Вы находитесь на странице: 1из 16

Original Article

International Journal of Electrical


Engineering Education
A flexible robotic 2016, Vol. 53(4) 341–356
ß The Author(s) 2016
control experiment for Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav

teaching nonlinear DOI: 10.1177/0020720916631159


ije.sagepub.com

adaptive control
Bidyadhar Subudhi1 and
Santanu Kumar Pradhan2

Abstract
Teaching a subject on system identification and adaptive control to Master’s students is a
challenging task. In view of providing a platform to achieve the teaching-learning on
adaptive control effectively, this paper proposes a real-time robotic adaptive control
experiment. The robotic system consists of flexible links which is subjected to varied
payloads. The link flexure of this robot and other parameters conditions are uncertain
thus necessitating an adaptive control to handle the payload variations. A simple regres-
sor-based adaptive control is designed by linear parametric representation of a highly
nonlinear and coupled dynamics this robot with flexible links (flexible robot). To rein-
force the student’s concept of design, development, and implementation of an adaptive
control, experiments were undertaken in real-time for different payloads. Attempts
have been made to make the students’ understanding more clear how an adaptive
controller performs well for situations with varied parameters which is not possible
to obtain by a fixed gain controller.

Keywords
Adaptive control, regressor-based adaptive control, flexible robotic control

Introduction
It is obvious that unlike a fixed gain controller an adaptive controller adapts its
gains, which in turn provides appropriate control action in response to the changes
happening in dynamics of the plant. However, many real-world systems such as

1
Department of Electrical Engineering, National Institute of Technology Rourkela, Rourkela, India
2
National Institute of Science and Technology, Berhampur, 761008 India
Corresponding author:
Bidyadhar Subudhi, Department of Electrical Engineering, National Institute of Technology Rourkela, Rourkela
769008, India.
Email: bidyadhar@nitrkl.ac.in
342 International Journal of Electrical Engineering Education 53(4)

power system, robotic system, etc. are nonlinear and are subjected to parametric
variations.2 One of the simplest approaches to design an adaptive controller is
based on a regressor model.
A flexible robot (FR) is different than a rigid robot in the construction, i.e. its
links are thin and longer thus offers several advantages over rigid-link robots such
as achieving high-speed operation, lower energy consumption, and increase in pay-
load-carrying capacity.1 These robots find applications requiring large workspace
like assembly of free-flying space structures and hazardous material management
from safer distance.3 However, owing to the distributed link flexure and parametric
variations, fixed gain controller may not provide desired trajectory tracking per-
formance. But the advantages and applications offered by flexible robots motivate
towards designing adaptive controllers such that in spite change in the dynamic
behavior of the robot accurate tip position tracking can be achieved.
In order to achieve good tip trajectory tracking while suppressing tip deflection
with varied payloads, adaptive control should be employed, which can provide
appropriate control torques to the actuators to achieve the above two-control
tasks (good tip trajectory tracking and suppression of tip deflection). Most of the
courses on adaptive control system are related to linear model of the system to
design an adaptive controller.3 But in reality, the plant dynamics in many physical
systems such as the flexible robot is nonlinear. Further, the motivation of this paper
is to design an experimental platform for teaching adaptive control course to Master
degree students in control and automation. The adaptive controller is a nonlinear
regressor-based model adaptive controller (RBAC), which is designed to capture the
change in FR dynamics due to change in payload. As the manipulator is expected to
maneuver with unanticipated payload at the end effector, thus payload variability is
also an important concern. Furthermore, due to sudden change in payload, there
may be large variation in manipulator parameters and that in turn adds further
complexities to the FR dynamics. Hence, the torque applied to the actuators of an
FR to control the tip position and its deflection with changes in payload should be
adaptive in nature. To adapt to the varied payload situations, the controller par-
ameters are tuned adaptively.4 Thus, through the experiment, the students feel the
necessity of adaptive control, situation where it is needed and how to design a
regressor-based adaptive control to cope up with the changing dynamics.
This paper proposes a flexible robot lab project for learning nonlinear model
based adaptive control using a regressor-based nonlinear adaptive control for post
graduate students. The goal of this project is to make students aware of the import-
ance of nonlinear-based adaptive control systems, and to equip them with the
knowledge to design a parametric representation of any nonlinear complex coupled
model. A sample project solution is offered for a variable payload task with a
flexible robot using Simulink platform. This experiment-based learning approach
was adopted for a Master’s degree course in the National institute of Technology
(NIT), Rourkela, India, and its results analyzed in comparison to a fixed gain PD
controller. In this paper, we are going to describe how an experimental study can be
exploited to demonstrate the student’s easy understanding of the subject.
Subudhi and Pradhan 343

The paper is organized as follows. Forthcoming section presents the detail


design of the flexible robot project. Next, the dynamics of the flexible robot is
described and the experimental set up is discussed. Later, a regressor-based adap-
tive controller applied to the FR is presented and the real-time build up process is
given. Finally, project results are discussed and a remark on students’ feedback is
provided. The paper concludes in the last section.

