High speed PM motor with hybrid magnetic bearing for kinetic energy storage

D. Johnson, P. Pillay* and M Malengret

Department of Electrical Engineering, University of Cape Town, Cape Town, South Africa
Tel: +27 21 650 3088, Fax: +27 21 650 3088, E-Mail: [email protected]
*ECE Department Clarkson University,Potsdam, NY, 13699

Abstract-This paper describes the design essentials of an out- signs, which was housed in a vacuum enclosure.
side rotor Permanent Magnet Halbach array motor with a hy-
brid magnetic bearing, intended for kinetic energy storage. The The objective of this exercise was to develop and evaluate
theoretical estimations are compared to Finite element analysis the viability of a unit economical enough for rural area and
as well as an approximated model using standard shape magnets.
stand alone applications in the developing world.
A dipole Halbach array produces a uniform flux distribution in-
side the cylindrical stator, where straight windings on an Iron-
less stator are placed near the inner boundary. The position of 2. MOTOR DESIGN
the windings inside this field does not affect the efficiency of the
motor. When the motor is operated continuously, these varia- The heart of the motor, with an outside rotor design, is a di-
tions become insignificant. pole Halbach array, establishing a uniform flux within the sta-
tor area, as shown in figure 1. A uniform flux is needed for
The motor is suspended on a hybrid magnetic bearing without operating the motor on a magnetic bearing, where slight
active control. Two radial repelling magnetic bearings are used movement of the stator inside the air gap is inevitable. Hal-
in combination with an axial journal bearing. The theoretical
bach [3] describes the flux density for a dipole with the equa-
values were found to be consistent to that of a working model.
tions :
For high speed kinetic energy storage it is essential that the
flywheel, which incorporates the motor and hybrid magnetic
bearing, is operated in a vacuum. Under these conditions, the
motor / bearing system approaches near 100% efficiencies e.g.
the approximated Halbach array motor model showed an electri-
B,B rem4log
_`
r2
r1
4Ç (1)

[ \
cal to mechanical efficiency of 97%. The flywheel target storage 2Í
capacity was 150 W-h with a minimum power rating of 40 W at sin
a maximum operating speed of 48000 rpm. It is intended that M
multiple modules of this basic unit can be used in parallel to in- Ç, (2)
crease the overall energy storage capacity for rural and isolated 2Í
power supplies. M
1. INTRODUCTION
B = Resultant uniform flux, Brem = Remnant flux in perma-
Kinetic energy storage in general requires a high efficiency nent magnets, r2 = outer radius, r1 = inner radius and M =
motor, taking into account that this method of storage is still number of poles.
expensive, and the energy has to move through the device at
least twice, i.e. charging and discharging. Inspection of (1) reveals that the wall thickness of the array
has a direct impact on the resultant flux density. Plotting the
Being able to store the energy for a reasonable time further flux density versus the wall thickness results in figure 2. This
imposes that the frictional losses should be low. This consists wall thickness is limited by commercial magnet dimensions,
mainly of bearing and air friction losses. The air friction is strength considerations due to centrifugal forces, cost of ma-
adequately catered for by removal, i.e. operating in a vac- terials and dimensional constraints.
uum. Limiting bearing losses can be achieved by combining a
passive magnetic bearing and dry lubricated journal bearing. When the number of magnets approaches infinity in (1), it
has a limit value of :

[ \
The design of this particular development was based
mostly on the work of R.F Post [1] and J. D. Steinmier [2], 2Í
which were both developed for vehicle applications. There sin
M
are few commercially available kinetic energy storage units lim M l  ,1 (3)
and are very expensive. The commercial units are in the hun- 2Í
dreds of kW range and intended for power quality applica- M
tions. New costs can be justified for power quality applica-
tions in the developing world, but not for rural applications.
This result is not practical, unless one such magnet could
The above designs are both 1 kWh units, suspended on be manufactured. To use a large number of magnets is costly,
magnetic bearings. The Steinmier [2] model used a hybrid because each magnet has to be magnetised in a unique direc-
magnetic bearing with an active bearing backup. Both incor- tion, the wedges become smaller and assembly becomes in-
porates filament wound multiple rim graphite composite de- creasingly cumbersome.

0-7803-7114-3/01/$10.00 (C) 2001

 The influence of the number of magnets used for an array is
shown in figure 2.

