International Journal of Thermal Sciences 130 (2018) 406–415

Contents lists available at ScienceDirect

International Journal of Thermal Sciences

journal homepage: www.elsevier.com/locate/ijts

Experimentally validated numerical model of thermal and flow processes T

within the permanent magnet brushless direct current motor
B. Melkaa,∗, J. Smolkaa, J. Hetmanczykb, Z. Bulinskia, D. Makielab, A. Ryfaa
a
Silesian University of Technology, Institute of Thermal Technology, Konarskiego 22, Gliwice, 44-100, Poland
b
Silesian University of Technology, Department of Power Electronics, Electrical Drives and Robotics, B. Krzywoustego 2, Gliwice, 44-100, Poland

A R T I C LE I N FO A B S T R A C T

Keywords: In this paper, a numerical model describing the heat transfer and air flow inside and outside the casing of the
Electric motor permanent magnet brushless direct current (PM BLDC) motor is presented. In the model, a conjugate heat
Permanent magnet brushless DC motor transfer including the heat conduction with heat sources in windings, magnets, bearings, natural and forced
Thermal analysis convection and radiation phenomena within internal and external air domain were analysed. The complex
CFD
geometrical model included windings, a magnet circuit, a rotor with neodymium magnets, a PCB electronic
Power loss
plate, plastic covers, internal air between solid parts as well as a portion of the external air. The validation
LDA
process of the developed model was realised on the basis of the velocity and temperature fields. The experi-
mental data were recorded by 7 constant temperature anemometers at 4 levels above the motor housing, Laser
Doppler Anemometry sensors in two axes inside the rear portion of the motor and 25 thermocouples fixed inside
and outside the motor. The experimental and numerical tests were performed for 4 motor loadings and 3 ro-
tational speeds. The average temperature error for the internal point was of 9 K, while that of the external points
was of 2 K. The velocity field was very well predicted close to the housing wall in terms of horizontal and vertical
components. A slightly worse agreement was found for the vertical component, especially in the vicinity of the
shaft. The results were consistent for all the considered motor loadings and rotational speeds.

1. Introduction with a mechanical commutator requires a special maintenance monitor

Environmental, political and industrial tendencies have directed a
reduction of non-renewable sources of energy consumption [1]. One
way to achieve this goal is to increase the use of electrical power drives
in vehicles and other machines to replace internal combustion engines
of lower efficiency. Hence, a direct current electric motor is expected
have wider application as a direct source of drive [2]. Today, the pro-
blem with implementing this concept, on the large scale, is the lim-
itation of electric power storage in vehicles and small systems. Never-
theless, the prospective for electric power storage technology and its
improved maintenance is promising [3–6]. An additional challenge to
solve with increased electric vehicle usage is the integration of charging
stations into the electric grid. Nevertheless, many studies and proposed
solutions can be found [7,8]. Therefore, in the near future, increased
use of small power electric motors powered by direct currents can be
expected.
The most popular technologies for motors powered by direct current
are brush or brushless solutions. Brush motors are cheaper to produce
and their construction is simpler [9]. Nevertheless, the brush solution
and requires systematic brush replacements. Moreover, the efficiency of
brush motors decreases with additional losses of the mechanical com-
mutation and higher electrical contact resistance [10]. The brushless
solution is free of the above-mentioned disadvantages [11]. A perma-
nent magnet brushless motor powered by direct current (PM BLDC
motor) is characterised by high-efficiency, high-power density and lack
of mechanical commutation [12]. The main disadvantages of this ma-
chine are the high market price of the motor with an electronic com-
mutator and the electronic commutation complexity. Nonetheless, they
are becoming ever more popular due to the mentioned advantages.
Therefore, this paper focuses on this type of motor.
In electric motor design, thermal analysis plays an important role
[13]. It helps estimate the value and location of the maximum tem-
perature in the machine. The specific material and components of the
motor should be protected from exceeding the maximum safe working
temperature. Hence, proper estimation of the maximum temperatures
occurring during exploitation could help reduce the risk of overheating
and failure of the machine [14]. A precise thermal analysis allows for
decreasing the machine size by reducing the safety margin and,

∗
Corresponding author.
E-mail address: Bartlomiej.Melka@polsl.pl (B. Melka).
URL: http://www.itc.polsl.pl/bmelka (B. Melka).

