I. INTRODUCTION
2
TABLE I
ROTOR FREQUENCIES FOR FORWARD FIELD COMPONENT IN UNBALANCED MACHINE
mN 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
1 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36
3 !2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34
5 !4 !2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
7 !6 !4 !2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
9 !8 !6 !4 !2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
11 !1 !8 !6 !4 !2 0 2 4 6 8 10 12 14 16 18 20 22 24 26
13 !1 !1 !8 !6 !4 !2 0 2 4 6 8 10 12 14 16 18 20 22 24
15 !1 !1 !1 !8 !6 !4 !2 0 2 4 6 8 10 12 14 16 18 20 22
17 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4 6 8 10 12 14 16 18 20
19 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4 6 8 10 12 14 16 18
21 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4 6 8 10 12 14 16
23 !2 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4 6 8 10 12 14
25 !2 !2 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4 6 8 10 12
27 !2 !2 !2 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4 6 8 10
29 !2 !2 !2 !2 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4 6 8
31 !3 !2 !2 !2 !2 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4 6
33 !3 !3 !2 !2 !2 !2 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2 4
35 !3 !3 !3 !2 !2 !2 !2 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0 2
37 !3 !3 !3 !3 !2 !2 !2 !2 !2 !1 !1 !1 !1 !1 !8 !6 !4 !2 0
Table II
ROTOR FREQUENCIES FOR BACKWARD FIELD COMPONENT IN UNBALANCED MACHINE
mN 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
1 -2 -4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38
3 -4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40
5 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42
7 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44
9 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46
11 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48
13 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50
15 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52
17 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54
19 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56
21 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58
23 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58 -60
25 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58 -60 -62
27 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58 -60 -62 -64
29 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58 -60 -62 -64 -66
31 -32 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58 -60 -62 -64 -66 -68
33 -34 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58 -60 -62 -64 -66 -68 -70
35 -36 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58 -60 -62 -64 -66 -68 -70 -72
37 -38 -40 -42 -44 -46 -48 -50 -52 -54 -56 -58 -60 -62 -64 -66 -68 -70 -72 -74
3
Circumferential segmentation; residual suppression—Fig. 3
shows a magnet of width $ in the circumferential direction, with
a segment J1 of the current-density wave that would be induced
by a harmonic field rotating relative to the magnet, if the magnet
was a complete ring. J1 is obtained from the earlier field
solution. Fig . 3 is drawn for one instant in time. The position of
the magnet relative to the J-wave is defined as the phase > of the
centre-line of the magnet relative to the J-wave: thus > = (21 +
22)/2 and $ = (22 21); or 21 = > $/2 and 22 = > + $/2. The
angles 21, 22, > and $ are all measured in electrical radians
relative to the harmonic wave length 8. Thus if $m is the magnet
width in actual mechanical radians, $ = $m × 2B/8.
If the J-wave is represented as J1 sin 2, the integral
22
J1 sin 2 d 2 (12)
21
represents the net current that would flow in a section of a
Fig. 3 Residual current suppression
complete ring of magnet between the angles 21 and 22. This
integral is not in general zero. However, the net current in a 7 2 1 $ 1
magnet block of finite width $ must be zero, because there is no ' (q 2 )$ 4 q sin sin > cos 2 > sin $ (16)
70 $ 2 2 2
connection by which it can find a return path in another magnet.
The net current must therefore return through the magnet block In Fig. 3 the magnet width is $ = 90E relative to the harmonic
itself. The distribution of return current-density across the cross- wavelength, and its phase position is > = 80E. Thus 21 = 80
section of the magnet block is not known a priori; but it can be 90/2 = 35E and 22 = 80 + 90/2 = 125E. Equation (14) gives q =
supposed to be uniform on the grounds that any non-uniformity J0/J1 = 0@88664, and with $ = B/2 radians (16) gives the loss
would imply that there were loops in the return current path reduction factor 7/70 = 0@0207. While this seems a very
containing induced EMFs, when the induced EMFs have in fact substantial reduction, it needs further comment.
already been completely accounted for by the solution of the field First, this value applies only at one instant corresponding to the
equations leading to the J-wave. (In finite-element programs the phase angle > = 80E. As the cycle progresses and the harmonic
imposition of zero net current is achieved by a related process — wave moves relative to the magnet, 7/70 varies and there are
often without proof — in which the end-faces of the conducting instants when it is equal to 1. Therefore we can expect the
regions are shorted together. Thus with a uniformly-distributed average loss reduction factor to be greater than 0@0207.
