You are on page 1of 12

Applied Thermal Engineering 31 (2011) 4024e4035

Contents lists available at ScienceDirect

Applied Thermal Engineering


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

Zonal thermal model of distribution transformer cooling


Jon Gastelurrutia a, Juan Carlos Ramos a, *, Alejandro Rivas a, Gorka S. Larraona a, Josu Izagirre b,
Luis del Río b
a
TECNUN (University of Navarra) e Dept. of Mech. Engineering, Thermal and Fluids Engineering Div., Paseo de Manuel de Lardizábal 13, 20018 San Sebastián, Spain
b
O.C.T. (Ormazabal Corporate Technology), Parque Empresarial Boroa, 48340 Amorebieta-Etxano, Spain

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

Article history: The present paper proposes an algebraic zonal model describing the cooling process, via internal oil and
Received 20 December 2010 external air natural convection, of distribution transformers during a standardised heating test. The
Accepted 2 August 2011 conception and adjustment of the presented algebraic zonal model rely on the oil flow and thermal
Available online 10 August 2011
results obtained by a more complex numerical differential model that has been verified and validated by
the authors. The domain of the original differential model is divided into several interrelated control
Keywords:
volumes or zones where mass and energy conservation laws are applied in conjunction with the
Distribution transformer
necessary boundary conditions. The generated nonlinear algebraic equation system is solved, obtaining
Thermal model
Zonal model
in the process the top oil temperature, the internal and external surface temperatures and the heat
Cooling dissipation distribution. The results of the zonal model have been validated using the previous results
Natural convection from the differential model as well as experimental measurements from heating tests with three
different transformers under various power loads. The model developed has proven to be an efficient tool
in improving and optimising the thermal design of these devices.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction factors) of each transformer are unknown values, requiring esti-


mations or complex internal temperature measurements.
One of the most common devices found in an electrical power Unizone or network models, based on a thermal-electrical
delivery network is the ONAN (Oil NaturaleAir Natural) cooled analogy [5,6], have been created to improve the previous models
three-phase distribution transformer. These transformers convert with new assumptions and adjustment methodologies. These types
electrical power from medium to low voltage, and this conversion of models are quite useful when the focus is on monitoring the
leads to some power losses in the internal core and coils. These thermal performance of distribution transformers for control
power losses are transformed into heat, and they have to be purposes.
adequately removed in order to avoid undesirable temperature Otherwise, if the goal is to analyse and optimise the thermal
rises inside the transformer [1]. design of a given transformer, the use of other types of mathe-
The top oil temperature rise obtained during a standard heating matical models with a more exhaustive treatment of the physical
test and an indirect measurement of the winding temperature rise phenomena that is taking place inside is required. The principal
by means of the evaluation of the electrical resistances [2] can be effort in this direction is by means of numerically solved models
used to approximate the hot-spot temperature reached in the using the differential expressions of the NaviereStokes and energy
internal insulation of the transformers under different standardised conservation equations. Examples of this approach can be found in
loading levels [3]. Mufuta and Van den Bulck [7], El Wakil et al. [8] and Smolka et al.
Based on the standardised heating test, Aubin and Langhame [4] [9,10], whose contributions are described in a previous work of the
proposed an improved thermal model to study the transient authors [11].
behaviour of distribution transformers, but their predictions may In the above mentioned paper, the authors presented a differ-
be regarded only as approximations because the thermal charac- ential model describing the cooling of ONAN distribution trans-
teristics (thermal degradation rate, time constants and adjustment formers and obtaining accurate flow and thermal results. This
information about the inner driving heat transfer phenomenon
could barely have been extracted experimentally. The model was
* Corresponding author. Tel.: þ34 943219877; fax: þ34 943311442. validated for three heating tests on the same transformer at
E-mail address: jcramos@tecnun.es (J.C. Ramos). different power losses, and additionally for two other transformers

1359-4311/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.applthermaleng.2011.08.004
J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035 4025

Nomenclature b thermal expansion coefficient, [1/K]


dT thickness of a thermal boundary layer, [m]
ε emissivity, [-]
Latin Letters l thermal conductivity, [W/m$K]
a semi-length of a rectangular duct, [m] n kinematic viscosity, [m2/s]
A surface area, [m2] r density, [kg/m3]
b semi-width of a rectangular duct, [m] s StefaneBoltzmann constant, [W/m2$K4]
c length of a side of a triangular duct, [m] Ø diameter, [mm]
d width of the domain of flow [m]
cp specific heat, [J/kg K] Subscripts
C suction proportionality coefficient, [-] Air thermophysical air properties
D orthogonal distance, [m] Amb ambient
e thickness of the casing, [m] Casing external casing
E rate of flow thermal energy, [W] Channel cooling channels
h heat transfer coefficient, [W/m2$K] Coil coils of the transformer
H height, [m] Cond heat transfer by conduction
L length, [m] Conv heat transfer by convection
m_ oil mass flow rate, [kg/s] Core tri-phase core of the transformer
n number of cooling ducts in a channel, [-] Ext external
N number of elements, [-] fExt external face of a control volume
Nu Nusselt number, [-] fFluid face of fluid between control volumes
P power losses, [W] fInt internal face of a control volume
q" heat flux, [W/m2] Fins hollow fins in the casing
Q heat transfer, [W] HV high voltage side of the coils
Ra Rayleigh number, [-] In oil getting into a control volume
S relative difference in percentage [%] Int internal
T temperature, [ C] Losses generation due to power losses
DT temperature rise over ambient, [ C] LV low voltage side of the coils
V vertical velocity, [m/s] N nominal value
Vol solid volume, [m3] OileAir from the oil to the air
W width, [m] Out oil getting out from a control volume
x local normal coordinate, [m] Rad radiation heat exchanges
y local vertical coordinate, [m] Vert vertical direction
z-d zonal minus differential
Greek Letters & Symbols z-e zonal minus experimental
a thermal diffusivity, [m2/s]

