See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3111060

Optimal design of a coreless stator axial flux permanent-magnet generator

Article  in  IEEE Transactions on Magnetics · February 2005

DOI: 10.1109/TMAG.2004.840183 · Source: IEEE Xplore

CITATIONS READS
158 8,081

4 authors, including:

Rong-Jie Wang Jacek F Gieras

Stellenbosch University University of Technology and Life Sciences in Bydgoszcz
98 PUBLICATIONS   1,590 CITATIONS    291 PUBLICATIONS   3,472 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

High performance electric motors for electrified transportation View project

What are you working on right now? View project

All content following this page was uploaded by Jacek F Gieras on 03 July 2013.

The user has requested enhancement of the downloaded file.

IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005 55

Optimal Design of a Coreless Stator Axial Flux

Permanent-Magnet Generator
Rong-Jie Wang, Member, IEEE, Maarten J. Kamper, Member, IEEE, Kobus Van der Westhuizen, and
Jacek F. Gieras, Fellow, IEEE

Abstract—This paper describes a hybrid method for calculating

the performance of a coreless stator axial flux permanent-magnet
(AFPM) generator. The method uses a combination of finite-ele-
ment analysis and theoretical analysis. The method is then incor-
porated into a multidimensional optimization procedure to opti-
mally design a large power coreless stator AFPM generator. The
measured performance of the manufactured prototype compares
favorably with the predicted results. The design approach can be
applied successfully to optimize the design of the coreless stator
AFPM machine.
Index Terms—Axial flux, design, finite-element methods, opti-
mization methods, permanent-magnet generator.

I. INTRODUCTION

A XIAL FLUX permanent-magnet (AFPM) machines with

coreless stators are regarded as high-efficiency machines
for distributed power generation systems [6], [13], [19]. Because
Fig. 1. Basic structure of the AFPM machine with a coreless stator. 1: Stator
winding. 2: Steel rotor. 3: PMs. 4: Frame. 5: Bearing. 6: Shaft.

of the absence of core losses, a generator with this type of de- representation of the magnetic field inside the machine, enabling
sign can potentially operate at a higher efficiency than conven- nonlinearity to be accounted for with great accuracy, it has been
tional machines. Besides, the high compactness and disk-shaped pointed out in the literature [5], [7], [20] that two-dimensional
profile make this type of machine particularly suitable for me- (2-D) FEM should be incorporated into design optimization of
chanical integration with wind turbines and internal combustion AFPM machines. However, there are no published works de-
engines (ICE), e.g., as integrated starter-generators. scribing the implementation of this approach to the design of
A schematic drawing of a typical coreless stator AFPM ma- AFPM machines in detail.
chine is shown in Fig. 1. The machine consists of two outer In this paper, the equivalent circuits of a coreless AFPM gen-
rotor disks and one coreless stator in the middle. On the two op- erator are first established (Section II). The calculation of cir-
posing rotor disks, there are surface-mounted permanent mag- cuital parameters by using both FEM (Section III) and classical
nets (PMs). The coreless stator winding consists of a number theory (Section IV) are then described. The performance calcu-
of single-layer trapezoidal-shaped coils. These coils have the lation of the coreless stator AFPM generator is explained in Sec-
advantages of being easy to make and having relatively short tion V, which is then incorporated into a multidimensional op-
overhangs. The coils are held together and in position by using timization procedure to optimally design a large power coreless
a composite material of epoxy resin and hardener. stator AFPM generator (Section VI). Some important mechan-
So far, most published works regarding design optimization ical design aspects are discussed in Section VII. The measured
of AFPM machines have been limited to maximizing (mini- performance of the manufactured prototype are compared with
mizing) an objective function with respect to a single variable the predicted results in Section IX. It is shown that the proposed
[3], [4], [9], [10], [15] using analytical methods. The finite-ele- design approach can be applied successfully to optimize the de-
ment method (FEM) is, in many instances, used merely to inves- sign of the coreless stator AFPM machine.
tigate certain design aspects. Since FE models give an excellent
II. EQUIVALENT CIRCUITS
Manuscript received May 25, 2004; revised October 29, 2004. This work was
supported by the University of Stellenbosch and SA industry. To calculate the performance of the AFPM machine, it is es-
R.-J. Wang and M. J. Kamper are with the Department of Electrical En- sential to consider the equivalent circuits of the machine. The
gineering, University of Stellenbosch, Matieland 7602, South Africa (e-mail: fundamental per phase equivalent circuit of a coreless AFPM
[email protected]; [email protected]).
K. Van der Westhuizen is with the Department of Mechanical Engi- machine with the same reluctance for the magnetic flux in the
neering, University of Stellenbosch, Matieland 7602, South Africa (e-mail: and axis may be represented by the electric circuit shown
[email protected]). in Fig. 2(a). In this circuit, is the stator resistance, is the
J. F. Gieras is with the United Technologies Research Center, East Hartford,
CT 06033 USA (e-mail: [email protected]). stator inductance, is the induced electromotive force (EMF)
Digital Object Identifier 10.1109/TMAG.2004.840183 due to the fundamental air-gap PM flux linkage, and and
0018-9464/$20.00 © 2005 IEEE
56 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005

