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

net/publication/352643116

Scaling laws for the FE solutions of induction machines

Article in Archives of Electrical Engineering · September 2019

DOI: 10.24425/aee.2019.129350

CITATIONS READS

13 81

3 authors, including:

Martin Nell Jonas Lenz

RWTH Aachen University Ulm University
30 PUBLICATIONS 126 CITATIONS 2 PUBLICATIONS 22 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Martin Nell on 01 September 2021.

The user has requested enhancement of the downloaded file.

ARCHIVES OF ELECTRICAL ENGINEERING VOL. 68(3), pp. 677–695 (2019)

DOI 10.24425/aee.2019.129350

Scaling laws for the FE solutions of induction machines

MARTIN NELL1 , JONAS LENZ2 , KAY HAMEYER1

1 RWTH Aachen University, Institute of Electrical Machines (IEM)

Schinkelstraße 4, 52062 Aachen, Germany
e-mails: [email protected]
2 Ulm University, Institute of Energy Conversion and Storage
Albert-Einstein-Allee 47, 89081 Ulm, Germany

(Received: 22.05.2019, revised: 23.05.2019)

Abstract: In this paper a scaling approach for the solution of 2D FE models of electric
machines is proposed. This allows a geometrical and stator and rotor resistance scaling as
well as a rewinding of a squirrel cage induction machine enabling an efficient numerical
optimization. The 2D FEM solutions of a reference machine are calculated by a model
based hybrid numeric induction machine simulation approach. In contrast to already known
scaling procedures for synchronous machines the FEM solutions of the induction machine
are scaled in the stator-current-rotor-frequency-plane and then transformed to the torque-
speed-map. This gives the possibility to use a new time scaling factor that is necessary to
keep a constant field distribution. The scaling procedure is validated by the finite element
method and used in a numerical optimization process for the sizing of an electric vehicle
traction drive considering the gear ratio. The results show that the scaling procedure is
very accurate, computational very efficient and suitable for the use in machine design
optimization.
Key words: evolutionary strategy, finite element method analysis, induction machine, in-
duction motor, loss calculation, multi-objective optimization, scaling laws

1. Introduction
Energy optimization performed for example by an improvement in the efficiency of electrical
equipment is the global trend today [1]. In developed countries the industrial induction machines
(IMs) are the major consumers of electric energy and globally account for about 40% of overall
power consumption [2].
To lead manufacturers to design and build more efficient IMs the European Union specified
the new premium efficiency standard (IE3) for IMs operated at 50 Hz or 60 Hz by the IEC
© 2019. The Author(s). This is an open-access article distributed under the terms of the Creative Commons Attribution-
NonCommercial-NoDerivatives License (CC BY-NC-ND 4.0, https://creativecommons.org/licenses/by-nc-nd/4.0/), which per-
mits use, distribution, and reproduction in any medium, provided that the Article is properly cited, the use is non-commercial,
0 and no modifications or adaptations are made.
678 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

60034-30 standard [3]. Since January 1, 2015 this standard is valid for IMs with a rated power
of 7.5 to 375 kW and since January 1, 2017 for IMs with a rated power of 0.75 to 375 kW. The
standard also reserves an IE4 class for the future.
Due to the low-cost, ruggedness and fault tolerance, frequency-inverter-driven IMs are used
as the main workhorse in the rising market of electrical and hybrid drive trains [4]. Here too,
however, the main aim is to reduce the energy consumption or, in other words, to improve the
efficiency of the drive train and particularly of the IM [15]. Moreover, in electric or hybrid vehicles
minimum cost and weight are further goals [15]. Proficient manufacturers have the experience
to design high premium efficient electrical machines. But only mathematical optimization can
handle the complexity of the relations between the machine’s geometry and its performance and
nonlinearity [5] in a short time. With the mathematical optimization tool the limits can be pushed
toward more efficient designs. For the design of high efficient IMs, particularly for frequency-
inverter-driven ones, a detailed loss analysis is required. This requires local and temporal highly
resolved nonlinear field computation and can be performed in the post processing of nonlinear
transient finite element simulations of the magnetic circuit [16].
For the IM, the finite element method (FEM) takes a large number of simulation time steps
to build up the machine’s rotor flux matrix [16]. Hence, using the time-consuming FEM in
a mathematical optimization procedure would end up in an extremely time-consuming calculation
and therefore is not suitable. Von Pfingsten, Nell and Hameyer [6, 16] proposed a hybrid simulation
approach for the IMs 2D finite element (FE) calculation that drastically decreases the simulation
time by shortening the transient build-up of the rotor flux. Nevertheless, this hybrid approach,
that needs about 2 000 core hours for an efficiency map, is still not sufficient for the application
in a mathematical optimization procedure.
Due to the fact that the FEM and other numerical methods are very time-consuming scaling
laws are a popular method in physics and engineering. They are often used in numerous examples.
Wood [7] described the general scaling laws for electromagnetic systems. His work was motivated
by the constraint of the system’s thermal stability. By using the electromagnetic and thermal
diffusion equation, as well as the momentum and kinematic equation Hsieh and Kim [8] presented
a detailed derivation of scaling laws for electromechanical systems.

