Optimized Modeling Process for Air Core Reactors

using Finite Element Analysis

Henrique B. Zaninelli Edson C. Bortoni
Electric and Energy Systems Institute Electric and Energy Systems Institute
Itajubá Federal University - UNIFEI Itajubá Federal University - UNIFEI
Itajubá, Brazil Itajubá, Brazil
[email protected] [email protected]

Abstract— This paper presents an optimized modeling process various voltages and currents are applied to the conductors,
2021 IEEE Power & Energy Society General Meeting (PESGM) | 978-1-6654-0507-2/21/$31.00 ©2021 IEEE | DOI: 10.1109/PESGM46819.2021.9638197

for air core reactors using a finite element-based software and computing the energy stored in those fields. To evidence the
with the deployment of this process, proposes a correction for correspondence between the methods proposed in [2] and [3] a
proximity-effect losses calculation method. A 2-D equivalent full factorial design of experiments (DOE), with four factors
geometry method is proposed and the validation is performed by and three levels is performed.
running a full factorial design of experiments, screening four
design parameters of coils: winding diameter, winding height, The proper evaluation of eddy current winding losses,
number of strands and strand diameter, as these parameters including skin-effect losses and proximity-effect losses, has
significantly affect air core reactors’ two main design been subject of several studies because of their importance and
characteristics: inductance and power losses. The results of the as a general analytical solution is difficult to be found due to
finite element simulations are compared to the results of the complexity of the winding geometries, the interactions
analytical calculations using formulations well known by the among the conductors in the windings and the proper
literature. The optimized modeling provides very good distribution of the magnetic field. The Dowell’s method [4]-
consistency with the analytical calculations and allows for a [6], which consists on analytical solutions for equivalent foil
huge reduction in the simulation time in the finite element conductors and the Ferreira’s method [7]-[9], based on the
simulations as it is not necessary to model every and each strand Bessel-function, used to the analytical field solution of a
of the coil.
single isolated round conductor exposed to an external
Index Terms— Air core reactors, design of experiments, finite-
uniform magnetic field, are analyzed in detail in [10]-[11] as
element analysis, response surface method. their limitations are explored and improvements are provided
based on a 2-D finite element analysis (FEA), presenting a
I. INTRODUCTION very reasonable solution for windings with round conductors.
Air core reactors (ACR) are widely used in power systems The ACR can be manufactured with different types of
in several different series applications and shunt compensation conductive material (aluminum, copper) and conductors’
systems with voltage levels up to 500 kV. Applying this type arrangement like bars, single wires, stranded cables and flat
of equipment in extra high voltage levels demands very transposed cables (FTCs).
accurate calculation models to establish the proper design A comprehensive study for the eddy-current winding
characteristics like inductance, power losses and temperature. losses in stranded cables is presented in [12], where an
For the inductance, Grover [1] has provided the basic analogy between the earlier presented methods (for round
formulas and the main calculation methods for specific cases, wires) and the cables is proposed.
which had certain limitations regarding accuracy and As the demand for reactive power compensation has been
applicability for different geometries of coils. In [2] Fawzi increasing so has the demand for equipment which would
and Burke have proposed the application of the Bartky’s provide lower power losses. In that sense, the use of FTCs in
transform for solving the elliptical integrals that results from the construction of ACRs has been a good alternative to
the calculation of the self and mutual inductances for round comply with that demand.
coils allowing for very accurate results despite of the coils’
dimensions and arrangement. Many studies [13]-[14] about the eddy-current winding
losses calculation can be found for the so-called Rutherford-
In a finite-element based software [3] the inductances are cable (FTC construction) and its application for
calculated by simulating the magnetic field that arises when

