IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 34, NO.

1, JANUARY/FEBRUARY 1998 133

Model-Based Current Control of AC Machines

Using the Internal Model Control Method
Lennart Harnefors, Member, IEEE, and Hans-Peter Nee, Member, IEEE

Abstract— In this paper, the internal model control (IMC) 1) Synchronous-frame PI or PI-type current controllers are
method is applied to ac machine current control. Permanent obtained.
magnet synchronous machines and induction machines are con- 2) The controller parameters (gain and integration time) are
sidered. The result is synchronous-frame proportional integral
(PI) or PI-type controllers, the parameters (gain and integration expressed directly in certain machine parameters and
time) of which are expressed directly in certain machine param- the desired closed-loop bandwidth. This simplifies the
eters and the desired closed-loop bandwidth. This simplifies the design procedure and trial-and-error steps are avoided.
controller design procedure, eliminating or reducing the need for Another benefit is that the cross coupling which exists between
trial-and-error steps, and is the main purpose for using IMC.
the and components of the stator voltage and current in
Index Terms—Current control, internal model control, inverter- synchronous coordinates can be compensated for easily. For
fed machines. standard synchronous-frame PI control, this cross coupling has
the following effect. When the setpoint (reference) for is
I. INTRODUCTION changed, a transient error in results, since the PI controllers
work in two single loops and are capable of suppressing the
T HERE have been several strategies proposed for ac
machine current control. Some well-known and sim-
ple methods are hysteresis control, stator-frame proportional
cross coupling only at steady state. To combat this, decoupling
networks in the current controllers have been suggested [8].
However, as pointed out in [6], a transient error in may
integral (PI) control, and synchronous-frame PI control [1,
not be very serious. For a field-oriented IM, it goes unnoticed
ch. 4, 5]. Of these, the latter has been acknowledged as
due to slow flux dynamics. Neither does it affect the torque
being superior, as unlike hysteresis control, it allows a fixed
for a round rotor PMSM (there is no reluctance torque). The
switching frequency and, unlike stator-frame PI control, it
approach in [6] is, therefore, to allow a fairly large transient
yields zero control error at steady state.
error in in exchange for a faster response in . Yet, there are
More advanced schemes—in the sense that improved per-
certainly cases where removal of cross coupling is desirable
formance in some respect, e.g., faster response or lower
and/or extremely fast response of is not required. We shall,
harmonic content is gained at the expense of higher controller
therefore, aim at removing the cross coupling, but not make
complexity than for the above ones—have also been proposed,
this endeavor a major issue. (A good topic for future research
including predictive control [2], generalized predictive control
would be to try to combine the simplicity of IMC design with
[3], state-variable regulator [4], state feedback [5], and the
the good properties of the control scheme in [6].)
promising new concept of minimum time control [6].
IMC has been considered for current control previously [9].
In this paper, the internal model control (IMC) method
The main difference between [9] and this paper is stated in
[7] is introduced and applied to ac machine current control.
Section III-A.
A permanent magnet synchronous machine (PMSM) is the
This paper is organized as follows. The PMSM model
working example. It is also shown how IMC can be applied
used for controller design is introduced in Section II. IMC
to induction machines (IM’s). The main benefits of IMC are
is introduced in Section III, and it is shown how the method
the following.
can be applied to controller design for the PMSM and the IM.
IMC is also analyzed with respect to sensitivity. Discrete-time
implementation issues, including selection of the sampling
Paper IPCSD 97–62, presented at the 1995 Industry Applications Society and switching frequencies, are considered in Section IV.
Annual Meeting, Lake Buena Vista, FL, October 8–12, and approved for
publication in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Simulations and experimental results are presented in Sections
Industrial Drives Committee of the IEEE Industry Applications Society. This V and VI, respectively.
work was supported in part by the NUTEK Center of Excellence in Electric
Power Engineering. Manuscript released for publication September 12, 1997.
L. Harnefors is with the Division of Electrical Machines and Drives,
Department of Electric Power Engineering, Royal Institute of Technology, II. THE MACHINE MODEL
SE-100 44 Stockholm, Sweden (e-mail: [email protected]) and with the
Department of Electrical Engineering, Mälardalen University, SE-721 23 In this section, we consider a model for the permanent
Västerås, Sweden. magnet synchronous machine which is suitable for controller
H.-P. Nee is with the Division of Electrical Machines and Drives, Depart- design. In Section III-C, it will be shown how the obtained
ment of Electric Power Engineering, Royal Institute of Technology, SE-100
44 Stockholm, Sweden. model can be modified and applied also to the induction
Publisher Item Identifier S 0093-9994(98)01210-9. machine. The PMSM is modeled in rotor coordinates (the
0093–9994/98$10.00  1998 IEEE