2. Scaling laws of electrical machines

Žarko, Stipetič and Ramakrishnan published several papers about the scaling laws for syn-
chronous machines (SM). In [17] and [9] the procedures of radial and axial geometrical scaling
and of rewinding for the SM are introduced. In [18] the efficiency maps of the SM are calculated
by using a scalable saturated flux linkage and loss model of the SM. In [15] and [19] the scalable
SM models are used to find the optimal sizing of a SM traction motor. Moreover, Žarko presented
a method to design a premium efficiency IM, using scaling laws for its equivalent circuit param-
eters in [2]. He assumed that the temperature rise in the slot of the original and the scaled motor
are similar.
Another method to improve the IM efficiency classes, using the method to scale the core axial
lengthening, was described by Alberti, Bianchi, Boglietti and Cavagnino in [10]. The influence of
the rotor diameter and the length on the rating of IMs was presented by Bone in [11]. His scaling
Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 679

laws for IMs are not as exact as the ones derived in this paper because the field solutions change.
Nonetheless, they are well applicable for machine designers. In this paper a more sophisticated
approach for a mathematical optimization of squirrel cage IMs is discussed. It is based, on the one
hand, on the hybrid simulation approaches for induction machine calculation from von Pfingsten,
Nell and Hameyer [6, 16], and on the other hand, on the scaling laws for the IMs proposed in this
paper and in [20]. The procedure of this method is illustrated in Fig. 1. The stator-current-slip-
frequency-operation-planes (I1 – f 2 -planes) of a reference machine design are calculated with the
hybrid simulation approach of von Pfingsten, Nell and Hameyer [6, 16].

Fig. 1. Machine scaling scheme

To obtain a new scaled motor design the solutions are scaled in the I1 – f 2 -plane without
changing the field solution of the IM. Taken into account requirements, such as dc-link voltage,
maximum inverter current, and the operation strategy, such as Maximum Torque Per Ampere
(MTPA) or Maximum Torque Per Electrical Losses (MTPELs), the I1 – f 2 -planes are transformed
to torque-speed-maps (T-n-maps).
680 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

3. Modeling of an induction machine

3.1. IM operating points in terms of the I1 - f 2 -plane
The fundamental wave T-equivalent-circuit of an IM is presented in Fig. 2. It demonstrates
the allocation of the stator current I1 into the magnetizing current Iµ and the rotor current related
to the stator side I2S . Moreover, it illustrates that all reactances and the rotor resistance R2S are
proportional to the synchronous angular frequency ω1 . According to von Pfingsten, Nell and
Hameyer in [6, 21] and [16] the allocation of the stator current I1 into the magnetizing current
Iµ and the rotor current I2S is independent of the stator frequency f 1 by subtracting the voltage
drop on the stator resistance R1 . As a result, this allocation only depends on the rotor frequency
f 2 and the saturation of the main inductance L M . The saturation has to be considered in highly
utilized traction drives and occurs at high values of Iµ that is reached at low values for f 2 and
high values for I1 [21]. Therefore, the current allocation of I1 into Iµ and I2S only depends on the
amplitude of the stator current I1 and the rotor current frequency f 2 , as long as the rotor resistance
and the inductances are assumed to be constant. All torque speed operating points of an IM with
a constant rotor resistance R2S can be mapped in the I1 - f 2 -plane.

Fig. 2. Equivalent circuit diagram of a squirrel cage induction machine

3.2. Induction machine calculation applying the hybrid simulation approach

To accelerate the FE calculation of the IM, the hybrid simulation approach presented in
[6] and [16] is used. With the hybrid simulation approach, the entire I1 - f 2 -plane and T-n-map,
respectively, can be calculated 50% faster than with the transient FEM. In combination with the
hereafter introduced scaling scheme it provides a fast procedure to calculate and scale IMs that
leads to the possibility of the use for numerical optimization. The main aspects of it will be
described in the following.
In the first step, a non-linear no-load static finite element analysis (FEA) with one simulation
time step is conducted k = 1, . . . , K times, where k marks a certain saturation state [6, 16]. The
stator current I1 is the only excitation. For each non-linear no-load FEA the inductance matrix
L(k) of the IM is extracted in accordance with [12]. From these extracted matrices L(k) the
rotor current I2S is calculated with the analytical fundamental wave equations derived from the
equivalent circuit diagram in Fig. 2 for every saturation state k. With the stator current vector ⃗I 1
S →−
and the saturation dependent rotor current vector ⃗I , the stator flux linkage vector Ψ (k) for every
2 1
Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 681

