losses of the transformers and the the substation width wise. To implement this simplification in the
LV boards were checked at intervals of 20 min. computational domain of both substations, the personnel access
The maximum uncertainties of the variables measured have metallic door and its ladder have been eliminated and the MV
been estimated based on a Student-t distribution with a confidence boards have been centred. Moreover, in substation TS02, the posi-
interval of 95% [13], obtaining the following values: 4% in the tion of the inlet protruding vents (9 in Fig. 1) has been assumed to
power measurements; 0.8 K in the ambient temperature; 1.6 K be symmetric as well. These simplifications can be checked by
in the measurements of the rest of the temperature rises over the comparing Figs. 1 and 3. The external boundary conditions on the
ambient temperature and 0.15 m/s in the velocity measurements. walls of the enclosure have also been assumed to be symmetric, as
the results of the experimental tests have confirmed.
A high quality hexahedral mesh was used in the discretization of
Table 1 both computational domains. The number of cells is of the order of
Main characteristics of the experimental tests.
31 106. The resolution of the mesh is higher in two zones of the
Test name Ventilation condition % of transformer domain where it is required in order to obtain more reliable results:
power losses in the zone surrounding the transformer (Fig. 4), especially
TS01-E00 Outside the substation 100 between the fins (z7  106) and in the zones where the ventilation
TS01-E01 Without grilles 100
grilles are situated (z5  106 cells in the inlet and z7.7  106 in the
TS01-E02 Standard 100
TS01-E03 Standard 50 outlet concrete vents of substation TS01 and z4  106 cells in the
TS01-E04 Reduced 50 protruding vents of substation TS02). All this modularity and
TS02-E05 Without grilles 100 meshing quality is achieved through the use of a non-conformal
TS02-E06 Standard 100 grid type, splitting the meshed domain into different coupled
TS02-E07 Standard 50
TS02-E08 Reduced 50
zones. This meshing strategy is numerically possible due to the use
of a code developed explicitly for unstructured meshes [14].
856 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863
To analyse the mesh quality, the values of the dimensionless
distance of the centroid of the cells to the nearest wall, yþ, have
been checked for the first cell of the walls of the transformer and
the substation. Among the eight simulations, it has been obtained
that the greatest average values are 1.85 and 2.48, for the trans-
former and the substation walls, respectively. Moreover, in all the
simulations, at least the 94% of the values of the yþ in the trans-
former walls are below 5 (recommended upper limit). For the
substation walls this percentage is 95% in all the simulations, except
TS02-E05 and TS02-E06 where it is 89% and 87%, respectively.
These values permit to assure that the resolution of the mesh close
to the walls is fine enough to characterize correctly the heat
transfer phenomena.
4.2. Governing equations
The equations that represent the air flow and the heat transfer
inside the substation are the steady Reynolds Average Naviere
Stokes (RANS) equations, (1) and (2), used to include the turbu-
lence effects in the mean flow variables, and the energy Eq. (3).
v
ðrui Þ ¼ 0
vxi
" ! # using the RNG keε turbulence model, the normal gradient of the
v
 vp v vui vuj 2 turbulent kinetic energy, k, is set to zero and its rate of dissipation,
rui uj ¼  þ ðm þ mT Þ þ  rkdij
vxj vxi vxj vxj vxi 3 ε, is calculated with an algebraic expression. In the simulations
 gi ðr  r0 Þ
" #
v
 v vT
rcp Tui ¼ ðl þ lT Þ
vxi vxi vxi
The buoyancy term has been included in the right side of Eq. (2)
to take into account the movement provoked by density variations
of the air. The Ideal Gas equation is used to model the variation of
the air density with the temperature. The rest of the air properties,
like molecular viscosity, thermal conductivity and specific heat, are
also temperature dependent.
The same two-equation turbulence models tried in [11] have
been assessed in the present case, because in the previous case all
of them provided acceptable results in the modelling of a similar
natural convection problem. Although those turbulence models
have not been specifically developed to study natural convection
due to the fact that they consider an isotropic relation between
temperature gradients and turbulent heat fluxes in conjunction
with a constant turbulent Prandtl number that relates the turbulent
eddy and thermal diffusivity, their adequateness for the modelling
of turbulence in natural convection problems [15] has been verified.
In summary, in the simulations corresponding to the tests
without ventilation grilles (TS01-E01 and TS02-E01 in Table 1), the
Low-Reynolds-number keε model by ChangeHsieheChen [16] re-
ported in the results was closer to the experimental ones. In the rest
of the simulations, which correspond to the tests with ventilation
grilles, a Two-Layer approach combining a RNG keε model [17] and
the one-equation model by Wolfstein [18], which deals with
turbulence modelling near the walls, has been adopted.
As our goal is not to analyse the influence of the modelling of
turbulence in natural convection, the choice of the turbulence
models has been based on obtaining the better correlation between
the numerical results of the main mean flow and thermal variables
provided by the mathematical model and the experimental results.
Taking into account this premise, it can be said that the Change
HsieheChen turbulence model seems to be more appropriate for
problems of natural convection in open domains (as in the case
where the substation’s ventilation grilles are removed) and the
RNG keε turbulence model for closed cavities (as in the case where
the substation’s ventilation grilles are installed). The latter
conclusion has been also pointed out in [11].
The radiation heat exchange has been calculated by means of
the Discrete Ordinates (DO) model [19]. Each octant of the angular
space has been discretized into 9 solid angles, which determine 72
directions where the radiation intensity is computed. We assumed
that the air does not participate in the radiation and that the
surfaces inside the domain are grey and diffuse.
4.3. Boundary conditions
This section reports the flow, turbulence and thermal boundary
conditions that are needed for the air at the limits of the proposed
computational domains.
A non-slip condition is ascribed to fluid velocity at all the solid
walls, and the shear stress is calculated considering smooth
surfaces. The normal velocity component and the normal gradients
of all the other variables are set to zero at the plane of symmetry.
On the inflow surfaces of the flow domain the total pressure is set to
zero and the direction of the velocity is imposed normal to the
(1) surface. On the outflow surfaces the static pressure is set to zero.
Regarding the turbulence at the solid walls, in the simulations
using the ChangeHsieheChen turbulence model, the turbulent
(2)
kinetic energy is set to zero and its rate of dissipation is calculated
as proportional to the second normal derivative of k. At the inflow
surfaces the turbulent variables are calculated from the imposition
(3) of a low level of turbulence by means of a turbulent intensity of 1%
and a turbulent viscosity ratio of 1.
The grating grilles of substation TS01 and the expanded metal
grilles presented in both substations are modelled as plane surfaces
where the air flow suffers a head loss. The curve relating the head
loss with the normal velocity of the flow has been obtained
numerically for the grating grilles and from a experimental corre-
lation from the literature [20] for the expanded metal grilles. The
louvered grilles in the windows of TS01 and in the protruding vents
of TS02 are geometrically modelled in the computational domain
because they modify the trajectory of the air flow.
The external walls of the enclosure have been modelled as solids
with one-dimensional heat conduction and the thermal conduc-
tivity of the concrete. In the outer side of these external walls mixed
(convection and radiation) boundary conditions are imposed. The
heat transfer coefficients have been obtained from empirical
correlations for natural convection presented in the literature [21],
and the radiation exchanges are calculated between a grey diffuse
surface (with the known values of the emissivities of the external
surfaces) and a black body at the ambient temperature.
The surface representing the protection plate has been modelled
as adiabatic. The surfaces of the MV cubicles are considered adia-
batic, and a uniform heat flux is imposed (384.4 W/m2) on the
surfaces of the LV boards.
On the surfaces of the transformer temperature distribution
maps extrapolated from the experimental measured temperatures
have been imposed. In Fig. 5, the temperature distribution maps
imposed on the transformer for the simulations of the temperature
rise test TS01-E01 are shown.
4.4. Discretization and resolution
The Finite Volume Method (FVM) is applied to discretize the
differential equations of the mathematical model described above,
 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863 857
Table 3
Comparison between the experimental results and the model results for the simulations of the temperature rise tests.

