Brushless

Permanent Magnet
Motor Design
Second Edition

Dr. Duane Hanselman

Electrical and Computer Engineering
University of Maine
Orono, ME 04469
USA
brushlessmotor@ieee.org
www.eece.maine.edu/motor

THE WRITERS’ COLLECTIVE

 Cogging Torque Relationships 209

Motor constant as given by (9.11) ignores the winding end turns. Since the end
turns create ohmic losses but do not produce torque, they directly diminish motor
constant. This fact suggests that motor constant increases at a slightly greater rate
with respect to Nm than that described in the preceding paragraph. This occurs
because the length of the end turns are inversely proportional to the number of mag-
net poles since the angular coil pitch is 2π/Nm radM. The presence of end turns also
suggests that motor constant decreases somewhat with increasing outside rotor
radius. This occurs because end turn length is directly proportional to the distance
between slots, which increases somewhat with respect to Rro if Rso is fixed.
In summary, increasing the air gap flux density is the most significant way to
increase motor constant. When the outside stator radius is fixed, the outside rotor
radius plays an important but much less significant role in maximizing motor con-
stant. Increasing the number of magnet poles plays a more significant role than the
outside rotor radius but a less significant role than increasing the air gap flux density.

9.2 Cogging Torque Relationships

Cogging torque, as illustrated previously in Chapter 4, describes the desire of the
permanent magnets on the rotor to align with a maximum amount of ferromagnetic
material. In chapter four it was shown that integral slot motors have greater cogging
torque than fractional slot motors. This occurs because the cogging torques created
by the magnets add in phase alignment in integral slot motors, whereas they are out
of phase with each other in fractional slot motors. In addition to identifying a num-
ber of fundamental ways to minimize cogging torque, skewing the rotor magnets or
stator slots was shown to minimize cogging torque.
The fundamental properties of cogging torque described in Chapter 4 can be used
to quantify the relationship among magnet pole count Nm, slot count Ns, and skew.
The cogging torques experienced by all stator teeth have the same shape, but are off-
set from each other in phase by the angular slot pitch. Furthermore, cogging torque is
periodic with respect to each magnet pole since South magnet poles create the same
cogging torque as North magnet poles. As such, the fundamental frequency of the
cogging torque is twice the fundamental electrical frequency whose period is one
magnet pole pair. As a result, the cogging torque experienced by the kth stator tooth
Tck(θ) can be written as the Fourier series

∞
Tck (θ ) = ∑Te
n =−∞
n
jn 2 (θ − kθ s )
(9.13)

Brushless Permanent Magnet Motor Design, © Duane Hanselman

210 Chapter 9 Performance

where Tn are the Fourier series coefficients, θ is in electrical measure, θs is the angular
slot pitch in electrical measure, and the factor of two in the exponent reflects the fact
that the fundamental cogging frequency is twice the electrical frequency.
Since the cogging torque of each tooth adds to create the net cogging torque of the
motor, the motor cogging torque can be written as

N s −1
Tcog (θ ) = ∑T
k =0
ck (θ ) (9.14)

Substituting (9.13) into (9.14) and simplifying leads to

∞
Tcog (θ ) = ∑ (T Θ
n =−∞
n n ) e jn2θ (9.15)

where the term in parentheses on the right hand side are the net cogging torque Fou-
rier series coefficients and

N s −1
Θn = ∑ek =0
− j 2 nkθ s
(9.16)

where θs= πNm/Ns radE.

The presence of skew modifies the net cogging torque by the skew factor (7.41),
which becomes

 nπ N mα sk 
S2 n = sinc   (9.17)
 Ns 

when expressed in terms of twice the electrical frequency to match (9.15). Applying
(9.17) to (9.15) as described in Chapter 7 gives

∞
Tcog (θ ) = ∑ (T Θ S
n =−∞
n n 2n ) e jn2θ (9.18)

This expression describes the net cogging torque. The tooth Fourier series coeffi-
cients Tn are determined by the magnetic field distribution around each tooth, the air
gap length, and the size of the slot opening between teeth. Minimizing or eliminating