return current density J0 we have Secondly, a magnet width of $ = 90Eis one-quarter of the
22 harmonic wavelength. Consider the 5th space-harmonic of the
( J0 J1 sin 2 ) d 2 ' 0 , (13) stator winding, whose wavelength at fundamental frequency is
21
1/5th of a pole-pitch. $ = 90E for this harmonic implies that the
from which magnet is segmented into 20 blocks per pole — rather a large
cos 22 cos 21 sin ($ / 2) number. A magnet with 180E arc not segmented would have $ =
J0 ' J1 ' J1 sin > . (14)
22 21 $/2 5 × 90 ' 450E, and it can be verified from (16) that the loss
reduction factor does not fall much below 1. A magnet with 80%
The loss in the magnet block is proportional to (J0 + J1 sin 2)2
pole-arc (144E) would have $ = 360E relative to the 5th space-
integrated over $ : thus if q = J0/J1 we can write a loss function
harmonic, and in this case the loss reduction factor would be 1.
22
7 ' J12 ( q sin 2 )2 d2 Of course the magnet “region” is only 80% filled with magnet, so
21
there is a reduction of 0@8 relative to a full-ring magnet; but there
(15)
is no reduction due to the “residual current suppression” effect.
1 $ 1
2
' J1 ( q 2
)$ 4 q sin sin > cos 2 > sin $ . The upper graph J1 in Fig. 3 is the “original” current
2 2 2
distribution and the lower graph J is the net current distribution
When $ ' 360kE (for integer k), the magnet width is an integral with “residual suppression”. Also shown are the squares of
multiple of the wavelength 8 of the harmonic field component, these two graphs. At this phase position the reduction in squared
and (13) is automatically satisfied, with J0 = 0, regardless of the current-density is noticeable.
phase or position of the J-wave relative to the magnet. The The arrows in Fig. 3 show the direction and magnitude of the
harmonic loss in a magnet whose width is equal to an integral current density in the magnet block, “seen from above”. At the
multiple of the harmonic wavelength is therefore unaffected by instant > = 80E it is asymmetrical. When $ is an integral multiple
segmentation. The integral 7 in (15) then degenerates to 70 ' of the harmonic wavelength, the distribution is symmetrical, and
J12$/2, so we can characterize the effect of circumferential as > advances the arrows at one end leave the block and re-appear
segmentation of the magnets by the loss reduction factor at the other end, while J0 remains zero at all times.
4
The loss reduction factor (16) is specific to one field harmonic.
This means that simplified methods of allowing for segmentation
(such as equivalent resistivity) are hardly likely to be successful.
For similar reasons, (16) cannot be used to introduce the effect of
segmentation into the results of a 2-dimensional finite-element
calculation.
The loss reduction factor u = 7/7o in (16) must be integrated
over one harmonic cycle to get the average value: thus
sin $ 2 $
u ' 1 $a 2 4 a sin sin $ sin2 > . (17)
$ $ 2
where a = (2/$) sin($/2) Then the average value of u = 7/7o is
2B 2
1 sin($ / 2)
U ' u (>) d> ' 1 . (18)
2B 0 $/2
Fig. 4 Estimation of finite-length effect
..a remarkably simple function. For the example considered
earlier with $ ' 90E or B/2, we get U = 0@18943. When $ ' 2B
E my my m 2y
or a multiple thereof, U = 1 as expected. ' 1 ' 1 ' 1
DJz d b y m h (b y) (24)
h
Axial segmentation — Fig. 4 shows half a block of magnet m
and so
rolled out in rectangular coordinates. The direction of rotation is E/D
in the y direction and z is the axial coordinate. The axial length Jz ' with J y ' m J z . (25)
2
1 m y / [m h (b y)]
of the block is 2h and its width is 2b, which can be taken as the
wavelength of the exciting harmonic. The current-density Jz is decreased by the factor in brackets.