TS01-E01 TS01-E02 TS01-E03 TS01-E04

Mod. Exp. Dif. Mod. Exp. Dif. Mod. Exp. Dif. Mod. Exp. Dif.
P [W] 12,620 12,407 213 12,356 12,407 51 6502 6204 298 6056 6204 148
T1 [ C] 16.1 16.1 e 15.9 15.9 - 14.6 14.6 e 14.6 14.6 e
T2 [ C] 46.0 48.0 2.0 51.9 53.9 2.0 37.7 39.0 1.3 49.8 49.1 0.7
T3 [ C] 46.1 40.0 6.1 50.1 41.7 8.4 36.3 30.4 5.9 39.2 16.9 22.3
T4 [ C] 39.5 37.9 1.6 55.9 46.3 9.6 35.0 33.3 1.7 46.5 47.3 0.8
T5 [ C] 43.5 43.0 0.5 49.4 50 0.6 37.2 36.0 1.2 44.0 44.9 0.9
T6 [ C] 26.8 26.0 0.8 30.0 27.6 2.5 23.5 22.2 1.3 29.7 25.4 4.3
T7 [ C] 30.8 33.4 2.6 35.3 37.8 2.5 26.5 27.5 1.0 35.3 33.0 2.3
Vinlet 0.423 0.299 0.124 0.359 0.385 0.026 0.291 0.376 0.085 0.138 0.248 0.110
[m/s]
Vinside 0.050 0.120 0.070 0.220 0.274 0.054 0.173 0.234 0.061 0.172 0.129 0.043
[m/s]
Voutlet 0.437 0.436 0.001 0.323 0.334 0.011 0.249 0.240 0.009 0.204 0.100 0.104
[m/s]

