Proceedings of the 17th World Congress

The International Federation of Automatic Control

Seoul, Korea, July 6-11, 2008

Optimal State Space Control of DC Motor

M. Ruderman, J. Krettek, F. Hoffmann, T. Bertram

Chair for Control Systems Engineering, Technische Universität

Dortmund, D-44221 Germany (Tel. +49/231/755-2496, e-mail:
[email protected])

Abstract:
In comparison to classical cascade control architecture of DC motors, the state feedback control
offers advantages in terms of design complexity, hardware realization and adaptivity. This
paper presents a methodic approach to state space control of a DC motor. The state space
model identified from experimental data provides the basis for a linear quadratic regulator
(LQR) design. The state feedback linear control is augmented with a feedforward control for
compensation of Coulomb friction. The controller is successfully applied and the closed loop
behavior is evaluated on the experimental testbed under various reference signals.

Keywords: model-based control; LQR control method, motor control; state-space models;
feedforward compensation

1. INTRODUCTION earities and uncertainties in the model is compensated by

an integral error feedback signal. The proposed controller
DC motors provide an attractive alternative to AC servo is evaluated for high and low velocity reference profiles
motors in high-performance motion control applications. including velocity reversal to demonstrate its efficiency
DC motors are in particular popular in low-power and high for high-performance servo applications. The proposed
precise servo applications due to their reasonable cost and scheme attempts to bridge the current gap between the
ease of control. Traditionally motor controls in industrial advance of control theory and the practice of DC actuator
applications employ a cascade control structure. The outer systems.
speed and inner current control loops are designed as PD
or PI controllers. However, the cascaded control structure This paper is structured as follows: Section 2 describes the
assumes that the inner loop dynamics are substantially state-space model of the DC motor derived from electro-
faster than the outer one (Chevrel et al. (1996)). mechanical relationships including friction. The model
identification is described in Section 3. Section 4 details
In recent years several publications propose alternative the LQR design of the optimal state feedback control with
approaches to identification and control of DC motors. an integrator and the feedforward friction compensation.
Umeno and Hori (1991) describe a generalized speed Section 5 analyzes the closed performance on the experi-
control design technique of DC servomotors based on mental testbed under different velocity reference profiles.
the parametrization of two-degrees-of-freedom controllers Finally, Section 6 summarizes the major conclusions of the
and apply the design method of a Butterworth filter to paper.
determine the controller parameter. Chevrel et al. (1996)
present a switched LQR speed controller, designed from
the linear model of the DC motor, and compare its 2. DC MOTOR MODEL
performance with a cascade control design in terms of
accuracy, robustness and complexity. Rubaai and Kotaru 2.1 Linear state-space model
(2000) propose an alternative way to identify and control
DC motors by means of a nonlinear control law represented Under the assumption of a homogenous magnetic field,
by an artificial neural network. Yu and Hwang (2004) the direct current (DC) motor is modeled as a linear
present an LQR approach to determine the optimal PID transducer from motor current to electrical torque. The
speed control of the DC motor. classical model of the DC motor, described by Isermann
(2002) is composed of a coupled electrical and a mechanical
This contribution proposes a systematic approach to ve-
subsystem.
locity controller design of a DC motor based on model
identification and LQR design augmented with a nonlinear The angular velocity ω is controlled by the input voltage u
feedforward compensator. The electrical and mechanical with a constant voltage drop attributed to the brush and
parameters of the DC motor, i.e., resistance, inertia, back- rotor resistance, and a back-electromotive force (EMF)
EMF, damping are identified from observations of the caused by the rotary armature. The motor inductance
open loop response. Coulomb friction is considered as contributes proportional to the change in motor current
the main cause of the nonlinear motor behavior and is i. The motor current couples the electrical component
adequately compensated by a feedforward control signal. with the mechanical one, as it generates the driving
The residual steady state error caused by minor nonlin- torque. This torque is antagonized by the motor inertia,