Consider a filamentary loop of eddy-current. If the block was However, the losses are proportional to
infinitely long in the axial direction, the current density in the w ' D (J z2 d ) y J y2 y ) z ) (26)
filament would be Jz0 = FE = jTFA, where A is the solution of
the 2-dimensional field at the circumferential position y. The cut and this can be expanded by writing
edge of the magnet at z = h forces all the current to veer into the w )y
' d ) y m 2 y ) z ' (m h (b y) m 2 y (27)
circumferential direction and rejoin its return path. In Fig. 4 the D J z2 m
current is assumed to be symmetrical about the centre-line of the
block. It is further assumed that the current in the filament )y Substituting for Jz from (25), and simplifying, we get
turns abruptly into the circumferential direction and flows in a E2 [ m h (b y) ] 2
circumferential filament of width )z such that w ' )y. (28)
D m [ m h (b y) m 2 y ]
J y )z ' J z )y . (19)
If there is no end-effect we have simply
The filament widths are further assumed to be related by
)y E2
' tan ( ' m , w0 ' h )y. (29)
)z
(20) D
where ( is an arbitrary angle defining the slope of the “break” Hence it is possible to define a loss reduction factor 8 for the
line shown dashed in Fig. 4. Hence filament at y :
Jy ' mJz. (21) w [m h (b y)] 2
8(y) ' ' . (30)
Considering just one quadrant of the loop, the driving EMF can w0 mh [ mh (b y) m 2 y ]
be identified as Ed. In the case where the magnet is infinitely
long in the z direction we have Ed = DJz0 h, where D = 1/F is the If we normalize h and y to b by writing h = h/b and y = y/b, this
resistivity. When the length is finite, the filament acquires a can be written
second segment of length y in the y-direction, while the segment w [m h (1 y)] 2
in the z-direction is shortened from h to d, where 8(y) ' ' . (31)
w0 mh [ mh (1 y) m 2 y ]
b y
d ' h a ' h . (22) If ( = 45E, m = 1 and the expression simplifies further:
m
The shortening of the axial length reduces the EMF w [h (1 y)] 2
proportionally, while the addition of the second segment 8(y) ' ' . (32)
w0 h[h 1 2y]
increases the total resistance. This is expressed by the equation
A magnet which is long in the axial direction has h >> 1, so that
Ed ' D Jzd Jyy (23) 8(y) ÷ 1 for all values of y. For long magnets ( is not critical, but
which can be expanded by writing in short magnets ( will have a minimum value.
5
Fig. 5 End-effect factor: 1 proposed method; 2 Russell and
Norsworthy's method [7]
Fig. 6 Integration of slot-modulated airgap flux
For example a “square” magnet has h = 1 or h = b. In this case B. Interior-magnet Machines (IPM)
( must not be less than 45E, giving m > 1. A magnet for which h Losses caused by time-harmonics in the stator current — The
= 0@5 (or h = b/2) restricts ( to values greater than arc tan(2) = IPM presents a problem in that the magnet region inside the rotor
63@4E. These limits of course have no physical basis, but are is not a plain cylinder and so it does not conform to the earlier
constraints imposed by the modelling assumptions. solution of the diffusion equation. However, an approximate
The loss reduction factor for the whole block is obtained as the estimate of harmonic losses can be made with the aid of the
average of 8(y) for all the filaments, or equivalent-circuit model and the frequency-dependent
1
1 synchronous reactance Ld(jT), which for the moment is assumed
7 ' 8( y ) dy . (33) known, [4]. We know that the phase angle of Ld(jT) is always
1 0
negative, so we can write
This can be integrated numerically or formally, giving Ld ( j T) ' LdR j LdX , (35)
(A C B 1 / C) ln (1 C) (B 1/2) C 1 where the real and imaginary components LdR and LdX are both
7 (h) ' (34)
C 2 k (k 1) functions of frequency T.
A d-axis current Id(jT) which alternates at the radian frequency
where k = mh, A = (k 1)2; B = 2 (k 1); and C = (1+m2)/(k 1). T produces a voltage
Fig. 5 shows examples of 7(h) calculated by (34) for various
V d ( j T) ' [ Rd j T Ld ( j T ) ] I d ( j T ) (36)
values of the magnet length/width ratio, h. The parameter k is
arbitrary and should be adjusted to match test data. where Rd is the armature resistance. Substituting the real and
The overall factor 7 in (34) is so far simply an end-effect imaginary components of Ld(jT),
factor for the losses in magnets of different length/width ratio h.