Inner v.s. Outer Diameter ratio

5
4.5
4
3.5
3
2.5
2
1.5
1
Fig.1: Magnet orientations for 8-magnet Halbach array
0.5
0
1 2 3 4 5 6 7 8 9 10
Number of Magnets 1.5 2.5 3.5 4.5 5.5 6.5 7.5 8.5 9.5

24 r2/r1
Number of Magnets

20 Fig. 5: Effect of the ratio between the inner- and outer radii
16
The resulting design therefore reduces to a steel cylinder
12 with mass produced magnets oriented in the preferential di-
8 rection as shown in figure 6.
4
To estimate the value of the remnant flux, Brem, in (1) for
0 the approximation, the flux has to be reduced in proportion to
0.1 0.3 0.5 0.7 0.9 the ratio of the actual volume of magnetic material to the total
0 0.2 0.4 0.6 0.8 1
volume of the material. For unity depth, (2) becomes :

_`
Factor
Amagnet r2
Fig 2: Effect of the number of magnets on resultant flux B,B rem 4log 4Ç (4)
Asteel r1
An eight magnet array gives an acceptable result, shown in
figure 3, which was used for this model. Amagnet = cross sectional area of the magnets and Asteel =
cross sectional area of the steel.

A prototype was constructed of readily available grade 28

NdFeB magnets of dimensions 10 x 50 x 4 mm and mounted
in an enclosure. A field strength of 0.39T was calculated. This
correlates exactly with that of the finite element analysis re-
sults and the measured values of the physical model. This
confirms the validity of using the magnet volume fraction in
(4).

Fig. 3: Finite Element analysis of 8-segment Halbach array

In order to reduce the cost of manufacture, the segments

may be replaced by commercially available rectangular mag-
nets, embedded in a steel matrix, as shown in figure 4.

N N
Fig. 6: Finite Element results of an approximated 8-seg-
ment Halbach array

The three phase coils consists of copper windings wound

onto an iron-less stator to eliminate iron losses. With the per-
Fig. 4: Magnet segment and approximation with angled manent magnetic field moving, eddy currents will be gener-
magnetisation direction ated in magnetic and conductive circuits in the vicinity of the

0-7803-7114-3/01/$10.00 (C) 2001

field. The induced voltage can be estimated by the the induction
law according to Fitzgerald [4] :
The torque generated is directly proportional to the ampere
turns and radius from the rotational centre in a Halbach array. d»
V i ,N (7)
dt

d»
Vi = induced voltage, N = number of turns and =
dt
rate of change of magnetic flux.

For a conductor loop the induced voltage becomes :

V i ,Õ» max sin ‘Õ t’ (8)

ω = angular velocity in radians per second and Φmax= mag-

netic flux given by the relationship[4] :
Fig. 7: Motor winding position
» max ,B A (9)
The armature windings are shown as single wires as oppos-
ing pairs, as shown in figure 7. When a current is flowing in a
B = calculated flux density in the Halbach array and A =
phase, it exerts a force on the winding and the magnetic field,
area enclosed within a conductor loop.
which are equal and opposite. Having a conductor in a uni-
form flux, the air gap between the coils and the rotor inside
For this motor running at 48000 rpm with a flux density of
the perimeter is of no significance for this motor`s efficiency.
0.39 Tesla and a mean area 0.022 X 0.050 m per winding, the
The torque generated by the motor is proportional to the am-
pere turns and radius from the rotational centre in a Halbach induced Voltage reduces to: V i ,2.43 N sin ‘Õ t’ per
array and are described by [4] : winding, with N = number of turns. The RMS value for a si-
3
T ,B4l4i4r (5) nusoidal wave shape has a value of [5] ,1.7 volts
2
per phase per winding at maximum speed.
F = force, B = magnetic flux, L = length of conductor, i =
current, r = radius and T = torque. TABLE 2
ESTIMATED VOLTAGES FOR DIFFERENT NUMBERS OF TURNS
A full bridge converter acts as a solid state commutation
device in this brushless DC machine, shown in figure 8. Windings VL-L (Delta Con- VL-L (Star Con-
nected) nected)
TABLE 1
14 24 41.6
BRIDGE CONVERTER SWITCHING SEQUENCE FOR 3-PHASE 20 34.3 59.4
OUTPUT
25 42.9 74.3
Angle 0 60 120 180 240 300
Positive TA+ TA+ TB+ TB+ TC+ TC+
Negative TB- TC- TC- TA- TA- TB-

