Investigation of acoustic noise sources in medium frequency, medium voltage transformers

Transcription

1 Investigation of acoustic noise sources in medium frequency, medium voltage transformers Peng Shuai and Jürgen Biela, Member, IEEE, This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of ETH Zürich s products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to pubs-permission@ieee.org. By choosing to view this document you agree to all provisions of the copyright laws protecting it.

2 Investigation of Acoustic Noise Sources in Medium Frequency, Medium Voltage Transformers Peng Shuai, Jürgen Biela Laboratory for High Power Electronic Systems, ETH Zürich Keywords <<Transformer>>, <<Acoustic noise>> Abstract Medium voltage, medium frequency transformers (MFTs) are much smaller in size and weight compared to conventional low frequency transformers. The MFTs are very attractive for applications where full control of the power flow and high power density are required, such as power electronic interfaces in smart grids and traction converter systems. Due to the limitation of high voltage semiconductor switches, the MFTs are usually operated in the khz range, which results in acoustic noise radiated from the transformer. In this paper, the origins of acoustic noise associated with MFTs are investigated based on vibration and acoustic measurements. The investigation focuses on the core materials widely used for MFTs, i.e. the amorphous and the nanocrystalline alloy. Experimental results show that the nanocrystalline core features lower level of vibration and acoustic noise emission than the amorphous core. Comparing core shapes, the toroidal core has better vibration and acoustic performance than the U-shape core. The existence of air gap in case of a cut core leads to excessive vibrations and therefore higher acoustic noise level due to increased magnetic forces compared with uncut core. Based on analysis of measurement results, recommendations for low noise MFT design are proposed. Introduction In modern electric power systems, renewable energy sources are increasingly integrated. Since these power sources are inherently fluctuating, power electronic interfaces are essential for dynamic control of power flow and power quality improvement. The typical configuration of the interfaces are usually based on DC-DC converters isolated with medium frequency transformers as shown in Fig. 1. As the medium frequency power conversion systems are operating in the khz range, significant reduction of weight and size can be achieved compared to conventional line frequency transformers. For medium voltage high Figure 1: Medium frequency power conversion system. Main specifications of designed DAB converters are: MV DC-link voltage 2400 V, LV DC-link voltage 400 V, nominal power 25 kw and switching frequency 4 khz. Figure 2: Volume reduction for optimal MFT design with VITROPERM500F by increasing frequency from 1 khz to 16 khz.

3 power applications, the DC-DC converters are usually cascaded and connected to the medium voltage (MV) grid. The design challenges are mainly high efficiency, thermal management and high insulation voltage in case of reduced size and higher operating frequency. The optimal design of the MFT for a DC-DC converter has been discussed in [1], where the isolation voltage level aims for 24 kv. The volume dependency on switching frequency as shown in Fig. 2 indicates that the volume reduction with increasing frequency is not significant above 4 khz. The benefit of reducing the size by increasing operating frequency is limited by isolation requirement and thermal stress above 4 khz. Furthermore, the switching losses of the converter increases significantly with the frequency, especially for the MV side switches. Therefore, in the considered design the operating frequency is selected as 4 khz, which is in the audible frequency range. In the past few decades, the acoustic noise emission from electromagnetic components has been constantly investigated. There exist extensive studies for electrical machines, e.g. [2, 3, 4] and a large number of research focusing on power transformers operating at line frequency can be found [5, 6, 7]. However, in the medium frequency range, literature regarding this topic is rare, especially for MFTs. The investigation on acoustic performance of electromagnetic components operating in the medium frequency range are presented in [8, 9, 10, 11, 12, 13] for inductors. In [8], a noise reduction method is proposed for inductors with distributed air gaps in iron cores by handling the frequency of