IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 65, NO.

2, FEBRUARY 2018 1711

Cooling Fin Optimization on a TEFC Electrical

Machine Housing Using a 2-D Conjugate Heat
Transfer Model
Stefan Ulbrich , Jeanette Kopte, and Jens Proske

Abstract—The housing of totally enclosed fan-cooled I. INTRODUCTION

electrical machines is usually finned to maximize the heat
N INDUSTRIAL applications, not only the energy efficiency
dissipation over its surface. Complex models involving heat
and fluid flow simultaneously are necessary to evaluate a
single design. An optimization over a wide parameter range
I but the steady-state temperature level of an electrical ma-
chine as well is of significant importance. Given the same
is, therefore, not possible. In this paper, a novel approach on amount of losses and similar loss distribution, the cooling system
how to model this particular conjugate heat transfer prob-
lem in 2-D without the help of lumped circuits is presented.
determines the rating into one of the thermal classes defined in
The proposed method is applied to an optimization process IEC-60034-11. In the case of a given maximum allowable tem-
controlled by an evolutionary algorithm. Comparative, more perature rise, the cooling system determines the power/torque
detailed simulations in computational fluid dynamic (CFD) to mass or volume ratio as well as the relative starting current
verify the tendencies obtained in the optimization. In an- for grid-connected machines. It is, therefore, quite beneficial
other case, the method is used to evaluate the allowable de-
gree of engineering tolerance for a given design. Although to maximize the capability to dissipate heat without lowering
the method does not provide results as accurate as CFD, it energy efficiency or raising the noise level at the same time.
is very helpful to study a large parameter set and determine A totally enclosed fan-cooled (TEFC) electrical machine
the CFD candidates. emits most of its losses via its finned outer surface that is blown
Index Terms—Cooling fins, electrical machine, multi- over with air as coolant. Therefore, it is necessary to consider
physics, thermal analysis, totally enclosed fan cooled both the fluid flow over the surface and the heat flow in the
(TEFC). housing simultaneously. This is often referred to as conjugate
heat transfer (CHT) problem [1]. Moon et al. [2] studied four
NOMENCLATURE cooling fin designs using computational fluid dynamics (CFD)
simulation and found notable differences between their perfor-
H Heat transfer coefficient (W/m²/K).
mances. Although they simplified the model using geometrical
D Channel hydraulic diameter (m).
symmetries, they stated that the high computational effort pro-
L Channel axial length (m).
hibited the consideration of more cases. A similar approach was
A Channel cross-sectional area (m²).
used in [3]. In [4], the heat transfer coefficient between finned
v Fluid average velocity (m/s).
housing and air was measured for different numbers and dimen-
v̇ Fluid volumetric flow rate (m³/s).
sions of the fins. The results obtained differ by as much as 20%,
Δp Static pressure loss (Pa).
which underlines the importance of a good housing design. Fur-
ka Fluid thermal conductivity (W/m/K).
ther studies [5]–[7] focused on the fluid flow in the channels. It
μ Fluid viscosity (kg/s/m).
was pointed out that the air velocity at the inlet as well as the
ρ Fluid specific density (kg/m³).
amount of air leaving before the end differs among the chan-
cp Fluid specific heat capacity (kJ/kg/K).
nels, depending on their location (top, side, etc.) and design.
Ra Housing surface roughness (mm).
The authors stated that complex relationships including, e.g.,
Re Reynolds number.
the cowling, fan, and rotational speed influence these aspects,
Nu Nusselt number.
but did not provide analytical descriptions. The work presented
Pr Prandtl number.
in [8] focused on the design optimization of a single fin to limit
Manuscript received February 27, 2017; revised June 30, 2017 and the computational cost. It was, therefore, possible to conduct
July 31, 2017; accepted August 7, 2017. Date of publication September an optimization run and study the influence of fin height and
4, 2017; date of current version December 8, 2017. (Corresponding thickness as well.
author: Stefan Ulbrich.)
The authors are with VEM Sachsenwerk GmbH, Dresden 01257, Detailed theoretically and experimentally verified studies
Germany (e-mail: sulbrich@vem-group.com; jkopte@vem-group.com; concerning the heat flow inside the cooling fins can be found
proske@vem-group.com). in [9]. The work also includes the fluid flow and relates cross-
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. sectional area, axial length, and fluid velocity to the heat transfer
Digital Object Identifier 10.1109/TIE.2017.2748051 coefficient.