978-1-6654-0507-2/21/$31.00
Authorized licensed use limited to: BC ©2021 IEEE
Hydro Library and Archives. Downloaded on November 05,2024 at 21:24:10 UTC from IEEE Xplore. Restrictions apply.
superconductive systems, but none could be found for its characteristics (inductance and power losses) of the ACR
application in ACRs. when comparing the analytical model and the proposed 2-D
FEM.
The equivalent geometry method using the 2-D FEA is
then proposed to solve for the eddy-current winding losses The analysis is performed by varying four main design
calculation for ACRs using the FTC technology. The parameters of the ACR: average winding diameter, total
validation of the method is analyzed by the above-mentioned winding height, number of strands and strand diameter. Those
DOE as well. parameters are set in three different levels (-1 0 +1), so the full
factorial design would lead to 81 different experiments.
II. THE EQUIVALENT GEOMETRY METHOD
In the FEM software, either in a 2-D or 3-D FEA it is IV. INDUCTANCE CALCULATION
possible that any said cable be represented by an encircling The 81 experiments were simulated with both analytical
equivalent geometry, as shown in Fig. 1, being possible to model and 2-D FEM and the first design characteristic
specify how many strands compose the cable, the strand evaluated was the total inductance of the windings.
diameter, material and all the relevant characteristics.
A. Analytical model
The inductances of the coils were obtained by using (1)
and (2) for self (L) and mutual (M) inductances respectively,
in (H).
ʹ
‫ ܮ‬ൌ Ͷߨߤ଴ ݊ଶ ܴଷ ൤‫ܥ‬௜ ሺܴǡ ܴǡ ݄ሻ െ ൨ (1)
͵ߨ
where n (turns/m) is the number of turns per unit length of the
Figure 1. 37 strands cable on the left and encircling geometry on the right. coil (turn/m), Ci is the function introduced by [2], R (m) is the
radius of the coil, h (m) is the height of the winding.
As windings usually present radial symmetry along the z ଷൗ
axis the above concept is further studied so not only a cable is ‫ ܯ‬ൌ Ͷߨߤ଴ ሺܴଵ ܴଶ ሻ ଶ ሾ‫ܥ‬௜ ሺܴଵ ǡ ܴଶ ǡ ‫ݖ‬ଵ ሻ െ ‫ܥ‬௜ ሺܴଵ ǡ ܴଶ ǡ ‫ݖ‬ଶ ሻሿ (2)
represented by the encircling equivalent geometry, but all the where R1 (m) and R2 (m) are the radii of the concentric coils,
turns of a coil will be modeled as a single 2-D structure, z1 (m) and z2 (m) are the height of the windings.
avoiding the need of modeling all the strands, as can be seen
in Fig. 2 and Fig. 3. B. 2-D Finite element model
In FEM inductance values are obtained through the
average energy Wav (J) of the system after a magnetic field
solution is completed as in (3).
ଶగ
‫ͳ ܮ‬ ଶ
ܹ௔௩ ൌ ൬ ൰ ൬ ൰ න ‫ܫ‬௉௘௔௞ ሾܿ‫ݏ݋‬ሺ߱‫ ݐ‬൅ ߠሻሿଶ ݀߱‫ݐ‬ (3)
ʹ ʹߨ ଴

C. Results
The deviations between the 2-D FEM and the analytical
model are presented in the Table I.
V. EDDY-CURRENT WINDING LOSSES
Figure 2. Example of a 2-D FEM of a FTC coil with 20 strands and 30
turns modeled strand by strand. Winding losses are composed by the dc components plus
the effect of a conductor carrying an ac current, which
corresponds to the skin-effect and the effect of the conductor
being immersed in a total equivalent varying magnetic field,
which corresponds to the proximity effect.
For each of the 81 experiments the dc components were
designed to have the same value for both models (analytical
and 2-D FEA) so they would not add any additional concern
when evaluating the results.

Figure 3. Example of a 2-D FEM of a FTC coil with 20 strands and 30

turns modeled with equivalent geometry.

III. THE DESIGN OF EXPIREMENTS

In this study the concept of the design of experiments [15]
is used to evaluate the variability of the desired design