using a segregated implicit solver to solve the generated algebraic
equation system. Equations are linearized and then sequentially
solved using the GausseSeidel algorithm accelerated by an alge-
braic multigrid method [22]. The pressureevelocity coupling is
achieved through the use of the SIMPLE algorithm [23]. Diffusive
terms of the equations are discretized using a second-order centred
scheme, and the convective terms are discretized using a second-
order upwind scheme [14]. A body force weighted scheme [24] is
chosen in the discretization of pressure to deal with this buoyancy-
driven flow. All this numerical procedure has been implemented in
the unstructured CFD code Fluent V.6.3 [25].
The mathematical model has been solved in an HP Proliant
DL585 G6 server with 4 Opteron 8439 SE dual core processors and
with 128 GB of RAM memory.
4.5. Convergence criteria
Three main convergence criteria have been applied to deter-
mine when the numerical procedure described in the previous
paragraph has converged to a solution. The first criterion consists of
reaching stable values for the power dissipated by the transformer
and for the temperatures of both the internal surfaces of the walls
of the enclosure and of the air at the outlet of the substation,
meaning that a converged steady state has been reached. Variations
of less than 1% for the power and of less than 0.3  C for the
temperatures in 1000 iterations are required. The second criterion
is the fulfilment of global continuity and energy balances: the
equality between the air mass flow rate at the inlet and at the outlet
of the flow domain, and the equality between the power dissipated
by the transformer and the power that leaves the domain through
the walls of the enclosure and with the air flow. Differences of less
than 0.1% for the former and 0.5% for the latter are required. The
final criterion is to check that the values for the scaled residuals of
the equations are below certain magnitudes: 103 for the mass,
momentum and turbulent equations, and 106 for the energy
equation and the DO radiation equation.
4.6. Grid independency
The grid independency of the mathematical model results has
been checked by means of the so called Grid Convergence Index
(GCI), based on the Richardson Extrapolation method [26]. This
value is used to determine the discretization error by comparing
the results for three different meshes. The first mesh (N1: 31, 107,
372 elements) is the finest, the second case (N2: 27, 675, 514)
represents an intermediate grid level and the third case (N3: 21,
824, 429) is for the coarse mesh. These three meshes have been
chosen taking into account the fact that it was not possible to
increase the finest mesh due to hardware limitations and that the
coarse mesh cannot be reduced in order to assure that this grid
level is inside the asymptotic range of convergence. The resulting
refinement factors are r21 ¼1.04 and r32 ¼ 1.08.
Three results from the model’s simulation for the conditions of
the experimental test TS01-E02 have been selected in order to
evaluate the discretization error: the air mass flow rate, the power
dissipated by the transformer, and the air temperature rise over the
ambient temperature at a point situated over the transformer. The
results of the evaluation are presented in Table 2. As can be seen,
the values of the GCI are acceptable for the three checked variables;
the maximum value is 6.0% for the air mass flow rate. These results
show that the assumed discretization error is low enough if the
finest mesh, N1, is chosen for the mathematical model.
5. Validation of the model
The mathematical model described in the previous section has
been validated by comparing the results obtained in the simula-
tions under the conditions of the 8 heating tests mentioned in
Section 3 with the experimental results.
The comparison of some of the results is presented in Table 3.
The correspondence between the symbols used in the table and the
measured variables is as follows: P represents the transformer
power losses; T1 corresponds to the ambient temperature; T2
corresponds to a thermocouple measuring the air temperature over
the transformer; T3 is the air temperature in a location at the outlet
ventilation grilles; T4 and T5 correspond to two thermocouples sited
between the central fins of the long and short side, respectively, of
the transformer; T6 and T7 are the temperatures measured from
two thermocouples located on the external and internal surfaces of
the right wall of the enclosure of the transformer zone, looking to
the transformer from the protection plate; Vinlet is the average air
velocity measured by the hot wire anemometer sited at the inlet
external ventilation grilles; Vinside is the average air velocity
measurement from the omni-directional anemometer located over
the transformer inside the substation; and Voutlet is the average
speed measurement from the omni-directional anemometer at the
outlet grilles.
From the results in Table 3 it can be concluded that the valida-
tion of the model is acceptable. For the transformer power losses,
the maximum difference between the model and the test occurs for
the test TS02-E08 and is of 5.5%. For the air temperature over the
transformer, T2, in all cases the difference is less than 2  C, except
for test E06. Taking into account that we are comparing tempera-
tures measured at a concrete point rather than average
858 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863

