IEEE TRANSACTIONS ON MAGNETICS, VOL. 42, NO.

4, APRIL 2006 967

Investigation of the Harmonic Behavior of Three

Phase Transformer Under Nonsinusoidal Operation
Using Finite Element and Wavelet Packets
O. A. Mohammed, Fellow, IEEE, N. Y. Abed, and S. Liu

Electrical and Computer Engineering Department, Energy Systems Laboratory, Florida International University,
Miami, FL 33174 USA

This paper investigates the harmonic behavior of three phase power transformers under nonsinusoidal operations. The terminal be-
havior of the transformer was obtained by coupling the transformer transient finite-element (FE) model and external electric circuits.
Such a technique would allow the physical representation of the nonlinear magnetization behavior of the transformer as well as the strong
frequency dependence of the transformer parameters. The harmonic behavior of the transformer currents and the dc load current were
analyzed using wavelet packet transform algorithm (WPT). The key findings are, uniform frequency bands resulting from WPT decom-
position process of the current waveforms can be used for identification of harmonic components. The flattened input voltage distorted
by 3rd harmonic component has the highest impact on the dc load current harmonics (with respect to the sinusoidal case). The peaking
input voltage distorted by the 5th harmonic component has the highest impact on the transformer secondary current and the magne-
tizing current harmonics.

Index Terms—FE, harmonic behavior, total harmonic band distortion, transformer, wavelet packets transform.

I. INTRODUCTION Harmonics and discontinuities generated in the power system

T RANSFORMERS are ubiquitous in all parts of the power
system, between all voltage levels, and exist in many dif-
ferent sizes, types, and connections. The increased utilization
of nonlinear power electronics components coupled with the
more prevalent use of microprocessor-based equipments gener-
ates harmonics and introduces power quality issues on the dis-
tribution networks.
Power quality problems have negative effects on the energy
system components. Transformers are most affected with this
problem and may need to be derated to as much as 50% capacity
when feeding loads with extremely distorted current waveforms
and may cause reduction in their service life [1].
There are different models for transformers for various con-
tingencies. However, no transformer model, currently available
for the study of electromagnetic transient and harmonics, ade-
quate for a wider range of frequencies [2]. Also, the frequency
dependency of the transformer parameters is not easy to include.
There are still a number of problems associated with frequency
dependent models in terms of relative complexity, accuracy, and
numerical stability.
Finite-element (FE) modeling can include all of these effects
and provide an efficient way for analysis, simulation, and opti-
mization for transformers. The FE analysis has been coupled to
circuit simulation. This technique allows us to simulate the op-
erating conditions of the transformer with the real power supply
connections. We will utilize a transformer coupled FE-circuit
model to obtain the transformer terminal behavior under nonsi-
nusoidal operation.
can have a wide frequency bandwidth, from high-frequency
transients and edges to slowly varying harmonic components.
Hence, analysis only in the frequency or time domain alone
is not sufficient to capture features that are spread within
a wide bandwidth. The wavelet transform provides a local
representation (in both time and frequency) of a nonstationary
signal. Therefore it is suitable for analyzing a signal where time
and frequency resolution is needed, unlike FFT which gives a
global representation of the signal.
This paper investigates the effect of the supply harmonics and
nonlinear loads on the transformer terminal behavior with the
help of coupled FE-circuit model and wavelet packet transform
(WPT). This study provides us with an efficient way to evaluate
different transformer designs, diagnose the incompatibilities of
the transformer with the load, predict the future performance of
load equipment or power quality mitigating devices, or trans-
former protection system, as well as design and evaluate filter
topologies.
II. THE COUPLED FE-CIRCUIT MODEL
A. Field Equations
The magnetic field inside the transformer is governed by the
following nonlinear partial differential equations:
(1)
(2)

where is the vector potential, is the total current density,

is the magnetic reluctivity, is the electric scalar potential, and
Digital Object Identifier 10.1109/TMAG.2006.872467 is the electric conductivity.
0018-9464/$20.00 © 2006 IEEE
968 IEEE TRANSACTIONS ON MAGNETICS, VOL. 42, NO. 4, APRIL 2006

B. Electric Circuit Equations

In general, any electric circuit can be represented by the fol-
lowing equations:

(3)

where , and represent the vector of

voltages source in each electric mesh m, the vector of the cur-
rent in each mesh, the matrix of resistances, the matrix of induc-
tances, the matrix of the reciprocal of the capacitance, and the
matrix of the nonlinear voltage drop, respectively.