Vd ' ( Rd T LdX ) j TLdR I d . (37)
Now suppose we start with a full-length magnet with a certain
value of length/width ratio h1 and end-effect factor 71. If this The term TLdX represents the resistance of the conductive
magnet is divided into n segments in the axial direction, the end- elements on the rotor, referred to the d-axis circuit of the stator.
effect factor for each segment becomes 7n, calculated with hn = Generally these elements are just the magnets. The losses in the
h1/n. Although 7n operates on only (1/n) of the losses, there are magnets are therefore given by
now n segments, so the overall effect is that the end-effect factor
is 7n for the whole array, instead of 71. Wm(d) ' T LdX Id2 , (38)
As an example, suppose we start with a full-length magnet where Id is the RMS current at the harmonic frequency T.
having h = 3, and divide it into 3 segments. The end-effect factor In a balanced 3-phase machine the simplest cases of such a
with ( = 60E is 0@699 for the undivided magnet, and 0@2935 for current arise from the (6k ± 1)th time-harmonics interacting with
the divided magnet. Thus the division into 3 segments reduces the the fundamental electrical space-harmonic of the winding
loss by a factor of 0@2935/0@699 = 0@419. distribution. For example the 5th harmonic produces a rotating
It should be said that this end-effect analysis is a rough-and- ampere-conductor distribution rotating backwards relative to the
ready estimate. It is compared in Fig. 5 with Russell and rotor at six times the fundamental synchronous speed, and this
Norsworthy's method, which is somewhat more “analytical” than can be resolved into d- and q-axis components which are
the present method, and was tested experimentally on an stationary with respect to the rotor, but which pulsate at six times
induction motor. Without detailed tests and finite-element the fundamental frequency. (See Tables I and II.). The analysis
calculations, it is impossible to say which is better for PM for higher-order space-harmonics is more complex, but these
machines, but the two methods support each other very roughly. should be attenuated relative to the fundamental.
6
Losses of this type can in principle also occur in the q-axis,
but here they are ignored on the grounds that the q-axis armature-
reaction flux does not pass through the magnets, and in any case
the circuits formed by induced currents in the magnets will be far
less effective in the q-axis.
7
This technique requires a voltage-driven solution and a special IV. RESULTS AND CONCLUSION
formulation in which the circuit equations are incorporated in the
field-solution matrix, [12]. Notwithstanding the excellent The methods presented in this paper are intended for fast
mathematical literature on the finite-element method, a review of calculation of rotor losses in permanent-magnet brushless
ten books on finite-elements in electrical engineering fails to machines. Together with iron-loss calculations [5], they are part
provide any practical reassurance of the validity of this method. of a complete set of calculating methods proposed to deal with
Fig. 7 shows the problem of calculating eddy-currents with a this very complex calculation, and combined with the necessary
conventional 2D solver using current excitation (and no external simulation tools [2] required to handle the external circuit and
circuit connection). The problem is to calculate the no-load eddy- controls. Validation of the 2-dimensional solution of the complex
current losses in the magnets at high speed. The particular diffusion equation is not an issue, because it is a widely used and
example in Fig. 7 uses the time-stepping algorithm of Crank- long established method [1]. Validation of the separate methods
Nicholson, and predicts a total loss of 2@06 W, but this result is for analysing the effect of magnet segmentation is almost
incorrect because the “infinite length” assumption permits impossible by practical measurement because of the difficulty of
induced current in magnet A to return in magnet B, and likewise segregating several coincident phenomena, but the total losses
current in magnet C to return in magnet D. Since there is no can of course be compared with test data. Comparison with
electrical connection between the magnets, this is a false result. finite-element calculations may also give some reassurance but is
(It is not the only example of the possibility of a false result not a satisfactory substitute for physical testing. The methods
obtained from the finite-element method by the unwary user). developed here are used in the companion paper [4] to help
The simple finite-element formulation satisfies (43) only over calculate subtransient parameters for short-circuit analysis.
the whole solution domain, and not individually in each magnet
region. An improved result can be obtained by suppressing the ACKNOWLEDGMENT
conductivity of, say, magnets B, C, and D, and simply calculating
the loss in magnet A. In a situation where the losses must be We would like to thank M. Olaru and the companies of the
equal in all magnet blocks, the final result is obtained by SPEED Consortium for support and discussions.
