Sensorless three-phase induction motor direct torque

control using sliding mode control strategy

laboratory set-up for motor speed control teaching
Marcos Vinicius Lazarini and Ernesto Ruppert Filho
Electrical and Computer Engineering School
University of Campinas (UNICAMP) – Brazil
lazarini@dsce.fee.unicamp.br, ruppert@fee.unicamp.br

Abstract - A three-phase induction motor direct torque very useful as a student learning environment and a non-
control laboratory set-up for simulation and linear control laboratory set-up is presented in this article.
experimental activities is presented in this paper. It The proposed set-up allows simulation activities of the
includes both PI controllers and sliding-mode controllers complete system using developed Matlab SimPowerSystems
and uses a sensorless method to estimate rotor speed. The models, where a concept of modular control algorithm
subject of this set-up is to present to the students a permits that a block diagram can be easily replaced by other.
simulation tool based on Matlab SimPowerSystems Using the simulation block diagram students can program the
toolbox with the possibility to check simulation results DSP and use the DSP based experimental set-up to control a
against a DSP based experimental system. The set-up real machine. The DSP software was developed with
provides to the electrical engineering students an modules concept in mind, providing a rich framework where
excellent learning tool for non-linear control studies students can use various implemented modules in
using as example the variable speed three-phase experiments, closely related to simulations.
induction motor control. Using this constructed set-up the students can study, for
example, the response of the various DTC controllers’
Index Terms - induction motor, sensorless, sliding mode, schemes and how they behave when the estimator has
torque control, simulation and experimental system detuned parameters. The subject of the experimental set-up
and the simulations developed is to show to the students
INTRODUCTION many aspects of the non-linear systems using linear and non-
linear controllers. This paper is going to show the theoretical
Among many control methods of induction machines, one of foundations of the sensorless DTC of three-phase induction
the most important today is the Direct Torque Control (DTC) motor using sliding mode controllers and MRAS estimator.
method introduced by [1] and [2]. It can provide a very fast,
accurate, reliable flux control and torque responses, and it is INDUCTION MOTOR CONTROL THEORY FUNDAMENTALS
today one of the most important three-phase induction motor
control method. Fundamentals of the induction motor control theory applied
Sliding Mode Control (SMC) is presented today as a to the proposed students’ activities are summarized below.
practical alternative to implement a discontinuous control
I. Direct Torque Control
and has some interesting advantages over the traditional
control theory. As a discontinuous control, it has key From the induction machine mathematical model above, the
advantages like the ability to be a very robust control, in main equation of the direct torque control method [7] defines
many cases invariant to uncertainties and disturbances [5]; it a relationship between the electromagnetic torque and the
has also properties of order reduction, decoupling design angle between the stator and rotor magnetic fluxes as shown
procedure and simple implementation in electric drives, since in the relation
they have a natural “on-off” operating mode [8]. 3 P Lm
Sensorless drives are becoming more and more Tem = Ψs Ψr sin(δ ) (1)
important as they can eliminate the speed sensor maintaining 2 2 Lls Llr
accurate response. Monitoring only the stator current and where Tem is the electromagnetic torque, P is the number of
stator voltages, it is possible to estimate the necessary control pole pairs, Ψs and Ψr are the stator and rotor windings
variables. The observer type used here, a model-reference linkage fluxes, Lls and Llr are the stator and rotor winding
adaptative system (MRAS) [4], has presented good accuracy. leakage inductances, and δ is the angle between the stator
Simulation tools like Matlab/Simulink are becoming and rotor winding magnetic flux vectors. Ignoring parameter
more and more important, following the computational variations, as long as the stator winding magnetic flux
power growth. Complex systems can be easily simulated, but magnitude remains constant, the rotor winding magnetic flux
they do not give the real experience to the students. magnitude will be constant too and, from (1), the
Experimental systems are still complex and costly, being a electromagnetic torque is related only to the flux angle
real challenge to build them. However such activities are difference δ.