Fig. 2. Per-phase equivalent circuits of an AFPM machine.

Fig. 3. d- and q -axis equivalent circuits and phasor diagram of the AFPM machine.

are the fundamental instantaneous phase voltage and current, equivalent circuit as shown in Fig. 2(b). In this way, the part of
respectively. The shunt resistance is the stator eddy-current the equivalent circuit marked by the dotted lines in Fig. 2(b) can
loss resistance. be accurately calculated by directly using FEM instead of ap-
The synchronous inductance consists of the armature re- proximate inductance equations.
action (mutual) inductance and the leakage inductance The corresponding steady-state - and -axis equivalent
as shown in Fig. 2(b), where , , and circuits of the AFPM machine in the rotor reference frame
are leakage inductance, differential leakage inductance about are shown in Fig. 3. The flux linkages and are the
the radial portion of conductors, and end winding leakage in- - and -axis total stator flux linkage components. These
ductance, respectively. Unlike conventional slotted machines, flux linkages include the flux linkage due to the permanent
there is no clear definition for main and leakage inductances magnets, , and the flux linkage due to stator currents,
in a coreless or slotless machine [1], [11], [17]. It is generally , but exclude the end-winding flux linkage,
difficult to derive accurate analytical expressions for , , . The parameter is the electrical speed of the rotor
and . With 2-D FE analysis, both mutual and leakage flux reference frame. In the phasor diagram (unity power factor
linkages can be readily taken into account. The only remaining was assumed), the space phasors and represent the
part is the end-winding leakage flux. The synchronous induc- stator terminal voltage and current, respectively. Note that
tance of the coreless machine may thus be split into two terms: includes the equivalent eddy-current loss component . It has
1) and 2) end connection leakage inductance . also been assumed that the eddy-current losses in the PMs
As an approximation, may be shifted to the left of in the and rotor disks are negligible.
WANG et al.: OPTIMAL DESIGN OF A CORELESS STATOR AFPM GENERATOR 57

Fig. 5. Flux distribution in a coreless stator AFPM machine.

to obtain the nodal magnetic vector potentials of the model.

Fig. 4. 2-D FE model of a coreless stator AFPM machine. Fig. 5 shows the flux plot of a coreless stator AFPM machine.
The total three-phase flux linkages , excluding end-winding
flux linkages, are then computed in the FE program as follows
III. CALCULATION OF EQUIVALENT CIRCUIT [18]:
PARAMETERS USING FEM
This section describes how the equivalent circuit parameters (2)
of Fig. 3 are calculated by using FEM.