Detail design of the flexible robot project


Experiment on studying adaptive control
The major motivation for this work is to provide a practical approach to demon-
strate the students about adaptive controller design in real-time for a robotic
system whose dynamics change due to parameter variation and payload variation.
Adaptive control design requires insight into the state of the system, so that experi-
ment-based learning for an appropriate practical exercise for MTech students.
Before describing about the experiment a pedagogical design is undertaken.
First, a through survey from the student’s aptitude on adaptive control was done.
In the past, several papers on design of adaptive controllers for FRs with variable
payloads have been reported. A simple decoupled adaptive controller comprising
the estimation of link’s natural frequency for a single-link flexible manipulator
under variable payload is proposed in Moudgal et al.5 Further advancement
adaptive controller has been proposed in Gutierrez et al.,6 where a discrete-
time nonlinear adaptive controller for a single-link flexible manipulator using
RLS-based payload estimation is used. Intelligent controllers based on supervised
learning using neural networks7 and fuzzy logic have been designed by some
investigators8 for FLMs under parametric uncertainty. However, neural net-
work-based controllers require training of the synaptic weights to an optimal
value which consume considerable amount of time and computational complex-
ity. Fuzzy logic-based adaptive controller design depends upon proper formula-
tion of control rule base. A hybrid neuro-fuzzy-based adaptive controller has
been proposed in Pradhan and Subudhi.9 Although, the above hybrid neuro-
fuzzy controller shows better performance compared to neural network and
fuzzy logic-based adaptive controllers but it needs a priori information about
the input–output relationship, i.e. supervised and offline learning are essentially
required. Hence, a simple regressor-based nonlinear adaptive control is chosen
for a FR working under payload variation. Second, the number of assume modes
to be chosen for the dynamic model of the FR is a critical issues as it adds to the
complexity.

Learning pedagogy
A robotic control is chosen because it has complex nonlinear dynamics with para-
metric uncertainty. The control objective is to achieve both tip trajectory tracking
344 International Journal of Electrical Engineering Education 53(4)

and suppression of tip deflection while handling variable payloads. The learning
outcomes from the experiments are the following:

. Controlling a complex nonlinear system handling variable payload;


. Able to design, implement, and test a nonlinear model-based adaptive controller
for a given system.

Flexible robot dynamics


The schematic diagram of a planar FLMS is shown in Figure 1, where  1 and  2 are
the actuated torque of the joint-1 and joint-2 respectively, d1(l1, t) represents the
deflection along link-1 and d2(l2, t) represents the deflection along link-2. The outer
free end of the FLMS is attached with payload mass, Mp.
The dynamic equation of the FLMS is derived using the standard Lagrangian
approach.5 The total Lagrangian (L) is difference of total kinetic energy (KT) and
total potential energy (UT) of the FLM system. The total KT of the can be expressed
as (Total kinetic energy due to joints) þ (Total kinetic energy due to links) þ (Total
kinetic energy due to payload Mp) and in absence of gravity. The links are modeled
as Euler–Bernoulli beams with deformation d1(l1, t) and d2(l2, t) for link-1 and link-2
respectively satisfying the link partial differential equation given as

@4 d1 ðl 1 , tÞ @2 d1 ðl 1 , tÞ
ðEI Þ1 4
þ 1 ¼0 for link-1 ð1Þ
@l 1 @t2

@4 d2 ðl 1 , tÞ @2 d2 ðl 2 , tÞ
ðEI Þ2 4
þ 2 ¼0 for link-2 ð2Þ
@l 2 @t2

Figure 1. Schematic diagram of a planar FLMS.


