Design Optimization of Coreless Axial-flux PM Machines with

Litz Wire and PCB Stator Windings

Murat G. Kesgin, Student Member, IEEE, Peng Han, Member, IEEE, Narges Taran* , Member, IEEE,
Damien Lawhorn, Student Member, IEEE, Donovin Lewis, Student Member, IEEE, and Dan M. Ionel, Fellow, IEEE
SPARK Laboratory, ECE Dept., University of Kentucky, Lexington, KY, USA

[email protected], [email protected], [email protected],

[email protected], [email protected], [email protected]

Abstract—Coreless axial-flux permanent-magnet (AFPM) ma- suitable for design modifications and accurate manufacturing
chines may be attractive options for high-speed and high-power- techniques [13], and with good capabilities for heat dissipation
density applications due to the elimination of core losses. In order [14].
to make full use of the advantages offered by these machines
and avoid excessive eddy current losses in windings, advanced A prototype motor for low-cost household applications
technologies for winding conductors need to be employed to was designed with a PCB stator employing a wave winding
suppress the eddy effect, such as the Litz wire and printed circuit pattern and described in [13]. A non-overlapping concentrated
board (PCB). In this paper, the best practices for designing Litz winding with hexagonal concentric shapes was utilized in the
wire/PCB windings are discussed and a brief survey of state of the PCB stator proposed in [14]. Multiple concentrated spiral-
art PCB winding technology is provided. Three coreless AFPM
machines are mainly considered. A design optimization procedure shaped winding variants were analyzed and a procedure for
based on the multi-objective differential evolution algorithm and back electromotive force (EMF) calculation and PCB winding
3-dimensional (3D) finite element analysis (FEA) is proposed to designs was proposed in [15]. A comparison of theoretical
take into account the ac winding losses of Litz wires and PCB goodness predictions for wave windings [13] and hexagonal
traces in the machine design stage. Selected designs are being concentric patterns [14] led to the adoption of the wave
prototyped and will be tested with a customized test fixture.
Index Terms—Axial-flux, coreless, electric machine, multi- winding pattern for the PCB stator design studied in the current
objective optimization, PCB, permanent magnet, Litz wire paper.
II. C ORELESS AFPM S WITH L ITZ W IRE AND PCB
I. I NTRODUCTION
S TATOR W INDINGS
Coreless axial-flux permanent-magnet (AFPM) machines Three coreless AFPM machines are considered for study
typically employ a special disk-type construction with a stator, (Fig. 1). Two-phase wave windings are used in the two
which does not have a ferromagnetic core, positioned in machines shown in Fig. 1a and Fig. 1b to reduce the number
between two rotors that ensure the magnetic return path. As a of wire joints. Two-phase windings are advantageous in terms
consequence, these machines may be, in principle, lighter and of balance. The minimization of the mutual coupling provides
more efficient, because the stator core losses are eliminated better fault tolerance. The third machine employs a special
[1]. Designs of coreless AFPM have been recently proposed fractional-slot concentrated winding configuration, which is
and studied by the authors’ extended research group [2]–[5] suitable to be fabricated with modular coil groups. All the
and by others, e.g. [6]–[8]. machines are required to produce 12 Nm up to a speed of
Because in a coreless machine the windings are directly 3,000 r/min.
exposed to the time-varying main magnetic field the eddy The higher the operating frequency is, the smaller the
current losses in the stator conductors may be significant nominal single wire diameter has to be. To consider the
and special mitigation measures are required [9], [10]. Such interaction between several bundle diameters with skin depth,
constructive techniques include the use of special Litz wires, δ, in a simplified way, the maximum single wire diameter is
especially if the machines operate at high frequency [5], [11], recommended to be smaller or equal to nearly δ/3 [16].
[12]. The efficiency of coreless machines with PCB stators
An alternative approach is to employ printed circuit boards is heavily dependent on cross-sectional area of the copper
(PCBs) for the stator, such that copper traces serve as planar traces. Reducing the conductor’s height or width increases the
winding conductors between layers of laminated material. dc copper loss due to lower current carrying capacity unless
PCB stators have been reported as being compact, flexible, the number of tracks (turns) increases. Outside of a typical
* Dr. Narges Taran was with the SPARK Laboratory, ECE Department,
range, available copper trace height differs from manufacturer
University of Kentucky, Lexington, KY and is now with BorgWarner Inc., to manufacturer. The minimal copper trace width is limited
Noblesville Technical Center, IN, USA. by the precision of machines used in current manufacturing