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.
134 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 34, NO. 1, JANUARY/FEBRUARY 1998

frame). From [10], we obtain

(1)

(2)

where and are the applied stator voltages, and the

stator currents, the stator resistance, the rotor speed, and Fig. 1. IMC structure.
and the and direction inductances, respectively. The
term is the back EMF. In (1) and (2), we clearly see that
cross-coupling (interaction) terms exist, namely,
and .
The system (1) and (2) is multivariable; there are two inputs
and outputs ( and voltages and currents, respectively).
We shall rewrite (1) and (2) using a transfer function matrix
[11], which is a generalization for multivariable systems of
the standard transfer function description. This technique is
rarely used for electrical machines and drives. However, we Fig. 2. Classic control system structure.
shall shortly see that matrix descriptions are very convenient.
Let us apply the Laplace transform to (1) and (2), regarding III. INTERNAL MODEL CONTROL
the permanent magnet flux term as a quasi-constant auxiliary
input signal which is included in the direction voltage: IMC was originally developed for chemical engineering
, while letting . The rotor applications [7] and is considered as a robust control method.
speed is considered as a slowly varying parameter as seen Before applying IMC to the current control problem, a general
from the electrical dynamics.1 Then presentation of the method is given.
The IMC structure is depicted in Fig. 1. The structure
(3) uses an internal model in parallel with the controlled
(4) system (plant) . For an ac machine, and are, thus,
the stator voltage and current vectors, respectively, while
Introducing Laplace transformed input and output vectors as is the current setpoint (reference) vector.
follows (where and are chosen instead of and for a The control loop is augmented by a block ; the so-called
notation consistent with that used in automatic control): IMC controller. , , and are all transfer function
matrices.
(5) There are some interesting facts worth noting about the IMC
structure.
1) IMC can be considered as a special case of the classic
we obtain
structure shown in Fig. 2. The controller in this
(6) system is related to the internal model and the
IMC controller in the following way:
where
(9)
where is the identity matrix. Thus,
. To obtain integral action, which
(7) is a necessity if steady-state errors are to be avoided,
should be zero. We thus need to let
.
and . We 2) It should be noted that, if the internal model is perfect,
also introduce the following notation, which will be used in i.e., , there is no feedback in Fig. 1, and
the following: the closed-loop system has the transfer function matrix
(10)
Hence, the closed-loop system is stable if and only if
and each are stable, implying that we have
(8) found all stabilizing controllers —given by (9)—for
the stable plant . This is a special case of the Youla
1 Current rise times of about 1 ms are common, whereas speed rise times as
parameterization [11]. In this case, it is tempting to
short as 10 ms are rare. Hence, there is normally a separation in dynamics of
at least one decade between stator current and rotor speed, so the mechanical let , giving , i.e., all plant
dynamics can be disregarded, as seen from the electrical dynamics. dynamics would be canceled and the output signal

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.
HARNEFORS AND NEE: MODEL-BASED CURRENT CONTROL OF AC MACHINES 135

would attain the setpoint instantaneously. It is clear

that this optimal result cannot be accomplished due to
a variety of reasons:
a) If is not minimum phase, i.e., there are zeros
in the right-half plane, will be unstable.
b) is hardly ever proper, i.e., the degrees of
the numerators of the elements are higher than the
Fig. 3. Decoupling and diagonal IMC (DIMC).
degrees of the denominators. cannot then
be implemented.
c) Any attempt to cancel the plant dynamics will give suitable controller. Using (9), with , we find that the
large magnitudes of the control signals. equivalent classic controller becomes
d) The method is highly sensitive to model errors.
However, with a few modifications, the idea can be used. Let
us factor as