Authorized licensed use limited to: BC Hydro Library and Archives. Downloaded on November 05,2024 at 21:24:10 UTC from IEEE Xplore. Restrictions apply.
 TABLE I. TOTAL INDUCTANCE DEVIATION, LT (%), BETWEEN 2-D ʹߨ
FEM AND ANALYTICAL MODEL FOR THE 81 EXPIREMENTS. ܴ௣௥௢௫ ൌ ߛሺ‫ܤ‬௥ ଶ ൅ ‫ܤ‬௭ ଶ ሻ݂௫ (5)
ߪ
where fx is the normalized unitless proximity-effect loss
LT LT LT factor as stated in [10].
Model Model Model
deviation deviation deviation
ACR1 -0.084% ACR28 -0.048% ACR55 -0.055% B. 2-D Finite element model
ACR2 -0.093% ACR29 -0.067% ACR56 -0.068% For the 2-D FEM when the equivalent geometry method is
ACR3 -0.114% ACR30 -0.047% ACR57 -0.046% considered the eddy-current losses are post-processed so a
formulation would be necessary. The same formula as in (5) is
ACR4 -0.091% ACR31 -0.065% ACR58 -0.043%
proposed to be used for the proximity-effect losses calculation.
ACR5 -0.086% ACR32 -0.054% ACR59 -0.051%
With this approach it is expected that simulation results for
ACR6 -0.076% ACR33 -0.087% ACR60 -0.057%
the that model will provide correction for the Ferreira’s
ACR7 -0.015% ACR34 0.040% ACR61 0.036% method in higher frequencies as the magnetic field will be
ACR8 -0.072% ACR35 -0.044% ACR62 -0.009% calculated for basically all the points across the cross section
of the windings.
ACR9 -0.092% ACR36 -0.049% ACR63 -0.063%
ACR10 -0.086% ACR37 -0.080% ACR64 -0.038% VI. SIMULATION RESULTS
ACR11 -0.061% ACR38 -0.091% ACR65 -0.068% In Fig. 4 the magnetic field distribution along the
equivalent geometry of the ACR81 is evidenced. The Fig. 5
ACR12 -0.055% ACR39 -0.021% ACR66 -0.018%
and Fig. 6 present the total losses values (kW) of the ACR79
ACR13 -0.122% ACR40 -0.100% ACR67 -0.083% and ACR81 calculated with the analytical model and the 2-D
ACR14 -0.098% ACR41 -0.041% ACR68 -0.069% FEM. The difference between the two ACR is basically the
strand diameter, this one being bigger in the ACR79 than in
ACR15 -0.063% ACR42 -0.027% ACR69 -0.070%
the ACR81. The results are consistent as the errors of most of
ACR16 -0.187% ACR43 -0.078% ACR70 -0.068% the models mentioned before increase with the increase of the
ACR17 -0.162% ACR44 -0.075% ACR71 -0.065% conductor diameter and frequency as well.
ACR18 -0.103% ACR45 -0.050% ACR72 -0.041%
ACR19 -0.062% ACR46 -0.044% ACR73 -0.021%
ACR20 -0.055% ACR47 -0.023% ACR74 -0.021%
ACR21 -0.035% ACR48 -0.018% ACR75 -0.014%
ACR22 -0.125% ACR49 -0.059% ACR76 -0.060%
ACR23 -0.090% ACR50 -0.036% ACR77 -0.026%
ACR24 -0.057% ACR51 -0.021% ACR78 -0.018%
ACR25 -0.212% ACR52 -0.080% ACR79 -0.069%
ACR26 -0.169% ACR53 -0.083% ACR80 -0.075%
ACR27 -0.101% ACR54 -0.063% ACR81 -0.065%
Figure 4. Magnetic field distribution using 2-D FEM for the ACR81.
The skin effect was also considered the same for both
models as provided in [9] and presented in (4). 10
ᇱ ሺߛሻ ᇱ ሺߛሻ
ܴ௦௞௜௡ ߛ ܾ݁‫ݎ‬ሺߛሻܾ݁݅ െ ܾ݁݅ሺߛሻܾ݁‫ݎ‬ 8
ൌ ቈ ቉െͳ (4) Analytical Model
ܴௗ௖ ʹ ܾ݁‫ ݎ‬ᇱ ሺߛሻ ൅  ܾ݁݅ ᇱ ଶ ሺߛሻ
ଶ
Total losses (MW)

2-D FEM
6
where Rdc (Ω) is the dc resistance of the winding, γ is the
frequency dependency [8], ber, bei, ber’ and bei’ are Kelvin 4
functions [7].
A. Analytical model 2

The proximity effect considered in this study follows what 0

was proposed in [7], with the magnetic field components in 0 2 4 6 8 10
radial and axial direction, Hr (A/m) and Hz (A/m) Frequency (kHz)
respectively, being calculated at the average radius of each
Figure 5. Total power losses comparision between analytical model and
winding, along the z axis.
2-D FEM for the ACR79.
The additional losses due the proximity effect (Rprox) for
each section of the winding can be calculated by (5).

