Scalable Electric-motor-in-the-Loop Testing for Vehicle Powertrains
Thomas D’hondt
1 a
, Yves Mollet
1,2 b
, Arthur Jacques Joos
2
, Leonardo Cecconi
1
,
Mathieu Sarrazin
1
and Johan Gyselinck
2 c
1
Test Division, Siemens Industry Software NV, Researchpark 1237, Interleuvenlaan 68, B-3001 Leuven, Belgium
2
BEAMS Department, Universit
´
e Libre de Bruxelles, Avenue Franklin Roosevelt 50, CP165/52, B-1050 Bruxelles, Belgium
Keywords:
Electric Vehicles, e-Powertrain, Model-Based System Testing, Scaling, Real-time Control, X-in-the-Loop.
Abstract:
Model-Based System Testing (MBST) combines physical testing and simulation models to enable the valida-
tion of complex systems early-on in their design cycle. Therefore, it shows great potential for the validation
of increasingly complex Electric Vehicle (EV) powertrains. In this work, the MBST methodology is applied
to a downscaled powertrain, including a Permanent-Magnet Synchronous Machine (PMSM) and a 3-phase
switch-mode inverter. This System-under-Test (SuT) is integrated into an X-in-the-Loop (XiL) test bench,
where real-time simulation models of the rest of the vehicle are used to impose realistic boundary conditions
to the SuT. These include the emulation of the vehicle inertia, its friction losses and the regenerative braking
controller. Both hardware and software architectures required to achieve this setup are presented. Subse-
quently, a methodology used for computing scaling factors that match the power levels of the full vehicle to
the miniature test bench is proposed. Finally, the combined physical-virtual system is evaluated on a driving
cycle to validate its behaviour. The usage of a downscaled SuT constitutes the first step towards full-scale
E-powertrain-in-the-loop testing, as well as a valuable multi-purpose didactical XiL setup.
1 INTRODUCTION
In Model-Based System Testing (MBST), simulation
models and physical testing are combined to inves-
tigate, improve or validate complex multi-physical
or mechatronic systems (Siemens Digital Industries
Software, 2019). In this framework, X-in-the-Loop
(XiL) testing is more specifically used to test one or
more physical components, while simulating their en-
vironment, such that the physical presence of the sur-
rounding hardware is not required to create realistic
working conditions (Van der Auweraer et al., 2017).
As a consequence, XiL-based prototyping can be per-
formed at early development stages and requires lim-
ited hardware compared to its physical counterpart,
permitting a shorter time-to-market and lower devel-
opment costs (Fathy et al., 2006).
This technique is particularly interesting for re-
search and development in Electric Vehicles (EVs),
especially focusing on drivetrain (Popp et al., 2015;
Petersheim and Brennan, 2009; Yang et al., 2015) and
a
https://orcid.org/0000-0003-2554-332X
b
https://orcid.org/0000-0001-6170-0148
c
https://orcid.org/0000-0003-2259-8560
consumption (Ciceo et al., 2016; Williamson et al.,
2006) to address their present shortages in efficiency,
autonomy and acoustic comfort, in the challenging
context of more and more restrictive ecological and
safety regulations (Fiori et al., 2016; Yang et al.,
2014). In this domain, XiL testing also allows for
high repeatability compared to real driving and may
avoid increased computation time or resources in case
of full-vehicle simulation (Fathy et al., 2006). This is
of paramount interest as more complex powertrains
and increasing amounts of electrical accessories can
result in integration issues if combined late in the de-
sign process. Therefore, the application of the MBST
methodology on EVs is currently being studied in the
OBELICS (Optimization of scalaBle rEaltime mod-
eLs and functIonal testing for e-drive ConceptS) re-
search project (Obelics consortium, 2019).
A surrogate System-under-Test (SuT) can be used
to reduce prototyping costs when the original hard-
ware is not required for the investigations of inter-
est (Petersheim and Brennan, 2009). Adequate scal-
ing factors allow then to match the characteristics of
the SuT to the model. This principle is used in the
present paper to significantly downscale the surrogate
SuT compared to similar setups (Popp et al., 2015;
594
D’hondt, T., Mollet, Y., Joos, A., Cecconi, L., Sarrazin, M. and Gyselinck, J.
Scalable Electric-motor-in-the-Loop Testing for Vehicle Powertrains.
DOI: 10.5220/0009887405940603
In Proceedings of the 17th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2020), pages 594-603
ISBN: 978-989-758-442-8
Copyright
c
2020 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
Real-time simulation model
Test bench
Traction
inverter
Load
inverter
˜
τ
m
e
LoadTraction
Downscale
Upscale
˜
τ
m
l
˜
ω
m
˜
ω
r
˜
τ
r
e
Vehicle
dynamics
Vehicle
Control Unit
τ
m
e
ω
r
τ
r
e
τ
b
ω
m
Driver
Reference
profile
˙x
r
˙x
u
t
u
b
Figure 1: Overview of the test bench layout and of the con-
nected simulation models. The variables with tilde corre-
spond to the reduced-scale test bench.