Brushless Permanent Magnet Motor Design, © Duane Hanselman

 Cogging Torque Relationships 211

cogging torque requires setting all Fourier series coefficients (TnΘnS2n) in (9.18) to
zero or minimizing the amplitude of the largest coefficients. Once the tooth Fourier
series coefficients Tn are set by the chosen motor dimensions and magnet properties,
minimizing or eliminating cogging torque requires study of the last two terms Θn
and S2n. For those harmonics where either of these terms are zero, the net cogging
torque harmonic is zero. For example, if Θn is nonzero for some n, the nth cogging
torque harmonic is zero if S2n is zero for this n.
The zeros of S2n are given by the zeros of the sinc function. Because this function is
defined as sinc(x)=sin(x)/x, sinc(x) is zero whenever x is a nonzero multiple of π. For
S2n this occurs when

nN mα sk
q= (9.19)
Ns

where q is any nonzero integer. When there is zero skew, i.e., αsk =0, (9.19) does not
hold for any q or n. This makes sense since S2n =1 for all n when there is zero skew.
The zeros of Θn are not as straightforward to determine. Through careful analysis of
many cases, it can be shown that in general

 lcm ( N s , N m )
Ns for n = q
Θn =  Nm
(9.20)

0 otherwise

where lcm(x,y) is the least common multiple of its arguments and q is any nonzero
integer that results in an integer n. Based on this result, all cogging torque harmonics
are zero except those for which Θn is nonzero. In this case, the tooth cogging torque
harmonic of all teeth simply add, making the net cogging torque a factor Ns greater
than the individual tooth cogging harmonic.
For skew to be effective at eliminating cogging torque the zeros of the sinc function
(9.17) must be zero at the same harmonic indices where Θn is nonzero. There is no
need for the sinc function to be zero where Θn is already zero.
To investigate this possibility, consider the four pole, twelve slot motor illustrated
in the cogging torque analysis conducted in Chapter 4. If a one slot pitch skew is
used, i.e., αsk =1, then the zeros of S2n (9.17) appear at the harmonic indices

n=3q

Brushless Permanent Magnet Motor Design, © Duane Hanselman

212 Chapter 9 Performance

where q is any nonzero integer. For this case, Θn (9.20) is nonzero at the same har-
monic indices. Therefore, the zeros of the sinc function appear exactly at the correct
harmonic indices to cancel cogging torque harmonics that would otherwise be non-
zero. In this case, a one slot pitch skew eliminates the cogging torque.
As another example, consider the four pole, fifteen slot motor also considered in
Chapter 4. If a one slot pitch skew is used, then the zeros of S2n (9.17) appear at the
harmonic indices
n=15q

where q is any nonzero integer. For this case Θn (9.20) is nonzero at the same har-
monic indices. Therefore, a one slot pitch skew eliminates the cogging torque for this
case as well. In fact, a one slot pitch skew always eliminates the cogging torque. For
this particular four pole, fifteen slot case, skews of one half and one quarter slot pitch
also eliminate the cogging torque.
Because the indices in both (9.19) and (9.20) are harmonically related, i.e., if n satis-
fies both relationships, then all multiples of n satisfy them also. Therefore, the mini-
mum skew required to eliminate cogging torque can be determined by finding the
first index n where both relationships hold. Doing so gives the minimum skew to
eliminate cogging torque as

∗ Ns
α sk =
lcm ( N s , N m ) (9.21)

This minimum skew may be easier to implement mechanically and will lead to less
smoothing of the tooth flux and resulting back EMF waveforms.
When skew is not used, (9.20) can be used to identify the harmonic index of the first
cogging torque harmonic. If this index is ncog, then the cogging torque has a funda-
mental frequency 2ncog greater than the electrical frequency of the motor. The higher
this harmonic frequency is, the less objectionable the cogging torque will be for two
reasons. First, because the Fourier series coefficients of smooth functions generally
decrease in amplitude as harmonic index increases, the higher ncog is the lower the
resulting cogging torque should be. And second, the higher the frequency is, the
more the motor inertia filters or smoothes out the cogging torque ripple.
Study of (9.20) shows that for motors having an integral slot pitch, i.e., Ns/Nm =q
where q is an integer, the first cogging torque harmonic index is ncog =q. This repre-
sents the worst case situation. On the other hand, for motors having a fractional slot
pitch, where Ns and Nm share no common factor, ncog =Ns. This is the best case situa-

Brushless Permanent Magnet Motor Design, © Duane Hanselman

 Cogging Torque Relationships 213

tion. Finally, when Ns and Nm share a common factor, i.e., when the greatest common
divisor gcd(Ns,Nm) is equal to some integer p, structural periodicity exists around the
air gap every 360/p°M and ncog =Ns/p.
Given the net cogging torque Fourier series (9.18), computing the cogging torque
requires computation of the cogging torque experienced by a single tooth (9.13)
under zero skew conditions. Once this is known, the Fourier series coefficients Tn are
easily computed. Computation of the cogging torque requires knowledge of the mag-
netic field entering the shoe tips in the tangential direction. That is, Bθ(r,θ) is required
in the slot openings between the stator teeth. Given this field, the tangential force
2
density in N/m on the stator teeth can be shown to be

