336 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO.

2, MARCH 2004

Analysis and Validation of a Real-Time

AC Drive Simulator
Roger Champagne, Louis-A. Dessaint, Senior Member, IEEE, Handy Fortin-Blanchette, and Gilbert Sybille

Abstract—This paper presents the analysis and validation of a eventually be installed on the grid, by interconnecting an actual
real-time ac drive simulator. Real-time simulation has been used controller to a simulated grid. This in turn opens the possibility
for over a decade in power systems engineering to test controllers of testing the various controllers much more thoroughly, in ways
thoroughly, efficiently, and safely. The development and testing
of large ac drives could benefit from this type of simulation, but that are impossible on a real power system, either for safety rea-
real-time power system simulators have modeling restrictions sons or simply because of the high cost of performing “live”
preventing stable and accurate simulation of isolated drives tests. Such simulators are used to test, among others, static reac-
switching at high frequencies. These restrictions can be overcome tive power (VAR) compensation systems (SVS) and controllers
by combining in a single model both the power converter and for HVDC systems [1].
motor models. After discussing the salient issues related to the
real-time simulation of ac drives, a prototype real-time simulator Engineers responsible for designing and prototyping high-
for drives is described. Its ability to accurately emulate the power ac drives could benefit from a real-time drives simulator
behavior of large drives is demonstrated through two case studies. just as the power engineering community has. Although there
A first case study demonstrates the feasibility of simulating a drive are numerous simulation tools dedicated to the simulation of
of typical complexity in real-time. The second case study demon- drives, to our knowledge there currently exists no real-time sim-
strates how the controller for a PWM VSI fed induction motor,
switching at 4 kHz, is designed, implemented in a microcontroller ulator that allows interconnecting an ac drive’s controller to a
and tested in real-time using a simulated inverter, motor and load. simulated plant (i.e., power source, converter, machine and load)
Comparisons against results obtained with another simulation and characterized by a converter switching frequency in the
tool, which uses extremely accurate variable-step integration kHz range. Although current real-time power system simulators
algorithms, demonstrates the validity of our approach. include power converter and electric machinery models, limi-
Index Terms—AC motor drives, hardware-in-the-loop testing, tations of these models and/or the overall simulation strategy
real time simulation, state space methods. prevents their use in the study of drives. The main limitation
(with respect to simulating drives) of existing real-time power
I. INTRODUCTION system simulators resides in the fact that machines are indepen-
dent models, as is the case in most simulation packages that

A S A CONSEQUENCE of major technological advance-

ments, variable speed ac drives have developed rapidly in
the past two decades. Switching devices such as the IGBT are
have electric machinery models. In such a context, the machine
models usually appear as current sources to the rest of the power
system. Consequently, it is often required to have some kind of
able to convert increasing amounts of power at higher frequen- noninductive passive load across the machine’s terminals (or be-
cies than the previous switching technologies. At the same time, tween its terminals and the reference node) in order to prevent
the availability of low-cost computing hardware, such as digital having only current sources and inductive branches connected
signal processors (DSPs) and microcontrollers led to more com- to a node, which translates into attempting to force a current
plex control schemes being incorporated in modern drive con- into an inductance and violates basic electric laws. In order to
trollers. Such systems today incorporate sensorless operation, obtain a stable simulation, the value of the passive load (usu-
fault detection, advanced diagnostics, and self-tuning capabili- ally a resistor) must be chosen in such a way that this parasitic
ties. load consumes at least as much as five or ten percent of the ma-
It is interesting to note that simulation tools made available chine’s nominal power. In a power system context, where ma-
to engineers designing modern drives have not advanced quite chines are usually included to model part of a manufacturing
as fast. Fully digital real-time simulation has been used for over plant, these passive loads surrounding the machine appear nat-
a decade now in the field of large power systems. This type of urally as lighting or heating equipment. However, if one wants
simulation allows engineers to emulate the dynamics of large to study the behavior of a drive in isolation, the presence of this
power systems and study the behavior of controllers that will parasitic load is unacceptable, yet necessary to keep the simu-
lation stable.
Nevertheless, real-time simulation of drives is discussed in
Manuscript received July 3, 2002; revised October 10, 2003. Recommended the literature. Le-Huy et al. [2] present a real-time model of a
by Associate Editor A. M. Trzynadlowski.
R. Champagne, L.-A. Dessaint, and H. Fortin-Blanchette are with the Elec- four-quadrant 200 HP dc drive in the context of a large power
trical Engineering Department, École de technologie supérieure, Montréal, QC system study. Sureshbabu et al. [3] consider various aspects
H3C 1K3, Canada (e-mail: [email protected]). of modeling induction motors in real-time, focusing their at-
G. Sybille is with the Power System Testing and Network Simulation Labo-
ratory, Institut de recherche d’Hydro-Québec, Varennes, QC J3X 1S1, Canada. tention on the choice of reference frame and integration algo-
Digital Object Identifier 10.1109/TPEL.2003.823242 rithm. Zhao et al. [4] present real-time simulation results of a