Authors’ manuscript version accepted for publication. The final published version is copyrighted by IEEE and available as: Kesgin M. G., Han P., Taran N., Lawhorn D., Lewis
D., and Ionel D. M., ”Design Optimization of Coreless Axial-flux PM Machines with Litz Wire and PCB Stator Windings,” Proceedings, IEEE Energy Conversion Congress and
Exposition (ECCE), Detroit, MI, Oct 2020, 5p. ©2020 IEEE Copyright Notice. “Personal use of this material is permitted. Permission from IEEE must be obtained for all other
uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution
to servers or lists, or reuse of any copyrighted component of this work in other works.”
 (a) (b) (c)
Fig. 1. Exploded views of coreless AFPM machines under study: (a) 2-phase AFPM with wave windings wound with Litz wires, (b) 2-phase AFPM with
PCB wave windings, (c) 3-phase AFPM with fractional-slot concentrated windings.

practices. However, the number of traces can be increased

axially by stacking and connecting more PCBs, creating more
layers in the PCB, or by reducing the conductor width to allow
for more traces on the same plane. Since higher numbers of
printed boards increases the electromagnetic airgap, ideally,
more traces should be placed on each single board.
The fill factor of PCBs is restricted by the material prop-
erties of the insulators between copper traces wherein voltage
creepage threatens the integrity of the PCB if gaps are created
below a minimum defined in current standards for PCB
Fig. 2. Cross sections of equivalent macro coils of Litz wire and PCB stator
manufacturing. For a coreless machine operating at a very low windings used in 3D parametric models for optimization. Dimensions of PCB
fundamental frequency of 25 Hz, the copper trace width of 1.2 traces and the gap between them depend on the number of layers, overall
mm was employed [14]. Another group considered trace width thickness, finnished copper, etc., and are closely related to PCB manufacturers.
of 0.2-0.3 mm and trace height of 0.07-0.105 mm with the gap
between traces set at 0.23-0.3 mm for a machine operating at In contrast with Litz wire windings, PCB windings are
a fundamental frequency of 33.3 Hz [15]. In a PM machine comprised of large numbers of copper traces, which are
designed for operation at 1 kHz, a copper trace width of 0.3 typically straight and without complex transposition. Eddy
mm was adopted, with a 0.3 mm circumferential gap between current loss in PCB traces can be estimated by 2D or 3D
traces [17]. The generic standard on PCB design, IPC-2221, FE models, as described, for example, in [13], [17], [20].
recommends to have a minimum gap of 0.13 mm between For coreless machines, the eddy current loss needs to be
traces at our intended voltage and a minimum trace width of considered during the design and optimization stage. However,
0.15 mm was chosen to allow for that gap for the design target the calculation of eddy current loss in conductors, including
under study. Litz wires and PCB traces, if solved directly using FE solvers,
is time-consuming and challenging for large-scale design opti-
III. E STIMATION OF E DDY C URRENT L OSS IN M ACHINE mizations [21]. This paper proposes to use a hybrid analytical–
W INDINGS numerical approach that can be directly incorporated into the
multi-objective optimization algorithm to quickly estimate the
Litz wires consist of multiple strands insulated electri-
eddy current loss in windings based on the variation of average
cally from each other, which are usually twisted or woven
flux density in equivalent macro coils from 3D transient FEA.
to minimize the skin effect and proximity effect. Accurate
predictions of eddy current loss in Litz wires involve detailed Based on the guidelines from Section II, the wire gauge
3D wire models for which the scale of the problem is usually for each strands needs to be at least AWG 38. A Litz wire
prohibitive for solving without using the high performance consisting of 100 strands of AWG 40 magnet wires was
computing system. To reduce the burden of modeling and selected, which is expected to be able to reduce the ac winding
computation, numerous reduced-order methods were proposed, losses for this study to negligible levels. The equivalent macro
such as the fast 2.5-dimensional (2.5D) partial element equiv- coil model for Litz wire windings is illustrated in Fig. 2.
alent circuit [18], the combined numerical and analytical Assuming that all the conductors in coils are straight and di-
approach called the squared-field-derivative method [19], the rectly exposed to a uniform magnetic field varying sinusoidally
hybrid method that combines the analytical equations and FE with time, neglecting the eddy current loss in end coils, the
simulations [9], [10], etc. eddy current loss in Litz wire windings with round conductors
 TABLE I
O PTIMIZATION VARIABLES AND T HEIR R ANGES .

