www.ietdl.

org

Published in IET Electric Power Applications

Received on 8th March 2008
Revised on 15th May 2008
doi: 10.1049/iet-epa:20080055

ISSN 1751-8660

Heat transfer in power transformer windings

with oil-forced cooling
M.A. Taghikhani A. Gholami
Electrical Engineering Department, Iran University of Science and Technology (IUST), Narmak 16846, Tehran, Iran
E-mail: taghikhani@ee.iust.ac.ir

Abstract: Power transformer outages have a considerable economic impact on the operation of an electrical
network. One of the most important parameters governing a transformer’s life expectancy is the hot-spot
temperature (HST) value. The classical approach has been to consider the HST as the sum of the ambient
temperature, the top-oil temperature rise and the hot-spot-to-top-oil temperature gradient. Temperature
distribution is solved using the heat conduction equation. Finite element method is used in the numerical
solution. The transformer selected for simulation is 32 MVA transformer with non-directed oil-forced cooling
and directed oil-forced cooling. Comparing the results with those obtained from finite integral transform and
experimental test checks the validity and accuracy of the proposed method.

Nomenclature k2 thermal conductivity at outer cylinder surface

a inner radius of the annular disc or layer (m) (W m21 K21)

b outer radius of the annular disc or layer (m) k3 thermal conductivity at bottom cylinder surface
(W m21 K21)
Cp specific heat of transformer oil ( J kg21 K21)
k4 thermal conductivity at top cylinder surface
Dm mean diameter of annular disc or layer of (W m21 K21)
windings (m)
kcu thermal conductivity of copper (W m21 K21)
f1 boundary function at inner cylinder surface
kin thermal conductivity of insulation (W m21 K21)
f2 boundary function at outer cylinder surface
koil thermal conductivity of transformer oil (W m21 K21)
f3 boundary function at bottom cylinder surface
kr thermal conductivity in radial direction
f4 boundary function at top cylinder surface (kr ¼ k1 ¼ k2)
fr relative index of free to forced convection kz thermal conductivity in axial direction
g acceleration due to gravity (m s22) (kz ¼ k3 ¼ k4)
Gr Grashof number based on temperature difference l thickness of the disc or height of layer winding (m)
Grhf Grashof number based on constant heat flux m oil temperature gradient along winding height
Gz Graetz number (K m21)
h1 heat transfer coefficient at inner surface (W m22 K21) n temperature rise exponent to bottom oil
h2 22
heat transfer coefficient at outer surface (W m K ) 21 Nu local Nusselt number
h3 heat transfer coefficient at bottom surface (W m22 K21) Num mean Nusselt number
h4 heat transfer coefficient at top surface (W m22 K21) Pr Prandtl number
K referred to as equivalent thermal conductivity Q volumetric heat source function (W m23)
k1 thermal conductivity at inner cylinder surface Q0 volumetric heat source at ambient temperature
(W m21 K21) (W m23)

IET Electr. Power Appl., 2009, Vol. 3, No. 1, pp. 59 – 66 59

doi: 10.1049/iet-epa:20080055 & The Institution of Engineering and Technology 2009
www.ietdl.org