Bθ2 ( r ,θ )
fθ ( r ,θ ) = (9.22)
2 µo

The torque density associated with this force density is given by the product of the
force density and the radius at which it acts, i.e., T=Fr, or

Bθ2 ( r ,θ )
tθ ( r ,θ ) = r (9.23)
2 µo

Integration of this torque density over the cross-sectional area of the shoe tips on
both sides of a tooth gives the cogging torque. Since the force experienced on oppo-
site sides of the tooth act in opposite directions, the cogging torque is given by the
difference in force experienced by each tooth side.
The simplest way to approximate Bθ(r,θ) is to use the circular-arc, straight-line flux
flow model as shown in Fig. 9-3. With this assumption, Bθ(r,θ) is equal to the corre-
sponding radial magnetic field Bar(Rs,θ) as given in (7.1) as modified by the slot cor-
rection factor (7.9), which is illustrated in Fig. 7-3.
Implementation of this technique for predicting cogging torque is straightforward,
but the details involved are cumbersome. Because cogging torque is highly depend-
ent on the exact form of Bθ(r,θ), which is influenced by magnetic saturation of the
shoe tip material, the accuracy of this prediction may not agree well with experimen-
tal measurements. As a result, the preceding analysis identifying harmonic content
relationships of the cogging torque provides significant insight that is independent of
the exact magnetic field distribution in the stator slots.

Brushless Permanent Magnet Motor Design, © Duane Hanselman

214 Chapter 9 Performance

Figure 9-3. Circular-arc, straight-line flux path approximation.

In summary, the cogging torque analysis conducted here provides guidance in the
selection of the number of magnet poles Nm, the number of stator slots Ns, and in the
amount of skew required in a motor design. Fractional slot motors where Nm and Ns
do not share a common factor exhibit the lowest cogging torque under no skew con-
ditions. In addition, as Nm and Ns increase, the higher the fundamental cogging
torque frequency becomes, making it easier for the motor inertia to filter it out.

9.3 Radial Force Relationships

In addition to undesirable tangential force, i.e., cogging torque, a motor may experi-
ence an undesirable radial force between the rotor and stator that varies as the rotor
rotates. As in the preceding cogging torque derivation, the force per tooth provides
the basis for further analysis. In this case, the radial force experienced by the kth sta-
tor tooth Frk(θ) can be written as the Fourier series

∞
Frk (θ ) = ∑Fe
n =−∞
n
jn 2 (θ − kθ s )
(9.24)

where Fn are the Fourier series coefficients, θ is in electrical measure and θs is the
angular slot pitch in electrical measure. The factor of two in the exponent reflects the
fact that the fundamental radial force frequency is twice the electrical frequency
because the radial force is the same over both South and North magnet poles.
The radial force on the kth tooth is directed at an angle kθsm, where θsm is the angu-
lar slot pitch in mechanical measure. As a result, the net force experienced by the

Brushless Permanent Magnet Motor Design, © Duane Hanselman

 Radial Force Relationships 215

rotor due to all teeth can be found by summing the x- and y-direction components for
each tooth. Doing so leads to a net x-direction force of

N s −1 ∞
Fx = ∑
k =0
Frk (θ ) cos ( kθ sm ) = ∑FX e
n =−∞
n n
j 2 nθ
(9.25)

where

N s −1
Xn = ∑ cos ( kθ
k =0
sm ) e − j 2nkθ se
(9.26)

The net y-direction force is

N s −1 ∞
Fy = ∑F
k =0
rk (θ ) sin ( kθ sm ) = ∑FYe
n =−∞
n n
j 2 nθ
(9.27)

where

N s −1
Yn = ∑ sin ( kθ
k =0
sm ) e − j 2nkθ se
(9.28)

The effect of skew (9.17) can be included in (9.25) and (9.27). However, skew does not
eliminate radial force but introduces an axial variation in the radial force and reduces
its amplitude. Since this phenomenon is secondary to this discussion, skew is not
considered further.
Based on (9.25) and (9.27) the rotor experiences zero radial force if both (9.26) and
(9.28) are zero for all harmonic indices n. These two expressions are zero whenever
there is geometrical symmetry between the rotor and stator around the air gap.
Stated mathematically, the rotor experiences zero net radial force whenever

gcd ( N m , N s ) > 1 (9.29)

where gcd(x,y) is the greatest common divisor of its arguments. More specifically,
when gcd(Ns,Nm)=p where p>1, symmetry exists between the rotor and stator that
repeats every 360/p°M. For example, in the four pole, twelve slot case considered ear-
lier, gcd(12,4)=4 and the symmetry between the rotor and stator repeats every 90°M.

