A THERMAL MODEL FOR OPTIMIZED

DISTRIBUTION AND SMALL POWER TRANSFORMER DESIGN

Linden W. Pierce, Member IEEE Tom Holifield
Consulting Engineer Howard Industries
P.O. Box 2367 P.O. Box 1588
Rome, GA 30164 Laurel, MS 39441

Abstract: A thermal model for the design of rectangular layer has been to add 15 'C to the average winding rise to obtain the
windings for transformers is summarized. The model calculates hottest spot rise. For small overhead distribution transformers this
thermal gradients across layers, to and from the core, and from the results in a pessimistic estimation of the loading capability because
no-duct region under the core iron to the ducted region. The the hottest spot rise on many units is below 80 'C. An exact
model evaluates the effects of cooling duct quantity, location, and estimation of the hottest spot rise would improve the accuracy of
size on the thermal performance. The model includes temperature the user's loading predictions. Three phase distribution
dependent fluid properties and electrical resistivity. Fluids include transformers for industrial installations may extend to 5000 kVA .
mineral oil, high temperature hydrocarbons, and silicone. The The capability with nonsinusoidal load currents could be better
overall computer program also includes the tank and radiator estimated as given in [6]. The next revision of the IEEE test code
cooling with provisions to evaluate solar heating and wind effects. [7] will require that the hottest spot rise be reported on the test
The model allows the transformer designer to control the hottest report. The current loading guide [8] assumes that the user knows
spot temperature rise and optimize the thermal performance of the the hottest spot rise at rated load.
transformer to meet IEEE standards and user overload
requirements. C. Prior Work

Keywords: Distribution Transformer, Layer Windings, Heat Taylor, Berger, and Western [9]reported on an experimental
Transfer, Thermal Model, Hottest Spot Temperature study of the heat-transfer properties of cooling ducts in transformer
coils under natural convection. Data for circular layer windings of
various vertical duct sizes and coil heights were generalized in
I. INTRODUCTION terms of empirical winding-to-oil temperature gradients. Allen and
Allan [lo]used a resistance-network and an analogue computer to
A. Hottest spot Temperature Rise Requirements solve the thermal equations for a circular winding of a large power
transformer. Preiningerova and Korinek [ 1 11 extended the work
IEEE Std. C57.12.00-1993 [l) requires that the hottest spot using a digital computer.. A study of the heat transfer in a
temperature rise not exceed 80 C when the average winding rectangular layer type winding cooling duct was given in [12] for
temperature rise is 65 'C. Although this is a performance mineral oil, high temperature hydrocarbon, and silicone fluids.
parameter to be met by the manufacturer, there are currently no Tests were performed on a model duct assembly. A finite element
requirements to substantiate the value. The next revision of IEEE analysis of one 10 kVA test transformer is given in [13].A finite
Std. (37.12.00 [2]will require that the hottest spot temperature element analysis of a pole mounted distribution transformer was
rise be determined either by direct measurementor by calculation. reported in [14] and [15]. The thennal gradient between the
The calculation method shall be based on fundamentalloss and heat average winding temperature and hottest spot temperature was very
transfer principles and substantiated by tests on production or small. In [16]predictions are made for the temperature rise of a
prototype transformersor windings. A guide [3]will describe the small distribution transformer for nonsinusoidal load currents.
requirements for a manufacturer's test program or mathematical Tests with nonsinusoidal load currents were dso reported in [ 171
model to demonstrate compliance with IEEE Std. C57.12.00. One and [18]however the number of thermocouples was insufficientto
manufacturer [4]of small distribution transformers has proposed plot temperature distributions. Pierce [19]developed an extensive
that the 65 'C average winding rise could be exceeded and only finite difference computer program for air cooled round layer
meet the 80 'C hottest spot rise. windings for dry type transformers.