at their nominal power losses. The principal handicap of this optimise geometric parameters (for example, the number and
differential model is the high computational resources that are dimensions of the fins) in order to reduce the top oil temperature
needed to perform a single simulation, and the long time period rise. Moreover, simulation time and computer requirements are
that is required to obtain results. These characteristics invalidate critically reduced compared to differential models.
the use of these models for design and optimisation purposes.
At this stage, a zonal model that would help with the design task 2. Description of distribution transformers
would be desirable, filling the gap between the limitations of the
previous unizone or network models and the more computation- Fig. 1 shows the internal and external geometry of the generic
ally expensive differential models. These kinds of models can be ONAN distribution transformer that is modelled in the present
subdivided into two different blocks: the prefixed and non-prefixed paper, showing its principal constructive characteristics and
models. In a prefixed model the flow movement pattern is always dimensions. Some of the most representative values for the three
the same [12], while in the non-prefixed models it can change different transformers that have been used to check the validity of
depending on other variables in the simulation [13]. the proposed model are given in Table 1 for comparative purposes.
Based on the thorough information obtained from the previ- The complete description of the components and their task in
ously validated differential model, an algebraic prefixed zonal the performance of the transformer can be found in [11].
model was developed and set up, as shown in the present paper. Five different tests will be used to check the validity of the
The main objective of the model is to obtain an accurate top oil developed zonal model, as will be shown later. In tests E01, E02 and
temperature rise for a given transformer design at the end of E03 the power losses of Transf-01 are varied to study the influence
a heating test, but additional information such as internal and of different power loads on the results of the model. Nominal
external surface temperatures and heat distribution will be also power losses, PN, are considered for test E03, three quarters of them
provided. The main advantage of the model presented here is that for test E02, 3/4 PN, and finally a half for test E01, ½ PN. In tests E04
when compared with the unizone models it is predictive because and E05 nominal power losses are considered for Transf-02 and
a specific adjustment is no longer needed depending on the type of Transf-03, respectively, to analyse the influence of considering
transformer tested and on the test conditions. The model devel- different transformer configurations and dimensions in the thermal
oped can be used by transformer designers to redesign and performance of the model.
4026 J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035

Fig. 1. Geometry of an ONAN distribution transformer with main dimensions and components.

3. Mathematical zonal model model reported in [11]. The experimental validation of that differ-
ential model guarantees an adequate representation of the cooling
The algebraic zonal model that is shown in the present paper of ONAN distribution transformers, constituting an excellent
can be considered a simplified version of the original differential starting point for both developing and adjusting the zonal model.

Table 1
Main characteristic of the commercial ONAN distribution transformers analysed.

Transf-01 Transf-02 Transf-03


PCore (W) 1450 993 1728
PLV (W) 3290 2801 5284
PHV (W) 3360 2860 5395
PN (W) 8100 6654 12407
WCore (m) 0.25 0.28 0.272
LCore (m) 0.9 0.871 0.989
HCore (m) 0.746 0.761 0.797
ØCoil, a (m) 0.425 0.425 0.466
HCoil (m) 0.37 0.39 0.39
NChannel 1 LV LV/HV 2 HV 1 LV LV/HV 2 HV 2 LV 2 HV 2 HV
Cross-Sectiona

2aa (mm) 15 e e 15 e e 15/15 e e


2ba (mm) 4 e e 3 e e 4/4 e e
ca (mm) e 6.5/4.5 4.5/4.5 e 8.5 4.5/4.5 e 6.5/4.5 4.5/4.5
da (mm) 40 40 40 39 39 39 40 40 40
WCasing (m) 0.5 0.5 0.54
LCasing (m) 1.275 1.1 1.218
HCasing (m) 1.005 0.927 0.976
HFins (m) 0.8 0.7 0.7
LFins (m) 0.23 0.2 0.31
NFins 64 78 84
a
See Fig. 5.
J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035 4027