(11)

where is the allpass part of , including all zeros in (17)

the right-half plane and all time delays. We could then make

(12) It is, thus, more straightforward to implement the classic struc-

ture (Fig. 2) than the IMC structure (Fig. 1). A comparison
which is the choice of obtained by the optimization
with two standard PI controllers (one each for the and
procedure [7]. However, this resolves only the first item on
loops)
the above list. We therefore detune the optimal controller with
a low-pass filter
(18)
(13)

This is generally chosen diagonal shows that (17) is an extension of PI control with integrators
added in the antidiagonal elements of in order to remove
the cross coupling, with

(14)

where the positive integer is chosen sufficiently large, so and

that becomes proper. Now, the closed-loop system can (19)
be made arbitrarily robust simply by making smaller.
(The controller (18) is, for obvious reasons, also called a
A. Controller Design for the PMSM diagonal controller, see below.)
In [9], the IMC structure was implemented rather than the
Let us now use the IMC method to design a suitable current
classic structure. Hence, the connection between IMC and PI
controller for the PMSM. Since has no right-half-plane
control was lost.
zeros and behaves as a first-order system for high frequencies,
we can let
B. Decoupling and Diagonal IMC (DIMC)
(15) An alternative to direct implementation of IMC is to first
decouple the machine dynamics by letting
where all diagonal elements of may be selected equal,
(20)
(16)
where and are the outputs of the and direction
Herein lies the main benefit of using IMC. PI controllers. This decoupling can be regarded as an inner
The tuning problem, which for a PI controller involves feedback loop, as shown in Fig. 3. The decoupling matrix
adjustment of two parameters, is reduced to the selection of is defined as in (8). This yields
one parameter only, the desired closed-loop bandwidth .
Since, for a first-order system, the 10%–90% rise time
(21)
is related to as , a specification of the rise
time immediately yields the desired bandwidth and, in turn, a

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.
136 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 34, NO. 1, JANUARY/FEBRUARY 1998

Now, the standard PI controller (18) can be used in the outer identical to that for the PMSM with the following parameter
feedback loop with the same parameter selection as for IMC, substitutions:
i.e., (19). If decoupling is not necessary, the inner loop can be
(28)
left out and standard synchronous-frame PI control is obtained.
Even in this case, (19) can be used for parameter selection. Due to this similarity, the notation for the PMSM is used
IMC and DIMC are compared in Section III-D. exclusively in the following. All results derived hold for the
IM as well, with the above substitutions.
C. Controller Design for the Rotor-Flux-Oriented IM
Consider the complex space vector equations for the IM in D. Sensitivity Analysis
synchronous coordinates [12]: We analyze, for IMC and DIMC, the relative sensitivity of
the loop transfer function matrix (which is the transfer
function of the controller–plant cascade) with respect to a
small deviation of the th model parameter, , from its
(22) true value, i.e.,

as (29)
(23)
We can express this as
Here, and are the stator and rotor current and flux
space vectors, respectively, while is the impressed stator (30)
voltage space vector. and are the stator and
rotor resistances and self-inductances, respectively. is the
That is, after the partial derivative is taken with respect to the
magnetizing inductance, while and are the
model parameter , the parameters of the model are made
stator and slip frequencies, respectively. Eliminating and
equal to those of the machine.
among the above equations, we obtain
For IMC, the loop transfer function is
, so we immediately obtain

(24) (31)

where . But rotor flux orientation implies For DIMC, we have

that is real and constant. Even though perfect field ori- where , giving
entation never can be accomplished in practice, will at
least vary slowly and can be considered as a quasi-constant (32)
“disturbance” which is compensated for by integral control
action. Hence, including the flux term in a modified stator
Differentiating the relation im-
voltage vector , we can write
plicitly yields
(25)

(33)
The complex-valued transfer function corresponds to the
inverse transfer function matrix and

(26)