B. User Needs D. Objective of Present Work

Utility users of distribution transformers require a value for the The objective was to develop computer subroutines which could
hottest spot temperature rise to predict loading capability due to a be utilized in the overall transformer design program to permit
varying load profile. Past practice [5] for distribution transformers thermal optimization and prediction of the hottest spot temperature
on every transformer design. Finite difference approachesutilizing
digital computers programmed in FORTRAN were considered the
best approach. A programming approach utilizing simple nodes is
illustrated in [20]. The computer program, TWIND, described in
this paper is more complex and utilizes variable node sizes to
accurately model the geometry of the transformer windings.

0-7803-5515-6/99/$10.00 0 1999 IEEE 925

Authorized licensed use limited to: General Electric Global Research. Downloaded on March 01,2021 at 13:07:25 UTC from IEEE Xplore. Restrictions apply.
 11. SUMMARY OF MATHEMATICAL MODEL B. Nodal Equations
FOR WINDING HEAT TRANSFER
Two dimensional finite difference nodal equations were used
A. Winding Description to describe the heat flow within and from the winding and core.
A co-ordinate system was used with the X coordinate along the
Distribution and small power transformers utilize rectangular conductor and the Y coordinate across the layers of conductor.
layer windings. A quarter section of a winding is illustrated in The subscripts I, and J describe the nodal positions in the X and Y
Figure 1. Cooling ducts are located in the portion of the winding directions. The winding layout showing the direction of the 1 and
outside the core iron. Fo:: this reason some refer to this winding J indices is illustrated in Figure 1. A similar procedure was used
as a "collapsed duct rectangular winding". That is, the cooling in [19]. As shown in [19] A generalized equation describing the
ducts do not extend compktely around the winding like round layer heat transfer at any node is given by;
windings used in power transformers. To reduce the hottest spot
tempera-, some transformers may contain a few ducts (termed
annular ducts) which extend completely around the winding.
The winding assembly may consist of secondary and primary
coils. For small distribution transformers, the winding
arrangement may consist of an inner secondary winding, primary where,
winding, and an outer secondary winding. The windings are K conductance from node (1,J) to an adjacent node,
wound tightly over each other with insulation between the watts/*C;
windings, although ducts are sometimes used between windings. Q, heat generated at a node, watts;
The space between the inner secondary winding and the core is T temperature riy over adjacent bulk fluid for an
small and the insulation under the winding may be in contact with adjacent node, C;
the core. The adjacent windings of three phase units, or two T,, temperature rise over adjacent bulk fluid for
legged single phase units, usually touch and may contact the outer node (I,J), 'C;
core legs of shell type or three phase five legged core type units. U overall nodal heat transfer coefficient from node
Cores may be either stacked or wound. (1,J) to ducts above or below the node and from the
The collapsed duct arrangement causes a circumferential top or bottom of the winding, wattsl'c.
temperature gradient from the center of the winding, which has no
ducts under the core iron, to the center of the winding outside core At nodes where specific conductances or heat transfer
iron where cooling ducts are. located. In addition, thermal coefficients are not present, they are omitted from the generalized
gradients are present between layers. The vertical temperature nodal equation to give a specific nodal equation. Many different
gradient is due to the oil flow within the cooling ducts and the specific nodal equations were required to describe all possible
outer surface of the windings. Fluids other than mineral oil are combinations of heat transfer components within the windings and
also used for distribution transformers in industrial applications. from the windings to the core. The computer program assigns the
correct nodal equations to the various nodes based on the geometry
and duct locations. Where applicable, either quarter (Figure 1) or
i half symmetry is used to minimize the number of nodes. This
results in a temperature matrix that is a two dimensional
representation of the temperature rise over adjacent fluid in the
main tank at the same elevation. The temperature rises can be
UNDER IRON considered over average bulk oil for determining the average rise
and over top fluid for determining the maximum (hottest spot rise).
A model of the core was used to calculate heat flow from the
windings to the core and from the core to the fluid.
CORE C. Heat Generation

The amount of heat generated at a winding node due to the

current is dependant on the current magnitude, conductor resistivity
(which depends on temperature), and per unit eddy losses. The
# tapped out turns of the primary are considered. For the low tap
8 position the heat generated at the nodes located in the tapped out
turns is zero. Heat generated within the various nodes include
eddy loss. Although eddy losses are extremely low the program
code may by easily modified to pennit a non-uniform distribution
of eddy losses throughout the windings. The core model allows
for the heat generated due to core loss.
-J

Fig. 1. Quarter section of a collapsed duct rectangular winding.