Although both models share the same flow domain and boundary The cooling performance of several transformers, under
limits, the discretisation and the definition of the equations that different testing conditions (E01-E05), has been numerically
have to be solved in each of them are completely different, gener- simulated using the differential model, identifying the principal
ating a new mathematical model. All these aspects are described parameters that govern the heat transfer phenomena. The
and analysed in the following paragraphs. velocity vectors in the mid-plane (with a schematic of the flow
pattern) and external surface temperatures (showing the
3.1. Description of the original differential model predominant vertical thermal stratification) obtained for E03 test
for Transf-01 are shown in Fig. 3. The same oil flow movement
In this section the geometry and the flow domain of the and thermal distribution can be ascertained for the rest of
differential model developed in [11] are briefly described, and the simulated cases allowing the creation of the proposed prefixed
principal numerical and resolution procedure issues are summar- zonal model.
ised. The results of the flow pattern and thermal distributions As it can be seen in Fig. 3, at the bottom of the transformer the
predicted by the differential model are also presented because they temperatures are relatively low and the oil movement is practi-
have been used to define the algebraic zonal model. cally inappreciable. The movement is focused on the upper hot
The differential model originally developed takes advantage of zone of the model, close to the walls, while in the central zones the
several simplifications of the real complex geometry of a distribu- oil has low velocity. The oil mass flow rates increase with the
tion transformer (details can be extracted from [11]). Only the oil vertical coordinate because convective boundary layers are being
confined in a 3D vertical cut of the real transformer centred with generated in the presence of internal hot walls. This phenomenon
respect to the second winding and containing one fin in conjunc- is especially remarkable inside the cooling channels of the coil
tion with the corresponding part of casing is considered in the flow where several ascending oil flows, (Oi) in Fig. 3, are generated.
domain of the model, as shown in Fig. 2. The oil flow domain is Their values (which vary from a lowest value of 0.297 g/s in the
internally limited by the solid parts of the transformer (core and first HV channel for test E01, up to 0.79 g/s in the first LV channel
coils), and externally by the surface of the casing and the fin, by for test E03) depend on the thermal properties of the oil, the heat
a couple of lateral symmetry planes, each of which is located in the flux dissipation on the walls, and the hydraulic diameters of the
middle between two consecutive fins on the longest side of the channels. In fact, an approximate proportionality between the
transformer, and by a symmetry plane normal to the lateral dissipated power value and the ascending oil mass flow can be
symmetry planes. The used boundary conditions are completely ascertained.
described later when explaining the boundary conditions of the In the image in the left part of Fig. 3, arrows have been added in
algebraic zonal model developed in the present paper because they order to help describe the main oil flow pattern. The flow move-
are shared by both models. ment by natural convection inside the casing is clearly influenced
The previously described flow domain has been discretised by by the presence of the coils and the oil getting in and out from its
means of a high quality hexahedral mesh of 1.25$106 elements. The cooling channels, (Oi). With all these contributions a notable rising
Finite Volume Method (FVM) has been used to discretise all the plume is generated close to the upper vertical wall of the core, at
governing equations (Reynolds Averaged NaviereStokes, energy the outlet of the channels, (A). This main oil flow rises up to the top
and turbulent magnitudes) of the differential model, obtaining the part of the domain where it is deflected by the lid and always
mean flow and thermal variables in the whole domain. Details of redirected to the fins, (B). The oil flow inside the hollow fins is
the turbulence model, the discretization of the equations, the determined by their relative position to the coils: the movement is
numerical procedure of resolution, the convergence criteria and the very active in the upper part and decreases when the oil moves
verification and validation of the model can be found in [11]. down. The main oil flow enters the hollow fins from the highest
zone of the casing, (C). When it arrives up to the height of the upper
zone of the coils, the flow is divided into two parts. The principal
fraction, (D, 1st Suction), is redirected to the outlets of the coil
channels and exits the hollow fin and the remaining fraction, (E),
moves down and is divided again. A part exits at the height of the
bottom inlets of the coil channels, (F, 2nd Suction), and a small
portion, (G), is derived to the base. The 1st Suction is always bigger
in comparison with the 2nd Suction (from a lowest value of 1.52 g/s
of 2nd Suction for test E01, up to 8.19 g/s of 1st Suction for test E03),
and their magnitude is again in direct relation to the dissipated heat
power value.
The vertical thermal stratification that is typical in these kinds of
closed cavities with natural convection [14] can be observed in
right-hand image in Fig. 3, but it is influenced by the previously
described Suction phenomena. Focusing our attention again on the
fins, it can be concluded that the external surface temperature is
almost uniform on the hollow fin over the 1st Suction, (1) in the
image to the right in Fig. 3, then it decays linearly between the 1st
and the 2nd Suction, (2), and finally the temperature drops sharply
on the bottom part of the fins, (3). This predominant vertical
temperature variation can be easily checked by comparing the
different oil temperature rises in the fins from the experimental data
shown later in Table 2: (DT02DT03)E03 ¼ 1 K, (DT03DT06)E03 ¼ 3.8 K,
(DT06DT07)E03 ¼ 21.4 K. As with the oil mass flow rates, the highest
Fig. 2. Flow domain of the differential model in comparison with the complete temperature variation is always focused on the 1st LV channel due to
geometry of Transf-02. the heat concentration in that zone.
4028 J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035

Fig. 3. Velocity vectors in the mid-plane in mm/s (left) and external surface temperatures in K (right) for Transf-01 (Test E03).

All these data and observations will be used in the next section Fig. 4 (left). These interchanges completely reflect the oil mass flow
to choose an adequate division of the flow domain of the zonal pattern that was previously described and shown in Fig. 3 (left). As
model, and later to validate an effective equivalence in the cooling seen in Fig. 4 (right), the second discretisation criterion is not
performance compared with the differential model. completely fulfilled in zones like the cooling channels (zones
10e13) or in the lower part of the fins and the casing (zones
3.2. Flow domain and control volumes of the zonal model 06e09), where a representative mean temperature has been
considered. Therefore, each control volume is characterised by
The same flow domain of the differential model described in the a unique oil temperature, and in addition the different walls
previous section has been used in the algebraic zonal model. The limiting each control volume are defined by a constant surface
control volumes of the zonal model has been defined bearing in temperature. Bearing in mind the geometric configuration shown
mind the fixed oil flow pattern and the predominant vertical in Table 1, thirteen control volumes are needed in the zonal models
thermal variation in the oil observed in the differential model and for Transf-01 and Transf-02: nine correspond to the oil in the casing
presented in Fig. 3. In fact, two main criteria have been used in and the fins, and four to the oil inside the cooling channels of the
dividing the domain into zones: the first one is that the oil mass coils. For Transf-03, an additional control volume has to be included
flow moving in and out of each zone has to be well defined, and the to represent the 2nd LV channel.
second one is that the oil temperature has to be uniform enough. Some additional criteria were also considered, taking into
Fig. 4 shows the zones defined for Transf-01, superimposed over the account geometrical restrictions and practical issues. For example,
oil flow pattern (left) and the oil temperature distribution (right) there is a by-pass in the oil flow from zone 06 to 08, or from zone 07
obtained by the differential model. to 09 (zone 05 only represents the boundary layer developing over
All the oil mass flow interchanges between the different control the outer surface of the HV windings), to assure the connection
volumes of the zonal model under consideration can be seen in between the fin and the bottom part of the casing for transformers,
where the bottom part of the fins is in a higher position than the
bottom part of the windings. The height of zones 04 and 07 is fixed
Table 2
at 50 mm, as is the minimum height of zones 02 and 06, guaran-
Comparison of oil temperature results: zonal (z) vs. differential (d).
teeing their existence in all the cases. Another example is that, as
Transf-01 the 1st LV channel (10) exists in all distribution transformers, the
E01 E02 E03 suction from zone 04 to zone 03 is related to the mean velocity in
z ( C) d ( C) Sz-d (%) z ( C) d ( C) Sz-d (%) z ( C) d ( C) Sz-d (%)
the outlet of this channel, as will be shown later.
TAmb 22.9 e 24.2 e 24.8
DT01 30.0 31.6 5.3 40.8 42.8 4.9 50.8 53.0 4.3 3.3. Governing equations
DT02 30.2 31.2 3.3 41.0 42.3 3.2 51.0 52.7 3.3
DT03 31.1 31.0 0.3 42.0 41.6 1.0 52.0 51.8 0.4
The mass and energy conservation equations in their steady
DT04 30.0 30.3 1.0 40.9 41.5 1.5 50.9 52.1 2.4
DT05 28.1 26.3 6.4 38.9 37.0 4.9 49.0 47.1 3.9 formulation, Eqs. (1) and (2), are applied to the control volumes or
DT06 26.7 27.1 1.5 37.2 37.9 1.9 47.1 48.0 1.9 zones into which the flow domain has been divided. Eq. (1) says
DT10 36.2 36.1 0.3 46.8 45.9 1.9 56.5 55.8 1.2 that the sum of the oil mass flow rates entering one control volume
DT11 34.0 31.9 6.2 45.4 43.6 4.0 55.9 53.6 4.1 through its faces that are in contact with other zones equals the
DT12 32.6 33.5 2.8 43.4 43.6 0.5 53.3 52.9 0.8
DT13 32.4 33.7 4.0 43.1 43.4 0.7 52.9 52.6 0.6
sum of the oil mass flow rates leaving it. Eq. (2) establishes that the
sum of the heat in the form of power losses entering one zone
J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035 4029

Fig. 4. Definition of the control volumes of the zonal model of Transf-01 based on the mass flow interchanges (left) and oil temperature distribution (right) in the differential model.

through the internal faces that are in contact with the core and the 3.3.1. Mass conservation
coils, plus the sum of the net oil flow thermal energy rates entering In practice, the application of Eq. (1) to each control volume
or leaving one zone through its boundaries in contact with other implies that the different oil mass flows shown in Fig. 4 (left) are
zones, minus the heat transferred by conduction to other zones, interrelated, describing the previously sketched oil movement. Eq.
minus the heat transferred by convection and radiation through the (3) shows the application of mass conservation to control volume
faces that are in contact with the external surface of the trans- 03 of Transf-01 as an example:
former equals zero.
_ 1003 þ m
m _ 1103 þ m
_ 1203 þ m
_ 1303 þ m
_ 0503
X X
_ In 
m _ Out ¼ 0
m (1) _ _
þ m0403  m0302 ¼ 0 ð3Þ
fFluid fFluid
In zonal models where the fluid movement is prefixed there is
0 1 no need to solve the pressure field to find the eventual direction-
X X X X
QLosses þ @ EIn  EOut A  QCond ality and magnitude of the flow [12,13]. In the present model
fInt fFluid fFluid fFluid several independent oil mass flows (the ones generated in the
0 1 different vertical cooling channels and those characterising both
X X suctions) are defined, based on the different phenomena that cause
@ QConv þ QRad A ¼ 0 ð2Þ
fExt fExt
them to appear. The rest of the unknown oil mass flows are derived
from equilibrium equations similar to (3).
Additionally, both principles also have to be fulfilled in the In the cooling channels and close to the internal hot walls, oil
whole domain: no oil is getting in or out the closed domain shown mass flows have been correlated with the internal heat fluxes using
in Fig. 4, and all the energy that is introduced from the interior faces equations from the literature. From this point of view, the heat
due to thermal losses has to be equal to the energy dissipated from transfer inside the cooling channels can be considered the driving
the external faces by convection with the air and radiation inter- force of the entire natural convection movement inside the trans-
changes with the surroundings. former, as has been verified with the results of the differential
The thermophysical properties of the solid material (the stain- model. For example, in the LV channels, Eq. (4), which is a modifi-
less steel of the external casing) have been considered constant. In cation of the expression for rectangular vertical channels from [15],
the case of the oil, its density, molecular viscosity and the specific is applied. In a similar way, Eq. (5) is applied to the triangular
heat vary with temperature, and the thermal conductivity has been channels of the HV side [15]. These formulas assume a fully
considered constant, as shown in [11]. developed flow inside the channels, as shown in Fig. 5.
4030 J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035

Fig. 5. Velocity vectors (top) and contours (bottom) inside cooling channels at mid-height of the coils obtained from the differential model of Transf-03.

Kaminski et al. [18], is used as shown in Eq. (10). This hypothesis


4 r$b$q00 $dT a postulates that the horizontal component of the velocity in a flow
_ 1003 ¼ n10 $ $b$a3 $
m $f (4)
3 n$l b suctioned by an ascending natural convection plume is propor-
pffiffiffi tional to the vertical component of the velocity of that thermal
c4 $ 3 r$b$q00 $dT plume. The mean velocity of the first LV channel flow has been
_ 1303 ¼ n13 $
m $ (5)
320 n$l selected as a representative value for the ascending plume. This
mean velocity is calculated by simply dividing the ascending oil
where:
mass flow rate in the channel by the oil density and the rectan-
gular cross-sectional area.
dT ¼ H$Ra*H (6)
_ 0403 ¼ ðC$V10 Þ$r$AFluid;
m Vert;0403 (10)
g$b$q00 $H 4
Ra*H ¼ (7)
a$n$l All the oil properties in Eqs. (4)e(10) are evaluated at the
characteristic temperature of the particular control volume that
For the vertical walls shown in Fig. 5, where both boundary is generating the specific oil flow. All the equations shown were
layers do not merge, i.e. on both sides of the LV/HV channel or in the originally developed for specific flow conditions, none corre-
external vertical wall of the HV winding, Eq. (8) has been selected. sponding exactly to those existing inside the transformer and for
This formula was developed from the expression presented by that reason some adjustment coefficients are required. The
Kimura and Bejan in [16] for the vertical velocity profile in differences with the driving oil mass flow rates of the differential
a boundary layer generated by natural convection over a vertical model vary from a smallest value of merely 0.5% for the low
wall with a uniform heat flux inside a closed thermally stratified power test E01, up to a highest value of 12.7% for the high power
domain: test E03. The global adjustment is quite good with a mean value
ZdT   64$a$r of 4.5%. These data indicate that the selected formulas and
*1=5 adjustment coefficients describe the oil mass flow generation
_ 12=0103 ¼ d$
m r$VðxÞ$dx ¼ d$ 1 þ ep $ $RaH (8)
z4 inside the different cooling channels of the coils adequately
0
enough, taking into account that very small magnitudes are being
where: reproduced. Both suctions are also well captured, with
  a minimum difference of 0.3% and a maximum of 16.1%,
z ¼ f Ra*H (9) showing that the adopted approximation is adequate for this
phenomenon as well.
The horizontal flow movements identified as suctions in the
description of the oil flow pattern obtained by the differential 3.3.2. Energy conservation
model are an indirect effect of the driving buoyant forces char- The different terms from Eq. (2) are developed and applied to
acterised by Eqs. (4)e(9). The horizontal flow movements appear control volume 01 of Transf-01 as a representative example. This
to maintain a momentum and mass balance with the oil flowing equation postulates an equilibrium in each control volume for (a)
into and out of the coil channels. In this case, the Taylor the energy that is lost from the internal walls (Eq. (11)), (b) the flow
Hypothesis, as referenced by Sreenivas and Prasad [17] and thermal energy rates going into and out of the control volume
J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035 4031