0278-0046 © 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution
requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
1712 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 65, NO. 2, FEBRUARY 2018

Many authors have worked on multimodal, multiobjective, TABLE I

HEAT TRANSFER COEFFICIENT h FOR DIFFERENT HYDRAULIC DIAMETERS D
and/or multiphysics approaches for the optimization of electrical AND FLUID FLOW VELOCITIES v
machine designs, deploying various different types of gradient-
based or stochastical algorithms. These algorithms generate
v in D = 10 D = 30 D = 60 D = 90 D = 120
parameter sets that are subsequently analyzed in models of a m/s mm mm mm mm mm
more or less detailed degree, depending on the physical domain,
5 8.25 15.15 16.90 16.89 16.74
desired accuracy, or computational effort. For electromagnetic 10 16.09 27.53 29.83 29.71 29.22
problems, the use of finite-element analysis (FEA) is common 15 23.71 38.93 41.57 41.21 40.43
standard [10]–[12]. Heat- and fluid-flow-related issues are usu- 20 31.16 49.72 52.59 51.96 50.89
25 38.47 60.07 63.08 62.18 60.83
ally considered via lumped parameter network models [13]–[15] 30 45.67 70.07 73.17 71.99 70.37
or by simplified assumptions concerning, e.g., a maximum al-
lowable current density. Only a very small number of solutions
is studied using CFD [16] simulation, which is more often used
to look into details. This is due to the computational effort and
the complexity of the model setup, which can easily have a
strong influence on the results [17].
However, considering only a few designs contradict the goal
of optimization because the probability of not finding the op-
timum solution is high. Therefore, it is necessary to employ
an automatable method that is more detailed than lumped cir-
cuits but not too computationally expensive in order to solve
the CHT problem of cooling fins on a housing. One possible
approach is presented in this study. At first, the model reduction
to 2-D is motivated (see Section II). In Section III, the actual
model and the used evolutionary algorithm are introduced. The
2-D-FEA heat flow model with analytically calculated bound-
ary conditions is then used to optimize the finned structure of
an exemplary housing (see Section IV). Ten of these results are Fig. 1. Exemplary machine (IC411, frame size 400).
subsequently analyzed again with CFD, proving the account-
ability of the proposed method (see Section V). Finally, the diameter is defined as four times the cross-sectional area of
reduced model is used to evaluate whether a fine or a coarse the channel divided through its perimeter. Table I summarizes
engineering tolerance is required for the casting of the housing the results.
(see Section VI). Small hydraulic diameters result in low values for h in this
case, with the maximum at D = 60 mm. Not considered in the
II. PROBLEM DESCRIPTION above is that the hydraulic diameter also defines the flow resis-
tance. Higher values of D, therefore, result in more fluid flow.
The heat flow through rectangular fins of certain height and
Especially interesting is that the column of D = 60 mm includes
width to a surrounding fluid can well be described analytically.
the highest values of h, regardless of the flow velocity. This mo-
To do this, it is necessary to assume a constant heat transfer
tivates to only consider a 2-D cross-sectional area instead of
coefficient between fin surface and fluid. If the total surface area
the full length, neglecting the influence of the geometry on how
is held constant, optimal heat dissipation is achieved by having
much fluid will leave the channel before the end.
more but shorter fins. This is due to the limited heat flow inside
The losses of an electrical machine are not evenly distributed
the material the fins are made of. Yet a change in the geometry
between its different components and most of the heat is dissi-
also results in different fluid flow conditions so that the heat
pated via the finned housing surface. This means that thermal
transfer coefficient is not the same in both cases. Therefore,
energy needs to be transported from the active parts to the hous-
an additional analytical approach to model the fluid flow con-
ing and then to the surrounding air. For many larger machines,
ditions was implemented in [9]. Based on experimental results
there is usually an additional inner cooling cycle whose warm
obtained by Nusselt for similar arrangements and turbulent flow,
air has to recirculate and dissipate heat throughout the housing
(1) and (2), shown below, are used to calculate the heat transfer
as well. This happens in the “inner air channels” (see Fig. 1),
coefficient h with the help of the dimensionless parameter m
where the airflow is in the same or opposite direction as the
 0.214
L0.946 ka outside flow through the channels.
m = 0.1448 · 1.16 · (1) Fig. 2 depicts the surface temperature of a frame 400 machine
D ρ · cp · v
at its rated point as a result of a 3-D CHT simulation on the left
ρ · cp · v · D  
h= · 1 − e−m . (2) side. Next to it is the surface temperature of such a machine as
4·L measured with an infrared camera is shown. Table II contains
If the axial length L of a finned configuration is set to a fixed the temperatures simulated and measured on a few specific lo-
value of 1, 4 m, it is possible to calculate h for various different cations. To achieve this high level of agreement between simu-
fluid flow velocities v and hydraulic diameters D. The hydraulic lation and measurement, it is necessary to model the boundary
ULBRICH et al.: COOLING FIN OPTIMIZATION ON A TEFC ELECTRICAL MACHINE HOUSING USING A 2-D CONJUGATE HEAT TRANSFER MODEL 1713