926

Authorized licensed use limited to: General Electric Global Research. Downloaded on March 01,2021 at 13:07:25 UTC from IEEE Xplore. Restrictions apply.
 D. Thermal Conductance G . Convergence

In the X direction the heat flow is considered to flow along the The program has successfully computed the temperature
coil conductor. In the Y direction the primary resistance to heat distribution in a winding assembly with 405 layers comprising
flow is through the tum and layer insulation. The program 4058 nodal temperatures. The upper limit for the number of nodes
accounts for different sizes of insulation within the winding. An has not been established. Temperature distributions are computed
additional gap between layers is assumed when round wire is used. for each layer. A tolerance between successive iterations or two
previous iterations for the difference in temperatures at all the
E. Heat Transfer Coefficients nodes is used to determine convergence. Experience has shown
that a value of .002 "C results in an accurate calculation without
The program considers the number, size, and location of excessive iterations. The convergence is accelerated by comparing
cooling ducts. The program is highly flexible in accepting duct temperatures of two successive iterations and predicting
placement by accommodating end ducts or annular ducts in any temperatures to try for the next iteration using an overrelaxation
combination. The inner or outer barrier may be solid, an annular factor. Similar methods were reported in [31]. The optimum
duct, or solid under iron and an a duct outside iron. The outer value of the overrelaxation factor was determined by trial and
wrap on winding may be solid or a duct assembly. The program mor. Acceleration of the convergence was found to decrease the
allows different duct sizes for the inner secondary, inner barrier, number of iterations by a factor of 5 to 10.
primary, outer barrier, and outer secondary, however within a
winding they are assumed to be the same size.
Windings utilizig sheet conductor have edges exposed at the 111. SUMMARY OF LIQUID RISE MODEL
top and bottom of the winding. For short distribution transformer
windings significant heat transfer occurs from the top and bottom In the past empirical methods correlatingtested fluid rises have
coil surfaces. For these surfaces and the exposed core, the heat proven satisfactory. However to comply with future changes to
transfer coefficients were determined using correlations for IEEE Std. C57.12.00 a finite difference computer program,
horizontal plates as given by McAdams [21] and others. TTANK, was developed to determine the fluid rise and
The outer surface of the outer winding was considered similar temperatures of major components of the tank, cabinet, and inside
to a vertical plate with uniform heat flux and cooled by natural air of the pole type, single phase pad, three phase pad, and
convection. The correlation of Sparrow and Gregg 1221 was used substation transformers. It is based on fundamental heat transfer
to determine the local heat transfer coefficient for the outer surface principles and uses a finite difference heat balance approach.
of the outer winding. Nodal equations were written describing the steady state heat gain
The heat transfer within a duct is the case of unequal heat flux; by convection and radiation and the heat loss by convection and
that is, the heat flux in watts per square inch is different on the radiation at each node. The analysis also considers the heat
two surfaces. There is little published information on this case, transfer in the gas space in the top of the tank and the radiation and
however the correlation of Miyatake and Fujii [23] was modified convection heat exchange within and from the cabinet of pad
for use in liquid immersed transformer windings. This correlation mounted transformers. They are important considerations for the
was used in the investigationby Pierce and described in [191. The small distribution transformers. The heat generated by electrical
equations give local heat transfer coefficients within the duct. components in the cabinet of pad mounted transformers is
These local heat transfer coefficients were averaged to give the included. The temperatures of the nodes are determined using an
average heat transfer coefficient in the duct. iterative approach. The program will predict steady state
Flow rates in the ducts are determined by the frictional performance under thermal test conditions as defined in IEEE Test
resistance of flow in the ducts and the available thermal head. The Code (37.12.90 [7]. In addition the program will predict steady
bulk bottom fluid to top fluid rise was determined by summing the state performance under field conditions including ambient
flow rates and temperature rises for all the ducts and the outer temperature variations, solar heating, and cooling effects of wind.
surface of the windings. The heat transfer coefficient in each duct
for temperature rise over the duct fluid was transformed into a heat
transfer coefficient for temperature rise over average fluid for use IV. TWIND PROGRAM RESULTS
in the nodal equations.
A. Temperature Profiles
F. Material Properties
Temperature gradients within a winding assembly are shown by
Material properties of insulation and core steel are given in the temperature profiles illustrated in Figures 2 and 3 for a 300
[24] and [25]. Conductor properties are obtained from readily kVA three phase transformer. The secondary winding consisted of
available sources. Fluid properties are considered in the heat 22 layers of aluminum sheet conductor with end ducts located over
transfer equations. TWIND evaluates the fluid properties in each layers 6, 11 and 17. The primary winding consisted of 22 layers
duct at the average film temperature. This film temperature is the of aluminum wire with end ducts located over layers 7 and 15. An
average of the fluid temperature at the mid-height of the duct and annular duct was located within the barfier that separates the
the average nodal temperatures on each side of the duct. Fluids secondary winding from the primary winding. Figure 2 shows
considered are mineral oil, high temperature hydrocarbon, and temperature profiles for each layer outward from the core for the
silicone. Properties are given in [12], [ X I , [27], [281, 1291 and under iron and outside iron regions. The hottest spot temperature
[30]. Other fluids could be considered by adding code to the fluid is located in the under iron region. The outside iron profile shows
property subroutine. the effect of the cooling ducts. Figure 3 shows the temperature
profile along several layers illustrating the heat conduction from
the under iron region to the outside iron region where ducts are
located.