free modes vibration. In [9], the noise reduction of electrical steel based inductors is achieved by reducing the flux density and modification of the air gap filler. The measurements of sound pressure level (SPL) are performed for inductors built with grain-oriented steel and ferrite cores under diverse excitation conditions in [10]. The magnetostriction and acoustic noise of Si-Fe powder compressed cores are measured in [11] for inductors employed in photovoltaic generators. In the extensive studies carried out for single-phase and three-phase inductors in [12, 13], the authors propose to reduce the noise by improvement of the air gap distribution in grain-oriented steel core and by optimizing the hardness of the air gap filler. Regarding transformers, in [14], the SPL of several spacecraft power transformers built with Si-Fe and Ni-Fe-Mo cores are measured and compared. The study is focusing on the influences of the core types and air gap and recommendations for low noise transformer design are proposed. The dependency of acoustic noise on frequency and magneto-mechanical resonance is investigated in [15] for 3-phase 3- limb transformer. Finally, the acoustic characteristics of 3-phase transformers built with amorphous cores operating at 60 Hz are introduced in [16, 17]. The work focuses on the influence of bending structure of cores on vibration and acoustic noise. As aforementioned, acoustic noise emission from electromagnetic components operated in medium frequency range is still a problem to be dealt with. A model for predicting the acoustic noise level needs to be established and better methods to reduce the noise need to be found. Regarding MFT design, the significance of each acoustic noise source is still unclear and the vibration related properties of core materials used for MFTs are seldom investigated. Therefore, in this paper the acoustic noise sources are investigated based on vibration and acoustic measurements with focus on amorphous and nanocrystalline tape wound cores, which are widely used for MFT design in high power applications. Sources of acoustic noise in MFTs The electromagnetic forces induce vibrations in the transformer structure which cause vibrations of particles in the surrounding air. These vibrations produce the acoustic waves which propagates through medium (usually air) to the ears and generate acoustic noise. In some cases, noise may also be caused by the auxiliary equipments such as fans and oil pumps, which is out of the scope of this paper. In MFTs, the structural vibrations are excited by three types of electromagnetic forces: Maxwell force, magnetostrictive force and Lorentz force. The Maxwell force is acting on the boundaries (surfaces) between two magnetic media with different magnetic properties (reluctivity), which is typically associated with the air gap regions and the joints of cores in transformers. The Maxwell force can be calculated by the surface integration of the local force density based on Maxwell stress tensor as F = ( H ( B n) 1 2 ( H B) n) ds, where F is the force, H is the magnetic field strength, B is the magnetic flux density and n is the vector in normal direction to the surface S. To reduce the eddy current losses, the magnetic cores of MFTs are usually constructed with laminated sheets. The effects of the Maxwell force at the joints of core lamination sheets and between the sheets are introduced in [5]. In case of MFT, tape wound cut cores are widely used which normally do not have any joints. Accordingly, three mechanisms due to Maxwell force can be assumed for U-shape tape wound core (cf. Fig. 3): repulsive forces between lamination sheets; in-plane attractive forces between the sheet ends (air gap region); off-plane attractive forces between the lamination sheets near the air gaps. The Magnetostrictive force is caused by magnetostriction (MS), which represents the deformation of magnetic materials under the effect of magnetic field. In case of core materials for transformer design, the major deformations are due to the Joule magnetostriction [2]. This effect is anisotropic and causes an elongation or contraction of the lamination sheets in the direction of the applied magnetic field. At the same time, shrink or expansion can be observed in the orthogonal directions to the magnetic field, as the total volume of the sheet remains constant. Magnetostriction is quantified as the mechanical strain λ Considering 5 modules cascaded and connect to 6.6 kv AC grid on MV side. For safety reason, the isolation voltage is defined as two times of the MV DC-link voltage level.

