2015 IEEE 10th International Conference on Industrial and Information Systems,ICIIS 2015, Dec.

18-20,2015,Sri Lanka

Low Cost Torque Estimator for DC Servo Motors

A.H. Z Farnaz, H. S. Sajith, P. J. Binduhewa, M. P. B. Ekanayake, B. G. L. T. Samaranayake
Department of Electrical and Electronic Engineering
Faculty of Engineering, University of Peradeniya,
Peradeniya, Sri Lanka
ahzfarnaz@hotmail.com, sajith299b@gmail.com, prabathb, mpb.ekanayake, lilantha@ee.pdn.ac.lk

Abstract- Electromagnetic torque of a DC motor can be It can be measured directly by a torque sensor attached to
measured directly by a torque sensor attached to the shaft. the shaft. However, this is an expensive solution requiring
However, this is an expensive solution requiring additional additional power and interfacing. In addition, sometimes the
power, interfacing and sometimes such a torque sensor can be too
torque sensor can be too bulky to be implemented for certain
bulky to be implemented in a compact environment.
applications.
Alternatively, the armature current measurement can be used to
derive the electromagnetic torque of the DC motor, provided the Another alternative would be to use a current sensor to
motor torque constant is known and does not vary over the time.
measure the armature current and use the simple linear
In addition, this method also needs an additional current sensor
relationship to find the electromagnetic torque. Requiring an
and its integration circuits. However, in general, DC servo
additional current sensor and its integration are the drawbacks
motors are equipped with either absolute or incremental encoder
connected to the shaft. This can be used to estimate the torque
of this approach. In addition, the torque constant can vary
without any additional sensor. with time and therefore requires correction [1]. The most cost
effective approach would be to use the existing shaft
In this paper, a method to estimate the electromagnetic connected position encoder or the tacho-generator feedback
torque of a DC servo motor in real-time is proposed using the and use an estimation technique to fmd the electromagnetic
shaft speed which is derived using the position from the encoder torque of the DC motor.
and the supply voltage measurement to the motor. The dynamic
equations of the linearized DC motor have been used to derive In this study, we propose a method to estimate the
the State Space model. The motor parameters have been obtained electromagnetic torque of a DC motor using the speed
experimentally through a system identification. The identified measurement and the input voltage measurement in real-time.
model is used in the closed loop observer design, where the motor For the input voltage measurement, a simple voltage divider is
speed and electromagnetic torque are the selected states. The used
observer is used to estimate the electromagnetic torque.
A Fuzzy control together with a PID control has been used
A Kalman filter has been used to reduce the estimation errors in [2] to estimate the torque, which performs better than a PID
caused by the presence of process noise and plant noise In controller alone. However this method is not simple to
™
addition to the simulation studies using MATLAB , both the implement in a single microprocessor. Sliding mode observer
closed loop observer and the Kalman filter have been based methods have been used together with extended Kalman
experimentally verified using a DC motor test rig which can filter for electromagnetic torque estimation of nonlinear
easily be replaced with a single microprocessor in the future. brushless DC motors [3] and induction motors [4],[6], which
hints the possible directions this work can be extended.
Keywords- Closed loop observer, DC motor, Kalman filter,
Torque estimation. A sensorless safety power window control system
consisting of a DC motor in an automotive application where a
I. INTRODUCTION Kalman filter is used to estimate the electromagnetic torque can
Despite the introduction of attractive AC motor substitutes, be found in [5]. Optimal feedback control of DC motors using
the DC motor still remains popular among control, robotics and speed and torque feedback can be found in [7].
automation community as an ideal solution for precise control This paper is organized as follows: Section II presents the
using simple controller and simple power electronics. It has a modeling of the DC motor in state space with torque and speed
wide range of applications in automotive, aerospace, material as states and it also gives the closed loop observer design and
processing, manufacturing, etc., industries as well as precise Kalman filter design. The section III describes the
instrumentation. implementation of the torque estimators both in simulations
Typically the DC motor is used for position servo, velocity and hardware. Results have been presented in Section IV and
servo, or torque control applications. Depending on the target the paper is concluded with proposed future work in Section V.
application, the DC motors for control, robotics and automation
II. MODELLING AND DESIGN
applications have been equipped with shaft connected encoder
for position feedback or a tacho-generator for velocity This section presents the modeling details of the motor and
feedback. For torque control applications and fast and precise the design of the closed loop observer Kalman filter for the
state feedback control, it is required to measure the servo motor.
electromagnetic torque of the DC motor.