Fig. 2. Surface temperature in CFD (left: Red hottest and blue coldest)
and test (right: Yellow hottest and violet coldest).
Fig. 3. Simulation process for each solution.

TABLE II
SIMULATION AND TEST RESULTS AT SPECIFIC LOCATIONS

Point Temperature CFD Temperature IR

1 86 °C 88 °C
2 40 °C 43 °C
3 75 °C 72 °C

layers of the fluid part detailed enough. This is expressed by the

dimensionless factor y+, which should be around 1. Needless
to say that this was computationally expensive and required a
manual mesh setup. Both pictures show that the surface temper-
ature increases along the flow path, but changes abruptly at the
end of the stator iron package. The surface temperature rise in
axial direction differs among the channels though. Those which
are nearer to the closed inner air channels are generally hotter Fig. 4. Smallest recurring segment and its location on the housing.
than those further away. Blockings, such as the terminal box,
worsen the flow conditions, and, thereby, the heat dissipation,
but are usually unavoidable due to various reasons. Referring to
layouts. The goal of the optimization is to minimize the mean
the 2-D model, this means that it is not sufficient to look at only
temperature on the inside of the housing (min f1 ). In order to
one channel and the additional heat coming from the inner air
make the solutions comparable, the heat flow into the studied
circle needs to be considered as well if present.
segment is fixed (Ptotal = Pstator + Piair ). Fig. 3 shows the se-
Additionally, the fluid flow velocity in the channels was eval-
quence of steps for each individual solution. First, the parameter
uated and measured. It was found that an average of 40% of
set is generated by an evolutionary algorithm. In the next step,
the airflow going into the channel went out before the end. This
the convective boundary conditions for the 2-D simulation are
does not automatically mean that closed channels would be bet-
determined (see Section III-B), making use of assumptions. To
ter. Test results with totally closed channels on the same machine
correct these, the results of the initial FEA simulation are used
showed higher winding temperatures than in the case with open
to recalculate the boundary conditions. The results of the second
channels. According to considerations mentioned in [7] and test
run are then saved and used to evaluate the objective function
results presented in [18], this might be due to a reduced air flow
(3), shown below, of each solution. However, the main goal for
rate because of a larger pressure drop along the channel. The
the optimization process is to create a number of promising can-
measured air velocities in the corresponding channels showed
didates for further analysis or selection due to practical reasons
that a turbulent flow is to be expected during rated operation of
 βm a x
the machine. Thus, natural convection and radiation have only 1
f1 = · T (β) dβ. (3)
a little influence and can be neglected. βm ax − β0 β 0

III. MODELING CHT IN 2-D A. Geometry

Based on the explanations given in Section II, it is possible The smallest recurring segment of the finned housing is stud-
and necessary to model heat and fluid flow simultaneously in ied. It is located at the top right part and defined by geometrical
order to optimize the finned structure of a housing. Due to the constraints for a given design, as Fig. 4 shows. Between the
complexity of the geometry, a 2-D stationary heat flow FEA is cooling fins lie the channels through which the air flows in axial
used. Its boundary conditions are being determined by analyti- direction. Only half of the leftmost channel is included in the
cal equations. In contrast to a CFD model setup, the relatively considered geometry, but fully in the calculation of the boundary
low computational effort allows us to study a large number of conditions.
1714 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 65, NO. 2, FEBRUARY 2018

TABLE III
GEOMETRIC PARAMETER OVERVIEW

Parameter Unit Description

α deg Opening angle of fin flanks

βm a x deg Angle from center to begin of inner air channel
δ deg Angle between adjacent fins
dh o mm Outer diameter of housing
wf mm Width of a fin in the top
wh mm Housing wall thickness
df mm Minimum distance between fins
Y0 mm Maximum height of housing outer shape
F – Parameter contour function
n – Parameter contour function
Nf – Number of fins in segment
Fw f – Factor for increasing fin width
Fδ – Factor increasing angular distance

Fig. 5. Heat and fluid flow conditions.

A parameter set is defined in Table III to describe all possible
modifications of the geometry and to allow the use of an auto- The Reynolds number needs to be calculated with an assumed
mated algorithm. Each fin may have a different length, but the average speed v. Using the dimensionless coefficient ξ, it is
top of each fin must be located on an outer contour which is possible to include the roughness Ra
defined by a function. By variation of the parameters F and n, ρvD
it is possible to consider many different shapes from circular to Re = (5)
μ
almost rectangular. The dimensionless factors Fwf and Fδ con-
trol the change of fin width and angular distance between two 0.25
ξ=  2 (6)
adjacent fins from fin to fin or channel to channel, respectively. 15 Ra
lg Re + 3.715·D
This shall allow for a broader variety in the created geometry
L
combinations by allowing the distance df , and height between R = 0.5 · ρ · ξ · . (7)
the fins to increase or decrease. D·A
After determining the volumetric flow rate v̇, the Nusselt
B. Conjugate Heat Transfer number Nu and the heat transfer coefficient h may be calculated
for the considered channel under the assumption of turbulent air
To model the problem, two sources of heat and a number of flow. The temperature-dependent Prandtl number Pr is used as
sinks according to the number of channels need to be set. An well
evenly distributed heat flow of constant total value is applied to  0.67

the inner side of the housing (Pstator ). This is the major source of D
Nu = 1 +
heat coming from the stator. Additional heat is inserted from the L
warm air of the inner circuit (Piair ). The sum of both fractions
is the total amount of heat dissipated (Ptotal ). The heat transfer 0.125[1.82lg (Re) − 1.64]−2 · (Re − 1000) · Pr
·  (8)
into the channels is defined by convective boundary conditions. 1 + 12.7 0.125[1.82lg (Re) − 1.64]−2 · (Pr0.67 − 1)
These are set for each channel individually, according to their
geometry. For the evaluation of their geometry, only the cross ka · Nu
h= . (9)
section is regarded and any influence on the in- or out-flow D
conditions is neglected. Fig. 5 depicts the mentioned boundary Finally, h is applied to the limiting sides of the channel, except
setup. for the opening. After a first simulation, the results are used to
In order to define a convective boundary condition, it is neces- recalculate the average temperature and correct h.
sary to set a heat transfer coefficient and an average temperature. In FEA, the mesh quality may have a strong impact on the
It is the determination of these two values which links heat and results. A mesh study was, therefore, conducted to determine
fluid flow and makes it possible to consider both at the same the parameters for the automatic mesh generation of the used
time. Seen from the direction of the airflow, all channels are par- software finite element method magnetics. Fig. 6 shows a fine
allel to each other. It is assumed that changes in their geometry and coarse variant, whose results differed by merely 2%. To
do not have a significant impact on the operating point of the fan avoid any mesh-dependent problems, the parameters that created
and a constant pressure loss Δp in each separate channel can the fine mesh resolution where chosen for this study.
be set. The average air temperature in each channel is initially
set to 40 °C, defining the properties such as specific density ρ, C. Algorithm and Parameter Setup
viscosity μ, and thermal conductivity ka . The flow resistance R
and the volumetric flow rate v̇ are then calculated iteratively to To find the optimum solution, a high number of variable pa-
fulfill the condition set in rameters and a reasonable range need to be set. Some constraints
like a minimum fin thickness or minimum distance between ad-
Δp = R · v̇ 2 . (4) jacent fins need to be regarded as well. An efficient way to
ULBRICH et al.: COOLING FIN OPTIMIZATION ON A TEFC ELECTRICAL MACHINE HOUSING USING A 2-D CONJUGATE HEAT TRANSFER MODEL 1715