Authorized licensed use limited to: BC Hydro Library and Archives. Downloaded on November 05,2024 at 21:24:10 UTC from IEEE Xplore. Restrictions apply.
 10 10

Analytical Model
8 8
2-D FEM

Total losses (MW)

Analytical Model
Total losses (MW)

2-D FEM Corrected Analytical Model

6 6

4 4

2 2

0 0
0 2 4 6 8 10 0 2 4 6 8 10
Frequency (kHz) Frequency (kHz)

Figure 6. Total power losses comparision between analytical model and Figure 7. Total power losses comparision between analytical model, 2-D
2-D FEM for the ACR81. FEM and the corrected analytical model for the ACR79.

VII. CORRECTION BY THE RESPONSE SURFACE METHOD 12

Among the 81 simulations, the minimum percentage of Analytical Model
10
deviation for the proximity-effect losses between the 2-D 2-D FEM

Total losses (kW)

FEM and the analytical model was a positive value of 34.5% 8 Corrected Analytical Model
and the maximum value was 109.4%, which is also in line
with the other discussed analysis. 6

The DOE method [16] facilitates the understanding of 4

how the design parameters, winding diameter, winding height,
2
number of strands and strand diameter, affects the design
characteristics, inductance and power losses. 0
0 2 4 6 8 10
Once it is evident that the design parameters have a major
Frequency (MHz)
impact on the accuracy of the loss’s calculation for high
frequencies, it is possible to draw a response surface which Figure 8. Total power losses comparision between analytical model, 2-D
correlates and quantify how much those parameters affect the FEM and the corrected analytical model for the ACR81.
design characteristics.
With the correction proposed with (6), the minimum
A correction formula (6) can be obtained by applying the percentage of deviation for the proximity-effect losses
response surface method [16] on the results of the study driven between the 2-D FEM and the analytical model is now -7.2%
by the DOE, and a correction factor (kprox) can be obtained. and the maximum value is 8.4%.
݈݊൫݇௣௥௢௫ ൯ ൌ  െͲǤ͸ͳ͵ ൅ ͲǤ͵ͳ͸ʹ݀௔௩ െ ͲǤͳͷ͵Ͷ݄௪ When adding to that the dc component and the values for
െ ͲǤͲʹ͸ʹʹ݊௦௧௥ ൅ ͲǤͲʹ͵͹݀௦௧௥ the skin-effect losses, composing then the total losses for the
ଶ ଶ
െ ͲǤͲ͵ͳͷͺ݀௔௩ ൅ ͲǤͲͲ͹Ͷͺ݄௪ windings, the minimum deviation observed is -0.8% and the
ଶ ଶ
െ ͲǤͲͲͲͲͳ݊௦௧௥ െ ͲǤͲͲͲ͸Ͷ݀௦௧௥ maximum is 3.1 %.
൅ ͲǤͲͲͶ͵Ͷ݀௔௩ ݄௪ (6)
൅ ͲǤͲͲͲͻͲ͹݀௔௩ ݊௦௧௥ VIII. CONCLUSION
െ ͲǤͲͲͳͳ͸͹݀௔௩ ݀௦௧௥ The application of the equivalent geometry method in the
൅ ͲǤͲͲʹͲͷͻ݄௪ ݊௦௧௥ 81 experiments has shown that there is a perfect
െ ͲǤͲͲʹͷʹ͸݄௪ ݀௦௧௥ ൅ ͲǤͲͲͻʹʹ correspondence between both simulation models for the
calculation of the total inductance.
where dav is the average diameter of the winding, hw is the
total winding height, nstr is the number of strands in the cable The analytical model applied in the loss’s calculation for
of the winding, dstr is the strand diameter. flat transposed cables also carries the lack of accuracy for
certain topologies of winding constructions, being more
The direct application of (6) in the proximity-effect losses critical for higher frequencies, as expected.
calculation using analytical model will provide the necessary
correction for the distribution of magnetic field as can be seen The expression for the kprox has proved to be very accurate
in Fig. 7 and Fig. 8, which presents the curves of the total and ease of use, dropping down the maximum error for the
losses calculated by the corrected analytical model. calculation of the proximity-effect losses from 109.4% to
8.4%.