(Eq. (12)), (c) the energy transmitted to the surrounding control  


X TExt;Lid;01  TAmb $ALid;01
volumes by conduction (Eq. (13)), and (d) the energy that is dissi- QConv ¼
1=hExt;Lid
pated through the external casing, which is a sum of the convective fExt
and radiation heat exchange terms (Eq. (14) and (15) respectively).  
TExt;Cas;01  TAmb $ACas;01
This energy balance is graphically shown in Fig. 6 as an equivalent þ
1=hExt;Cas
thermal circuit.  
In Eq. (11) each constant heat flux on the internal faces of the TExt;Cas* ;01  TAmb $ACas* ;01
domain is multiplied by its respective area and summed, obtaining þ (14)
1=hExt; Cas*
all the rate of energy that is introduced in the control volume from
the transformer’s power losses: X  
4 4
X QRad ¼ εLid $s$ TExt;Lid;01  TAmb $ALid;01
QLosses ¼ q00Core $ACore;01 (11) fExt
fInt  
4 4
þ εCas $s$ TExt;Cas;01  TAmb $ACas;01 (15)
In Eq. (12), a net balance between the flow thermal energy rate
that is going into the control volume from the upwind volumes and As can be seen in Eqs. (14) and (15), the definition of the
the rate that is leaving it through the outgoing fluid faces is convective and radiation heat exchange terms needs the external
obtained: surface temperatures to be introduced as new unknowns in the
equations. To close the system of equations, an extra expression,
0 1
X X like the one shown in (16) for the lid, is required for each external
   
@ EIn  EOut A ¼ m_ 0201 $cp $T02  m_ 0102 $cp $T01 (12) surface temperature of every control volume.
fFluid fFluid  
T01  TExt;Lid;01
QOilAir;i ¼ QConv;i þ QRad;i 0
Eq. (13) is a summation of the small energy fraction that is 1 eLid
D
transferred by conduction to the surrounding control volumes, in hInt;Lid $ALid;01 lLid $ALid;01
the horizontal and vertical direction, formulated as one-
¼ QConv;Lid;01 þ QRad;Lid;01 (16)
dimensional heat transfer following Fourier’s law:

X l$ðT01  T02 Þ$AFluid;Horiz;0102


QCond ¼
DVert;0102 Additionally, internal surface temperatures can be easily calcu-
fFluid
lated after solving the main equation system by using Eqs. (17)
l$ðT01  T02 Þ$AFluid;Vert;0102
þ (13) and (18):
DHoriz;0102
!
QLosses;fInt ;i
Eq. (14) evaluates the energy dissipated from the external TInt;fInt ;i ¼ þ Ti (17)
hInt;fInt ;i $AfInt ;i
surfaces to the surrounding air by natural convection following
Newton’s law of cooling. Finally, Eq. (15) represents the energy !
transmitted by radiation heat exchanges to the surrounding QOilAir;fExt ;i
surfaces (considered black bodies at the ambient temperature) TInt;fExt ;i ¼ Ti  (18)
hInt;fExt ;i $AfExt ;i
following StefaneBoltzmann law.

3.4. Boundary conditions

All the developed equations depend explicitly or implicitly on


the thermal boundary conditions at the internal and external wall
limits of each control volume. The same boundary conditions that
previously proved their adequateness in the differential model are
also used in the present zonal model. A complete analysis of these
thermal boundary conditions can be found in [11], but the most
relevant information is given in the following paragraphs.
Starting from the inside, uniform heat fluxes are imposed on the
internal surfaces of the model, as shown in Fig. 4, representing the
heat generation inside the core and in the zones in which the LV
and HV coils are divided by the cooling channels. These heat fluxes
are calculated by dividing the power generated in each solid
portion by its total surface area and they are used, for example, in
Eqs. (4),(5),(8) to evaluate different oil mass flow rates or to
calculate the heat portion that is generated on the internal surfaces
of each control volume in Eq. (11).
The internal heat transfer coefficients have been obtained from
the differential model, and they are in general mean constant
values, except in places like the lid or the fins, where linear corre-
lations dependent on the total power losses have been considered.
Fig. 6. Equivalent thermal circuit of the application of the energy conservation A sensitivity analysis was carried out with the zonal model, and it
equation to a generic control volume or zone of the model. proved to be quite insensitive to modifications in these coefficients
4032 J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035