saturation state k is calculated. By comparing the amplitude of this stator flux linkage and the
→−
no-load stator flux linkage of the no-load FEA Ψ 1,nl (k) the valid saturation state is found. With
numerical interpolation, the value of the rotor current for each operating point in the I1 - f 2 -plane
is found and used as the excitation for the second, now transient, FE simulation [6, 16]. The
calculation steps are performed in the I1 - f 2 -plane with the fixed stator frequency f 1 and fixed
rotor resistance R2S , as well as the rotor conductivity σ2 , respectively.
By considering an operation strategy such as MTPEL and by scaling the loss power of the IM
according to different synchronous speeds f 1 , as described in [4], the I1 - f 2 -plane is transformed
to the T-n-map. A variation of the rotor resistance R2S by temperature can be considered with the
scaling laws of the rotor resistance due to the temperature variations described in section 4.6.

4. Scaling laws for induction machines

The previous scaling laws for IMs introduced by Bone in [11] are not exact due to the fact
that the field solution is changed. The IM scaling laws of Žarko in [2] deal with scaling the IM’s
equivalent circuit parameters assuming equal temperature rise in the slots of the original and
reference motor. For the SM, Stipetič, Žarko and Popescu derived scaling laws that consider the
same field solution. This is the basis for the IM scaling laws in this paper. In the following the
scaled parameters are marked with (′).

4.1. Geometrical scaling

The geometrical scaling in cylindrical systems is subdivided into radial and axial scaling with
the radial scaling factor kr and the axial scaling factor k a . The effect of the geometrical scaling
is pictured in Fig. 3 and described with

ρ′ = ρkr , (1)
l ′ = l ka , (2)
′
Across = Across kr2 , (3)
′
Asurface = Asurface kr k a , (4)
V ′ = V kr2 k a , (5)

Fig. 3. Scaled machine parameters

682 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

where l is the axial length, ρ is the radial distance to the center, Across is the cross-section areas,
Asurface is the radial surface area and V is the volume of the active part of the machine. The
cross-section of the short-circuit ring also increases with kr2 and is independent of k a to preserve
a constant relation to the bar cross-section. The conductor cross-section also changes with kr2 .
For the end windings it is assumed that the conductor length increases quadratically with kr since
the conductor cross-sections increase, thus the axial extent becomes larger and the arc length
increases linearly with kr .

4.2. Centrifugal force

The rotation or the angular velocity ω respectively of the rotor of an IM causes a centripetal
force, orthogonal to the motion and towards the center of the rotor. The centripetal force

Fρ = m ρ ω2 (6)

is proportional to the mass of the rotor m and the radial distance to the center ρ. The fictitious
centrifugal force that is directed away of the rotation axis, is a reaction to the centripetal force.
The centripetal force causes a tension
m ρ ω2
σ= , (7)
A
where A is the area at which the force acts. To avoid plastic deformations in the materials of the
rotor at high rotational speeds the tension in the materials must not exceed the maximum tension
σmax . Therefore, the angular velocity has to be limited to
√
σmax A
ωmax = . (8)
mρ

With the geometric scaling relations in (4) and (1) and the mass, scaled with k a kr2 , the scaling
law for the maximum rotor speed
′ 1
nmax = nmax (9)
kr
is derived.

4.3. Scaling of the electrical and magnetic parameters

Due to the fact that the magnetic permeability µ is in a non-linear relation to the magnetic
→
−
field strength H, one assumption of scaling the FE solutions is that the magnetic field strength
distribution inside the IM does not change. Therefore,
→
− ′ ( ′ ′) →
−
H ρ , φ = H( ρ, φ) (10)

is applicable.
In accordance with Ampère’s law
[ ]
→
−′ →− 1 ∂ ∂Hρ′
J = ∇′ × H ′ = ( ρ ′ ′
Hφ ) − e⃗z , (11)
ρ ∂ ρ′ ∂φ ′
Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 683

with the polar coordinates ρ and φ, as well as the unit vector e⃗z the scaling dependence of the
electric current density
→−′ 1 →
− 1→ −
J = ∇×H = J (12)
kr kr
follows.
In [8] it is described that a scaling factor k t1 has to be used to satisfy that the magnetic flux
density B is kept constant B( ρ, φ) = B ′ ( ρ′, φ ′ ) and therefore the equality of the magnetic field
strength (10) is satisfied. With the assumption that the geometrical dimensions are clearly shorter
than the magnetic wave length Ampère’s circuital law with Maxwell’s addition is simplified to
Ampère’s law:
→− →
−
∇ × H = σ E. (13)
With Faraday’s law of induction
→
−
→− ∂B
∇× E = (14)
∂t
and mathematical transformations
→
− →
−
∂B′ 2 ′ 1 * ′ B+
k t1 ′ + kr ∇ × ∇ × =0 (15)
∂t σ, µ-

→−
is derived, where σ is the electric conductivity and E is the electric field strength. The assumption
that the magnetic field strength, and therefore, the magnetic flux density do not change during
scaling, leads to
k t1 = kr2 . (16)
Hence, the time scaling factor k t1 corresponds to the square of the radial scaling factor kr . As
a result, all time depending parameters have to be scaled by the total time scaling factor k t , which
is the product of the first time scaling factor k t1 and the second one k t2 . The second time scaling
factor is a result of the rotor resistance scaling and will be discussed in detail in section 4.5 and
4.6. The time scaling leads to the proportionality of the reactances to the reciprocal time scaling
factor shown in Fig. 4.

Fig. 4. Scaling factor dependencies of the elements of the equivalent circuit diagram
of a squirrel cage induction machine
684 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

The magnetic flux Ψ, which is proportional to the inductance, scaled with k a and the current,
scaled with kr , changes with k a kr :
Ψ ′ = kr k a Ψ. (17)
From (3) and (12) it follows that the current changes with kr :

I ′ = kr I. (18)

The machine’s torque Telec is dependent on the flux and current. Hence, the torque is scaled
in accordance to:
′ 3 ( )
Telec = p Iq,1 kr Ψd,1 kr k a − Id,1 kr Ψq,1 kr k a = Telec kr2 k a . (19)
2

4.4. Stator resistance scaling

The stator resistance of an IM has to be scaled separately according to the winding head and
the axial length of the stator iron l Fe , that is independent of the radial scaling kr and proportional
to k a . It is assumed that the conductor length in the winding head l WH increases linearly with kr ,
since firstly, the conductor cross-sections increase and thus the axial expansions of the winding
head become larger and secondly, the arc length increases linearly with kr . The conductor cross-
sections Awire themselves increase with kr2 . This leads to the scaled stator resistance:
l Fe k a + l WH kr
R1′ = 2N1 , (20)
σ1 Awire kr
where σ1 is the conductivity of the winding material and N1 is the number of stator slots.

4.5. Rotor resistance scaling due to geometric variations

The equivalent circuit of the IM in Fig. 2 shows that the rotor resistance related to the stator
side R2S and therefore the rotor resistance R2 has a major impact on the machine’s behavior.
In contrast to the scaling of the SM done by Žarko, Stipetič and Ramakrishnan in IMs the
scaling of the rotor resistance is an important step. In addition to the varied rotor resistance
due to the geometrical scaling the resistance can vary because of temperature, material and
electric conductivity variations, respectively. The consequence of the geometrical scaling has to
be considered separately in terms of the bar resistance Rbar and the resistance of the short-circuit
∗ . The resistance of the rotor bar can be described with
ring Rring

l Fe ka
Rbar = ∝ 2, (21)
σ2 Abar kr
where l Fe is the active length of the IM, Abar is the area of the rotor bar and σ2 is the conductivity
ka
of the rotor conductors, and it is proportional to 2 . The resistance of a short circuit ring segment
kr
∆Rring can be described by:

2πr ring kr 1
∆Rring = ∝ 2 = , (22)
σ2 Aring Q2 kr kr
Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 685

where r ring describes the middle radius of the short-circuit ring, Aring is the area of the short-circuit
ring and Q2 is the rotor bar number. According to [13] this resistance can be transformed to an
∗
equivalent series resistance ∆Rring described by

∗ 1 1
∆Rring = ∆Rring · ( ( ))2 ∝ k , (23)
πp r
2 sin
Q2
with the number of pole pairs p, which is used in the 2D FEM. As outlined in section 4.1 the
cross-section area of the short-circuit ring increases with kr2 and is independent of the axial
∗
scaling. Therefore, ∆Rring is proportional to the inverse of the radial scaling factor kr as in (23).
The addition of the bar resistance and the equivalent series resistance of the short-circuit ring
lead to the total resistance of the rotor
∗
R2 = Rbar + 2∆Rring . (24)

By defining a compensating conductivity

l Fe ka
σ2,comp = ∝ 2 , (25)
Abar R2 kr · k (R2 )
it leads to a rotor resistance scaling factor
1
∗
Rbar + 2∆Rring
σ2,comp
k R1 = ′ = , (26)
σ2,comp ka
( )
ka ∗ 1
kr2 Rbar + 2∆Rring
kr2 kr
which can be simplified to
( )
σ2,comp kr
k R1 = ′ = 1 + − 1 κ2 , (27)
σ2,comp ka

with
1
κ2 = ( )
l Fe Aring πp
Q2 sin2 +1
πring Abar Q2
as the rotor geometry constant that is defined for the unscaled machine. The scaling factor k R1
describes the variation of the rotor resistance due to a variation in the relation of the axial
kr
length and radial length, expressed by of the machine. If the relation is kept constant k R1 is
ka
equal to one.