TS02-E05 TS02-E06 TS02-E07 TS02-E08

Mod. Exp. Dif. Mod. Exp. Dif. Mod. Exp. Dif. Mod. Exp. Dif.
12,556 12,407 149 12,442 12,407 35 6477 6204 273 6547 6204 343
14.6 14.6 e 16.4 16.4 e 17.1 17.1 e 18.2 18.2 e
46.7 48.0 1.3 50.3 56.9 6.6 40.6 42.6 2.0 49.6 49.8 0.2
48.5 45.3 3.2 51.4 56.8 5.4 37.8 42.4 4.6 49.2 50.6 1.4
43.3 34.4 5.9 53.7 45.7 8.0 38.9 36.0 2.9 45.8 44.4 1.4
42.2 43.0 0.8 50.6 52.0 1.4 38.2 39.5 1.3 46.0 46.8 0.8
26.2 26.0 0.2 30.2 31.4 1.2 26.6 26.4 0.2 31.5 30.0 1.5
30.5 32.5 2.0 35.5 39.3 3.8 30.1 30.9 0.8 36.5 36.2 0.3
0.534 0.892 0.358 0.202 0.068 0.134 0.167 0.040 0.127 0.524 0.559 0.035

0.054 0.257 0.203 0.148 0.107 0.041 0.151 0.150 0.001 0.233 0.203 0.030

0.742 0.718 0.024 0.445 0.734 0.289 0.216 0.104 0.112 0.999 0.461 0.538

temperature of a zone, it can be considered a good result. For the
temperatures at the outlet ventilation grilles, T3, the results are not
so good. However, this is because in these outlet grilles there is
a zone of reverse air flow; the thermocouple is located at the
boundary between the outflow and the backflow air. It is also
the case that the model is not predicting exactly the same area for
the reverse flow zone as the experimental tests. This will be
explained in detail in the next section. For the air temperatures
between the fins, T4 and T5, and for the wall temperatures, T6 and T7,
the results are, in general, acceptable.
Regarding the average air velocities, the results of the model are
also acceptable. For the eight simulations, the results are in general
better for the air velocities inside the substation and at the outlet
grilles. However, these results must be considered in a qualitative
way because the velocity measurements showed a high fluctuation
(as it is reflected in the uncertainty value presented in Section 3).
6. Analysis of the results and discussion
The first result to be analysed is the difference between the
model and the experiments regarding the temperature of the air at
the outflow ventilation grilles. It must be taken into account that
the result corresponds to the point where the thermocouples were
located, not to an average temperature on all the surface of the
outflow grilles.
Fig. 6 shows an infrared image of the outlet grating ventilation
grilles during the running of heating test TS01-E02. The white-
yellow colours represent high temperatures and the purple-black
colours low temperatures. The white dot shows the location of
the thermocouple measuring temperature T3 in Table 3. In this
picture it can be seen that there are zones of the grating grilles that
have the purple colour, which means that there is no hot air leaving
the substation through these zones, but instead external air at

Fig. 2. Location of some of the probes during the experimental tests: (a) photo of the thermocouples on the transformer surface; (b) diagram of the thermocouples on the
substation walls; (c) photo of one anemometer inside the substation; and (d) photo of two anemometers at the outflow grille.
 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863 859

Fig. 3. Boundaries and main components of the computational flow domains of substations TS01 (left) and TS02 (right).

