Theoretical and Experimental Implementation of DC Motor Nonlinear

Controllers
D.R. Espinoza-Trejo ∗ and D.U. Campos-Delgado †
∗ Facultad de Ingenierı́a, CIEP, UASLP, espinoza trejo dr@uaslp.mx
† Facultad de Ciencias, UASLP,
Av. Salvador Nava s/n, Zona Univ., C.P. 78290, S.L.P., México
ducd@fciencias.uaslp.mx

Abstract— The development of two nonlinear control strate- includes a PID compensation according with the reference
gies for the velocity regulation of a DC motor are detailed in error and a load torque estimation. This technique is based on
the paper. The parallel (shunt) connection of the DC motor differential geometric control [1], [9]. It is observed that the
is studied. First, a parameter identification was carried out
using experimental input-output data of the motor. One control integral action in conjunction with the torque estimation add
algorithm involves a nonlinear cancelation law (input-output robustness to the control algorithm. Another control method-
linearization) with a PID velocity reference error compensation, ology that has been used extensively for nonlinear systems
and a Luenberger observer to estimate the load torque. In is variable-structure control [4], [11]. The main advantages
this algorithm, the integral compensation and the torque of this algorithm is its robustness against noise and model
estimation improve the robustness of the overall control scheme.
In addition, a variable-structure control (sliding mode) was uncertainty. For comparison in the paper, a variable-structure
developed also for velocity regulation, that uses the information controller is also developed for the velocity regulation of
of the Luenberger observer to include the estimated load torque. the DC motor. The paper is organized as follows. Section
Experimental results in a 2 HP test-bed corroborate the analysis 2 presents the mathematical model of the motor, and the
and designs presented. parameters identification. The control algorithms are detailed
Index Terms— DC motor, variable-structure control, input-
output linearization. in Section 3, and the experimental implementation is shown
in Section 4. The paper ends in Section 5 with final remarks
and conclusions.
I. I NTRODUCTION
Electrical motors are a key piece in almost any automatic
process. They convert the energy from electrical into me- II. DC M OTOR M ODELING
chanical in order to produce movement. There are two basic
types of electrical machines: DC and AC motors [6]. The
DC motors are pretty common in industrial processes due to In the following derivations, consider the states of the
their operational properties and control characteristics. They system as x1 = ia the armature current, x2 = if the field
are used for traction, cranes, mills, etc. According with the current, and x3 = ω the angular velocity. It is assumed
connection between armature and field in the DC motor, three that they are all measured on real-time. The control variable
connection area devised: (i) parallel (shunt) connection, (ii) u = V is the variable DC voltage delivered to the motor.
series connection, and (iii) independent excitation [6],[8]. The parameters of the DC motor are defined as:
The first two have the advantage that they only need one Ra Armature resistance
variable DC power supply to control the motor. However, La Armature inductance
they present the disadvantage that the corresponding control M Mutual inductance
algorithms are more involved, since the mathematical model Rf Field resistance
of both systems are nonlinear [2], [3], [7]. The control of Lf Field inductance
the parallel configuration will be addressed in the paper. B Mechanical friction
In some applications, it is necessary to adjust the motor J Inertia
angular velocity constantly despite the motor load. For this Tl Load torque
purpose, it is needed the appropriate hardware to be able In this study compared to previous ones [1], it is not
to make the adjustments in the input voltage to the motor, neglected La , since for medium to large size motors, this
according with a control algorithm (variable speed drive). parameter could be comparable in magnitude to Ra .
Hence it is necessary to keep the angular velocity of the
motor regulated and adjust the motor electrical torque to
compensate the load. This paper addresses the problem A. Parallel (Shunt) Connection
of velocity regulation with load torque compensation for
the parallel DC motor. In order to achieve this goal, an The mathematical model of this configuration is presented,
input-output linearizing controller is designed [5], [10], that assuming that it is available a variable resistor Radj to adjust
 TABLE I
the maximum velocity in the motor (field weakening) [2]:
DC M OTOR PARAMETERS .
Ra M 1
ẋ1 = x1 − x2 x3 + u − Parameter Value
La La La
Ra 0.699 Ω
Rf + Radj 1
ẋ2 = − x2 + u (1) La 0.297 H
Lf Lf M 2.134
B M 1 Rf 445 Ω
ẋ3 = − x3 + x 1 x 2 − Tl Lf 56 H
J J J J 2.79 × 10−3 kg · m2
B 4.45 × 10−3 N · m/rad/s
Note that Radj has to be selected according to the desired
maximum angular velocity, and the rated maximum field
current in the motor, in order to calculate the maximum
instant. Thus, assuming that the data is sampled at period Ts ,
power dissipated by this resistor. Figure 1 shows the electri-
denote the matrices W (nTs ) and y(nTs ), and the regressor
cal diagram of the parallel (shunt) connection.
Θ in (5), then to compute a solution a summation for N
time instants is used [9]. Therefore, define the summation
+ matrices:
Ra Radj
N
ia
X
RW , W T (nTs )W (nTs ) ∈ R5×5 (3)
La Rf n=1
V
if N
J, B + X
Tl

