Anda di halaman 1dari 5

Co-Simulation of an Electric Traction Drive

Christoph Schulte and Joachim Böcker

Abstract—For the simulation of electrical drives, reduced- of the design and the system set up of the coupled structure, the
order models or simple look-up tables are often used in order basic principle is described. A co-simulation generally consists
to decrease the computation time. In the latter case, the quality of several components, each equipped with a bidirectional
of the results is reduced due to interpolation. This publication
introduces a coupled simulation structure which includes the information interface. This means that data or information can
control, the power electronics and also a permanent magnet be received from and sent to the linked software tools. This
synchronous machine. Each of these components is simulated results in a coupled simulation environment which works as
in different software environments. The coupling between the a combined system. The system model of the electric drive
different software tools is realized by a defined data flow discussed in this work consists of the following components:
structure. Such a structure can be useful for accurate design and
system optimization. Comparison between the simulation results 1) Control model in MATLAB/Simulink
and experimental measurements shows that a good agreement is 2) Power Converter in ANSYS Simplorer
obtained. 3) Motor model in ANSYS Maxwell 2D
Figure 1 shows a system level representation of the entire drive
scheme.
I. I NTRODUCTION
ONVENTIONALLY reduced order models are being
C used for the simulation of electric drives, where the
inverter for the motor is replaced by a sinusoidal voltage
A. Control (MATLAB/Simulink)
The control structure used here is based on flux-oriented
source and the motor model is replaced by a simple look-up- control (FOC) which ensures good dynamics, stability and
table. While such an approach can give reasonable information high efficiency in the whole operation range as reported in
about the system level performance of the drive, it cannot [5].
identify and investigate certain transient and harmonic effects, udc
switching- and increased iron losses etc., which are necessary udc
Control Scheme
for optimization of the drive [4]. In order to maximize the
ua sa
quality of simulation results of an electrical drivetrain it is i∗d
T∗ ub sb
therefore essential to use a model which is as detailed as
Operating Point i∗q Field Oriented
PWM
Selection Current Control uc sc
possible. But, making such a detailed model is not feasible
using a single software tool [2]. This is because, the model ia
ψmax ib
which consists of a control structure, power electronics and
ic
a machine, is a heterogeneous model. The machine model is a
replaced by a finite element model to achieve high accuracy. Modulation
udc
Control
Although the simulation effort is increased by the finite IPMSM

element analysis [1], the present day computing hardware is ε


Position Signal
ωel
capable of handling this. The requirement of a detailed model, Processing
coupled with the capabilities of modern computers, propel us T ,n
towards realizing a more sophisticated simulation structure.
This publication presents a coupled simulation structure (co- Fig. 1. Overview of a complete drive system containing of field-oriented
simulation) for an electric drive, where the control structure, control structure, power converter and motor model [5]
power electronics and motor are modeled in different environ-
ments. A data flow between each model is realized using a The operation characteristics for the different working
defined structure. points are defined in the Operating Point Selection block by
several parameters including speed of the motor and predeter-
II. S IMULATION STRUCTURE mined d- and q-current shapes or desired torque T ∗ . The entire
control scheme was implemented in MATLAB in accordance
The main focus is the development of a co-simulation struc- with [5]. The inputs to this control scheme are the three motor
ture of an electrical traction drive. Before going into the details currents ia,b,c , the motor angle ε and the DC link voltage uDC
while it outputs the pulse-width-modulation (PWM) signals
Christoph Schulte and Joachim Böcker are with the department for Power
Electronics and Electrical Drives, University of Paderborn, D-33095 Pader- sa,b,c for the power electronic inverter.
born (Germany), E-Mail: {schulte, boecker}@lea.upb.de
This work was developed in the course of the Collaborative Research Center
614 Self-Optimizing Concepts and Structures in Mechanical Engineering
University of Paderborn, and was published on its behalf and funded by the
Deutsche Forschungsgemeinschaft.
B. Power Converter (ANSYS Simplorer)
q
Figure 2 shows the three-phase inverter [3] of the simulation
structure which is implemented in ANSYS Simplorer. Here,
ideal switches are used, but it is also possible to replace them
with exact models of IGBTs. The PWM signals (sa , sb , sc )
generated from the control scheme in MATLAB are received d
through an interface (which will be described in the next
section) and fed to the inverter switches S1 to S6 built up
in Simplorer.