https://doi.org/10.1016/j.ijthermalsci.2018.04.029
Received 11 October 2017; Received in revised form 20 April 2018; Accepted 20 April 2018
Available online 26 May 2018
1290-0729/ © 2018 Elsevier Masson SAS. All rights reserved.
B. Melka et al.
consequently, decreasing the material costs. The optimisation of the
thermal motor behaviour could also minimise the copper losses of the
machine by decreasing the electrical resistivity of the windings de-
pending on the temperature. Due to these facts, thermal analysis should
always be included in the electric motor design process.
One of the simplest and the most popular models to predict the
temperature distribution in electric machines is the lumped parameter
model (LPM) [15]. LPM in the thermal analysis of electric machines is
also known as the thermal network [16]. Complex thermal networks
contain dozens of the elements - nodes [17]. Usually, LPM allows the
temperature of the entire element to be calculated, e.g., one tempera-
ture for winding, when it is based on the zero-dimensional model
without spatial resolution. Moreover, it does not contain precise in-
formation about the coolant flow. The essential parameters used in the
thermal networks, e.g., convective heat transfer coefficients, are as-
sumed or calculated usually from the empirical equations, which are
presented in the literature for the general simplified shapes and geo-
metries. These equations were derived from the heat transfer field
analysis, e.g., in Ref. [18]. The main advantage of thermal networks is
their sufficient accuracy with short computational times. Even simple
thermal networks produce reasonable temperature response for short-
time thermal transient [19]. It is possible, especially for the machines
with high thermal inertia, while the thermal capacitance of applied
materials was correctly assumed. The temperature prediction in the
electric motors with permanent magnets can also be realised by means
of numerical simulation and on-line measurements together [20].
A more complex and accurate method for estimation of the tem-
perature distribution is Computational Fluid Dynamics (CFD). In this
method, many characteristic parameters, such as the heat transfer
coefficient, can be calculated directly and locally within a selected
computational domain. This is possible by extending the model geo-
metry to space outside the machine and, consequently, by considering
the heat generated inside the motor and then dissipated directly to the
ambient air [21,22]. As a result, a more detailed analysis leads to more
accurate heat dissipation predictions. In the literature, there are many
studies that compare CFD and LPM with respect to experimental data
[23]. One of the most important works about thermal modelling of the
electric motor operation describes the CFD model formulation and the
usage of the CFD field results in the LPM analysis [24]. In the literature,
there are also studies based on the Finite Element Method, which is
usually applied when researchers focus only on the thermal conduction
in solids. One of the studies with thermal conduction analysis in the
motor housing is the study of [25]. In the cited work, the authors as-
sumed heat flux from the internal surface of the housing frame and
analytically calculated the heat transfer coefficient. Consequently, the
estimated temperature field of the finned housing was compared with
the experimental study conducted. Zhang et al. [26] presented a cou-
pled analysis between the electromagnetic and thermal solvers. How-
ever, the external heat transfer coefficient was calculated analytically,
while radiation between internal elements was assumed to be negli-
gible. An interesting work was presented by Jungreuthmayer et al.
[27], where thermal analysis was conducted for a water-cooled electric
motor. Oil spray cooling of internal parts of electric motor was in-
vestigated experimentally in Ref. [28] and also numerically in Ref.
[29]. Heat storage mechanism allowing cooling intensification of the
transient mode of electric motor using phase change material was in-
vestigated in Ref. [30]. The multi-physics simulation connecting
thermal analysis with mechanical analysis was presented for case of
aircraft actuator in Ref. [31].
Nevertheless, studies showing the flow field inside and outside the
motor casing are not prevalent in the literature. The velocity mea-
surements were made in a water-filled axial machine in the work of
Aubert et al. [32]. Preliminary results of the air velocity validation were
also described in our previous work [33]. Moreover, the velocity vali-
dation caused by natural convection from a small heat source using
Particle Image Velocimetry technique in the closed cavity was
407
International Journal of Thermal Sciences 130 (2018) 406–415
presented in Ref. [34]. The authors obtained satisfactory agreement
between the model and experimental results. They suggested the RNG
k− ε model for turbulence modelling in the heat dissipation process via
natural convection.
The main aim of the presented paper was to determine the hot-spot
temperature and flow fields in terms of a safe and long-term operation
of the analysed low-power PM BLDC motor. To achieve this goal, the
temperature and the velocity fields were determined within and around
the machine housing under various loadings and rotational speeds. The
fields were computed on the basis of the developed thermal model that
included the computational domain of a very complex internal struc-
ture of the considered motor. In addition, the block of the surrounding
ambient air was formulated.
To validate the model, a test rig was built to measure temperature
and velocity fields at specified locations inside and outside the motor
housing. A set of calibrated thermocouples was used to record the va-
lues at the stator magnetic core, housing and in the vicinity of the
windings. Moreover, the temperatures were also captured at a number
of locations above the machine housing. The velocity field was mea-
sured by means of the Laser Doppler Anemometry (LDA) technique in
the rear part of the motor. The velocity validation inside the motor
housing and around the motor is a challenging task. Therefore, this
problem is rarely presented in the literature. In the presented study, the
LDA technique was used showing its strengths and weaknesses in this
application. In addition, the anemometers were used to find the velocity
fields above the motor at four levels. All the mentioned parameters
were measured for the motor working at four loading conditions at
various rotational speeds reaching the maximum value of 3500 rpm.
Namely, the considered PM BLDC motor was experimentally tested at
11 operating conditions.
A temperature comparison of the solid motor parts shows very good
agreement between the numerical and experimental results. Similarly,
the velocity profile inside and the field outside the housing was pre-
dicted accurately for most measurement points. The study also shows
some problems in the velocity predictions at some near-wall locations.
2. Test rig
2.1. Motor description and test rig components
The studied machine was a low-power electric motor with perma-
nent magnets on the rotor. The motor was supplied by the electronic
commutator transforming direct current. The hall effect sensor was
used to control the rotor position in the control process. The rated
parameters of the rotor were as follows: the output power of 431 W, the
input voltage of 24 V and the current of 21 A.
The test rig consisted of two PM BLDC motors connected to each
other by a universal coupling. The first motor worked in the motor
mode, while the second one worked in the generator mode. The ma-
chines were mounted with aluminium brackets and the natural rubber
distances to the aluminium base. The distances were used to reduce
heat conduction towards the base. The test rig was covered by an ac-
rylic glass cover. The cover was applied to reduce velocity fluctuations
caused by the laboratory space. The cover was a square-shaped channel
open from the bottom and closed at the top. In addition, an 8 mm hole
was drilled through the top wall for natural air flow around both mo-
tors. The above-mentioned elements were the main parts of the test rig
and all of them were taken into consideration for the formulated CFD
model. The picture of the test rig is presented in Fig. 1 with schema-
tically marked boundary conditions set in the model. The motor con-
struction with measuring probes positions were schematically pre-
sented in Fig. 2.
The remaining elements used in the test rig but not included in the
model were the main power supply unit, an electronic commutator
consisting of mosfet transistors, a control system for the PM BLDC
motor, and system of resistors used to dissipate electric power produced
B. Melka et al.
Fig. 1. Test rig with schematically marked boundary conditions.
by the generator. These devices were connected by wires and con-
nectors.
2.2. Measuring devices
Temperature was measured using 22 calibrated T-type thermo-
couples. The thermocouple calibration was performed using a Fluke
9100S unit with an accuracy of ± 0.25 °C. The first six thermocouples
were fixed in the region of the motor windings. Thermocouple #T7 was
mounted between the stator magnetic core and the aluminium motor
housing. Thermocouples from #T8 to #T11 were fixed within the rear
portion of the motor. The positions of thermocouples mounted inside
the motor are schematically presented in Fig. 2 as dots with prefixes T
Fig. 2. Simplified scheme of the studied motor with thermocouple and anemometer positions.
408
International Journal of Thermal Sciences 130 (2018) 406–415
and numbers coloured red. Thermocouples from #T12 to #T17 were
mounted on the external part of the motor housing. Thermocouples
#T18 - #T21 were mounted above the analysed motors at heights of 15,
25, 38 and 48 cm from the housing wall, respectively.
The vertical component of the velocity vector outside the motor was
measured by seven constant temperature anemometers at four height
levels. The positions of these sensors are presented in Fig. 2 as green
square markers with prefixes A and numbers in green colour. The
horizontal distance between anemometers was approximately 0.04 m,
while vertical distance was at the level of 0.047 m.
The velocities within the motor housing in its rear part were mea-
sured using a LDA system from Dantec Dynamics. The LDA system was
used to measure velocity components in two directions: horizontal and
vertical. The velocity measurements within the housing were performed
in two sections.
The first section was on the height of shaft axis. This section con-
tained the distance from the housing wall to the shaft wall. Its height is
marked by a blue triangle with the label of LDA1 in Fig. 2. The second
section, where the velocity measurements were also conducted, in-
cluded the distance from the wall to the centre of the motor. The height
of this section is marked by a blue triangle with the label of LDA2 in
Fig. 2. The distance between each recorded point was 1 mm.
The electric measurements were carried out using a Sanwa PC5000a
digital multimeter, Chauvin Arnoux E3N current probes and a
Tektronix MSO 3014 oscilloscope.
2.3. Measurement procedure and motor losses calculation
The temperature, velocity and electrical measurements were exe-
cuted for three rotational speeds at four different loads. The load was
modified by changing a connection of four resistor sets located on the
generator side. This allowed the examination of 11 different operational
states presented in Table 1. In the table, the calculated losses are also
presented. The copper losses were calculated according to the multi-
plication of square current in A and the winding resistance in Ω. The
resistance of the motor windings was estimated using the DC mea-
surement technique. The windings resistance was determined each time
when the steady state conditions were obtained in the experimental
test. The resulting volumetric heat sources were obtained based on the
Joule power losses equation (see Eq. (12)). In this way, the tempera-
ture-dependent resistance and the resulting power losses were
B. Melka et al. International Journal of Thermal Sciences 130 (2018) 406–415