Variable Description Min. Max.

Rir Inner radius of rotor [mm] 80 120
kcw Coil width / Max. coil width 0.50 0.95
kry Rotor yoke thickness / PM axial length 0.5 1.5
ks Stator axial length / PM axial length 0.5 2.0

is estimated by:
Ba2 πd4
Peddy = σLc Nc Nt Ns ω 2 , (1)
128
where σ is the conductivity of conductors, ω the electrical
angular velocity, Lc the length of conductors, Nc the number
of coils, Nt the number of turns per coil, Ns the number of
strands per turn, d the diameter of each strand (conductor).
Ba is the amplitude of the flux density [9].
For PCB stator windings, the traces are treated as rectan-
gular wires, and the equation for eddy current loss estimation
is: 2
(Baz w2 + Baφ
2
h2 )wh
Peddy = σLc Nc Nt Ns ω 2 , (2)
24
Fig. 3. Parametric 3D model and main geometric variables.
where Baz and Baφ are the axial and tangential components
of the flux density, and w and h are the width and height of
each strand, respectively.

IV. PARAMETRIC M ODELING AND O PTIMIZATION

The material properties and dimensions of magnets and Litz
wires are kept constant during the optimization since they
are predetermined by available suppliers. The dimensions of
the stator and rotor back iron are optimized to achieve the
minimum total axial length and total electromagnetic loss,
including the dc winding loss, ac winding loss and PM eddy
current loss. The width and height of equivalent macro coils
are m and n times the diameter of Litz wire for Litz wire
windings, where m and n are positive integers. They can be
equal or unequal. In PCB stator windings, the width and height
Fig. 4. The Pareto front and all the other designs evaluated by 3D FEA
of the equivalent macro coil are controlled similarly to keep throughout the differential evolution for the 3-phase AFPM with fractional-
the fill factor constant. slot concentrated windings.
Four independent variables are identified for the optimiza-
tion, as tabulated in Table I. The geometry of the coreless V. R ESULTS AND D ISCUSSION
machine using Litz wire and PCB stator windings is illus- The Pareto front for the optimization and all the other
trated in Fig. 3. The fill factor for equivalent macro coils is designs using Litz wires are shown in Fig. 4. The highest
determined by the wire layout. It is 0.386 and 0.303 for the electromagnetic efficiency achieved by the optimum designs
Litz wire winding and PCB stator winding shown in Fig. 2, from the Pareto front is 98.8%. The box plot as shown in
respectively. Fig. 5 indicates that independent variables of all the Pareto
The optimization method is a variation of the 2-level front designs are within their ranges and the optimization
surrogate-assisted algorithm which combines the multi- is acceptable. The ac winding loss accounts for a small
objective differential evolution (DE) and kriging meta-models fraction of the total electromagnetic loss, as shown by the
to accelerate the generation of the Pareto front [22]. This loss breakdown in Fig. 6.
optimization algorithm requires a reduced number of design The PCB version is also optimized following the same
evaluations and can be implemented in powerful workstations procedure except the Litz wires are replaced by rectangular
or high performance computing systems, enabling the accurate copper traces. The equation used for ac winding loss esti-
performance calculation of each candidate design in less than mation and fill factor are changed according to the different
10 minutes. constructions. One design example with PCB stator windings
 Fig. 5. Box plot for the Pareto front shown in Fig. 4.