By switching the MOSFETS / IGBTS of a full bridge con-

verter, figure 8, in the sequence shown in table 1 results in a
square wave voltage fed to the 3-phase windings. The in-
duced back EMF in the windings is sinusoidal. With the rotor
being a fixed magnetic field and the windings on an iron-less
core, the voltage equation for a DC generator, on the DC side
of the converter, applies [4] i.e.:

E,K » Õ0I4R (6)

E = induced motor EMF, Φ = flux density, fixed for a per-

Fig. 8: Full bridge converter
manent Halbach array, ω = angular velocity in radians per
second, I = winding current, R = resistance measured over the
MOSFETS and the windings and K = motor constant. Substituting KΦ in (6), we get :

0-7803-7114-3/01/$10.00 (C) 2001

 K1 (12) it is clear that the radial stress generated in the rotating
E, Õ0I4R (10) body is a function of the cylinder wall thickness. This is re-
900 quired for multiple rims, as the fibres in the composite are
aligned in the hoop direction and only the matrix binds the
K1= maximum rated voltage of the motor and R = com- laminates in the radial direction.
bined resistance of the semiconductor switches and the wind-
ings. The constant, "900", is the maximum speed of the ma- The rims are inter-connected with a rubber compound to
chine in Hertz. transmit the rotational-, but not the radial forces. The outer
rim was constructed from a filament wound glass fibre/epoxy
3. FLYWHEEL DESIGN
composite, while the inner rim was moulded from chopped
strand short glass fibres. This is allowable because the
Benham et al. [6] defines the hoop and radial stresses in a
stresses are much lower closer to the centre of the flywheel.
rotating cylinder by:

30Ê 11Ê The maximum stress that an aligned continuous fibre com-
Ï ×, 4Î4Õ24•‘r o ’20 4‘r ’2 – (11) posite material can withstand, as in the case of the filament
4 30Ê i wound outer rim of the flywheel, can be estimated by [8] :

Ï r,
30Ê
4Î4Õ 4‘r o1r i ’
2 2
(12) Ï ,Ï m ‘11 f ’0Ï f (13)
8
σ = composite tensile stress in the direction of the fibres,
σθ = maximum hoop stress, σr = maximum radial stress, ρ = σm = tensile strength of a matrix, σf = tensile strength of a
density of the cylinder, ro, ri = cylinder outer and inner radii, fibre and f = fibre volume fraction.
ω = angular speed and υ = poisons ratio.
Philips [8] describes the strength of a short fibre composite.
This depends on the critical fibre length, i.e. the length at
which fibre and the matrix fails at the same strain. This length
is calculated by :

Ïfd
l c, (14)
Ï xy

lc = critical length, d = fibre diameter and σxy = shear

strength of the matrix.
Fig. 9: Piecewise differential disk profiles

Berger and Porat [7] have shown that it is possible to ob-

tain much higher specific energies when using a combination
of piecewise differential disk profiles, as shown in Fig 10.

By maximising the specific kinetic energy, it was shown

that an optimum is reached when points p3,p4 and p5 coincide,
resulting in a profile disk as depicted in figure 10.

Fig. 11: Dual Rim Composite flywheel

Fig. 10: Optimal flywheel shape For a short fibre composite with fibres longer than the criti-
cal length, the material strength can be estimated by [8]:
This shape results in a specific energy of 1.92 times that of
a thin cylinder whereas an exponential profile disk only yields lc
an energy density of 1.49 times that of a thin cylinder. Ï , f Ï f ‘l1 ’0‘11 f ’Ï m (15)
2l
For the experimental model, a dual rim approach was taken
in order to simplify construction, shown in figure 11. From l = average fibre length.

0-7803-7114-3/01/$10.00 (C) 2001

 The resulting flywheel parameters are listed in table 3.