978-3-902661-00-5/08/$20.00 © 2008 IFAC 5796 10.3182/20080706-5-KR-1001.3625

17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

structure damping, friction, and the external load. The

motor dynamics are described by: pareto−optimal solutions
di 10
u(t) = L + Rm i(t) + Ke ω(t) , (1)
dt 9.8
dω

ISE current
Km i(t) = J + Kd ω(t) + τl + τf , (2) 9.6
dt
9.4
where Km , Ke and Kd denote the motor torque, the back- 9.2
EMF and the damping constants. J denotes the mechan-
ical inertia including the motor armature and shaft. L 9
and Rm represent the inductance and the total connection
8.8
resistance of the motor. The system load and friction are 12.542,6 12.542,7 12.542,8 12.542,9
ISE velocity
denoted by τl and τf .
However, for many applications this structure is not suffi- Fig. 1. Pareto-optimal compromises between ISE for cur-
cient. The main drawback of the linear state-space model rent and angular velocity
is a negligence of nonlinear effects, whose properties can
significantly affect the dynamic behavior of a modeled sys- 3. MODEL IDENTIFICATION
tem. To complete the representation of essential physical
phenomena effecting in the motor structure the frictional The signals for identification are generated from the open
nonlinearity must be included. loop step response of the DC motor at different ampli-
tudes. The identification yields the set of optimal parame-
ters that minimize the squared error between model output
2.2 System nonlinearity and data. 
According to Paduart et al. (2006)) the linear state- Eω = (ω(t) − ω̂(t))2 dt
space model with a multivariable nonlinear input function  (5)
f (x(t), u(t)) assumes the general form: Ei = (i(t) − î(t))2 dt .
ẋ = A x + B u + H f (x, u) , (3)
As the model is linear in the unknown parameters, these
in which A is the system matrix, B is the input vector, and are identified by means of least squares. The remaining
the coupling vector H links the nonlinearity with the linear choice is the trade-off between the two errors. This trade-
part. In context of permanent magnet DC motor Coulomb off is specified by their relative weight w in
friction constitutes the major source of nonlinear behavior E = wEω + (1 − w)Ei . (6)
(Knudsen and Jensen (1995)). Additional nonlinearities
emerge from the inhomogeneity of the stator magnetic In the context of state feedback control the model should
field and transfer characteristics of the amplifier and IO not only reflect the input-output behavior but also accu-
elements as well as motor cogging and ripple effects (see rately describe the dynamics of internal states, in our case
Proca et al. (2003)). the motor current.
Tjahjowidodo et al. (2005) describe advanced friction mod- Fig. 1 visualizes the set of pareto-optimal solutions ob-
els which introduce auxiliary internal states to capture tained from variations of the weight w ∈ [0, 1]. The squared
friction dynamics. As these auxiliary states are not observ- error in the angular velocity is rather insensitive to pa-
able customized identification techniques are required for rameter variations as the stick slip effect at low velocities
the identification of their associated parameters. For many causes an oscillation in the angular velocity (see upper left
applications a static friction model that includes Coulomb graph in Fig. 2) not captured by the model. This deviation
and viscous parts suffices to capture the main frictional causes a large offset in squared error compared which the
phenomena. residual error contributions in the rising edge and steady
state are negligible. The compromise solution is marked
The linear viscous friction is already comprehended in the
by an arrow in Fig. 1.
damping term in equation (2). Considering the nonlinear
Coulomb friction which depends on the rotation direction The actual step responses are compared with the model
and introducing the state vector x = [i, ω]T results in: output in Fig. 2 for a subset of six out of sixteen signals

R m Ke
 taken into account for identification. The graphs show
1 0
   
− − that the identified parameters correctly capture the steady
ẋ =  L L x +  L u + 
Fc  sgn(x2 ) (4) state behavior as well as the characteristic time constants

Km Kd  −
− 0 J in the rising edge. The oscillatory behavior at low fre-
J J quencies does not correspond to an eigen frequency of the
system but merely reflects the variation of friction during
in which Fc denotes the Coulomb friction coefficient. The
a complete rotation of the motor shaft.
overall model has six independent parameters, of which
the inductance L = 25 × 10−6 H is obtained from the The identified parameters are listed in Table 1 and com-
manufacturer datasheet and the remaining five parameters pared with the nominal values provided by the manufac-
are identified from experimental data. turer. The differences between the nominal and the iden-