a lower temperature is entering back into the substation. (For
interpretation of the references to colour, the reader is referred to
the web version of this article.) In other words, there is a reverse air
flow.
The same behaviour has been obtained from the simulation of
the mathematical model. Fig. 7 (left) represents the temperature
contours at the surface of the flow domain representing the
outflow grating grilles for the simulation of test TS01-E02. Fig. 7
(right) shows the z-component of the velocity vectors at the
same surface. In both figures the white dot represents the position
where temperature T3 in Table 3 was measured. It can be seen in
Fig. 7 (right) that there are zones on the outlet grating grille where
the z-component of the velocity vectors is negative; that is, the air
flow is not going out of the substation but rather entering it. This is
confirmed by the non-uniform temperature distribution of Fig. 7
(left). In the zones on the grating grille where there is a reverse
flow the temperature of the air is lower (green-blue colour), rep-
resenting a mixing temperature between the outflow and the
ambient air. It can be seen in Fig. 7 that the white dot representing
the position of thermocouple T3 is in the zone where the air
velocity passes from positive (leaving the substation) to negative
values (entering the substation). (For interpretation of the refer-
ences to colour, the reader is referred to the web version of this
article.)
The above explanation could be the reason for the not so good
correlation between the model and the experiments for T3: the
zones of reverse flow in the real grating grilles and in the modelled
grilles do not exactly coincide. Moreover, in reality the thermal
plumes of the outflow air are not static structures; instead, they are
changing with time, so the mean thermal plume predicted by the
model is not in exactly the same position as in the experimental test
and the area of the zone of the grating grilles with reverse flow is
not always the same. In summary, the mathematical model is
predicting the reverse flow in the outlet grating grilles in the same
zone as in the experimental tests, though it does not have quite the
same surface area, resulting in a difference in temperature T3. In
test TS01e04, the difference is much more pronounced than in the
rest: the empirical temperature T3 is 16.9  C, just 2.5  C over the
ambient temperature, confirming the existence of the reverse air
flow.
Similar results have been obtained for substation TS02, the one
with protruding louvered vents. In this case the reverse flow
appears in the lower louvers of the vent. For this substation the
differences are smaller and in some tests they are negative,
implying that the model is predicting that fewer louvers of the
grille have reverse flow.
Regarding the air temperatures inside the substation, T2, T4 and
T5 in Table 3, the model appears to predict them acceptably. T2

Fig. 4. Computational mesh in the surface of the transformer.