Table 1 presented in Fig. 3 on a cross-section built on the surface rotated at an

Operational points tested on the test rig. angle of 45° in the domain centre. It is presented in the figure that most
Var. No Res. of the Rot. Av. Copper Iron Mech. of the motor components were discretised using structured HEX ele-
generator speed, Current losses, W losses, losses, ments. The same type of the grid was employed in the gap between the
circuit, Ω RPM per phase, W W stator and the rotor. Due to complex and irregular shape of the front
A and rear air region, the TET element mesh was used in those zones.
I 0.25 1415.1 3.41 10.22 3.98 0.88
Outside the motor housing, in the machine vicinity the TET mesh was
II 0.25 3489.0 14.37 43.10 13.22 2.94 also used, while in the further region the HEX elements dominated
III 0.33 1415.1 2.31 6.92 3.98 0.88 (region not presented in Fig. 3). In the meshing strategy, the size of
IV 0.33 2500.5 6.25 18.74 8.33 1.85 elements was gradually increased. The worst element characterised by
V 0.33 3571.5 10.53 31.59 13.67 3.04
the smallest orthogonal quality was on the level of 0.07 and was located
VI 0.5 1415.1 1.00 2.99 3.98 0.88
VII 0.5 2500.5 3.45 10.35 8.33 1.85 in some cell in the air outside the machine in the TET zone. Orthogonal
VIII 0.5 3571.5 6.32 18.95 13.67 3.04 quality of the rest of elements was higher than 0.2 and in the most
IX 1 1388.9 0.46 1.39 3.89 0.86 regions did not fell below 0.5. The minimum value of the Y+ parameter
X 1 2500.5 1.19 3.58 8.33 1.85
in the numerical domain was at the level of 0.09 located within motor
XI 1 3489.0 2.26 6.78 13.22 2.94
in its rear part and the maximum value of Y+ reached 14.6 located in
the air outside the machine.
estimated. In Fig. 3, the global coordinate system is presented. The gravita-
The core losses, also known as the iron losses, were calculated from tional body force was set in the opposite direction of the x-axis of this
the characteristics collected during the idle state measurements. They system.
were dependent on the rotational speed. The mechanical losses were
first included in the characteristics. However, in the next step, they 3.1. Governing equations
were separated. The mechanical losses were calculated using metho-
dology presented by the bearing manufacturer. All the measurements The CFD model was built on the basis of the governing equations
were performed until a thermal steady state had been reached. describing the mass, momentum and energy conservation principles
that were expressed by Eq. (1), through Eq. (5). A motion of the motor
rotor was modelled in accordance with the Multiple Reference Frame
3. Numerical model approach [35].
∂ρ
To simulate the phenomena occurring during the experiment, which + ∇⋅(ρ→
v)=0
∂t (1)
was executed on the test rig, a numerical model was proposed. The
numerical mesh generated within the computational domain included where t is the time, ρ is the density defined as a function of temperature
more than 8 million elements. The visualisation of the mesh is in the fluid zone and therefore air is treated as ideal gas and → v is the

Fig. 3. Mesh displayed in the centred vertical cross-section.

409
B. Melka et al. International Journal of Thermal Sciences 130 (2018) 406–415

velocity vector. coefficient, n is the refractive index, σs is the scattering coefficient, σ is

∂ → the Stefan-Boltzmann constant of 5.669⋅10−8 W⋅m−2⋅K−4 , Φ is the phase
(ρ v ) + ∇⋅(ρ→→ →
v v ) = −∇p + ∇ (τ ) + ρg function and Ω′ is the solid angle.
∂t (2)
where p is the pressure, τ is the stress tensor that can be calculated from 3.2. Source terms
→
Eq. (3) and g is the gravitational acceleration vector.
2 The power losses in the motor were implemented as a homogenous
τ = μ ⎡ (∇→
v + ∇→
v ) − ∇⋅→
T
v ⎤ volumetric heat source in the case of copper and iron losses. These heat
⎣ 3 ⎦ (3)
sources were set in the volume of windings, laminated core and bear-
where μ is the dynamic viscosity and  is the unit tensor. ings. The heat source values were set individually for the load and the
∂ specific speed. The heat sources were calculated on the basis of the
(ρh) + ∇⋅(→
v ρh) = ∇⋅(kij ∇T ) + Sh
∂t (4) electrical measurements. The heat sources in the winding were esti-
mated as the copper losses from Eq. (12).
where h is the specific enthalpy, kij is the thermal conductivity in the
specific direction due to the anisotropic conductivity of the windings I 2⋅R (T )
Sh =
and core and Sh is the volumetric heat source. The anisotropic thermal VCu (12)
conductivity and the heat sources are described in the following sec-
where I is the electric current, R (T ) is temperature-dependent re-
tions.
sistance and VCu is the windings volume.
∂ In the model, the mechanical losses were set on the external bearing
(ρE ) + ∇⋅(→
v (ρE + p)) = ∇⋅(keff ∇T + (τeff ⋅→
v ))
∂t (5) surfaces where balls were in contact with bearing walls.
where the total energy E can be calculated using Eq. (6). Moreover, in The estimated copper, iron and mechanical loss values are presented
the former equation, keff is a sum of k and kt , where k is the laminar for every studied load in Table 1.
thermal conductivity and kt is the turbulent thermal conductivity.
Analogically, the stress tensor τeff is calculated. 3.3. Material properties

