The electric machine, the target of this design, was assumed to be integrated inside an electromagnetic shock absorber for full active suspension control. Given the nature of suspension motion, a mechanism is necessary to convert linear displacements into rotation. Many options are possible in this context [
3]. However, for the purpose of this study, an ideal transmission ratio can be defined as
where
is the mechanical angular speed of the electric machine, and
is the actuator speed. For this application,
; this value was obtained from a previously designed actuator [
5]. Similarly, the actuator force
is directly proportional to the motor torque
:
These simple relationships allow mapping suspension force and linear velocity requirements into torque and angular speed.
2.1. System Requirements from Vehicle Dynamics
The performance of a regenerative damper was analyzed through the vertical dynamics of a passenger vehicle using a quarter-car model (
Figure 1). In this case, the traditional damper was replaced with an ideal active actuator that represents the electromagnetic shock absorber. The expressions that govern this system are
where labels s,u denote the sprung (chassis) and unsprung (wheel hub) bodies, respectively. Generalized coordinates are denoted with
, whereas body masses are given by
. Parameters
and
are the stiffness of the suspension spring and the tire, respectively. The input
is the road roughness vertical displacement profile, while
is the active force provided by the damper.
The quarter-car model was simulated in MATLAB/Simulink™ using the parameters listed in
Table 1. The vehicle features belonged to the C-class vehicle template in the automotive software CarSIM™. To ensure proper sizing, the road class and vehicle longitudinal speed were selected to generate worst-case road unevenness. This choice leads to severe activity on the electric machine. A classical suspension control based on mixed skyhook–groundhook strategies [
8] was applied. Thus, the active force reference is given by
where
and
are the tunable skyhook and groundhook damping coefficients, respectively.
The sprung mass acceleration was used to assess passenger comfort; it was filtered by a third-order bandpass filter that amplifies human-sensitive frequencies [
9], according to the ISO 2631 standard [
10]. Likewise, tire road holding was determined by the so-called road-holding index, i.e., the ratio between the tire elastic force and the corner weight [
11]:
where
is the gravity acceleration.
The simultaneous optimization of comfort and road holding is not possible because there is an intrinsic trade-off between them in a passenger car [
8]. Therefore, the optimal gains for
and
were found by minimizing the root mean square (rms) weighted (filtered) chassis acceleration
or the road-holding index
. In both cases, the suspension stroke
was constrained to its mechanical limits, whereas the road-holding index was restricted below unity to avoid wheel detachment from the ground. The control gains and their respective metrics are outlined in
Table 2.
Then, the actuator design parameters were based on the following considerations. Firstly, continuous actuation force was obtained from
in the worst-case condition (road holding); thus,
. This force can be related to the continuous operation of the electric machine. Considering an air-cooled device through natural convection, the actuator should exert this force while supplied by a current density
[
12]. Secondly, the motor speed was selected by analyzing the effect of field weakening on vehicle dynamics, unlike previous studies in the literature omitting flux-weakening operation [
5].
To this end, the ideal force actuator on the quarter-car model was dynamically saturated according to
where
is the actuator speed. The actuator base speed
represents the limit between the constant force and the constant power regions. A coefficient
was included to account for the overload capability of the machine in transient conditions (less than ten seconds). Hence, a suitable base speed must be selected to set the field-weakening region boundary to the machine without hindering the dynamic response of the vehicle. Note that the machine sizing is based on current density limitations following the empirical thermal performance available in the literature [
12]. A dedicated thermal model is needed to properly assess this point. However, in the given application, the set up of such model is very complex due to the uncertainty of heat sources, such as the vehicle engine, the brake calipers, and the airflow running through the wheel arch, among others. Although these details are of relevance and interest, they are beyond the scope of the present study.
The results in
Figure 2 illustrate the effect of changing
on both the suspension stroke and theroad-holding index limits. These values should not exceed the established restrictions:
. Base speed values above
led to suspension stroke and road-holding index values that fulfilled the aforementioned constraints. Thereby, this value was selected as the base speed for the electric machine design.
Finally, applying (
2) and (
1), the force and linear speed values were converted to torque and angular speed targets:
,
.
Figure 3 and
Figure 4 show the electric machine activity within the suspension when controlled with the optimal comfort and road holding strategies. Remember that the obtained suspension activity comes from a worst-case simulation where the vehicle runs at
on an ISO-C road profile. In this condition, the actuator rarely works in the field-weakening region. The comfort strategy operates in field weakening
of the time. For the road-holding strategy, field weakening takes place only
of the time. It is worth noting that although field weakening is rarely used in normal vehicle operation, rare situations, such as traveling over a bump or pothole, push the suspension to speeds in excess of
. Then, the machine should be able to enter the field-weakening region in these extreme cases.