978-1-4799-1876-8/15/$31.00 ©2015 IEEE 187

 2015 IEEE 10th International Conference on Industrial and Information Systems,ICIIS 2015, Dec. 18-20,2015,Sri Lanka

��"J1 ':�;)l JL;,a LlJ . [Tv(t)L(t)]

[
A. Dynamic model of the DC motor
-
A schematic diagram of a linearized lumped parameter
model of a DC motor is shown in Fig 1, where, o

v(t) - supply voltage (V); L

e(t) - back electromotive force (V); (6)

0] [��;)] 0] [���]
La - armature inductance (H);
Ra - armature resistance (0); +
yet) = [1 [0 (7)
oYt) - motor shaft speed (rad/s);
ia(t) - armature current (A);
Tnlt) - motor electromagnetic torque (Nm);
C. Closed loop Observer design
Tdt) - load torque (Nm);
Em - viscous friction coefficient (kgm2/s); Since the observer is implemented digitally, its design has
Jm - rotor inertia (kgm2). also been carried out in the discrete domain. The state space
model in Section II-B can be discretized with sampling period
h as in (8) and (9);

x[k + 1] = <1>x[k] + ru[k] (8)

+

.�
J � "-�� w
y[k] Cx[k] + Du[k], (9)
Tm(!)
=

eel) where QJ is the discrete system matrix; r is the discrete input

matrix; C [l, 0}, D [O, O}, x[k} is the discrete state vector;
= =

Jm :' y[k} is the discrete output vector; u[k} is the discrete input
0------' Bm ' vector. QJ and r can be obtained by (10) and (11).

(10)
Fig 1 DC motor schematic diagram
r = foh eAS Bds (11)

[
It is assumed that (a) symmetrical armature winding is
present, (b) there is no magnetic saturation, (c) hysteresis and where
eddy current losses are zero, (d) air-gap is uniform, (e) mutual B �
m

1 [
inductance is negligible, and (f) armature reaction is 0
negligible. Hence the linear model of the motor can be A -
- _
Jm
Ke·Kt _
Jm
Ra '
B =
Kt

achieved by equation (1) to (3). Ke and K, are back La La La

electromotive constant and motor torque constant. Substituting For the above discrete time system, a closed loop observer
(2) and (3) in (1) yield equation (4). Applying Newton's second with discrete observer state i and the closed loop observer gain
law of motion to the rotational motion of the motor gives matrix L can be designed as in (12) and (13).
equation (5).
i[k + 1] <1>i[k] + ru[k] + L (y[k] - Ci[k]) (12)
ia t
=

vet) Rai(t) + La d ( ) + e(t) (1)

dt
=
y[k] = Ci[k] . (13)
e(t) Kew(t),
= (2) The estimation error of the close loop observer x given by
Tm(t) Ktia(t)
= (3) (14).
vet) :;
Tm(t) +
= �; d T (t) +
; t
Kew(t). (4)
x[k] x[k] - i[k] ,
= (14)
Tm(t) - TL(t) - Bm(t) = 1m d :�t) (5) which is used to design L, because the error dynamics of the
observer is given by (15)

Hence the armature current measurement is eliminated x[k + 1] = (<1> - LC)x[k]. (15)
from the dynamic equations. This leads to the derivation of the
state space, where the speed oYt) and the electromagnetic To make observer error dynamics to be stable, the
torque Tnlt) will be the states. eigenvalues of should be located within the unit circle of the
B. State space model of the DC servo motor complex plane. Value of observer gain matrix L have been
calculated from the pole placement technique such that the
By selecting the states as above, the supply voltage and the
error dynamics are a decade faster than the system, so that the
load torque as the inputs and the speed as the output, the
derived linearized model equations of the DC motor can be transient response will be smooth. The software
used to obtain the state space model as given in equation (6) implementation of the observer using Simulink™ blocks is
and (7). shown in Fig. 2.