p →
v
2
In the steady state, which was assumed in the model, only thermal
E=h− +
ρ 2 (6) conductivity influences the results obtained. In the electric motor, most
of the generated heat is released in the materials which have aniso-
Moreover, the standard k-ε model for turbulence was employed in
tropic thermal conductivity. It directly influences the temperature field.
the solver in a form expressed by Eqs. (7) and (8).
Therefore, anisotropic thermal conductivity was set for the windings
∂ ∂ ∂ ⎡⎛ μ ∂k and laminated core. The following thermal conductivities were as-
(ρktur ) + (ρktur →
vi ) = ⎜ μ + t ⎞ tur ⎤ + Gk + Gb − ρε − YM
⎟
sumed in the model:
∂t ∂x i ∂x j ⎢
⎣⎝ σ ⎥
k ⎠ ∂x j ⎦

+ Sk (7)
• windings in the axial direction (along the wires) 167.3 W⋅m ⋅K −1 −1

[36];
• windings in the radial direction 1.4 W⋅m ⋅K [36];
∂ ∂ ∂ ⎡⎛ μ ∂ε ⎤ ε
(ρε ) + (ρε→
vi ) = ⎜ μ + t⎞ ⎟ + C1ε (Gk + C3ε Gb) −1 −1
∂x j ⎢ ⎥
∂t ∂x i ⎣ ⎝ σε ⎠ ∂x j ⎦ k tur
• the laminated core in the radial direction 22.2 W⋅m ⋅K [36]; −1 −1

− C2ε ρ
ε2
+ Sε • the laminated core in the axial direction 4.9 W⋅m ⋅K [36];−1 −1

ktur (8) • aluminium housing (the most popular aluminium alloy 6061) 167
−1 −1
W⋅m ⋅K ;
where ktur is the turbulence kinetic energy, ε is the turbulent dissipation
rate, Gk is the generation of turbulence kinetic energy due to the mean • electronic plate in a form of the Printed Circuit Board (PCB) 0.55
W⋅m−1⋅K−1 [37].
velocity gradients, Gb is the generation of turbulence energy due to
buoyancy, Ym is the contribution of the fluctuating dilatation in com-
The air properties were defined as the temperature-dependent
pressible turbulence to the overall dissipation rate, C1ε , C2ε , Cμ , σk , σε are
properties including the density that was defined according to the in-
the model constants of 1.44, 1.92, 0.09, 1.0, 1.3, respectively. The
compressible ideal gas equation.
turbulent viscosity μt was calculated using Eq. (9). The constant C3ε was
Thermal contact resistance was added in the crucial parts where it
calculated by Eq. (10).
could play an important role. The thermal contact resistance was de-
k2 fined at the interface between windings and plastic layer which covered
μt = ρCμ
ε (9) the iron core. In this case, the thickness of the equivalent material was
→ of 8.5e-5 m. The thermal conductivity of the equivalent contact re-
v sistance material was assumed on the level of 0.0242 W⋅m−1⋅K−1 and its
C3ε = tanh →
v⊥ (10) value was close to the air properties.
→ →
where v and v⊥ are components of the velocity vector parallel and
3.4. Boundary conditions
perpendicular to the gravity direction, respectively.
The heat transfer by radiation was modelled using Discrete
To solve the governing equations presented in Section 3.1, the
Ordinates (DO) [35]. In the presented model, the angle division was set
thermal and flow boundary conditions were defined on the domain
to 4 for each octant of space, while the pixilation of the control angle
boundaries. The constant-pressure inlet located at the bottom surfaces
was defined as 3. The DO model can be expressed by Eq. (11).
under the plastic cover, where the air could freely be exchanged with
σT 4 the laboratory room, was set equal to 0 Pa. This boundary condition
∇⋅(IR (→
r,→
s )→
s ) + (a + σs ) IR (→
r,→
s ) = an2
π was schematically marked by blue arrows in Fig. 1. The constant-
σ 4π
+ ∫ IR (→
r,→
s ′)Φ(→
s,→
s ′) dΩ′ pressure outlet was set on the top wall where an 8 mm hole was drilled.
4π 0 (11)
Pressure outlet was presented in Fig. 1 by orange arrow. On the vertical
→
where IR is the radiation intensity depending on the position r and and top walls, the ambient air temperature and local heat transfer
direction →
s vectors. In the former equation, a is the absorption coefficient (HTC) were calculated based on the empirical equations for