2.2. Machine Geometry and Winding Scheme
In a suspension, compactness is key to guaranteeing proper kinematics without interference. In fact, suspension elements share space with other assemblies, such as the brakes and the wheel. For the front axle, the steering must also be taken into account. These constraints are considered in the electric machine design, where the main goal is to maximize the output torque to limit the actuator volume and mass.
To yield a compact electric machine, a fractional-slot motor topology with a double-layer winding was selected. This choice brings multiple advantages. First, these layouts minimize end-turn length, thus leading to a compact machine in terms of the axial envelope and low end-turn Joule losses. Second, fractional-slot layouts exhibit low cogging torque ripple without needing to apply additional mitigation techniques [
13].
Different pole–slot combinations are possible for fractional-slot PMSM topologies. These choices lead to different values of the winding factor, which is a measurement of the effectiveness of the magnetic flux linked in the coils [
14]. Accordingly, the torque and back electromotive force (EMF) are proportional to the winding factor. In this regard, a twelve-slot, ten-pole PMSM configuration was selected as the base architecture for all the machines in this study. This combination leads to a relatively high winding factor with respect to other alternatives, maintaining a low harmonic content on the back EMF, which has a positive impact on the machine efficiency.
The stator geometry was the same for all the machine designs, although numerical dimensions varied among solutions. As shown in
Figure 5, the stator geometry is defined by outside radius
, back iron radius
, back iron width
, tooth base width
, slot base width
, slot inner width
, tooth width
, shoe opening
, shoe depth
, tooth-to-shoe depth
, tooth base depth
, and air gap length
g. Its winding obeys the fractional-slot topology with two layers per slot, as also illustrated in
Figure 5.
On the contrary, rotor topologies did change for the tested machines. IPM geometries present flat and V-shaped magnets. They are depicted in
Figure 6. Common parameters are shared among rotor geometries: magnet length
, outer rotor radius
, and inner rotor radius
. The SPM topology is further specified by rotor back iron
and magnet fraction
, where
is the magnet pitch, and
is the pole pitch.
IPM designs share three parameters: magnet opening angle
, web thickness
w, and bridge width
b. The IPM V-shape has an additional parameter, which is the magnet depth
in the radial direction. Dimensions
w and
b define an iron structure to hold the magnets into the rotor. The presence of this iron material is relevant, especially when the rotor is subject to high centrifugal stress. However, oversizing this structure might lead to a short circuit of the PM paths, which would decrease the air gap flux and, consequently, reduce the output torque of the machine. Therefore, the proposed design procedure focused exclusively on magnetic performance. However, a preliminary centrifugal stress analysis was carried out at
on initial IPM cross-sections using Solidworks™. Conservative values of
and
were used. These numbers are compliant with the tolerances of the steel lamination stamping process. Specifically, the minimum distance between cut features must be between one and one and a half times the material thickness (
for the employed M230-35A steel) [
15].
2.3. Optimization Procedure
The motor design process starts by fixing the parameters listed in
Table 3. The listed values belong to common design choices and constraints. The magnetization curve for the M250-35A steel is depicted in
Figure 7. Then, each machine cross-section (stator and rotor) is optimized to maximize the torque-to-length ratio. In a force actuator, this design choice aims to minimize the mass of the PMSM, which is a desirable feature in automotive mechatronics. Since the actuator mass is added to the wheel hub (unsprung mass), it may hinder the dynamic vertical vehicle performance. This approach differs from those addressing traction applications where the machine performance is evaluated in a wide speed range [
16].
Torque was evaluated at continuous current density in the maximum torque-per-ampere (MTPA) condition. The average torque value was calculated from the simulation of five rotor positions equally spaced over a stator slot pitch
. The first position was randomly selected within one-fifth of
to minimize fundamental and third-order ripple harmonics [
17]. A Dirichlet-type condition was applied to the outer boundary of the air surrounding each machine geometry. No periodicity in the analysis domain was used in this case. A sliding band air gap model in FEMM was used to allow movement without changing the rotor or stator meshes [
18]. To speed up computations, a coarse mesh of approximately 6000 triangular elements was used for each simulation. During optimization, the average flux density in the stator teeth and back iron was limited to
. At this stage, the winding was simplified as a single turn with the slot fill factor
specified in
Table 4. The number of turns is defined at a later stage when setting the base speed of the machine. The total length of the machine was normalized to one meter.
The design variables of each topology are summarized in
Table 4. They were varied at each iteration due to their significance and impact on the optimization objective, i.e., the output torque.
Table 4 also lists the optimized values for these variables. The known effects of the relevant variables are listed next.