C. Coupling Fig. 1. WPT two decomposition levels with successive filtering and
downsampling by 2.
Equations (1) and (2) are discretized using Galerkin method
[3]. The resultant equations are coupled with the circuit equation
(3) and solved simultaneously. The external circuit connections B. Wavelet Representation of RMS and Total
are used to describe the electrical connectivity between the con- Harmonic Distortion
ducting regions, external loads, and power supplies. During the In DWT the signal can be represented in terms of shifted
coupling of the electric circuit with the FE domain, we need to and dilated versions of a bandpass wavelet function and
associate the coil conductors with the corresponding entities of lowpass scaling function as follows:
the FE domain.
The simulation were carried out on a 3-phase, 150-kVA,
240/120 V, 60 Hz, transformer. The conductors have constant (4)
conductivities. The primary and the secondary windings are
represented by rectangles of corresponding materials in the
FE domain. The magnetic core is an isotropic nonlinear mag- where are the
netic material defined by analytic saturation curve. The FE scaling function coefficients.
model contains 5083 second order elements with 13 645 nodes. are the DWT coefficients of at
Dirichlet boundary condition was adopted on the external circle node at level .
of infinite region surrounding the transformer. The node is the lowest frequency band of the original
waveform. The nodes include the waveforms of higher
III. WAVELET PACKET TRANSFORM ALGORITHM frequency order bands.
A. Introduction The RMS of a waveform with period can be ex-
The Wavelet transform is a time-frequency DSP technique, pressed in terms of wavelet at a certain level as follows:
which decompose a signal in terms of oscillations (wavelets)
localized in both time and frequency. The main advantage of
wavelet over the short time Fourier transform (STFT) is that it
uses a variable-sized regions windowing technique, while STFT
uses a fixed time window size for all frequencies. For nonsta- (5)
tionary signals, STFT does not track the signal dynamics prop- Using the wavelet the scaling function and the wavelet basis
erly because of the fixed window width. On the other hand, orthogonality property, (5) becomes [4]
wavelet uses long time intervals where we want more precise
low-frequency information, and short time intervals where we
want high-frequency information. (6)
The discrete wavelet transform (DWT) is computed by suc-
cessive lowpass and highpass filtering of the discrete time-do-
main signal together with changes in sampling rates. where is the RMS value of frequency band at node .
A signal can be successively approximated by DWT with The total harmonic band distortion (THBD) is defined by the
different scales. Each step of the decomposition of the signal ratio of the RMS value of the harmonic bands at (i.e.,
corresponds to a certain resolution. The decomposition process excluding the lowest band) to the RMS value of the distorted
can be iterated, with successive approximations being decom- waveform. The THBD for the signal is given by
posed in turn. Therefore one signal is broken down into many
lower-resolution components. This is called the wavelet decom-
position tree. (7)
WPT method is a generalization of wavelet decomposition.
In wavelet packet analysis, the details as well as the approxima-
tions can be split. Fig. 1 shows a two level WPT decomposition The algorithm was then implemented to extract the different
levels. waveform features.
MOHAMMED et al.: INVESTIGATION OF THE HARMONIC BEHAVIOR OF THREE PHASE TRANSFORMER 969

Fig. 2. The system used for the study with the electronic rectifier load.