Authorized licensed use limited to: BC Hydro Library and Archives. Downloaded on November 05,2024 at 21:24:10 UTC from IEEE Xplore. Restrictions apply.
 ACKNOWLEDGMENT [8] J. A. Ferreira, “Analytical computation of ac resistance of round and
rectangular litz wire windings,” IEE Proceedings-B Electric Power
The author would like to thank GE Renewable Energy for Applications, vol. 139, no. 1, pp. 21–25, Jan. 1992.
allowing this work to happen, and give thanks to FAPEMIG, [9] J. A. Ferreira, “Electromagnetic modeling of power electronic
INERGE, CAPES, and CNPq for their continued support in converters,” Norwell, MA: Kluwer, 1989.
[10] C. R. Sullivan, “Computationally efficient winding loss calculation
conducting research. The authors also would like to thank the with multiple windings, arbitrary waveforms, and two-dimensional or
engineer Renan Schwinden from ESSS Company for the key three-dimensional field geometry,” IEEE Transactions on Power
support in using of the FEA software. Electronics, vol. 16, no. 1, pp. 142–150, Jan. 2001.
[11] X. Nan and C. R. Sullivan, "Simplified high-accuracy calculation of
REFERENCES eddy-current loss in round-wire windings," IEEE 35th Annual Power
Electronics Specialists Conference, Aachen, Germany, 2004, pp. 873-
[1] F. W. Grover, “Inductance Calculations – Working Formulas and 879 Vol.2.
Tables,” 1st edition, New York: Dover Publications, 1946. [12] J. Acero, R. Alonso, J. M. Burdio, L. A. Barragan and C. Carretero, "A
[2] T. H. Fawzi, M. K. Gohar and F. Abdelaul, “The Accurate model of losses in twisted-multistranded wires for planar windings used
Computation of Self and Mutual Inductance of Circular Coils,” IEEE in domestic induction heating appliances," APEC 07 - Twenty-Second
Transactions on Power Apparatus and Systems, vol. PAS-97, no. 2, Annual IEEE Applied Power Electronics Conference and Exposition,
1978. Anaheim, CA, USA, 2007, pp. 1247-1253.
[3] ANSYS Maxwell. Help-Search. Ansoft Corporation. [13] E. Barzi, G. Gallo and P. Neri, "FEM Analysis of Nb-Sn Rutherford-
[4] J. Lammeraner and M. Stafl, “Eddy Currents,” London: Hiffe Books, type cables," in IEEE Transactions on Applied Superconductivity, vol.
1966. 22, no. 3, June 2012, pp. 4903305-4903305.
[5] P. L. Dowell, “Effects of eddy currents in transformer windings,” [14] S. Stavrev, F. Grilli, B. Dutoit and S. Ashwroth, “Finite-element
Proceedings of the IEE, vol. 113, no. 8, pp. 1387–1394, Aug. 1966. analysis and comparison of the ac loss performance of BSCCO and
[6] A. M. Urling, V. A. Niemela, G. R. Skutt, and T. G. Wilson, YBCO conductors,” EUCAS 05 – The 7th European Conference on
“Characterizing high-frequency effects in transformer windings-A Applied Superconductivity, vol. 43, Vienna, Austria, September 2005.
guide to several significant articles,” in Proc. APEC’89, pp. 373–385, [15] M. Abramowitz and I.A. Stegun, “Handbook of Mathematical
Mar. 1989. Functions,” New York: Dover Publications, 1970.
[7] J. A. Ferreira, “Improved analytical modeling of conductive losses in [16] General Electric Company, “Lean Six Sigma Book of Knowledge,” ver.
magnetic components,” IEEE Transactions on Power Electronics, vol. 5.4, Brazil, 2019.
9, no. 1, pp. 127–131, Jan. 1994.

Authorized licensed use limited to: BC Hydro Library and Archives. Downloaded on November 05,2024 at 21:24:10 UTC from IEEE Xplore. Restrictions apply.