The effect of the radius ratio
is bi-fold. The rotor arm length and PM surface are directly proportional to the rotor outside radius. In contrast, an increase in this radius decreases the slot area and reduces the magneto-motive force of the machine [
19]. The back iron and tooth base width modify the slot area, but their selection is crucial in defining an unsaturated working point of the stator iron paths. Since the rotor is part of the magnetic circuit path, saturation is crucial therein as well. For this reason, the rotor and stator back iron widths are set as equal in the SPM design. Similarly, for IPM designs, a space between the inner rotor radius
and the closest PM point is set to
. The magnet fraction
changes the magnet flux toward the stator and the magnet-to-magnet leakage flux [
19]. The role of the angle
is similar in IPM geometries.
The magnet length in the SPM rotor fixes the PM in a safe operating point to avoid demagnetization. To this end,
was selected to set the so-called permeance coefficient at a value of five. Since analytically solving the magnet operating point is difficult for IPM geometries, the magnet length
was included as a design variable. Also, the web thickness
w was included in both IPM designs because it affects the magnet leakage flux [
20]. Lastly, the magnet depth
directly affects the air gap flux density [
20].
The particle swarm optimization (PSO) algorithm was selected as it is a readily available choice from the Global Optimization Toolbox in MATLAB™. Previous efforts have shown positive results when using this optimization algorithm for electric machinery [
21,
22]. A custom cost function was designed to perform the following steps:
Define a motor geometry in FEMM by using the parameters listed in
Table 4.
Simulate the motor performance in FEMM.
Extract the torque output, the stator back iron, and the tooth base magnetic flux density values.
Discard solutions where magnetic flux density values are aboved .
This level of integration is only possible through the OctaveFEMM function set. The PSO algorithm was set up to have a swarm size of 40, self- and social-adjustment weights of , adaptive inertia in the range , and the termination criterion was set to a relative error of .
2.4. Motor Mapping
The PMSM model in the direct (d) and quadrature (q) axes and steady-state conditions can be written in terms of flux linkages as follows:
where
denotes the axis flux linkages,
R is the phase resistance,
is the rotor mechanical speed,
p is the number of pole pairs,
indicates the axis voltages,
represents the axis currents, and
T is the electromagnetic torque. In the linear case, flux linkages can be written as
with inductance terms
and PM flux linkage
. However, flux linkage representation allows mapping the performance of the machine in the presence of saturation. This nonlinear behavior inevitably occurs in overload conditions.
To account for leakage effects, the end turn inductance contribution was added to the flux linkage terms in d, q. The end-turn inductance
is calculated through an analytical model [
23]:
where
is the mean coil pitch, and
is the slot area. For the concentrated coil winding,
is computed as
Before mapping the machine operation, the number of turns per coil
N was considered in the model. In particular, the following scaling relations are valid:
The superscripts denote that the parameters were computed with a single turn per coil, as presented in
Section 2.3. The resulting resistance, flux linkage, and current are included in Equations (
8)–(
10).
The machine was controlled using field-oriented control (FOC) with field weakening, based on the maximum torque-per-ampere (MTPA) and maximum torque-per-volt (MTPV) approaches [
24]. The process to obtain the FOC maps of the machine is described next.
The dq flux linkage maps were calculated by applying the rms current density vectors to the machine in the range
. These current density vectors led to specific values of
. Then, the voltage and torque contributions were computed using the dq model (Equations (
8)–(
10)). Finally, the maximum torque was found at different mechanical speeds of the machine under the following current and voltage constraints:
where
is the maximum current vector norm, and
is the DC link voltage.
The characterization of the motor losses is relevant to estimating its overall efficiency. Hence, copper, iron, and PM loss components were computed using FEMM, automated with MATLAB™. The resistive losses in the windings were computed as
where
is the stator length,
is the total turn-length of a single coil at both ends,
is the rms current density, and
is the number of coils per phase.
The core and magnet losses were computed with FEMM to measure flux density and vector potential in the dq current grid from the FOC maps. The results were then postprocessed with the Fourier transformation to obtain the net component from signal harmonics [
25]. Iron losses were computed according to Bertotti’s loss separation model [
26]:
where
,
and
for the M250-35A steel. The first term inside the summation corresponds to the hysteresis loss (
); the second one represents losses due to eddy currents (
); the third term describes the excess or anomalous losses (
).
is the amplitude of the
mth harmonic of the flux density,
is the evaluated frequency, and
is the total number of harmonics considered. The number of harmonics was determined from a preliminary sensitivity analysis to account for the most significant loss components. For each motor topology, the frequency was evaluated at
, i.e., the nominal speed design requirement (see
Section 2.1). The results can be extrapolated to any mechanical speed through
The magnet losses were calculated from the expression
where
is the
mth harmonic of the constraint current density calculated with the model from [
27],
is the
mth harmonic of the PM magnetic vector potential out-of-plane component,
, and
. PM losses can be extrapolated to any speed through
Finally, total power losses at any speed are calculated as
The torque contribution in (
10) was corrected by accounting for core and magnet losses:
and the net shaft mechanical power is given by
The efficiency of the machine can be calculated according to its operating mode: