3, July 1988

Abstract Winding losses increase when a transformer sup urements for a similar unit have already been published by the
plies nonsinusoidal load currents. The corresponding increase in authors [7], and these were repeated for this unit. The calcula
winding temperature is usually assumed to increase propor tions were performed using a finite element based numerical algo
tionately with loss. Also, the position of the winding hotspot is rithm. This technique is now expanded to combine electrical and
generally presumed to be fixed, regardless of harmonic current thermal models and to provide an overall evaluation of the impact
content. The purpose of this paper is to compare winding tem of harmonic currents on small distribution transformers.
perature predictions from a finite element based method with The purpose of this paper is to compare winding tempera
actual measured values for a 10 kVA oilimmersed wirewound ture predictions from a finite element based iterative algorithm
single phase distribution transformer. The results confirm the with actual measurements for the test transformer at common
prediction algorithm and illustrate that the position of the hotspot lowordered harmonics. An iterative algorithm is used to solve
remains basically unchanged when harmonics are present. both electrical and thermal problems simultaneously, thereby
allowing complete interaction between temperature and winding
INTRODUCTION resistivity in every turn. The results compare favorably. In addi
The widespread use of solid state controlled loads has tion, it is observed that the position of the winding hotspot
increased harmonic levels significantly during the past several remains basically unchanged for typical harmonic levels.
years. Harmonics affect both power system operation and power
system components, thereby necessitating a better understanding APPLICATION OF THE FINITE ELEMENT METHOD
of their overall impact. Since distribution transformers are usu Use of the finite element method to calculate harmonic
ally the interface between a power system and a harmonics pro impedances of transformers has been described in the authors
ducing load, they are components of major interest in this regard. previous publication [7]. The procedure has been shown to pro
Although analytical and numerical methods for analyzing har duce very accurate results. Since this procedure has been
monic problems in transformers exist [l41, an overall evaluation described in detail already, only the thermal finite element prob
of the effect of harmonics on transformer temperatures and loss lem is presented here.
of life is needed.
Let the temperature rise @(x,y) in the windings be defined
Increases in power loss and temperature rise are the primary as
harmonicsrelated concern in transformers. The presence of har
monic currents requires transformer deration [ 5 ] . Usually, two @(x,y) = T(x,y)  To,, 9 (1)
assumptions are made during this procedure:
where T(x,y) is the temperature at winding position (x,y), and
. the increase in temperature is proportional to losses To,, is the temperature at the bottom of the oil duct. The steady
the position of the winding hotspot is fixed. state twodimensional heat conduction equation [891 within the
Using these, increased losses are calculated with the frequency winding cross section is
squared rule [5], and the transformer is derated to maintain con
stant winding loss. In actuality, winding eddycurrent losses are
inversely proportional to winding temperature rise [6]. Hence, it
is important to consider the interaction between winding losses where
and temperature when making an evaluation of the overall impact K,, K ~ : Thermal conductivities in the x and y directions for
of harmonic currents on transformers. the conductors or insulation materials
For the purpose of this research, a speciallyequipped 10 q: Heat source in the conductors, zero in the insulation
kVA, 72001240 volt single phase distribution transformer was materials.
constructed and used to conduct thermal tests with harmonic Note that the heat source, or power loss density, exists only in
currents through the 25th multiple of 60 Hz. This wirewound the winding conductors, and not in the insulation materials or air
unit contains 36 thermocouples implanted in the windings, in the gaps. Applying the finite element method to (2) produces
oil, and on the tank. Electrical impedance calculations and meas
KSO=Q, (3)
where
K: Thermal conductivity of the conductors and insulation
87 SM 5366 A paper recommended and approved
materials
by the I E E E Transformers Committee of the IEEE Power S: Square matrix with elements defined by the finite ele
Engineering Society for presentation at the I E E E / ment method
PES 1987 Summer Meeting, San Francisco, California,
July 12  17, 1987. Manuscript submitted 0: Unknown column vector of temperature rises for the
January 27, 1987; made available for printing winding finite elements
April 17, 1987. Q: Known column vector of heat loss per unit length in
each element.
08858977/88/07001074$01.00@1988 IEEE
1075
The discrete temperature rise in the windings is obtained by solv The geometric configuration of the 10 kVA shelltype test
ing (3). transformer is shown in Figures 1(A) and 1(B) (without tank and
In order to solve (3), boundary conditions for the winding insulating oil). Winding conductors in both cross sections are
surfaces must be determined using arranged as Nparallel sets. The gaps between windings and
K
a0 + a
an E  e,] = 0, (4)
winding layers represent oil ducts. Insulating oil circulates
upward through these ducts, and heat created within the windings
passes through the conductors and insulating material to the cir
where n is the direction normal to the winding surface, a is the culating oil.
heat transfer coefficient, and 0,is the oil duct temperature rise. Discretization of the thermal finite elements in the test
Since the winding surface is surrounded by insulating oil, a transformer is performed with linear rectangular elements. For
represents the oil duct heat transfer coefficient. Complicated each winding set, boundary grid lines are established at winding
thermal tests are required to find the true a,which is a nonlinear surfaces which contact insulating oil. The temperature of hor
function of temperature and oil duct height [lo]. However, for izontal boundary lines is assumed to be constant and equal to
long narrow oil ducts like those in the test transformer, this pro either the top or bottomoil duct temperature. Temperature in the
cedure can be simplified by assuming that the oil duct tempera vertical oil ducts is assumed to vary linearly.
ture varies linearly between the bottomoil duct and topoil duct. Due to device symmetry, only onehalf of each cross section
In other words, a is assumed to be constant and is calculated
shown in Figure 1 need be modeled. Likewise, since the wind
from the average oil duct and winding temperatures.
ings are assumed to have the same temperature distribution
The bottomoil duct, topoil duct, and average winding tem whether inside or outside the winding window, only one cross
perature rises are. calculated using a thermal analog circuit model section need be modeled. These assumptions have been verified
(described in the next section). Therefore, there are two steps with thermal test results.
needed to calculate winding temperature distribution at any
specified power loss level: THERMAL ANALOG CIRCUIT MODEL
A fifthorder thermal'analog circuit model for the 10 kVA
1. Find the oil duct heat transfer coefficient and establish the
test transformer with natural oil cooling is shown in Figure 2.
boundary conditions using the analog model
Nodes 15 represent high voltage windings, low voltage windings,
2. Solve the finite element conduction equation (3) using the core, oil, and tank, respectively. T1 and T2 are the average tem
boundary conditions. perature rises in the high and low voltage windings. T3T5
denote the temperature rises above ambient in the core, bottom
oil duct, and tank. Heat inputs Q1, 4 2 , and 43 are the power
losses in the high voltage winding, low voltage winding, and
core. Power loss in the tank wall is neglected. Parameters G1
G8 are the heat conductances between thermal nodes.
Parameters (3148 must be determined from thermal tests.
The procedure used is similar to that proposed by Lindsay [ 111.
First, several sets of temperature rises in nodes 15 are measured
by energizing heat source Q1 (using DC current injection into the
high voltage winding). Temperatures are measured, and the
results are shown in Figure 3, where a fitting function of the form
f(x) = is used to smooth and interpolate the results and to
reduce the number of required thermal tests (note  each thermal
test requires approximately one to three hours). Next, an AC
10.6 in.
short circuit test is performed, which energizes heat sources Q1
and 42. Thus, the problem of injecting a very large DC current
Figure l(A): Cross Section of the 10 kVA Test Transformer into the low voltage winding is avoided. After performing each
(Inside Window) test twice at different power loss levels, the results for source Q2
are obtained by subtracting the temperature contribution from
source Q1 (based on Figure 3) from the measurements. The
results shown in Figure 4 are obtained using the f(x) fitting func
tion,
'6
t
7.0 k 1
I
Ql

.._
1. LV Whdino
2. HV WhdhO
Figure l(E3): Cross Section of the 10 kVA Test Transformer Figure 2: Thermal Analog Circuit Model for the Test
(Outside Window) Transformer
1076
LV Winding
Core
Bottom Oil Duct G8 I 7.26704 I 0.11327
Tank
Gn(T4) = a, + alT4
T4 : Temperature of BottomOil Duct
50 I1 m=0.8 /
/ Lv Winding
resistance method.
The principal need for the thermal analog model is to deter
mine heat transfer coefficient CL in the oil ducts as well as
boundary conditions for the thermal finite element solution.
Temperature rises TlTS in nodes 15 can be calculated using
this analog model for any given load current pattern. Fitting
function f(x), which is based on measured temperatures, is used
to predict topoil duct temperatures at different power levels.
The average temperature of bottom and topoil ducts, along with
the average high and low voltage winding temperatures, are used
to predict a,which is needed in the finite element solution.
I
I
I
I x34 I
HV AND LV POUER LOSSES
I I
I
I
I
I
t
Figure 6(A): Arrangement of Thermal Sensors in the 10 kVA
NO
> Test Transformer (Inside RightHand Window)
i
I
I . ELECTRICAL F I N I T E UODEL
I
POWER DISTRIBUTION I
I
I
I
Figure 5 :
1 B . OUTPUT
Using the same short circuit thermal tests at 60 Hz, a com The iterative algorithm confirms that the position of the
parison of several measured and predicted actual winding tem hotspot within each winding is relatively fixed, regardless of the
perature rises is shown in Figures 7(A) and 7(B). The predicted harmonic pattern. Under certam circumstances, however, it is
values are calculated using the iterative algorithm. For the 250 possible that the transformer hotspot could move from one wind
W case, the iterative algorithm with resistivity correction is ing to the other. However, for low voltage wirewound distribu
approximately 2" C more accurate than the thermal finite model tion transformers subjected to typical distorted loads, it is reason
only (which ignores loss variation with temperature). Although able to assume that the position of the transformer hotspot is
there is a 4.2' C maximum error between measured and predicted fixed.
temperatures at the hotspot (position 11) for the rated power loss
case (250 W), this is considerably better than the 15'C error
found by applying standard calculation procedures [12]. This
serves to validate use of the iterative algorithm for determining

 Predicted Temperature of High Voltage Winding
CONCLUSIONS
An algorithm which combines electrical and thermal finite
element models in an iterative loop is used to predict total wind
ing loss and temperature distributions in a specially equipped
wirewound 10 kVA distribution transformer which is subjected
to harmonic currents through the 25th order. The iterative algo
rithm illustrates that the dependence between winding power loss
and temperature rise is important in evaluating the overall impact I I I I I I I
of harmonic currents on transformers. Since the winding eddy 1 2 3 4 5 6 7
current loss is observed to be inversely proportional to tempera Winding Height  Inches
ture, the overall temperature rise is not exactly proportional to
Figure 7(B): Comparison of Measured and Predicted Tempera
total winding loss, as would be calculated by superimposing indi
ture Rise ("C) Outside the Window of the Test
vidual harmonic losses.
Transformer
1079



Predicted Temperature of High Voltage Winding
Measured Temperature of High Voltage Winding
Predicted Temperature of Low Voltage Winding
................ Measured Temperature of Low Voltage Winding
g:
=PI
~
12
1 Figure 8(A): Fundamental
Frequency
,
1 .o
1 5 9 13 1 7 21 25
h  Harmonic Multiple of 60 Hz
Pu(l): Measured = 27.9W, Redicted = 28.6W
Figure 9: Deration Curve for the Test Transformer When Sub
5
0 1 2 3 4 5 6 7
jected to Individual Harmonic Currents
:$
Winding Height  Inches
REFERENCES
1. S . Crepaz, "Eddy Current Losses in Rectifier Transformers,"
Figure 8(B): 15th Harmonic
IEEE Transactions on Power Apparatus and Systems, Vol.
PAS89, No. 7, pp. 16511656, September/October 1970.
13
2. A.E. Emanuel, Xiaoming Wang, "Estimation of Loss of Life
of Power Transformers Supplying Nonlinear Loads," IEEE
Transactions on Power Apparatus and Systems, Vol. PAS
104, NO. 3, pp. 628636, March 1985.
3. E.F. Fuchs, D.J. Roesler, K.P. Kovacs, "Aging of Electrical
Appliances Due to Harmonics of the Power System's Vol
tage," IEEE Transactions on Power Delivery, Vol. PWRD1,
No. 3, pp. 301307, July 1986.
4. S.V. Preiningerova, V. Kahous, "Thermal Analysis of Large
OilImmersed Transformer Windings," Elecrric Machines
and Power Systems, Hemisphere Publishing Corporation,
Pu(15): Measured = 33.4W, Redicted = 32.6W
Washington, D.C., pp. 89102, 1983.
5 5. IEEEPES Transformer Committee, "Recommended Practice
0 1 2 3 4 5 6 7
Winding Height  Inches for Establishing Transformer Capability When Supplying
:
Nonsinusoidal Load Currents," C57.110, 1986.
Figure 8(C): 25th Harmonic 6. S.A. Stigant, The J . and P. Transformer Book, Johnson and
Phillips LTD., London, 1941.
7. M.S. Hwang, W.M. Grady, H.W. Sanders, Jr., "Distribution
Transformer Winding Losses Due to Nonsinusoidal
Currents," IEEE Transactions on Power Delivery, Vol.
PWRD12, NO. 1, pp. 140146, January 1987.
8. J.P. Holman, Heat Transfer, Fifth Edition, McGrawHill,
Inc., New York, 1981.
9. S.V. Preiningerova, M. Pivmec, "Temperature Distribution in
Coils of a Transformer Winding," Proceedings of the IEE,
Vol. 124, NO. 3, pp. 218222, March 1977.
10. R.L. Grubb, M. Hudis, A.R. Traut, "A Transformer Thermal
Duct Study of Various Insulating Fluids," IEEE Transactions
P,(25): Measured = 40.2W, Predicted = 39.3W on Power Apparatus and Systems, Vol. PAS100, No. 2, pp.
466473, February 1981.
5
0 1 2 3 4 5 6 7 11. J.F. Lindsay, "Temperature Rise Of An OilFilled
Winding Height  Inches Transformer With Varying Load," IEEE Transactions on
Power Apparatus and Systems, Vol. PAS103, No. 9, pp.
Figure 8: Comparison of Measured and Predicted Tempcrature 25302536, September 1984.
Rise (OC) in the Inner Low and High Voltage Wind 12. American National Standards Institute, "Guide for Loading
ings with Constant Current Magnitude (0.551 A = MineralOil Immersed Overhead and PadMounted Distribu
0.397 pu) tion Transformers," Standard C57.91, 1981.