4 Figure 3: Mechanisms of Maxwell force in tape wound cut core. Figure 4: Mechanisms of magnetostriction in tape wound cut core. Table I: In-plane saturation magnetostriction of core materials without mechanical stress Material λ s [µm/m] 2605SA1 (Amorphous) 27 VITROPERM500F (Nanocrystalline) % Silicon steel % Silicon steel 0 TDK PE90 (ferrite) -0.6 a a Negative value means contraction. The quantification of magnetostriction for ferrite is related to the crystal directions. induced in the material in the absence of a restraint as λ = l/l(µm/m or ppm), where l is the length of the sample and l is the change of the length. In general, λ is proportional to the square of magnetic flux density B and reaches the maximum value at the saturation point of the material. This maximum value is defined as the saturation magnetostriction λ s, which is a material parameter. Typical values of λ s for several core materials are listed in Table I. However, it should be noticed that magnetostriction is also dependent on the mechanical stress. The Lorentz forces are acting on the current-conducting windings and the volume force density f v is calculated as f v = J B, where J is the current density in the conductor and B is the magnetic flux density. In literature, the contribution of Lorentz force is usually considered to be less significant than the other two sources. Both magnetostriction and Maxwell force are assumed to be more pronounced sources of acoustic noise emission [5] associated with transformers. However, it is concluded in [8] that the winding is always a source of mechanical stress, so it can not be ignored in the mechanical analysis. The impact of the Lorentz force on vibrations will be discussed later in this paper based on simulation and measurement results. At the air gap region, the in-plane attractive forces pulls the sheet ends in two core halves towards each other. This mechanism is more significant than the repulsive forces between lamination sheets. The inplane magnetostriction causes the elongation (or contraction) of the lamination sheets along the magnetic flux line. Due to the magnetic fringing field near the air gap, the off-plane attractive forces arise and act together with the off-plane magnetostriction on the sheet ends. As illustrated in Fig. 3 and Fig. 4, these two mechanisms enhance with each other and cause expansions of the sheet ends. Therefore, the air gap may lead to excessive vibrations in MFTs due to increased Maxwell and magnetostriction forces. Moreover, a coincidence of the vibration frequencies with the eigenfrequencies of the transformer structure will amplify the vibration and therefore generate more acoustic noise. In order to identify the contribution of each source as well as the relevance of electromagnetic and mechanical properties of the transformer to the acoustic noise emission, both vibration and acoustic measurements are performed on several magnetic cores. The details of the measurements are introduced in the following sections. Measurement setup In this work, the measurement is mainly performed on a pair of cut and uncut standard nanocrystalline cores (VITROPERM500F T60102-L2157-W159), which is selected for the MFT prototype design in [1]. For comparison, an amorphous core and a ferrite core with similar size are also chosen for measurement. The size of these cores are shown in Fig. 5 and listed in Table II. The measurement setup as shown in Fig. 6 has been built for both the vibration and the acoustic noise measurements. The excitation voltage/current for the test unit (TU) is generated by a signal generator, amplified with a power amplifier and then fed to the primary winding of the TU. A capacitor bank with a total capacitance of 5 mf is connected in series with the TU to decouple the DC component which might be induced by the power amplifier. To control the magnetic flux density, the voltage on the secondary winding of the TU is measured by the digital multimeter (NI PXI-4071) and sent to PC as feedback signal to regulate the output voltage of the signal generator by using LabVIEW software. The vibration of TU is measured with a laser scanning vibrometer (Polytec PSV-400) controlled with controller (Polytec OFV-5000), which allows a non-contact measurement without altering the structure of TU as in case of using a contact type accelerometer. The TU is placed on a vibration isolated table to

5 Table II: Dimensions of cores for measurement Dimensions[mm] Core VAC VTROPERM500F T60102-L2157-W159 Metglas 2605SA1 PS0509CA Kaschke K2008 U93/30/76 A B C D W H Figure 5: Symbol of core dimensions. avoid the disturbance from the environment. The measured signal is decoded to velocity and acquired by the data recorder and sent to PC for further analysis. The acoustic measurement is done by measuring the sound pressure level through a single 1/4 inch microphone with integrated pre-amplifier (G.R.A.S. 40PH), where the measured signal is acquired by the dynamic signal analyzer (NI PXI-4462) and then sent to PC. The TU is located in the center of an anechoic room, the size of which is much larger than the TU. The microphone is located 1 m away from the surface of the TU. The vibration measurements are performed on three surface areas on the magnetic cores as indicated in Fig. 7, i.e. front, side and top. Accordingly, the acoustic measurement are also performed from these directions separately. During both vibration and acoustic measurements, the sensor (vibrometer or microphone) is fixed at the same location while the TU is moved in order to measure each surface/direction. Analysis of measurement results The vibration and acoustic measurement are performed separately on different cores excited with sinusoidal voltage/current. The results are analyzed and discussed below. Eigenfrequencies and mode shapes of uncut core To investigate the eigenmode of the magnetic cores, the