You are on page 1of 4

IEEE TRANSACTIONS ON MAGNETICS, VOL. 24, NO.

I, JANUARY 1988 463

NUMERICAL COMPUTATION OF TORQUES IN PERMANENT MAGNET MOTORS


BY MAXWELL STRESSES AND ENERGY METHOD
*
M. Marinescu and N.Marinescu
Ing. -BOro fOr Magnet�echnik. Mail�nder Str.I9. D-6000 FranKfurt 70. FRG
*Institut fOr Theoretische PhysiK. Univ, Heidelberg . D -6900 Heidelberg.FRG

Abstract in determining the derivotion. Nevertheless.


the energy method appears useful, particularly
We present a numerical computation of torques if an examination of the torque's angle depen­
for three different permanent magnet excited dance is of interest anyway. which is always
motors. In all cases the no-load torques have the case when determining the cogging torques,
been derived from the vector potential. compu­ The preparation worK can be minimized if a
ted by using the Finite Element Method. both program for the automatical rotation of the
by directly apPlying the Maxwell Stresses for­ rotor is available, Under these circumstances
mula and as the derIvative of the magnetosta­ the energy method can be considered equivalent
tic energy with respect to the rotatIon angle. to the other methods in respect of calculation
In one Of the cases the on-load torques have effort, Besides, it pravides valuable informa­
also been computed. Instead of the energy, func­ tion on the magnetic processes in the system,
tion we have used a "total work function' at which for instance can lead to a correct in­
constant current. The agreement between the terpretation of the mechanism responsible for
torque values obtained by the two methods is the cogging torques, thus opening ways to
very good, their reduction,
1. Introduction In this paper we present the numerical results
which we obtained for the 3 motors represented
The precise Knowledge of the no-load torque is in the Figs,2a, �, 5h and 6a. The vector po­
of great importance for the design of low tentials have been cOTculated with our FEM pro­
noise permanent magnet excited motors, To this gram MANI running on a 32/16 Bit microcomputer,
aim the FEM can be used as a powerful methOd.
However. it is not always clear if the preci­ 2, Basic Formulas
sion of the numerical results is high enough
In order to be able to make accurate state - a) The Maxwell Stress Tensor
ments about the magnetic noise phenomena of
the machine. It is therefore interesting to A general formula for the torque IS:
compute the forces from the vector potential T =\reX)'BCX)'P(J()'dl' (1)
by using two different methods, The difference
in the results may be taken as a criterion for
the overall precision of the methOd,

In a 2D treatmen the surface integral is re­
duced to a line integral along an integration
In part 2 we shali deduce the formulas applied path that is placed In the air gap [21,
in the computation of energy and torque; part In (1) r(x) is the distance between tne point
3 shows the examined motors and the respective X on the integration path and the rotation
calculation results. axis, e(x) the unit vector in x in aZimuthal
direction and pix) the magnetostatic pressure
Several basically equivalent methods for the on the surface, The latter can be obtained
determination of energy, forces and torques in from Maxwell's Stress Tensor by using the for-
electromagnetic systems have been dis u�sed in
� mula:
detail in various textbooks (see e. g, Ij), p
-

H· (1m) 1-
(FIB)·n (2)
Z
= - J

For implementation In a Finite Elemen context


those methOdS that are suited best derive the with n the unit vector normal to the integra­
force from a system's material and field va­ tion path I in X,
lues in one 'single relative position of the In the case of an exact (ideal) solution the
movable versus the fixed parts. These methods integral (1) is independent of the integration
include the valume integral of the force den­ path chosen. Due to unavoidable discretIzation
sity. the surface integral of Maxwell's Stress errors this characteristic is only rOU hlY �
Tensor f21 or the r�centlY proposed Virtual preserved in a FEM solution (see also 2J,[3]1
Work Ap�roach [3. 4J. Except for the different It always appears reasonable to place .he i�­
discretization errors. these methOds lead to tegration path in the middle Of the grId trI­
essentially the same results in forces and angles' edges, where the values of B are ap­
torques, proximated best ,
In comparison with these methods, which only
require a single numerical treatment of the b) The Energy Method
system, the computation of the force by diffe­
rentiating the energy with respect to a gene­ Forces and torques can be derived both from
ralized caordinate (e. g, displacement or rota­ the energy and the co-energy of a system. The
tion angle) entails the disadvantage of having co-energy seems better suited when the ap­
to determine the energy in several positions proach of the scalar potential is applied,
of the movable parts, This normally requires whereas in the case of a vector PQtential
additional work on the preparation of the grid. approach it is the energy (see r 3j) ,
On top of that. evaluation errors could occur In a closed' system the torque is calculated as
OOI8·9464/88/0100·0463$Ol.OO© 1988 IEEE
464

