System Identification of a DC Servo Motor Using

ARX and ARMAX Models

Salah-eddine Mokhlis Said Sadki Bahloul Bensassi
Laboratory of Industrial Engineering, Laboratory of Industrial Engineering, Laboratory of Industrial Engineering,
Information Processing, and Logistics Information Processing, and Logistics Information Processing, and Logistics
Faculty of Science Ain Chock, Faculty of Science Ain Chock, Faculty of Science Ain Chock,
University Hassan II. University Hassan II. University Hassan II.
Casablanca, Morocco. Casablanca, Morocco Casablanca, Morocco
[email protected] [email protected] [email protected]

Abstract — System identification is a major preoccupation in the experience in their determination (black box, gray box,
most scientific disciplines. Its use in the robotics field has grown etc.).1
exponentially in recent years, due to the high interest shown to
movement accuracy in the industry. Therefore, this article As mentioned in [5], in system identification of experimental
describes an application of system identification to a DC models, we distinguish between:
Servomotor widely used in robots. The aim of such an
• Non-parametric models where the structure is non-
experiment is to reproduce a mathematical model of the system
using experimental data. To collect the inputs and output and to
defined and the number of parameters is infinite. We
meet the application requirements, we have built a usually represent them on a graph or a table of
microcontroller-based acquisition card. Afterward, we focused numerical values (e.g. step response).
our study on identifying two model structures ARX and • Parametric models where the structure is well defined
ARMAX of the system to test and compare their results based and have a finite number of parameters. The
on validation criterions. Finally, the ARMAX model has
description is given about the system often contain
reproduced a better adjustment to the measurements compared
equations (e.g., transfer function, differential
to ARX model.
equations).
Keywords— System identification, acquisition card, ARX, In this article, Section II highlights the identification of the
ARMAX, PEM, DC Servomotor. parametric models, especially for ARX and ARMAX models.
Section III describes the different steps carried out for the
I. INTRODUCTION
implementation of the application. In IV, we present and
Identifying a process or a system is looking for a discuss the results obtained from the experiment.
mathematical model from a known model class, which
reproduces a static and dynamic output response to input test
signals as close as possible to the real system [1].
According to [2], system identification is about to follow
a specific procedure containing many steps to get a valid
model that describes the system behavior. In chart fig.1, we
cover the experimental organization of this procedure, going
from experiment preparation to validation. The main aim of
the resulting models from this operation can be either to
understand the behavior for study purposes, e.g. [3], or to
build adaptive control loops for the system, e.g. [4].
The notion of a mathematical model of a system, a process,
or a phenomenon is a fundamental concept in physics and
much more in other disciplines. There is a multitude of model
types in literature, each one is for a particular application, but
we can group them into two major classes [2] [5].
A. Theoretical models:
These are physical and mathematical laws based models.
They give a complete description of the system and are used
for the theoretical study and the simulation of the processes.
These models are often complex models difficult to define [6].
B. Experimental models
These are models based on experimental input/output data. Figure 1: Experimental approach for defining a parametric model of a
They can take many forms depending on the involvement of system

978-1-7281-4216-6/19/$31.00 ©2019 IEEE

 II. PARAMETRIC MODELS IDENTIFICATION
( ) ( )
A. Generalities Studied system
The identification of the parametric models is about ( )
estimating the parameters using proper algorithms (e.g., less
square, instrumental variable, maximum likelihood) from ()
sampled process data with the help of a numerical calculator. Model
The process of estimating those parameters from input and
output data can be done with non-recursive algorithms using
offline data or recursive algorithms using online live data from Criterion
the experiment.
For our case, we focused our study on two model structure Figure 2: Prediction Error Method
ARX (AutoRegressive with an exogenous variable), and
ARMAX (AutoRegressive with Adjusted moving Average
and exogenous variable). Many real applications from To use the PEM method, we follow the following steps:
different disciplines use this type of identification, e.g., [7] [8] • Choose the model structure (ARX, ARMAX, BJ, and
[9]. OE).
• Choose the predictor ( | − 1, ):
This model can be defined by [2]: • Choose the Criterion function ( , ) In general, the
( ) criterion to be minimized is given by :
( )= ( )+ ( ) (1) ( , )=ℎ ( , )
( ) ( )
Where:
( ) ( )
( )= ( )+ ( ) (2)
( ) ( )
( , )= ∑ ( , ) ( , ) (9)
Where: Where y(k) is the general representation of the input-
output model described in Fig.2, and ( | − 1, ) is the 1-
( )= 1+ +⋯+ (3) step ahead prediction of the output.
C. Model validation
( )= + ⋯+ (4) The validation of the model is the final step which
consists of checking, validating and comparing the model
( ) = 1+ + ⋯+ (5) estimated with the real model, in general, it is based on the
test of several criterions like those used in this paper [10]:

These models are derived from the generalized transfer = ∑ ( , )( ( , )) (10)

function represented by:
( ) ( )
( ) ( )= ( )+ ( ) (6) = 100 1 − ∑
∑ ( )
(11)
( ) ( )
( )