vibration of the VITROPERM uncut core is measured by means of a frequency sweep. The core is hanging as shown in Fig. 9 to enable a free vibration condition. The core is turned over for 90 degrees to measure the top surface. The excitation voltage on the core feeding to a 4-turn winding is a chirp signal amplified 20 times with the power amplifier. To avoid the LC-resonance between the capacitor bank and the magnetizing inductance, the frequency sweep is started from 300 Hz and performed up to 20 khz. Since the flux density is lower at higher frequency with the same amplitude of exciting voltage, the induced magnetic force is also reduced with the increase of the frequency. In order to have a clearer mechanical response of the core in the whole measuring frequency range, the frequency sweep is divided into 3 sub ranges with different amplitude of excitation voltage: from 300 Hz to 5 khz with 2 V, from 5 khz to 10 khz with 20 V and from 10 khz to 20 khz with 40 V. The scanning vibrometer measures a set of points located on the measured surfaces. The vibration in the direction perpendicular to the surface at these points are measured. In Fig. 8, the frequency spectrum of measured average velocity of each selected surface area are shown together. To have a better visibility, the magnitude of velocity from 300 Hz to 5 khz and from 5 khz to 10 khz are multiplied by a factor of 20 and 2 respectively. As can be seen from the frequency response in the figure, several eigenmodes of this core are within the audible frequency range and may coincide with the frequency of the excitation voltage Figure 6: Schematic of measurement setup. Figure 7: Surface areas of vibration measurement.

6 Figure 8: Frequency spectrum of measured average surface velocity of VIT- ROPERM uncut core by means of frequency sweep. Figure 9: Vibration measurement setup for eigenmode analysis. Figure 10: First 3 dominant eigenmodes of VITROPERM uncut core identified from vibration measurement. and its harmonics and result in mechanical resonance. The comparison of the measured amplitude on different surfaces evidently shows that above 5 khz the top surface has the most significant vibration, followed by the side surface while the vibration on the front surface is relative weak. This indicates that the tape wound core is more prone to vibrate in the directions perpendicular to the lamination layers. Since the tape wound core is composed of magnetic material together with epoxy resin as isolation between the lamination layers as well as varnish outside of the core, the mechanical properties of the core is anisotropic. As introduced e.g. in [4], the laminated core can be modeled as orthotropic structure where the in-plane (lamination layer) material property is assumed to be the same while the material exhibits different properties in the directions perpendicular to the plane. However, an accurate model requires accurate material parameters which usually need to be determined by experimental measurements. In case of the measured core, the vibration measured on side and top surfaces demonstrates the out-ofplane behavior while the vibration on front surface is related to the in-plane properties. As illustrated in Fig. 10 for the first 3 dominant eigenmodes identified by measurement, similar mode shapes are observed on the side and top surfaces which are different as the mode shapes on the front surface. The first and third mode shapes illustrate the flexure deformation of the core limbs and yokes while the second mode shape shows the torsional deformation. At higher frequency, the mode shapes are mixed with low order vibration modes and can not be clearly distinguished.

7 Figure 11: Maxwell force and Lorentz forces distribution at the same magnetic flux density in inductors and transformers with or without air gaps. The inductor winding has 4 turns and the transformer winding has 40 turns (turns ratio 1:1). If no air gap exists, the Maxwell force and Lorentz forces are comparable. In case that an air gap is present, large attractive forces are acting on the two core halves and pull them together. The flux density in the middle of core limbs is approximately 0.5 T in all cases with a peak value at around 1.5 T at the corners. Figure 12: Comparison of measured velocity of the winding and the VIT- ROPERM uncut core excited to 0.5 T at 4 khz on the front surface. Contribution of Lorentz force As mentioned before, the Lorentz force is considered to be less significant than the other two sources of acoustic noise. However, the simulation results shown in Fig. 11 indicate that in case of uncut core, the Maxwell force and Lorentz force are actually comparable. To compare the vibration of winding and core simultaneously during operation, the VITROPERM uncut core is excited to 0.5 T with a 4-turn winding fed with 4 khz sinusoidal voltage. This induces the magnetizing current with the peak value of approximately 0.6 A in the winding. The vibration measurement is performed on part of the front surface of the core limb and on the surface of one turn of the winding. As the measurement result shows, in this case the measured surface velocity of the winding and the core are in the same range, which is similar as the simulation result shown in Fig. 11 (a). As will be revealed in following sections, the vibration on the front surface of the core is relative small compare to the other two surfaces. In case of a cut core excited to the same flux density, the magnetizing current will be larger, and the Lorentz force acting on winding will also increase. Measurement of a pure winding fed with 20 A peak sinusoidal current shows that the maximum surface velocity can reach 300 µm/s. However, compared with the core surface, on which some areas could reach a velocity of over 10 mm/s, this value is still much lower. Therefore, the vibration of winding due to Lorentz force can be