188
 2015 IEEE 10th International Conference on Industrial and Information Systems,ICIIS 2015, Dec. 18-20,2015,Sri Lanka

X'[k)
SPEED 1

TL

SPEE D
1

L
Fig. 2 Simulink model of the closed loop observer

D. Kalmanfilter design
In the presence of plant noise 1J[k] and measurement noise
v[k], the discrete time state space of the DC servo motor plant
gets modified as in (16) and (17).
x[k + 1] ctJx[k] + ru[k] + T][k]
= (16)
y[k] = Cx[k] + Du[k] + v[k] (17)
The Kalman filter has two steps as Predictor step and
Corrector step. Accordingly the Kalman filter algorithm for
the above system can be obtained as in (18) - (22). Fig 3 Simulink model of the Kalman filter

Predictor Step: Time update

Project the state estimates ahead:
x-[k] = ctJx-[k - 1] + ru[k] (18) STM32F4
Discovery
Project the error covariance ahead: De Motor development
kit
r[k] = ctJP[k - l]ctJT + Q (19)
Corrector Step: Measurement Update
Compute the Kalmanfilter gain K:
K[k] r[k]CT(Cr[k]CT + R)-l
= (20)
Update estimate x with measurement: Fig. 4. Block diagram of STM32F4 Discovery board and motor connectivity
x[k] = x-[k] + K[k](y[k] - Cx-[k]) (21)
Update the error covariance P:
P[k] (I - K[k]C)P-[k]
= (22) A. Hardware implementation
The input to the motor is the supply voltage. The position
where, Q = E [1J1JT] is the process noise covariance and output of the shaft is obtained from the optical encoder which
R E[vV] is the measurement noise covariance.
=
comes with the DC servo motor. The supply voltage, which is
The software implementation of the Kalman filter using
in the range of 40 V, is scaled down using the voltage divider.
Simulink™blocks is shown in Fig. 3.
A voltage follower circuit is used to obtain the voltage signal
measurement to the observer and Kalman filter. The observer
III. IMPLEMENTATION OF TORQUE ESTIMATOR and Kalman filter have been implemented in the STM32F4
DiscoveryTM development kit, which is a state of the art low
This section describes simulation and hardware cost controller implementation platform for testing purposes.
implementation used in this study. The first part of the section The STM32F4 Discovery development kit consists of
describes the hardware modules used in this work and the STM32F407VGT6 ARM Cortex-M4 32-bit microcontroller
second part describes the complete Simulink model of the featuring 1 M bytes of flash memory, 192 k bytes of RAM,
simulation. A block diagram of hardware configuration and which enables to port the implementation easily to a standalone
Laboratory setup of hardware implantation used for this study single microcontroller in the future.
shown in Fig 4 and Fig 5 respectively.

189
 2015 IEEE 10th International Conference on Industrial and Information Systems,ICIIS 2015, Dec. 18-20,2015,Sri Lanka

C. Data acquisition platform

For the purpose of observation and recording, the (a)
calculated speed from position measurements, (b) input
voltage to the DC servo motor (c) estimated speeds and (c)
estimated electromagnetic torques from the closed loop
observer and the Kalman filter are transmitted to the host PC.
This is done using PL2303 USB to RS232 TTL converter
using the UART protocol. The Simulink™ model shown in
Fig. 7 was used to receive the transmitted UART signals from
the STM32F4 Discovery development kit. [n this Simulink™
model, a host serial receiver module used to receive UART
signals to the host PC, TORQUE and SPEED scopes used to
Fig. 5 Hardware implementation at laboratory
compare the readings and other four scopes used to observe
the readings separately.