A. Finite-Element Model In (2), is the nodal value of the magnetic vector potential
of the triangular element , or indicates the
The 2-D FE modeling of an AFPM machine is usually carried
direction of integration either into the plane or out of the plane,
out by introducing a radial cutting plane at the average radius,
is the area of the triangular element , is the total number
which is then developed into a 2-D flat model. Owing to the
of elements of the meshed coil areas of the phase in the pole
symmetry of an AFPM machine, each half of the machine from
region, is the number of parallel circuits (current paths) per
the center plane mirrors the other half in axial direction. It is
phase, is the total number of elements of the in-going and
possible to model only half of the machine comprising the rotor
out-going areas of the coil, , and are the number of turns,
disk, the air-gap clearance, and half of the stator. The air-gap
length, and area of a coil, respectively.
region is modeled using the Cartesian air-gap element (CAGE),
From a machine design perspective, it is of main interest to
as described in [27]. By assigning negative periodic boundary
find the fundamental components of the total flux linkages. For
conditions to the left and right boundaries, it is sufficient to
a coreless stator AFPM machine, the flux linkage harmonics
model only one pole-pitch of the machine. Fig. 4 shows an FE
due to iron stator slots and magnetic saturation are absent.
mesh coupled with a CAGE for such a model, which spans one
Owing to a large air gap, the harmonics caused by stator
pole-pitch of the AFPM machine. For an AFPM machine with
winding MMF space distribution are negligible. The most
coreless stator, there is no tangential field component on the
important flux linkage harmonics needed to account for are
center plane of the stator so that the Neumann boundary con-
those due to the flat-shaped PMs.
dition can be assigned to the top boundary.
Given these considerations, the flux linkage wave of an
AFPM machine is nearly sinusoidal, though, for a nondis-
B. Calculation of Flux Linkages
tributed winding, an appreciable third and less significant
To calculate the flux linkage using the FEM, it is necessary fifth and seventh harmonics are still present in the total flux
to specify the phase current of the AFPM machine. linkage waveform. If the fifth, seventh, and higher harmonics
The amplitude of the current space phasor may be determined are ignored, the fundamental total phase flux linkages can be
from a given copper loss , which is predetermined based on calculated by using the technique given in [18], i.e.,
the thermal analysis of the machine, by using
(3)
(1) where the co-phasal third-harmonic flux linkage, including the
higher order triple harmonics, can be obtained from
in which a current angle of has been assumed for
balanced resistive load. With the and current component (4)
amplitudes known, the instantaneous three-phase currents ,
which need to be put in the FE program according to the rotor The use of (3) and (4) is of great importance in the optimiza-
position, can be calculated using the inverse Park transforma- tion process as it enables the fundamental total phase flux link-
tion. The defined FE model is solved by using a nonlinear solver ages of the AFPM machine to be determined by using just one
58 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005

set of field solutions. The fundamental flux linkages are the basis main flux and fundamental operating frequency of the machine.
of subsequent performance calculation of the machine. The eddy-current losses (for round conductors) are calculated
With the fundamental total phase flux linkages and rotor po- by using [8]
sition known, the flux linkages are calculated using Park’s
transformation as follows [12]: (10)
(5)
where is the conductor length, is the fundamental
where frequency, is the diameter of the conductor, is the total
number of conductors in the machine, and and are the pe-
ripheral and axial components (peak values) of the fundamental
(6) flux density wave, respectively. The values of the flux density
components can be obtained from the FE field solution.

From this, the speed dependent voltages and C. End-Winding Inductance

of the equivalent circuits are determined.
The end-winding inductance is calculated by using an
analytical approach based on
IV. EQUIVALENT CIRCUIT PARAMETERS CALCULATED BY
CLASSICAL THEORY
(11)
In this section, the calculation of the remaining equivalent cir-
cuit parameters of Fig. 3 such as winding resistance, eddy-cur- where is the number of pole pairs, is the number of coils
rent resistance, and end-winding inductance by using classical per pole per phase, is the length of the single-sided end con-
theory is described. nection, and can roughly be estimated from the following
semianalytical equation [14]:
A. Stator Winding Resistance
The temperature-dependent stator winding resistance per (12)
phase is calculated as

V. PERFORMANCE CALCULATION
(7)
From the current components and the end-winding leakage
where is the number of turns in series per phase, is the inductance , the end-winding leakage flux linkage speed de-
electric conductivity of the wire at temperature , and is the pendent voltages and of Fig. 3 are determined.
cross section area of the wire. The skin effect has not been taken The terminal voltage components, and , and the voltage
into account in (7) as thin parallel wires (0.42 mm diameter) amplitude are calculated from
were used to minimize this effect in the design.

B. Eddy-Current Resistance (13)

For an AFPM machine with a coreless stator, associated iron
losses are absent. The core losses in the ferromagnetic rotor The power factor is easily calculated from the voltage
disks (back irons) are also negligible due to low flux variation. components, and , and current components, and ,
However, the eddy-current losses in the stator winding are sig- as follows:
nificant due to the high pole number rotor that may spin at rel-
atively high speeds. The shunt resistance may be calculated (14)
in the same way as that of the core loss resistance described in
[16], [18] to account for the eddy-current losses, i.e.,
The generated kVA of the machine is calculated as

(8) (15)

where is the rms value of phase EMF (see Figs. 2 and 3) and The steady-state electromagnetic torque of the AFPM ma-
is given by chine can be calculated by using the following relation:

(16)
(9)
The total input shaft power of the generator can then be cal-
A detailed treatment of the calculation of eddy-current losses culated by
in AFPM machine has been given in [24]. As an approxima-
tion, one may consider only the eddy-current losses due to the (17)
WANG et al.: OPTIMAL DESIGN OF A CORELESS STATOR AFPM GENERATOR 59

where is the angular speed of the machine, and are the uation. Each bit of the probability vector is updated based on
windage and friction losses, which can be estimated from [23] update rule of competitive learning, i.e.,