4.6. Rotor resistance scaling due to rotor conductivity variations

The scaling of the rotor resistance in accordance to the scaling of the compensating rotor
conductivity leads to further possibilities of rotor resistance scaling. The rotor conductivity and
686 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

resistance, respectively, can vary due to a variation of the material or by temperature. Hence,
a second rotor resistance scaling factor

σ2 1 + αnew (ϑsim,new − ϑref,new )

k R2 = (28)
σ2,new 1 + α(ϑsim − ϑref )

is introduced. It is dependent on the old and new conductivity, σ2 and σ2,new , the old and new
temperature coefficients α and αnew , the old and new reference temperatures ϑnew and ϑref,new
and the old and new simulation temperatures ϑsim and ϑsim,new . The resulting scaling factor for
the rotor resistance can be calculated by the total scaling factor of the rotor resistance k R and
leads to the scaled rotor resistance:
ka
R2′ = R2 k R , with k R = k R1 · k R2 · . (29)
kr2
In (29) the first part of the total rotor scaling factor k R1 considers a variation in the relation
of the axial to the radial length of the machine. The second part k R2 considers a variation in
ka
the temperature or the conductivity of the rotor bars and rings. The last part 2 considers the
kr
geometric variation of the rotor bars due to an axial and radial scaling of the IM. Fig. 4 shows that
the rotor resistance is dependent on the temperature, material and geometry of the IM. To satisfy
(10) the allocation of I2S and Iµ must not vary. As long as the rotor resistance is kept constant
and the calculation of the machine is done in the I1 - f 2 -plane, as described in section 3.1, (10) is
valid. With a variation in R2 the allocation changes. To ensure the same current allocation, even
in the event of a variation in the rotor resistance, the condition that the ratio of the axial to the
time scaling factor must correspond to the total rotor resistance scaling factor
ka !
=k R (30)
kt
has to be fulfilled. As a result the time scaling is supplemented by a second time scaling factor
ka 1
k t2 = = . (31)
k R k t1 k R1 k R2
Thus, the total time scaling factor results in

k t = k t1 k t2 . (32)

With these rotor scaling factors it is possible to scale the IM due to a variation of the rotor
resistance. This variation can be a result of a variation in the conductivity of the rotor conductor
by different materials, such as copper or aluminum, or by their different qualities, as well as by an
alternating temperature. Differences in the machine’s behavior due to a varying quality of the rotor
bar material can be calculated very fast with the proposed scaling process. In addition, a variation
of the rotor resistance due to the skin effect can be taken into account by using analytical formula
to recalculate the rotor resistance. This scaling process can also be used in combination with
a thermal model of the IM to simulate the machine in different operating points with varying
temperature conditions.
Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 687

4.7. Scaling the number of turns of the winding

In addition to a geometric scaling of the IM, a variation in the number of turns of the stator
winding NW can be useful. In an axially extended machine the induced voltage increases with k a .
As a result, the maximum voltage of the machine is reached for lower speed and thus the corner
point is at lower speed. To compensate this, the induced voltage can be reduced by reducing the
number of turns of the stator winding. This can be done in the post processing of the FE simulation
but a scaling factor of the number of turns of the winding k N enables shorter computation times
as one complete execute of the post processing procedure. Since the number of turns is an integer
number k N , it is defined with:
′
NW = NW k N , (33)
N′
kN = W . (34)
NW
The proportionals for the scaling of the number of turns of the winding are collected in
Table 1.
Table 1. Parameter proportionals for the scaling of the number of turns of the stator winding

Parameter Variable ∝
Number of turns of stator NW kN
Transformation ratio a kN
1
Stator wire cross sectional area Awire
kN
1
Stator current I1
kN
Stator flux density B kN
Stator resistance R1 k 2N
Stator copper losses PL,ohm,1 1
1
Stator voltage V1
kN

4.8. Scaling of the IM losses

The losses of an IM can be distinguished in Ohmic losses PL,ohm and iron losses PL,Fe . The
Ohmic losses of the rotor PL,ohm,2 are proportional to kr2 k R as in:
′
PL,ohm,2 = PL,ohm,2 kr2 k R . (35)
′
The scaled Ohmic losses of the stator PL,ohm,1 have to be calculated with the scaled stator
′ ′
current I1 and the scaled stator resistance R1 . This leads to the scaled Ohmic losses of the stator:
′
PL,ohm,1 = 3 · I1′2 R1′ , (36)
( )2
′ kr l Fe k a + lW H kr 2
PL,ohm,1 = 3 · I1 · 2N1 kN . (37)
kN σ1 Awire kr
688 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

The iron losses, in turn, can be separated into hysteresis, eddy current and excess losses [21].
All three parts have a different dependency on the frequency f . Hence, different scaling relations
′
for the three iron loss components exist. The scaled iron loss power density pL,Fe is:
( )
′ f f2 f 1.5
pL,Fe = khyst B α + keddy B2 2 + kexcess B 1.5 1.5 , (38)
kt kt kt
where khyst, keddy and kexcess are the hysteresis loss, eddy current loss and excess loss coefficient,
respectively. With the iron loss power density and the iron mass mFe , scaled with kr2 k a , the scaled
′
iron loss power PL,Fe is:
′ ′
PL,Fe = pL,Fe mFe kr2 k a . (39)
All scaling laws for the IM are summarized in Table 2.