qw heat flux (W m22) another system of voltage and current by electromagnetic

R loss ratio ¼ load loss/no-load loss induction. Large power transformers belong to the most
valuable assets in electrical power networks and an outage
Ra Rayleigh number based on temperature difference
affects the stability of the network. In a power transformer
Rahf Rayleigh number based on constant heat flux operation, a part of the electrical energy is converted into
Re Reynolds number heat. Although this part is quite small compared with total
ri’s i ¼1, 2, . . . radius of insulation and conductor layers electric power transferred through a transformer, it causes
(m) significant temperature rise of transformer constructive
parts, which represents the limiting criteria for possible
Tamb ambient temperature (K)
power transfer through a transformer. That is why the
Tb temperature at the bottom of the winding (K) precise calculation of temperatures in critical points (top oil
tcu thickness of conductor (m) and the hottest solid insulation spot) is of practical interest.
tin thickness of insulation (m) Thermal impact leads not only to long-term oil/paper-
insulation degradation, but is also a limiting factor for the
Tm temperature average of oil and winding surface (K)
transformer operation [1].
Ttop temperature at the top of the winding (K)
The basic criterion for transformer loading is the
temperature of the hottest spot of the solid insulation (hot-
Abbreviations spot). It must not exceed the prescribed value in order to
DOF directed oil-forced cooling avoid insulation failures. The hot-spot temperature (HST)
HTC heat transfer coefficient depends on the load loss (i.e. on the current) diagrams and
on the temperature of the external cooling medium. A
HV high voltage
HST calculation procedure is given in the International
LV low voltage Standards [2 –4]. An algorithm has been given for
NDOF non-directed oil-forced cooling calculating the HST of a directly loaded transformer in [5,
p.u. per unit 6]. They propose improvements in the modelling of
thermal processes inside the transformer tank. Some
attempts are exposed for the heat transfer calculations from
Greek symbols winding to oil and oil to air in [7]. The usage of average
heat transfer coefficient (HTC) is typical in a transformer
b coefficient of volumetric expansion of oil (K21)
designing process to calculate needed number (area) of
uu ultimate bottom-oil temperature rise over ambient cooling surfaces [8 – 13].
temperature (K)
ufl ultimate bottom-oil temperature rise over ambient In this paper, a procedure is proposed for obtaining the
temperature at rated load (K) temperature distribution in the transformer windings. The
d characteristic dimension (m) procedure requires calculation of the HTC in the energy
(thermal) equation. The authors use finite element method
r density of transformer oil (kg m23)
in the numerical solution. For this purpose, they have
f( j) correction factor to cylindrical curvature developed a code using MATLAB version 6.5. In this
ac temperature coefficient of electrical resistance (K21) paper, Section 2 deals with mathematical formulation for
m viscosity of oil (kg m21 s21) heat conduction equation, whereas its solution using finite
element method for steady state is provided in Section
mb viscosity of oil at oil bulk mean temperature
3. Discussion of the proposed work has been provided in
(kg m21 s21)
Section 4.
mw viscosity of oil at winding wall temperature
(kg m21 s21)
2 Mathematical formulation
Subscripts The geometry of a transformer winding is complex and is not
typical geometry use in the analytical solutions of partial
b bulk property differential equations (PDEs). Under general conditions,
m mean value the authors assume that the transformer winding is
w wall value cylindrical; hence, a layer or a disc winding is a finite
annular cylinder [10]. The thermal and physical properties
of the system would be equivalent to a composite system of
1 Introduction insulation and conductor. It has been assumed that heat is
generated throughout the body at a constant rate, and oil in
The power transformer with two or more windings the vertical and horizontal ducts take away the heat
transforms a system of alternating voltage and current into through the process of convection. However, in an actual

60 IET Electr. Power Appl., 2009, Vol. 3, No. 1, pp. 59– 66

& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-epa:20080055
 www.ietdl.org

transformer winding, the conductor is the only heat source. law of cooling, are of the forms (7) and (8)
Later in this section, formulations are given for calculating
different thermal and physical properties of the system. It f1 (y) ¼ h1  (Tb þ m1 :y) (7)
has been assumed that temperature is independent of space
variable, because winding structure is symmetrical (i.e. f2 (y) ¼ h2  (Tb þ m2 :y) (8)
presence of spacers has been ignored, thus reducing three-
dimensional problem to a two-dimensional). Dielectric loss
The term Tb is the temperature at the bottom of the disc or
in insulation is assumed small compared with copper losses
layer, as applicable. Terms m1 and m2 are the temperature
in the conductor. We assume that surface of disc or layer
gradients along the winding height (for layer) or along disc
is flat.
thickness for a disc. Similarly, functions f3 and f4 represent
temperature profiles across bottom and top surfaces, having
The generalised system of non-homogeneous heat
the same form as shown in (7) and (8), where temperature
conduction equation with non-homogeneous boundary
gradient term has been taken as zero.
condition in Cartesian coordinate system is written thus
[10, 14]
f3 (x) ¼ h3  Tb (9)
!
@2 T @2 T
k þ 2 þQ¼0 (1) f4 (x) ¼ h4  Ttop (10)
@x2 @y

Transformer winding has unequal thermal conductivities in

in the region a , x , b and 0 , y , l. different directions. However, in an actual case, the thermal
conductivities in radial directions (k1 and k2) are equal and
At the inner cylindrical surface conductivities in axial direction will be the same (k3 and
k4). Thermal conductivity has been treated as a vector
@T quantity, having components in both radial and axial
k1 þ h1 T ¼ f1 (y) (2)
@x directions. Resultant thermal conductivity of the system can
be estimated as [10]
At the outer cylindrical surface
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
K ¼ k2r þ k2z (11)
@T
k2 þ h2 T ¼ f2 (y) (3)
@x
where
At the bottom flat surface
log (rn =r1 )
kr ¼
@T ( log (r2 =r1 )=k1 þ log (r3 =r2 )=k2 þ    þ log (rn =rn1 )=kn )
k3 þ h3 T ¼ f3 (x) (4)
@y
kcu kin (tcu þ tin )
kz ¼
At the top flat surface (t in kcu þ tcu kin )

@T The term K represents resultant thermal conductivity of

k4 þ h4 T ¼ f4 (x) (5)
@y
Equations (1)– (5) represent the general heat conduction
equation with convection at all four boundary surfaces. In
the above equations, temperature T is a function of space
variables x and y. The term Q is the heat source function
and has been modified here to take care of the variation of
resistivity of copper with temperature. The heat source term
Q can be of the form
Q ¼ Q0  [1 þ ac  (T  Tamb )] (6)
ac is the temperature coefficient of electrical resistance of
copper wire. With this representation, the function Q
becomes temperature-dependent, distributed heat source.
Boundary functions f1( y) and f2( y), derived from Newton’s
insulation and conductor system. HTC h1 – h4 are different
across all the four surfaces. To determine boundary
functions f1 – f4 , it is necessary to calculate HTC across the
four surfaces. Difficulty has been encountered in the
calculation of HTC. It is reported elsewhere [10] that it
depends on as many as 13 factors (e.g. winding size, type,
duct dimensions, oil velocity, type of oil circulation, heat
flux distribution, oil thermal properties etc.). In this work,
corrections have been given for temperature dependence of
the thermal and physical properties of oil, such as viscosity,
specific heat, volumetric expansion and thermal
conductivity. It was found that there is negligible effect of
specific heat, coefficient of volumetric expansion and
conductivity in the working temperature range (110–
1608C). We use some of the heat transfer relations in
natural cooling (ON) mode. These equations have been
used to calculate the HTC [14]. Local Nusselt and

IET Electr. Power Appl., 2009, Vol. 3, No. 1, pp. 59 – 66 61

doi: 10.1049/iet-epa:20080055 & The Institution of Engineering and Technology 2009
www.ietdl.org

Rayleigh number for laminar flow over vertical plates are thus
 4=9
Nu ¼ 0:6Ra0:2
hf (12) @T 2 qw l 1=9
¼ 4:25  10 Rahf (23)
@y koil pDm
Rahf ¼ Grhf Pr (13)
where l is the winding height, Dm is the mean diameter of
where Rahf and Grhf are the local Rayleigh and Grashof
annular disc or layer of windings. Determination of
numbers based on heat flux (qw) at characteristic dimension boundary conditions in forced convection [oil-forced (OF)
(d). Pr is the Prandtl number of transformer oil. Expression
mode] too requires calculating HTC. The expression when
of Rayleigh number based on constant heat flux is
the oil velocity is of lower values, the mean Nusselt number
based on temperature difference is of the form of (24),
g bCp r2 qw d4 corresponding mean Nusselt number based on constant
Rahf ¼ (14)
k2oil m heat flux is of the form of (25)
 
Mean Nusselt number in this case can be computed as mb 0:14
1=3 4=3 1=3
Num ¼ 1:75[Gz þ 0:012(Gz Gr ) ]  (24)
mw
Num ¼ 1:5[Nu]d¼l (15)
 0:125
0:9 1=3 1:16 1=3 mb
However, correction to formula (12) has to be given for Num ¼ 1:63[1:8Gz þ 0:02(Gz Grhf ) ] 
mw
cylindrical curvature. The correction factor in this case is of
the following form (30 , Pr , 50): (25)
D
f (j)  1 þ 0:12j (16) Gz ¼ RePr (26)
l
where Gz is called Graetz number. The terms mb and mw are
pffiffiffi viscosities of oil computed at oil bulk mean temperature
2 2 d and at winding wall temperature, respectively. Re is the
j¼ 0:25
 (17)
Gr r Reynolds number and Pr is the Prandtl number. The
relative importance of natural and forced cooling is
Here Gr is Grashof number based on temperature difference. indicated by the factor fr ¼ Gr/Re2. If fr 1 then both the
Local HTC can be computed as cooling modes have to be considered. At a lower value of
this factor, natural cooling can be ignored. The oil
Nu  koil viscosity, which depends on temperature, is given in (27)
h¼ (18)
d from [9, 10]

Mean coefficient (hm) can be calculated from mean Nusselt  

g
number, as in (15). After knowing hm for a particular m ¼ a  exp (27)
Tm
surface, temperature difference between the winding surface
and oil can be found, dividing the hm by the heat flux where
through the surface. Mean Nusselt number of top surface
of annular cylindrical winding for laminar and turbulent a ¼ 0:0000013573(kg m1 s1 )
regime will normally be of the form
g ¼ 2797:3(K)
Num ¼ 0:54Ra0:25 ¼ 0:61Ra0:2
hf (19)
The viscosity was calculated at the mean oil and wall
Num ¼ 0:15Ra1=3 ¼ 0:24Ra1=4
hf (20) temperature. Initially, the winding surface temperature is
not known, so a starting guess for the winding surface
ba temperature has to be made, after calculating the value of h,
d¼ (21)
2 the temperature difference (Tw 2 Toil ) is to agree with the
assumed value. To calculate the winding wall temperature
where a and b are the inner and outer radius of annular disc or at different surfaces, only the bottom-oil temperature is
layer. Nusselt number of bottom surface of annular cylindrical necessary. In this paper, the bottom-oil rise over ambient
winding for laminar and turbulent regime is in (22) temperature has been calculated as

Num ¼ 0:27Ra0:25 ¼ 0:35Ra0:2

hf (22)
!n
Ir2 R þ 1
The axial oil temperature gradient in the presence of cooling uu ¼ uf l  (28)
Rþ1
by a constant heat flux can be found by using (23), from [14],

62 IET Electr. Power Appl., 2009, Vol. 3, No. 1, pp. 59– 66

& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-epa:20080055
 www.ietdl.org

where uu is the bottom-oil temperature rise over ambient

temperature, ufl is the full-load bottom-oil temperature rise
over ambient temperature obtained from an off-line test, R
is the ratio of load loss at rated load to no-load loss. The
variable Ir is the ratio of the specified load to rated load:

I
Ir ¼ (29)
Irated

The exponent n depends upon the cooling state. The loading

guide recommends the use of n ¼ 0.8 for natural convection
and n ¼ 0.9 – 1.0 for forced cooling.

Assumptions have been made in the calculation of HTC in

the ducts provided in the disc-type windings under OF
modes of heat transfer [directed oil-forced (DOF) and Figure 1 Temperature distribution of LV layer 1 winding at
non-directed oil-forced (NDOF)]. The cooling in OF Tamb ¼ 25(8C) with Re ¼ 750
mode is due to mixed mode (natural and forced)
convection. While the oil-flow velocity in both vertical and
horizontal ducts has been assumed equal, we are in the
DOF mode. If the flow velocity in horizontal ducts is
assumed negligible compared with velocity in the vertical
ducts, we are in NDOF mode. In the case of DOF mode,
the HTC is assumed as a function of both heat flux
through the surface and the oil-flow velocity. The
mechanism of heat transfer in this mode of cooling is same
in both axial and radial directions of the disc. In the case of
NDOF mode, the HTC in the vertical duct has been
estimated by the same formula as for DOF. However, the
convection of heat in the horizontal ducts has been
assumed a purely natural type.