Petersheim and Brennan, 2009; Ciceo et al., 2016).
This reduction in size does not only reduce material
costs, but also has a two-fold objective: propose a safe
validation step between pure simulations and full-size
test environments, as well as a valuable and trans-
portable didactic / demonstration setup. Indeed, it can
be used to evaluate control strategies on a state-of-
the-art test bench controller and inverter, while stay-
ing in a low-risk environment. The bench comprises
two low-power Permanent-Magnet Synchronous Ma-
chines and their respective inverters. One Permanent-
Magnet Synchronous Machine (PMSM) and its in-
verter constitute the surrogate SuT, receiving a torque
reference from the throttle, whereas the other emu-
late the behaviour of a real car via a speed refer-
ence. It is obtained by a real-time model of the rest of
the car combined with adequate scaling factors (Fig-
ure 1). Furthermore, although this article considers an
EV context, the test bench may be used for XiL tests
where other environments are simulated.
This paper first describes in detail the real-time
simulation models which are used to provide realis-
tic boundary conditions to the SuT and how they are
integrated into the overall controller of the test bench.
The physical layout of the test bench is then investi-
gated and scaling factors are computed to match the
power level in the simulation model to that of the
physical components. Finally, the first experimental
results and conclusions are presented.
2 REAL-TIME SIMULATION
MODELS
The real-time simulation model used to provide real-
istic boundary conditions to the SuT (Figure 1) con-
sists of three major subsystems: (i) the driver, (ii) the
Vehicle Control Unit (VCU) and (iii) the dynamics of
the rest of the vehicle. This last part of the model
includes the behaviour of the car body, gearbox and
brakes. All of them are implemented considering a
BMW i3 as a full-scale reference EV (BMW, 2017).
Their overall implementation in the Simcenter
Amesim software (Figure 2), as well as the corre-
sponding parameters, will be further explained in this
section. The computation of the scaling factors used
to match the characteristics of the reference vehicle
to the physical limitations of the test bench will be
investigated in Section 5.
2.1 Driver
The aim of the driver is to compute the adequate throt-
tle and brake commands to be applied to the vehi-
cle, u
t
and u
b
respectively. This closed-loop con-
troller therefore takes the measured longitudinal ve-
hicle speed ˙x and tries to make it track a pre-defined
velocity profile ˙x
r
(Figure 1). Those profiles can,
for instance, be the New European Driving Cycle
(NEDC) or World harmonised Light vehicles Test Cy-
cle (WLTC) for assessing the energy consumption of
the vehicle. In the remainder of the paper, the super-
scripts m and r will refer to measured and reference
Figure 2: Vehicle simulation model generated in Simcenter
Amesim.
Scalable Electric-motor-in-the-Loop Testing for Vehicle Powertrains
595
˙x
r
˙x
˙x
r
Anticipative
K
a
t
K
a
b
PI throttle
+
u
t
PI brake
+
1
u
b
Figure 3: Driver controller layout.
quantities respectively. Similarly, the subscripts t and
b correspond to the throttle and brake actions. Finally,
tilde variables are referred to the reduced-power test
bench, whereas regular variables correspond to the
full-scale EV.
Two independent Proportional Integral (PI) con-
trollers with an additional anticipative action are used
to compute the vehicle inputs (Figure 3). The antici-
pative action is calculated using an approximated ref-
erence acceleration:
u
a
=
˙x
r
(t + T ) ˙x
r
(t)
T
, (1)
where T is the anticipative time constant. It is then
multiplied by throttle- and brake-specific gains and
summed to the outputs of the regular PI controllers.
Finally, both the throttle and braking action are satu-
rated in the [0,1] interval.
2.2 Vehicle Control Unit
The VCU converts the computed throttle and brake
commands into a reference torque for the E-motor τ
r
e
and for the mechanical brakes τ
b
, taking regenerative
braking into account (Figure 1). Therefore, a user-
requested torque τ
r
is first computed by transforming
the normalised user inputs into a torque value:
τ
r
= u
t
τ
max
e
+ u
b
τ
min
e
τ
max
b
r
, (2)
where τ
max
e
> 0 and τ
min
e
0 are the maximum E-
motor torques in motoring and braking operation. The
maximum mechanical braking torque τ
max
b
> 0 of the
vehicle is referred to the shaft of the E-motor by di-
viding it by the gearbox ratio r. The user-requested
torque τ
r
is then split according to a fixed repartition
strategy (Figure 4). The blue and green areas rep-
resent the contribution of τ
r
e
and τ
b
to the total ve-
hicle torque referred to the E-motor shaft. During
the braking operation, the fixed ratio is held until τ
r
e
reaches the braking torque limit of the E-motor τ
min
e
.
Although not further studied in this paper, the regen-
erative action can also be inhibited at too low or too
high speeds and depending on battery state-of-charge.