because the predominant thermal resistance is introduced by the This widely-used programme offers a user-friendly environment
external air. and includes a lot of programming options through use of the so-
Two different heat transfer phenomena have to be described on called Visual Basic for Applications. The Standard Solver proved to
the outer surfaces of the model: radiation heat exchanges with the be a valuable tool for analysing and optimising systems and
surroundings, and convective heat transfer with the air. processes of medium complexity (expressed by means of smooth
The radiation heat exchange can be easily stated in open zones, nonlinear equation systems) with acceptable efficiency [20,21]. To
such as the lid or the base as was previously shown in Eq. (15). It is use the Solver, all the previous equations are reordered and the
assumed that these surfaces exchange radiation with black body system is expressed as an optimisation problem with several
surroundings at ambient temperature, except for the ground where restrictions. There is a nonlinear objective function representing
an approximated surface temperature was considered. In the the overall energy balance that has to be minimised to zero. The
vertical region surrounded by the fin arrangement, the net value of rest of the equations (as explained, some of them are also
heat exchanged by radiation is relatively small compared to the nonlinear) are expressed as restrictions. There are three resolution
convective one. In fact, only on the more exposed extremes of the types for the zonal model, obtained by changing the objective
fins it is relevant. In this region, the radiation will be introduced by function and some of the changing variables but maintaining the
means of a mixed heat transfer coefficient mentioned in the rest of the mathematical model as restrictions:
following paragraph.
The convective heat transfer coefficient for vertical surfaces (fins 1) The calculation mode: The model is solved for a specific
and casing) is expressed by means of a correlation between a local geometry by means of changing all the oil and surface
Rayleigh number (Ray) and a local Nusselt number (Nuy) that has temperatures (changing variables) to fulfil the objective func-
been numerically obtained. An adjustment coefficient has been tion (overall energy balance equals 0) and obtaining all the
introduced in this correlation to take into account the low radiation output elements previously specified.
heat exchanges and the small influence the horizontal thermal 2) The redesign mode: An objective top oil temperature value is
variations have on that zone in a mixed or averaged way. The prefixed and mass and energy conservation equations are ful-
specific value of this adjustment coefficient, increasing the net filled by means of changing the rest of the oil and surface
convective heat transfer by 22%, is explained and fully justified in temperatures and a geometric parameter of the transformer
[11] for the differential model. (which is redesigned): the number of fins, their vertical posi-
The convective heat transfer coefficients in horizontal places (lid tion or some of their principal dimensions.
and base) are expressed by means of correlations between a mean 3) The optimisation mode: In this case the objective function to be
Rayleigh number (Ra) and a mean Nusselt number (Nu) obtained minimised can be the cost of the transformer, the heat transfer
from the literature [19]. surface area (amount of material) or the volume of oil inside
All the equations corresponding to the previously described the transformer. As in the previous mode, the top oil temper-
boundary conditions can be found in [11]. The values of the different ature is fixed, the mass and energy equations are set as
heat transfer coefficients that are obtained with those equations restrictions and some dimensions of the transformer must be
depend on the value of the surface temperature, which is a result of added to the changing variables.
the simulation. This fact introduces an additional non linearity in
the zonal model through the convective terms, Eq. (14), of each zone After checking different options, an enhanced version of the
energy equation because the external thermal boundary conditions Generalized Reduced Gradient algorithm (GRG2) [22] was consid-
will change during the resolution of the problem. ered to be the best choice among all the resolution engines that are
available in the Solver to deal with this problem. The initial values
4. Implementation of the model of the independent variables are calculated using quadratic
extrapolation, the derivatives are evaluated by central differentia-
The mathematical zonal model that has been described in the tion, and the searching direction is found by means of the conju-
previous section has been implemented in software application to gated gradient method.
check its real performance. The generated nonlinear algebraic
equation system is numerically solved using an efficient resolution 5. Validation of the model and discussion of the results
procedure. The mentioned nonlinearity is a consequence of the
dependence of the oil properties in terms of oil temperature, the The zonal model developed was run for the three transformers
definition of external radiation heat exchanges and the dependency described in Table 1 under different loading conditions, emulating
of the external convection coefficients on surface temperature, as five standardised heating tests (E01-E05). A double validation
explained earlier. strategy was employed to check the flow and thermal results from
The input to the model consists of the ambient temperature, the the proposed simplified model. Firstly, the oil temperatures and the
most relevant geometrical characteristics of the transformer that is external heat distributions obtained by the zonal model for Transf-
going to be simulated (some of the principal values can be seen in 01 under different loading conditions (E01, E02 and E03) were
Table 1), and finally, the power losses that are generated in compared with those obtained by the original differential model.
conjunction with their distribution (PCore, PLV and PHV can be easily This was done to evaluate whether all the assumed modelling
evaluated by means of standardised tests, and they are usually simplifications allowed for an adequate representation of the
known values at rated conditions as shown in Table 1). The output cooling phenomenon under study. Secondly, the top oil tempera-
of the model consists of the top oil temperature rise (this charac- ture and the external surface temperatures obtained with the zonal
teristic value is checked by the manufacturers for homologation model for different transformers (Transf-01, Transf-02 and Transf-
purposes), which is considered equal to the oil temperature rise in 03) were compared with the experimental results, evaluating the
the control volume 01, the rest of the oil temperatures, the values of final validity of the proposed model. A complete description and
the inner and outer surface temperatures, and the external total assessment of the differential model and heating tests, whose
heat flux distribution. results are shown in Table 2, can be found in [11].
The Microsoft ExcelÒ spreadsheet has been selected, with its If the ambient temperature is taken as reference and the oil
built-in SolverÒ, to implement the mathematical model developed. temperature rises over this temperature are evaluated, relative
J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035 4033

Fig. 7. Oil temperature rises over ambient temperature: comparison between the zones of the zonal model and the corresponding points of the differential model for test E01.

differences can be established for comparison in Table 2. Addi- can be found (see Fig. 4), it would be necessary to divide the zonal
tionally a graphical comparison of the oil temperatures predicted model in more control volumes to obtain a more accurate
by the differential model (local values and their vertical variation temperature distribution. Actually, this is not a critical fact because
represented by means of vertical bars) and the zonal model (local these lower cold zones have a very small influence on the overall
value in each zone) is shown in Fig. 7 for test E01 for Transf-01. The cooling process as will be shown later when analysing heat dissi-
differences for the top zones of the model (zones 01e06) are lower pation (see Fig. 8).
than 6.4% in all the cases, showing a great similarity in this key area Taking a global perspective, all these observations indicate that
in the cooling process. In the case of the cooling channels (zones 10- the selected subdivision in control volumes is adequate enough to
13), it can be seen in Fig. 7 that the zonal model is clearly predicting capture all the key oil temperatures that have been observed in the
the temperatures at the exit of the channels. If these upper values differential model.
are taken as reference, the differences are quite low again, with
values between 0.3% and 6.2%. So, although the zones 10e13 are
representing the whole length of the cooling channels of the coils,
they are providing the temperatures of the outlet of the channels.
This can be explained by means of the rate of thermal energy
conservation Eq. (2) applied to one coil cooling channel. As this is
an internal zone, there are no external faces and neglecting the
conduction heat transfer with other zones, Eq. (2) reduces to the
three first terms. Taking into account Eqs. (11) and (12), it can be
said that this equation establishes that the total power losses of the
corresponding coil are used to increase the enthalpy of the oil mass
flow rate flowing through the cooling channel, so it is clear that the
temperature representing this zone must be the outlet tempera-
ture. In this sense this can be considered an advantage of the zonal
model because the maximum channels outlet temperature can be
approximated to the hot-spot temperature reached in the internal
insulation of the coils, which are of high interest for the designers of
distribution transformers. For the same reason, in the top zones of
the model (01e06) where the temperature is more uniform, as can
be seen in Fig. 4, the temperatures given by the zonal model are
closer to the mean temperature in each zone obtained in the
differential model.
The differences increase at the bottom zones of the model
(zones 07e09), as can be seen in Fig. 7, obtaining values between
41.1% and 77.9% (if the central point temperature is taken as refer-
ence). But, again the temperatures of the zonal model are closer to
the maximum temperatures of these zones in the differential
model. The explanation is the same pointed out for the cooling
channels: the temperature provided by the zonal model for each Fig. 8. Comparison of the external heat dissipation between the differential and the
zone is the maximum temperature of that zone. So in the lower cold zonal model: convection versus radiation heat exchanges (top) and distribution into
zones of the transformer, where a high vertical thermal variation the different parts of the transformer (bottom).
4034 J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035