3 Solution of heat conduction Figure 2 Temperature distribution of LV layer 2 winding at

equation
We have developed a code using MATLAB version 6.5 for
finite element solution of heat conduction equation. The
model has been validated on transformer of rating
32 MVA. The Cu-loss (at Tamb ¼ 258C) in the winding
per disc/layer of the above transformer has been tabulated
in Table 1. The result of temperature distribution in x,y
plane from low-voltage (LV) layer 1 winding in 1 p.u. load
with Re ¼ 750 has been shown in Fig. 1. It can be pointed
out that for LV layer 1 winding, maximum temperature
location is around 90– 95% of winding height from the
bottom and at about 50% of radial thickness of the layer.
The result of temperature distribution from LV layer 2
winding has been shown in Fig. 2. It can be pointed out
Tamb ¼ 25(8C) with Re ¼ 750
that for LV layer 2 winding, maximum temperature
location is the same as LV layer 1 winding. Temperature
distribution from a disc of high-voltage HV winding has
been shown in Figs. 3 and 4. It may be observed that the
maximum temperature occurs in the neighbourhood of 30–
35% of the axial and 50% of the radial thickness of the disc
in NDOF mode and 50– 55% of the axial and 50% of the
radial thickness of the disc in DOF mode. Magnitudes of
HST at different loading and OF modes are given in
Table 2. In this work, different rates of forced oil
circulation are taken into account. Figs. 5 – 8 show
variations of local HST with load and Reynolds number in
different windings and different OF modes. Tables 2 and 3

Table 1 Characteristics of transformer windings

Transformer rating, MVA Winding type Cu-loss, W Dimensions, mm a, b, l Remarks

32 LV layer 1 7579 263,287.5,1460 Per layer
32 LV layer 2 8536 301.5,326,1460 Per layer
32 HV disc 570 371,438,21 Per disc

IET Electr. Power Appl., 2009, Vol. 3, No. 1, pp. 59 – 66 63

doi: 10.1049/iet-epa:20080055 & The Institution of Engineering and Technology 2009
www.ietdl.org

Figure 3 Temperature distribution of HV disc winding in Figure 5 HST (local) distribution of LV layer 1 winding at
NDOF mode at Tamb ¼ 25(8C) with Re ¼ 750 Tamb ¼ 25(8C)

Figure 4 Temperature distribution of HV disc winding in Figure 6 HST (local) distribution of LV layer 2 winding at
DOF mode at Tamb ¼ 25(8C) with Re ¼ 750 Tamb ¼ 25(8C)

show comparison of the proposed method with boundary

value method using finite integral transform used in [10]
and experimental test in [9].
It is clear from Table 2 that the HST in the case of NDOF
mode is greater than that of the corresponding DOF mode,
so is the average temperature rise. From Figs. 5, 6 and 8 it
is also borne out that by an increase in the amount of
cooling by the way of increased oil-flow speed would
provide a reduced winding temperature, which is obvious.
However, an increase oil flow has a lesser effect on HST in

Table 2 HST (8C) magnitudes at Tamb ¼ 25(8C) with Re ¼ 750

Load, p.u. HST, 8C – LV layer 1 HST, 8C – LV layer 2 HST, 8C – HV disc Global HST (proposed) HST – Ref. [10]
NDOF DOF
0.8 95 94 87 67 95 92
0.9 108 107 99 74 108 104
1.0 123 122 112 82 123 117
1.1 140 138 126 91 140 134
1.2 158 156 141 101 158 152
1.3 177 175 157 112 177 183
1.4 199 197 174 124 199 200
1.5 222 220 192 137 222 229

64 IET Electr. Power Appl., 2009, Vol. 3, No. 1, pp. 59– 66

& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-epa:20080055
 www.ietdl.org

Figure 7 HST (local) distribution of HV disc winding in
NDOF mode at Tamb ¼ 25(8C)
4 Conclusions
In this paper, an attempt has been made to suggest a method
to improve the accuracy of prediction of the temperature of
the hottest spot in power transformer by solving the heat
transfer partial differential equation (PDE) numerically.
The purely numerical approach for evaluating hot spot and
its location followed in this paper seems to correspond
reasonably well with the results of calculations and actual
tests and on site measurements [9, 10]. The authors wish
to point out that the IEEE loading guide and other similar
documents offer relations for the calculation of the HST
based on p.u. load. The formulations tend to ignore the
possibilities of two transformers that are rating identical but
have a different winding structure and varying heat loss/
unit volume. The method suggested by the authors gives
due representation for this omission and, hence, is believed
to give more accurate estimates. The thermal model
presented here can predict the hot-spot location, with a
reasonable degree of accuracy. The authors are currently
working on a 3D model of transformer for the estimation
of temperature at different points of transformer knowing
the load conditions.