860 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863
Fig. 5. Contours of temperature [K] imposed on the transformer for the simulation of heating test TS01-E01.
corresponds to the temperature at a point over the transformer
where a thermocouple was sited and, as reported previously, the
differences with the experiments are less than 2 degrees for all the
tests, except test E06. T4 and T5 are the temperatures of the top air
between the central fins of the long and the short side of the
transformer, respectively. For T4 the differences between the model
and the test are around 2  C in all the tests, except in cases E02, E05
and E06. A possible explanation for this difference is that, as the fins
of the long side of the transformer are in front of the main venti-
lation air flow, the model is predicting a hot thermal plume that
does not exist in the real experimental tests. The reason why the
hot thermal plume appears only in these three tests and not in the
rest stays hidden. On the other hand, the differences in T5 are less
than 1.5  C for all the tests. In this case, the fins of the short side of
the transformer are in front of the lateral walls of the substation,
a zone where the ventilation air flow is not so important, and the
model seems to give a better prediction.
The simulations of the model can be used to analyse the air
thermal and flow patterns inside the substation. Figs. 8 and 9
represent the temperature contours (left) and the velocity vectors
(right) for the simulations of tests TS01-E02 and TS02-E06,
respectively.
Analysing the temperature and velocity fields shown in Fig. 8
together, it can be said that the fresh air entering the substation
through the inflow grilles in the MVeLV board zone goes directly to
Fig. 6. Thermographic image of the outlet grating grilles during the temperature rise
test TS01-E02 and location of the thermocouple measuring temperature T3.
the bottom part of the substation for two reasons. First, fresh air has
lower density and it weighs more than hot air. Second, the “engine”
behind the air flow is the heat dissipated by the transformer and so
the air passes to the transformer zone through the free space below
the protection plate that separates the MVeLV board zone from the
transformer zone. After passing between the fins of the trans-
former, the refrigeration air is redirected to the outflow windows
and from here enters the prefabricated concrete vent and leaves the
substation through the horizontal grating grilles. In this zone, the
reverse flow mentioned previously can be observed once again.
It can be also observed that in the top part of the MVeLV board
zone of the substation there is warm air at an intermediate
temperature (about 35e37  C), which is not participating in the
ventilation of the substation (low velocities in Fig. 8 (right)). This air
is heated by the power dissipated by the LV boards (1583 W).
In the transformer zone and above it, the air is at a temperature
of around 50  C, and the plumes of hot air (z60  C) coming from
the fins of the transformer and going to the outflow windows can
be observed.
In analysing the temperature and velocity fields shown in Fig. 9,
the same conclusions pointed out in the above paragraphs for
substation TS01 hold for substation TS02. The only difference in this
case is that there is an acceleration of the inlet fresh air because the
inlet protruding vent is just over the MV cubicles and the inflow air
collides with them, and most of the flow tends to pass through the
small space between the MV boards and the walls of the enclosure.
Another interesting result that has been obtained with the
simulation of test TS01-E02 is that the two central windows of the
six windows presented in the inlet prefabricated concrete vent are
not contributing to the ventilation of the substation. This can be
seen in Fig. 10, where, with the same time of integration, only a few
pathlines released from the central window of the domain (in blue
colour) arrive at the outlet grating grille, meanwhile the most of the
pathlines from the other windows (in green, yellow and red
colours) exit the substation. The most of the pathlines from the
central window goes to the bottom part of the MVeLV board zone
of the substation, a region where the air velocities are low, being an
indication of its small contribution to the cooling of the substation.
To verify this result a simulation closing the central window of
the model was carried out. The air mass flow rate circulating
 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863
Fig. 7. Contours of temperature in  C (left) and z-component of the velocity vectors coloured by its magnitude in m/s (right) at the outlet grating grille for the simulation of the
heating test TS01-E02.
through the substation was diminished by 8% with respect to the
test TS01-E02. Taking into account that the reduction in the
ventilation area by the closing of one window is 33%, it can be
concluded that the influence of the ventilation air flowing through
the central windows is very limited.
The most important results of the simulations of the model are
those that cannot be experimentally determined but can be used in
the development of a zonal model: the mass flow rate of the air
circulating through the substation and the heat transfer coefficients
over the surfaces of the transformer and of the walls of the
enclosure.
Table 4 shows the eight temperature rise tests carried out with
their ventilation conditions and transformer power losses, the
experimentally measured transformer top-oil temperature rise
over the ambient temperature, and the results of the air mass flow
rates in the simulations of the eight tests with the mathematical
model.
Several interesting conclusions can be drawn from Table 4. First
of all, the direct relationship between the reduction of the venti-
lation or of the transformer power losses and the increment of the
top-oil temperature rise and the reduction of the air mass flowing
through the substation can be seen. For substation TS01, when the
ventilation grilles are installed (test E02 versus test E01) there is
a relative increase in the oil temperature rise of 12.3% and a relative
reduction in the air mass flow rate of 18.3%. Comparing the same
tests for substation TS02 (test E05 versus test E06) the increase in
the oil temperature rise is 8.6% and the reduction in the air flow rate
is 16.4%. Thus, it can be concluded that the grilles of substation TS02
handicap the ventilation to a lesser extent than the grilles of TS01.
Once the grilles are installed, reducing the transformer power
Fig. 8. Contours of temperature in  C (left) and velocity vectors coloured by velocity magnitude in m/s (right) at two perpendicular planes for the simulation of the heating test
TS01-E02.
861
losses by 50% (test E02 vs. test E03 and test E06 vs. test E07) gives
similar results in both substations: a reduction of 38.7% and 38.4%,
respectively in DToil and a reduction of 16.9% and 17.9%, respec-
tively, in the air flow rate.
When comparing test TS01-E03 vs. test TS01-E04 and test TS02-
E07 vs. test TS02-E08, although the reduction in ventilation is
dissimilar for both substations, it is greater in TS01; the increase in
DToil is similar between the compared tests for both substations:
13.8% and 13.1%, respectively. However, the reduction in the air flow
rate is greater for TS01: 42.9% versus 28.9% for TS02. This would be
an indication that ventilation is better in substation TS01.
In Table 4 and in the analysis of it presented in the previous
paragraph, it has been demonstrated that the most influencing
parameters governing the quantity of air flowing through the
substation are the ventilation grilles and the heat losses from the
transformers. Hence, from the results presented a correlation
relating these three parameters can be deduced. The ventilation
grilles are characterized by the discharge coefficient, Cd,grilles, and
the surface area, Agrilles, and the transformer losses only by the
convection heat losses, qconv,transf, because the radiation heat losses
do not directly influence the air flow rate. The adjusted correlation
is shown in Eq. (4) and the numerical results with the adjusted
correlation and the 10% error lines are shown in Fig. 11.
 0:7001
_ air ¼ 0:0358$q0:297
m conv;transf $ Cd;grilles $Agrilles (4)
total
Regarding the heat transfer coefficients on the surfaces of the
transformer and on the walls of the enclosure, for the lid and the
base of the transformer and for the walls of the enclosure empirical
862 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863

Fig. 9. Contours of temperature in  C (left) and velocity vectors coloured by velocity magnitude in m/s (right) at two perpendicular planes for the simulation of the heating test
TS02-E06.