927

Authorized licensed use limited to: General Electric Global Research. Downloaded on March 01,2021 at 13:07:25 UTC from IEEE Xplore. Restrictions apply.
 TABLE 1
COMPARISON OF WINDING INTERACTIONS
25 W A Single Phase Shell Type Design

Secondary I, Amp 104.17 0.0 104.17

Primary I, Amps 0.0 3.47 3.47
INDEX I. 1 IS UNDER IRON, 10 IS OUTSIDE IRON
I = 10
--I 1
---_ IS Ave Rise, 'C 10.91 8.11 17.66
OS Ave Rise, 'C 7.49 3.19 9.87
I IS + OS ave, 'C 8.74 4.99 12.73
Pri. Ave. Rise, 'C 5.57 9.51 13.90
IS HS Rise, 'C 11.19 8.93 18.04
OS HS Rise, 'C 8.26 4.48 11.73
pri. HS Rise, 'C 9.24 11.19 17.75

1 S = I ~ e Secondary,
r OS=Outer Secondary
HS=Hot spot rise over top fluid
Note: Core geometry was considered but no core loss, i. e., short
circuit heat run.

NODAL INDEX J TABLE 2

1 TO 22 IN SECONDARY COMPARISON OF CORE GEOMETRY EFFECTS
23 TO 44 IN PRIMARY 25 W A Single Phase Shell Type Design

Core Geometry no' Yes

Core loss no no
Fig. 2. Winding temperature rise profiles across layers.
IS Ave. Rise, 'C 24.63 19.42 17.66
OS Ave. Rise, :C 11.33 10.21 9.87
+
IS OS ave., C 16.20 13.59 12.73
Pri. Ave. Rise, 'C 15.98 14.42 13.90
INDEX sec. HS Rise, 'C 25.21 19.67 18.04
J=7
-
J=36 5-38
.......... pri. HS Rise, 'C 22.82 19.03 17.75