(19)
(18)
where is the probability of generating a one in the bit position
where is the rotation speed in revolutions per second (r/s), is , is the th position in the solution vector that the probability
the friction coefficient, and is the density of cooling medium. vector is being pushed toward, and is the learning rate, which
The active output power is calculated by using . is the amount the probability vector is changed after each cycle,
The efficiency is then given by . The learning rate has a significant effect on the convergence
This concludes the calculation of the equivalent circuit and speed.
performance parameters of the AFPM machine. This calculation After each update of the probability vector, a new set of so-
method is used by the optimization algorithm described in the lution vectors is created. As the search progresses, the values in
next section. the probability vector start to move toward either 0 or 1 repre-
senting a high evaluation solution vector. The use of mutation
in PBIL is for the same reason as in the GA, i.e., to prevent pre-
VI. OPTIMIZATION mature convergence.
3) Constrained Optimization: To transform constrained op-
This section describes the design optimization of the AFPM
timization problems into unconstrained ones, the penalty func-
machine. The aim of the optimization procedure is to minimize
tion is used together with Powell’s method. The objective func-
the amount of PM material used or maximize the efficiency of
tion is modified by adding terms or functions that penalize any
the machine, while ensuring a rated output power, acceptable
increased constraint violation. The resultant objective function
current density, and desired phase voltage.
is then
A. Optimization Algorithms
(20)
Two different optimization algorithms, i.e., Powell’s method
and the population-based incremental learning (PBIL) algo-
rithm, are used in this paper for the unconstrained design where is the function to be minimized, are weighting
optimization of the AFPM machine. The reasons for using factors, and are functions which penalize increased con-
these methods are: straint violation.
• to compare the effectiveness of the linear maximization Owing to the nature of the stochastic search, PBIL al-
(minimization) method (Powell’s method) with that of a gorithms do not require the use of penalty functions in the
stochastic method (PBIL); objective functions.
• to verify the optimum design results by using two com-
B. Variables
pletely different algorithms.
1) Powell’s Method: Powell’s method is basically an iter- The geometric layout of an AFPM machine with a coreless
ative method. Each th iteration of the procedure maximizes stator is shown in Fig. 6. Only five variables of the machine are
(minimizes) the objective function along linearly independent selected. These are the PM thickness , magnet width to pole
directions, . The initial set of vector directions pitch ratio , stator winding thickness , rotor disk inner
are the coordinate directions. After each iteration, a new direc- radius , and the air-gap clearance . For the specific applica-
tion is defined which is used to form the vector directions for the tion, the rotor outer radius is limited to 360 mm and the typ-
next iteration. After iterations, a set of mutually conjugate ical operating speed is about 2000 rpm. The number of parallel
vector directions are obtained so that the maximum (minimum) circuits per phase and the number of poles are
of a quadratic function is found. predefined. The comparison done in previous studies [21] re-
To avoid linear dependence and premature termination in the veals that the design of an AFPM machine using purely electro-
optimization, specially designed tests have been incorporated magnetic calculations without taking into account mechanical
into the algorithm. A detailed explanation of this method is strength requirements may lead to an unrealistically thin rotor
given in [18] and [22]. disk. To rectify this problem, mechanical strength analysis is of
2) PBIL Algorithm: PBIL is a method combining genetic al- great importance in determining the thickness of the rotor disk
gorithms (GA) and competitive learning for function optimiza- . The FE analysis of the mechanical strength of the rotor disks
tion [2], [14]. The algorithm attempts to generate a probability is described later.
vector, which is then sampled to produce the next generation’s
population. Unlike GAs, operations of PBIL act directly on the C. Objective Functions
probability vector instead of population. To maintain the most The copper losses are kept constant in the design optimization
diversity, each bit position of the probabilities is set to 0.5 at the program. An iterative procedure, making use of the thermo-fluid
beginning. A number of solution vectors are generated based model established in [25], has been used to determine the max-
on the probabilities of the probability vector. The probability imum allowable losses that the machine can handle. The esti-
vector moves toward the solution vector with the highest eval- mated allowable full-load copper loss is about 2.5 kW for a rated
60 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005

Fig. 6. Geometric layout of AFPM machine showing design variables: (a) a linearized section of the radial cutting plan, and (b) a rotor disk with PMs.