2.3 Vehicle Dynamics
The vehicle body is simulated as a point-mass moving
in the longitudinal dimension x. The equivalent mass
of the vehicle includes both the vehicle mass M and
its wheel inertia J
w
:
M + 4
J
w
R
w
2
¨x =
rτ
m
e
τ
b
R
w
F
r
, (3)
with R
w
, τ
m
e
and τ
b
the wheel radius, the measured
and upscaled E-motor torque and the braking torque.
Both torques are transformed into longitudinal forces
considering the fixed and unitless reducer gear ratio r
and the wheel radius R
w
. Finally, the resistive force
exerted by the environment on the car F
r
comprises
both aerodynamic drag and rolling resistance:
F
r
=
ρ
air
C
x
S
2
˙x
2
+ Mg(a + b ˙x), (4)
with ρ
air
the air density, C
x
the air penetration coeffi-
cient and S the vehicle active area. No external wind
is currently considered in the aerodynamic drag term.
The rolling resistance is supposed to be proportional
to the vehicle mass M and the gravity g with a con-
stant term a and a term proportional to the vehicle
speed b ˙x. It should be noted that no slope, nor stic-
tion effects are considered in the present paper.
A friction torque compensation mechanism is in-
cluded in the simulation model due to the non-
negligible friction in the test setup compared to the
nominal torque of the machines. Indeed, the elec-
tromagnetic torque
˜
τ
m
l
estimated by the load inverter
based on current measurements differs from the elec-
tromagnetic torque
˜
τ
m
e
from the SuT motor expected
by the vehicle simulation model. The difference be-
tween both corresponds to the friction of the loading
τ
r
τ
0
0
τ
max
e
τ
min
e
τ
max
b
r
τ
max
e
τ
min
e
τ
max
b
r
τ
min
e
Electrical motoring and
braking torque τ
r
e
Mechanical braking
torque at motor shaft τ
b
/r
Figure 4: Split between regenerative and mechanical
torques during braking manoeuvres.
ICINCO 2020 - 17th International Conference on Informatics in Control, Automation and Robotics
596
4 2 0 2 4
40
20
0
20
40
Rotation speed [krpm]
Friction torque [mNm]
Measured points
Interpolated data
Figure 5: Test bench friction
˜
τ
f
as a function of rotational
speed
˜
ω
m
.
machine
˜
τ
m
e
=
˜
τ
m
l
+
˜
τ
f
, which is identified during pre-
liminary tests at constant speed and approximated us-
ing a first-order relation:
˜
τ
f
(
˜
ω
m
) = sign(
˜
ω
m
)
τ
f ,0
+ K
f
|
˜
ω
m
|
, (5)
where τ
f ,0
and K
f
are the static and dynamic friciton
coefficients. The measured and interpolated data used
in the simulation model are represented in Figure 5.
3 CONSTRUCTION OF THE XiL
TEST BENCH
The downscaled XiL test setup created for this work
consists of three main subsystems: (i) the power cab-
inet, (ii) the real-time computing unit and (iii) the E-
motor testbed. All three are visible in Figure 6.
The power cabinet and the testbed contain two
motor and inverter pairs: the powertrain-under-test
Figure 6: The power supply and inverters are located inside
the electrical power cabinet (1). The real-time controller (2)
is fixed on a frame, which also supports the electric motors
using a small vibration absorber (3). A DAQ is mounted on
the side of the setup (4).
PSU
230 V
Traction
inverter
Load
inverter
Braking
resistor
˜
i
dc
˜
i
a
˜
i
b
˜
i
c
Traction Load
˜
ω
m
˜
τ
m
l
Real-time controller
CAN EC
DAQ
Operator computer
ETH ETH
Figure 7: Electrical scheme of the test setup.
and its load (Figure 7). Both traction and load motors
use the same low-power PMSM, whose specifications
are provided in Tables 1 and 2. Their rotor shafts are
connected through a spider coupling, while their sta-
tors are rigidly mounted and aligned with a metallic
bracket.
The traction motor is controlled by a Texas In-
struments (TI) inverter development kit, whose torque
control algorithms can be freely tuned by the test
engineer (Texas Instruments, 2017). Its controller
chip is similar to the ones used in commercial road
car powertrain applications; only the power levels
are reduced. The load motor is controlled by a
maxon EPOS 3 industrial inverter, which operates in
speed-controlled mode. Finally, a 37.3 V DC-bus is
provided by a regular industrial Power Supply Unit
(PSU), which is protected from over-voltage during
sudden braking manoeuvres by a braking resistor and
a chopper. Both inverters share their DC-bus to enable
the recirculation of the generated power.
The test bench is controlled from a real-time com-
puting unit whose objectives are twofold: (i) execute
the above detailed real-time simulation model of the
vehicle and of the driver and (ii) handle the commu-
nication between the simulation model and the phys-
Table 1: Specifications of the PMSM used on the testbed.