Brushless Permanent Magnet Motor Design, © Duane Hanselman

216 Chapter 9 Performance

Since the number of magnet poles is always an even number, there is zero net radial
force whenever the number of slots is also an even number. Equation (9.29) is always
met in this case. On the other hand, when the number of slots is an odd integer, the
rotor experiences a net radial force whenever (9.29) is not met. For example, in a four
pole, fifteen slot motor, gcd(15,4)=1 and a net radial force exists. However, in a six
pole, twenty-one slot motor, gcd(21,6)=3, and zero radial force exists.
In those cases where a net radial force appears, it is beneficial to know the harmonic
indices that contribute to the net force since these indices identify the relative speed
at which the net radial force rotates with respect to the rotor speed. The harmonic
indices that contribute to the net radial force satisfy

( 2 q − 1) N s ± 1
n= (9.30)
Nm

where q is a positive integer. The radial force at all other harmonic indices is zero.
The harmonic indices that satisfy (9.30) do not exist for all q, nor are they uniformly
spaced. For example, in a four pole, fifteen slot motor, the harmonics that satisfy
(9.30) are n=4,11,19,26,…, where the difference between indices alternates between
7 and 8. For a ten pole, thirty-three slot motor, the harmonic indices that satisfy (9.30)
are n=10,23,43,56,76,…, where the difference between indices alternates between
13 and 20. In both of these examples, the first contributing harmonic index is equal to
Nm. However, this is not always true. For example, in an eight pole, twenty-slot
motor, the harmonic indices that satisfy (9.30) are n=10,17,37,44,…, where the first
contributing index is 10 rather than 8.
In addition, it is true in general that the sum of the two numbers describing the
alternating differences between indices for which nonzero radial force exists always
equals the number of slots Ns. For example, in the four pole, fifteen slot motor the
two numbers are 7 and 8, which sum to Ns =15.
When Xn in (9.26) is nonzero, it always equals Ns/2. Similarly, when Yn in (9.28) is
nonzero, it equals Ns/2 for those indices for which (9.30) is satisfied with the minus
sign. And it equals –Ns/2 for those indices for which (9.30) is satisfied with the plus
sign. For those cases in which radial force appears, it can be computed by determin-
ing the radial force experienced by a single tooth. From the Fourier series coefficients
of this force Fn, equations (9.25) through (9.28) give the desired net radial force in the
x- and y-directions. The radial force on one tooth can be computed by integrating the
2
radial force density in N/m

Brushless Permanent Magnet Motor Design, © Duane Hanselman

 Core Losses 217

2
Bar ( Rs ,θ )
fr (θ ) = (9.31)
2 µo

over the surface of a tooth. In (9.31), Bar(Rs,θ) is given by (7.1) evaluated at the stator
radius Rs at the air gap.
In summary, motors that do not have symmetry between the rotor and stator
around the air gap will exhibit a net radial force having harmonic content above the
fundamental electrical frequency of the motor. Since cogging torque is generally
worse for motors having this symmetry, there is a tradeoff between minimum cog-
ging torque and the presence of net radial force. In most applications, net radial force
has little impact on performance. However, in low noise applications such as hard
disk drive spindle motors, the presence of net radial force can limit motor acoustic
performance.

9.4 Core Losses

Basic Concepts
2
In most motor designs, I R losses are the dominant contributor to reduced energy
conversion efficiency. Core losses in the ferromagnetic portions of the motor are usu-
ally the next largest contributor to motor losses. As described qualitatively in Chapter
2, core loss is the sum of hysteresis and eddy current losses. These two core loss com-
ponents have been studied extensively for many years. Some studies have focused on
understanding and modeling the phenomena at the atomic level. Others have
focused on developing core loss expressions that facilitate core loss prediction in
actual devices such as motors and transformers. Still others have focused on develop-
ment of new material testing procedures that facilitate accurate curve fitting of meas-
ured material properties to core loss expressions.
While the physical mechanisms that create core losses are well understood, applica-
tion of this knowledge to make accurate core loss predictions remains difficult for a
number of reasons. These reasons include:
• Core losses are created on a microscopic scale within a material, whereas
core loss prediction uses a macroscopic scale based on assumptions of mag-
netic field uniformity throughout regions of the device. Therefore, the accu-
racy of core loss predictions depends on how well the chosen macroscopic
regions model the material loss properties on a microscopic scale.