considered to have very limited contribution to acoustic noise. Comparison of various core materials and shapes To compare the vibration and acoustic performance of magnetic cores with various materials and shapes, several cores are measured under same excitation conditions. During the measurement, all the cores are placed directly on the table without any additional mechanical fixation. In Table III, the measurement results of VITROPERM cut and uncut cores together with the Metglas cut core are compared. All the cores are excited with 4 khz sinusoidal voltage to the flux density levels given in the table. Since only the harmonics in audible range are of interest, all the results are RMS values calculated with the 1st to 5th harmonics. The comparison of both measured surface velocity and sound pressure among different surfaces / directions for each core indicate similar results: the most significant vibration exists on the top surface of the core while the measured acoustic noise emission in the normal direction to this surface is also the largest. The side surface is subject to less vibration as well as noise emission compared to the top surface but more than the front surface. The relationship of degree of vibration / noise emission among the measured surfaces, i.e. top surface > side surface > front surface, is consistent with the results obtained by frequency sweep measurement for the VITROPERM uncut core. The comparison between the VITROPERM cut and uncut cores indicates that with air gaps, vibrations and noise emission increase by a factor of more than 10. The increase of vibration is more significant on the side and top surfaces, i.e. in the direction perpendicular to the lamination layers. In case of cut cores, the difference of vibration intensity between the perpendicular and parallel directions of the lamination layers is more apparent due to the increased Maxwell force induced by the air gaps. It should be pointed out that the measured surface velocity exclusively indicate the vibration on the measured surfaces. On the other hand, the measured sound pressure in one direction is mainly radiated from the surface perpendicular to this direction, but the noise emission from other surfaces also has impact and can not be completely excluded from the measurement results. In Fig. 13, the areas with relative high surface velocity on the measured surfaces of each core at various flux density levels are indicated. Here the vibration of the highlighted areas are not in phase, i.e. the

8 deformation of these areas do not appear at the same time. The numbers only represent the highest amplitudes of measured surface velocity in the areas. On the top surface, the area with largest vibration is located in the middle of the measured surface, which is the same for all three cores. This is related to the bending mode of the yoke as shown in Fig. 10. On the front surface, the intensive vibration appears on the corner as shown for VITROPERM cores and is related to the torsion mode. View on side surface, the bending areas (near the corners) are subject to high degrees of vibration which is related to the bending mode of the core limb. In case of VITROPERM cut core, the air gaps amplify the vibrations in these areas. The air gaps of Metglas cut core directly cause strong vibrations in the nearby regions similar as shown for front surface of VITROPERM cut core. It is noticed that the vibration measured on the upper side and lower side half cores are not symmetric. The reason is that the vibration of the lower side half core is influenced by the gravity of the upper side half core and also the friction with the surface of the table. Also, this phenomenon can be observed by the measurement results of other cut cores which will be shown later. For comparison, the measurement results of the ferrite core at 0.2 T is also listed in Table III. In spite of the non-laminated structure and the right angle at the joints of core limbs and yokes (rectangular shape), the ferrite core shows similar relationship of vibration intensity and noise emission on different surfaces as the U-shape lamination core. At the same flux density level, ferrite core presents less vibration/noise emission than VITROPERM and Metglas cut cores. For further comparison, a pair of cut and uncut toroidal cores made of VITROVAC 6025F amorphous alloy together with the ferrite core are measured at 0.4 T. The dimensions and the measured areas of the toroidal cores are shown in Fig. 14 and the results are given in Table IV and Fig. 15. As can be seen, the areas with high degrees of vibration on the surface of measured ferrite core is quite similar as the other U-shape cores. In case of toroidal cores, there exists no part with large curvature as the bending region of the U-shape cores. As a result, the magnetic flux in toroidal cores is relative well distributed, which lead to relative equal distributions of local magnetic forces. Similarly, with laminated structure, the VITROVAC toroidal cores also present larger vibration / noise emission in the direction perpendicular to the lamination layers (the radial direction). Furthermore, the VITROVAC uncut toroidal Table III: Comparison of vibration and sound pressure measurement results for nanocrystalline and amorphous cores under excitation of 4 khz sinusoidal voltage to different flux density levels. Core Flux density [T] Average surface velocity [µm/s] Sound pressure [µpa] Front Side Top Front Side Top VITROPERM uncut VITROPERM cut Metglas Ferrite VITROPERM uncut VITROPERM cut Metglas VITROPERM uncut VITROPERM cut Metglas Figure 13: Areas with relative high velocity on measured surfaces of nanocrystalline and amorphous cores excited to different flux density levels with 4 khz sinusoidal voltage. The numbers indicate the points with maximum surface velocity (RMS value) in [µm/s] calculated with 1st to 5th harmonics.