Table 2. Scaling factors for the machine’s parameters

Parameter Variable ∝
Length l ka
Lateral surface Asurface k a kr
Cross sectional area Across kr2
Volume V k a kr2
Magnetic field strength H 1
Magnetic field density B 1
Magnetic field linkage Ψ k a kr
1
Current density J
kr
Current I kr
Time T kt
1
Frequency f
kt
1
Speed n
kt
Torque T k a kr2
k a kr
Voltage V
kt
Inductance L ka
ka
Reactance X
kt
Rotor resistance R2 kR
k a kr2
Mechanical power Pmech
kt
1 kR
Mechanical power density pmech =
kt ka
Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 689

4.9. Validation of the scaling laws

To validate the scaling laws for IMs a reference motor is calculated with the FEM in the first
path via the hybrid simulation approach described in section 3.2. In the second path, the geometry
and the rotor resistance of the IM are scaled and an FE simulation is performed with the scaled
machine. The FE solutions of the scaled IM are rescaled in the I1 - f 2 -plane to the parameters of
the reference machine regarding the procedure described in Fig. 1. Finally, the solutions in the
I1 - f 2 -plane are transformed to the T-n-map, considering the same requirements. The procedure is
shown in Fig. 5. The scaling factors in this simulation are kr = 1.2, k a = 1.1 and k R2 = 1/1.05.

Fig. 5. Validation procedure of IM scaling

The results of the FE solutions of the reference machine (first path) and the rescaled FE
solutions of the scaled machine (second path) are compared. The comparison is done in the
I1 - f 2 -plane and the T-n-map as described in Fig. 5. The calculated deviation of the loss power in
the I1 - f 2 -plane has a maximum error of 0.2 · 10−3 % that proves the correctness of the proposed
scaling scheme. The comparison of the reference loss power and the loss power of the IM rescaled
in the I1 - f 2 -plane and transformed to the T-n-map is presented in Fig. 6. It also shows a very
accurate performance of the proposed scaling procedure.

Fig. 6. Total loss deviation of the reference and the rescaled FE solutions in the T-n-map
690 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

In Fig. 7 the total losses transformed to the T-n-map of a reference IM machine and in Fig. 8
the total losses for a scaled machine, with the scaling factors kr = 1.2, k a = 1.1 and k R2 = 1/1.05,
are shown. For the transformation to the T-n-map the same requirements, such as the maximum
frequency, the maximum inverter current and dc-link voltage and the operation strategy MTPEL,
are taken into account. The maximum speed of the reference machine is set to 20 000 rpm which
results in a maximum surface velocity of the rotor of 110 m/s. By increasing the radial dimension
of the IM the speed limit has to be scaled according to (9). Thus, the maximum speed in Fig. 8
is about 16 666 rpm. In Fig. 9 the percentage deviation of the magnetic flux density ∆B in % of
a reference IM and an IM whose geometrical dimensions are scaled with kr = 1.2 and whose
rotor bar conductivity is scaled with k R2 = 1/1.2 are shown. Here, the input parameter, such
as the stator current I1 , the stator frequency f 1 or the rotor current frequency f 2 are not scaled
resulting in a non-constant field distribution. Therefore, ∆B reaches values of more than 50%.
Fig. 10 shows the percentage flux density deviation in the case that the input parameter are scaled
due to the proposed scaling laws in Table 2. The maximum value of ∆B is 0.02%.

Fig. 7. Total losses of the reference IM machine

Fig. 8. Total losses of the scaled machine calculated with the scaling laws

It shows that with the proposed scaling laws the assumption of a constant flux distribution is
fulfilled. For the calculation of ∆B the magnetic flux density B of the reference machine in each
point ( ρ, φ) is compared with the magnetic flux density of the scaled machine (B ′ ) in each point
( ρ′, φ ′ ) and mapped into the coordinates of the reference machine ( ρ, φ).
Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 691

(a) (b)
Fig. 9. Deviation of the flux density ∆B in % for the stator (a) and rotor (b)
of the reference and scaled IM without scaling of electrical parameters

(a) (b)
Fig. 10. Deviation of the flux density ∆B in % for the stator (a) and rotor (b)
of the reference and scaled IM with scaling of electrical parameters