0885-8993/04$20.00 © 2004 IEEE

CHAMPAGNE et al.: ANALYSIS AND VALIDATION OF A REAL-TIME AC DRIVE SIMULATOR 337

vector controlled induction motor, but their simulation consists under study, which can be a tedious task [9]. However, we will
in a simulated motor connected to a real controller and power describe in subsequent sections a fully automated process which
converter, therefore bypassing the difficulties associated to the obtains the state equations of any drive.
simulation of high frequency switching devices contained in
the converter. Abourida et al. [5] present a real-time simulator B. Machine Equations
which they claim can simulate drives in real-time, but the exam- Electric machinery modeling is a topic that has been ex-
ples provided don’t support this claim. Moreover, they specifi- tensively covered in the literature (see for instance [10]).
cally state that for PWM simulations (which use passive loads However, in the context of a real-time simulation, there are a
instead of motors), their system is limited to a switching fre- few important points which must be considered carefully. In
quency of 2 kHz for a 50 time step and uses averaging sim- this paper, we consider a three-phase three-wire wye connected
ulation models for switching frequencies beyond this limit. squirrel-cage induction machine. The fact that we are consid-
This paper presents the analysis and validation of a proto- ering a squirrel-cage machine implies that the rotor voltages are
type real-time ac drive simulator that offers new possibilities zero, while the three-wire connection implies that there is no
to drive designers. The rest of this paper is organized as fol- homopolar component in the voltages or currents. Moreover,
lows. Section II discusses various modeling issues related to there are only two independent stator voltages, for instance
the individual central components of a drive, namely the power and , and two independent currents, for instance and ,
converter and the electric motor. An innovative method of up- in both the stator and rotor.
dating converter-related state equations is described. Section III In order to simplify the differential equations describing the
presents a global model created by Davat [6], based on the state dynamic behavior of machines, it is common practice to con-
variable approach (SVA), which includes both the machine and sider machine equations in a set of transformed coordinates
inverter. This model is a key factor in our ability to simulate . The most common reference frames are
ac drives in real-time. Section IV demonstrates the feasibility • the rotor reference frame, associated to the Park transfor-
of the proposed approach by providing results from two case mation;
studies. The first case study, where all drive components are • the stationary reference frame, associated to the Clarke or
simulated (including the controller), is used to show how drives transformation;
with multiple power converters (for instance a rectifier and an • the synchronous reference frame.
inverter) are handled in our approach. In the second case study, The choice of a reference frame transformation depends on
a drive consisting of a PWM VSI inverter fed induction motor the application [10]. In our case, the rotor voltages are zero but
is simulated. The (real) controller uses a PI speed regulator the stator will be fed by a power converter. This encourages
and space vector PWM with a switching frequency of 4 kHz. us to use either the stationary or synchronous reference frame
This case study provides a quick walkthrough of the process in- transformation. Because it yields simpler (constant) transforma-
volved in designing and prototyping a drive’s controller using tion matrices, we have chosen the stationary reference frame.
our approach. Validation of both case studies’ results is per- The transformation equations for stator voltages and currents
formed by comparing them to those obtained with another simu- are then given by (3) and (4), where the subscript indicates
lation tool well adapted for the simulation of drives, namely the stator quantities
Power System Blockset (PSB) [7]. Finally, we provide a few
concluding remarks and discuss future work. (3)

II. MODELING DRIVE COMPONENTS

(4)
A. State Variable Approach
In this paper, we demonstrate the use of the SVA to simulate With stator and rotor currents as state variables, the induction
drives in real-time. A system is described in the SVA by machine’s state equations are given by

(1)
(2)

where is the vector of state variables, is the vector of inputs,

and is the vector of outputs. There are several reasons why
we choose to model electric drives with the SVA. First, it is an (5)
intuitive method for our purpose since power electronics con-
verters and electric machinery are both naturally expressed in (6)
the form of first-order differential equations. There is a wealth
of mathematical knowledge on solving such equations and on In (5) and (6), the and subscripts are used to indicate
the properties of their solutions [8]. SVA is also easily extended -axis and -axis quantities, respectively. Subscripts and in-
to time-varying and nonlinear networks and is often the method dicate rotor and stator quantities. Note that all rotor variables are
of choice in such cases. The main drawback of the SVA is that it viewed from the stator, which is indicated in (5) and (6) by the
requires that the analyst derive the state equations of the system prime symbol next to all rotor variables. and are the total
338 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 2, MARCH 2004

rotor and stator inductances, respectively, while is the mu-

tual inductance. Finally, is the electrical angular speed of the
rotor (mechanical speed multiplied by number of pole pairs).

