TRANSFORMER CORE NOISE MEASUREMENT AND PREDICTION R. HAETTEL*,1, A. DANERYD1, C. PLOETNER2 and J. ANGER3 1ABB Corporate Research Sweden, 2ABB Transformers Germany, 3ABB Transformers Sweden
*Corresponding author: 721 78 Västerås Sweden, [email protected]
ABSTRACT
Today, lowering noise pollution is a matter of increasing importance worldwide. Power transformers, as any
other industrial products, shall comply with various requirements on noise levels. It is thus essential to predict
transformer sound levels with a sufficient accuracy at an early stage of the product design to select the most
appropriate strategy for noise control.
The paper will focus on core noise which is a typical multiphysics phenomenon involving electromagnetism,
mechanics and acoustics. A finite element model coupling the different physical fields will be used to study the
vibro-acoustic behaviour of a specific transformer operating in no-load conditions. The prediction model
provides a fairly accurate tool to carry out various parametric studies. Furthermore, some data obtained from
sound and vibration measurements will be presented to explain the mechanism implying core noise generation
and to show the correlation between experimental and predicted results.
KEYWORDS: Transformer, Measurement, Prediction, No-Load, Core Noise and Vibration.
1. INTRODUCTION
Low noise levels are nowadays requested for power transformers in order to comply with customer
specifications and environmental legislations [1,2]. Consequently, manufacturers have to improve the acoustic
performances of their products, while, at the same time, limiting the costs. It is thus of great importance to
predict sound levels with a sufficient accuracy at an early stage of the product design.
Three main sources of sound can be identified in transformers: 1) No-load noise or core noise generated by
magnetostriction in the core steel laminations [3], 2) Load noise produced by electromagnetic forces acting on
windings and tank walls [4,5] and 3) Noise due to auxiliary equipment such as fans and pumps used in the
cooling system.
The paper describes a method to predict core noise which constitutes a typical multiphysics phenomenon
involving electromagnetism, mechanics and acoustics. When an alternating voltage is applied to one or more
windings of a transformer, a magnetic flux is generated in the transformer core laminations made of grain
oriented electrical steel. This material has a non-linear anisotropic property called magnetostriction implying
alternating changes of the core dimensions due to the varying magnetic flux in the laminations. Those
magnetostrictive forces cause core vibrations which are transmitted to the tank via the insulation oil and the core
clamping points. Part of the mechanical energy is eventually radiated by the tank walls as noise in the
surrounding environment.
Due to the complexity of the structure and the strong coupling with oil, an analytical model cannot be used to
predict the sound radiation. In general, empirical methods based on statistics and dimensional basic parameters
are used by most transformer manufacturers. This approach presents limitations when applied to new designs
and does not enable accurate parametric studies. Therefore, prediction models based on finite element
formulations shall be utilized to describe accurately the complex interactions of the various design parameters
and the coupling of the physical fields.
In this paper, experimental tests performed in air on the scale model of a large power transformer core are
presented in section 2. In section 3, the equations governing the electromagnetic, mechanical and acoustic fields
in a transformer core are implemented in a multiphysics software using a finite element approach. This FE-
model is then applied to the scale model of the core and the results obtained by computation can be compared to
the experimental data in order to validate partially the FE-formulation. Finally, some data obtained by
performing vibration measurements on the core of a full-scale power transformer are reported in section 4.
2. ACOUSTIC TESTING ON SCALE MODEL OF TRANSFORMER CORE
Acoustical measurements were performed in laboratory conditions on a specific test core in order to identify and
describe clearly the mechanisms implying no-load noise in transformers. As a matter of fact, a scale model was
designed to be approximately one fourth of a core as used for a typical medium power transformer. The core
was manufactured by stacking grain oriented steel sheets clamped by two pressplates made of glass-reinforced
plastic. Both pressplates were joined together by means of bolts and nuts to clamp the core at a suitable pressure.
The main characteristics of the test core are given in Table I.
Table I. Test core dimensions and characteristics.