5. Optimization of an induction machine for traction application

Due to its very fast performance, the proposed machine scaling scheme can be used in a numer-
ical optimization procedure. In [22] Stipetič and Žarko present an overview of the methodology
using mathematical optimization procedures to achieve an optimal design of an electrical machine.
They suggest a metaheuristic algorithm, such as Evolution Strategy or Differential Evolution, for
the complex electrical machine design. In [19], mixed integer distributed ant colony optimization
is used to optimize a traction drive with a permanent magnet motor, calculated by using the FEM
and geometrical scaling. In [14] multi-objective optimization in combination with the scaling
laws of an SM is used to find the optimal size of the traction motor, as it is done in [15], with the
gear ratio as an additional design variable.
692 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

5.1. Methodology
To show the potential and usability of the proposed IM scaling scheme an IM in an electric
vehicle is optimized. Changes of the machine configuration, such as the number of stator slots,
are not considered to focus on the radial and axial scaling. The gear ratio, the axial length and
the radius of the IM are used as the design parameters. The objective function that is minimized
in the optimization process, considers the costs of the machine and the produced loss energy of
the IM in the worldwide harmonized light vehicles test procedure (WLTP). Here, the costs of the
machine and the loss energy are weighted with different factors. The Evolutionary Strategy is
used as an optimization strategy. The procedure of it is shown in Fig. 11.

Fig. 11. Evolutionary strategy

Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 693

g
It starts with the initialization of the parameter set ΘE,1−µ and the step size of the parameters
g
δ E,1−µ , where µ is the number of the parents and g the generation number. For the initial parameters
the radial and axial scaling factor are set to kr = k a = 1 and the gear ratio is set to i gear = 8. The
initial step size is set to a fourth of the initial parameter values. The step size is averaged in the
second step and adapted by a logarithmic distributed random number. With the normal distributed
g
random number z and the step size the new parameter sets Θ N,k , named progenies, are calculated
in the variation step. With the new parameters the machine is scaled, the gear ratio changed and
the objective function is calculated. In the selection the best progenies become the new parents
parameter and the calculation of the next generation is started. The process ends after a certain
number of generations.

5.2. Results
The results of the design optimization with the Evolutionary Strategy is shown in Fig. 12.
It shows the machines with different parameters of the radial scaling factor kr , the axial scaling
factor k a and the gear ratio i gear that results in the minimum objective function. All different
parameter sets are marked with a gray point. The optimum parameter sets are marked by black
dots. It can be seen that the optimization algorithm varies the three variable parameters in a wide
range and converge fast towards an optimum region. The calculation of the objective function
for every single parameter set, including the scaling of the IM’s FE solutions, in Matlab takes
a processor time of ca. 0.8 s, using an Intel(R) Core(TM) i7-6500U CPU @ 2.5 GHz and an
8 GB RAM.

Fig. 12. Pareto front of the proposed optimization process

6. Conclusions
In this paper an IM scaling procedure for the machine’s 2D FE solutions is proposed. Besides
the geometrical scaling of the IM, the scaling laws for variations in the stator and rotor resistance,
for the maximum speed and for the number of turns of the stator winding are presented. The
reference FE solutions of the IM are scaled in the I1 - f 2 -plane and transformed to the T-n-map, by
considering boundary conditions, such as maximum current, and taking into account an operation
694 M. Nell, J. Lenz, K. Hameyer Arch. Elect. Eng.

strategy. The results of the scaling process show a very good agreement with the FE results of
the scaled IM. The scaling procedure is tested in the optimization of a traction drive. The results
of the optimization show that the proposed scaling process is suitable for the use in numerical
optimization processes.
The rotor resistance scaling can be used to analyze the machine’s behavior for different rotor
bar materials, their quality differences or different rotor temperatures. It can also be used in
combination with a thermal model of the IM to calculate the IM’s thermal behavior in difference
operation conditions. The proposed scaling methodology is a rapid and very accurate tool to scale
entire operation maps of IMs. In further work, the scaling laws for variations in the electrical
steel due to temperature differences or the quality of the material will be studied. Furthermore,
the limits of this method will be analyzed and the scaling will be validated for different machine
configurations and designs, such as closed rotor slots. A validation with experimental results will
be conducted. The focus of further publications can also be the optimization process itself.