Fig. 3. Sketch of an permanent magnet synchronous motor with embedded


magnets and corresponding direct and quadrature axes
S1 S2 S3
RS Lend
uDC
S4 S5 S6 III. C O -S IMULATION
In order to execute the co-simulation, continuous data
exchange needs to take place among all the individual compo-
nents (MATLAB, Simplorer and Maxwell models) as shown
sa sb sc in Fig. 5. The interface between MATLAB and Simplorer has
PWM been realized and implemented using an S-function (as shown
in figure 4) provided by the software manufacturer ANSYS.
Fig. 2. 3-Phase Power Converter with additional winding resistance and
end-winding inductance

imported from Simplorer


sa ε
exported to Simplorer

Besides the inverter, the model includes the concentrated


phase end-winding inductances Lend and phase winding resis- ia

S-Function
tances RS . These components are included in this model as it sb
is not possible to integrate them in a two dimensional Finite
ib
Element Method (FEM) based motor model.

C. Motor (ANSYS Maxwell 2D) sc ic


The permanent magnet synchronous machine with interior
magnets (IPMSM), which is exemplarily illustrated in Fig. Fig. 4. Embedded MATLAB S-Function block showing exported and
3, is realized as a two-dimensional Finite Element model in imported values
Maxwell 2D. A two-dimensional model is chosen, as it can
generate almost all the information at a reduced computational
burden. Such two dimensional models are extensively used
The objective of this S-function is to transfer the switching
in motor design analysis and optimization. IPMSM has been
signals sa,b,c to the inverter model in Simplorer and to receive
chosen for this particular co-simulation analysis, owing to
the motor currents ia,b,c and rotor position ε from the motor
the increasing interests of automotive manufactures on it.
model in Maxwell. Even other variables (like flux linkages,
Such a motor presents advantages like an improvement in
losses, etc.) can be exchanged if required. It acts as data in-
maximum torque to weight ratio as well as efficiency. The
terface between MATLAB and Simplorer as there is no direct
design of motors with embedded magnets results in a non-
data exchange between Maxwell and MATLAB. Information
uniform reluctances leading to inductances Ld and Lq . For a
exchange between the models in MATLAB and Maxwell (like
given permanent magnet flux Ψp and number of pole pairs p
speed and torque information) therefore must be realized via
the torque T can be calculated as
Simplorer. This is because there exists only a configurable
3 data interface module between Maxwell and Simplorer, which
T = p(Ψp iq + (Ld − Lq )id iq ) (1)
2 is provided by the software manufacturer ANSYS.
where, id is the direct axis current component and the iq is
the quadrature axis component of the current. The torque T With all links established, the co-simulation process can
can be divided into two parts: the reaction and the reluctance be started by executing the MATLAB model with a given
torque. The reaction torque is a function of current iq as well initial position of the motor. All other software tools are
as the permanent magnet flux Ψp , while the reluctance torque then started automatically. The corresponding voltage pulse
can be defined as a function of Ld , Lq , id and iq . A sketch of pattern is calculated by the control and forwarded to the power
a motor model with embedded magnets is shown in Fig. 3. inverter model in Simplorer. In Simplorer, this pattern is then
Case 1: n = 6000 min−1 , id = −100 A, iq = 0 A → 100 A
MATLAB
Simulink

sa,b,c 120 120


ia,b,c , ε
100 100
ANSYS ANSYS

iq in A
80 80
Simplorer Maxwell 60 60
ua,b,c
40 40
Fig. 5. Dataflow of the co-simulation between all software tools 20 Co−Sim 20
Meas.
0 Ref. 0

transmitted to switches of the three phases. The resulting 0 2 4 6 8 50 52 54 56 58


voltage waveforms of the inverter are then passed to the finite- -80 -80
element model in Maxwell. The FEM model calculates the