Core considered thermally insulated from windings

IS=Inner Secondary, OS=Outer Secondary
HS=Hot spot rise over top fluid

C . Core Heating Effects

-------- -- -....I.
Another important consideration for modeling of small
distribution transformer windings is to allow for heat flow to and
from the core. This was first suggested by Piqce in the discussion
of [15]. A significant portion of the heat generated within the
NODAL INDEX I windings may be dissipated by the core exposed to fluid. The
1 IS UNDER IRON program was coded to either consider the core thermally insulated
10 IS OUTSIDE IRON or to consider the effects of core geometry. The final version
considers the core geometry. Table 2 gives the results of this
Fig. 3. Winding temperature rise profiles along layers. comparison.

D. Comparison of Hottest Spot and Average Rises

B. Winding Interactions
A listing of the program predictions for various designs is
Many small distribution transformers have few or no cooling given in Table 3. These results were from designs availableat the
ducts. For these units it is extremely important that the thermal conclusion of the program development but prior to thermal
model allow for heat flow between the various windings. optimization. The difference between the hottest spot and the
Simplistic models which assume that one winding is thermally average rise for small distribution transformers is usually very low.
insulated from the other are- not valid. Table 1 illustrates the This agrees with the test data and predictions given in [ 141 and
influence of the various windings on each other. The program was 1151.
run to compare temperature rises assuming zero current in each of The thermal gradient between hottest spot and average rise is
the windings with temperature rises for current in both windings. greater for larger distribution transformers. The IEC Loading
The results are for a 25 W A single phase shell type unit with an Guide 1321 gives equations with an H factor. The average winding
inner secondary, primary winding, and outer secondary winding. rise over average liquid is multiplied by this factor and added to
Two end ducts were located within the primary winding. The the top oil rise to give the estimated hottest spot rise. This factor
inner and outer secondary windings had no ducts. is intended to account for higher eddy losses at the top of the