Core Height 650 mm
Core Length 750 mm
Limb Pitch 300 mm
Core Sheet Thickness 158 mm
Core Material Grain-oriented steel of HiB type
Sheet Thickness 0,27 mm
Core Construction 6 sheet step lap with single sheet stacking
Pressplate Material Glass-reinforced plastic
Pressplate Thickness 25 mm
The acoustical measurements were carried out on the test core suspended elastically in a reverberant room to
create free boundary conditions, as shown in Fig. 1 and 4. Prior to the tests, a well-defined static pressure was
applied by the pressplates on the core by using bolts and nuts in combination with Belleville washers.
Figure 1. Scale Model of the transformer core suspended by elastic ropes in the reverberant room.
The vibration levels were measured by 22 accelerometers placed at different locations on the core laminations
and on the core clamping plates, as shown in Fig. 3. The sound pressure level was recorded by one microphone
mounted on a rotating boom, as shown in Fig. 4. The sound power level radiated by the test core was estimated
by using a calibrated reference sound source to determine the room characteristics, as shown in Fig. 4. In
addition, the voltage was controlled under testing by using four search coils wound around the three limbs and
the upper yoke.
Figure 2. Dimensions of core and clamping plates given in mm.
The time signals provided by the 22 accelerometers, the microphone and the three search coils were acquired
simultaneously by 26 input channels of the measurement system.
The vibration measurements were performed for different positions of the accelerometers on the core, as shown
in Fig. 3. The accelerometers were mounted on core laminations to measure vibrations in all directions. The
expression "in-plane" and "out-of-plane" are used to describe vibrations of the core in the YZ-plane and in the
X-direction, respectively. The Y and Z-directions are represented in Fig. 3.
750
650
150
158 800
700
25
200
Figure 3. Accelerometer positions selected on the core laminations and the clamping plates.
A tailor-made excitation system was used to generate a magnetic field with three phases in the core at varying
frequencies, providing a realistic electromagnetic excitation due the magnetostriction. Therefore, the acoustical
behaviour of the test core in operation could be investigated thoroughly by means of a frequency sweep at
constant magnetic flux density.
To produce a magnetic field with three phases, a winding with several turns was wound around each limb of the
core, as shown in Fig. 1 and 3. The three windings were Y connected and the voltage supplied to each winding
was delayed with a determined time constant to simulate the no-load testing conditions used for a three-phase
transformer.
Figure 4. The Experimental set-up with the reference sound source and the microphone mounted on a rotating
boom used for the acoustic calibration of the reverberant room.
The excitation system mainly consisting of a frequency generator, a controller, a digital crossover, three
amplifiers and power resistors was connected to the winding cables wound around the core limbs to form an
electrical circuit with an adjustable frequency. The voltage applied to the circuit was kept sinusoidal and
sufficiently large to produce a magnetic flux density in the core generating measurable levels of sound and
vibration. The voltage was increased linearly with increasing frequency to keep a constant magnetic flux density
in the core during the entire sweep, as expressed in Equation (1). It is here assumed that the magnetostrictive
excitation varies exclusively as a function of the voltage on the entire frequency range for the tests. Therefore,
by using a constant magnetic flux density, the magnetostrictive forces can also be considered as constant for the
entire frequency sweep, implying that sound and vibration levels can be compared directly without a
normalization procedure.
MNO
P
V
U
A
R
C
DEF
G
H
I
J K L
QB
S
T
Y
Z
Rotating boom with microphone
Reference sound source
Figure 5. Voltage spectrum measured by the search coil on limb 1 for electrical excitation at 200 Hz.
During the tests, the voltage levels supplied to each of the three windings were checked at each frequency step.
In addition, the phase currents were observed and limited to a maximum value to control losses in the core and
the power electronic feeding system.
The voltage level applied to the windings and the magnetic flux density in the core are related according to the
expression
U = 2πf NBA (1)
where U is the supplied voltage, B the flux density, N the number of winding turns, A the cross area of the core
limb and f the frequency of the electrical circuit.
Figure 6. Acceleration spectrum measured on upper yoke at point E for electrical excitation at 200 Hz.