410
B. Melka et al.
vertical and horizontal plates in natural convection [18]. One of the
walls, where this boundary condition was set, was marked by green
lines in Fig. 1. The average HTC value for the horizontal external walls
of the plastic cover was calculated and reached approximately 3.4
W⋅m−2⋅K−1 and 2.62 W⋅m−1⋅K−1 for vertical walls in Variant V. The
emissivity of the plastic cover was set equal to 0.95. The rotational
speed was assigned to walls which were moving according to the data
listed in Table 1.
The internal parts of the motor working as the generator (the ma-
chine on the right-hand side in Fig. 2) were not included in the model to
reduce the number of elements in the computational domain. The
measured temperature collected from the generator housing were used
to build the temperature profile and implemented in the model as a
boundary condition.
4. CFD results and validation
The following results are presented for the temperature and velocity
fields. First, the temperature field in a vertical cross-section in the
centre of the domain is illustrated. Next, a comparison of measurements
and CFD results at the specific points inside and outside the motor is
discussed. Then, the velocity measurements and CFD results above the
studied machines are presented. Finally, the velocity measurements
collected inside the motor are compared with the CFD results calculated
for some turbulence models. The following results are presented for
Variants IV and V, but the calculations were executed for all the var-
iants described in Table 1. The two variants, presented in the results
section, vary between each other by the rotational velocity and the
losses level.
4.1. Temperature field inside and outside the motor housing
The temperature field of the load described as Variant V is presented
in Fig. 4 in the middle vertical cross-section of the computational do-
main. In the figure, the motor is on the left-hand side, and the generator
is on the right-hand side.
The highest temperature, approximately 78 °C, occurred in the
windings region. The presented cross-section does not cut the model
directly through the windings. The external shape of the machines are
visible in the presented temperature field because the surrounding air
temperature was notably lower. Moreover, it is observed that the air
Fig. 4. Temperature field in °C displayed in the domain cross-section through
the motor axis at the load of 3 resistors and the rotational speed 3500 rpm
(Variant V).
411
International Journal of Thermal Sciences 130 (2018) 406–415
Fig. 5. The temperature comparison between the CFD and experimental results
at the load of 3 resistors and the rotational speed of 3500 rpm - Variant V.
temperature in this view was higher directly above the end caps of both
machines. The highest air temperature outside the motor occurred
above the back-end cap of the generator. The remaining temperature
field located 0.1 m above the machines was practically uniform and was
present at half of the domain height.
A comparison of the temperature fields obtained using CFD and
experimental measurements is presented in Fig. 5. The results collected
at the specific locations are presented in four groups, which are sepa-
rated by the dotted green lines. Those locations denoted as red coloured
points are schematically shown in Fig. 2:
• inside at the front part of the motor #T1 -#T7;
• inside motor and in the rear part of the motor #T8 -#T11;
• on the external part of the motor housing #T12-#T19;
• air above the motor #T20 -#T24.
In that figure, the thermocouples fixed on the windings recorded the
highest temperature values between 80 and 85 °C. The last thermo-
couple in the first group shows a value of 71.4 °C. The CFD results
present more uniform temperature distribution in the first group of
thermocouples and reach approximately 84 °C. Thermocouples #T8 and
#T9 fixed in the back part of the motor show an internal air tem-
perature of 68 and 70 °C, respectively. One of largest differences be-
tween measurements and CFD results occurs in those points. In parti-
cular, the CFD results are underestimated by approximately 5 K. Point
#T10 fixed inside the motor on the housing reached 65 °C. The last
measured point #T11 located on the PCB recorded a temperature of
84 °C. This temperature was similar to the temperature occurring on the
windings. It was caused by the thermocouple directly fixed where one
of the phase electric paths was located. The current flows through that
path caused a local heat generation. The next seven points, where the
temperature was measured, were attached to the external surface of the
housing. The temperature for these points was approximately 65 °C.
Slightly higher temperatures were observed in the middle of the motor
length. The maximum difference between these temperatures was 1 K.
Hence, temperature distribution was uniform on the housing walls
mainly due to the high thermal conductivity. The last four temperatures
were measured for different heights of the air above the considered
motor. The temperature of these points was at a level of 35 °C. The
differences between the measurements and CFD results at these points
are in the range of 1 K.
The average heat transfer coefficient obtained on the basis of the
average convective heat flux, the outer casing wall temperature and the
B. Melka et al.
Fig. 6. Vertical velocity component comparison between CFD and experimental
results at the load of 3 resistors and a rotational speed of 3500 rpm (Variant V).
average air temperature was in the range of 4–6 W⋅m−2⋅K−1. These
values are consistent with the heat transfer coefficient obtained from
the Nusselt number correlation for the natural convection flow around a
cylindrical shape [18].
4.2. Velocity field outside the motor housing
A comparison of the velocity results from CFD and measurements,
collected outside the motor, is presented in Fig. 6. The chart describes
the vertical component of the velocity vector in 28 points measured by
7 anemometers at 4 height levels above both the motor and generator.
The values were grouped according to these levels with the group de-
scription above the chart. The sensor numbering at the bottom of the
chart is consistent with the numbers with prefixes A in Fig. 2. The error
bars were added to the measured values based on the standard devia-
tion analysis. As shown in this figure, the error bars are wide for many
points. This can be explained by the natural convection flow that
caused high instability of the measurement. In the case of the external
air flow, that was the main mechanism of the heat dissipation from the
motor walls.
According to Fig. 6, in the first group on the lowest presented level,
the CFD model results were in the range of the estimated velocity errors
for 5 out of 7 measured points. In the next group, there were only 3