Where:
III. APPLICATION
( )=1+ + ⋯+ (7)
A DCSM is a DC motor with position angle feedback.
This information is usually in the form of an analogic voltage
( )=1+ + ⋯+ (8) from a potentiometer processed through a controller that
ensures the control of the position according to the instruction
reference it receives from an external member (fig.3). This
B. Parameters estimation
reference must be in the form of a PWM signal. For the
Parameter estimation is a mathematical process, which servomotor used in this application, a time of 20 ms between
consists of estimating the model parameters from the two rising edges must not be exceeded.
observed data; this can be done by minimizing a criterion. However, there is a major problem in achieving a
Therefore it’s necessary to choose the most suitable method satisfactory dynamic performance using these servomotors,
of estimation. In this paper, parameter estimation is done with and this is due to the existence of an integrated circuit that
the prediction error method described in fig.2 and [2], the aim serves as servo-position controllers [11]. Therefore the main
of such method is to minimize the error between the predicted objective will be to estimate the ARX and ARMAX models
system and the measured one. containing all the system components presented in fig.3 to be
able to compare and exploit the results obtained in more
complex systems.
 The table below presents the estimated parameters of the
proposed structures with the calculation of the validation
criterions (Best fit and FPE).
As presented in table bellow, we can conclude that the
ARMAX model provides a better fit of 94.28, a smaller FPE
and lower parameters order than the ARX model who provide
a higher parameters order with lower fit and higher FPE, and
this result is confirmed graphically in the two figures 7 and 8.

Models Best FPE Parameters

fit
(%)

− .
Figure 3: Internal structure of a DC Servomotor. = .
.
To extract reliable input and output data, we have .
developed a microcontroller-based acquisition card, which ARX 74.45 0.37
will be the interface between the computer and the
servomotor. It contains eight analog inputs of 10bits . −
resolution, eight digital inputs, eight digital outputs, and two =
.
PWM outputs. (Fig.4). .

1
= −1.98562856547525
0.985745253359852

ARMAX 94.28 0.36 0

=
0.000116342858167619

1
=
−1.00000001489413

Figure 4: Data acquisition board schema

The acquisition card is linked to the computer through a

serial communication RS232 with a speed of 115200 bit/s to
send the command to system input and receive system output.
In order to know the sampling time Ts of the acquisition
card, which related to the delay caused by instructions
execution time as well as the uncertainty of the internal and
external components of the microcontroller, the analogic Figure 6: Comparison between the estimated model (ARMAX) and
input is subjected to a periodic signal of known frequency, measured output
the samples collected from the card are represented on a
graph and the frequency of this test signal is adjusted until a
similar signal to that of the input is obtained. In the case of
this application Ts = 0.0034 which allows having a speed of
300sps/s.
Next, the analysis and parameter estimation process will
be applied to the collected input and output data.
IV. RESULTS AND DISCUSSION
Based on the experiment on ARX and ARMAX models
for the DC servo motor, we tried to estimate the linear model
as close as possible to the real system. It is done by
calculating over 100 ARX models and 125 ARMAX models
and applying the parsimony principle consisting of choosing
the model with fewer parameters and a higher fit. Figure 7: Comparison between the estimated model (ARX) and measured
output
 V. CONCLUSION Control,» chez American Control Conference, Seattle,
In this article, we were able to carry out a practical 2017.
identification of two simple and realistic autoregressive [5] I. Rolf et M. Marco, Identification of Dynamic Systems
models of the DC servomotor from the real input/output data An Introduction with Applications, Springer, 2012.
of the system. We have developed a microcontroller-based
[6] L. Ljung, «Perspectives on System Identifiction,»
acquisition board, which serves as an interface between the
system and the computer to deal with data collection. A Annual Reviews in Control, vol. 34, n° 11, pp. 1-12,
comparison between the two resulting models was made using 2010.
the validation criterions Best fit and FPE. However, the final [7] R. Sofia, B. Bahloul et N. Benyad, «System
results show that the ARMAX model structure has reproduced Identification of Inventory System Using ARX and
a step response that is close enough to the real system ARMAX Models,» International Journal of Control
compared to the one given by the ARX model. and Automation, December 2015.
In future work, we plan to apply other identification [8] L. C. Sudarshan et B. T. Dhananjay, «System
techniques to more complex systems, which contain nonlinear identification black box approach for modeling
physical behavior to develop proper algorithms for the control performance of PEM fuel cell,» Journal of Energy
loops. Storage, vol. 18, pp. 327-332, 2018.
VI. REFERENCES [9] W. Araydah, T. A. Tutunji et A.-N. Ibrahim, «System
identification for a liquid flow process,» chez Applied
[1] M. VILLAIN, Automatique 2 - Systèmes Asservis Electrical Engineering and Computing Technologies
Linéaires, Ellipses, 1996. (AEECT), Jordan, 2017.
[2] L. LJUNG, System Identification: Theory For The [10] L. Ljung, System Identification Toolbox™ Getting
User, Prantice Hall, 1987. Started Guide, The MathWorks, Inc., 2016., 2016.
[3] J. Miroslav, O. Lucie, J. Boril et J. Rudolf, «Parameter [11] W. Takashi, I. Masato, K. Ryohei, M. Ichiro et S.
identification for pilot behaviour model using the Toshiharu, «Practical Modeling and System
MATLAB system identification toolbox.,» chez Identification of R/C Servo Motors,» chez 18th IEEE
International Conference on Military Technologies, International Conference on Control Applications,
Brno, 2017. Part of 2009 IEEE Multi-conference on Systems and
[4] R. Sandeep, Y. Shih-Yu et T. Tsu-Chin, «Wind Control, 2009.
Turbine System Identification and Individual Pitch