Brushless Permanent Magnet Motor Design, © Duane Hanselman

218 Chapter 9 Performance

• Core loss data for many common materials is only available for 50 or 60Hz
operation using sinusoidal excitation. This 50 or 60Hz excitation does not
produce an assumed ideal sinusoidal magnetic field when using the data.
• Core loss data obtained from the commonly accepted Epstein Square Test is
often of questionable value because standard test conditions do not match
those of an actual motor.
• Material properties can vary by as much as 30% from batch to batch and
within the same batch. In addition, they can vary within individual lamina-
tions.
• Core loss is significantly influenced by the mechanical stress and strain
experienced both within the material and on its surfaces and edges. For this
reason, motor laminations are often annealed after being stamped or cut.
Since core loss predictions uniformly ignore this material influence, core
loss predictions for nonannealed or improperly annealed laminations are
invariably inaccurate.
• Since the time variation of the magnetic field distribution within a motor is
seldom sinusoidal, the accuracy of core loss predictions depends on how
sinusoidal excitation data is used to generate core loss expressions that
apply to nonsinusoidal magnetic fields.
• Core losses are easier to predict with accuracy in regions where only the
amplitude of the magnetic field changes with time, e.g., in the stator tooth
bodies and stator yoke sections. In regions where both the amplitude and
angular direction of the magnetic field varies with time, an additional core
loss component appears. That is, in addition to traditional hysteresis and
eddy current loss components, the material exhibits additional rotational
losses. This additional loss mechanism is highly dependent on the trajectory
taken by the amplitude and angular motion of the magnetic field in a
region. In a motor, these additional rotational losses are created in the tran-
sition area between the stator teeth and the stator yoke. Studies have shown
that rotational losses can double the core losses in a region. Because of the
significant work required to identify magnetic field trajectories versus time
and to convert this knowledge into viable core loss expressions that vary
with the trajectory shape, rotational losses are often neglected or are crudely
estimated.
• Some materials exhibit an additional loss component in addition to hystere-
sis and eddy current components. This loss component, called the excess or

Brushless Permanent Magnet Motor Design, © Duane Hanselman

 Core Losses 219

anomalous loss, is least understood, is difficult to determine from traditional

core loss data, and is commonly ignored when predicting core losses.
Given all of these issues, core loss prediction using relatively simple modeling may
indicate the correct trends from one motor design to the next but will not likely pro-
duce accurate estimates of core losses at any given operating point.
Using knowledge of the fundamental principles that cause core losses, they can be
reduced by:
• Reducing the lamination thickness. Ideally, eddy current losses are directly
proportional to the square of the lamination thickness. Therefore, if lamina-
tion thickness is reduced by a factor of two, eddy current losses decrease by
a factor of four.
• Increasing the resistivity of the lamination material. Eddy current losses are
directly proportional to material resistivity. Adding silicon to lamination
steel is the most commonly adopted approach to increasing material resis-
tivity.
• Annealing laminations after they have been stamped or cut. This eliminates
the influence of mechanical stress on core loss.
• Reducing the amplitude of the magnetic field within the material. Hystere-
sis losses are directly proportional to the amplitude of the magnetic field
raised to a power between 1.5 and 2.5. Eddy current losses are directly pro-
portional to the square of the magnetic field amplitude. Using this property
to reduce core loss is in direct conflict with maximizing torque production.
As a result, other techniques for minimizing core losses are often imple-
mented first.
• Reducing the number of magnet poles Nm. Hysteresis losses are directly
proportional to the fundamental electrical frequency. Eddy current losses
are directly proportional to the square of the fundamental electrical fre-
quency. Since the fundamental electrical frequency is Nm/2 times greater
than the motor shaft speed, reducing the magnet pole count allows one to
reduce core losses significantly without lowering the motor shaft speed.

Core Loss Modeling

With the preceding information in mind, this section illustrates one approach to esti-
mating core losses in a motor. The process develops a core loss expression that is fit
to standard core loss data. Then, using expressions for the stator tooth and yoke flux
densities, core losses are estimated.

Brushless Permanent Magnet Motor Design, © Duane Hanselman