e Lf Ry , W T (nTs )y(nTs ) ∈ R5 (4)
-

- n=1

ω
RW is always positive semi-definite, but if N is large enough
and under an appropriate excitation of u then RW > 0, and
Fig. 1. Parallel Configuration of DC Motor.
a solution can be deduced Θ = RW −1
Ry . The parameters
identified are shown in Table I.
B. Experimental Identification III. N ONLINEAR C ONTROL S TRATEGIES
First of all, it is needed to obtain an approximation of The proposed nonlinear control strategies followed in this
the parameters of the model in (1), and a least squares paper are detailed next. It is important to mention that this
approximation is carried out [9]. Since the armature electrical control problem is very demanding since the mathematical
and mechanical parameters are the most difficult to estimate, model is nonlinear; There is intrinsically some uncertainty in
the DC motor is considered in a separated armature-field the identified parameters; There is an unknown perturbation
configuration. Hence a constant DC voltage source is con- acting on the system (load torque); and in the test-bed,
nected to the field, without load torque applied to the motor. there are noisy measurements. As a result, a simple PID
Meanwhile, the armature is supplied with a square voltage or other types of linear controllers can not achieve the
signal in order to provide excitation to the system and achieve control objective, and a more complex algorithm must be
the identification. Note that in this configuration the DC mo- pursued. Other model-based control strategies as Lyapunov-
tor presents linear dynamics with respect to the parameters. based design or adaptive nonlinear control could also be
The parameters to identify are (La , Ra , Kb , J, B), where the viable tools, however they are not explored in this work.
electromagnetic constant Kb is related to the constant field
current If and the mutual inductance M , i.e. Kb = M If . A. Nonlinear Cancelation Law
The field parameters Rf and Lf could be identified applying This control scheme consists of three parts:
a step voltage, and measuring the resulting current to identify 1) Input-output linearizing law,
its characteristic time and peak response, or directly by a 2) PID reference compensation, and
LCR Multimeter. The latter approach was adopted in this 3) Luenberger observer for load torque estimation.
paper. Now, a regressor formulation was used to identify the The control block diagram is presented in Figure 2, where
motor parameters using integral relations to avoid derivatives the estimated variables are denoted by (ˆ·). In the next
and improve noise robustness, i.e. subsections, these three parts are detailed.