Coimbra, Portugal September 3 – 7, 2007

International Conference on Engineering Education – ICEE 2007
 The stator flux linkage vector depends directly on the
stator voltages [9] and can be calculated as
(
Ψs = ∫ Vs − rs i s dt ) (2)
where Vs is the phase terminal voltage of the machine, rs is
the stator winding resistance, and is is the stator current. The
underline denotes a 2-coordinate vector variable, the dq
coordinate system. In most cases the ohmic voltage drop is
small and can be neglected so the flux variations can be
approximated considering rs=0. Equation (2) shows that the
flux variation is then due only to the applied stator voltage.
In other words, to control the stator winding linkage flux
vector, the inverter generates a stator voltage vector that
moves the stator flux to a convenient position; this voltage
vector is recalculated at each switching period. FIGURE 1
DTC BLOCK DIAGRAM
The control algorithm calculates the magnitude and
angle to move the stator flux winding linkage vector to a
position producing torque according to (1). Flux vector
magnitude is usually hold at rated value to generate the
fastest possible response, but it is also possible to work with
an underrated flux vector magnitude, in order to have lower
acoustic noise and a better energy efficiency machine in
specific situations at the cost of a slower response [1].
In the classical DTC control, the traditional two-level
inverter can only produce 6 active and 2 inactive voltage
vectors. This is an important constrain to the system and it is
the cause of torque ripple in steady-state situations. Instead FIGURE 2
of the classical approach [1], using the space vector DTC-SLIDING MODE CONTROL LAW
modulation strategy for a given fixed-time switching period,
it is possible to generate a much wider range of average
voltage vectors, producing a much more smooth and accurate
response.
In order to find the correct magnitude and the new stator FIGURE 3
TYPICAL EVAL FUNCTIONS: A) SIGN, B) RELAY, C) LINEAR WITH SATURATION
winding linkage flux vector position, the induction machine
stator voltage equations [3] can be re-written. From [10], the
dq components in the stationary reference frame are II. Sliding Mode Control
dΨ*
dΨ *
Sliding Mode Control is presented today as a feasible
Vsd = rs isd + s
≈ s
alternative to implement a robust control, taking advantage
dt dt (3) of the “on-off” inverter switches characteristics and its
r
Vsq = rs isq + ω s Ψ = ω s Ψ + s * Tem
* * inherent discontinuous algorithm to control the electrical
Ψs
s s
machine. However this causes a major disadvantage that is a
where superscripts * means reference value and ωs is the variable switching frequency. In order to get a fixed
synchronous angular speed of the stator flux vector. The switching frequency operation it is used the space vector
above equations show, for a constant magnitude stator flux modulation strategy that preserves the good characteristics of
vector, that the d component of the stator voltage only affects the sliding mode control. Implementation in discrete-time
the stator flux and can be used to control it directly. The q digital signal processor also demands changes to avoid
component of the stator voltage affects the torque variable, unwanted chattering as explained below.
and if the term ωs Ψs* is decoupled, it can be used to control The sliding mode controller was designed to operate
the produced torque. over the same variables used in the DTC method. The main
Figure 1 illustrates the control block diagram of the goal of this approach is to obtain a sliding mode controller
implemented DTC scheme. Of the two PI controllers shown, with all the qualities of DTC, which calculates the optimal
although recommended by some authors [11], one controller stator voltage vector to maintain the stator winding linkage
can be removed without significant penalty to the dynamics flux magnitude and angle within the desired ranges, and
of the system. Because Vsd can be approximated to the outputs this reference to the SVM inverter. The implemented
variation of the stator flux according to (3), the stator block diagram is shown in Figure 2.
In this figure, the eval block usually is any function of
Ψs − Ψ
*
winding linkage flux output error ( ˆ ) already the following family: sign, relay or linear with saturation as
s

gives the proper variation and can drive Vsd directly. shown in Figure 3. Both the sign and the relay functions do
not perform accurately in a discrete-time system, resulting in
oscillations and undesired chattering. A linear function with