B. Implementation of Simulation using Matlab™ISimulinkTM

In order program the STM32F4 Discovery development
ISerial_Serupl
kit, it is necessary to convert the closed loop observer shown Port:COM5
Baud: 115200 (SoNone·1)
in Fig. 2 and the Kalman filter shown in Fig. 3, which are in
Simulink™ blocks into their corresponding C program code.
Host Seri. Serup
In general this is a tedious process to convert one by one.
Instead "Waijung" is a freely available Simulink™ block set (single) ".I
[Seri.1 Recei""1
that can be used to generate the C code automatically from the Port: COM5 (singlel'I.It---�+-� D
Simulink™ simulation models. Packet: Ascii
Transfer. Blocking
(5ing lel'I.I I----
r-+--------'
-1
Hence the closed loop observer model and the Kalman Baud: 115200 (8·t.b�1e1 'I' TORQUE
filter model were implemented in the Simulink™ environment Ts (secl: 0
(singlel'l.l
together with the "Waijung" Simulink™ block set as shown in
Host Seri.1 Rx
Fig. 6. In this implementation, a regular analog to digital
module is used to measure the supply voltage to the motor and SPEED
a pulse width modulation capture module is used to measure
Fig. 7 Simulink model on the host PC to receive UART signal transmitted
the pulse frequency generated from the encoder. This
from the Discovery.
frequency value is converted to speed. Measuring the load
torque is not feasible. Thus estimated torque from the closed IV. RESULTS AND DISCUSSION
loop observer and Kalman filter were feed in to the respective
Fig. 8 compares the electromagnetic torque estimated
module in real-time by considering the assumption of both
using closed loop observer (green) and the Kalman filter (red)
estimated torque and load torque are same in this method.
with the measured torque (blue), when the load torque of the
A UART transmission module is used to transmit the
motor is changed randomly by external means. For
measured values to the host PC. The closed loop observer
comparison purpose, the torque is measured by connecting a
block and Kalman filter block have been used to estimate
current sensor to measure the armature current and then using
electromagnetic torque developed by the motor.
equation (3). The results show that the electromagnetic torque
estimated using both methods follow the measured torque. The
estimates depend on the speed calculated using the position
measurements. [t is observed that from time to time, for
example at time 2s, 4s and 16.5s in the figure, there exist some
glitches in the torque estimate. As can be seen in Fig. 9, this
Ace Module: 1
originates from the speed.
OutputD�tlIl'1pe: Silgle ANT
l5($ec):O
L-r.=:-L....J""""1lingle
Fig. 9 compares the measured speed (blue) with the
ReguluAOC
estimated speed using the closed loop observer (green) and the
Kalman filter (red) respectively. It can be observed that both
estimates follow the general trends in the speed, but the
limer:2{J2bir:)
Input Pin: AS Kalman filter follows it more accurately. Since the estimates
Edge: Faling
T5(ltc):O
depend on the speed output, it can be anticipated that the
electromagnetic torque estimated using Kalman filter is more
UARTTI

PWMCapture
accurate than the one using the closed loop observer.

Fig. 6 Simulink model embedded on the STM32F4 Discovery board.

190
 2015 IEEE 10th International Conference on Industrial and Information Systems,ICIIS 2015, Dec. 18-20,2015,Sri Lanka

-- Measured
-- Estimated: Closed Loop Observer
0.1
0.25
-- Estimated: Kalman Filter

0.2

E
j
z
0.15

>-

0.1

-D.05 - Error of Closed Loop Observer

- Error of Kalman Filter

1
-D , 0L------'------L---..J.,------!--'-----..J.1
- O,------"
-- 12
2 4 6 10 12 14 16 18 20 Time (s)
Time (s) .
Fig. 10 Comparison of the error between estimated torque and actual torque

Fig. 8 Estimated torque using closed loop observer and Kalman filter.

0.6

0 .4
90

� 0 2.
80 00
:a
� 0
70

� �2
�
60

t �4.
]; 'C
I

50
"""
"
� �6
� 40 ID
C.
(() If) 8
�.
30

20
·1 2.