the derivative of the magnetastatic energy u· and I· are the valtage and the current res­
with respect to the rotation angle 0: pectivel� (the sum inclUdes all windings), A
. a Wm is the vector potentiol, j the current density,
T = - - -- , (3) The magnetastatic energy in (3) and (4) is
a0 calculated as the integral of the energy den­
If the torque is to be determined in rota­ sity w(x) in the entire domain:
tion at a constant current, the motor can no
longer be regarded as a closed system: in or­ W
m
= \
w(x) ·dxdy , (6)
der to keep the total current constant during
the rotor movement, energy has to be supplied
1
with w in princiP e consisting of 3 components:
from outside. The torque in thIS case is the w(x) W (X) + W (X) + W (x).(7)
a f m
=

derivative af a "work function" L with respect The energy density w in media with � =l is
to the rotatian angle 0: formulated as:
a
1 B- H .
-
r

T=-.d..L • w = L
a ' (8)
a0 In soft-magnetic materials (without hystere-
with
(4) sis) w is:
f
�(X)
L = W - W '
Sm wf = � H· dB . (9)
W is - as in (3) -
the magnetostatic energy,
w�ile W stands for the energy supplied from
s � shows w as a hatched area.
autside in order to maintain the tatal cur­
rent constant: TnIhOfd magnet f c materials the magnetic (free)
energy density is:
6w =
s
L' Ui'Ii·5t - ' j(Xl-8A(X),dXdY 2
[ ]
=

i ) (5) w
m = 2� UII BII (x)
; - B
rem
(X) +

W - \j(X)'A(X).dXdY ,
s 1 2
=

+
2�0�1 B1.(x) (10)

�.-
BII and B1. are the companents of B, � and �1.
the relative permeabllities parallel"and per­
pendicular to the easy axis direction.
B em is the residual flux density which has
t5 be defined separately for each different
type of hard magnetic material, as Shown in
FIG.1b,Ctd. Ih these figures the free energy
densIty s dePlctea as hatched areas.
I a) b)

3.Discusslon of the Results


We intended to use the Maxwell Method together
with the Energy Method in the calculation of
several different permanent magnet excited
motars (FfG.2n4.5,6)
and later compare the
results. 0 t IS aim, one has to enable the
inner part of the configuration to rotate au­
tomatically. We define a "rotation line" in
the air gap, divided by the radial lines of
the grid into equidistant azimuthal angle sec­
tions. A specifically developed program can
rotate the inner configuration (grid and mate­
rial distribution) in multiples of one angle
section. In the 2-pole B8 SCH motor (E.lG...)..2..!l
the discretization was 3 so that rotor
Positions could be considered in the determi­
, 6
nation of the cogging torques. These reluc­
tance tsrques recur every half slot pitch,
0
i,e. 15 , In the symmetrical positions 0 0
and 0 = 19
(slot axis opposite pole axis,
respectively tooth axis oPposite pole axis)
=

the reluctance torque disappears, The Maxwell


Method cannot give any information here. The
Energy Method requires the additional compu­
tation of the total energy at t 8e edges of
the interval (0
derivotive is zero.
o
0 and 0
= 5 ), where its = 1
Magnetization characteristics for a The same additional computation is also needed
saft-magnetic material (a), for a per­ for the calculation of the on-load torque.
In the case of the 4-pole motor with Inner
manent magnet with rigid magnetization
(b) and for Alnico (c,d), showing the excitation (FIG.6)
the absolute values of the
torque at-loCidfi'Ove an angle seriodicity of a
definitions of w and w .
f m holf of a pole pitch, 1. d. 45 . AlthOugh in
the functioning motor Doth edges of thIS in­
terval are not reached at full current, the
entire characteristic (�) can be experi­
mentally verified.
465