multiplying the result for one block by 8. When this is done in
Fig. 8, the total calculated loss is 0@97 W when scaled up to
include all magnets. This is slightly less than half the incorrect REFERENCES
first estimate. The method can be described as a crude form of
residual current suppression without using external circuits. [1] T.J.E. Miller and P.J. Lawrenson, Penetration of Transient Magnetic Fields
Through Conducting Cylindrical Structures, with Particular Reference to
This method of single-region eddy-current calculation
Superconducting AC Machines, Proc. IEE, Vol. 123, No. 5, 1976, 437-443
obviously relies on the assumption that the eddy-currents in any [2] T.J.E. Miller, SPEED's Electric Machines 2009, available from the
magnet do not affect the eddy-currents in any other magnet. One SPEED Laboratory
could assume that this is characteristic of “resistance-limited” [3] P.J. Lawrenson, P. Reece and M.C. Ralph, Tooth-ripple losses in solid
poles, Proc. IEE., Vol. 113, No. 4, April 1966, pp. 657-662
eddy-currents, but it is more correct to describe it as the neglect [4] (in press) K.W. Klontz, T.J.E. Miller, H. Karmaker and P. Zhong, Short-
of proximity effect, which in other situations is known to be circuit Analysis of a Permanent-Magnet Generator, IEMDC 2009
dangerous. Nevertheless, engineers in all disciplines have long [5] D.M. Ionel, M. Popescu, M.I. McGilp,T.J.E.Miller,S.J. Dellinger, and R.J.
relied on simplified calculations combined with other means of Heidemann, Computation of Core Losses in Electrical Machines Using
Improved Models for Laminated Steel, IEEE Trans. Ind. Appl. vol. 43, No.
verification than simply throwing time and money at the problem 6, Nov/Dec 2007, pp. 1554-1564.
through very expensive calculating tools; and if this approach is [6] A. Hughes and T.J.E. Miller, Analysis of fields and inductances in air-cored
still permitted, the method is surely worth trying. and iron-cored synchronous machines, Proc. IEE, Vol. 124, No. 2, February
A “refinement”, if it can be so described, is to suppress the 1977, pp. 121-126
[7] R.L. Russell and K.H. Norsworthy, Eddy Currents and Wall Losses in
conductivity judiciously in pairs or patterns of magnets, and Fig. Screened-Rotor Induction Motors, Proc. IEE, Vol. 105A, pp. 163-175, 1958
9 shows an example where magnets A and C are treated as [8] K. Oberretl, Eddy Current Losses in Solid Poles of Synchronous Machines
conductive while magnets B and D are not. The total loss at No-Load and On Load, IEEE Trans., Vol. PAS-91, 1972, pp. 152-160
calculated in this case is 1@02 W, again scaled up to include all [9] N.T. Irenji, S.M.Abu-Sharkh, and M.R. Harris, Effect of rotor sleeve
conductivity on rotor eddy-current loss in high-speed PM machines, ICEM
magnets. This is reasonably close to the 0@97 W calculated using 2000, Espoo, Finland, August 2000, pp. 645-648
the single-block approximation, and the difference may give [10]Z.Q. Zhu, K. Ng, N. Schofield and D. Howe, IEE Proc.-Electr. Power Appl.,
some idea of the accuracy of the method. The only way to be Vol. 151, No. 6, November 2004, pp. 641-650
sure is to measure it, although some confidence might be [11]B. Heller and V. Hamata, Harmonic effects in induction machines, Elsevier
Scientific Publishing Company, Amsterdam, 1977
obtained through more sophisticated calculation tools. [12]S.J. Salon, Finite-Element Analysis of Electrical Machines, Kluwer
An interesting observation in Figs. 7,8 and 9 is the “lateral” Academic Publishers, 1995
diffusion of the eddy-currents towards the edges of the magnets, [13]Jung Jae-Woo et al, Optimum Design for Eddy Current Reduction in
with little variation in current density in a direction parallel to the Permanent Magnet to Prevent Irreversible Demagnetization, ICEMS 2007,
Oct. 8~11, Seoul, Korea, pp. 949-954
flux. This property is used in [4] in the formulation of simplified [14 N. Boules, Impact of Slot Harmonics on Losses of High-Speed Permanent
loss calculations for the IPM, and the related calculation of the Magnet Machines with a Magnet Retaining Ring, Electric Machines and
frequency-dependent synchronous inductance mentioned earlier. Electromechanics, Nov. 1981, pp. 527-539.