5797
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

is obtained from the solution P of the algebraic Riccati

equation:
AT P + PA − PBR−1 BT P + Q = 0 . (9)

The weight matrices are specified such that the closed loop
system is able to track the reference signal with a control
signal that does not significant violates the saturated
actuator limits. For a fixed weight matrix Q, the control
penalty R is chosen such that for the maximum state error,
the feedback control signal
u = −Kx + V ωr (10)

is in accordance with the actuator bounds. To compensate

the steady state error of the closed control loop a feedfor-
ward term is included in the control:
V = −(CT (A − BK)−1 B)−1 . (11)

To compensate disturbances, and model uncertainties of

the DC model the integral output error
t
ε= (ωr − ωa ) dt (12)
Fig. 2. Comparison of real and model step-responses for 0
velocity and current at different voltages
tified inertia and friction are explained by the additional is introduced as an additional state variable, in which ωr
inertia and bearing of the rotary encoder. The increase of and ωa denote the reference and actual velocities. The
identified resistance is explained by additional contacts of linear part of the state space model is augmented by the
the motor connection and structure changes of the motor auxiliary integral state variable:
        
brushes and commutator. The other identified parameters ẋ A 0 x B 0
are in accordance with the nominal values. = + u + ωr . (13)
ε̇ −CT 0 ε 0 1
Table 1. System parameter identification
Correspondingly the weight matrix is augmented with a
nominal identified small weight for the integral error.
total motor resistance Rm (Ω) 0.35 0.98 100

torque constant Km (N m/A) 0.0296 0.0274 Q= 0 1 0 , R = 10 . (14)
EMF constant Ke (V s/rad) 0.0296 0.0297 0 0 0.001
damping constant Kd (N s/rad) 6.7 × 10−5 7.2 × 10−5
total system inertia J (kg m2 ) 2.9 × 10−6 3.2 × 10−5
4.1 Feedforward friction compensation
Coulomb friction Fc (N m) 0.0200 0.0593
The closed loop behavior is further improved by a feed-
inductance L (H) 25 × 10−6 forward control for immediate compensation of Coulomb
friction. The friction is constant and it sign is opposite
4. CONTROL DESIGN to the direction of rotation. The discontinuity at velocity
reversal is smoothed by replacing the step at ωr = 0 by a
State feedback controller design is accomplished either linear segment for small velocities in the range of σ = ±1
by pole placement or in the context of optimal control rad/s. The add-on feedforward control gain
by means of linear quadratic regulator (LQR) design R m Fc
(see Anderson and Moore (2007)). In pole placement the Kf = , (15)
designer specifies the desired eigenvalues of the closed Km
loop system in the left half plane. LQR design minimizes
determined from equations (1) and (2) by the elimination
a weighted squared state error and control effort. The
of all dynamic terms compensates the static friction phe-
optimal feedback state regulation, minimizes the quadratic
nomenon.
cost function
∞
T Table 2. Controller parameters
x (t)Q x(t) + uT (t)R u(t) ,

J= (7)
friction gain Kf 1.06
0
feedforward gain V 0.3166
in which Q and R are symmetric, positive semi-definite feedback gain Ki 0.0984
respectively positive definite weight matrices. The optimal feedback gain Kω 0.3003
feedback gain
feedback gain Kε −0.01
K = R−1 BT P , (8)

5798
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

4.2 Control law 5.2 Experimental setup

The overall control law including feedforward compensa- The experimental testbed of the DC motor for identifica-
tion tion and control is shown in Fig. 5. The sample rate of
the real time controller onboard the host computer is 5

 Kf sgn(ωr ), if |ωr | > σ
Γ= (16) kHz. The control signal uout with a range of ±5 V applied
ω
 Kf r , else to the DC motor is amplified. The motor is an AXEM
σ DC servo motor with a shrunk-on-disk rotor, F9M2 with
becomes rated power output of 63 W and rated speed 3000 r.p.m.
    (=314.1593 rad/s).
x 1
u = −[Ki,ω Kε ] + [Γ V ] , (17)
ε ωr

with the corresponding gains listed in Table 2.