Fig. 6. Automatically generated mesh with fine (left) and coarse (right)
resolution.

TABLE IV
VARIABLE PARAMETERS
Fig. 7. Mean temperature at the inside of the housing dependent on
Parameter Minimum Maximum the number of fins (each cross is one solution).

df [mm] 10 20
F 0.008 0.011
n 1.4 1.6
Nf 5 10
Fw f 1 1.05
Fδ 1 1.05
w f [mm] 7 10

TABLE V
FIXED PARAMETERS

Parameter Value

α [deg] 5 Fig. 8. Comparison of five exemplary results.

β [deg] 38
dh o [mm] 417
Y 0 [mm] 458 be at least seven to ten times the number of variable parameters,
w h [mm] 17
Δp [Pa] 300
100 possible solutions per generation were considered here.
Ra 0, 1
P to ta l [W] 4000
IV. RESULTS
Multiple runs of 26 generations were conducted. Each run
took about 4 h on a single standard computer, without the help
search the optimum in the solution space defined by the variable of parallelization. During the runs, the minimum, maximum,
parameters is the use of an evolutionary algorithm. In this case, and average value of Tm ean in every generation were observed.
a differential evolution algorithm of the type DE/rand/1/bin (ac- The decision to stop after 26 generations was made because
cording to [19]) has been implemented. Main advantages of the only small changes in the mentioned values were observed.
algorithm are its high robustness against convergence into local The initial assumption of turbulent flow was confirmed by the
optima and its little number of control parameters. Addition- range of the Reynolds number and fluid flow velocity. While the
ally, test results in other studies related to electrical machine Reynolds number ranged from 30 000 to 65 000, the velocity
optimization (e.g., [20]) proved its competitiveness. The im- was found to be between 20 and 28 m/s in the channels.
plementation into other stochastic algorithms, such as particle Many of the 2600 solutions lie closely together regarding
swarm optimization or genetic algorithms, would be an option the goal of the optimization, the lowest mean temperature at the
as well but was not considered for this study. Although a multi- inside of the housing. Nevertheless, the difference between the
objective optimization would be possible, only the minimization best and worst result is 19 K, as Fig. 7 shows. The gap between
of the mean temperature at the inside of the housing was chosen the results of the same number of fins appears to be smaller,
as objective (3). This is because the optimization run converges making it less probable to pick a “bad” variant. However, this
faster the fewer objectives are considered [21]. Tables IV and V is mainly due to the algorithm setup where only competitive
show the parameter setup for an exemplary run for the design results were saved and others discarded early.
of a housing for a frame size 450 machine, which has to dis- Designs with nine fins show the best result (see Fig. 7), being
sipate a total of about 32 kW in losses via its casted housing. much better than those with up to eight fins. Interestingly, a
According to the recommendations given in [22], where Price further increase in the number of fins has no positive effect. In
et al. found that the number of individuals per generation shall Fig. 8, five exemplary designs are shown. Analytically derived
1716 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 65, NO. 2, FEBRUARY 2018

Fig. 9. Mean temperature at the inside of the housing dependent on Fig. 11. Comparison between results of 2-D and 3-D simulations.
the mean hydraulic diameter.

hydraulic diameter is the average value across all channels,

while the total hydraulic diameter is the sum of all channels.
Both values have to be looked at differently. Obviously, a better
heat dissipation can be achieved with any of the two values
being maximized, but regarding only the maximum mean value
could lead to a false result as the groups determined by the
number of fins in Fig. 9 make clear. However, a maximum total
hydraulic diameter as the sum of all channels seems to be more
important to reach the optimum and could, therefore, be chosen
as a universal criterion. The position of results with ten fins in
Fig. 10 confirms this thought.