id in A
torque, the phase currents and the field quantities. In order to -90 -90
close the simulation cycle the current motor position ε and
currents ia,b,c are transmitted from the motor model to the -100 -100
control scheme in MATLAB via Simplorer as shown in Fig.
5. -110 -110
The advantage of selecting such a co-simulation concept -120 -120
is that e.g. field quantities or the occurring iron losses,
which are mainly influenced by the switching of the power 0 2 4 6 8 50 52 54 56 58
t in ms t in ms
converter, can be studied in detail. Furthermore, every
part of the co-simulation can be replaced by a new model
Fig. 6. Responses of iq and id for the step change in iq at 6000 min−1
without affecting the simulation structure as long as the
inputs and outputs of the interface remain the same. Hence,
performance comparisons between several combinations of Case 2: n = 9000 min−1 , id = −200 A, iq = 0 A → 50 A
motors, control and power electronic circuit models can be
generated easily. The behavior of such a model will be very
close to that of a hardware setup as shown in the next section. 60
50 50
iq in A

40 40
30 30
20 20
IV. E XPERIMENTAL VALIDATION
10 Co−Sim 10
0 Meas. 0
-10 Ref. -10
In order to validate the concept of a coupled co-simulation
model, co-simulation results were compared with the measure- 0 2 4 6 8 50 52 54 56 58
ment results from an equivalent test setup. Several operating
-180 -180
points n ∈ {6000 min−1 , 9000 min−1 , 12000 min−1 } were
examined. At each case (refer to Fig. 6-8), the response
id in A

-190 -190
to a step change in the q component of the current iq
was observed at t = 0 ms. Then, 50 ms later, the iq step -200 -200
demand was reduced to 0 A. During these operations the
d component of the current id was kept unchanged at a -210 -210
constant value. To ensure the comparability of both the results,
the simulation results were temporally adjusted according -220 -220
to the measurement hardware equidistant sampling rate of 0 2 4 6 8 50 52 54 56 58
100 µs. In the following the focus of the several profiles t in ms t in ms
lies in the observation of the several currents during the
step changes. While the red dashed profile shows the de- Fig. 7. Responses of iq and id for the step change in iq at 9000 min−1
manded current, the blue and green profiles indicate the
test bench measurement and simulation results respectively.
Case 3: n = 12000 min−1 , id = −300 A, iq = 0 A → 25 A Case 2: n = 9000 min−1 , id = −200 A, iq = 0 A → 50 A

30 30 1.2
1
iq in A

TMeas,avg
20 20 0.8

TCo-Sim
0.6
10 10
0.4
Co−Sim 0.2
0 Meas. 0
0 Co−Simulation
Ref. Avg. Measurement
-10 -10 -0.2
0 2 4 6 8 50 52 54 56 58 0 10 20 30 40 50
-270 -270
t in ms
-280 -280
Fig. 10. Torque profile of the co-simulation at 9000 min−1
id in A

-290 -290
Case 3: n = 12000 min−1 , id = −300 A, iq = 0 A → 25 A
-300 -300
-310 -310
-320 -320 1.4
-330 -330 1.2
0 2 4 6 8 50 52 54 56 58 1
t in ms t in ms
TMeas,avg
TCo-Sim

0.8
0.6
Fig. 8. Responses of iq and id for the step change in iq at 12000 min−1 0.4
0.2
0 Co−Simulation
After adaption of all simulation results and comparison with -0.2 Avg. Measurement
the test bench measurements profiles a good agreement can -0.4
be observed with the given plots. Due to the cross-coupling 0 10 20 30 40 50
of the two current components, under- and overshoots can be t in ms
identified in the measurement plots as well as in the profiles
Fig. 11. Torque profile of the co-simulation at 12000 min−1
of the co-simulation results.
Similar to the current response, comparison from the ex-
perimental and simulated results, the measured TMeas,avg and The dashed blue lines indicate the average steady state
simulated torque profiles TCo-Sim were examined for all the values of the measured torque of the equivalent test setup. The
three cases. The following plots show the normalized results measured value from the torque sensor is only an indication of
of the co-simulation. the average torque measured once the steady state has reached.