La
 1) Input-Output Linearization: The control scheme
· R R ¸  Ra  · R ¸ adopted is based on differential geometric methods [5], [10].
ia ia Rω 0 R0 
 Kb  =
 u
(2) For this purpose, the output of interest is the angular velocity
0 0 − ia ω ω  
J 
 0 ω, and it can be proved that the system presents a relative
B degree of two. The relative degree is well-defined if the
condition β(x) 6= 0 is satisfied [7], where β(x) is given
The parameters are then identified from collected data of ia , by
ω and u in order to construct the linear algebraic equations
· ¸
M x1 x2
β(x) = + (6)
in (2). Note that the equations must be satisfied at each time J Lf La
 · R R ¸
ia (nTs ) ia (nTs ) Rω(nTs ) 0 0
W (nTs ) , R ∈ R2×5
0 0 − ia (nTs ) ω(nTs ) ω(nTs )
· R ¸
u(nTs )
y(nTs ) , ∈ R2 (5)
0
¤T
, ∈ R5
£
Θ L a R a Kb J B

ia 3) PID Reference Compensation: A constant reference for

if the angular velocity ωref is assumed. In order to provide
ω good reference tracking, the desired dynamic induced in the
u V
Input-Output
Linearizing Law
Control
Actuator
DC MOTOR law (7) has a PID action:
v
v = ω̈ref + Kd (ω̇ref − ω̇) + Kp (ωref − ω) +
^ B
Z
ω
PID Reference Velocity and Load
Ki (ωref − ω)dt − 2 T̂l (11)
Compensation Torque Observer
J
^
Tl where the estimated torque T̂l is used to cancel this term and
ωref avoid a steady-state error. However, the previous equation
(11) can be simplified, since the reference is assumed to
be constant or to change slowly ω̈ref ≈ 0 and ω̇ref ≈ 0.
Fig. 2. Overall Nonlinear Cancelation Law. Moreover, it is desirable to avoid the derivative of the angular
velocity, so the mechanical equation in the motor is used
instead. Therefore, the resulting PID law is proposed:
Hence, the relative degree is well defined if the sum of
armature and field currents is different from zero. In a
µ ¶
B M 1
v = Kd − ω̂ + ia if − T̂l + (12)
practical setting, this condition is satisfied, assuming that J J J
the DC motor is operated in just one rotating direction. B
Z
Therefore, a linearizing control law is given by Kp (ωref − ω̂) + Ki (ωref − ω)dt − 2 T̂l
J
−α(x) + v where the estimated value of the velocity ω̂ is used in the
u= (7)
β(x) proportional and derivative actions to avoid the noise effects.
for β(x) 6= 0, where v is a desired dynamic added to The selection of the PID gains Kp , Kd and Ki must pursue
the system, and the function α(x) is defined as in (8). that the characteristic equation:
The zero dynamics for this configuration were proved to
λ3 + K d λ2 + K p λ + K i = 0 (13)
be asymptotically stable (minimum-phase system), and the
proof is not included for brevity. had its roots in the left-half plane to provide closed-loop
2) Velocity and Load Torque Luenberger Observer: In stability.
order to compensate the load torque, it is necessary to
estimate this quantity to avoid steady-state error between the IV. S LIDING -M ODE C ONTROLLER
velocity and its reference. A Luenberger observer [9], [10] is
proposed assuming that the load torque is roughly constant On the other hand, a control scheme using a variable-
Ṫl ≈ 0. The observer reproduces the mechanical equation structure philosophy (sliding mode) [4], [11] consists on
in the DC motor with an error correction term due to the defining an sliding surface that reflects the performance
velocity estimation: objectives, and obtaining the equivalent and approximation
control laws. The control block diagram is presented in Fig-
dω̂ B 1 M ure 3, where the estimated variables are again denoted by (ˆ·).
= − ω̂ − T̂l + ia if + l1 (ω − ω̂)
dt J J J In the next subsections, these three parts are detailed. First,
dT̂l consider the following general structure of the mathematical
= −l2 (ω − ω̂) (9)
dt model without external perturbations:
Note that the input variables to the observer (9) are the ẋ = f (x) + g (x) · u (x, t) (14)
angular velocity ω, and the armature and field currents ia , if .
The observer gains l1 and l2 must be selected such that the where for the DC motor (see original model in (1)) f (x) and
characteristic equation: g(x) are given by:
λ2 + (B/J + l1 )λ + l2 = 0 (10)
 R 
− Laa x1 − LMa x2 x3
 1 
La
R
has its roots in the left-half plane, in order to guarantee f (x) =  − Lff x2 g(x) =  L1f  (15)
 

convergence to the real values. −B M
J x3 + J x1 x2
0
 B2 M Rf + Radj M2 2
·
B Ra
¸
α(x) = x3 − + + x1 x2 − x x3 (8)
J2 J J Lf La JLa 2