power of 150 kW [26]. The maximum allowable current density vector potentials. The machine performance parameters
is set to 10 A/mm in the design program. It has been confirmed are calculated using flux linkages and circuital equations in post-
experimentally that the AFPM machine can withstand this cur- processing as described in Section III.
rent density. The performance parameters to be optimized have Powell’s method requires an initial value for each of the vari-
been selected as the mass of the PM material and the efficiency. ables. If it is too far from the real optimum, then the optimization
1) Optimize for Minimal Mass of PM Material: The opti- may end up being trapped in a local optimum in the vicinity of
mization problem for minimizing the total mass of the PM mate- the initial value [28], which will lead to the necessity of testing
rial, , can be expressed as subject to the following with different sets of starting values to verify the optimum point.
constraints: When the PBIL algorithm is used, it does not really matter
what starting values are used. A total of 30 sample bits, 6 bits per
independent variable, were used in the optimization. The step
(21)
sizes for the variables are 0.05 mm for the air-gap , 0.01 mm
for the PM height , 0.1 mm for the stator thickness, 0.5 mm
where is the total mass of the PM material used, is the for the rotor inner radius , and 0.005 for the PM width to pole
desired output power, is the maximum allowed current den- pitch ratio. The number of bits and step size per variable were
sity, and is the maximum rms phase voltage at rated output chosen to ensure the largest feasible range. The stopping rule of
power. This criterion is almost equivalent with the minimization the PBIL algorithm is that the optimization cycles have to reach
of cost as PMs are expensive parts in a machine. a preset number of generations.
2) Optimize for Maximum Efficiency: The optimization During the optimization process, the mesh of the FE model
problem for maximizing the machine’s efficiency, , can be changes as the optimization progresses. Occasionally, some of
expressed as subject to the constraints the elements may be badly shaped or ill conditioned resulting
in poor accuracy or even no solution. It is therefore necessary
to check that the model dimensions are reasonable before the
FE mesh is constructed. A thermo-fluid model described in [25]
(22) is also incorporated in the optimization process to predict the
temperature distribution in various parts of the machine and to
thus check the validity of the design.
where is the machine’s efficiency at rated output power, ,
and is the maximum allowable mass of the PM material E. Results of the Optimized AFPM Machine
used at . Starting from the same initial design, the optimization was
done using both Powell’s method and the PBIL algorithm ac-
D. Optimization Procedure cording to the two different design objectives, i.e., maximum
The overall design methodology presented in the paper is to efficiency or minimum PM material, respectively. A compar-
use a combination of the classical circuit model and FE field so- ison of the effectiveness between the two methods was also
lutions directly in a multidimensional optimization procedure. done. Table I shows the results of a maximum efficiency ma-
The basic structure of the approach is shown in Fig. 7. The opti- chine design of a 150 kW machine. Both optimization algo-
mization algorithm searches for the machine variables that rithms give similar results. For the design of minimum PM ma-
minimizes (maximizes) the function value . In each itera- terial (Table II), the PBIL optimization came up with a design
tion, a new FE mesh is generated according to machine dimen- using less PM material. It can be seen that Powell’s method re-
sion input , a nonlinear solver is called to find the magnetic quires a total of 106 field solutions while the PBIL algorithm
WANG et al.: OPTIMAL DESIGN OF A CORELESS STATOR AFPM GENERATOR 61

Fig. 7. Basic structure of the optimization procedure.

TABLE I TABLE III

OPTIMIZATION RESULTS FOR MAXIMUM EFFICIENCY PERFORMANCES OF DIFFERENT FE OPTIMIZED DESIGNS

TABLE II
OPTIMIZATION RESULTS FOR MINIMUM PM MATERIAL The calculated performances of the above optimized designs
are shown in Table III. It was found that the inner to outer diam-
eter ratio is about 0.68 for the maximum efficiency design
and around 0.7 for minimum PM mass and/or volume design.
By minimizing the PM material, the cost and the mass of the
machine are also reduced.

VII. MECHANICAL STRENGTH ANALYSIS

The deflection of the rotor disks due to the strong magnetic
pull may have undesirable effects on an AFPM machine such
as: 1) closing the running clearance between the rotor disk and
needs 5562 field solutions. Obviously, Powell’s method is a lot the stator; 2) breaking the permanent magnets due to bending;
more efficient than the PBIL algorithm as it used only a fraction 3) reducing air-flow discharging area, hence deteriorating the
of the CPU time that the PBIL required. The design optimization cooling capacity; and 4) a nonuniform air gap causing a drift in
was carried out on a 1670-MHz Intel PC running RedHat Linux the electrical performance from the optimum. Besides, the rotor
operating system. On average it takes less than 2 s to solve one disks account for roughly 50% of the total active mass of an
field solution. AFPM machine. Hence, the optimal design of the rotor disks is
62 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005

Fig. 8. Deflection (blown up) and Von Mises stress distribution of a rotor disk.

of great importance to realize a design of high power-to-mass TABLE IV

COMPARISON OF DIFFERENT DESIGNS OF ROTOR DISK
ratio. All these aspects make the mechanical stress analysis of
the rotor disk a necessity.