C. Initial Computation of Converter State Equations

The initial set of state equations for the converter and source
are obtained by considering all switches open. Linear graph
theory provides a systematic and easily programmable means
of modeling the converter and power source, using SVA. The
method used in our work is well described in [8]. It consists
in finding a proper or modified proper tree and constructing Fig. 1. Converter-related state-space representation with switches modeled
outside the state-space matrices.
the nodal incidence and fundamental cut-sets matrices.
Then, from these and basic components matrices (resistors, ca-
pacitors and inductors), the state-space representation of the Fig. 2 illustrates the overall state-space representation ob-
source and converter is obtained in the form of matrices , , tained for the source and converter, with the switches included
, and , where subscript is used to indicate the original in the state-space matrices. It is a reduction of Fig. 1 using
values of the various matrices, obtained with all switches open.
In our scheme, switches are modeled as binary resistors. There-
fore, converter-related matrices , , , and will change
as the simulation progresses and our system is said to be time-
varying. The source and converter’s state variables are inde- (7)
pendent capacitor voltages and independent inductor currents.
These quantities are referred to as twig capacitor voltages and
link inductor currents in [8]. III. SIMULATING THE COMPLETE DRIVE
This method of computing state equations for electrical sys-
tems uses matrix computations extensively. When obtained with A. Modeling
a computational engine which treats sparse matrices efficiently, Since the power converter and electric machine are of
such as Matlab, state equations for power systems can be com- different nature, it seems logical to simulate these two compo-
puted very fast [11]. nents of a drive separately, exchanging various signals between
both parts. As an example, the machine can be modeled as a
D. Converter Equation Updates voltage-controlled current source, with the converter computing
the voltage across the machine’s terminals and the machine
Every time a switch changes state, we must reformulate the computing its currents from these voltages and feeding the
state equations for the converter. This could be done by using the currents back to the converter at the next integration step,
same method as the one described above. However, this method as shown in Fig. 3. However, this approach doesn’t work, if
involves a fairly large number of matrix computations and is not we consider only fixed-step integration algorithms [6]. When
realizable in real-time. A more efficient method of updating the solving differential equations with fixed-step algorithms, we
state matrices must be used. assume that the inputs are constant during the time step. As a
Such a method was recently developed and implemented in consequence of separating the simulations, the converter now
version 2.0 of the Power System Blockset, a Matlab toolbox sees two inputs: the independent voltage source and the stator
used for simulation of power systems and drives. It consists in currents of the machine. The voltage source can indeed be
computing updated matrices , , , and from the ma- considered constant during the time step if it is chosen appro-
trices of original values , , , and (obtained with all priately. However, the stator currents depend on the interaction
switches open) and the switch states (on or off). Fig. 1 illustrates between the converter and machine. Therefore, they cannot
in the form of a control diagram the starting point of our ap- be assumed constant during the time step. Consequently, we
proach. The original state-space representation , , and propose to use a global modeling technique which is presented
has for inputs a vector of switch currents followed by here. The complete development can be found in [6]. The
all other inputs (ideal voltage and current sources) contained in system shown in Fig. 4 will be used to describe the proposed
vector . The output of the state-space representation con- method.
sists in a vector of switch voltages and a vector of various First, we consider the machine’s stator currents as fictitious
other measurements (voltages and currents) . The feedback current sources injected back to the converter. The converter
loop simply contains matrix , which relates switch currents now has two sets of inputs: vector and vector
to switch voltages. is a diagonal matrix in which the first , as shown in Fig. 5. Next, we need to consider the
diagonal elements (for a system containing switches) system in three distinct parts: the converter part, the machine
are the conductance values of the switches in the system and part and the part which is common to both converter and ma-
have either a zero (open switch) or high (closed switch) value. chine, as illustrated in Fig. 5. We will soon develop interface
The remaining elements of are zero. equations for the latter. In Fig. 5, subscript is used to indicate
CHAMPAGNE et al.: ANALYSIS AND VALIDATION OF A REAL-TIME AC DRIVE SIMULATOR 339

Fig. 2. Equivalent state-space representation of Fig. 1.