The entire control structure in state space representation
is depicted in Fig. 3.

Fig. 5. Laboratory testbed: DC motor with a rotary

encoder (left), and system overview (right)
The motor current is measured by the voltage drop uI
across a shunt Rm , and the motor shaft position ϕ and
Fig. 3. Block diagram: proposed control structure with a direction of rotation are provided by a digital single turn
state-space system representation rotary encoder S with 13 bit resolution. The angular
velocity is obtained from time derivation of the shaft
5. CONTROL SYSTEM BEHAVIOR position. A low pass filter with window size 3 is applied
to smooth the velocity signal and reduce the quantization
5.1 Disturbance rejection errors.

5.3 Velocity control

The designed velocity controller is validated on the nom-
inal model in simulation for different reference signals. The tracking behavior of the velocity controller is evalu-
To analyze the robustness of the controller an external ated for a sinusoidal signal with amplitude of 200 rad/s
periodic disturbance torque with amplitude of 0.06 Nm and period of 0.4 s and a sequence of up and down step
and pulse width 0.01 s is applied to motor. The disturbance signals with reference velocities in the absolute range of 5
has the same magnitude as the constant DC motor friction. up to 220 rad/s at intervals of 0.4 s.
3
contol voltage 30 reference The reference and observed velocities as well as simulated
disturbance actual
2 response of the closed loop system are depicted in Fig.
20
6 a) and b). The controller is able to track the reference
velocity (rad/s)

1 10 signal and with no residual steady state error at high as

(V) / (Nm)

0 0 well as low velocities. The closed loop system exhibits a

−10
lag characteristic with a time constant determined by the
−1
slowest eigenvalue attributed to the mechanical subsystem.
−20
−2 To recognize is the overlap in acceleration phase of the
−30 closed loop behavior and the open loop response (by satu-
−3
0 0.5 1 0 0.5 1 rated control signal) depicted in Fig. 6 c), which indicates
simulation time (s) simulation time (s) a maximal achievable controller performance bounded by
actuator properties. Plotting the signals from Fig. 6 at
Fig. 4. Simulation results: torque disturbance and control larger scale in Fig. 7 reveals that the actual velocity oscil-
voltage (left), reference and actual velocity (right) lates with small amplitude around the reference velocity.
This jitter is caused by the limited resolution of the rotary
The controller responds immediately to the disturbance encoder. The amplitude of the jitter corresponds to the
resulting in a rapid compensation of the error in output magnitude equivalent to a single bit and is independent of
velocity as shown in Fig. 4. Notice, that the designed
progress of the angular velocity.
controller is specifically adopted to the unloaded operation
of the DC motor. The identification and control system The control signal saturates during the acceleration and
design has to be repeated if the DC motor is utilized deceleration phases, such that the ability to track the ref-
to drive an permanent external load. If the applied load erence is effectively dominated by the actuator limitations
is not known in advance, an adaptive control scheme is rather than the control. Fig. 8 compares the observed cur-
advocated. rent with the current predicted by the model in case of step

5799
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

exceeded. The high-frequency jitter of the current signal at

a) 200 actual steady state is attributed to the cogging and ripple effects
simulation and the stick-slip motion.
reference
100
velocity (rad/s)

30
a)
0 20

10
−100

current (A)
0

−200 −10

0.5 1 1.5 2 2.5 3 −20

time (s) actual
−30 simulation
b) 200 actual −40
simulation 0.5 1 1.5 2 2.5 3
time (s)
reference
100
velocity (rad/s)

30
b)
0 25 actual
simulation
20
−100

current (A)
15
−200 10
0 0.5 1 1.5 2 5
time (s)
250 0
c)
−5
200 0.4 0.5 0.6 0.7 0.8
time (s)
velocity (rad/s)