The magnetic field induced in the core varies at the same frequency as the voltage provided by the electrical
circuit. The magnetostrictive forces are generated by the magnetic field at even multiples of the current’s
fundamental frequency for a purely AC excitation [6,7,8]. Those forces imply core lamination vibrations which
eventually are transmitted to the air surrounding the core as sound. The electrical frequency of the circuit was
varied stepwise from 200 Hz to 1220 Hz, in general, with a resolution of 10 Hz. The procedure thus provided an
acoustical frequency sweep extending from 400 Hz to 2440 Hz with a resolution of 20 Hz.
Acceleration, sound pressure and voltage levels were recorded for a period of 60 seconds at each of the 102
frequency steps of the frequency sweep.
Measurement: Freq_Sweep M_01_200Hz
0 500 1000 1500 2000Frequency - Hz
10µ
100µ
1m
10m
100m
1
10
100
V peak Ch. 24 Volt_Limb1
200.0 [Hz]
11.0 [V]
400.0 [Hz]
0.004 [V]
600.0 [Hz]
0.039 [V]
Measurement: Freq_Sweep M_01_200Hz
0 500 1000 1500 2000 2500 3000Frequency - Hz
20
30
40
50
60
70
80
90
dB(lin) [1e-6m/s²] peak Ch. 5 Acc5_MidUpYoke
400.0 [Hz]
82.4 [dB(lin)]
800.0 [Hz]
60.9 [dB(lin)]
1200.0 [Hz]
59.2 [dB(lin)] 1600.0 [Hz]
37.2 [dB(lin)]
2000.0 [Hz]
38.4 [dB(lin)]
2400.0 [Hz]
56.4 [dB(lin)]
The time signals provided by the 22 accelerometers, the microphone and the four search coils were processed by
means of the analyzer to obtain vibration, sound and voltage levels with respect to frequency. The general
parameters used to perform the Fast-Fourier Transform (FFT) are given in Table II.
Table II. The FFT parameters used for the measured data analysis.
Sampling Frequency 8192 Hz
Frequency Resolution 1 Hz
Number of Frequency Lines 3201
Average Type Linear
Number of Averages Variable
Window Types Hanning
Overlap 66,7%
Function Class Autospectrum and Cross-spectrum
When a magnetic field is induced in the core for a given frequency, a number of peaks can be observed in the
frequency spectra for voltage, vibration and sound pressure obtained from the FFT-analysis, as shown in Fig. 5,
6 and 7, respectively.
Figure 7. Sound pressure spectrum recorded by the microphone mounted on the rotating boom for electrical
excitation at 200 Hz.
During the measurements, the voltage supplied to the circuit is kept sinusoidal. Therefore, the voltage spectrum
measured for example at side limb 1 presents mainly a single dominating peak at the fundamental frequency of
the electrical circuit, as shown in Fig. 5. The sound pressure and vibration spectra exhibit usually a main peak at
twice the fundamental electrical frequency and other peaks at multiples of this fundamental acoustical
frequency, as presented in Fig. 6 and 7. In fact, for an AC-excitation, the magnetostrictive forces are generated
at even multiples of the fundamental electrical frequency as long as there are no additional disturbances such as,
for example, a DC-component in the exciting voltage. Hence, the peaks identified in the sound pressure and
vibration spectra at even multiples of the fundamental electrical frequency correspond to the mechanical
responses of the core. The fundamental acoustic frequency is defined as twice the double fundamental electrical
frequency.
The peaks observed at odd frequencies in the acoustic spectra can be considered mostly as electromagnetic
disturbances affecting various parts of the measurement chain. The voltage signal can be somewhat distorted
due to the non-infinite stiff power source and the ohmic resistance of the leads and windings.
The levels recorded by the accelerometers and the microphone were selected in the vibration and sound pressure
spectra at the fundamental acoustical frequency for each step of the frequency sweep while the voltage levels
were recorded at the fundamental electrical frequency. That means that the higher harmonic components were
not considered further on. This procedure allows an accurate evaluation of the acoustical behaviour of the core.
As a matter of fact, for a low magnetic flux density the sound and vibration spectra present mainly a dominating
peak at the fundamental acoustical frequency whereas the peak amplitudes of the harmonics are significantly
lower.