Parameter Symb. Value
Pole pairs p 4
Voltage constant K
e
4.75 V
phph
/krpm
Torque constant K
t
0.04533 N m/A
Stator resistance R
s
0.36 (25
C)
d-axis inductance L
d
0.201 mH
q-axis inductance L
q
0.201 mH
Rotor inertia J
r
7 kg mm
2
Scalable Electric-motor-in-the-Loop Testing for Vehicle Powertrains
597
Co-simulation master
FMI
slave
EC
slave
CAN
slave
Data
logger
Configuration
Data
file
Figure 8: Structure of the real-time controller. Blue, green
and purple blocks correspond to hard real-time processes,
non-real-time processes and output files respectively.
ical components on the testbed (Figure 7). As in a
real automotive use case, the torque
˜
τ
r
e
requested by
the VCU and the inverter state transitions are sent to
the TI inverter over Controller Area Network (CAN).
The maxon load inverter communicates the measured
load torque
˜
τ
m
l
and speed
˜
ω
m
to the controller and
receives its new reference speed
˜
ω
r
using the indus-
trial field bus EtherCAT (EC) (EtherCAT Technology
group, 2019). Finally, a Simcenter SCADAS Data-
Acquisition System (DAQ) is included in the setup to
measure the speed
˜
ω
m
and two phase currents
˜
i
a
and
˜
i
b
(the last current
˜
i
c
being computed as
˜
i
c
=
˜
i
a
˜
i
b
) of
the surrogate PMSM, as well as the current
˜
i
dc
flow-
ing from the DC-bus to the traction inverter during
the execution of the tests. Both this DAQ and the
real-time computing unit are configured via Ethernet
(ETH) from an operator computer.
4 REAL-TIME CONTROL
STRUCTURE
The real-time computing unit is responsible for the
execution of simulation models and the interfacing
between them and with the physical components on
the test bench. For this purpose, a real-time co-
simulation architecture is set up within the unit, which
is functionally divided into a master and several slaves
(Figure 8).
The co-simulation master orchestrates the execu-
tion and data exchange of the different co-simulation
slaves. The master execution is scheduled by a low
jitter real-time timer, firing every master timestep t
M
i
.
At the beginning of each timestep, the master triggers
the execution of the slaves and waits for them to return
outputs. Co-simulation correction algorithms are then
run on the resulting outputs and the corrected values
are propagated to the respective slave inputs. Finally,
the master waits for the end of the current timestep
unless an overrun occurred (Figure 9).
In the present application, a co-simulation slave
exposing a Functional Mock-up Interface (FMI) is
used to execute the Amesim model (Blochwitz et al.,
2012). Apart from the execution of the simulation
models, co-simulation slaves are also responsible for
specific functionalities of the controller, such as the
handling of physical I/O. Indeed, both CAN and
EtherCAT slaves are used to communicate with the
traction and load inverter respectively.
In general, the computation rate of the master can
differ from the one of the slaves. Indeed, the co-
simulation slaves can run their models once or more
per master timestep t
M
to ensure a specific virtual rate
t
M
n
S
i
for the numerical integration, where n
S
i
is the num-
ber of micro-steps of the ith slave. Subsequently, their
results are returned to the master for synchronisation
and exchange (Gomes et al., 2017). Due to the mas-
ter pacing at a lower rate, the slaves see their inputs
updated only once every n
S
i
steps. This could intro-
duce artefacts in the co-simulation and generate non-
physical frequency content in the signals. To miti-
gate this effect, the co-simulation master implements
numerical techniques aiming to reconstruct the be-
haviour of the input signals locally to the slave during
isolation. An example of such method is provided in
(Stettinger et al., 2014).
Each of the previously mentioned slaves, as well
as the master, are executed in separate processes and
on different CPU cores of the real-time computing
unit to minimise computational time jitter. In fact, the
most important requirement for a real-time system is
time determinism over the carried-out operations. For
this reason, the real-time unit is based on a real-time
patched (Real-Time Linux Wiki, 2016) Linux kernel
that runs the operating system as a fully preemptive
process, hence increasing scheduling and computa-
tional time determinism.
CAN slave
EC slave
FMI slave
Master
t
M
0
t
M
1
Figure 9: Timing diagram of the real-time controller.
ICINCO 2020 - 17th International Conference on Informatics in Control, Automation and Robotics
598
5 INTRODUCTION OF SCALING
FACTORS
Scaling factors are used to match the full-scale real-
time model to the reduced-scale test bench. These co-
efficients are calculated following the approach used
in (Petersheim and Brennan, 2009). The relevant pa-
rameters for motor scaling are first identified, based
on the torque-speed curve and the limitations of the
motor: the maximum motor torque τ
max
, the maxi-
mum power P
max
, the maximum speed ω
max
, the ro-
tational speed ω and the motor torque τ. The values
of τ
max
, ω
max
and P
max
are listed for the EV motor
and for the surrogate PMSM, both with and with-
out considering friction, in Table 2. Three dimen-
sionless variables are then deduced based on the the-
ory of Applied Dimensional Analysis and Modelling
(Szirtes and R
´
ozsa, 2007): π
1
=
ω
ω
max
, π
2
=
τ
τ
max
and
π
3
=
P
max
ω
max
τ
max
. The detailed calculation of the dimen-
sionless variables can be found in (Joos, 2019).