^ As a result, in order to guarantee the existence of an

TL Velocity and Load equivalent control law, it is required:
2 Torque Observer 3
ω^ ∂σ
· g(x, t) 6= 0 (20)
Approximation + u Control V ∂x
ωref DC Motor
Law Actuator Therefore, for the DC motor in a parallel configuration, the
+ existence condition is given by:
Equivalent ia
if
1 ∂σ 1 ∂σ
Law + 6= 0 (21)
ω La ∂x1 Lf ∂x2
3
· ¸
M x2 x1
=⇒ + 6= 0 (22)
J La Lf
Fig. 3. Overall Variable-Structure Control Strategy.
C. Approximation Law
Now, consider the following approximation control law
Therefore, a general control structure following the sliding [4], [11]:
mode methodology is given by: ½ ¾−1
∂σ
uR (x, t) = · g(x) · ûR (23)
u (x, t) = ueq (x, t) + uR (x, t) (16) ∂x
where,
where ueq denotes the equivalent control law, and uR the 1/2
ûR = −K |σ| sign(σ) K>0 (24)
approximation law [4].
In the last equation, the parameter K will affect the speed
A. Sliding Surface convergence of the trajectories to the sliding surface. Note
The sliding surface specifies the desired characteristics of that the approximation law (24) will be large if the system
the control system as: stability, tracking, regulation, etc. It is is far away from the sliding surface. In a practical imple-
then proposed an sliding surface for velocity regulation: mentation, there exists noise in the measurements, therefore
to reduce the chattering phenomenon a boundary layer [12]
σ = γe + ė = 0 (17) approach can be followed for the approximation control law.
Hence, the control (24) is modified as
where e = x3 − ωref denotes the regulation error and ωref ³σ´
1/2
the velocity reference. Note that when the surface has been ûR = −K |σ| sat (25)
²
reached (σ = 0), the parameter γ define convergence speed
where the parameter ² defines the size of the boundary layer.
of the regulation error to the origin. Then, substituting the
state ẋ3 in (17), and considering that the reference velocity V. E XPERIMENTAL I MPLEMENTATION
is a constant (ω̇ref = 0), it is obtained the sliding surface The control diagrams shown in Figures 2 (nonlinear can-
as: celation law) and 3 (variable-structure control) were imple-
mented experimentally in a dSPACE DS1103 system running
µ ¶
B M 1
σ (x) = γ − x3 + x1 x2 − Tl − γωref (18) at a sampling frequency of 10 kHz. The control parameters
J J J
for both strategies are presented in Table II. The test-bed
Therefore, in order to avoid the velocity derivative in the shown in Figure 4 was used, and it consists of a 2 HP Shunt
sliding surface, it is needed the load torque Tl information DC Motor that is connected to a 2 HP Permanent Magnet
in σ(x). DC Motor utilized as a load. A tacogenerator measures the
angular velocity of the shaft, at a proportion of 50 V/RPM
B. Equivalent Law with an error of ±10 %. There are measurements of the
The equivalent control constitutes a control input which, armature and field currents through hall-effect sensors. It
when exciting the system, produces that the resulting trajec- is important to mention that the three measurements are
tories remain on the sliding surface whenever the initial state noisy during the experiments, as it will be observed in the
is on the surface [4]. Based on the existence of a sliding implementation plots, and this issue presents a challenge for
mode, and using the chain rule, it is define the equivalent the control algorithm to show good robustness.
control ueq for systems of the form (14) as: The motor voltage V is controlled by DC-DC chop-
per working under a PWM scheme (switching frequency
10kHz), where the control parameter is the duty cycle u
½ ¾−1 ½ ¾
∂σ ∂σ
ueq = − · g(x) · · f (x) (19)
∂x ∂x [6]. The chopper was selected as control actuator due to its
 three-phase
rectifier bridge
transformer