(a)

Fig. 6. The loss breakdown of Pareto front designs shown in Fig. 4. The
three loss components are normalized relative to the values for the Design 1.

from the optimization was selected and modeled in detail, as (b)

shown in Fig. 7. There are two PCBs, one for each phase, Fig. 7. Detailed modeling of a PCB stator design example, (a) top view, (b)
stacked with a phase shift of 90 electrical degrees. Each PCB unrolled 2D view from a cylindrical cut plane shown together with rotors.
has 6 layers with 50 traces per layer. The terminals are shown
in the zoomed-in view in Fig. 7a.

VI. P ROTOTYPING AND T ESTING

Three selected optimal designs are being fabricated. The
customized test fixture shown in Fig. 8 is able to adjust the
airgap length during the testing. The structure developed has
been constructed using tight tolerance aluminum jig plate and
a 20 mm keyed steel shaft, as well as a dual bearing system for
proper alignment. This design allows flexibility for mounting
of various AFPM machines and precise calibration of their
airgaps. The fixture is also equipped with a vertical back plate
with an opening for the shaft such that an encoder can be
mounted for speed feedback to be used in the control loop for
Fig. 8. Customized test fixture with the coreless machine with PCB stator
the machine. installed.
To assemble the coreless machine under test to the fixture,
the first of the two rotors is mounted to a flanged clamping
shaft collar and then placed onto the shaft into position and present between the second rotor and stator, an adjustment
locked into place using the collar’s locking nuts. Nylon spacers tool is used which allows precise, controlled placement of the
are then bolted to the vertical plate of the fixture, which second rotor on the shaft. Once the airgap is satisfactory, the
the stator is then bolted to. This procedure sets the airgap second rotor is locked into place using the same flanged shaft
between the first rotor and stator. To combat axial forces collar approach as the first.
 VII. C ONCLUSION [8] Y. Liu, Z. Zhang, C. Wang, W. Geng, and T. Yang, “Design and anal-
ysis of oil-immersed cooling stator with non-overlapping concentrated
Coreless AFPM machines have special constructions and winding for high-power ironless stator AFPM machines,” IEEE Trans.
their windings can be designed with additional degrees of free- Ind. Electron., Mar. 2020, DOI: 10.1109/TIE.2020.2978694.
dom as there are no geometrical constraints imposed by teeth [9] N. Taran, V. Rallabandi, D. M. Ionel, G. Heins, and D. Patterson,
“A comparative study of methods for calculating ac winding losses in
and slots. Such machines may achieve very high efficiency permanent magnet machines,” in 2019 IEEE Int. Electric Mach. Drives
by employing Litz wire or PCB windings and through design Conf. (IEMDC), 2019, pp. 2265–2271.
optimization. [10] N. Taran and D. M. Ionel, “A hybrid analytical and fe-based method
for calculating ac eddy current winding losses taking 3d effects into
Three coreless AFPM machines, two with Litz wire wind- account,” in 2019 IEEE Energy Conv. Cong. and Expo. (ECCE), 2019,
ings and the other with PCB stator windings, were optimized pp. 4867–4872.
taking into account the ac winding loss by the proposed [11] C. Zwyssig, S. D. Round, and J. W. Kolar, “An ultrahigh-speed, low
power electrical drive system,” IEEE Trans. Ind. Electr ., vol. 55, no. 2,
optimization procedure. It is shown that high efficiency can pp. 577–585, 2008.
be achieved through the proposed design optimization method. [12] H. Hämäläinen, J. Pyrhönen, J. Nerg, and J. Talvitie, “AC resistance
The prototyping of three selected designs are underway, with factor of Litz-wire windings used in low-voltage high-power generators,”