Subudhi and Pradhan 345

Applying proper boundary conditions in order to solve these partial differential


equations, the boundary conditions used are clamped-free, clamped inertia, and
pinned inertia. Also, by applying the finite-dimensional model of link-flexibility
using assume mode method.5 The link deflection for link-1 and link-2 can be writ-
ten as

d1 ðl 1 , tÞ ¼ ’11 ðl 1 Þ11 ðtÞ þ ’12 ðl 1 Þ12 ðtÞ for link-1 ð3Þ


d2 ðl 2 , tÞ ¼ ’21 ðl 2 Þ21 ðtÞ þ ’22 ðl 2 Þ22 ðtÞ for link-2 ð4Þ
where ’11 represents mode shapes (spatial coordinate) of the link-1; ’12, mode
shapes (spatial coordinate) of the link-2; ’21, modal coordinates (time coordinate)
of the link-1; ’22, modal coordinates (time coordinate) of the link-2.
To formulate the dynamic model of the FLMS the finite expression using equa-
tions (1) to (4) for link deflection along with proper boundary conditions as des-
cribed in Moudgal et al.5 The Lagrangian–Euler dynamic equation can be written as

d @L @L
 ¼ ð5Þ
dt @q_ @q

where  is the control torque applied; q is the generalized coordinates; L is KT  UT.


In equation (5), we put the value of total kinetic energy (KT) and total potential
energy (UT) of the FLM system and solve for q generalized coordinates i.e. (1, 2,
11, 12, 21, 22). As a result we can write the dynamic model in matrix form as
 " #  " #  " #  
Mrr Mrf q€r Crr Crf q_ r 0 0 qr 
þ þ ¼ ð6Þ
Mfr Mff q€f Cfr Vff q_ f 0 Kff qf 0

where r, f is the rigid-mode and flexible-mode part of the FLMS, M is the positive
definite symmetric inertia matrix; C is the matrix containing of Coriolis and centri-
fugal forces; Kff the Stiffness matrix ¼ diag{kff1, kff2, kff3, kff4}; qr ¼ 1, 2; qf ¼ 11,
12, 21, 22.
The dynamic equation of the FLMS in equation (6) is used to design and imple-
ment a direct adaptive control. Let a be the parameter in the FR dynamics. Then a
linear parameterization of the robot dynamics of equation (1) can be made using
the new output (2) as

 i ¼ ai YTi ð7Þ

where Yi is the regressor matrix for ith link.

Experimental setup
The robot control set-up (Quanser Make) used to conduct adaptive control
experiments.8
346 International Journal of Electrical Engineering Education 53(4)

Figure 2. Interfacing of hardware and different components of the robot set-up.

Hardware
It has two links and two joints with an end effecter to carry payload. These two
joints are excited by two DC brush motors powered with  15 V DC power supply.8
The DC drives have gear ratios of 100:1 and 50:1, respectively.
Digital encoders are used to measure the angular positions of the each joint and
two stain gauges are used to measure the link deflection. A/D and D/A signals are
processed using an in built hardware-in-the-loop (HIL) board.

Software
Figure 4 shows the considered robotic control set-up which works on MS
Windows operating system with the Matlab/Simulink. The interfacing of different
hardware and software is components are shown in Figure 2. There is an arrange-
ment available in the set-up to add additional payload to the robot as shown in
Figure 3.

Regressor-based adaptive controller


Adaptive controller design procedure
The design of an adaptive controller usually involves choosing a control law with
tunable parameters, and then an adaptation law is to be developed using the closed-
loop error dynamics. The objective here is to develop an adaptive control law such
that even there is a change in robot dynamics due to payload variation the good tip
trajectory tracking is achieved.
Thus, in order to develop an adaptive controller the dynamics in equation (3) is
rewritten as a closed-loop error dynamics
  T
i ¼ ~ _ y_ r, y€ r,
a Y y, y, ð8Þ

where
 T
yr ¼ y_  e and y ¼ y1 , y2 ð9Þ
Subudhi and Pradhan 347

Figure 3. Block diagram of the experimental set-up showing hardware and payload
arrangement.

where
 Positive definite matrix.
e Tip trajectory error (yd  y).
ã Parameter estimation error.
With parameters a defined as
a1 Link-1 inertia.
a2 Link-2 inertia.
a3 Joint-1 inertia.
a4 Joint-2 inertia.
a5 Total (hub þ link)inertia of link
a6 Total mass of link (including the payload)
a7 Total coupling mass
The adaptive control law for this robotic system is derived as follows8
h  iT
¼~ _ y_r, yr,
a Y y, y, KD s ð10Þ

where

s ¼ e_ þ e ¼ y_ þ yr ð11Þ
348 International Journal of Electrical Engineering Education 53(4)

and KD is the positive definite matrix. The parameter adaptation rule is given by