l.!.0tal magnet 1 c energy (J/m2J

I I !
I

{,!:oli1

I�
, I II
I�
[orr II ,
1

U II
li II II "I Ii ;mirm �
II lit Hillill1J1111111111 1111 III
I I I
o 3 6 9 12 15

tlM
Jfr'.'\
(Nm/lm)
1/f\ IMOTOR BOSCHJ
1'61
, � I
l-t--
\
Computed from tne Energy
" ,
COIllPuted b� uSing tne
V,I ''''"k"t1 Maxwell tresses I

J \\
!j) /:
\
\
'=
I �
J' \
11 I\,
11: � ""<l;
II J:o-. -�
o 3 6 9 12 15
ill.....2. 2-pole motor BOSCH
al Flux distribution at no-load in the
o
rotor Position 0 = 3
bl Cogging torque at no-load

FIG,3 shows the variation of the total energy,


depending on the rotor position in the BOSCH
motor (FIG.2al, The characteristic curve was
obtained from the 6 energy values at hand
using a Spline Interpolation Method where the
dlsappearance of the energy derivotive at the
edges of the interval was made a condition.
One sees that the energy variations are very
small. a fact that led many to the conclu­
sion that the Energy Method allows only an
inexact computation of the torque.
We, though, found a good agreement of the re­
sults of the Maxwell and the Energy Method in
all motors examined, both at-load and no-load
(see FIG,2b. 4b, 5c and 6bl, Had an interpola­ o 5 10 15 20 25 30
tion method been used for the Maxwell values
rn.....!± 2-pole motor JOHNSON
as well. almost the same curve of the torque
al Flux distribution at no-load in
would have been achieved.
the rotor position 0 = 100
b) Cogging torques at no-load
466

Computed from tne Energy

Maxwe II tresses
FIG,6 4-pole motor with inner permanent
magnet excitation
aJ Flux distribution at full load
bJ Torques at full load

5,References

, [1 ] J,A,Stratton, Electromagnetic Theory,


Mc Grow Hill, 1941,
[2 ] K,Reichert, H.Freundl and W,VOgt, "The
1 . 5 3 4. 5 6 7 .5 calculation of forces and torques within
4-pole motor numerical magnetiC field calculation me­
aJ Grid triangulation thods", COMPUMAG 1976, Oxford,
bJ Flux distribution at no-load in J.L,Coulomb and G,Meunier, "Finite element
the rotor position 0 = 3 0 implementation of virtual work principle
c) Cogging torques at no-load for magnetiC or electric force and torque
computation", IEEE Trans, on Magnetics,
4.Conclusions Vol,MAG-20, No,S, Sept,1984, pp,1894-6,
If the interdependence between torque and an­ J,L,Coulomb, "A methodology for the deter­
gle is to be investigated and several FE-col­ mination of global electromechonical quan­
Cuiotions are carried aut in a number of rotor tities from a finite element analysis and
positions, one should olso determine the ener­ its application to the evaluation of ma­
gy in the different motor parts and then cal­ gnetic forces, torques and stiffness",
culate the torque with these data. IEEE Trans, on Magnetics, Vol,MAG-19,No,6,
The variation of the magneticol energy in the Nov,1983, pp,2514-9,
magnet, iron and air can bring helpful iaeas E,M,H,Kamerbeek, "On the theoretiCal and
for reducing the cogging torques. In addition, experimental determinatian of the elec­
a means is available for verifying the torque tromagnetic torque in electrical ma­
values gained according to Maxwell and drawing chInes·, Philips Res, Repts, Suppl"
conclusions about the accuracy of the perfor­ 1970, No,4, pp,102-156.
med calculations,

You might also like