Measurement: Freq_Sweep M_01_200Hz
0 500 1000 1500 2000Frequency - Hz
-20
-10
0
10
20
30
40
50
60dB(lin) [2e-5Pa] peak Ch. 23 Mic23
400.0 [Hz]
33.2 [dB(lin)]
800.0 [Hz]
17.0 [dB(lin)]
The vibration data collected at the fundamental acoustic frequency are averaged for in-plane, out-of-plane and
both directions. Moreover, the sound pressure data selected at the fundamental acoustic frequency are used to
estimate the sound power level radiated by the test core.
Figure 8. Averaged vibration levels measured on the test core for in-plane, out-of-plane and both directions.
The averaged vibration levels for in-plane, out-of-plane and both directions are plotted in Fig. 8. The curve for
the averaged in-plane vibration levels displays a shape with distinct parts over the frequency range. The curve
starts with a rather linear increase of level from 400 Hz to 1300 Hz followed by a steep rise and decrease of
level between 1300 Hz and 1600 Hz implied by the peaks at 1380 Hz and 1520 Hz. The curve then continues
with some level variations about a line more or less horizontal up to a new rise due to a peak at 2340 Hz.
The averaged out-of-plane vibrations present higher levels at low frequencies and generally lower levels at
higher frequencies than the in-plane vibrations. As a result, the total average of the vibration levels is dominated
mostly by the contribution of the in-plane vibrations from approximately 600 Hz, as it can be observed in Fig. 8.
The out-of-plane vibrations provide a contribution to the total vibration average at low frequencies and are the
result of global modes which occur because the core stiffness in X-direction is significantly lower than in Y and
Z-directions.
In general, the larger contribution of in-plane vibrations to the core motion can be attributed to the
magnetostrictive excitation acting mainly in the YZ-plane. In addition, the clamping plates may have a
restraining effect on the vibration transmission in X-direction.
Figure 9. Averaged vibration levels for in-plane and out-of-plane directions and sound power levels radiated by
the test core.
The sound power level determined from the sound pressure measurements made with the microphone mounted
on the rotating boom are plotted in Fig. 9. It can be observed that the sound power radiated by the core follows
closely the same trend as the averaged vibration level, as clearly displayed in Fig. 9. The discrepancies observed
between sound power and vibration levels can be attributed especially at low frequencies to the unknown
radiation efficiency which is lower than one at frequencies lower than 800 Hz. In addition, the limited number
30
40
50
60
70
80
90
100
110
200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100 2200 2300 2400 2500
Vib
rati
on
Le
ve
ls d
B R
ef.
5e
-8 m
/s
Frequency Hz
Averaged Velocity Levels on Core
Averaged Velocity In-Plane Averaged Velocity Out-of-Plane Averaged Velocity Both Directions
30
40
50
60
70
80
90
100
110
200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100 2200 2300 2400 2500
Le
ve
ls d
B P
ea
k
Frequency Hz
Sound Power and Averaged Velocity Levels
Averaged Velocity In-Plane and Out-of-Plane Ref. 8e-5 m/s Sound Power dB Ref. 1 pW
of measurement points used to measure vibrations on the core may have an impact on the averaged value of the
vibration levels at certain frequencies.
Figure 10. Out-of-plane motion at 500 Hz with
middle limb bending in X-direction.
Figure 11. In-plane motion at 2380 Hz with side limb
bending in YZ-plane.
An Operational Deflection Shape (ODS) was performed by using the vibration signals acquired by the
accelerometers at the main acoustic frequencies corresponding in general to the main peaks in Fig. 9 in order to
identify the motion patterns of the core. For example, two clear motion shapes, one out-of-plane at 500 Hz and
another one in-plane at 2380 Hz, are shown as fixed pictures in Fig. 10 and 11, respectively.
To confirm the presence of global mode shapes, an Experimental Modal Analysis (EMA) was carried out on the
test core suspended by elastic ropes. The mechanical excitation was provided by an electrodynamic shaker
attached to the laminations at an oblique angle of 45º, as shown in Fig. 12.
The input excitation from the shaker in the YZ-plane was measured by a force transducer and the vibration
levels generated by the shaker excitation were recorded in X, Y and Z-directions by a roving tri-axial
accelerometer positioned at 60 different points on the test core, as shown in Fig. 13.
Figure 12. Electrodynamic shaker attached to the core
laminations by a rod at an angle of 45º to provide the
mechanical excitation.