Table 3
Comparison of oil temperature results: zonal (z) vs. experimental (e).

Transf-01 Transf-02 Transf-03

E01 E02 E03 E04 E05


       
z ( C) e ( C) Sz- e(%) z ( C) e ( C) Sz-e (%) z ( C) e ( C) Sz-e (%) z ( C) e ( C) Sz-e (%) z ( C) e ( C) Sz-e (%)
TAmb 22.9 24.2 24.8 22.0 14.6
DTOil 29.6 31.5 6.2 40.3 41.8 3.6 50.3 51.4 2.2 47.8 46.6 2.5 54.9 56.2 2.2

The final validation of the zonal model is done using experi- practically complete. The experimental adjustment in the upper hot
mental temperature measurements. As our main interest is focused zone is again very good, decreasing in the lower cold zone, by the
on the accurate prediction of the top oil temperature, Table 3 shows reason explained in a previous paragraph.
the relative differences that were found between the predicted and In order to show the potential and utility for design purposes of
the measured values in the top oil temperature for all the analysed the proposed zonal model a brief parametric study is given taking
transformers and rating conditions. It can be seen that the as reference Transf-01 under nominal test conditions, test E03. Six
maximum difference is always lower than 6.2%, showing high different parameters have been changed independently in order
agreement in all cases. not to introduce cross influences: total power losses, power losses
In analysing the external heat dissipation, it can be concluded of the core, position of the core, number of fins, thickness of the fins
that in both models the convection phenomenon is always greater and depth of the fins. All the parameters have been modified in
in comparison to the radiation heat exchanges with the a realistic range and the result of the model chosen to compare the
surroundings at an approximate ratio of 9 to 1, as shown in Fig. 8 variations is the top oil temperature.
(top). The heat distribution predicted by both models for each The parameters less influencing the cooling of the transformer
zone is also similar, as seen in Fig. 8 (bottom). Approximately 83% of are the core power losses, the position of the core respect to the
the total heat is dissipated at the surfaces of the fins, 8% by the base of the transformer and the thickness of the fins. For the first
casing, 7% by the lid and 2% by the base. This indicates the impor- one, a relative reduction of a 25% produces a relative reduction in
tance of the hollow fins in cooling these distribution transformers the top oil temperature of only a 3.4%. Reducing or increasing the
due to the concentrated dissipation area in that zone, as pointed out distance from the bottom of the core to the base of the trans-
in [23], and illustrates the relatively small influence of the lower former in a 100% (from 20 mm to 0 and 40 mm), only produces
cold zones, as has been justified previously. a 0.4% of drop or rise, respectively, in the top oil temperature. And
In checking the external surface temperatures over the fins as finally, the top oil temperature only varies in a 0.6% when the
shown in Fig. 9, a triple assessment (experimental vs. differential vs. thickness of the fins changes in a 25% (from 8 mm to 6 and
zonal) can be made, representing the evolution of this modelling 10 mm).
work at different stages. Experimental measurements show that On the other hand, the main parameters influencing the cooling
the vertical variation is always much higher than the horizontal of the transformer are the total power losses, the number of fins
one: there are different points for the same y coordinate, almost and the depth of the fins. Using the zonal model it has been
perfectly overlapped. As previously explained, in the zonal model obtained that, for the same design and dimensions of the trans-
these horizontal temperature variations have been completely former, the variation of the top oil temperature is not linear with
discarded, focusing attention on the vertical thermal variations that the total power losses: 17.1% and 40.6% of temperature drop for
show quite adequate adjustment both with experimental and power reduction of 25% and 50%, respectively.
differential model results. The similarity between the surface As it has been pointed out in the previous paragraphs relative to
temperature prediction for the differential and zonal model is the validation of the model, the fin arrangement is the key zone in

Fig. 9. External vertical surface temperature variations over a fin for three tests: experimental (points) vs. differential (continuous lines) vs. zonal (discontinuous lines).
J. Gastelurrutia et al. / Applied Thermal Engineering 31 (2011) 4024e4035 4035