A. Mechanical Stress Analysis of Rotor Disk

The structure of the rotor disks of the designed AFPM ma-
chine was analyzed with the aid of an FEM structural program.
The aim was to find a least thickness for the rotor disk, which
satisfies the critical strength requirements of a rotor disk. The costs are taken into account for small production volumes, it
maximum tolerable deflection of the rotor disk was set to be is justifiable to use a steel disk with uniform thickness. The
0.3 mm. This is to ensure that the PMs would not suffer any constructed machine described in the next section uses 17 mm
excessive forces that have the potential to break the magnets or straight disks.
peel them off from the steel disk.
By taking into account the symmetry of the machine, only one VIII. PROTOTYPE MACHINE
sixteenth of the rotor disk was analyzed using 4-node shell-ele- To verify the optimization design and performance described
ments, with symmetrical boundary conditions applied. The axial in Section VI, an AFPM machine optimized for minimal
magnetic pull between PM disks at zero current state was calcu- PM material (design option C in Table III) has been built.
lated as 14.7 kN while the magnetic pull between the PM disk Fig. 9 shows the constructed air-cooled coreless stator AFPM
and stator under load due to tangential flux is rather insignifi- machine. The single stator is mounted on one side of the
cant (about 45 N). In the FE program, the magnetic pull-force external frame. There are 20 parallel connected coils per phase,
is applied in the form of a constant 69.8 kPa pressure load over as shown in Table V. To facilitate making connections, four
the total area that the PMs occupy. The stiffness provided by the circular bus-bars are used as shown in Fig. 9(b). Rare-earth
magnets was not included so as to keep the design on the con- sintered NdFeB magnets are used, which has a remanent flux
servative side. density of 1.18 T and a maximum allowable working tem-
Based on the analysis, the rotor disk thickness was chosen as perature around 130 C. The hub structure shown in Fig. 9(c)
17 mm with a maximum deflection of 0.145 mm. Fig. 8 shows serves as both air intake and supporting structure for the rotor
the deflection (blown up) and the Von Mises stress distribution disks. Furthermore, it also acts as a centrifugal fan improving
of the laboratory prototype 17 mm disk. The maximum stress of the air cooling of the AFPM machine.
35.6 MPa is much lower than the typical yield strength of mild
steel that is in the region of 300 MPa. Previous studies [21] show
IX. PERFORMANCE
that the bending of the rotor disk decreases toward its outer pe-
riphery. The rotor disk may be machined in such a way that the The performance tests on the prototype AFPM machine were
disk becomes thinner toward the outer periphery. As shown in carried out in the laboratory and were analyzed. The tests fo-
Table IV, the tapered disk uses approximately 10% less iron than cused on its generation mode. A reconfigurable water-cooled
the straight disk. The maximum deflection increases by only bank of resistors was configured into a balanced three-phase
0.021 mm with the tapered disk, which is negligible. This can load and then connected across the AFPM machine terminals.
effectively save the active mass of the machine without com- An induction machine was used as prime mover. The water
promising the mechanical strength. However, if manufacturing cooling system consists of a water tank, pipe system, and a
WANG et al.: OPTIMAL DESIGN OF A CORELESS STATOR AFPM GENERATOR 63

Fig. 9. The designed single-stage synchronous AFPM machine: (a) rotor disk Fig. 10. Testing setup of the designed AFPM machine. 1: 600 kW induction
with surface mounted PM segments, (b) coreless stator with busbars, and (c) the machine drive. 2: AFPM generator. 3: Water-cooled resistive load. 4: Measuring
assembled machine. equipment.