To perform a validation of the simulation results a com-


Case 1: n = 6000 min−1 , id = −100 A, iq = 0 A → 100 A parison with test bench measurements was done. While the
co-simulation current profiles are directly compared with the
test setup measurements the torque profiles are compared in
the time-range 20 ms - 50 ms with the steady state torque of
1.2 the measurement. The resulting difference for every case is
1 given in Table I.
TMeas,avg

0.8
TCo-Sim

TABLE I
0.6 AVERAGE DIFFERENCE OF THE CURRENT COMPONENTS AND THE TORQUE
0.4
Working Point Diff. in id Diff. in iq Diff. in T
0.2
Co−Simulation 6000 min−1 1.01% 1.49% 2.45%
0
Avg. Measurement 9000 min−1 0.51% 2.91% 0.78%
-0.2
12000 min−1 4.84% 1.33% 3.01%
0 10 20 30 40 50
t in ms
The average difference of the co-simulation results com-
Fig. 9. Torque profile of the co-simulation at 6000 min−1 pared to the measurement lies in the rage of 0.5% to 4.8%.
Therefore, it can be seen that both results are in close R EFERENCES
coordination with each other. [1] S. Kanerva, J. Kaukonen, A. Szucs, and T. Hautamaki. Coupled fem-
The other benefit of a co-simulation is the estimation of control simulation in the analysis of electrical machines and converters.
In 12th International Power Electronics and Motion Control Conference,
losses with any inverter fed motor. The iron loss PC,Co-Sim 2006.
calculated with the co-simulation includes all harmonic losses [2] U. Knorr and R. Juchem. A complete co-simulation-based design
due to the high switching frequencies of the inverter. In environment for electric and hybrid-electric vehicles, fuel-cell systems
and drive trains. Technical report, Ansoft Corporation, Pittsburgh, 2003.
contrast the loss calculation PC,FEM of the common Finite [3] R. Krishnan. Permanent Magnet Synchronous and Brushless DC Motor
Element Analysis (FEA) only the fundamental frequency is Drives. CRC Press, 2010.
considered. Thus a holistic estimate of losses can be obtained [4] F. Leonardi and B. Ionescu. Advancements in tools and methods for
the design of permanent magnet integrated starter alternators. In IEEE
from a co-simulation structure. This is also one of the main International Conference on Electric Machines and Drives, 2005.
features of the structure. Table II gives an overview over the [5] W. Peters, T. Huber, and J. Böcker. Control realization for an interior
simulated iron losses for both cases. permanent magnet synchronous motor (ipmsm) in automotive drive trains.
In PCIM Europe, 2011.

TABLE II
C OMPARISON OF I RON LOSS FOR COMMON FEM AND CO - SIMULATION

PC,Co-Sim
Working Point PC,FEM PC,Co-Sim ∆ PC,FEM

6000 min−1 370.04 W 525.37 W 41.97%


9000 min−1 558.25 W 781.67 W 40.02%
12000 min−1 1124.07 W 1519.52 W 35.18%

When considering the losses in both cases, a clear difference


is apparent. The iron loss calculated with all voltage harmon-
ics exceed the fundamental losses of the common FEA by
approximately 35%. This result can be handful when the drive
efficiency has to be considered.

V. C ONCLUSION AND F UTURE WORK

The proposed co-simulation structure presents advantages in


terms of accuracy and modularity. Such a coupled model can
not only be useful for accurate motor design and optimization
but, also in the optimization of power electronics. Further, it
is possible to accurately study the system performance before
realization of an expensive hardware. Other benefits include
its usage for accurate design and optimization while being
modular. As shown in three exemplary cases the difference
lies in the range of 0.3% to 4.9%. However, the proposed co-
simulation structure suffers with a problem of same simulation
step size for all the simulation components. This will greatly
slow down the simulation because, the step time is usually
decided based on the component with highest dynamics.
Ideally, it would be greatly beneficial to increase the step
time for the motor model in order to speed up the simulation,
without affecting the results. Future work will be concentrated
towards finding a co-simulation method in which individual
components are simulated with differing timesteps.
Nevertheless, the proposed co-simulation structure provides
when compared with the usual individual simulations the
following benefits:
• Analysis of impact factors
• High degree of modularity
• Ability to optimize on system level

Anda mungkin juga menyukai