IEEE Trans. Ind. Electron., vol. 61, no. 2, pp. 693–700, 2014.
testing and experimental validations to follow. [13] F. Marignetti, G. Volpe, S. M. Mirimani, and C. Cecati, “Electromagnetic
design and modeling of a two-phase axial-flux printed circuit board
ACKNOWLEDGMENT motor,” IEEE Trans. Ind. Electron., vol. 65, no. 1, pp. 67–76, 2018.
The support of the National Science Foundation, NSF Grant [14] X. Wang, C. Li, and F. Lou, “Geometry optimize of printed circuit board
#1809876, of University of Kentucky, the L. Stanley Pigman stator winding in coreless axial field permanent magnet motor,” in IEEE
Veh. Power Prop. Conf. (VPPC), 2016, pp. 1–6.
endowment and ANSYS Inc., is gratefully acknowledged. [15] D. Gambetta and A. Ahfock, “Designing printed circuit stators for
brushless permanent magnet motors,” IET Electr. Power Appl., vol. 3,
R EFERENCES no. 5, pp. 482–490, 2009.
[16] Elektrisola, [Online]. Available: https://www.elektrisola.com/en/hf-
[1] N. Taran, V. Rallabandi, G. Heins, and D. M. Ionel, “Coreless and
litz-wire-litz/products/terminology-basics/selection-of-litz-wire -
conventional axial flux permanent magnet motors for solar cars,” IEEE
parameters.html.
Trans. Ind. App., vol. 54, no. 6, pp. 5907–5917, Nov 2018.
[17] N. S., S. P. Nikam, S. Singh, S. Pal, A. K. Wankhede, and B. G.
[2] V. Rallabandi, N. Taran, and D. M. Ionel, “Multilayer concentrated
Fernandes, “High-speed coreless axial-flux permanent-magnet motor
windings for axial flux pm machines,” IEEE Trans. on Magn., vol. 53,
with printed circuit board winding,” IEEE Trans. Ind. Appl., vol. 55,
no. 6, pp. 1–4, 2017.
no. 2, pp. 1954–1962, 2019.
[3] V. Rallabandi, N. Taran, D. M. Ionel, and J. F. Eastham, “Coreless
[18] T. Guillod, J. Huber, F. Krismer, and J. W. Kolar, “Litz wire losses:
multidisc axial flux pm machine with carbon nanotube windings,” IEEE
Effects of twisting imperfections,” in IEEE Workshop Contr. Modeling
Trans. Magn., vol. 53, no. 6, pp. 1–4, 2017.
Power Electron. (COMPEL), 2017, pp. 1–8.
[4] N. Taran, V. Rallabandi, and D. M. Ionel, “Waved: A coreless axial flux
[19] C. R. Sullivan, “Computationally efficient winding loss calculation with
pm motor for drive systems with constant power operation,” in 2019
multiple windings, arbitrary waveforms, and two-dimensional or three-
IEEE Trans. Electric Mach. Conf. and Expo (ITEC), 2019, pp. 1–6.
dimensional field geometry,” IEEE Trans. Power Electron., vol. 16, no. 1,
[5] M. G. Kesgin, P. Han, N. Taran, and D. M. Ionel, “Overview of flywheel
pp. 142–150, 2001.
systems for renewable energy storage with a design study for high-speed
[20] X. Wang, W. Pang, P. Gao, and X. Zhao, “Electromagnetic design and
axial-flux permanent-magnet machines,” in 2019 8th Int. Conf. Renew.
analysis of axial flux permanent magnet generator with unequal-width
Ener. Research and App. (ICRERA), 2019, pp. 1026–1031.
PCB winding,” IEEE Access, vol. 7, pp. 164 696–164 707, 2019.
[6] Z. Zhang, A. Matveev, R. Nilssen, and A. Nysveen, “Ironless permanent-
[21] M. Rosu, P. Zhou, D. Lin, D. M. Ionel, M. Popescu, F. Blaabjerg,
magnet generators for offshore wind turbines,” IEEE Trans. Ind. Appl.,
V. Rallabandi, and D. Staton, Multiphysics Simulation by Design for
vol. 50, no. 3, pp. 1835–1846, 2014.
Electrical Machines, Power Electronics and Drives. Wiley-IEEE Press,
[7] B. Xia, J. Shen, P. C. Luk, and W. Fei, “Comparative study of air-cored
2017.
axial-flux permanent-magnet machines with different stator winding
[22] N. Taran, D. M. Ionel, and D. G. Dorrell, “Two-level surrogate-assisted
configurations,” IEEE Trans. Ind. Electron., vol. 62, no. 2, pp. 846–856,
differential evolution multi-objective optimization of electric machines
2015.
using 3-D FEA,” IEEE Trans. Magn., vol. 54, no. 11, pp. 1–5, 2018.