Coimbra, Portugal September 3 – 7, 2007

International Conference on Engineering Education – ICEE 2007
a proper gain provides much better results in reducing step response for example. A more detailed view of the
oscillations while still maintaining the properties of sliding background theory and sliding surface design is shown in [8].
mode [8].
The individual sliding surfaces are designed to behave in
III. Estimator Subsystem
a similar way to the DTC control. First it is defined the error
functions The adopted estimation structure is a model-reference
e Ψs = Ψ x − Ψ
* adaptative system MRAS [4], and it consists of three stage
ˆ
s system: two independent estimators and an adaptative
(4)
eTem = Tem − Tem
* ˆ mechanism to correct the estimations, producing the final
result. An improved model, discussed in [6], was the basic
where the hat symbol means estimated quantity and e is an
scheme. The adopted structure is presented in Figure 4.
error value to be minimized.
The sliding surface set S is defined from (4)

 1
s  e Ψ s + c Ψ s
d
(eΨs ) 
S= = dt
 (5)
 2  eTem + cTem d (eTem )
s
 dt 
where cΨs e cTem are constants to be defined according to the
desired dynamic response.
The surfaces s1 and s2 were defined according to (3),
using the same idea that the d stator voltage component is FIGURE 4
related with stator flux and q stator voltage component is MRAS BLOCK DIAGRAM
related with electromagnetic torque. A. Reference Model
The system control law is proposed in a similar way
The reference model discussed here is adaptative itself. The
 1
Vsd =  KPΨ + KI Ψ eval ( s1 ) rotor winding linkage flux is calculated from a closed loop
 s
(6)
system. This method, analyzed in details in [6], uses two
 1 distinct reference frames to get an improved estimation. The
Vsq =  KPT + KI T eval ( s 2 ) + ω s Ψ
ˆ
s stationary and rotor flux reference frames were used.
 s Using the rotor flux reference frame, the rotor flux
where KPΨ, KIΨ, KPT and KIT are PI gains. The eval function equations are simpler to be described: the d axis is aligned
is implemented as a linear gain with saturation with the rotor flux and receives all the flux contribution, with
 xk ev if lower limit < x < upper limit the q, perpendicular to the rotor flux, receiving no
 (7) contribution. Thus, the rotor flux in rotor flux reference
eval ( x) = upper limit if x > upper limit
lower limit frame can be written as
 if x < lower limit
Ψrdr   Lm (1 + sTr ) ⋅ isdr 
where kev is a constant related to the system dynamics. Ψ = r=
r
 (8)
 Ψrq  
r
The system's state can start outside the sliding surface 0 
but it will be driven in the direction of the sliding surface as
where Lm is the machine magnetizing self-inductance, Tr is
control effort will be produced according to (6) to reduce the
the rotor time-constant and superscript r denotates variable in
errors (4) and to reach S=0. This phase of the process is
the rotor flux reference frame. It is interesting to note that
named the reaching phase.
this model uses only stator currents and some machine
When the system state reaches the S=0 surface and
parameters.
enters in the sliding phase or sliding mode, the control law
Performing a coordinate system conversion of (8) to the
(6) restricts the state to the slide surface S and the system
stationary reference frame, the stator flux can be calculated
actions is governed by the dynamics imposed by S=0 only.
from the rotor flux and stator current as
The system state is not allowed to leave the surface,
ˆ i = Lm Ψ + Lls Llr − Lm i .
2
generating a quick and large control effort to keep the system
state very close to the sliding surface. This intense reaction,
Ψ s r s (9)
Llr Llr
besides producing a very fast response, can also generating
undesired ripple, as a side-effect of limit cycle in the state The superscript i denotates the stator flux calculated
space plane. Using the linear function from Figure 3 and from the stator currents.
with proper gains, the system response is fast and stable. The reference model also calculates the stator flux with
The equivalent control is a fundamental theory in another method, through a feedback system where the stator
variable structure systems that simplify the discontinuous flux estimation Ψs is the feedback variable. A PI
system analysis, replacing discontinuous equations with compensator - represented by (KPΨe + KIΨe / s) - dictates the
continuous equivalents, where its trajectory is the sliding error contribution between the flux estimations (9) and (10)