TABLE V
DESIGN DATA FOR THE AFPM MACHINE UNDER STUDY

Fig. 11. Comparison of predicted and measured no-load phase voltages

(2300 rpm).

cooling tower of 250 kW capacity. The testing setup is shown

in Fig. 10.
In Fig. 11, the no-load phase voltage of the prototype ma-
chine at rated speed calculated by a FE time-stepped model is
compared with the measured results. The details of the time-step
modeling of AFPM machine is given in [27]. The output power
and phase current were measured at different rotating speeds.
The same conditions were simulated using the FE computer
program. The load resistance value used in the computation
under various load conditions was compensated with an esti-
mated temperature factor. The results are presented and com-
pared in Fig. 12. Agreement between measured and predicted
output power and phase current is shown to be well within the
limits of experimental accuracy.
As shown in Table VI, the rated output power of the AFPM Fig. 12. Predicted and measured power and phase current for balanced
generator at unity power factor is measured to be 154 kW three-phase operation.
at rated speed. Taking into account the mechanical loss
(measured), eddy-current loss (measured), and copper loss 6989.7 W, of which 1732 W are eddy-current losses, 3509 W
(calculated), the total mechanical input power (ignoring the are mechanical losses, and 1748.7 W are copper losses. The
losses in PMs and rotor disks) is then 161 kW. This gives a stabilized machine winding temperature rise was measured
machine efficiency of 95.7% at that speed. The total losses are as 56 C, which is not high and shows the good air cooling
64 IEEE TRANSACTIONS ON MAGNETICS, VOL. 41, NO. 1, JANUARY 2005

TABLE VI [5] F. Caricchi, F. Crescimbini, E. Santini, and C. Santucci, “Influence of the