Figure
13. Shaker excitation measured by force
transducer and vibrations recorded by tri-
axial accelerometer.
The measurement method is designated as the Single Input Multiple Output (SIMO) procedure since the
excitation is applied at one fixed point and the vibration levels are recorded at 60 different measurement points
in three directions. As a result, 180 Frequency Response Functions (FRF) were measured as accelerances
defined as acceleration through force. The overlaid traces of all measured FRF are presented in Fig. 14. At some
frequencies, distinct peaks clearly indicate the presence of structural modes.
The formal EMA enabled the eigenfrequencies and the mode shapes of the test core to be identified
systematically. In fact, a great number of flexible modes could be found: 65 global and local modes were
Electrodynamic
Shaker
Force
Transducer
Tri-Axial
Accelerometer
detected in the frequency range from 200 Hz to 3200 Hz. For example, the deformation shapes in Fig. 10 and 11
could be related to two mode shapes presenting similar vibration patterns at the same frequencies.
Figure 14. Overlaid traces of the 180 FRF measured on the test core for the EMA.
The deformation shapes identified by ODS on the test core clearly corresponded to particular global mode
shapes found by performing the proper EMA. As a matter of fact, the peaks displayed by the plots for sound
power and the averaged vibration level, as shown in Fig. 9, could be associated with certain specific modes
obtained by the EMA.
The sound power radiated by the test core is strongly coupled to the averaged vibration velocity of the core.
Moreover, the vibration levels of the core excited by a realistic magnetic field are influenced by the structural
mode shapes of the test core. Therefore, the magnetostrictive excitation at the fundamental acoustic frequency,
but even at the corresponding harmonics for higher magnetic flux densities, can efficiently interact with the
mechanical resonances of the core to cause increased sound levels.
If the scale model is assumed to be roughly a one-fourth model of the core for a typical medium power
transformer, the first main resonance for the corresponding full-scale model would thus be found in the vicinity
of 350 Hz. This resonance frequency would be rather close to the frequency of 360 Hz corresponding to the
third harmonic of the magnetostrictive force generated by the typical network frequency of 60 Hz. In fact, the
resonance peak is fairly wide and may cause an amplification of the sound level at 360 Hz. This observation
highlights the need of a procedure at the design stage to identify and avoid core resonances.
The experimental studies carried out in laboratory conditions on the core scale model provide some insight into
the general mechanisms leading to transformer no-load noise. Moreover, the experimental data can be used to
validate and calibrate the models used to predict the core noise of power transformers.
3. PREDICTION MODEL AND SIMULATION RESULTS
3.1 Electromagnetic Model
At no-load condition, an alternating voltage is applied to the phase windings on one side of the transformer
while the windings on the other side (usually called primary side) are left open. The excited windings will carry
the magnetizing current and a magnetic flux will be induced in the core.
To obtain the magnetic flux density in the core at no-load, Maxwell’s equations are used with the additional
expressions.
𝐁 = 𝜇𝐇 (2)
where μ is the magnetic permeability in the core material and H the magnetic field strength.
𝐉 = 𝜎𝐄 (3)
where J is the current density and E the electric field density.
Out-of-plane modes
from 400 Hz to 600 Hz
In-plane modes
near 1400 Hz
In-plane modes
near 2300 Hz
Figure 15. Permeabilities in rolling and cross-rolling directions for the core steel sheets.
The resulting magnetic flux density B induced by the total currents (Ampere turns) of the windings can be
expressed as
B = ∇ × A (4)
where A is the magnetic vector potential.
Figure 16. Magnetic flux density distribution in Tesla for a three phase transformer core at one specific point on
wave.
By combining Maxwell’s equations and the vector potential formulation, the governing equation for the
magnetic field is given by the relationship
𝜎𝜕𝐀
𝜕𝑡+ ∇ × (
∇×𝐀
𝜇) = 𝐉 (5)
The model is investigated in two dimensions by considering one single core package at a time. The core is
surrounded by air. A magnetically insulating boundary condition is used on the outer boundary imposing the
constraint
𝐧 ∙ 𝐇 = 0 (6)
where the normal component of the magnetic field is set to zero.