ˆ = 1 V − r i −  KP e + KI Ψ Ψ ( )  . (10)
e
surface itself. Therefore it is possible to calculate traditional
Ψ  Ψ
ˆ −Ψ
ˆi
s 
coefficients, like damping factor and natural frequency for a s s s s s s
 s 

Coimbra, Portugal September 3 – 7, 2007

International Conference on Engineering Education – ICEE 2007
 FIGURE 5
MRAS REFERENCE MODEL

The superscript e is used to avoid confusion with others linear, and some small impact simplifications and
PI compensators gains. This model uses now stator currents linearizations are made. Thus, after gaining some experience
and voltages. This PI compensator is used to correct pure with the simulation, the students should have an experience
integrator errors and small stator resistance variation through with the real system, to test the controllers they have
the voltage model estimation. designed and simulated, during the experimental activities
The reference model final rotor flux estimation value Ψr using the proposed set-up.
is calculated from stator voltages and currents by The machine parameters used in the simulation are the

ˆ − Lls Llr − Lm i .
2 same got from the real machine and the system operates in
ˆ = Llr Ψ
Ψ (11) closed-loop with speed feedback from tacogenerator or from
r s s
Lm Lm the MRAS estimator.
Typical tests situations of a sensorless induction motor
B. Adjustable Model control include the load torque step change, reference speed
The adjustable or adaptative model equation is simpler and is step reversal, reference speed ramp reversal and low speed
obtained from the current model of the machine equations in operation tests.
stationary reference frame [4] using stator currents and rotor The proposed simulated scenarios shown in this paper
angular velocity covers the following situations: a step change in the speed

ˆ a = (−1 / Tr ) − ωˆ r  ˆ a Lm reference (from 0.5 pu to -0.5 pu) and a step change in torque
Ψ  ωˆ Ψ + i . (from 0 to 0.5 pu).
(−1 / Tr ) r Tr s
r (12)
 r
The results of step change in the speed reference are
presented in Figure 6, where the real speed and estimated
The superscript a denotates the stator flux calculated
speed are shown. Although with some oscillations, the
from the adaptative model.
estimation tracked the real speed very close and it was able
C. Rotor speed estimation to follow the rotor speed within 5% accuracy most of the
time.
With the rotor flux estimation from two methods - the
The stator flux is also estimated and is shown in
voltage model Ψr (reference model) and the current model
Figure 7. During the start up phase, the stator flux grows
Ψra (adaptative model) - the rotor speed estimation ωr can be
from zero to the rated value. The flux magnitude is held
calculated with a PI adaptation mechanism by
constant from this moment on, and this can be verified that
 KI ω 
ω̂ r =  KPω + ⋅e (13)
the stator flux locus in a xy plane is a circle.
 s 
where
e=Ψ
ˆ aΨ
rd
ˆ −Ψ
rq
ˆ aΨ
rq
ˆ
rd (14)
is the cross-error between the adjustable and reference
models.

SIMULATION ACTIVITIES

Simulation tools play a fundamental role in today’s

engineering educational projects. Using a mathematical
model of a given plant, the simulation provides valuable
information about the dynamic behavior of the plant,
avoiding the high costs of equipment and risk of accidents. FIGURE 6
SYSTEM RESPONSE TO THE REFERENCE SPEED STEP CHANGE ±0.5 PU
The three-phase electric machine model is complex and non-

Coimbra, Portugal September 3 – 7, 2007

International Conference on Engineering Education – ICEE 2007
 of the load torque using the voltages and currents produced
by the DC generator.