MEASURED PERFORMANCE AT RATED OPERATING CONDITION radial variation of the magnet pitches in slotless PM axial flux motors,”
in Proc. IEEE-IAS Annu. Meeting, vol. 1, 1997, pp. 18–23.
[6] F. Caricchi, F. Crescimbini, O. Honorati, G. L. Bianco, and E. Santini,
“Performance of core-less winding axial-flux PM generator with power
output at 400 Hz-3000 rev/min.,” IEEE Trans. Ind. Appl., vol. 34, no. 6,
pp. 1263–1269, Nov./Dec. 1998.
[7] F. Caricchi, F. Crescimbini, E. Santini, and C. Santucci, “FEM eval-
uation of performance of axial flux slotted PM machines,” in Proc.
IEEE-IAS Annu. Meeting, vol. 1, 1998, pp. 12–18.
[8] G. W. Carter, Electromagnetic Field in Its Engineering As-
pects. London, U.K.: Longmans, 1954.
[9] W. S. Leung and C. C. Chan, “A new design approach for axial field
capacity in an essentially self-cooled AFPM generator. For an electrical machine,” IEEE Trans. Power. App. Syst., vol. PAS-99, no. 4,
ambient temperature of 30 C–40 C, typical for ICE power pp. 1679–1685, 1980.
generation applications, the actual temperature of the stator [10] C. C. Chan, “Axial-field electrical machines: Design and application,”
winding will be in the range of 86 C–96 C. The relatively IEEE Trans. Energy Convers., vol. 2, no. 2, pp. 294–300, Jun. 1987.
[11] J. Engstrom, “Inductance of slotless machines,” in Proc. IEEE Nordic
high mechanical losses are mainly due to the windage losses. Workshop on Power and Industrial Electronics, Aalborg, Denmark, June
The power density of the machine is calculated as 4.43 MW/m , 2000.
which is relatively high when compared with that of conven- [12] A. E. Fitzgerald and C. Kingsley, Electric Machinery, 2nd ed. New
York: McGraw-Hill, 1961.
tional ac machines (typically 2.2–2.6 MW/m ). [13] J. F. Gieras, R. Wang, and M. J. Kamper, Axial Flux Permanent Magnet
Brushless Machines. Dordrecht, The Netherlands: Kluwer, 2004.
[14] J. F. Gieras and M. Wing, Permanent Magnet Motor Technology: Design
X. CONCLUSION and Applications, 2nd ed. New York: Marcel Dekker, 2002.
[15] C. Gu, W. Wu, and K. Shao, “Magnetic field analysis and optimal design
of DC permanent magnet coreless disk machine,” IEEE Trans. Magn.,
The overall design methodology presented in the paper is to vol. 30, no. 5, pp. 3668–3671, Sep. 1994.
use a combination of classical circuit analysis and FE field anal- [16] V. B. Honsinger, “Performance of polyphase permanent magnet
ysis in an optimization process. Both Powell’s method and the machines,” IEEE Trans. Power App. Syst., vol. PAS-99, no. 4, pp.
1510–1516, 1980.
PBIL algorithm have been applied in the optimization process [17] A. Hughes and T. J. Miller, “Analysis of fields and inductances in air-
of the AFPM machine. Powell’s method is more efficient than cored and iron-cored synchronous machines,” Proc. Inst. Elect. Eng.,
the PBIL algorithm as it needed only a fraction of the CPU time vol. 124, no. 2, pp. 121–126, Feb. 1977.
[18] M. J. Kamper, F. S. Van der Merwe, and S. Williamson, “Direct finite ele-
that the PBIL required. However, the PBIL optimization found ment design optimization of cageless reluctance synchronous machine,”
slightly better solutions in all the case studies. By minimizing IEEE Trans. Energy Convers., vol. 11, no. 3, pp. 547–555, Sep. 1996.
the PM material, an overall better design can be obtained with [19] N. F. Lombard and M. J. Kamper, “Analysis and performance of an iron-
less stator axial flux PM machine,” IEEE Trans. Energy Convers., vol.
lower eddy-current losses, high efficiency, high power-to-mass 14, no. 4, pp. 1051–1056, Dec. 1999.
ratio, and low cost. [20] H. C. Lovatt, V. S. Ramden, and B. C. Mecrow, “Design an in-wheel
One of the designed AFPM machines was built and tested. motor for a solar-powered electric vehicle,” Proc. Inst. Elect.
Eng.—Elect. Power Appl., vol. 145, no. 5, pp. 402–408, 1998.
Owing to a very low phase inductance in the coreless stator [21] D. Mbidi, K. vd Westhuizen, R. Wang, M. J. Kamper, and J. Blom,
AFPM machine, the output voltage varied almost linearly with “Mechanical design considerations of double stage axial-flux permanent
the load current. When operated with a balanced three-phase magnet machine,” in Proc. IEEE-IAS 35th Annu. Meeting, vol. 1, Rome,
Italy, 2000, pp. 198–201.
resistive loading, the waveform of the stator phase voltage and [22] M. J. D. Powell, “An efficient method for finding the minimum of a
current were found to be very close to sinusoidal. The measured function of several variables without calculating derivatives,” Comput.
performance of the prototype AFPM machine compares favor- J., vol. 7, pp. 155–162, 1964.
[23] J. Saari and A. Arkkio, “Losses in high-speed asynchronous motors,” in
ably with the predicted one. Proc. ICEM’94, vol. 3, Paris, France, 1994, pp. 704–708.
[24] R. Wang and M. J. Kamper, “Calculation of eddy current loss in axial
field permanent magnet machine with coreless stator,” IEEE Trans. En-
REFERENCES ergy Convers., vol. 19, no. 3, pp. 532–538, Sep. 2004.
[25] R. Wang, M. J. Kamper, and R. T. Dobson, “Development of a ther-
[1] K. Atallah, Z. Q. Zhu, D. Howe, and T. S. Birch, “Armature reaction mofluid model for axial field permanent magnet machines,” IEEE Trans.
field and winding inductances of slotless permanent-magnet brushless Energy Convers., to be published.
machines,” IEEE Trans. Magn., vol. 34, no. 5, pp. 3737–3744, Sep. 1998. [26] R. Wang, “Design aspects and optimization of an AFPM machine with
[2] S. Baluja, “Population-based incremental learning: A method for an ironless stator,” Ph.D. dissertation, Dept. Elect. Eng., Univ. Stellen-
integrating genetic search based function optimization and compet- bosch, Matieland, South Africa, 2003.
itive learning,” Carnegie Mellon Univ., Pittsburgh, PA, Tech. Rep. [27] R. Wang, H. Mohellebi, T. J. Flack, M. J. Kamper, J. Buys, and M. Feli-
CMU-CS-94-163, June 1994. achi, “Two-dimensional Cartesian air-gap element (CAGE) for dynamic
[3] P. Campbell, “Principle of a PM axial field DC machine,” Proc. Inst. finite-element modeling of electrical machines with a flat air-gap,” IEEE
Elect. Eng., vol. 121, no. 1, pp. 1489–1494, 1974. Trans. Magn., vol. 38, no. 2, pp. 1357–1360, Mar. 2002.
[4] F. Caricchi, F. Crescimbini, A. D. Napoli, and E. Santini, “Optimum [28] S. Williamson and J. Smith, “The application of minimization algo-
CAD-CAE design of axial flux permanent magnet motors,” in Proc. rithms in electrical engineering,” Proc. Inst. Elect. Eng. A, vol. 127, no.
ICEM’92, vol. 2, Paris, France, 1992, pp. 637–641. 8, pp. 528–530, 1980.

View publication stats