values within the error ranges. In the third group, 4 of the measured
points were consistent with the modelled values. Finally, the highest
level contained 6 points in which the CFD results were in the range of
the standard deviations calculated from the measurements. The average
Fig. 7. Horizontal component of the velocity vector within motor on a height of the shaft axis (LDA1) for Variant No. V (left) and Variant IV (right).
412
International Journal of Thermal Sciences 130 (2018) 406–415
value of the vertical velocity component from all the measured points
was 0.089 m⋅s−1, while the average of these values from the thermal
model was 0.131 m⋅s−1. In the figure, most of the CFD results are in the
error bar ranges estimated from the measurements. Considering that
calculations were conducted in a steady state mode, the agreement
between model and measurement can be treated as satisfactory.
4.3. Velocity field inside the motor housing
A reduced model was created to validate the velocity calculation.
This model was limited to the domain of air in the back part of the
motor, where the LDA apparatus was used to record velocity compo-
nents as described in Section 2.2. For this model, three numerical me-
shes were tested during the calculation process that contained between
0.7 and 3 million elements. The temperature profiles were implemented
for all investigated loads from the full model described in the previous
sections.
First, a comparison was made between the horizontal component of
the velocity vector for the CFD results and the measurements in section
LDA1 as shown in Fig. 7. The first line marked as continuous in red
colour represents a position where the LDA measurements were per-
formed. In this figure, the second line represents the velocity in the
section moved 5 mm to the front of the motor. A comparison of the two
positions was used to visualise the velocity field in the analysed region.
The results of this configuration are presented for Variants IV and V. For
both presented loads, the velocity obtained by the model and the
measurements is negative in the region near the wall. The first value
was measured at 0.001 m from the wall. That means the horizontal
velocity component is directed to the front of the motor in this region.
According to the measurements, the direction of the horizontal velocity
component changes at approximately 1 cm from the wall. The same
tendency can be observed in the model results. Moreover, a rapid in-
crease of the horizontal velocity in the direction of the motor front at a
distance of approximately 1 cm from the wall can be observed. The
experimental results show a sudden decrease in the direct shaft vicinity.
At this location, the CFD results reveal the changes between the two
presented lines. The second line moved 5 mm to the front of the motor
shows better agreement with the measured values than the first line
recorded in the middle of the back part of the motor.
In Fig. 8, the vertical component of velocity on the height of the
shaft axis (LDA1) is compared between results from the experiment and
the model for Variants IV and V. In this section, the vertical velocity
increases slightly from wall vicinity to shaft direction approaching the
end of the recorded range. In the distance behind recorded range from
measurements, i.e., more than 0.022 m, the values derived from the
CFD model increased rapidly to achieve the linear velocity of the shaft.
The values in the measured range are similar between experiment and
CFD results. The measured values of velocity were limited by the re-
solution (of 1 mm) in which each measured point could be tested. The
B. Melka et al.
Fig. 8. Vertical component of the velocity vector within the motor on a height of the shaft axis (LDA1) for Variant V (left) and Variant IV (right).
highest expected vertical velocity should be reached in the wall vicinity
as it was observed in the CFD results. Therefore, it seems that the values
recorded during the experiment in those locations are probably wrong
(the last non-zero measured value). The LDA device could reflect the
laser beam or did not cross laser beam in the region where the seed
(smoke) occurred. Therefore, the maximum velocity value could occur
in the distance between the measured points. In Fig. 8, the two lines
obtained from the CFD model were also presented. The differences
between the lines are negligible for the vertical velocity component
along the height of the shaft axis.
In Fig. 9, the horizontal component of the velocity vector is pre-
sented on the line directly above the shaft (LDA2). Similar to the pre-
vious figures, the CFD results were presented for the two mapped lines.
In the figure, the CFD results are consistent with the measured values
for the entire range of investigated section. Firstly, near the wall, the
horizontal velocity was directed in the front of the motor with negative
values. The maximum value in this direction reached approximately 0.1
m⋅s−1. Similar to the previously described section, the horizontal velo-
city changed its direction at a distance of approximately 0.01 m towards
the back of the motor (positive values). From this point, the measured
values began to gradually increase. The CFD results for load Variant V
were more in agreement with the measured range than in Variant IV,
especially in the part with negative values of the horizontal velocity.
Nevertheless, the CFD results of the horizontal velocity component
reasonably reflected the experiment in section LDA2 to a distance of
0.02 m from the wall. The first line of the CFD results still increased but
as quickly as the measured values. In this case, the second line behind
the distance of 0.02 m from the wall began decreasing and reached
approximately 0 m⋅s−1 in the shaft vicinity.
In Fig. 10, the vertical component of velocity on the line above the
shaft (LDA2) is presented. For Variant V, the measured velocity values
increased from 0 m⋅s−1 in the vicinity of the wall to approximately 0.2
m⋅s−1 at a distance of 0.02 m from the wall. The CFD results showed an
Fig. 9. Horizontal component of the velocity vector within the motor on a height above the shaft axis (LDA2) for load Variant V (left) and Variant IV (right).
413
International Journal of Thermal Sciences 130 (2018) 406–415
increasing tendency in the same section, but the values are approxi-
mately two times greater than the measured values. At distances farther
than 0.02 m from the wall, this velocity increased again achieving ap-
proximately 0.25 m⋅s−1 . In this region, the velocity from the experi-
ment and the model are similar. In the vicinity close to the shaft, the
vertical velocity component significantly decreases in the CFD results.
For Variant IV, the behaviour of this velocity component is similar to
that described with the higher load. Nevertheless, the values from the
CFD results are approximately three times greater than the values from