a_^ ¼ YT s ð12Þ

where  is a symmetric positive definite matrix. Robot dynamics described by


equation (1) with unknown parameters a. A nonlinear adaptive control law
given in equation (11) with the adaption law in equation (12) will provide the
desired tip trajectory tracking under payload variation. To prove this, we define
a Lyapunov candidate function, V(t) as
1 
VðtÞ ¼ sT Ms þ a~ T 1 a~ ð13Þ
2
where M is the inertia matrix in equation (1) and  is a symmetric positive definite
matrix. Differentiating equation (13) with respect to time leads to
  1
€ þ a~ T 1 a_~
V_ ðtÞ ¼ sT M€y  M€yr þ sT Ms ð14Þ
2

Substituting for M€y from the robot dynamics (1) and using the linear param-
eterization of the robot dynamics in equation (6), one obtains
 
V_ ðtÞ ¼ sT   aYT þ a~ T 1 a_~ ð15Þ

Substituting the adaptive control law  ¼ aY ^ T  KD s in equation (11), we obtain


 
V_ ðtÞ ¼ sT a^ YT  KD s  aYT þ a~ T1 a_~
 
¼ sT ða  a^ ÞYT  KD s þ a~ T 1 a_~
  ð16Þ
¼ sT a~ YT  KD s þ a~ T 1 a~_
h i
¼ sT KD s þ a~ T 1 a_^ YT s

a_ ¼ YT s is used which leads to


The parameter adaptation rule ^

V_ ðtÞ ¼ sT KD s  0 ð17Þ

It can be seen from equation (17) that for some positive values KD, the error
functions s converge to zero. The structure of the adaptive controller given by
equations (11) and (12) is sketched in Figure 4. An explicit form of the control
law for the ith link in terms of parameter vector a is given by

1 ¼ a1 Y11 þ a3 Y12 þ a4 Y13 þ a5 Y14 þ a6 Y15 þ a7 Y16  KD s1


ð18Þ
2 ¼ a2 Y21 þ a3 Y22 þ a5 Y23 þ a6 Y24 þ a7 Y25  KD s2
Subudhi and Pradhan 349

Figure 4. Adaptive control structure.

where Yij are the ijth element of regressor vectors for i ¼ 1, 2 and j ¼ 1, 2, . . . , m. The
adaptation law can be explicitly written as

^_ 1 ¼ 1 Y11 s1
a ^_ 4 ¼ 4 ðY13 s1 Þ
a
a_ 2 ¼ 2 Y21 s2
^ a_ 5 ¼ 5 ðY14 s1 þ Y23 s2 Þ
^
ð19Þ
a_ 3 ¼ 3 ðY12 s1 þ Y22 s2 Þ
^ a_ ¼  ðY s þ Y s Þ
^6 6 15 1 24 2

a_ ¼  ðY s
^7 7 16 1 þ Y25 s2 Þ

The feedback gain matrix KD and the adaptation gain matrix  are diagonal as
defined below
 
 ¼ diag 1, 2, 3, 4, 5, 6, 7, and KD ¼ diagðKD1 , KD2 , Þ

Real-time implementation procedure


. Ensure that the power amplifier, external DC source, and sensor input switches
are on.
. Ensure that all encoders and strain gage sensors of the robotic system are work-
ing properly before preceding the Simulink model build up process.
. Open the Simulink file named naldpt.mdl which is shown in Figure 5. Align the
two flexible links in their central position to start from the zero degree initial
positions.
. Run the setup script file called setup_2DSFL_robot.m prior to running the
model as this file sets the required parameters for proper operation of the
Simulink model and the hardware itself into the workspace of the current
directory.
. The Matlab code for adaptive controller is built up in the real-time set up by
using the real-time target logic code in C language. For this click on the top
Quarc button and slide down to tab called build, press build to start the model
build up process. The screen shot of the above procedure is shown in Figure 6.
. After the successful build process, a message that successfully target code is built
in the command prompt will appear. Next, connect the built model to the hard
ware set-up by clicking on connect to target button as shown in Figure 7.
350 International Journal of Electrical Engineering Education 53(4)

Figure 5. Simulink structure of the regressor-based adaptive controller.

Figure 6. Model build-up process.

Figure 7. Connection to load the software with the software.

. Once the integration of the software with the hardware is established, the model
runs by clicking on the run button as shown in the Figure 8. The experiment in
this problem is run for 10 s. It can be changed to different values by changing the
run time.
. While the experiment was running, snapshots were taken to obtain the real-time
readings of control input, tip deflection, tip position, and tip trajectory tracking
error as shown in Figure 9.
Subudhi and Pradhan 351

Figure 8. Running the experiment.