By using reduced variables π
1
and π
2
, the torque-
versus-speed curves of the EV motor and small
PMSMs used on the bench can be plotted along nor-
malised axes, as shown in Figure 10. However, the
shapes of the curves do not match: the curve of the
EV motor consists of a hyperbolic curve in the flux
weakening zone, i.e. where the maximum power is
the main limitation of the machine. On the contrary,
the ideal torque-versus-speed curve of the PMSM of
the test bench presents a wide linear part, as the max-
imum continuous power is only reached close to the
maximum speed. A better match of both curves could
be obtained by performing flux weakening on the test
bench motor, but falls out of the scope of this paper,
as safety margins must be defined to avoid demag-
netization of the permanent magnets. Furthermore,
considering the non-negligible friction torque on that
machine, its corrected curve shows a maximum shaft
torque decreasing with speed as the friction torque in-
creases.
Considering the curves in Figure 10, a small area
(at low speed and high torque) can be observed where
the curve of the surrogate PMSM with friction is lo-
cated below the one of the EV motor. Therefore, those
operating points of the EV motor correspond to an
overload of the test bench PMSM. However, when
plotting the scaled working points corresponding to
the NEDC and WLTC, no point falls in that region
and an important margin can even be seen. There-
fore, these scaling factors ensure a safe operation of
the bench for testing such driving cycles.
By equalling the dimensionless variables for both
machines, the scaling factors for torque and speed can
be easily computed as the ratios between maximum
torques
˜
τ
max
τ
max
and speeds
˜
ω
max
ω
max
of both machines. For
sake of simplicity, the scaling factors are chosen for
the XiL tests without considering friction in Table 2,
i.e.
˜
τ
max
τ
max
= 1.03 ×10
3
and
˜
ω
max
ω
max
= 0.517.
6 XiL TEST RESULTS
Closed-loop XiL tests are performed to validate this
scalable testing approach. Detailed results on the first
part of the NEDC (0 s to 30 s) are first discussed be-
fore showing the overall bench performance on the
whole cycle. The evolution of speed, currents and
torque on the surrogate PMSM are displayed in Fig-
ures 11, 12 and 13 respectively.
In Figure 11 the shaft speed
˜
ω
m
of the surrogate
SuT measured with the Simcenter SCADAS is com-
pared to the theoretical profile of the NEDC converted
in terms of speed of the EV motor.
Figure 11 shows that the speed profile of the
NEDC is globally well followed for the small-scale
SuT, similarly to the results presented in (Petersheim
and Brennan, 2009; Ciceo et al., 2016) for normal-
scale benches. Some delays can, however, be reduced
through an improved tuning of the driver parameters.
The more important delay in the decelerating slope
can be linked to the use of different controller pa-
rameters for acceleration and braking, as presented in
Figure 3. When the reference speed starts decreas-
ing, there is a small delay between the time the accel-
eration controller stops acting to counter the friction
(i.e. the driver releases the throttle) and the time the
braking controller starts acting (i.e. the driver starts
pressing on the braking pedal). The absence of anti-
wind-up loops on these controllers is the most proba-
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-0.2
0
0.2
0.4
0.6
0.8
1
EV motor
PMSM motor (w/o friction)
PMSM motor (w/ friction)
NEDC
WLTC
PMSM overload
Figure 10: Dimensionless plot of the torque-versus-speed
curves of the real and surrogate SuTs, with operating point
for two different driving cycles.
Scalable Electric-motor-in-the-Loop Testing for Vehicle Powertrains
599
Table 2: Maximum ratings of the EV and surrogate SuT motors.
Parameter Symbol Value (EV) Value (PMSM without friction) Value (PMSM with friction)
Maximum speed ω
max
11.6 krpm 6 krpm 6 krpm
Maximum torque τ
max
270 N m 279 mN m 263 mN m
Maximum power P
max
75 kW 171.7 W 151.2 W
10 15 20 25 30
Time [s]
0
100
200
300
400
500
600
Test bench speed [rpm]
0
200
400
600
800
1000
EV motor speed [rpm]
Measured scaled speed
Linear speed from NEDC profile
Figure 11: Speed profile
˜
ω
m
of the surrogate SuT and the
first part of the urban cycle of the NEDC.
ble explanation.
Looking in greater detail at Figure 11, a small
bump can be observed prior to the plateau speed cor-
responding to 15 s in the NEDC. It is probably due to
the anticipative action of the acceleration controller,
monitoring the difference between present and fu-
ture (0.1 s later) speed set-points. Furthermore, just
before the begin of the rising slope at 11 s, a very-
low-amplitude negative speed is measured, probably
due to some backwards movement of the rotor at the
bench start-up. The traction inverter does not seem to
be responsible for this, as the phase currents in Fig-
ure 12 only show noise before 11 s.