Thus, the current controller for the rotor-flux-oriented IM is (34)

given by
From (8), it follows that the two functions can be written as
(27) (35)

from which for DIMC or standard PI control is (36)

obtained by putting . The controller for the IM is, thus,

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.
HARNEFORS AND NEE: MODEL-BASED CURRENT CONTROL OF AC MACHINES 137

IV. DISCRETE-TIME IMPLEMENTATION

We consider DIMC only. A sample software implementation
of the algorithm derived here—suitable for a digital signal
processor (DSP)—is presented in Section C of the Appendix.
The controller is given by

(37)

The controller can be implemented on state space. A realiza-

tion is straightforwardly obtained as

Fig. 4. Singular values of the PMSM model for !r = {3, 0.5, 0.1, 0} p.u. (38)
(solid, dashed, dash–dotted, and dotted, respectively).

The sensitivity functions of the two cases are, hence, scaled

by the machine transfer function matrix at the operating speed where is the controller state vector, ,
and zero speed, respectively. and is the inner feedback for cross-coupling removal.
Let us make an evaluation in the frequency domain. As
is a transfer function matrix, the gain is undefined.
A. Discretization
Instead, singular values have to be utilized; see [11], [13] and
Section A of the Appendix. The singular values of for Backward difference discretization [15] is suggested, which
different are depicted in Fig. 4 for a PMSM having the yields a simple algorithm:
parameter values and p.u. We can
note the following characteristics. (39)
• For low frequencies ( ) and high speeds, IMC has (40)
significantly lower sensitivity than DIMC. Furthermore,
the low-frequency sensitivity of IMC decreases with where is the sampling period and the sampling instant (at
increasing speed. This is good, since the cross coupling time ).
increases with . IMC can, therefore, be expected to
remove cross coupling better than DIMC. On the other B. Inverter Saturation and Antiwindup
hand, integral control action takes care of low-frequency
cross coupling, so the difference between IMC and DIMC A problem which has to be dealt with is that the output
may not be overwhelming. signal in practice must be limited, as the inverter will saturate
• The drawback of IMC is the large sensitivity peak at whenever reaches the maximum available voltage. Let
, which is equal in magnitude to the sensitivity of be the limited output:
DIMC at . This large peak is due to the attempt to
cancel the machine dynamics (which are characterized by
a complex pole pair, the imaginary parts of which follow (41)
the rotor speed [14]), and shows that oscillatory behavior
can be expected at high speeds.
However, only limiting the controller output signal will in-
The sensitivity analysis can be concluded by the following evitably lead to poor performance.
recommendation.
• Integrator windup results if the integration of the control
error is not stopped when saturates. This is known to
DIMC is more desirable than IMC, as the sensitivity peak give large overshoots [16].
at is avoided. IMC has lower low-frequency • The controller output limitation can be considered as a
sensitivity than DIMC, but the performance improvement disturbance entering between the controller and
is not expected to be overwhelming. the plant. IMC is sensitive to such disturbances, since the
canceled plant dynamics are “activated” [7], [16]. This is
Hence, IMC is of interest mainly not for direct implementa- undesirable also for DIMC, as the poles of the decoupled
tion, but as a design method for standard PI controllers, with or system are located close to the origin and will give rise
without an inner cross-coupling removal loop added (DIMC). to slow transients.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.
138 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 34, NO. 1, JANUARY/FEBRUARY 1998

Fig. 5. Vector block diagram of the DIMC current controller with anti-
windup. (W = 0 for PI control.)

The above problems can be dealt with in the following way,

which is known as back-calculation [16].
Step 1: At the sampling instant , compute the ideal con-
troller output,

(42)

and limit the signal, . This is the com-

mand vector to the pulsewidth modulator (PWM) Fig. 6. Space vector modulation. T and Tsw are the sampling and switching
[1], [17]. periods, respectively.
Step 2: In order to avoid the problems discussed above, the
saturation disturbance has to be “moved” from the C. Selection of the Sampling and Switching Frequencies
controller output to the controller input. Hence, by
inverting (42), compute the error signal which It is important to select the sampling and switching fre-
would yield : quencies high enough, so that the system performance does
not degrade. Selecting the angular sampling frequency
(43) as at least 10 times the closed-loop bandwidth is a good
recommendation [15]:
Then use to update the controller state vector
(i.e., the integrators): (48)