928

Authorized licensed use limited to: General Electric Global Research. Downloaded on March 01,2021 at 13:07:25 UTC from IEEE Xplore. Restrictions apply.
 TABLE 3 The maior challenge in computer programming was to develop
COMPARISON OF HOT SPOT AND AVERAGE RlSE code to compute all the various parameters and to assign the proper
nodal equations for any combination of cooling duct arrangements
kVA-ph Nodes Wdg. HS Ave. HS-Ave. HS+Ave' and for many different core and coil geometries. Developing code
to correctly compute the nodal dimensions was also a major
15-1 559 S 7.16 7.03 .13 1.018 challenge. The program was coded to easily permit changes to
P 5.56 3.91 1.65 1.422 various constants and boundary conditions to adjust the predictions
25-1 650 S 18.04 12.73 5.31 1.417 to agree with test data obtained with direct reading sensors. A
P 17.75 13.90 3.85 1.277 later paper may compare predictions with test data.
50-1 450 S 13.39 11.16 2.23 1.200
P 17.96 13.43 4.53 1.377
100-1 430 S 16.18 14.34 1.84 1.128 VI. RECOMMENDATIONS
P 17.46 11.47 5.99 1.522
100-1 4050 S 12.03 9.25 2.78 1.300 The following factors were recommended for inclusion in the
P 14.55 10.34 4.21 1.407 IEEE Guide document [3] as important factors to consider in a
100-1 300 S 8.76 7.81 .95 1.122 winding heat transfer model for distribution or small power
P 7.87 5.45 2.42 1.444 transformers with rectangular collapsed duct layer windings:
150-3 676 S 12.28 8.29 3.99 1.481
P 12.33 9.98 2.35 1.235 1. Quantity, location, and size of cooling ducts should he
300-3 440 S 14.63 10.83 3.8 1.351 considered.
P 26.97 19.54 7.43 1.380 2. Heat flux from each side of the cooling duct is different. The
750-3 330 S 21.60 14.21 7.39 1.520 heat flux also varies with the particular cooling duct and
P 33.37 19.83 13.54 1.682 circumferential along a given duct.
2000-3 280 S 29.83 17.99 11.84 1.658 3. Heat flow across layers should be considered. Layer insulation
P 39.56 22.72 16.84 1.741 thicknesses may vary throughout the winding.
2500-3 210 S 26.22 16.05 10.17 1.634 4. Circumferential heat flow from the portion of the winding
P 35.03 21.43 13.60 1.635 under core iron to the portion of the winding outside core iron
should be considered.
Ave 1.405 5 . Heat flow between the various windings and the core occur and
IEC H Factor should be considered.
6. An accurate determination of the dimensions of all parts of the
winding is important to correctly predict temperature rises. To
winding. The Guide uses a typical number of 1.1 for distribution properly allow for conduction heat transfer and loss generation,
transformers and 1.3 for power transformers. A 1995 Cigre it is of particular importance to determine the dimensionsof the
Electra article [33] giving experimental results using fiber optic parts of the winding where cooling ducts are not present.
data showed that the hot spot factor ranged from 0.51 to 2.06 for 7. Layers and turns in the tapped out parts should be considered.
power transformers. The conclusion was that the factor cannot be 8. Fluid properties should be considered.
represented by a single constant. The calculated data shown in
Table 3 also indicates that the factor of 1.1 for distribution
transformers is not valid. A factor of 1.1 would indicated a VII. REFERENCES
temperature differential between hottest spot and average rise of
only 1 to 2 degrees! IEEE Standard General Requirements for Liquid-Immersed
Distribution, Power, and Regulating Transformers, IEEE
Std C57.12.00-1993.
V. DISCUSSION IEEE Draft Standard General Requirements for Liquid-
Immersed Distribution, Power, and Regulating
One method of predicting temperature distributions in Transformers, IEEE Std C57.12.00-199N, Clause 5.11.
transformer windings is finite element analysis (FEA). FEA, using IEEE Working Group on Hottest Spot Temperature Rise
commercialprograms, is a time consumingprocess of building and Determination, "Guide for Determination of Maximum
analyzing a mathematical model of a single proposed or existing Winding Temperature Rise in Liquid Filled Transformers",
design. Commercial programs utilizing FEA were not considered Draft 1.0, Nov. 4, 1997.
suitable for an automatedprogram for many different designs. The G. G. Gonzalez, A. Yerges, and G. Goedde, "Distribution
automatedfinite difference thermal subroutineTWIND reduced the transformer for the 21st century," I4rh fnt. Con$ on
three or four days required by a commercial FEA program to a Electricity Distribution (CIRED), 1997, IEE Conf. Pub.
few seconds. 438, Vol. 1, Paper 1.25.
With present computer technology it is not necessary to have IEEE Guide for Loading Mineral-Oil-Immersed Overhead
a main frame computer to solve large systems of equations. The and Pad Mounted Distribution Transformers rated 500 kVA
program was developed in FORTRAN code using a 200 MHZ and Less with 65 OC or 55 'C Average Winding Rise, IEEE
Pentium PC. The PC version gives a complete printout of the Std (37.91-1981, p. 28, eq. 7.
temperatures at all nodal locations. For the extreme case of 405 IEEE Recommended Practice for Establishing Transformer
layers comprising 4058 nodes the running time was about two Capability When Supplying Nonsinusoidal Load Currents.
minutes. The PC FORTRAN code was then adapted as a ANSIlIEEE (37.110-1986. (A revision was submitted to the
subroutine for the transformer design program on a main frame IEEE standards Board in 1998.)
computer.

929

Authorized licensed use limited to: General Electric Global Research. Downloaded on March 01,2021 at 13:07:25 UTC from IEEE Xplore. Restrictions apply.