Fig. 3. Separate simulation of converter and machine.

Fig. 5. Dividing the system in three parts.

longer considered in the following equations. The machine’s

state-space representation (9) is then re-written as

Fig. 4. Electric diagram of test system.

(12)
quantities relating to the converter only, while subscript The relation between the fictitious current sources and the
is used to indicate quantities relating to the machine only. The machine’s stator currents in the reference frame is given
rotor voltages and currents, and , are the only quanti- by
ties which fall in this last category. Finally, subscript is used
to indicate quantities which are common to both the converter (13)
and machine parts of the system. For the moment, the fictitious
current sources and voltages across them are the variables where is the reference-frame transformation matrix defined
common to both parts. in (4). The next step toward our final system of equations con-
We mentioned earlier how we obtain the state-space repre- sists in finding the relation between the machine’s stator volt-
sentation of the source and converter using linear graph theory. ages in the reference frame, , and the voltage across the
Using (1) and splitting the input vector into two vectors and fictitious current sources, . First, we define
, we obtain (8), where the state vector consists in twig ca-
pacitor voltages and link inductor currents , (14)
as described earlier
where matrices , and are obtained through linear graph
(8)
theory and are simply sub matrices of the converter output ma-
trices and . Finally, the machine’s stator voltages are re-
For the machine, the state-space representation of the elec-
lated to the fictitious current source voltages by
trical part was shown in (5). This equation can be written in
compact form as (9), where subscript indicates machine-re-
lated quantities
(15)
(9)
Combining (13) and (8), and then (15) and (12), we obtain
The machine’s state and input vectors are redefined as follows the global state-space representation (16) of the electrical part
to conform to the system in Fig. 5 of the entire drive, where the state variables are the converter
twig capacitor voltages and link inductor currents (in the
(10) reference-frame), and the machine’s stator and rotor currents in
(11) the stationary reference frame. In (16), subscript indicates
“final” state matrices. Because of the way the switch is modeled
Since we are considering a squirrel-cage induction machine, (two-valued resistor), matrices , , and will change
and are zero and therefore is zero and is no every time a switch changes state. Also, the terms in (16)
340 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 2, MARCH 2004

Fig. 6. First case study drive topology.

contain the machine’s rotor speed . This term is introduced A. Simulation Platform
by the reference-frame transformation Although this project concentrates on developing simulation
algorithms related to the real-time simulation of drives, the hard-
ware dimension of such simulations is also important. This sec-
tion therefore briefly describes the hardware aspect of imple-
menting a real-time drives simulator.
The simulator used in our work is not primarily meant for
the simulation of drives but is adequate for such use. This sim-
ulator is a Hypersim real-time power system simulator [13].
(16) Our setup consists of 15 computing nodes, each containing a
DEC Alpha 533 MHz processor and a SHARC DSP enabling
communications with six other processors. A set of digital and
analog inputs and outputs completes the simulator. This simu-
B. Numerical Integration
lator is perhaps too powerful for the simulation of drives. Un-
All the steps done so far involve continuous (not discrete) less the goal is to simulate a large number of interconnected
state matrices. Since we are seeking to implement a real-time drives, it is not required to have so many processors. Our first
simulation, we can’t consider variable-step integration algo- case study presented below uses three proces sors, while our
rithms. The iterative nature of these algorithms makes them second case study uses a single processor, as there is no rectifier
useless when considering real-time simulation of electrical in this case. Hence a much smaller (and less expensive) simu-
systems. Therefore, once and are computed, we must lator would be adequate for the simulation of one or perhaps a
compute their discrete equivalents, using some fixed-step few drives. We are currently investigating using PC clusters for
integration algorithm. our purposes, a trend which is also being investigated in the large
We have chosen to use the trapezoidal integration rule in our power system simulation community [14]. The main advantage
work. It is the most accurate implicit one-step method [12] and of using off-the-shelf PCs over custom built simulators is the
it is widely used in simulation of large power systems. Using substantially reduced cost, making the technology accessible to
this integration algorithm, we update the state vector at step a larger set of potential users. The main requirement for any
as in (17) [12], using a fixed time step ( is the identity hardware to be adequate for our purposes is the inter-processor
matrix) communication overhead and the I/O acquisition time, both of
which must be as low as possible, as they limit the smallest time
step that can be used for a real-time simulation.