(44) As for the switching frequency, two independent voltage space

vectors can be generated per switching period using space
If does not saturate, . The latter two vector modulation of a voltage-source inverter [1, ch. 4], as
equations can be slightly simplified by combining illustrated in Fig. 6. (This holds for the suboscillation PWM
them as follows: method as well.) The angular switching frequency should,
hence, not be lower than half the sampling frequency. This
yields

(45) (49)

The controller matrices and can be simplified by Using this formula and the relation , we can relate
making suitable approximations. We have the switching frequency (in hertz) to the rise time:

(46) (50)

and A 1-ms rise time thus requires a switching frequency of

at least 1.75 kHz. This is a low figure for insulated gate
bipolar transistor IGBT and MOSFET inverters [17]; it is
often desirable to switch faster for lower harmonic content
(47) and reduced acoustical noise. The recommendation
obviously does not put excessive demands on the switching
frequency. Neither is the recommended sampling frequency
This results in the block diagram depicted in Fig. 5. of 3.5 kHz too demanding.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.
HARNEFORS AND NEE: MODEL-BASED CURRENT CONTROL OF AC MACHINES 139

Fig. 7. PMSM simulation. Step responses for IMC (solid), DIMC (dashed),
and PI (dash–dotted). Fig. 8. IM experiment. Step responses for DIMC.

V. SIMULATION rms value correspondence; p.u. corresponds to rated

current and similarly for the stator voltage vector.
Consider a PMSM with the data
in per-unit values and with a base frequency of 50 Hz,
A. Experiment 1
i.e., 314 rad/s. The permanent magnet flux p.u. and
the maximum stator voltage p.u. The machine The motor is operated at no load at approximately 80% of
model is slightly incorrect: . rated flux and p.u. At ms, the setpoint for is
From the machine model data we design a current controller stepped up 0.2 p.u. and stepped down again at ms. The
for a closed-loop system rise time of 1 ms. setpoint for is held constant at 0.44 p.u. The step responses
Bandwidth Selection: The desired rise time corresponds to for DIMC are shown in Fig. 8. (It should be noted that
p.u., so the bandwidth should be selected has been adjusted vertically slightly, so that the step starts at
as p.u. 0.1 p.u. and the voltages and in stator coordinates are
Sampling Rate Selection: The “10 times bandwidth” rule denoted as va and vb, respectively.) The step responses for
gives a sampling frequency of 70 p.u., i.e., 3.5 kHz. A PI are virtually identical and are not shown. This was to be
one-sample delay is simulated to account for inverter and expected due to the low rotor speed. The rise time is slightly
computational dead times. shorter than the theoretical 0.88 ms. The settling time on the
Evaluation: At p.u. and , we simulate the other hand is fairly long, which is an indication of slightly too
response to a setpoint step change; goes from 0.6 to 1.0 low integral action.
p.u. and back again at ms, while (see Fig. 7).
At the first step, the inverter saturates during the transient, B. Experiment 2
hence, the rise time is much longer than the desired. IMC is, The field is weakened ( p.u.) and the speed
as seen, slightly better than DIMC at removing the decoupling, increased to 1.6 p.u. The same experiment is then repeated
which was predicted by theory. The difference is, however, (see Figs. 9 and 10). This time, DIMC performs slightly better
not significant. Standard PI control gives a control error in than PI. The slow decay of the control error in which was
which very slowly decays to zero. acknowledged in the simulation is also present here (although
the error is not particularly large).
VI. EXPERIMENTAL RESULTS
The DIMC algorithm is evaluated experimentally on an VII. CONCLUSION
induction motor drive, the data of which are given in Section In this paper, the concept of internal model control was
B of the Appendix. The IM is controlled using indirect rotor applied to synchronous-frame current control for permanent
flux orientation [1, ch. 5]. magnet synchronous machines and induction machines. It
It was found that p.u. is a good tradeoff between was shown that the controller resulting from the IMC design
speed and robustness. A further increase was found to give method corresponds to two PI controllers with cross-coupling-
too oscillatory a response. The nominal rise time is then removing integrators added. The additional cost of implement-
ms. The control algorithm has a ing IMC compared to PI control is negligible. An alternative
delay of one sampling period. Suboscillation method PWM structure, DIMC, using an inner decoupling loop and two
generation is used, so the inverter is not utilized to its full outer standard PI loops was introduced as an alternative.
capacity. Note in the following that the vectors are scaled for DIMC is the structure preferable to use. Although IMC has