150
Fig. 8. Model predicted and measured motor current by
100 the stair shaped reference signal a) and zoom in of
motor current b)
50 closed loop
open loop 6. CONCLUSIONS
0
0.2 0.25 0.3 0.35 0.4
This paper proposes a novel approach to control design
time (s) of a DC servomotor based on system identification and
LQR control design. The feedback controller is augmented
Fig. 6. Experimental results: a) stair shaped signal, b) with a feed-forward friction compensation. The mechanical
sinusoidal signal, c) step response of the open and and electrical parameters of the DC motor are identified
closed loops from the open loop responses with respect to motor current
and angular velocity. The LQR design provides an optimal
state feedback control minimizes the quadratic state error
50 −150
actual and control effort. The auxiliary integral error state and
40 −160
simulation
reference
feedforward compensation of the nonlinear friction reduce
velocity (rad/s)

velocity (rad/s)

30 −170
the residual error across the entire range of reference veloc-
ities. The experimental results demonstrate the feasibility
20
actual
−180 of the controller design for high precision servo applica-
10 simulation −190 tions. The proposed method is well suited for the controller
reference
design of highly dynamic DC motors. Future research is
0 −200
concerned with the design of adaptive controllers to han-
1.59 1.6 1.61 1.62 1.63 1.64 1.45 1.5 1.55 1.6
time (s) time (s) dle variations of load and the identification of periodical
disturbances and advanced friction models.
Fig. 7. Zoom in of angular velocity
ACKNOWLEDGEMENTS
responses. The current peaks in case of abrupt changes in
reference velocities causing large state errors. The actual The authors are grateful to the Kübler GmbH company
peak currents are about two to three times larger than the for providing the sensor. We also thank Heiko Preckwinkel
predicted ones whereas the steady state currents match. for controller implementation and experiments.
We assume that the excess in peak currents is explained
by the initial breakaway force (static friction) that the REFERENCES
motor has to overcome. Nevertheless, these peaks are not B. D. O. Anderson and J. B. Moore. Optimal Control:
critical as the maximal allowed motor current of 60 A is not Linear Quadratic Methods. Dover Publications, 2007.

5800
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

P. Chevrel, L. Sicot, and S. Siala. Switched LQ controllers

for DC motor speed and current control: acomparison
with cascade control. In Proc. Power Electronics Spe-
cialists Conference PESC’96 Record., pages 906–912,
Baveno, Italy, June 1996.
R. Isermann. Mechatronische Systeme. Springer, Berlin,
2002.
M. Knudsen and J. G. Jensen. Estimation of nonlinear
DC-motor models using a sensitivity approach. In Proc.
3. European Control Conference ECC’95, Rome, Italy,
1995.
J. Paduart, J. Schoukens, K. Smolders, and J. Swevers.
Comparison of two different nonlinear state-space iden-
tification algorithms. In Proc. International Conference
on Noise and Vibration Engineering ISMA’06, Leuven,
Belgium, September 2006.
A. B. Proca, A. Keyhani, A. El-Antably, W. Lu, and
M. Dai. Analytical model for permanent magnet motors
with surface mounted magnets. IEEE transactions on
energy conversion, 18:386–391, 2003.
A. Rubaai and R. Kotaru. Online identification and con-
trol of a DC motor using learning adaptation of neural
networks. IEEE transactions on industrial applications,
36:935–942, 2000.
T. Tjahjowidodo, F. Al-Bender, and H. Van Brussel.
Friction identification and compensation in a DC motor.
In Proc. 16th IFAC World Congress, Prague, Czech
Republic, July 2005.
T. Umeno and Y. Hori. Robust speed control of DC servo-
motors using modern two degrees-of-freedom controller
design. IEEE Transactions on industrial electronics, 38:
363–368, 1991.
G.-R. Yu and R.-C. Hwang. Optimal PID speed control
of brush less DC motors using LQR approach. In Proc.
IEEE International Conference on Man and Cybernet-
ics, pages 473–478, Hague, Netherlands, October 2004.

5801