+ Vcd -
Three-phase
Voltage AC
u DC-DC Chopper Source

control signal output

+ V
-
R
field-armature






L 










     

connection
     

     

     

F2 F1 hall effect
current sensors
          
A2
       
A1       

                         

                         

                         

T1 + 3 HP DC Motor Load
T2
-
Shunt DC Motor (permanent
Torque
tacogenerator
ω magnet)

Fig. 4. Experimental Configuration of the DC Motor.

TABLE II
extremely noisy. Nevertheless, the controller is capable of
C ONTROL PARAMETERS .
providing good tracking for a reference change. Due to this
Parameter Value change, the control signal decreases its value to compensate
Nonlinear Cancelation Law the lower reference. Also, the armature and field currents
Kp 227.24 update their value to accommodate the reference modifica-
Ki 932.46
Kd 24.62 tion. Similarly, for the variable-structure control law good
Load Torque Observer tracking is observed, and the plots are omitted due to space
l1 30.37 limitations. Finally, in Figure 6 the experimental results for
l2 512.15
Variable-Structure Control
Case B are illustrated, but now for the variable-structure
γ 150 control law. The top plot shows that the angular velocity is
K 250 correctly compensated, since only a transient effect is noticed
² 1.0
after the load change. On the other hand, the armature and
field currents increase due to the load effect, and recover
their previous value when the load is again removed. Similar
fast response and linear dynamics, and it was just modeled as results are derived for the nonlinear cancelation control
a scaling factor in the control system. The construction of the law. Consequently, both techniques show good performance
actuator was carried out in our lab, and it is designed such against a reference change and load perturbations.
that it is controlled by a voltage signal u in the interval [0, 10]
V. This saturation in the control signal u did not limited the VI. C ONCLUSIONS AND F INAL R EMARKS
performance of the system, due to the fast dynamics of the
In the present paper, a theoretical derivation and practical
actuator related to the time constants of the DC motor. The
implementation of two nonlinear control schemes for a
adjustment resistor during tests was not used (Radj = 0 Ω),
shunt DC motor are detailed. The experimental identification
since the open-loop maximum velocity was adequate and
proposed, with integral relations, was capable to provide
there was no need to apply field-weakening. Two tests were
good estimates of the real parameters. A nonlinear control
carried out for the control algorithm:
algorithm is designed departing from input-output lineariza-
Case A: tracking of an angular velocity reference change tion theory (differential geometric tools). This technique
from 1400 to 1200 RPM’s (see Figure 5). is recognized to have robustness issues during practical
Case B: load torque variation from 0 N m (no-load) to implementations. For this purpose, integral action to the
1.5 N m, and back to 0 N m (Figure 6). reference error and load torque estimation were added to
In Figure 5 the results for Case A are presented with improve the robustness of the overall structure. Despite
the nonlinear cancelation control. It can be observed that noisy measurements during the practical implementation, the
the tacogenerator measurement for the angular velocity is control scheme is able to adjust the motor voltage properly
 1450 1350

NO LOAD TORQUE Tl=1.5 N m NO LOAD TORQUE

1400 1300

1350 1250
ωref=1200 RPM
ω (RPM)

ω (RPM)
ω =1400 RPM ωref=1400 RPM
1300 ref 1200

1250 1150

1200 1100

1150 1050
0 5 10 15 20 25 0 5 10 15 20 25
time (seconds) time (seconds)