the cooling of a distribution transformer because its great dissi- technicians José Antonio Parada and Iker Barragán from O.C.T. are
pation area. In this sense, the zonal model predicts an almost linear also acknowledged.
variation of the top oil temperature with the number of fins (6.8%
for þ25%; 5% for þ17%; þ3% for 11% and þ5% for 17%) and with References
the depth of the fins (10.5% for þ25% and 18.3% for þ50%).
[1] IEC 60076-1, Power transformers Part 1: general, IEC Standard, Ed. 2.1 b, 2000.
6. Conclusions [2] IEC 60076-2, Power transformers Part 2: Temperature Rise, second ed., IEC
Standard, Ed. 2.0 b Cor.1, 1997.
[3] IEC 60076-7, Power Transformers Part 7: Loading Guide for Oil-Immersed
An algebraic zonal model representing the cooling of ONAN power transformers, IEC Standard, Ed. 1.0 b, 2005.
distribution transformers under different loading conditions was [4] J. Aubin, Y. Langhame, Effect of oil viscosity on transformer loading capability
at low ambient temperatures, IEEE Transactions on Power Delivery 7 (2)
developed. This model was implemented as a thermal simulation (1992) 516e522.
tool that allows the manufacturer to predict the top oil temperature [5] Z. Radakovi c, D.J. Kali
c, Results of a Novel algorithm for the calculation of the
obtained in a standardised heating test for a given transformer, to characteristic temperatures in power oil transformers, Electrical Engineering
80 (3) (1997) 205e214.
redesign one geometric parameter of the transformer or to opti-
[6] G. Swift, T.S. Molinski, W. Lehn, A Fundamental approach to transformer
mise the size or the cost of the transformer in order to obtain thermal modeling e Part I: theory and equivalent circuit e Part II: field
a desired value for the top oil temperature. All the geometric, verification, circuit, IEEE Transactions on Power Delivery 16 (2) (2001)
171e180.
loading and ambient conditions can be modified by the user in
[7] J.-M. Mufuta, E. Van den Bulck, Modelling of the mixed convection in the
order to obtain the top oil temperature, in order to be able to windings of a disc-type power transformer, Applied Thermal Engineering 20
improve the final thermal design. (2000) 417e437.
The development and adjustment of the simplified model [8] N. El Wakil, N.C. Chereches, J. Padet, Numerical study of heat transfer and fluid
flow in a power transformer, International Journal of Thermal Sciences 45
proposed here was done based on the previous results of a more (2006) 615e626.
detailed differential model [11]. Five different heating tests were [9] J. Smolka, D.B. Ingham, L. Elliott, A.J. Nowak, Enhanced numerical model of
simulated for three distribution transformers under several loading performance of an encapsulated three-phase transformer in laboratory envi-
ronment, Applied Thermal Engineering 27 (2007) 156e166.
conditions using this zonal model, showing the capacity of the [10] J. Smolka, A.J. Nowak, Experimental validation of the coupled fluid flow, heat
model to predict adequate thermal results for different geometries transfer and electromagnetic numerical model of the medium-power Dry-
and operating conditions. type electrical transformer, International Journal of Thermal Sciences 47
(2008) 1393e1410.
The results of this model are compared with those obtained [11] J. Gastelurrutia, J.C. Ramos, G.S. Larraona, A. Rivas, J. Izagirre, L. del Rio,
from the differential model and from experimental tests in order to Numerical modelling of natural convection of oil inside distribution trans-
validate it. A high degree of similarity between the performance of formers, Applied Thermal Engineering 31 (2011) 493e505.
[12] J.C. Ramos, A. Rivas, J. Biera, G. Sacramento, J.A. Sala, Development of
both numerical models was confirmed, guaranteeing that the
a thermal model for automotive twin-tube shock absorbers, Applied Thermal
entire cooling phenomenon was correctly captured after the Engineering 25 (2005) 1836e1853.
adopted simplifications. The relative differences between models in [13] J. Stewart, Z. Ren, COwZ e A subzonal indoor airflow, temperature and
contaminant dispersion model, Building and Environment 41 (2006)
the top oil temperatures were lower than 6.4%. The key features of
1631e1648.
the experimentally observed external surface temperature distri- [14] R. De C. Oliveski, M.H. Macagnan, J.B. Copetti, Entropy generation and natural
bution, especially the vertical variation, were quite adequately convection in rectangular cavities, Applied Thermal Engineering 29 (2009)
predicted by the zonal mode, and the differences in experimental 1417e1425.
[15] F.M. White, Viscous Fluid Flow 119e125, second ed. McGraw-Hill, 1991.
values in the predicted top oil temperature were lower than 6.2% [16] S. Kimura, A. Bejan, The boundary layer natural convection regime in a rect-
for all cases. angular cavity with uniform heat flux from the side, Journal of Heat Transfer
The simplified zonal model offers quick thermal results with 106 (1984) 98e103.
[17] K.R. Sreenivas, A.K. Prasad, Vortex-dynamics model for entrainment in jets
a higher computational efficiency than the previous differential and plumes, Physics of Fluids 12 (2000) 2101e2107.
model, discarding no essential cooling details and making the [18] E. Kaminski, S. Tait, G. Carazzo, Turbulent entrainment in jets with arbitrary
development stage of a new distribution transformer shorter. Using buoyancy, Journal of Fluid Mechanics 526 (2005) 361e376.
[19] F.P. Incropera, D.P. De Witt, T.L. Bergman, A.S. Lavine, Fundamentals of Heat
this simplified model, a complete optimisation study, taking into and Mass Transfer, sixth ed. John Wiley & Sons, New York, 2006, 559e619.
account economic issues, could be proposed. The zonal model was [20] A. Rivas, T. Gómez-Acebo, J.C. Ramos, The application of spreadsheets to the
implemented with a user-friendly software design that does not analysis and optimization of systems and processes in the teaching of
hydraulic and thermal engineering, Computer Applications in Engineering
require specialized skills or knowledge on the part of the final user, Education 14 (4) (2006) 256e268.
as is the case with the more complex differential models. [21] A. Rivas, I. Irizar, E. Ayesa, Model-based optimisation of wastewater treat-
ment plants design, Environmental Modeling & Software 23 (4) (2008)
435e450.
Acknowledgements
[22] L.S. Lasdon, A.D. Waren, A. Jain, M. Ratner, Design and testing of a generalized
reduced gradient code for nonlinear programming, ACM Transactions on
The authors of this research wish to acknowledge Ormazabal Mathematical Software (TOMS) 4 (1) (1978) 34e50.
Corporate Technology, O.C.T., and the Antonio Aranzabal Founda- [23] J.C. Ramos, A. Rivas, J.M. Morcillo 929e934, Numerical Thermal Modelling of
the Natural Ventilation of a Half-Buried Transformer Substation using CFD
tion e University of Navarra for their economic support and Techniques, Progress in Computational Heat and Mass Transfer, vol. 2, Lav-
collaboration. The experimental measurements carried out by oisier, Paris, 2005.

You might also like