Fig. 10. Mean temperature at the inside of the housing dependent on

the total hydraulic diameter.
V. VERIFICATION WITH CFD
statements from [9] concerning the thickness of fins are con- To verify the proposed method and rate its significance, ten
firmed: Above a certain—but low—-value, there is no benefit results were examined with 3-D-CFD CHT simulation using
in having more heat conduction by thicker fins because a larger Ansys CFX. Besides nine results from the run presented in
portion of the channel is blocked. This can be seen in the two Section IV, another one was selected from a different run that
examples on the top left. The cross-sectional area available for allowed for more fins. Additionally, to the housing itself, the
the airflow is also reduced when many fins are present, even surrounding air region and the cowling were modeled to in-
when they are thin. This is also the reason why results with ten clude the in- and out-flow area as well as the effect of air leav-
fins are not as good as those with nine for the given space. A ing the channel before the end. Not included were the rotating
further disadvantage for ten fin configurations is due to the geo- fan and the terminal box. To make both models comparable,
metrical constraints. Because of the minimum distance between the volumetric flow rate at the inlet, the heat flow from the
adjacent fins and the angle α, the height of thin fins is limited. inside of the machine, and the rightmost fin were set to the
None of the better results follow a circular shape of the outer same values.
contour, meaning that the height of the fins is steadily increasing Fig. 11 shows the results of this comparison, with the 2-D and
from the left to the right. The highest local temperature is to be correspondent 3-D design being encircled. In each pair, the 2-D
found at the bottom right, where the additional heat flow from solution has a lower temperature value, but the absolute value
the inner cooling circuit is inserted. of the difference is not constant. This is probably due to the
However, a sensitivity study on the impact of the parameters different amount of air leaving the channel which was found not
used was not performed and the results may, therefore, not be to be constant. Although the results of both modeling approaches
generalized. Different housing material or another arrangement differ, the general tendency remains the same in almost all cases.
of the cooling fins, e.g., in radial direction, could lead to a change A cooling fin design that is better in comparison to the others
in the results. in 2-D remains so also in 3-D, in most cases. It is, therefore,
According to the argumentation in Section II, the hydraulic concluded that the proposed method can be used to evaluate a
diameter has a strong impact on the obtained results. This is large number of designs before a few promising candidates are
underlined by the results depicted in Figs. 9 and 10. The mean selected for a more detailed approach.
ULBRICH et al.: COOLING FIN OPTIMIZATION ON A TEFC ELECTRICAL MACHINE HOUSING USING A 2-D CONJUGATE HEAT TRANSFER MODEL 1717