(17) B. First Case Study: Decoupling of Rectifier and Inverter

In order to discuss various aspects of real-time simulation of
drives, we will first use the system shown in Fig. 6. It consists of
IV. REAL-TIME SIMULATION
an inductive three-phase ac power source, a six-pulse diode rec-
In this section, we describe the simulation platform and the tifier, a dc bus capacitor, a three-leg IGBT inverter and a 50 HP
results of two case studies performed to assess the validity of induction motor driving a fan-type load. A snubber consisting
the proposed methodology. The goal of the first case study is to of a series resistor and capacitor is placed across each of the
demonstrate that it is feasible to simulate a drive of typical com- switches in the diode rectifier. This system has a total of thirteen
plexity in real-time, using our approach. The goal of our second independent state variables (two inductor currents in the source,
case study is to demonstrate the usefulness of the approach as a the voltages across the six snubber capacitors, the voltage across
prototyping tool. In this case, a controller is developed and im- the dc bus capacitor, and two rotor and two stator currents in the
plemented in a microcontroller and interfaced to a drive which is machine) and thirteen switches (six in the rectifier, six in the
simulated in real-time. In both case studies, the results obtained inverter and one for the dynamic brake). The closed-loop con-
with our approach are compared to simulations performed with troller uses the direct flux and torque control (DFTC) scheme
the PSB. It is worth noting that these reference simulations are [15], [16]. The values of all system components correspond to
performed using variable-step integration algorithms, yielding those used in the ABB ACS601 drive. The values of the motor
extremely accurate results. parameters are those of the 50 HP motor found in [10].
CHAMPAGNE et al.: ANALYSIS AND VALIDATION OF A REAL-TIME AC DRIVE SIMULATOR 341

Fig. 7. Decoupling of first case study system at the dc bus.

TABLE I The second validation phase is qualitative and consists in

PEAK SIMULATION TIMES FOR THE VARIOUS PARTS OF FIRST CASE STUDY
actually looking at waveforms produced by the simulation.
The goal here is not to demonstrate the performance of the
DFTC controller or to obtain the best possible response time.
The DFTC controller was tuned until acceptable performance
was obtained, and then real-time results were compared to a
simulation of the same system using a different simulation tool
and environment (the Power System Blockset). The validation
therefore consists in verifying that the same results are obtained
with both tools. In order to perform this qualitative validation,
the drive is started from stall and a speed reference corre-
sponding to the motor’s nominal speed is imposed. The motor
is driving a fan type load which develops nominal motor torque
Since the simulation of this system involves a number of ma-
at nominal speed. Once nominal speed is reached, the speed
trix operations, some of which have order of complexity
reference is reversed. In other words, full nominal speed is
(inversion, multiplication), it is more efficient to simulate two
requested, but in the opposite direction. Fig. 8 shows the speed
small systems rather than simulating a single big system. The
reference and the resulting rotor speed obtained during this test.
idea behind decoupling is to split a system in two or more parts.
Fig. 9 shows the motor line current in phase . The top figure
At the point of decoupling in our scheme, the voltage is mea-
is obtained by the real-time simulation. The detailed view
sured on one side and fed to the other while on the other side,
(bottom figure) shows results obtained with both the PSB and
the current is measured and fed back to the first side. Both the
our real-time simulation. In this particular figure, it is important
transmitted current and voltage arrive on the opposite side with
to note that the reference waveform, obtained with the PSB,
a one time step delay. Therefore, the point at which we split-up
uses variable-step variable-order integration algorithms. This
or “decouple” the system must be chosen with care. In the case
simulation is entirely continuous, except that the switching
of a drive consisting in two converters separated by a dc bus,
frequency is limited to a little more than 15 kHz (one over the
this dc bus provides a good decoupling point. Decoupling of the
fixed sampling time used in the real-time simulation). On the
system at the dc bus is shown in Fig. 7.
other hand, the results obtained in real-time use the trapezoidal
With the decoupled system, we now have two systems to sim-
integration rule, at a fixed time step of 65 . The two curves
ulate. The source and rectifier part contains eight state variables
are slightly offset (on purpose), in order to offer a good view
and is simulated using the scheme presented in Section II-D.
on the details. Fig. 10 shows the dc bus voltage obtained in the
The rest of the diagram (not including the controller) contains
real-time simulation.
five state variables and is simulated using the scheme presented
The third validation phase consisted in verifying that the sim-
in Section III. The simulation is performed on the simulator de-
ulation remains stable over a long period of time. Because we
scribed in the previous section [13]. The source, rectifier and dy-
are using very small time steps, a lot of samples are generated
namic braking switch are simulated on processor #1, while the
for each second of simulation (over 15000 steps per second at a
dc bus, inverter, machine and load are simulated on processor
time step of 65 ). The fact that a computer represents numbers
#2. Finally, the DFTC controller is simulated on processor #3.
using a finite number of bits causes round-off errors which can
The switching frequency of the controller is limited to about
accumulate over the duration of a long simulation and cause a
4 kHz. Execution times for each part are indicated in Table I.
simulation to become unstable. This was a major problem in the
These times do not include inter-processor communications and
first fully digital real-time power system simulators, which used
input/output acquisition, which together yield an overhead of 25
16-b processors. In order to test for this eventuality, the simula-
at each simulation step. The time step used in case study 1 is
tion was left running for a period of 12 h and remained stable.
65 .
Moreover, a few more speed reference steps were imposed after
The validation of this case study is done along three dimen-
this long period and the results remained as accurate as at the
sions: real-time feasibility, qualitative correctness of results, and
beginning of the simulation.
stability of the simulation. The first objective was to determine
if this method can be used to simulate drives in real-time. This is
clearly possible as demonstrated by the results shown in Table I. C. Second Case Study: Closing the Loop
Adding the aforementioned 25 overhead to the computation The goal of this second case study is to demonstrate how a
times in the table, all values remain within the 65 time step. real-time drive simulator can be used to help develop and test the
342 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 2, MARCH 2004