the experiments up to a distance of 0.02 m from the wall. This differ-
ence decreased by less than twice. In Fig. 10, the second line obtained
from the CFD results shows slightly smaller values than those from the
first line and values more similar to the experimental measurements.
5. Conclusions
In this paper, a validated numerical model of heat and flow phe-
nomena in a PM BLDC motor was presented. Experiments were per-
formed for 11 loadings at a thermal steady state. The experiments in-
cluded the following: temperature measurements with thermocouples
inside and outside the motor, velocity measurements above the motor
and generator with constant temperature anemometers, velocity mea-
surements inside the rear part of the motor using Laser Doppler
Anemometry, and electrical measurements by current probes, a digital
multimeter and oscilloscope.
The numerical model presented in this paper was based on the
complex geometry, which realistically duplicated the real machine. The
compared points where the temperatures were measured showed rea-
sonable agreement with the experimental data. A very high accuracy of
the temperature prediction was obtained for the ambient air and the
motor housing. In those regions, the maximum discrepancy did not
exceed 3 K. Inside the housing, the numerical and experimental tem-
peratures were also accurately simulated. One of the main reasons was
B. Melka et al.
Fig. 10. Vertical component of the velocity vector within the motor on a height above the shaft axis (LDA2) for Variant V (left) and Variant IV (right).
the temperature-dependence of the winding power losses that were
based on the experimental values for the electric resistance. The largest
difference between the computed results and measured results as re-
gistered in the region of the stator core was approximately 9 K. The
reason for the mentioned discrepancies could be caused by the assumed
values of the anisotropic thermal conductivities taken from the litera-
ture. The second reason of the result difference could be caused by the
temperature-dependent, but still homogeneous power losses defined for
the windings and the core. In the further research, the two-way coupled
model of thermal and electromagnetic processes is planned.
The validation of the velocity results inside the motor was one of the
novelties presented in the paper. The agreement of the velocity field
above the machines was also reasonable because the computed values
were in the error bar ranges for the majority of the compared points.
The velocity results obtained with the CFD model inside the motor
showed a similar tendency as in the experiment. Those observations are
especially justified for the horizontal component of the velocity vector.
In that case, the predicted values were close to those found in the ex-
perimental tests along the whole path from the wall to the shaft. In the
case of the vertical component of the velocity vector, the numerical
results were very close to the measured values in the central region of
the investigated region. In the wall and the shaft region, higher dis-
crepancies were noted.
Nevertheless, the CFD results from the second additional line prove
that the velocity field was more distorted in the tested machine than
that obtained from the CFD model.
Acknowledgements
Financial assistance was provided by grant no. DEC-2011/03/D/
ST8/04171 funded by the National Science Centre, Poland. The work of
BM was also partially supported by statutory research funds of the
Faculty of Energy and Environmental Engineering of the Silesian
University of Technology within grant no. BKM-559/RIE6/2016.
References
[1] C. Thiel, W. Nijs, S. Simoes, J. Schmidt, A. van Zyl, E. Schmid, The impact of the EU
car CO2 regulation on the energy system and the role of electro-mobility to achieve
transport decarbonisation, Energy Pol 96 (2016) 153–166, http://dx.doi.org/10.
1016/j.enpol.2016.05.043.
[2] J.H. Langbroek, J.P. Franklin, Y.O. Susilo, The effect of policy incentives on electric
vehicle adoption, Energy Pol 94 (2016) 94–103, http://dx.doi.org/10.1016/j.enpol.
2016.03.050 http://www.sciencedirect.com/science/article/pii/
S0301421516301550.
[3] K.G. Gallagher, S. Goebel, T. Greszler, M. Mathias, W. Oelerich, D. Eroglu, et al.,
Quantifying the promise of lithium - air batteries for electric vehicles, Energy
Environ Sci 7 (2014) 1555, http://dx.doi.org/10.1039/c3ee43870h.
[4] L. Chen, L.L. Shaw, Recent advances in lithium-sulfur batteries, J Power Sources
267 (2014) 770–783, http://dx.doi.org/10.1016/j.jpowsour.2014.05.111 https://
doi.org/10.1016/j.jpowsour.2014.05.111.
[5] S.M. Rezvanizaniani, Z. Liu, Y. Chen, J. Lee, Review and recent advances in battery
health monitoring and prognostics technologies for electric vehicle (EV) safety and
414
International Journal of Thermal Sciences 130 (2018) 406–415
mobility, J Power Sources 256 (2014) 110–124, http://dx.doi.org/10.1016/j.
jpowsour.2014.01.085.
[6] T.-F. Yi, S.-Y. Yang, Y. Xie, Recent advances of Li 4 Ti 5 O 12 as a promising next
generation anode material for high power lithium-ion batteries, J Mater Chem A 3
(11) (2015) 5750–5777, http://dx.doi.org/10.1039/C4TA06882C http://pubs.rsc.
org/en/content/articlehtml/2015/ta/c4ta06882c.
[7] H. Lund, W. Kempton, Integration of renewable energy into the transport and
electricity sectors through V2G, Energy Pol 36 (9) (2008) 3578–3587, http://dx.
doi.org/10.1016/j.enpol.2008.06.007.
[8] Lopes JAP, Soares FJ, Almeida PMR. Integration of electric vehicles in the electric
power system. Proc IEEE 99(1). https://doi.org/10.1109/JPROC.2010.2066250.
[9] X.D. Xue, K.W.E. Cheng, N.C. Cheung, Selection of electric motor drives for electric
vehicles, Power engineering conference, 2008. AUPEC '08. Australasian
Universities, 2008, pp. 1–6.
[10] K. Hameyer, R.J.M. Belmans, Permanent magnet excited brushed DC motors, IEEE
Trans Ind Electron 43 (2) (1996) 247–255, http://dx.doi.org/10.1109/41.491348.
[11] K.T. Chau, C.C. Chan, C. Liu, Overview of permanent-magnet brushless drives for
electric and hybrid electric vehicles, IEEE Trans Ind Electron 55 (6) (2008)
2246–2257, http://dx.doi.org/10.1109/TIE.2008.918403.
[12] E. a. Lomonova, E. Kazmin, Y. Tang, J.J. Paulides, In-wheel PM motor: compromise
between high power density and extended speed capability, COMPEL Int J Comput
Math Electr Electron Eng 30 (1) (2011) 98–116, http://dx.doi.org/10.1108/
03321641111091467.
[13] D. Staton, A. Boglietti, A. Cavagnino, Solving the more difficult aspects of electric
motor thermal analysis in small and medium size industrial induction motors, IEEE
Trans Energy Convers 20 (2005) 620–628, http://dx.doi.org/10.1109/TEC.2005.
847979.
[14] N. Putra, B. Ariantara, Electric motor thermal management system using L-shaped
flat heat pipes, Appl Therm Eng 126 (2017) 1156–1163, http://dx.doi.org/10.
1016/j.applthermaleng.2017.01.090 https://ac.els-cdn.com/