Figure 12 presents the measured currents
˜
i
a
, and
10 15 20 25 30
Time [s]
-1.5
-1
-0.5
0
0.5
1
1.5
Current [A]
Phase a
AC RMS
DC
Figure 12: DC-bus, phase and estimated RMS currents in
the surrogate SuT during the first part of the urban cycle of
the NEDC.
˜
i
dc
on the SuT. These signals are displayed after hav-
ing been filtered with a zero-phase second-order But-
terworth filter with a 2 kHz cut-off frequency. This
filtering is required to match the hypothesis of sinu-
soidal and balanced phase currents needed to estimate
the value of the AC RMS current (also displayed in
Figure 12) with the following formula:
˜
i
e
rms
=
r
1
3
˜
i
a
2
+
˜
i
b
2
+
˜
i
c
2
, (6)
where the e superscript reflects the estimated nature of
this quantity. It is then used together with the torque
constant of the machine K
t
to estimate the torque of
the surrogate motor
˜
τ
e
e
:
˜
τ
e
e
= K
t
2
˜
i
e
rms
·sign
˜
i
dc, f iltered
, (7)
where
˜
i
dc, f iltered
represents
˜
i
dc
after applying a
second-order Butterworth filter with a 5.12 Hz cut-
off frequency to prevent noise from affecting the esti-
mated torque direction.
˜
τ
e
e
is an estimate of
˜
τ
m
e
.
The electromagnetic torque
˜
τ
e
e
of the traction
PMSM of the bench, estimated from (7), is compared
in Figure 13 with the theoretical torque profile to be
applied to the EV motor to follow the NEDC (con-
sidering a constant 57 % ratio of the braking torque
is supplied via the EV motor). An estimate of the
loading-machine torque
˜
τ
e
l
using the measurements
acquired with the Simcenter SCADAS is also dis-
played in the figure. This value is computed from
˜
τ
e
e
using (5).
10 15 20 25 30
Time [s]
-40
-20
0
20
40
60
Electromagnetic torque [mNm]
-40
-20
0
20
40
60
EV motor torque [Nm]
SUT machine
Loading machine
EV motor
Figure 13: Torques of the surrogate SuT
˜
τ
e
e
and loading ma-
chine
˜
τ
e
l
(left scale), as well as the corresponding torque for
the real vehicle (right scale).
ICINCO 2020 - 17th International Conference on Informatics in Control, Automation and Robotics
600
10 15 20 25 30
Time [s]
0
2
4
6
8
10
12
14
16
Consumed energy (bench) [J]
0
1
2
3
4
5
6
7
8
9
Consumed energy (EV) [Wh]
DC bus (PMSM)
PMSM shaft (w/o friction)
DC bus (EV)
EV motor shaft (w/o friction)
Figure 14: Consumed energy by the surrogate SuT and the
corresponding energy requirement from the EV.
The resulting torque profile is globally well followed
with the expected scaling factor. However, some de-
lay is observed due to the driver’s dynamic, as already
shown for the speed profile in Figure 11. The evo-
lution of
˜
τ
e
l
shows the important part of the friction,
which is compensated for by feed-forwarding the es-
timated friction torque
˜
τ
f
in the real-time simulation
model.
Some torque oscillations can be noticed at the be-
ginning and the end of the test. This corresponds to
moments where the motors start or stop their rotation.
The ones of
˜
τ
e
l
before the acceleration ramp are prob-
ably nonphysical and linked to the small backwards
movement of the rotor and the way this torque is es-
timated. However, the later oscillations in the SuT
and loading machine torques could have a three-fold
origin. Firstly, the sensorless torque control of the
TI drive, which uses a speed estimator instead of a
sensor, tends to work badly at lower speeds. Indeed,
the estimated direction of rotation can be inaccurate
when its speed is around 0 rpm. Secondly, the load-
ing machine is controlled in block commutation by
the maxon drive, which leads to a jerky torque gen-
eration at lower speeds. Finally, the friction torque
compensation provided in (5) also shows a disconti-
0 2000 4000 6000 8000 10000
Speed [rpm]
0
50
100
150
200
250
Torque [Nm]
0.25
0.25
0.25
0.25
0.25
0.5
0.5
0.5
0.5
0.5
0.6
0.6
0.6
0.6
0.6
0.7
0.7
0.7
0.7
0.7
0.75
0.75
0.75
0.75
0.8
0.8
0.8
0.8
0.825
0.825
0.825
0.825
0.85
0.85
0.85
0.85
0.875
0.875
0.875
0.9
0.9
0.91
0.91
0.92
0.92
0.93
Figure 15: Estimated efficiency map of the EV motor.
0 200 400 600 800 1000 1200
Time [s]
0
500
1000
1500
2000
2500
3000
3500
4000
4500
Test bench speed [rpm]
Scaled actual speed
Scaled reference speed
Figure 16: Actual
˜
ω
m
and reference speeds of the surrogate
SuT for the full NEDC.
nuity around 0 rpm, which could result in jumps in τ
m
e
that propagate across the real-time simulation model.