correlations from the literature [21] for the average Nusselt number Table 4
have been fitted with the results of the eight simulations of the Experimental oil temperature rise and numerical air mass flow rates in the eight
tests.
temperature rise tests.
However, for the fins of the transformer the variation of the heat Test name Ventilation condition Transformer power DToil [K] _ air ½kg=s
m
transfer coefficients with the vertical direction is important and losses [%]

must be taken into account, as pointed out in [11]. For this reason, TS01-E01 Without grilles 100 65.2 0.3863
TS01-E02 Standard 100 73.2 0.3184
on the surface of the fins of the transformer a new correlation for
TS01-E03 Standard 50 44.9 0.2623
the local Nusselt number in the vertical direction has been fitted. TS01-E04 Reduced 50 51.1 0.1498
The correlation has been obtained as the average of the correlations TS02-E05 Without grilles 100 67.3 0.3646
fitted for the numerical results of a vertical line in the centre of the TS02-E06 Standard 100 73.1 0.3047
surface of the four central fins on each side of the transformer for TS02-E07 Standard 50 45.0 0.2501
TS02-E08 Reduced 50 50.9 0.1778
the eight experimental tests. That is, the final fitted correlation is
the average of thirty-two fitted correlations. Eq. (5) shows the
average fitted correlation and Fig. 12 shows the fitted correlation
with the 30% error lines and the numerical results for one fin for can be used to analyse and optimise the ventilation of underground
each of the eight temperature rise tests simulated. transformer substations by means of design improvements.
The results from the simulations of the developed mathematical
Nuz ¼ 1:3254$Ra0:1845 (5) model can be used to obtain parameters that are related to or that
z
influence ventilation, parameters that cannot be determined
The qualitative and quantitative results of the air flow patterns experimentally. This information will prove to be instrumental in
and temperature distributions shown and discussed in this section developing a simplified zonal thermal model of the natural venti-
lation of underground transformer substations.

0.50
0.45
Air mass flow rate [kg/s]

0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00
2 3 4 5 6 7
0.7001 0.297
(C d,grilles ·A grilles ) ·(q conv )