S1359431117305458/1-s2.0-S1359431117305458-main.pdf?{\_}tid=ff89478c-
a52d-11e7-8763-00000aacb35f{\&}acdnat=1506700444{\_}
dfefeb721193f0207dc927459f32d63f.
[15] R. Khlissa, S. Vivier, G. Friedrich, K. El Kadri Benkara, B. Assaad, Thermal modeling
of an asymmetrical totally enclosed permanent magnet integrated starter generator,
Math Comput Simulat 130 (2016) 32–47, http://dx.doi.org/10.1016/j.matcom.
2015.06.015 https://ac.els-cdn.com/S0378475416000094/1-s2.0-
S0378475416000094-main.pdf?{\_}tid=cc757612-a9e3-11e7-a144-
00000aab0f27{\&}acdnat=1507218332{\_}
b8d5be2a4402cae3fe49bd2ab47c545cΩhttp://linkinghub.elsevier.com/retrieve/
pii/S0378475416000094.
[16] P. Mellor, D. Roberts, D. Turner, Lumped parameter thermal model for electrical
machines of TEFC design, IEEE Proc B Elec Power Appl 138 (5) (1991) 205, http://
dx.doi.org/10.1049/ip-b.1991.0025.
[17] A. Boglietti, A. Cavagnino, D. Staton, M. Shanel, M. Mueller, C. Mejuto, Evolution
and modern approaches for thermal analysis of electrical machines, IEEE Trans Ind
Electron 56 (2009) 871–882, http://dx.doi.org/10.1109/TIE.2008.2011622.
[18] Y.A. Cengel, Heat transfer: a practical approach, 2nd ed., McGraw-Hill, London,
2007.
[19] A. Boglietti, E. Carpaneto, M. Cossale, S. Vaschetto, Stator-winding thermal models
for short-time thermal transients: definition and validation, IEEE Trans Ind Electron
63 (5) (2016) 2713–2721, http://dx.doi.org/10.1109/TIE.2015.2511170.
[20] Y.-Z. Li, S.-J. Zhu, Y. Li, Q. Lu, Temperature prediction and thermal boundary si-
mulation using hardware-in-loop method for permanent magnet synchronous mo-
tors, IEEE/ASME Trans Mechatron 21 (1) (2015), http://dx.doi.org/10.1109/
TMECH.2015.2443800 1–1 http://ieeexplore.ieee.org/document/7121001/.
[21] B. Melka, J. Smolka, Z. Bulinski, J. Hetmanczyk, An experimental and numerical
analysis of temperature and velocity field in PM BLDC motor, Przeglad
Elektrotechniczny 1 (3) (2016) 94–97, http://dx.doi.org/10.15199/48.2016.03.22
http://sigma-not.pl/publikacja-97014-2016-3.html.
[22] O. Meksi, A.O. Vargas, Numerical and experimental determination of external heat
transfer coefficient in small TENV electric machines, 2015 IEEE Energy Conversion
Congress and Exposition, ECCE 2015(3) (2015), pp. 2742–2749, http://dx.doi.org/
10.1109/ECCE.2015.7310044.
[23] M. Fasil, D. Plesner, J.H. Walther, N. Mijatovic, J. Holbøll, B.B. Jensen, Numerical
B. Melka et al.
and experimental investigation of heat flow in permanent magnet brushless DC hub
motor, SAE Int J Alternative Powertrains 4 (1) (2014), http://dx.doi.org/10.4271/
2014-01-2900 2014–01–2900 http://papers.sae.org/2014-01-2900/.
[24] D. Staton, E. So, Determination of optimal thermal parameters for brushless per-
manent\nmagnet motor design, Conference record of 1998 IEEE industry applica-
tions conference. Thirty-third IAS annual meeting, vol. 1, 1998, pp. 41–49.
[25] M. Grabowski, K. Urbaniec, J. Wernik, K.J. Wołosz, Numerical simulation and ex-
perimental verification of heat transfer from a finned housing of an electric motor,
Energy Convers Manag 125 (2016) 91–96, http://dx.doi.org/10.1016/j.enconman.
2016.05.038.
[26] Y. Zhang, J. Ruan, T. Huang, X. Yang, H. Zhu, G. Yang, Calculation of temperature
rise in air-cooled induction motors through 3-d coupled electromagnetic fluid-dy-
namical and thermal finite-element analysis, IEEE Trans Magn 48 (2) (2012)
1047–1050, http://dx.doi.org/10.1109/TMAG.2011.2174433.
[27] C. Jungreuthmayer, T. Bauml, O. Winter, M. Ganchev, H. Kapeller, a. Haumer,
et al., A detailed heat and fluid flow analysis of an internal permanent magnet
synchronous machine by means of computational fluid dynamics, IEEE Trans Ind
Electron 59 (12) (2012) 4568–4578, http://dx.doi.org/10.1109/TIE.2011.
2176696.
[28] T. Davin, J. Pell, E.,S. Harmand, R. Yu, Experimental study of oil cooling systems for
electric motors, Appl Therm Eng (2015), http://dx.doi.org/10.1016/j.
applthermaleng.2014.10.060 https://ac.els-cdn.com/S1359431114009314/1-s2.0-
S1359431114009314-main.pdf?{\_}tid=ee8b7ad4-a52f-11e7-89dd-
00000aab0f27{\&}acdnat=1506701274{\_}
248c775d636751144956e0823059cc8b.
[29] D. Hyun Lim, S. Chul Kim, Thermal performance of oil spray cooling system for in-
wheel motor in electric vehicles, Appl Therm Eng (2014), http://dx.doi.org/10.
1016/j.applthermaleng.2013.11.057 https://ac.els-cdn.com/
S1359431113008612/1-s2.0-S1359431113008612-main.pdf?{\_}tid=f7089b5c-
a52e-11e7-ab8b-00000aacb35d{\&}acdnat=1506700859{\_}
7e66e286db5bc5ee982e3cb4843b11cb.
[30] S. Wang, Y. Li, Y.-Z. Li, J. Wang, X. Xiao, W. Guo, Transient cooling effect analyses
415
International Journal of Thermal Sciences 130 (2018) 406–415
for a permanent-magnet synchronous motor with phase-change-material packaging,
Appl Therm Eng (2016), http://dx.doi.org/10.1016/j.applthermaleng.2016.08.036
https://ac.els-cdn.com/S1359431116313825/1-s2.0-S135431116313825-main.
pdf?{\_}tid=a7e65828-a530-11e7-8511-00000aacb361{\&}acdnat=
1506701585{\_}14b77c6e1ead2795eb38eed0d755bf1c.
[31] Y. Li, Z. Ji, L. Yang, P. Zhang, X. Bing, J. Zhang, Thermal-fluid-structure coupling
analysis for valve plate friction pair of axial piston pump in electrohydrostatic ac-
tuator (EHA) of aircraft, Appl Math Model 47 (2017) 836–858 https://ac.els-cdn.
com/S0307904X16304395/1-s2.0-S0307904X16304395-main.pdf?{\_}tid=
7334fb28-f84d-11e7-a73d-00000aacb362{\&}acdnat=1515839899{\_}
bbdeeba2a35f55686f6b08ab53c02c5e.
[32] A. Aubert, S. Poncet, P. Le Gal, S. Viazzo, M. Le Bars, Velocity and temperature
measurements in a turbulent water-filled Taylor-Couette-Poiseuille system, Int J
Therm Sci 90 (2015) 238–247, http://dx.doi.org/10.1016/j.ijthermalsci.2014.12.
018 https://doi.org/10.1016/j.ijthermalsci.2014.12.018.
[33] B. Melka, J. Smolka, Z. Bulinski, J. Hetmanczyk, D. Makiela, A validated numerical
model of heat and mass transfer in a PM BLDC electric motor, (2016), pp.
1409–1413, http://dx.doi.org/10.1109/SPEEDAM.2016.7525885 http://
ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7525885.
[34] X. Zhang, J. Yu, G. Su, Z. Yao, P. Hao, F. He, PIV measurement and simulation of
turbulent thermal free convection over a small heat source in a large enclosed
cavity, Build Environ 90 (2015) 105–113, http://dx.doi.org/10.1016/j.buildenv.
2015.03.015 https://doi.org/10.1016/j.buildenv.2015.03.015.
[35] ANSYS® Academic Research, Release 17.0, help system: ANSYS, Inc.
[36] R. Wrobel, P.H. Mellor, A general cuboidal element for three-dimensional thermal
modelling, IEEE Trans Magn 46 (8) (2010) 3197–3200, http://dx.doi.org/10.1109/
TMAG.2010.2043928.
[37] K. Azar, J. Graebner, Experimental determination of thermal conductivity of printed
wiring boards, Twelfth annual IEEE semiconductor thermal measurement and
management symposium. Proceedings, IEEE, 1994, pp. 169–182, , http://dx.doi.
org/10.1109/STHERM.1996.545107 http://ieeexplore.ieee.org/lpdocs/epic03/
wrapper.htm?arnumber=545107.