Figure 14 shows the amount of energy absorbed
by the traction inverter and delivered to the shaft of
the surrogate PMSM excluding friction. For compar-
ison, their theoretical counterparts computed directly
from the NEDC speed profile for the EV motor are
displayed as well, considering the assumptions made
in (Joos, 2019) regarding the EV motor and converter
efficiencies. While a constant 95 % converter effi-
ciency is assumed, the efficiency map of the EV mo-
tor is computed by adding a 15 % constant offset to
the one of the surrogate PMSM, leading to the map
displayed in Figure 15.
It appears that the consumed energies at the shaft
match relatively well, even if an increased difference
can be seen during the decelerating ramp, due to the
delay between the measured speed and the theoretical
one. However, the energies taken from the DC-bus or
battery differ, due to the different motor efficiencies.
The scaled reference and actual speeds of the load-
ing PMSM extracted from the Amesim model over
the whole NEDC are shown in Figure 16. They show
0 200 400 600 800 1000 1200
Time [s]
0
20
40
60
80
100
120
140
160
180
Energy consumption [Wh/km]
Figure 17: Estimated energy consumed at battery terminals
for the real vehicle over the NEDC based on torque and
speed measurements on the surrogate SuT.
Scalable Electric-motor-in-the-Loop Testing for Vehicle Powertrains
601
a good tracking of the reference speed by the maxon
drive, as the NEDC profile is easily recognisable.
Finally, the estimated consumed energy per kilo-
meter at the EV battery terminals is displayed in Fig-
ure 17 (considering the same assumptions as for Fig-
ure 14). The final value of this plot (168 Wh/km) is
17.5 % higher than the claimed consumption of the
modelled EV, i.e. 143 Wh/km (BMW, 2017). Such
an overestimation can be expected, since the raw es-
timation of the motor and converter efficiencies are
very probably underestimations.
7 CONCLUSIONS
In this work, the MBST methodology has been suc-
cessfully applied to a downscaled XiL test setup used
for the validation of EV powertrains, thanks to ade-
quate scaling factors. Validation has been performed
on the NEDC. The results also show the influence of
the control strategy on the tracking of the speed pro-
file and suggests a possible future use of the small-
scale bench for control optimisation.
In future work, the authors will build on this work
to increase the power level of the SuT and show
the scalability of the proposed framework to real EV
components. Additionally, the didactic side of this
downscaled setup will be further exploited by further
extending its instrumentation. The present PMSM
used in SuT will also be exchanged with an induc-
tion motor of similar power to enable testing this type
of traction motors.
ACKNOWLEDGEMENTS
This project was partially funded by the European
Union’s Horizon 2020 research and innovation pro-
gram under grant agreement No 769506.
The content of this publication does not reflect the of-
ficial opinion of the European Union. Responsibility
for the information and views expressed therein lies
entirely with the authors.
REFERENCES
Blochwitz, T., Otter, M.,
˚
Akesson, J., Arnold, M., Clauß,
C., Elmqvist, H., Friedrich, M., Junghanns, A.,
Mauss, J., Neumerkel, D., Olsson, H., and Viel,
A. (2012). Functional mockup interface 2.0: The
standard for tool independent exchange of simulation
models. In Proceedings of the 9th International Mod-
elica Conference.
BMW (2017). Specifications. the new bmw i3. In BMW
media information, pages 1–8.
Ciceo, S., Mollet, Y., Sarrazin, M., Van der Auweraer, H.,
and Martis, C. (2016). Model-based design and test-
ing for electric vehicle energy consumption analysis.
Electrotehnic
˘
a, Electronic
˘
a, Automatic
˘
a, 64:46–51.
EtherCAT Technology group (2019). EtherCAT. www.
ethercat.org. Accessed: 2019-08-05.
Fathy, H. K., Filipi, Z. S., Hagena, J., and Stein, J. (2006).
Review of hardware-in-the-loop simulation and its
prospects in the automotive area. Proceedings of SPIE
- The International Society for Optical Engineering,
6228.
Fiori, C., Ahn, K., and Rakha, H. A. (2016). Power-
based electric vehicle energy consumption model:
Model development and validation. Applied Energy,
168:257–268.
Gomes, C., Thule, C., Larsen, P. G., and Vangheluwe, H.
(2017). Co-simulation: State of the art.
Joos, A. J. (2019). Implementation of a small-scale elec-
trical drive system in Hardware-in-the-Loop simula-
tions. Master’s thesis, Brussels Faculty of Engineer-
ing.
Obelics consortium (2019). Homepage - Obelics. obelics.
eu. Accessed: 2019-08-05.
Petersheim, M. D. and Brennan, S. (2009). Scal-
ing of hybrid-electric vehicle powertrain components
for hardware-in-the-loop simulation. Mechatronics,
19:1078–1090.