10
-- Estimated: Kalman Filter 6 10 12
0 Time (5)
0 2 4 6 10 12 14 16 18 20
Time (s)
Fig. II Comparison of the error between estimated speed actual speed
Fig. 9 Comparison of the actual speed with the estimated speed using closed
loop observer speed and Kalman filter.
Similarly the estimation error of the Kalman filter has been
calculated using difference between the estimated speed by
Fig 10 shows the comparison of the calculated error of the Kalman filter and actual speed evaluated as in the previous
Closed Loop Observer (blue), and the Kalman filter (red) case. It can be observed that the speed estimation error of the
respectively. The error of Closed Loop Observer estimation Closed Loop Observer is higher in presence of plant noise and
has been calculated taking the difference between the torque measurement noise. On the other hand, the Kalman filter
estimated using the Closed Loop Observer and torque estimates the speed more accurately.
measured using the current sensor arrangement. Similarly the
error of the Kalman filter estimation has been calculated using v. CONCLUSIONS AND FUTURE WORK
difference between the estimated torque by Kalman filter and A closed loop observer is used to estimate the
the torque measured using the current sensor arrangement. It electromagnetic torque of a DC servo motor using the velocity
can be observed that the torque estimation error of the Closed feedback and the input voltage, which replaces the need of a
Loop Observer is higher in presence the of plant noise and
bulky shaft connected torque sensor or a current sensor with
measurement noise. On the other hand, the Kahnan filter
additional interfacing electronics.
estimates the electromagnetic toque more accurately in
The estimation process is vulnerable to plant and
presence of plant noise and measurement noise. Fig 11 shows
measurement noise. Therefore a Kalman filter has been
the comparison of the calculated error of the Closed Loop
Observer (blue), and the Kahnan filter (red) respectively. The introduced to avoid estimation errors due to noise. Both
estimation error of Closed Loop Observer has been calculated solutions provide torque estimates sufficient for torque
by taking the difference between the estimated speed by Closed feedback and torque control applications.
Loop Observer and actual speed evaluated using the shaft Since the estimation algorithms are simple, they will be
encoder. implemented in a single microprocessor in the future, which is

191
 2015 IEEE 10th International Conference on Industrial and Information Systems,ICIIS 2015, Dec. 18-20,2015,Sri Lanka

proposed to be integrated to the DC motor, so that the servo Applications, IEEE Transactions on , voL42, no.5, pp.1275,1283, Sept.­
Oct. 2006
motor with velocity feedback comes with torque feedback
[4] Pai, D.; Mangsuli, P.R.; Rao, N.J., "Nonlinear observer based sensorless
with minimum additional cost.
direct torque control of induction motor," Power Electronics and Motion
Control Conference, 2000. Proceedings. IPEMC 2000. The Third
ACKNOWLEDGMENT International , yoU, no., pp.440,445 yoU, 2000
The authors would like to thank the Dept. of Electrical and [ 5] Jung Hoon Park; Ga Hyung Choi; Tae Sung Yoon; Jin Bae Park, "A
Electronic Engineering, Faculty of Engineering University of sensorless safety power window control system in automotive
application," Control, Automation and Systems, 2008. ICCAS 2008.
Peradeniya, Sri Lanka for facilitating this project. International Conference on , vol., no., pp.1457,1461, 14-17 Oct. 2008
[ 6] Pai A, D.; Umanand, L.; Rao, N.J., "Direct torque control of induction
REFERENCES
motor with extended Kalman filter," Power Electronics and Motion
Control Conference, 2000. Proceedings. IPEMC 2000. The Third
Je-Wook Park; Seon-Hwan Hwang; Jang-Mok Kim, "Sensorless Control International , yoU, no., pp.132,137 yoU, 2000
[I]
of Brushless DC Motors With Torque Constant Estimation for Home [7] Jiang, J.-P.; Chen, S.; Sinha, P.K., "Optimal feedback control of direct­
Appliances," Industry Applications, IEEE Transactions on , voL48, no.2, current motors," Industrial Electronics, IEEE Transactions on , voL37,
pp.677,684, March-April 2012 no.4, pp.269,274, Aug 1990
[2 ] Liem, D.T.; Ahn, K.K., "DC motor parameters identification and [8] Modern Control Engineering; Katsuhiko Ogata; Fifth Edition; Prentice
sensorless torque estimation using Fuzzy PID," Control, Automation and Hall, 201O.
Systems (ICCAS), 2012 12th International Conference on , vol., no., [9] Control Systems Principles and Design; M. Gopal; Second Edition; Tata
pp.76,81, l7-21 Oct. 2012 McGraw-Hili Education, 2002.
[3] Yong Liu; Zi Qiang Zhu; Howe, D., "Instantaneous Torque Estimation [10] MathWorks; Control System Design and Analysis in Simulink;
in Sensorless Direct-Torque-Controlled Brushless DC Motors," Industry http://in.mathworks.com/products/simcontrol/features.html#control­
system-design-and-analysis-in-Simulink (accessed: 02/03/2015).

192