TABLE 3
DUAL RIM FLYWHEEL SPECIFICATIONS
Outer diameter 210 mm
Length 160 mm
Safe operating speed 48000 rpm
Safety factor 1.5 (on the outer rim)
Total Weight 6.234 kg
Capacity 157.42 W-h
Inertia 26.55 x 10-3 kg m/s
Fig.13: Axial journal bearing using synthetic gemstones
Specific energy 25.25 W-h / kg
The remaining degree of freedom in the axial direction has
4. HYBRID MAGNETIC BEARING to be confined to allow rotation of the shaft only. Axial sup-
port is achieved by resting the rotating part of the motor with
Steinmier et al. [2] describes the operation of a stable hy-
a hard ball on top of a hard flat plate, as shown in figure 14.
brid magnetic bearing system. This system is constructed of
two passive radial magnetic bearings (figure 12) and a journal TABLE 4
thrust bearing, shown in figure 13. Steinmier [2] calculated
the rotational frictional moment at the point of contact as: STIFFNES OF RADIAL MAGNETIC BEARINGS

Parameter Steinmier model [2] Prototype

T ,0.9Í r F É
2
(16)
ØINNER [mm] 44.5 30

T = Torque, r = Ball diameter, F = vertical force and µ = ØOUTER [mm] 50.8 60

frictional coefficient ØDISK [mm] 19.2 22

Height [mm] 10 10
Yonnet [9] shows that opposing ring and disk shaped
Stiffness [N/m] 600 3000
NdFeB permanent magnets yield a stable radial bearing, but
unstable in the axial direction, shown in figure 12. Likewise,
opposing disk magnets will be stable in the axial direction, A 5 mm stainless steel ball on a stainless steel surface will
but not in the other five degrees of freedom. Using two pairs give a frictional moment of approximately 1x10-4 Nm per
of ring and disk magnets removes the remaining degrees of kilogram of flywheel. For this flywheel it amounts to a run-
freedom except for two i.e. one linear in the direction of the down time of about 200 days, excluding any other losses,
shaft and the rotation thereof. The radial stiffness in this ar- from a speed of 48000 rpm.
rangement is approximately linear with distance.

Determining the stiffness can be done analytically with

great difficulty. Generally the Finite element method is used
and is verified experimentally. Steinmer et al. [2] determined
a linear radial stiffness of 600 N/m for a disk and annular
magnet pair, shown in Table 1.

Fig. 12: Radial Magnetic Bearing Fig. 14: Hybrid Magnetic bearing
An experimental prototype magnetic bearing was con- For a flywheel exerting 60 N on a single ball the frictional
structed out of commercially available NeFeB magnets. The losses will be as indicated in Table 5 [10].
dimensions and stiffness are shown in Table 1. The total ra-
dial stiffness for two pairs is 6000 N/m or 6 N/mm.

0-7803-7114-3/01/$10.00 (C) 2001

 TABLE 5
Speed v.s. Power in Vacuum
FRICTION VALUES FOR BEARING MATERIALS 2550

2500
Ball Plate Coefficient Torque Power
of Friction [Nm] @ 50000 2450

Speed [rpm]
[10] rpm [mW] 2400
Ruby / sap- Sapphire 1.5 x 10-4
2350
phire 0.1 152
2300
Hard Steel Hard Steel 0.42 6.0 x 10-4 628
Cast iron Cast Iron 0.15 2.2 x 10-4 230 2250

Teflon Steel 0.04 6.0 x 10-5 63 2200

20.15 21 27.2 29.75 32.4

Input Power [Watt]

3. MECHANICAL LOSSES OF THE FLYWHEEL SYSTEM Fig. 16: Motor speed vs. Power in vacuum
To determine the losses in the hybrid magnetic bearing, a I R
2
rundown curve was obtained while the containment was kept Õ, (18)
under vacuum. The rundown curve is shown in figure 15. 0.9Í r 2 F4É
At the resonant frequencies, especially the first, a large de- Applying this to the prototype flywheel results in a fric-
celeration is visible. The largest vibrations occurs at the first tional coefficient of: µ = (82 * (2* 0.055 + 0.516))/(0.9π
resonant frequencies. (0.01)2*60) *261) = 0.16 , using 2500 rpm as the maximum
speed (see figure 16) and a current of 8 Amperes.
Rundown Curve in Vacuum
2500
This value correlates with the coefficient of friction of Cast
2250 Iron on Cast Iron and lies in between those of hard steel on
2000 hard steel and hard steel on Teflon. The slight increase over
Speed [rpm]