Popp, A., Sarrazin, M., Van der Auweraer, H., Fodorean,
D., Birte, O., Karoly, B., and Martis, C. (2015). Real-
time co-simulation platform for electromechanical ve-
hicle applications. 2015 9th International Symposium
on Advanced Topics in Electrical Engineering, pages
240–243.
Real-Time Linux Wiki (2016). Real-Time Linux Wiki.
rt.wiki.kernel.org/index.php/Main Page. Accessed:
2020-04-01.
Siemens Digital Industries Software (2019). Model-
based system testing: Efficiently combining test
and simulation for model-based development.
https://www.plm.automation.siemens.com/media/
global/en/Model-based%20system%20testing%
20WP tcm27-67978.pdf.
Stettinger, G., Horn, M., Benedikt, M., and Zehetner, J.
(2014). Model-based coupling approach for non-
iterative real-time co-simulation. In 2014 European
Control Conference (ECC), pages 2084–2089.
Szirtes, T. and R
´
ozsa, P., editors (2007). Applied Di-
mensional Analysis and Modeling, pages 133 161.
Butterworth-Heinemann, Burlington, 2nd edition.
Texas Instruments (2017). InstaSPIN-FOC and InstaSPIN-
MOTION User’s Guide. http://www.ti.com/lit/ug/
spruhj1g/spruhj1g.pdf. Accessed: 2019-08-05.
Van der Auweraer, H., Sarrazin, M., and Marques dos San-
tos, F. (2017). Model-based system testing: A new
ICINCO 2020 - 17th International Conference on Informatics in Control, Automation and Robotics
602
drive to integrating test and simulation. In 2017 In-
ternational Conference on Structural Engineering Dy-
namics.
Williamson, S., Lukic, M., and Emadi, A. (2006). Compre-
hensive drive train efficiency analysis of hybrid elec-
tric and fuel cell vehicles based on motor-controller
efficiency modeling. IEEE Transactions on power
electronics, 21(3):730–740.
Yang, Y.-P., Shih, Y.-C., and Chen, J.-M. (2014). Real-time
driving strategy for a pure electric vehicle with mul-
tiple traction motors by particle swarm optimization.
7th IET International Conference on Power Electron-
ics, Machines and Drives (PEMD 2014).
Yang, Z., Shang, F., Brown, I. P., and Krishnamurthy, M.
(2015). Comparative study of interior permanent mag-
net, induction, and switched reluctance motor drives
for ev and hev applications. IEEE Transactions on
Transportation Electrification, 1(3):245–254.
ACRONYMS
CAN Controller Area Network.
DAQ Data-Acquisition System.
EC EtherCAT.
ETH Ethernet.
EV Electric Vehicle.
FMI Functional Mock-up Interface.
MBST Model-Based System Testing.
NEDC New European Driving Cycle.
PI Proportional Integral.
PMSM Permanent-Magnet Synchronous Machine.
PSU Power Supply Unit.
SuT System-under-Test.
TI Texas Instruments.
VCU Vehicle Control Unit.
WLTC World harmonised Light vehicles Test Cycle.
XiL X-in-the-Loop.
SYMBOLS
C
x
Air penetration coefficient.
F
r
Longitudinal resistive force.
J
w
Wheel inertia.
M Vehicle mass.
R
w
Wheel radius.
S Vehicle active area.
˙x
r
Reference longitudinal vehicle speed.
˙x Longitudinal vehicle speed.
ω
m
Measured EV motor rotational speed.
ω
r
Reference EV motor rotational speed.
ρ
air
Air density.
τ
r
User torque request.
τ
max
b
Maximum vehicle braking torque.
τ
max
e
Maximum EV motor torque.
τ
min
e
Minimum EV motor torque.
τ
b
Mechanical braking torque.
τ
m
e
Measured EV motor torque.
τ
r
e
Reference EV motor torque.
˜
ω
m
Measured SuT rotational speed on the test bench.
˜
ω
r
Reference SuT rotational speed on the test bench.
˜
τ
m
e
Torque at the SuT shaft on the test bench.
˜
τ
e
e
Estimated SuT torque on the test bench, based on
Simcenter SCADAS measurements.
˜
τ
r
e
Reference SuT torque on the test bench.
˜
τ
f
Test bench mechanical friction torque.
˜
τ
e
l
Estimated load motor torque on the test bench,
based on Simcenter SCADAS measurements.
˜
τ
m
l
Measured load motor torque on the test bench.
˜
i
e
rms
Estimated RMS current in the phases of the SuT.
˜
i
a
Current in the phase a of the SuT.
˜
i
b
Current in the phase c of the SuT.
˜
i
c
Current in the phase c of the SuT.
˜
i
dc
Current in the DC-bus of the SuT.
g Constant of gravity.
r Gearbox ratio.
u
b
Brake command.
u
t
Throttle command.
x Vehicle longitudinal position.
Scalable Electric-motor-in-the-Loop Testing for Vehicle Powertrains
603