FIGURE 7
STATOR FLUX DURING THE SPEED STEP CHANGE TEST

FIGURE 9
EXPERIMENTAL SET-UP SHOWING THE ELECTROMECHANICAL SET
(INDUCTION MOTOR AND ITS LOAD) AND THE DRIVING SET

FIGURE 8
SYSTEM RESPONSE TO THE LOAD TORQUE STEP CHANGE 0.5 PU

The step change in the torque, from 0 to 0.5 pu at 50%

rated speed is shown in Figure 8. Operating at no load, the
speed estimation almost matches the real speed; with 0.5 pu
load, the system settle to a new steady state, where the
difference between the real and the estimated speed is less
FIGURE 10
than 2%. INVERTER (LEFT) AND THE CONTROL CIRCUITS TOWER (DSP AND
The learning activities can also include changes in the CONDITIONING CIRCUITS)
controllers’ parameters, detuned estimator gains, changes in
the motor parameters, autoclosing and changes in the The used DSP is a high-performance 32 bits RISC Texas
controller types to see their effects on the system response. Instruments, model TMS320F2812, and runs at 150 MHz. It
is able to do complex calculations in real-time, like
EXPERIMENTAL ACTIVITIES sophisticated speed estimation methods and digital motor
control, operating at a high switching frequency of 10 kHz.
The experimental activities are realized using the set-up The DSP has many built-in peripherals, as a 16 channel AD
proposed in this article. It includes electronic circuits and converter and 2 independent PWM modules, providing a lot
electro-mechanical devices. Figures 9 and 10 display the of resources to implement digital motor control without
built experimental setup during the development phase. many external devices.
The experimental set-up consists of a DSP (Texas Texas Instruments provides a very complete set of
Instruments TMS320F2812) connected to an induction digital motor control foundation libraries, supporting an easy
motor, driven by a 6 kW Semikron three-phase inverter (SKS and stable environment for the implementation of the
27F B6U + B6CI 10V06). algorithms in a fast pace, an appropriate condition to an
The induction motor has a moving frame proper for the educational experiment. Additional routines to control
electromagnetic torque measurement. Load torque can be specific sub-systems as, e.g., the conditioning signals
measured directly in the shaft using another load cell. The circuits, the AD converter and the inverter, were developed
load torque that is the torque available at the shaft can also and are provided. This way the student can focus on the
be calculated as the electromagnetic torque minus the losses control algorithm.
caused by ventilation and friction. The load is a DC machine A three-phase inverter produced by Semikron is driven
operating as a DC generator, delivering the produced power by the DSP. It is a 6 kW inverter, using modules SKS 27F
to a load rheostat. This configuration allows the calculation B6U and B6CI 10V06. The DSP has complete access to the
inverters IGBTs, allowing the engineer to choose between