Fig. 8. First case study: rotor speed reference and speed.

(a)

(b)
Fig. 9. First case study: motor current, phase A. (a) First 4.5 s of the simulation. (b) Detail of the last few cycles, showing real-time (thin curve) and PSB (thick
curve) results.

Fig. 10. First case study: dc bus voltage.

prototype of a drive’s controller in a safe and efficient manner. The first step consists in building an offline simulation that
In the previous case study, the controller was simulated. Here, will serve as reference. To perform this simulation, a model of
we describe a system where the controller was actually designed the controller is assembled in Simulink. This model was actu-
and implemented in a microcontroller. The controller is then in- ally used to design the controller and takes into account physical
terfaced to the real-time simulator, which emulates the entire limitations of the actual controller, such as the various sampling
high-power stage of the drive, namely the power source, con- rates used in different parts of the controller. The ac drive chosen
verter, machine and load. The rest of this section is a walk- for the study is a PWM VSI inverter fed induction motor. The
through of the process involved. PWM modulator is based on the space vector scheme [17] and
CHAMPAGNE et al.: ANALYSIS AND VALIDATION OF A REAL-TIME AC DRIVE SIMULATOR 343

Fig. 11. Second case study: offline simulation used as reference (Simulink/PSB).

Fig. 12. Second case study: experimental setup.

Fig. 13. Second case study, rotor speed: offline simulation with PSB (thin line), proposed real-time method (thick line).

the switching frequency is 4 kHz. The drive uses a 5 HP motor ulation. Fig. 14 shows the line current in phase of the motor,
and a fan-type load. The motor speed is regulated by a PI con- again using both simulation methods. There is a small difference
troller. Fig. 11 shows the PSB/Simulink block-diagram of the in rotor speed, resulting in the currents drifting away from each
complete drive. other. However, the envelopes of the two current waveforms are
Once the offline simulation model is built, a real-time ver- nearly identical and both curves exhibit the same frequency con-
sion is obtained simply by removing the controller part (since tent.
the intent is to interface a real controller to a simulated in- It is important to note that we are comparing waveforms from
verter and motor). The controller is implemented using a Mo- two very different simulations. On the one hand, the offline sim-
torola MC68332 microcontroller. The controller and simulator ulation uses a model of the controller, while on the other hand,
are electrically isolated from each other through phototransis- a real controller is interfaced to a simulated power stage. The
tors. The setup is shown in Fig. 12. small differences obtained in the speed and current waveforms
The test performed consists in starting the drive from stall, can probably be attributed to the fact that the controller model
imposing a speed reference of 300 rpm. After stabilization of used in the offline simulation does not represent all the sub-
rotor speed and motor current, the speed reference is stepped to tleties of the real controller. Moreover, the converters through
a new value of 600 rpm. Fig. 13 shows the rotor speed obtained which signals transit between the controller and the simulator
with both the reference offline simulation and the real-time sim- are simply modeled as zero-order holds in the offline simula-
344 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 2, MARCH 2004

(a)

(b)
Fig. 14. Second case study, motor phase A current using both simulation methods. (a) Overall view. (b) Detail of last few cycles using PSB (thin line) and
proposed real-time method (thick line).