9 core shows extremely low degree of vibration as well as low noise level while the air gaps of cut cores cause excessive vibration and noise emission. Frequency domain analysis The analysis of sound and vibration is often performed in frequency domain. Since the magnitude of both Maxwell force and magnetostriction are proportional to the square of flux density, the fundamental harmonic of vibration is twice of the excitation frequency. Theoretically, if the core is excited with a sinusoidal voltage of frequency f s, the magnetic flux has only one harmonic at f s in frequency spectrum. Therefore, the induced magnetic forces and then the vibration velocity as well as the radiated sound power have one harmonic at 2 f s in frequency spectrum. However, the excitation voltage usually contains harmonics. Furthermore, the mechanical response of the core is nonlinear. Therefore, additional harmonics of magnetic forces will be induced. The MFT is usually operated under excitation of rectangular voltage. The flux density is then with a quasi triangular waveform which contains a number of harmonics. If the frequency of one of these harmonics coincides with one of the mechanical resonant frequencies of the transformer structure, additional vibration as well as noise emission can be induced. Although the excitation voltage is expected to be purely sinusoidal, a small portion of harmonics still exist during the measurement in this work, which maybe generated from the power amplifier. As an example, the measured waveform of excitation voltage and the square of the voltage waveforms as well as their frequency spectra are shown in Fig. 16. As can be seen, both the voltage and its square contain several harmonics, where the fundamental harmonic (4 khz) and the 2nd harmonic are dominant in each waveform respectively. Consequently, the measured surface velocity and sound pressure also contain these harmonics. In Table V, the first 4 harmonics of the surface velocity and the sound pressure of the VITROPERM cut core under excitation with this voltage are listed. As example, the frequency spectra of average velocity and SPL of side surface are also shown in Fig. 16 (e) and (f) respectively. As expected, the 2nd harmonic has the most contribution to vibration and noise. Therefore, the operating frequency of MFT at half of the eigenfrequency of the structure should be absolutely avoided. If possible, by operating the transformer at over 10 khz, the dominant harmonic of magnetic forces will be out of the audible frequency range and the acoustic noise can be effectively reduced. Dependency of SPL on flux density and frequency In this section, the results of sound pressure measurements performed on the aforementioned cores under excitation with voltage of various frequencies and amplitudes are compared. In Fig. 17 and Fig. 18, the unweighted and A-weighted SPL averaged Table IV: Comparison of vibration and sound pressure measurement results for ferrite and VITROVAC cores under excitation of 4 khz sinusoidal voltage to flux density of 0.4 T. Core Average surface velocity [µm/s] SPL [µpa] Front Side Top Front Side Top Ferrite VITROVAC cut N/A N/A VITROVAC uncut N/A N/A Figure 14: Surface areas of vibration measurement for VIT- ROVAC toroidal core. Figure 15: Areas with relative high velocity on measured surfaces of ferrite and VITROVAC cores excited to flux density of 0.4 T with 4 khz sinusoidal voltage. The numbers indicate the points with maximum surface velocity (RMS value) in [µm/s] calculated with 1st to 5th harmonics.