Figure 9. Scopes showing the real-time readings of control input, tip deflection, tip position,
and tip trajectory tacking error.

Project results and discussions


The regressor-based adaptive controller (RBAC) and a fixed gain controller PD are
applied to the flexible robot. To validate the tip trajectory tracking performances,
the desired trajectory vector for two joints yd(t) is chosen as follows10,11
 5 
t t4 t3
yd ðtÞ ¼ i ðtÞ þ 6 5  15 4 þ 10 3 ðt ðtÞ  i ðtÞÞ ð20Þ
td td td

where yi(0) ¼ {0,0} are the initial positions of the links and yt(0) ¼ {p/4, p/6} are the
final positions for link-1 and link-2. td is the time taken to reach the final position
which is about 4 s and total simulation time is set as 10 s. The physical parameters
of the studied robot are taken from Quanser Consulting Inc.9 The feedback gain
matrix KD ¼ diag(5, 20) and the adaptation gain matrix are chosen as  ¼ diag(0.5,
0.5, 0.5, 0.75, 1.5, 0.5, 0.5).
352 International Journal of Electrical Engineering Education 53(4)

Controller performances
Experimental results for controller performance with an initial payload of
0.157 kg. Figure 10(a) compares the tip deflection trajectories (link-2) obtained
using RBAC and PD control carrying a nominal payload of 0.157 kg. It is seen
that the initial tip deflection amplitude is high (0.001 mm) in case of PD control
whereas in case of RBAC it is small (0.0005 mm), also deflection settles faster in
case of RBAC.
Figure 10(b) presents the tip trajectory tracking error curves obtained with
RBAC and PD control. From Figure 10(b), it is revealed that the tracking errors
are 0.08 for PD control whereas it is 0.01 in case of the RBAC.
Torque profiles generated by RBAC and PD control for joint-2 are shown in
Figure 10(c). The control torque for RBAC attains 0.06 Nm and reduces to almost
zero after 4 s but in case of PD control, the maximum control torque is 0.08 Nm
and it is maintained till the tip attains its final position and, consequently, reduces
to 0.03 Nm after 4 s.

Figure 10. Experimental results: Comparison of the controller performances RBAC (), PD
control (–) under nominal payload of 0.157 kg: (a) comparison of tip deflection performances;
(b) tip trajectory tracking errors; (c) torque profiles (joint-2).
Subudhi and Pradhan 353

Experimental results for controller performance with an additional payload of 0.3 kg. An
additional payload of 0.3 kg is added to the initial payload of 0.157 kg and the
overall payload becomes 0.457 kg. Now with this change in payload, the perform-
ances of RBAC and PD control are compared in Figure 11 and it also depicts the
tip deflection trajectories carrying additional payload of 0.3 kg. From Figure 11(a)
it is observed that the initial tip deflection amplitude is more (0.01 mm) in case of
PD control than that of RBAC (0.004 mm). Figure 11(b) compares the tip trajec-
tory errors yielded by RBAC and PD controllers. RBAC exhibits better tip track-
ing performance compared to PD control as tip tracking errors are 0.085 and
0.001 resulted by RBAC and PD control, respectively. Figure 11(c) presents con-
trol torque profile generated for joint-2 by RBAC and PD control. In case of
RBAC, it attains 0.065 Nm and reduces to almost zero after 4 s but in case of
PD control the maximum input control torque is 0.085 Nm and it persists till the
tip attains its final position and reduces to 0.035 Nm after 4 s.
Figure 12 shows the estimated parameters of the robot. As the parameters are
unknown, they are initialized with zero values and estimation process yield param-
eter values in real time during the control of the flexible robot.

Figure 11. Experimental results: Comparison of the controller performances RBAC (), PD
control (–) under additional payload of 0.3 kg: (a) comparison of tip deflection performances;
(b) tip trajectory tracking errors; (c) torque profiles (joint-2).
354 International Journal of Electrical Engineering Education 53(4)

Figure 12. Estimated robot parameters.

Table 1. Comparison of experimental results for the controllers (RBAC and PD control).

Controller schemes Payload (kg) Overshoot (%) Settling-time (ts) Reference figure

RBAC 0.157 35 4.5 Figure 10(b)


0.457 03 5.5 Figure 11(b)

PD Control 0.157 40 6.5 Figure 10(b)


0.457 05 7.0 Figure 11(b)
RBAC: regressor-based model adaptive controller; PD: proportional differential.