tion. Last but not least, the reference offline simulation uses For instance, there are many industrial drives in motion ap-
variable-step variable-order integration algorithms, while our plications based on the permanent magnet synchronous motor
real-time approach uses fixed-step integration. (PMSM), or dc brushless motor. Ongoing work is also being
performed in order to use more affordable PC-based simulators
V. CONCLUSION and to improve the communications and acquisition overhead,
We have described in this paper a promising tool, a real-time in order to reduce the time step of the simulations. Alternative
simulator for ac drives. After discussing simulation issues re- algorithms are also currently under investigation.
lated to the individual components of the drive, a global state-
space model containing both the inverter and the motor was pre- ACKNOWLEDGMENT
sented. The use of this model is a key factor in the success of The authors wish to thank the anonymous reviewers for
the proposed approach, as it eliminates the one time step delay their constructive comments, which undoubtedly improved the
inherently present when the two components are simulated in quality of this publication.
separate models. A first case study demonstrated how to handle
systems that can’t be simulated in a single processor, which is REFERENCES
the case when there is more than one power converter. This case
[1] H. Nakra, G. Sybille, C. Gagnon, and M. Toupin, “The IREQ simu-
study illustrates a limitation of the proposed approach, in the lator—a test bench for closed loop testing of controllers of high power
sense that some types of drives require a smaller time step than electronics apparatus,” in Proc. Int. Conf. Power Electron. Drives En-
the one we are currently able to use to represent the full dy- ergy Syst. (PEDES), New Delhi, India, 1996.
[2] H. Le-Huy, G. Sybille, P. Giroux, J.-C. Soumagne, and F. Guay, “Dig-
namics of the drive. However, the results obtained with our ap- ital real-time simulation of a four-quadrant dc drive for static transfer
proach in this case study are more than acceptable for the early switch testing,” in Proc. IEEE Power Eng. Soc. Winter Meeting, 1999,
design stage of drives. A second case study demonstrated how a pp. 761–765.
[3] N. Sureshbabu, S. Seshagiri, A. Masrur, and B. K. Powell, “On real-time
simulated plant, containing the entire high power stage, can be simulation on induction motors,” in Proc. Amer. Contr. Conf., 1999, pp.
interfaced to a real drive controller in order to assist in designing 719–723.
and testing it. This case study used an induction motor fed by [4] Z. M. Zhao, S. Meng, and X. N. Yue, “A flexible virtual system for real
time simulation and evaluation of motor drives,” in Proc. IEEE 1999 Int.
a PWM VSI switching at close to 4 kHz. The results from both Conf. Power Electron. Drive Syst. (PEDS), 1999, pp. 361–365.
case studies prove that we are able to simulate such a system [5] S. Abourida, C. Dufour, J. Bélanger, G. Murere, N. Léchevin, and B. Yu,
in real-time and obtain results that compare very well to those “Real-time PC-based simulator of electric systems and drives,” in Proc.
17th Annu. IEEE Appl. Power Electron. Conf. Expo. (APEC), 2002, pp.
obtained with a different offline simulation tool which uses vari- 433–438.
able-step variable-order integration algorithms. Although the [6] B. Davat, “Étude-Mise au point et applications d’une méthode de simu-
PSB is able to simulate drives using fixed-step algorithms, com- lation globale de convertisseurs statiques connectés à des charges élec-
triques complexes. Study—Design and applications of a global simula-
paring our real-time results to a simulation using very accurate tion method of static converters connected to complex electric loads (in
variable-step integration makes our results that much more cred- French),” Ph.D. dissertation, Inst. Nat. Polytech. de Toulouse, Toulouse,
ible. France, 1979.
[7] L.-A. Dessaint, K. Al-Haddad, H. Le-Huy, G. Sybille, and P. Brunelle,
Follow-up work of this project will include testing the pro- “A power system simulation tool based on Simulink,” IEEE Trans. Ind.
posed approach with other converter and motor combinations. Electron., vol. 46, pp. 1252–1254, Dec. 1999.
CHAMPAGNE et al.: ANALYSIS AND VALIDATION OF A REAL-TIME AC DRIVE SIMULATOR 345