Experimental data for a specific steel type are used in the model to describe the magnetic properties of the core
material. The core steel sheets are magnetically oriented and the permeabilities in the rolling and cross-rolling
directions present different non-linear virgin curves, as shown in Fig. 15.
The excitation of the simulation model is provided by voltages shifted by a phase of 120º applied to the
windings to simulate a three phase transformer. The analysis is then carried out in the time domain to obtain a
magnetic field density in the core, as displayed in Fig. 16.
Figure 17. Magnetostriction strains with respect to induction in rolling direction for the fundamental frequency
and nine harmonics.
By applying a Fast Fourier Transform to the time signals, the main frequency components of the magnetic flux
density in rolling and cross rolling directions can be calculated at each node point of the mesh. The fundamental
magnetic flux density components at each node can then be systematically associated with magnetostrictive
strains provided in tables gathering experimental data specific for each steel grade, as shown in Fig. 17.
Those values relating the magnetic flux density and the magnetostrictive strains are used as input data to the
mechanical simulations performed in the frequency domain.
3.2 Mechanical Model
In the mechanical model, the linear elasticity equations are used to describe the in-plane motion of the core
implied by the magnetostrictive strains. The Navier’s equation is given by the expression
∇ ∙ 𝜎 = −Fv (7)
where σ is the stress tensor and Fv the volume forces.
The stress tensor σ is a function of the strain tensor εmech as given in Hooke's law
𝜎 = 𝐸 ∙ 𝜀𝑚𝑒𝑐ℎ (8)
where E is the elasticity tensor.
Since the electric steel has different elasticity properties in rolling and cross rolling directions, the elasticity
tensor is specified for an orthotropic material. Then, the magnetostriction strain is introduced into Hooke’s law
𝜎 = 𝐸 ∙ (𝜀𝑚𝑒𝑐ℎ − 𝜀𝑚𝑎𝑔) (9)
where εmag is the magnetostriction strain.
The back-coupling effects corresponding to magnetostriction strain influencing magnetic field, are considered as
negligible.
In a finite element formulation, displacements are approximated in elements as
𝑢 = 𝑁𝑁𝑢𝑁 (10)
where NN is the array of interpolating functions and uN the nodal displacement.
The strain can thus be expressed as
𝜀 = 𝐷𝑁𝑢𝑁 (11)
where DN is the spatial derivative of NN.
The governing equation for the motion of the core subjected to magnetostrictive strain is given by the
relationship
𝑀�̈�𝑁 + 𝐶�̇�𝑁 + 𝐾𝑢𝑁 = 𝑓𝑁 (12)
where M, C and K represent the inertia, damping and stiffness terms, respectively and the term fN gives the load
array containing the forces applied at the nodes. In the present model, the core is excited by the magnetostrictive
strains. This yields
𝑓𝑁 = ∫𝐵𝑇𝐸𝜀𝑚𝑎𝑔 𝑑𝑉 (13)
The displacement field for each selected core induction can be finally obtained from the magnetostrictive strains
and for instance be analyzed in detail in the vicinity of the fundamental acoustic frequency as well as around the
first acoustic harmonics.
To obtain a better understanding of the core motion, a modal analysis is carried out with a suitable software to
calculate the core resonances and, as an example, an eigenmode is identified at 2390 Hz, as displayed in Fig. 18.
The resonance at 2390 Hz corresponds to the clear peak displayed in Fig. 9. This resonance could be easily
excited by the magnetostrictive strains and causes the high vibration amplification at this frequency, as shown in
Fig. 9.
Figure 18. Mechanical resonance of the test core at 2390 Hz obtained by modal analysis.
3.3 Acoustic Model
The core motion generates sound waves propagating in a surrounding fluid, such as air or oil, as described by
the equation
∇2𝑝 −1
𝑐2
𝜕2𝑝
𝜕𝑡2= 0 (14)
where p is the acoustic pressure and c the speed of sound. The fluid is considered here as non-viscous.
The fluid loading of the core gives a first boundary condition and is defined as
fp = −𝑝n (15)
where p is the fluid pressure and n a vector normal to the structure surface.