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.
140 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 34, NO. 1, JANUARY/FEBRUARY 1998

TABLE I
MACHINE DATA

Fig. 9. IM experiment. Step responses for DIMC.

in the implementation. With a state-of-the-art DSP, this is

usually no obstacle, however, since trigonometric operations
can be implemented using a lookup table or the CORDIC algo-
rithm [18]. If coordinate transformations are still not desirable,
the controller can be transformed to and implemented in stator
coordinates, as shown in [19].
For further information, see [14].

APPENDIX

A. Singular Values
For a multivariable dynamic system, one can obtain a
frequency function by substituting in any transfer
function matrix, as for any single variable (scalar) transfer
Fig. 10. IM experiment. Step responses for PI. function. However, the gain is not as straightforward
to express. It can be shown [11] that
lower parameter sensitivity, there is also a risk for oscillatory
behavior which is avoided with DIMC. (51)
The benefit of the IMC design method is that the controller
parameters are expressed directly in the machine parameters
where and are the minimum and maximum singular values.
and the desired closed-loop bandwidth. Hence, the design
The singular values for a frequency function matrix are defined
procedure is simple, and trial and error can largely be avoided.
as
It should be stressed that, even if removal of cross coupling
is not an important objective (see also [6]), IMC is still very
(52)
useful for standard PI controller design.
Rules for sampling and switching frequency selection were
where is the th eigenvalue. There are as many singular
given and discrete-time implementation issues were consid-
values as there are inputs/outputs. (We consider square
ered. A backcalculation algorithm was suggested in order to
only.) It is, hence, not possible to talk about a fixed gain but
avoid integrator windup (and, thus, degraded performance)
a gain spread, bounded by and . The actual gain depends
when the inverter saturates at transients.
on the direction of the input vector . See also [13].
We finally emphasize that the controllers are implemented in
synchronous coordinates, which has been shown to yield the
best performance among low-complexity current controllers B. Data for Laboratory Induction Motor Drive
[1, chs. 4, 5]. Therefore, coordinate transformations are needed 1) Machine: The data are given in Table I.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.
HARNEFORS AND NEE: MODEL-BASED CURRENT CONTROL OF AC MACHINES 141