Fig. 11. Numerical results of the air flow rate (red dots); fitted correlation (continuous
Fig. 10. Pathlines releasing from the inlet windows coloured by the particle number line) and 10% error lines (dashed lines). (For interpretation of the references to colour
for the simulation of the temperature rise test TS01-E02. in this figure legend, the reader is referred to the web version of this article.)
 J.C. Ramos et al. / Applied Thermal Engineering 51 (2013) 852e863
50
Nuz
5
1E+04 1E+05 1E+06 1E+07 1E+08 1E+09
Raz
Fig. 12. Numerical results of local vertical Nusselt number on the surface of the fins
(different dots); fitted correlation (continuous line) and 30% error lines (dashed
lines). (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.)
7. Conclusions
A mathematical differential model to represent the ventilation
and the thermal performance of underground transformer
substations has been developed and numerically solved using CFD
techniques. The grid independence of the model has been verified
and it has been experimentally validated with the results of eight
temperature rise tests carried out under different conditions of
ventilation and transformer power losses.
The simulations of the model have served to obtain the air flow
patterns and the air temperature distributions. The analysis of the
air flow patterns has served to conclude that there is an important
component of reverse flow in the external outflow ventilation
grilles for both substations and that in substation TS01 the two
central windows of the inlet ventilation windows are not contrib-
uting to ventilation. The analysis of the air temperature distribu-
tions has served to detect a zone of static air at an intermediate
temperature at the top part of the substation in the zone of the MV
and LV boards.
Two important parameters related to the ventilation perfor-
mance of the substation that are very difficult to determine
experimentally have been obtained from the simulations of the
model: the air mass flow rate through the substation and the heat
transfer coefficients on the surfaces of the transformer and on the
walls of the enclosure. Correlations for the air mass flow rate and
for the local heat transfer coefficient on the surfaces of the fins of
the transformer have been fitted with the results of the eight
simulations of the temperature rise tests.
All the information provided by the CFD detailed mathematical
model is going to be used to develop a simplified zonal algebraic
thermal model that can be used as a design and optimisation tool
for transformer substation manufacturers.
Acknowledgements
This research has been carried out as part of the research project
called CRISALIDA, with the support of Grupo Ormazabal. CRISALIDA
863
was funded by Ministerio de Industria, Turismo y Comercio through
the CENIT (Consorcios Estratégicos Nacionales en Investigación
Técnica) program. The financial support of Cátedra Fundación
Antonio Aranzábal e Universidad de Navarra is also gratefully
E01 Back Fins
acknowledged.
E02 Left Fins
E03 Front Fins
References
E04 Right Fins
E05 Back Fins
E06 Left Fins [1] IEC 60076-2:1997, Power Transformers. Part 2: Temperature Rise, second ed.,
IEC Standard, 1997.
E07 Back Fins
[2] IEC 62271e202:2006, High-Voltage Switchgear and Controlgear. Part 202:
E08 Back Fins High Voltage/Low Voltage Prefabricated Substations, IEC Standard, 2006.
Fitted Correlation [3] W. M. M. Menheere, Transformer stations and natural ventilation, in: CIRED
13th International Conference on Electricity Distribution, Brussels (1995),
+30%
Paper no. 1e23.
-30% [4] Z. Radakovic, S. Maksimovic, Non-stationary thermal model of indoor trans-
former stations, Electrical Engineering 84 (2002) 109e117.
1E+10
[5] Z. Radakovi c, S. Maksimovi c, Dynamical thermal model of oil transformer
placed indoor, in: CIRED 20th International Conference on Electricity Distri-
bution, Prague (2009), Paper no. 0304.
[6] I. Iskender, A. Mamizadeh, An improved nonlinear thermal model for MV/LV
prefabricated oil-immersed power transformer substations, Electrical Engi-
neering 93 (2011) 9e22.
[7] N. Loucaides, Y. Ioannides, V. Efthymiou, G. E. Georghiou, Thermal modeling of
power substations using the finite element method, in: Seventh Mediterra-
nean Conference and Exhibition on Power Generation, Transmission, Distri-
bution and Energy Conversion, 7e10 November 2010, Agia Napa, Cyprus,
Paper no. MED10/188.
[8] J.C. Ramos, A. Rivas, J.M. Morcillo, Numerical thermal modelling of the natural
ventilation of a half-buried transformer substation using CFD techniques, in:
R. Bennacer (Ed.), Progress in Computational Heat and Mass Transfer, vol. II,
Lavoisier, Paris, 2005, pp. 929e934.
[9] M. Musy, E. Wurtz, F. Winkelman, F. Allard, Generation of a zonal model to
simulate natural convection in a room with a radiative/convective heater,
Building and Environment 36 (2001) 589e596.
[10] J. Stewart, Z. Ren, Prediction of indoor gaseous pollutant dispersion by nesting
sub-zones within a multizone model, Building and Environment 38 (2003)
635e643.
[11] J. Gastelurrutia, J.C. Ramos, G.S. Larraona, A. Rivas, J. Izagirre, L. del Rio,
Numerical modelling of natural convection of oil inside distribution trans-
formers, Applied Thermal Engineering 31 (2011) 493e505.
[12] J. Gastelurrutia, J.C. Ramos, A. Rivas, G.S. Larraona, J. Izagirre, L. del Rio, Zonal
thermal model of distribution transformer cooling, Applied Thermal Engi-
neering 31 (2011) 4024e4035.
[13] T.-W. Lee, Thermal and Flow Measurements, Taylor & Francis Group, New
York, 2008.
[14] S.R. Mathur, J.Y. Murthy, A pressure-based method for unstructured meshes,
Numerical Heat Transfer e Part B: Fundamentals 31 (2) (1997) 195e215.
[15] R.A.W.M. Henkes, C.J. Hoogendoorn, Comparison exercise for computations of
turbulent natural convection in enclosures, Numerical Heat Transfer e Part B:
Fundamentals 28 (1) (1995) 59e78.
[16] K.C. Chang, W.D. Hsieh, C.S. Chen, A modified low-Reynolds-number turbu-
lence model applicable to recirculating flow in pipe expansion, ASME Journal
of Fluids Engineering 117 (1995) 417e423.
[17] D. Choudhury, Introduction to the Renormalization Group Method and
Turbulence Modelling, Fluent Inc., 1993, Technical Memorandum, TM-107.
[18] T. Jongen, Y.P. Marx, Design of an unconditionally stable positive scheme for
the keε and two-layer turbulence models, Computers and Fluids 26 (5) (1997)
469e487.
[19] G.D. Raithby, E.H. Chui, A finite volume method for predicting a radiant heat
transfer in enclosures with participating media, Journal of Heat Transfer 112
(1990) 415e423.
[20] I.E. Idelchik, Handbook of Hydraulic Resistance, third ed., Jaico Publishing
House, Delhi, 2006.
[21] F.P. Incropera, D.P. De Witt, T.L. Bergman, A.S. Lavine, Fundamentals of Heat
and Mass Transfer, sixth ed., John Wiley & Sons, New York, 2006.
[22] B.R. Hutchinson, G.D. Raithby, A multigrid method based on the additive
correction strategy, Numerical Heat Transfer 9 (1986) 511e537.
[23] S.V. Patankar, Numerical Heat Transfer and Fluid Flow. pp. 126e131, Hemi-
sphere Publishing Corporation, New York, 1980.
[24] J.H. Ferziger, M. Peric, Computational Methods for Fluid Dynamics. pp.
157e217, third ed., Springer-Verlag, Germany, 2002.
[25] Fluent Inc, Fluent 6.3 User’s Guide, Cavendish Court, Lebanon, NH 03766,
2006.
[26] P.J. Roache, Perspective: a method for uniform reporting of grid refinement
studies, ASME Journal of Fluids Engineering 116 (1994) 405e413.