The second boundary condition is given by the particle velocity continuity between structure and fluid, thus
yielding
vf =∂u
𝜕𝑡∙n (16)
where vf is the particle velocity of the fluid and u the displacement vector.
The relationship between pressure and velocity is obtained by using the linear equation of motion yielding
𝜌𝜕u
𝜕𝑡= −∇𝑝 (17)
where ρ gives the fluid density and u the displacement vector.
For the acoustic analysis, the 2D-model used for the mechanical simulations has to be extruded to a 3D-model.
In addition, perfectly matched layers (PML) are implemented to create a free space for sound propagation, as
shown in Fig. 19.
Figure 19. Sound pressure field radiated by the vibrating core given at a specific frequency.
4. TESTING ON FULL SCALE TRANSFORMER CORE
Measurements were performed on the core of a full scale transformer in order to detect potential effects of
resonances on the vibration levels. Sensors were mounted on the core of a power transformer to record the
vibration levels during Factory Acceptance Test. The 117 MVA transformer selected for the tests was
manufactured in a fairly standard way with a three limb core. Seven accelerometers were attached at different
positions on the upper yoke, as shown in Fig. 20.
Figure 20. Seven accelerometers mounted with magnets on the upper yoke of a 117 MVA transformer.
The vibration signals from the accelerometers were acquired simultaneously when the transformer was operated
at nominal voltage (100% Unom) in no-load conditions at 50 Hz.
Figure 21. Zoom on acceleration spectrum measured by one sensor placed on the upper yoke.
Accelerometers on Core
Potential resonance
near 480 Hz
The measurements performed on the transformer operating at no-load provided vibration data from the core
immersed in oil. For example, a potential resonance near 480 Hz implying an increase of the vibration level can
be observed clearly in the acceleration spectrum acquired on the operating core, as shown in Fig. 21.
Those valuable vibration data will be used to validate and calibrate the prediction tools under development.
5. CONCLUSION
Sound and vibration measurements were performed on the scale model of a transformer core in order to
investigate in laboratory conditions the multiphysics phenomena implying no-load noise. The main resonance
frequencies of the core could be identified experimentally by performing a frequency sweep of the input voltage.
Furthermore, prediction models based on a finite element formulation were developed to describe the complex
interactions in the core structure and the coupling of the physical fields. Measurements carried out on a full scale
transformer indicate the influence of resonances on the vibration levels of the core.
6. REFERENCES
1. G. Ramsis, J. Anger, D. Chu, The Sound of Silence Designing and Producing Silent Transformers, ABB
Review, 2/2008, Pages 47-51 (2008)
2. IEC 60076-10-1, Power Transformers – Part 10 – 1: Determination of Sound Levels – Application Guide
(2005)
3. B. Weiser, H. Pfützner, J. Anger, Relevance of Magnetostriction and Forces for the Generation of Audible
Noise of Transformer Cores, IEEE Transactions on Magnetics, Volume 36, Pages 3759-3777 (2000)
4. M. Rausch, M. Kaltenbacher, H. Landes, R. Lerch, J. Anger, J. Gerth and P. Boss, Combination of Finite
and Boundary Element Methods in Investigation and Prediction of Load-Controlled Noise of Power
Transformers, Journal of Sound and Vibration, Volume 250, Pages 323-338 (2002)
5. R. Haettel, M. Kavasoglu and C. Ploetner, Prediction of Transformer Load Noise, Proceedings of the
Comsol Conference, Paris (2010)
6. E. Reiplinger, Assessment of Grain Oriented Transformer Sheets with Respect to Transformer Noise,
Journal of Magnetism and Magnetics Materials 21 (1980)
7. J. Anger and A. Daneryd, Noise in Power Transformers – Model for Generation, Transmission and
Propagation, Proceedings of the Conference Noise in Machines, UK Magnetics Society, Wolfson Centre for
magnetics, Cardiff University (2009)
8. M. Mizokami, M. Yabumoto and Y. Okazaki, Vibration Analysis of a Three-Phase Model Transformer
Core, Electrical Engineering in Japan, Vol. 119, No. 1 (1997)
7. ACKNOWLEDGEMENTS
C.-G. Johansson, M. Kavasoglu and C. Lagerström are greatly acknowledged for their contribution to this work.