References
[1] Mallik S., Mallik K., Barman A., Maiti D., Biswas S.K., Deb N.K., Basu S., Efficiency and cost
optimized design of an induction motor using genetic algorithm, IEEE Transactions on Industrial
Electronics, vol. 64, no. 12, pp. 9854–9863 (2017).
[2] Žarko D., Design of premium efficiency (ie3) induction motors using evolutionary optimization and
scaling laws, Przeglad Elektrotechniczny, vol. 1, pp. 183–186 (2016).
[3] Einfuegen S., Efficiency and cost optimized design of an induction motor using genetic algorithm,
IEEE Transactions on Industrial Electronics, vol. 64, no. 12, pp. 9854–9863 (2017).
[4] von Pfingsten G., Steentjes S., Hameyer K., Operating point resolved loss calculation approach
in saturated induction machines, IEEE Transactions on Industrial Electronics, vol. 64, no. 3,
pp. 2538–2546 (2017).
[5] Žarko D., Stipetič S., Martinovic M., Kovačić M., Jercic T., Hanic Z., Reduction of computational
efforts in finite element-based permanent magnet traction motor optimization, IEEE Transactions on
Industrial Electronics, vol. 65, no. 2, pp. 1799–1807 (2018).
[6] von Pfingsten G., Hameyer K., Highly efficient approach to the simulation of variable-speed induction
motor drives, IET Science, Measurement Technology, vol. 11, no. 6, pp. 793–801 (2017).
[7] Wood R.W., Scaling magnetic systems, IEEE Transactions on Magnetics, vol. 47, no. 10, pp. 2685–2688
(2011).
[8] Hsieh K.T., Kim B.K., One kind of scaling relations on electromechanical systems, IEEE Transactions
on Magnetics, vol. 33, no. 1, pp. 240–244 (1997).
[9] Stipetič S., Žarko D., Popescu M., Ultra-fast axial and radial scaling of synchronous permanent magnet
machines, IET Electric Power Applications, vol. 10, no. 7, pp. 658–666 (2016).
[10] Alberti L., Bianchi N., Boglietti A., Cavagnino A., Core axial lengthening as effective solution to
improve the induction motor efficiency classes, IEEE Transactions on Industry Applications, vol. 50,
no. 1, pp. 218–225 (2014).
[11] Bone J.C.H., Influence of rotor diameter and length on the rating of induction motors, Electric Power
Applications, IEE Journal on, vol. 1, no. 1, pp. 2–6 (1978).
[12] Lange E., Henrotte F., Hameyer K., An efficient field-circuit coupling based on a temporary lineariza-
tion of fe electrical machine models, IEEE Transactions on Magnetics, vol. 45, no. 3, pp. 1258–1261
(2009).
 Vol. 68 (2019) Scaling laws for the FE solutions of induction machines 695

[13] Williamson S., Begg M.C., Calculation of the resistance of induction motor end rings, IEE Proceedings
B – Electric Power Applications, vol. 133, no. 2, pp. 54–60 (1986).
[14] Ramakrishnan K., Stipetič S., Gobbi M., Mastinu G., Optimal sizing of traction motors using scalable
electric machine model, IEEE Transactions on Transportation Electrification, vol. 4, iss. 1, pp. 314–321
(2018).
[15] Ramakrishnan K., Stipetič S., Gobbi M., Mastinu G., Multiobjective optimization of electric vehicle
powertrain using scalable saturated motor model, in 2016 Eleventh International Conference on
Ecological Vehicles and Renewable Energies (EVER), Monte Carlo, Monaco, pp. 1–6 (2016).
[16] von Pfingsten G., Nell M., Hameyer K., Hybrid simulation methods for induction machine calculation
reduction of simulation effort by coupling static fea with transient fea and analytic formulations,
in 2017 18th International Symposium on Electromagnetic Fields in Mechatronics, Electrical and
Electronic Engineering (ISEF) Book of Abstracts, Lodz, Poland, pp. 1–2 (2017).
[17] Stipetič S., Žarko D., Popescu M., Scaling laws for synchronous permanent magnet machines, in
2015 Tenth International Conference on Ecological Vehicles and Renewable Energies (EVER), Monte
Carlo, Monaco, p. 1–7 (2015).
[18] Stipetič S., Goss J., Calculation of efficiency maps using scalable saturated flux-linkage and loss
model of a synchronous motor, in 2016 XXII International Conference on Electrical Machines (ICEM),
Lausanne, Switzerland, pp. 1380–1386 (2016).
[19] Žarko D., Kovačić M., Stipetič S., Vuljaj D., Optimization of electric drives for traction applications,
in 2017 19th International Conference on Electrical Drives and Power Electronics (EDPE), Dubrovnik,
Croatia, pp. 15–32 (2017).
[20] Nell M., Lenz J., Hameyer K., Efficient numerical optimization of induction machines by scaled fe
simulations, in 2018 XIII International Conference on Electrical Machines (ICEM), Alexandroupoli,
Greece, pp. 198–204 (2018).
[21] von Pfingsten G., Steentjes S., Hameyer K., Transient approach to model operating point dependent
losses in saturated induction machines, in 2016 XXII International Conference on Electrical Machines
(ICEM), Lausanne, Switzerland, pp. 626–632 (2016).
[22] Stipetič S., Miebach W., Žarko D., Optimization in design of electric machines: Methodology and
workflow, in 2015 Intl Aegean Conference on Electrical Machines Power Electronics (ACEMP), 2015
Intl Conference on Optimization of Electrical Electronic Equipment (OPTIM) 2015 Intl Symposium
on Advanced Electromechanical Motion Systems (ELECTROMOTION), Side, Turkey, pp. 441–448
(2015).

View publication stats