REFERENCES
[1] A. Boglietti, A. Cavagnino, D. Staton, M. Shanel, M. Mueller, and C.
Mejuto, “Evolution and modern approaches for thermal analysis of elec-
trical machines,” IEEE Trans. Ind. Electron., vol. 56, no. 3, pp. 871–882,
Mar. 2009.
[2] S. H. Moon, Y. H. Jung, and K. W. Kim, “Numerical investigation on
thermal-flow characteristics of a totally enclosed fan cooled induction
motor,” in Proc. Int. Conf. Electr. Mach., 2016, pp. 1930–1935.
[3] J. Kuria and P. Hwang, “Optimizing heat sink geometry for electric vehicle
BLDC motor using CFD,” in Proc. Sustain. Res. Innov. Conf., 2011, vol. 3.
[4] M. A. Valenzuela and J. A. Tapia, “Heat transfer and thermal design of
finned frames for TEFC variable-speed motors,” IEEE Trans. Ind. Elec-
tron., vol. 55, no. 10, pp. 3500–3508, Oct. 2008.
[5] A. Boglietti, A. Cavagnino, and D. Staton, “TEFC induction motors ther-
mal models: A parameter sensitivity analysis,” in Proc. 39th Ind. Appl.
Fig. 12. Spread of results for the mean temperature at the inside of the Soc. Annu. Meet. Conf. Rec. IEEE Ind. Appl. Conf., 2004, vol. 4, pp. 2469–
housing for coarser and finer tolerance. 2476.
[6] D. Staton and A. Cavagnino, “Convection heat transfer and flow calcu-
lations suitable for electric machines thermal models,” IEEE Trans. Ind.
VI. TOLERANCE EVALUATION Electron., vol. 55, no. 10, pp. 3509–3516, Oct. 2008.
[7] D. Staton, A. Boglietti, and A. Cavagnino, “Solving the more difficult
According to the chosen engineering tolerance, the dimen- aspects of electric motor thermal analysis in small and medium size in-
sions of a casted housing may vary. This means that, e.g., single dustrial induction motors,” IEEE Trans. Energy Convers., vol. 20, no. 3,
pp. 620–628, Sep. 2005.
fins could be higher/smaller or thicker/thinner than they are [8] Y. Chen, B. Chen, and C. Chen, “CFD thermal analysis and optimiza-
meant to be. To evaluate how sensitive a design is to these vari- tion of motor cooling fin design,” in Proc. Heat Transf. Conf., 2005,
ations or to define which degree of engineering tolerance would pp. 625–629.
[9] F. Heiles, “Ueber die zweckmaessige gestaltung anordnung von
be acceptable, the 2-D model can be used. For every design that kühlrippen,” Elektrotechnik Maschinenbau, vol. 14, pp. 323–331, 1952.
shall be evaluated, new constraints for each tolerated dimen- [10] M. Amrhein, T. O’Connell, and J. Wells, “An integrated design process
sion are set, e.g., wf ± 0.2 mm. A Monte–Carlo algorithm then for optimized high-performance electrical machines,” in Proc. Int. Electr.
Mach. Drives Conf., 2013, pp. 847–854.
randomly creates new designs which are subsequently studied [11] S. Gunther, S. Ulbrich, and W. Hofmann, “Driving cycle-based design
in the way presented in Sections III-A and B. Fig. 12 shows the optimization of interior permanent magnet synchronous motor drives for
results of two runs with a finer and coarser engineering toler- electric vehicle application,” in Proc. Int. Symp. Power Electron. Electr.
Drives, Autom. Motion, 2014, pp. 25–30.
ance for one solution with eight fins. It is obvious that in this [12] J. Buschbeck, M. A. Vogelsberger, A. Orellano, E. Schmidt, and M.
case, the spread of results is larger with the coarser tolerance, Bazant, “Multi-physics optimization of high power density induction ma-
but the difference in the temperature range is less than 2 K. chines for railway traction drives,” in Proc. Int. Conf. Ind. Technol., 2015,
pp. 2656–2661.
If the total temperature range is regarded as too high even [13] N. Bracikowski, M. Hecquet, P. Brochet, and S. V. Shirinskii, “Multi-
with the finer tolerance, then other designs may be evaluated physics modeling of a permanent magnet synchronous machine by using
as well. lumped models,” IEEE Trans. Ind. Electron., vol. 59, no. 6, pp. 2426–2437,
Jun. 2012.
[14] W. Jiang and T. M. Jahns, “Coupled electromagnetic/thermal machine
VII. CONCLUSION design optimization based on finite element analysis with application of
artificial neural network,” in Proc. IEEE Energy Convers. Congr. Expo.,
A new method to simplify the simultaneous calculation of 2014, pp. 5160–5167.
a fluid and heat flow (CHT) problem was presented and ap- [15] J. Goss, D. Staton, R. Wrobel, and P. Mellor, “Brushless AC interior-
permanent magnet motor design: Comparison of slot/pole combinations
plied to find an optimized design for the cooling fins of a casted and distributed vs. concentrated windings,” in Proc. IEEE Energy Convers.
housing. The approach is especially useful to evaluate large Congr. Expo., 2013, pp. 1213–1219.
numbers of possible designs automatically in a short time. Due [16] Z. Huang and J. Fang, “Multiphysics design and optimization of high-
speed permanent-magnet electrical machines for air blower applications,”
to the applied simplifications, the results are not as exact as IEEE Trans. Ind. Electron., vol. 63, no. 5, pp. 2766–2774, May 2016.
3-D CHT simulations. Nevertheless, it was shown that ten- [17] P. H. Connor, C. N. Eastwick, S. J. Pickering, C. Gerada, and R. Rolston,
dencies found in 2-D remained the same in 3-D in most “Stator and rotor vent modelling in a MVA rated synchronous machine,”
in Proc. Int. Conf. Electr. Mach., 2016, pp. 573–579.
cases. Therefore, it is most promising to implement the method [18] A. Bierman and H. Ellerbrock, “The design of fins for air-cooled
prior to choosing a few candidate designs for a detailed CFD cylinders,” NASA, Washington, DC, USA, Tech. Rep. NACA-TR-726,
analysis. pp. 401–426, 1989.
[19] R. Storn, “On the usage of differential evolution for function optimiza-
The method may be extended by a more detailed approach tion,” in Proc. Biennial Conf. North Amer. Fuzzy Inf. Process. Soc., 1996,
to model the inner cooling circuit. It shall be implemented to pp. 519–523.
design a new housing for a frame 450 machine in the near future. [20] Y. Duan and D. M. Ionel, “A review of recent developments in elec-
trical machine design optimization methods with a permanent-magnet
Additionally, it can be used to compare different ways of how synchronous motor benchmark study,” IEEE Trans. Ind. Appl., vol. 49,
to fin a housing. no. 3, pp. 1268–1275, May/Jun. 2013.
Another interesting result is the importance of the total hy- [21] J. Kacprzyk and W. Pedrycz, Eds., Springer Handbook of Computational
Intelligence, 1st ed. Berlin, Germany: Springer, 2015.
draulic diameter. In case it could be taken as a universal param- [22] K. Price, R. Storn, and J. Lampinen, Differential Evolution: A Prac-
eter to maximize heat transfer, this could simplify the approach tice Approach to Global Optimization. New York, NY, USA: Springer,
even more. 2005.
1718 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 65, NO. 2, FEBRUARY 2018