Coimbra, Portugal September 3 – 7, 2007

International Conference on Engineering Education – ICEE 2007
pre-defined hardware imposed or DSP software controlled CONCLUSION
dead-time. This permits fine control over the quality of the
generate PWM signals, supporting future studies related to The main goal of this paper is to propose a laboratory set-up
unwanted harmonics and EMI effects. for simulation and experimental activities in a three-phase
Conditioning signal boards are necessary to acquire the induction motor direct torque control experiment, offering a
motor state variables from a high level of voltage and current great learning experience about non-linear control to the
to an appropriate voltage level to be sampled and converted students. Different controllers (PI and sliding-mode
by the internal AD converter. Software routines are provided controllers) and a sensorless method to estimate rotor speed
to remove the residual mean value of AC signals. can also be used and their responses can be analyzed. Results
An oversampling technique is used to sample the are very clear, illustrative and proper to prepare several types
voltages and currents. Thanks to the powerful DSP the of learning activities according to the instructor imagination.
sampling frequency can be defined as eight times the The proposed set-up and activities allow electrical
switching frequency (i.e. 80 kHz) and a moving average engineering students (undergraduate and graduate) to
filter with a uniform weight of 1/8 is used to smooth the experience the difficulties and particularities of non-linear
sampled signal. This method has many advantages over systems and non-linear controllers. Using as example the
traditional sampling and is also available to the students. variable speed three-phase induction motor control, there are
Figure 11 shows the speed reversal test, varying between plenty of good situations to be exploited by the faculties and
+30 Hz and -30 Hz each 2s. In this test, the estimator gave the assistant teachers enrolled with the teaching of control
good results, as both signals were almost superimposed; the and electrical machines.
steady state error is below 2%. The Figure 11 also shows that
the breaking is faster than the acceleration, because the ACKNOWLEDGMENT
friction force wasn’t considered in the equations of the motor
dynamic mathematical model, but it wasn’t negligible in this The authors are grateful to FAPESP – the State of São Paulo
machine. Research Foundation – and to CNPq – Brazilian National
In a much similar situation to the presented in Figure 8, Research Council – for the financial support and to Texas
the experimental set-up has produced the output shown in Instruments for the DSP donation.
Figure 12. Although the estimator was not able to track the
real speed closely, the oscillations were minimal. REFERENCES

[1] I. Takahashi and T. Noguchi, “A new quick response and high

efficiency control strategy of an induction motor,” Industry
Applications, IEEE Transactions on, vol. 22, pp. 820–827, 1986.
[2] M. Depenbrock, “Direct self-control (DSC) of inverter-fed induction
machine,” Power Electronics, IEEE Transactions on, vol. 3, no. 4, pp.
420–429, 1988.
[3] Y. Xue, X. Xu, T. Habetler, and D. Divan, “A low cost stator flux
oriented voltage source variable speed drive,” in Industry Applications
Society Annual Meeting, 1990., Conference Record of the 1990 IEEE,
1990, pp. 410–415 vol.1.
[4] C. Schauder, “Adaptive speed identification for vector control of
induction motors without rotational transducers,” Industry
Applications, IEEE Transactions on, vol. 28, no. 5, pp. 1054–1061,
1992.

FIGURE 11 [5] J. Hung, W. Gao, and J. Hung, “Variable structure control: a survey,”
REAL AND ESTIMADED ROTOR SPEED IN STEP CHANGE, ±30 HZ (10 HZ/DIV) Industrial Electronics, IEEE Transactions on, vol. 40, no. 1, pp. 2–22,
1993.
[6] P. Jansen, R. Lorenz, and D. Novotny, “Observer-based direct field
orientation: analysis and comparison of alternative methods,” Industry
Applications, IEEE Transactions on, vol. 30, no. 4, pp. 945–953, 1994.
[7] P. Vas, Sensorless Vector and Direct Torque Control. Oxford
University Press, 1998, ISBN: 0198564651.
[8] V. Utkin, J. Guldner, and J. Shi, Sliding Mode Control in
Electromechanical Systems. CRC Press, 1999, ISBN: 0748401164.
[9] M. P. Kazmierkowski, Control in Power Electronics: Selected
Problems (Academic Press Series in Engineering). Academic Press,
2002.
[10] G. Buja and M. Kazmierkowski, “Direct torque control of PWM
inverter-fed AC motors - a survey,” Industrial Electronics, IEEE
Transactions on, vol. 51, no. 4, pp. 744–757, 2004.
[11] C. Lascu, I. Boldea, and F. Blaabjerg, “Direct torque control of
FIGURE 12 sensorless induction motor drives: a sliding-mode approach,” Industry
SPEED RESPONSE TO A 0.5 PU TORQUE STEP CHANGE AT 30 HZ Applications, IEEE Transactions on, vol. 40, no. 2, pp. 582–590, 2004.

Coimbra, Portugal September 3 – 7, 2007

International Conference on Engineering Education – ICEE 2007