9.5 7.8
7.6
9
7.4
7.2
8.5
u (duty cycle)

u (duty cycle)
7
8 6.8
6.6
7.5
6.4
6.2
7
6
6.5
0 5 10 15 20 25 0 5 10 15 20 25
time (seconds) time (seconds)

8 8

7.5 7

7 6
REFERENCE UPDATE
6.5 5
Tl=0 N m Tl=0 N m
6 4
ia (A)

ia (A)

5.5 3

5 2 Tl=1.5 N m
4.5 1

4 0

3.5 −1
0 5 10 15 20 25 0 5 10 15 20 25
time (seconds) time (seconds)

0.3 0.195

0.19
0.28
0.185
0.26
0.18
0.24 0.175
if (A)
if (A)

0.22 0.17

0.165
0.2
0.16
0.18
0.155

0.16 0.15
0 5 10 15 20 25 0 5 10 15 20 25
time (seconds) time (seconds)

Fig. 5. Experimental Response for Case A with Nonlinear Cancelation Fig. 6. Experimental Response for Case B with Variable Structure Control
Control (TOP) Angular Velocity Measurement, (MIDDLE 1) Control signal (TOP) Angular Velocity Measurement, (MIDDLE 1) Control signal (duty
(duty cycle), (MIDDLE 2) Armature Current, (BOTTOM) Field Current. cycle), (MIDDLE 2) Armature Current, (BOTTOM) Field Current.

to follow a variable velocity reference and to compensate [4] J.Y. Hung, W. Gao and J.C. Hung, “Variable Structure Control: A
load torque changes. On the other hand, a variable-structure Survey”, IEEE Transactions on Industrial Engineering, Vol. 40, No.
1, pp. 2-20, 1993.
control was also proposed. The resulting control algorithm [5] A. Isidori. Nonlinear Control Systems. Springer-Verlag London Lim-
produced good performance against perturbations and noise ited, (1995).
measurement. As a result, both techniques are visualized as [6] R. Krishnan. Electric Motor Drives: Modleing, Analysis and Control.
Prentice-Hall, Upper Saddle River, NJ, (2001).
practical tools in this nonlinear setup. [7] P.D. Oliver. “Feedback Linearization of DC Motors”. IEEE Transac-
tions on Industrial Electronics, 38(1991), 498–501.
ACKNOWLEDGMENTS [8] W. Leonhard. Control of Electrical Drives. Springer-Verlag Berlin
(2001).
This research was supported in part by a grant from [9] S. Mehta and J. Chiasson, “Nonlinear Control of a Series DC Motor:
PROMEP. Diego Espinoza-Trejo acknowledges the financial Theory and Experiment”, IEEE Transactions on Industrial Electronics,
aid provided by CONACYT through a doctoral scholarship. 45(1998), 134–141.
[10] S. Sastry. Nonlinear Systems: Analysis, Stability, and Control.
R EFERENCES Springer-Verlag New York, Inc. (1999).
[11] V. I. Utkin, “Sliding Mode Control Design Principles and Applications
[1] M. Bodson and J. Chiasson, “Differential Geometric Methods for to Electric Drives,” IEEE Transactions on Industrial Electronics, vol.
Control of Elecrtical Motors”, International Journal of Robust and 40, No. 1, 1993.
Nonlinear Control, Vol. 8, pp. 923-954, 1998. [12] K.D. Young, V.I. Utkin and U. Ozguner, “A Control Engineer’s Guide
[2] J. Chiasson and M. Bodson. “Nonlinear Control of a Shunt DC Motor”. to Sliding Mode Control”, IEEE Transaction on Control Systems
IEEE Transactions on Automatic Control, 38(1993), 1662–1666. Technology, Vol. 7, No. 3, pp. 328-342, 1999.
[3] J. Chiasson. “Nonlinear Differential-Geometric Techniques for Control
of a Series DC Motor”. IEEE Transactions on Control Systems
Technology, 2(1994), 35–42.