10 Figure 16: Waveforms of measured excitation voltage (a), the square of the voltage (c) and their frequency spectra (b) & (d). The voltage is measured on the secondary winding (turns ratio is 1:1). Accordingly, frequency spectrum of measured average velocity (e) and SPL (f) on the side surface of VITROPERM cut core when excited to flux density of 0.5 T at 4 khz with this voltage. from measured results in three directions are shown in db values (referred to 20 µpa). The measured background noise of the anechoic room is 38.2 db unweighted or 30.4 db A-weighted. As expected, the increases of SPL with the flux density level can be observed on all the measured cores except for the VITROVAC uncut core. The reason is that the noise emission of this core is extremely low and can not be separated from the background noise. On the other hand, different cores exhibit different dependencies of SPLs on operating frequencies, which is related to the eigenfrequencies of each core. When excited at 1 khz and 2 khz, the two dominant harmonics of the noise are all in the most sensitive frequency range of human ears (1 khz to 6 khz), in which the A-weighting curve has a gain larger than 0 db. As can be seen in some of the results, the A-weighted SPL value may even be higher than the unweighted SPL value. If the transformer is operated over 3 khz, the second harmonic of the noise which is normally the dominant harmonic will be shifted in the less sensitive frequency range and lead to better performance in terms of acoustic noise. Naturally, further increase of operating frequency will lead to even lower noise, especially when over 10 khz as already explained in previous section. Unfortunately, the operating frequency of MFT is normally limited due to the switching losses of high voltage switches and / or requirements of efficiencies and power densities of the converter systems. Figure 17: Measured unweighted and A-weighted SPL of VITROPERM and Metglas cores under excitation of sinusoidal voltage of different frequencies (1 khz, 2 khz, 4 khz, 8 khz and 16 khz) to various flux density level.

11 Figure 18: Measured unweighted and A-weighted SPL of ferrite and VITROVAC cores under excitation of sinusoidal voltage of different frequencies (1 khz, 2 khz, 4 khz, 8 khz and 16 khz) to various flux density level. Table V: First 4 harmonics of measured surface velocity and sound pressure of VITROP- ERM cut core when excited to flux density of 0.5 T at 4 khz. Harmonics of Harmonics of Measurement surface velocity [µm/s] sound pressure [µpa] suface/direction 1st 2nd 3rd 4th 1st 2nd 3rd 4th Front Side Top Figure 19: Oval core with small curvature at the bending part. Conclusions In this paper, the acoustic noise sources in MFT for high power applications are investigated based on vibration and acoustic measurements. The measurements are performed on magnetic cores made of different materials and with various shapes with focus on the most widely used materials for MFT design. Measurement results confirmed that the winding has relative weak vibration compared to the core especially in case of cut cores and therefore can be considered to have minor contribution to acoustic noise. Regarding the core materials, ferrite features low noise but the application in MFT with requirement of high power density maybe limited due to the low saturation flux density. The nanocrystalline material VITROPERM500F has much better acoustic performance compared to amorphous material Metglas 2605SA1 and it also features other advantages, e.g. low loss density. Therefore, this material could be most suitable for low noise MFT design. Vibration measurements indicate that the direction perpendicular to the lamination layers of tape wound core is subject to higher vibration and more noise radiation compared to the direction parallel to the lamination layers. In case of the most popular U-shape cores for MFT application, the large curvature at the corners leads to intensive vibrations at the surfaces nearby which is related to the bending mode shapes of the core. On the other hand, a toroidal core has relative lower level of vibration and noise emission. The vibration measurement results confirm that the air gap can induce excessive vibration directly near the air gap region and enhance the vibrations near the bending parts. As a result, the cut core has much higher noise emission compared with its counterpart. The frequency spectra of the measured vibration and sound pressure show that the harmonic with double excitation frequency is dominant and therefore particular attention needs to be paid to avoid the frequency coincidence of this harmonic with the eigenmodes. Based on the analysis and experimental results, for a low noise MFT design, the nanocrystalline core with toroidal shape and without cutting would be preferred. The drawback is that the power density maybe reduced compared to a MFT constructed with the traditional U-shape core. As a compromise between acoustic performance and power density, an uncut oval core as shown in Fig. 19 maybe more suitable for MFT design. This core shape takes the advantage of toroidal core with small curvature at the bending part without increasing much of size compared to the U-shape core. However, this kind of core shape is not standard product and the availability is very limited. In future work, further investigation on the acoustic performance of this core shape is necessary.