IV. IMPLEMENTATION AND SIMULATION RESULTS

The common source of harmonics and harmonic problems,
especially in industrial networks, are the three-phase bridges
supplying motor drives. With this in mind, Fig. 2 shows the
system used for the study. It consists of a three phase voltage Fig. 3. Transformer secondary current for different supply harmonics.
supply, a transformer coupled FE-circuit model, and a three
phase bridge rectifier feeding resistive-inductive load (RL load).
To simulate the transformer open-circuit condition a very high
resistance load was used.
Usually in industrial networks the strengths of the odd har-
monics of the supply voltage are higher than the strengths of the
even harmonics. Therefore, we conducted a number of studies
with a wide variation of odd harmonic characteristics of the
supply voltage.
The study includes the following scenarios: a) Ideal sinu-
soidal supply voltage, b) Supply voltage distorted by the 3rd,
5th, 7th, 9th, 11th, 13th, and 15th harmonic components, respec-
tively with a peak value 10% of the fundamental peak. For each
harmonic component added to the voltage waveform, there are
two cases; peaking case and flattened peak case. In the peaking
case, the positive peak of the harmonic waveform coincides with
the positive fundamental peak. In the flattened peak case, the
negative peak of the harmonic component coincides with the
Fig. 4. Magnetizing current for different supply harmonic components.
positive fundamental peak, and c) Supply voltage distorted by
combination of the odd harmonics (up to the 15th harmonic), in
which the harmonic distortion of the 3rd, 5th, and 7th harmonics
is 3% of the fundamental and 1% for the higher odd harmonics.
In order to study the individual effect of the 3rd, 5th, and 7th
harmonics in the presence of other harmonics, their amplitude
is changed to 10% of the fundamental respectively, while the
other harmonics are kept unchanged.
A frequency study was performed using FE to investigate
the frequency dependence of the transformer parameters. The
supply frequency was varied from 60 Hz to 1 kHz. The trans-
former inductances and resistances were changed by a big per-
centage. For example, the primary resistance value increased by
49.6% (60 Hz–1 kHz), while the primary inductance value de-
creased by 76% (60 Hz–1 kHz).
Figs. 3, 4, and 5 show, respectively, the effect of supply
voltage waveforms, when distorted by different harmonics on
the transformer secondary current, the magnetizing current,
and the converter dc side current.
Fig. 5. DC side current for different supply harmonic components.
Here, “Mx” stands for a waveform that contains a number of
harmonics with “x” as the highest harmonic component (10%
of the fundamental). Here, “P” and “F” stand for peaking case wavelet was selected to perform our study. Wavelet db20 is from
and flattened case, respectively. Daubechies family of orthogonal wavelets with compact sup-
The current waveforms were then analyzed by the WPT al- port and highest number of vanishing moments. This will de-
gorithm. A five level decomposition with db20 as the mother compose the signal into 32 frequency bands.
970
Fig. 6. Wavelet transform coefficients of the dc side current under flattened
supply voltage distorted by 3rd harmonic for the first six bands.
TABLE I
WPT RESULTS FOR THE DC SIDE CURRENTS
TABLE II
WPT RESULTS FOR THE TRANSFORMER SECONDARY CURRENT
Fig. 6 shows the Wavelet transform coefficients (WTCs) of
the dc load current under flattened voltage supply distorted by
3rd harmonic component for the first six bands. Some bands
have very small or zero WTC amplitudes. This demonstrates the
wavelet property of time-frequency localization. In this figure,
the spikes at the beginning of each of the six bands correspond
to the starting disturbance.
Tables I, II, and III give the wavelet packet analysis results for
the dc side currents, transformer phase “a” secondary current,
and the magnetizing current, respectively.
Table I shows that the M5th, and 3rd F supply voltages have
the highest impact on the dc load current THBD. It will raise the
third harmonic component in the dc load current. Also, for Mx
voltage waveforms, there is a reduction in the effective dc cur-
rent (band 0), which indicates that the power transmitted to the
dc load side is reduced. Tables II and III, respectively, show that
the peaking voltage waveforms distorted by the 5th harmonic
has the highest impact on the THBD for the transformer sec-
IEEE TRANSACTIONS ON MAGNETICS, VOL. 42, NO. 4, APRIL 2006
TABLE III
WPT RESULTS FOR THE TRANSFORMER MAGNETIZING CURRENT
ondary current and magnetizing current. It will raise the fifth
harmonic component in the transformer secondary current. The
difference between the harmonic responses of the transformer
under sinusoidal and voltage supply distorted by 9th, 11th, 13th,
and 15th harmonics is relatively small.
In addition, from Tables I–III, one can conclude that the har-
monic response of the transformer is mainly dependent on the
phase angle of the voltage supply harmonics relative to the fun-
damental component, it also depends in lesser extent on the
magnitude of the harmonics components itself.
V. CONCLUSION
The effect of supply harmonics on the transformer harmonic
behavior under no-load and non linear load conditions is in-
vestigated. We propose a new approach to investigate this har-
monic behavior of the transformer by utilizing coupled transient
FE-circuit model and WPT.
The harmonic characteristics of the transformer magnetizing
current and load current are significantly different under nonsi-
nusoidal supply voltage. The transformer harmonic response is
mainly dependent on the phase angle of the voltage supply har-
monics, and to a lesser extent on the magnitude of the harmonics
components itself. The effect of supply voltage harmonics grad-
ually diminishes with the order of harmonics increase beyond
the 11th harmonic.
The results of the WPT analysis show an ability to quan-
tify different types of disturbances. It also shows high ability
of wavelets to extract the different harmonic components disre-
garding the length of their occurrence in time.
REFERENCES
[1] IEEE Recommendation Practice for Establishing Transformer Ca-
pability When Supplying Nonsinusoidal Load Currents, IEEE/ANSI
C57.110-1998.
[2] J. Pedra, F. Corcoles, L. Sainz, and R. Lopez, “Harmonic nonlinear trans-
former modeling,” IEEE Trans. Power Del., vol. 19, no. 2, pp. 884–890,
Apr. 2004.
[3] F. Piriou and A. Razek, “Finite element analysis in electromagnetic sys-
tems accounting for electrical circuits,” IEEE Trans. Magn., vol. 29, no.
2, pp. 1669–1675, Mar. 1993.
[4] O. A. Mohammed, N. Abed, S. Ganu, and S. Liu, “Wavelet analysis of
permanent magnet synchronous machine flux density harmonic content
with different pole number designs,” in Proc. ACES 2004.
[5] A. H. Chowdhury, W. M. Grady, and E. F. Fuchs, “An investigation of
the harmonic characteristics of transformer excitation current under non-
sinusoidal supply voltage,” IEEE Trans. Power Del., vol. 14, no. 2, pp.
450–458, May 1999.
[6] G. String and T. Nguyen, Wavelets and Filter Banks. Cambridge, MA:
Wellesley-Cambridge Press, 1996.
Manuscript received June 25, 2005 (e-mail: mohammed@fiu.edu).