Stefan Ulbrich received the Dipl.Ing. degree in Jens Proske was born in Dresden, Germany.
electrical engineering from Dresden University He received the Dipl.Ing. degree in electrical en-
of Technology, Dresden, Germany, in 2013. gineering from Dresden University of Technol-
For 2.5 years, he was with the Department ogy, in 1992.
of Electrical Machines and Drives, Dresden He worked until 1996 as an Engineer with
University of Technology. Since 2016, he has AEG Lloyd Dynamo, Bremen, Germany. After-
been with VEM Sachsenwerk GmbH, Dres- ward, he was a Scientific Assistant with the In-
den. His work focuses on the electromagnetic stitute of Electrical Power Engineering, Dresden
and thermal design of electrical machines for University of Technology, until 1999. He then
industrial purposes. joined VEM Sachsenwerk GmbH, Dresden, as
a Development Engineer. Since 2001, he has
been the Head of the Calculation and Development Department.
Mr. Proske has been a Member of the national standardization
Jeanette Kopte received the Dipl.Phys. degree committee “rotating electrical machines” since 2007.
in physics from Dresden University of Technol-
ogy, Dresden, Germany, in 1989.
Until 2011, she was with Dresden Univer-
sity of Applied Sciences, working in the field
of thermal analysis, e.g., of internal combus-
tion engines, after treatment of exhaust gas
and heat pipes. Since 2011, she has been
with VEM Sachsenwerk Gmbh, Dresden. She
is responsible for the thermal design of the
company’s products.