[8] N. Balabanian and T. A. Bickart, Electrical Network Theory. New Louis-A. Dessaint (M’88–SM’91) received the
York: Wiley, 1969. B.Ing., M.Sc.A., and Ph.D. degrees from the École
[9] O. Wasynczuk and S. D. Sudhoff, “Automated state model generation Polytechnique de Montréal, Montreal, QC, Canada,
algorithm for power circuits and systems,” IEEE Trans. Power Syst., vol. in 1978, 1980, and 1985, respectively, all in electrical
11, pp. 1951–1956, Nov. 1996. engineering.
[10] P. C. Krause, O. Wasynczuk, and S. D. Sudhoff, Analysis of Electric He was a Research Assistant at the Hydro-Québec
Machinery. New York: IEEE Press, 1995. Research Institute (IREQ) from 1980 to 1985, where
[11] R. Champagne, “Simulation en temps réel à l’aide de la représentation he worked on the simulation and control of a wind
d’état: application à un entraînement électrique basé sur une machine energy conversion system. He is currently a Professor
asynchrone. Real-Time simulation using the state variable approach: ap- of electrical engineering at the École de Technologie
plication to a drive based on an asynchronous machine,” Ph.D. disserta- Supérieure, Montréal. From 1992 to 2001, he was the
tion, École de Technologie Supérieure, Montréal, QC, Canada, 2001. Director of the Groupe de Recherche en Électronique de Puissance et Com-
[12] T. T. Hartley, G. O. Beale, and S. P. Chicatelli, Digital Simulation of mande Industrielle (GREPCI) (a research group on power electronics and dig-
Dynamic Systems: a Control Theory Approach. Englewood Cliffs, NJ: ital control). Since 2002, he has been the TransÉnergie Chair on Power Systems
Prentice-Hall, 1994. Simulation and Control. He is an author of the SimPowerSystems simulation
[13] V. Q. Do and J.-C. Soumagne, “Hypersim, an integrated real-time sim- software.
ulator for power networks and control systems,” in Proc. 3rd Int. Conf. Dr. Dessaint received the “Outstanding Engineer Award” from IEEE-Canada
Digital Power Syst. Simulators (ICDS), Västerås, Sweden, 1999. in 1997.
[14] C. Larose, S. Guerette, F. Guay, A. Nolet, T. Yamamoto, H. Enomoto,
Y. Kono, Y. Hasegawa, and H. Taoka, “A fully digital real-time power
system simulator based on PC-cluster,” in Proc. 7th Int. Conf. Modeling
Simulation Elect. Machines, Converters Syst. (Electrimacs’02), Mon-
tréal, QC, Canada, 2002.
[15] I. Takahashi and T. Noguchi, “A new quick-response and high-efficiency
control strategy of an induction motor,” IEEE Trans. Ind. Applicat., vol.
22, pp. 820–827, 1986.
[16] R. Champagne, L.-A. Dessaint, G. Sybille, and B. Khodabakhchian, “An Handy Fortin-Blanchette received the B.Eng. and M.Eng. degrees from École
approach for real-time simulation of electric drives,” in Proc. Canadian de Technologie Supérieure, Montréal, QC, Canada, in 2001 and 2003, respec-
Conf. Elect. Comput. Eng. (CCECE), Halifax, NS, Canada, 2000. tively, where he is currently pursuing the Ph.D. degree in electrical engineering.
[17] K. Zhou and D. Wang, “Relationship between space-vector modulation He has been working with the Groupe de Recherche en Électronique de Puis-
and three-phase carrier-based PWM: a comprehensive analysis,” IEEE sance et Commande Industrielle (GRÉPCI) (a research group in power elec-
Trans. Ind. Electron., vol. 49, pp. 186–196, Feb. 2002. tronics and industrial control) as a Research Assistant for the past three years.

Roger Champagne received the B.Eng. degree in

electrical engineering, the M.Eng degree in systems
technology, and the Ph.D. degree in electrical
engineering from École de Technologie Supérieure
(ÉTS), Montréal, QC, Canada, in 1991, 1997, and Gilbert Sybille received the M.Sc. degree from Laval
2001, respectively. University, Quebec City, QC, Canada, in 1978.
From 1991 to 1994, he worked in industrial In 1978 he joined Hydro-Québec Power System
automation. He then was a Research Assistant Simulation Department, Institut de Recherche
at ÉTS from 1994 to 1997, in collaboration with d’Hydro-Québec (IREQ), as a Research Engineer.
Hydro-Québec’s Research Institute (IREQ). He He has been project leader in many simulation
is an author of the Mathworks SimPowerSystems studies where he developed an expertise in real
simulation software. He is currently a Professor with the Department of time testing of FACTS controllers. He has also
Electrical Engineering, ÉTS, and also a Guest Worker with the National developed various models and software programs
Research Council’s Software Engineering Group. His research interests for the IREQ’s real time simulator. He is the
include real-time and offline simulation of drives, and software engineering, technical leader and one of the key developers of the
specifically software architectures, and software quality. MATLAB SimPowerSystems simulation software.