12 Acknowledgment The authors would like to thank ECPE, the European Power Electronics Research Network, for financial support of the research project and VACUUMSCHMELZE GmbH for providing the core samples and valuable information about the core materials. References [1] P. Shuai and J. Biela, Design and optimization of medium frequency, medium voltage transformers, in 15th European Conference on Power Electronics and Applications (EPE), 2013, pp [2] A. Belahcen, Magnetoelasticity, magnetic forces and magnetostriction in electrical machines, Ph.D. dissertation, Helsinki University of Technology, [3] J. Roivainen, Unit-wave response-based modeling of electromechanical noise and vibration of electrical machines, Ph.D. dissertation, Helsinki University of technology, [4] M. van der Giet, Analysis of electromagnetic acoustic noise excitations: A contribution to low-noise design and to the auralization of electrical machines, Ph.D. dissertation, RWTH-Aachen University, [5] B. Weiser, H. Pfützner, and J. Anger, Relevance of magnetostriction and forces for the generation of audible noise of transformer cores, IEEE Transactions on Magnetics, vol. 36, no. 5, pp , [6] R. S. Masti, W. Desmet, and W. Heylen, On the influence of core laminations upon power transformer noise, in Proceedings of ISMA, 2004, pp [7] A. Moses, P. Anderson, T. Phophongviwat, and S. Tabrizi, Contribution of magnetostriction to transformer noise, in 45th International Universities Power Engineering Conference (UPEC), 2010, pp [8] O. Barre, B. Napame, M. Hecquet, and P. Brochet, Acoustic noise emitted by passive components in magnetic devices and design of a low-noise industrial inductor, COMPEL: The International Journal for Computation and Mathematics in Electrical and Electronic Engineering, vol. 27, no. 5, pp , [9] S. Schmitt, Acoustic noise of sheeted electrical steel inductors in PWM operation causes and mitigation, in 13th European Conference on Power Electronics and Applications (EPE), 2009, pp [10] J. Mühlethalter, M. Schubiger, U. Badstübner, and J. W. Kolar, Acoustic noise in inductive power components, in 15th European Conference on Power Electronics and Applications (EPE), 2013, pp [11] P. Jang and G. Choi, Acoustic noise characteristics and magnetostriction of Fe-Si powder cores, IEEE Transactions on Magnetics, vol. 48, no. 4, pp , [12] Y. Gao, K. Muramatsu, M. J. Hatim, K. Fujiwara, Y. Ishihara, S. Fukuchi, and T. Takahata, Design of a reactor driven by inverter power supply to reduce the noise considering electromagnetism and magnetostriction, IEEE Transactions on Magnetics, vol. 46, no. 6, pp , [13] Y. Gao, M. Nagata, K. Muramatsu, K. Fujiwara, Y. Ishihara, and S. Fukuchi, Noise reduction of a threephase reactor by optimization of gaps between cores considering electromagnetism and magnetostriction, IEEE Transactions on Magnetics, vol. 47, no. 10, pp , [14] A. Kelley, Measurement of spacecraft power transformer acoustic noise, IEEE Transactions on Magnetics, vol. 26, no. 1, pp , [15] Y. G. Yao, T. Phway, A. Moses, and F. Anayi, Magneto-mechanical resonance in a model 3-phase 3-limb transformer core under sinusoidal and pwm voltage excitation, IEEE Transactions on Magnetics, vol. 44, no. 11, pp , [16] Y.-H. Chang, C.-H. Hsu, H.-L. Chu, and C.-P. Tseng, Magnetomechanical vibrations of three-phase threeleg transformer with different amorphous-cored structures, IEEE Transactions on Magnetics, vol. 47, no. 10, pp , [17] Y.-H. Chang, C.-H. Hsu, H.-W. Lin, and C.-P. Tseng, Reducing audible noise for distribution transformer with HB1 amorphous core, Journal of Applied Physics, vol. 109, no. 7, p. 07A318, 2011.