5 References
[1] PIERCE L.W.: ‘Predicting liquid filled transformer loading
capability’, IEEE Trans. Ind. Appl., 1994, 30, (1),
pp. 170– 178

[2] IEC354: ‘Loading guide for oil immersed transformers’,

1991
Figure 8 HST (local) distribution of HV disc winding in DOF [3] IEEE C57.91: ‘Loading guide for mineral oil immersed
mode at Tamb ¼ 25(8C) transformer’, 1995, pp. 18– 19, 46– 53

case of NDOF mode than the DOF mode (Figs. 7 and 8). [4] IEEE Std.1538-2000: ‘IEEE guide for determination of
From the results of this investigation, it is clear that DOF maximum winding temperature rise in liquid-filled
cooling is more effective than NDOF cooling mode, as transformers’, 2000
expected.
[5] RADAKOVIC Z.: ‘Numerical determination of characteristic
Table 3 HST (8C) locations (in LV winding) at Tamb ¼ 25(8C) temperatures in directly loaded power oil transformer’, Eur.
Trans. Electr. Power, 2003, 13, pp. 47– 54
Load, Proposed (radial Experimental [9] (radial
p.u. thickness%/height%) thickness%/height%) [6] RADAKOVIC Z., KALIC DJ.: ‘Results of a novel algorithm for
0.8 50/95 44/94 the calculation of the characteristic temperatures in
power oil transformers’, Electr. Eng., 1997, 80, (3),
0.9 50/94.8 45/95 pp. 205– 214
1.0 50/94.6 45/94
[7] SWIFT G. , MOLINSKI T., LEHN W., BRAY R. : ‘A fundamental
1.1 50/94.5 45/93 approach to transformer thermal modeling—Part I:
1.2 50/94.5 45/93 Theory and equivalent circuit’, IEEE Trans. Power Deliv.,
2001, 16, (2), pp. 171– 175
1.3 50/94.3 49/91
1.4 50/94.2 49/91 [8] TANG W.H., WU O.H., RICHARDSON Z.J.: ‘Equivalent heat circuit
based power transformer thermal model’, IEE Proc. Electr.
1.5 50/94 52/90 Power Appl., 2002, 149, (2), pp. 87– 92

IET Electr. Power Appl., 2009, Vol. 3, No. 1, pp. 59 – 66 65

doi: 10.1049/iet-epa:20080055 & The Institution of Engineering and Technology 2009
www.ietdl.org
[9] PIERCE L.W. : ‘An investigation of the thermal
performance of an oil filled transformer winding’, IEEE
Trans. Power Deliv., 1992, 7, (3), pp. 1347 – 1358
[10] PRADHAN M.K., RAMU T.S.: ‘Prediction of hottest spot
temperature (HST) in power and station transformers’,
IEEE Trans. Power Deliv., 2003, 18, (4), pp. 1275 – 1283
[11] RYDER S.A.: ‘A simple method for calculating winding
temperature gradient in power transformers’, IEEE Trans.
Power Deliv., 2002, 17, (4), pp. 977 – 982
66
& The Institution of Engineering and Technology 2009
[12] SWIFT G., ZHANG Z.: ‘A different approach to transformer
thermal modeling’. IEEE Transm. Distrib. Conf., New
Orleans, April 1999, pp. 12– 16
[13] RADAKOVIC Z., FESER K.: ‘A new method for the calculation
of the hot-spot temperature in power transformers with
ONAN cooling’, IEEE Trans. Power Deliv., 2003, 18, (4),
pp. 1 – 9
[14] INCROPERA F.P., DEWITT D.P.: ‘Fundamentals of heat and
mass transfer’ (J. Wiley & Sons, 1996, 4th edn.)
IET Electr. Power Appl., 2009, Vol. 3, No. 1, pp. 59– 66
doi: 10.1049/iet-epa:20080055