1750 that of the steel on Teflon lubricated steel surface, used in this
1500 prototype, is due to the steel surfaces not being entirely
1250 smooth, flat and aligned. This causes the ball to slide over the
1000
surface, in small circular movements, instead of rotating on
750
one point.
500
250
0
5. SYSTEM OVERVIEW
0 84 146 269 390 500 620 717 771 824 924
45 115 203 322 439 573 687 742 803 860 981 Shown in figure 17 is a schematic, illustrating a solar array
application of a kinetic energy storage system.
Time [s]
Fig. 15: Rundown Curve in a Vacuum

At these amplitudes the moving parts makes contact with

the stationary parts, resulting in large mechanical losses.

The almost linear sections between the natural frequency

and multiples thereof i.e. ±900 , 1800 etc. rpm, shows that the
only frictional force is the contact on the axial support, con-
firming that the air friction has become negligible.

The motor was run at a constant current, which represents a

torque, and allowed sufficient time to reach balance with the
frictional torque of the system. This speed which was reached
for a constant input power is depicted in figure 16.

To find the maximum speed of this curve, the mechanical

friction in (16) is equated to the electrical resistive loss , giv- Fig. 17: Discrete logic controlled Electromechanical stor-
ing: age schematic

0.9 Í r 2 F4É4Õ,I 2 R (17) The solar array provides DC power to a user, sharing a
common bus with a bidirectional DC-DC converter, perma-
The theoretical maximum speed from the motor can now be nently connected to a kinetic energy storage unit. The control
rewritten as: logic limits the current flow in and out of the motor and moni-

0-7803-7114-3/01/$10.00 (C) 2001

tors the voltage, representing the speed of the motor. and control of a flywheel storage system with a hybrid
magnetic bearing.", Transactions of the ASME, Vol. 119,
The 3-phase, full bridge converter is synchronised with the December 1997.
motor by an optical encoder.
[3] K. Halbach, "Design of permanent multipole magnets with
When the voltage of the solar array is higher than that of orientated rare earth cobalt material", Nuclear instruments
the DC bus, current flows into the flywheel system which ac- and Methods, Vol. 169, 1980, North Holland Publishing
celerates the flywheel. When a consumer starts demanding Co.
power, which the solar panel may not be able to supply at the
time, the DC bus voltage reduces slightly and the current [4] A.E Fitzgerald, D Higginbotham, A Grabel, "Basic Elec-
starts flowing out of the electromechanical storage system, trical Engineering", 5th edition, 1981, Mc Graw Hill Kuga-
slowing down the flywheel. kusha Ltd.
6. CONCLUSION [5] N. Mohan, T.M. Undeland, W.P. Robbins, "Power Elec-
tronics, converters, applications and design", Second Edi-
The design procedure of a high efficiency motor / generator tion, 1995.
with a hybrid bearing system, built with off-the-shelf compo-
nents, has been described. High efficiencies were obtained, [6] P.P. Benham, F.V Warnick, "Mechanics of solids and
using standard manufacturing techniques available in South structures", 1982, Pitman Publishing Inc.
Africa, for a kinetic energy storage application.
[7] M. Berger, I. Porat, "Optimal design of a rotating disk for
It is concluded that the manufacture of efficient, low cost kinetic energy storage", Transactions of the ASME, Vol
kinetic energy storage devices is not only possible, but should 55, March 1988.
be pursued in order to develop an affordable and competitive
commercial unit for rural and isolated applications, as com- [8] LN Philips, "Design with Advanced Composite Materi-
petitor to the electrochemical battery. als", 1989, Biddles Ltd.

7. REFERENCES [9] JP Yonnet, "Passive magnetic bearings with permanent

magnets", IEEE Transactions on magnetics, Vol 14, Sep-
[1] R.F Post, S.F. Post, "A high-efficiency electromechanical tember 1978.
battery", Proceedings of the IEEE, Vol. 81, No 3, March
1993. [10] E.A. Avallone, T Baumeister III, "Marks Standard hand-
book for Mechanical Engineers", 9th edition, 1987,
[2] J. D. Steinmier, S.C. Thielman, B.C. Fabien, "Analysis McGraw-Hill Inc.

0-7803-7114-3/01/$10.00 (C) 2001