2) Inverter: IGBT voltage-source inverter, 540-V dc link, [4] R. D. Lorenz and D. B. Lawson, “Performance of feedforward current
20-kW rated power, three-phase diode rectifier on mains side, regulators for field-oriented induction machine controllers,” IEEE Trans.
Ind. Applicat., vol. 23, pp. 597–602, July/Aug. 1987.
analog suboscillation method PWM generation with 5.3-kHz [5] D.-C. Lee, S.-K. Sul, and M.-H. Park, “High performance current
switching frequency. regulator for a field-oriented controlled induction motor drive,” IEEE
3) Digital Signal Processor: Texas Instruments TMS320C40 Trans. Ind. Applicat., vol. 23, pp. 1247–1257, Sept./Oct. 1994.
[6] J.-W. Choi and S.-K. Sul, “New control concept-minimum time cur-
floating-point DSP, sampling frequency 5.3 kHz, DSpace rent control in the three-phase PWM converter,” IEEE Trans. Power
development system on Pentium PC host. All algorithms are Electron., vol 21, pp. 124–131, Jan. 1997.
[7] M. Morari and E. Zafiriou, Robust Process Control. Englewood Cliffs,
implemented in C code. Stator current and resolver signal NJ: Prentice-Hall, 1989.
sampling is made synchronously with inverter switchings, at [8] I. Boldea and S. A. Nasar, Vector Control of AC Drives. Boca Raton,
the peaks of the PWM suboscillation triangular wave. This FL: CRC Press, 1992.
[9] J. L. Thomas and M. Boidin, “An internal model control structure in
gives significantly reduced noise levels in the samples. field oriented controlled v.s.i. induction motors,” in Proc. EPE, Florence,
Italy, 1991, vol. 2, pp. 202–207.
[10] P. Vas, Electrical Machines and Drives: A Space-Vector Theory Ap-
C. A Sample Software Implementation of DIMC proach. Oxford, U.K.: Clarendon, 1992.
A C program implementing the DIMC control algorithm fol- [11] J. M. Maciejowski, Multivariable Feedback Design. Reading, MA:
Addison-Wesley, 1989.
lows below. It should be self explanatory to anyone with some [12] W. Leonhard, Control of Electrical Drives. Berlin: Springer-Verlag,
knowledge of programming. Constants are written in capital 1985.
[13] L. Harnefors and H.-P. Nee, “On the dynamics of ac machines and
letters. Declarations of variables, constants, and functions are sampling rate selection for discrete-time vector control,” in Proc. ICEM,
not shown. Vigo, Spain, Sept. 1996, vol. 2, pp. 251–256.
[14] L. Harnefors, “On analysis, control and estimation of variable-speed
drives,” Ph.D. dissertation, Div. Electr. Machines and Drives, Dep.
& Elect. Power Eng., Royal Inst. Technol., Stockholm, Sweden, 1997.
[15] R. H. Middleton and G. C. Goodwin, Digital Control and Estimation:
A Unified Approach. Englewood Cliffs, NJ: Prentice-Hall, 1990.
[16] K. J. Åström and T. Hägglund, PID Controllers: Theory, Design and
Tuning, 2nd ed. Research Triangle Park, NC: Instrument Society of
America, 1995.
[17] J. G. Kassakian, M. F. Schlecht, and G. C. Verghese, Principles of Power
Electronics. Reading, MA: Addison-Wesley, 1991.
[18] S. Y. Kung, VLSI Array Processors. Englewood Cliffs, NJ: Prentice-
Hall, 1988.
[19] T. M. Rowan and R. J. Kerkman, “A new synchronous current regulator
and an analysis of current-regulated PWM inverters,” IEEE Trans. Ind.
Applicat., vol. 22, pp. 678–690, July/Aug. 1986.

Lennart Harnefors (S’94–M’98) received the

M.S., Licentiate, and Ph.D. degrees in electrical
engineering from the Royal Institute of Technology,
Stockholm, Sweden, in 1993, 1995, and 1997,
respectively.
Since 1994, he has been with the Department
of Electrical Engineering, Mälardalen University,
Västerås, Sweden, where he is a Senior Lecturer. He
is also an Affiliate Senior Lecturer with the Division
- of Electrical Machines and Drives, Department
of Electric Power Engineering, Royal Institute of
Technology, Stockholm, Sweden. His research interests include circuits,
systems, and control, particularly as applied to variable-speed drives.

Hans-Peter Nee (S’91–M’96) received the M.S.,

Licentiate, and Ph.D. degrees in electrical engi-
neering from the Royal Institute of Technology,
REFERENCES Stockholm, Sweden, in 1987, 1992, and 1996,
respectively.
[1] B. K. Bose, Ed., Power Electronics and Variable Frequency Drives. Since 1988, he has been with the Division of
Piscataway, NJ: IEEE Press, 1996. Electrical Machines and Drives, Department of
[2] J. Holtz and S. Stadtfeld, “A predictive current controller for the stator Electric Power Engineering, Royal Institute of
current vector of ac machines fed from a switched voltage source,” in Technology, Stockholm, Sweden, where he is a
Proc. IPEC, Tokyo, Japan, 1983, pp. 1665–1675. Research Associate and Director of the Permanent
[3] L. Zhang, R. Norman, and W. Shepherd, “Long-range predictive con- Magnet Drives Research Program. His research
trol of current regulated PWM for induction motor drives using the interests are permanent magnet and induction motor drives.
synchronous reference frame,” IEEE Trans. Contr. Syst. Technol., vol. Dr. Nee is the recipient of several awards, including a Best Paper Award
5, pp. 119–126, Jan. 1997. at the ICEM’94 Conference.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on March 03,2020 at 03:37:18 UTC from IEEE Xplore. Restrictions apply.