Comparison of time-domain performance indices such as settling time and max-


imum overshoot for tip trajectory tracking achieved with RBAC and PD control
are compared in Table 1. From this table it is observed that PD control yields a
35% maximum overshoot for tip trajectory error under a nominal payload of
0.157 kg and in case of RBAC the maximum overshoot percentage is 3%. When
an additional payload of 0.3 kg is attached to tip, the maximum overshoots in case
of PD control is 40% for tip position error, but the RBAC gives 5% overshoot. Tip
trajectory error settles at 6.5 s in case of PD control and corresponding settling time
for RBAC is 4.5 s under nominal payload of 0.157 kg. In case 0.457 kg payload, the
settling time for PD control is 7 s and for RBAC it is 5.5 s. From the comparison it
Subudhi and Pradhan 355

Table 2. Summary of experiment feedback.

Average ratings
Questionnaire (5 being the max. value)

Q1. Did the experiment help you in understanding basic concepts 4.59
of design and implementation of adaptive tracking control for
nonlinear system with parametric uncertainty?
Q2. How do you compare the experiment with previous 4.48
experiments done in the area of adaptive control?
Q3. Have you gained oral and written communication of results of 4.45
the experiments performed.
Q4. Has the experiment-based teaching helped you to get insight 4.67
in real-time about hardware control components?
Q5. How effective was the role of faculties in felicitating the 4.88
course in the whole semester towards learning?

is clear that RBAC provides appropriate control signal to cope with payload vari-
ation but the fixed gain PD controller does not perform well.

Assessment and students’ feedback


This experiment has been conducted by a group of 30 students of Master in
Technology in control and automation specialization.
For the analysis of feedback on the learning outcomes, feedbacks of the students
were collected which were designed according to the following criteria:

. Project report: Design and justification of the FR for finding the regressor
matrix. Clarity and technical quality of the document.
. Presentation: Content, structure, and delivery of the presentation. Answers to
questions posed by the teacher and audience.
. Software structure and hardware implementation: Effective implementation of
designed controller. Proper use of the Simulink platform and resources.
. Creativity for extra functionalities: Such as an additional number of
fixed gain controllers, uniqueness of the adopted approach, nonlinear
model, etc.

The feedbacks collected are summarized in Table 2. The project designed for
teaching of adaptive control helped the students in gaining hands on experience on
developing adaptive controller and applying it to a nonlinear dynamics system and
how a robot adapts itself to payload variation whereas a fixed gain controller was
not possible. They learned how an adaptive controller provides necessary torque
signals to the robot joints so as to track the desired trajectory in spite of variation
of payload.
356 International Journal of Electrical Engineering Education 53(4)

Conclusions
This paper shows a simple computer-controlled experiment for flexible robot
system has been used to teach one of the popular adaptive control technique i.e.
regressor-based adaptive controller. The students gained confidence in controlling
adaptively a plant when its parameters vary in this experiment (payload varies in
the flexible robot). The students feedbacks envisage that they are convinced them-
selves about the benefits of using an adaptive controller compared to the fixed gain
controller. From the feedback statistics, it is observed that the lab experiment has
been shown to yield two major learning outcomes: first concept of adaptive control
and second project-based teaching is a good approach to teach such a difficult
subject with ease considering a real-world problem.

Declaration of Conflicting Interests


The author(s) declared no potential conflicts of interest with respect to the research, author-
ship, and/or publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publication
of this article.

References
1. Tokhi MO and Azad AKM. Flexible robot manipulators: Modeling, simulation and con-
trol. London, UK: IET, 2008.
2. Lin C-M. A microcomputer-based system for teaching adaptive control. IEEE Trans
Educ 1988; 31: 224–227.
3. Feliu V, Rattan KS and Brown BH. Adaptive control of a single link flexible manipu-
lator. IEEE Control Syst Mag 1990; 10: 29–33.
4. Rokui MR and Khorsani K. Experimental results on discrete time nonlinear adaptive track-
ing control of a flexible-link manipulator. IEEE Trans Syst Man Cybern 2000; 30: 151–164.
5. Moudgal VG, Kwong WA, Passino KM, et al. Fuzzy learning control for a flexible-link
robot. IEEE Trans Fuzzy Syst 1995; 3: 199–210.
6. Gutierrez LB, Lewis FL and Lowe JA. Implementation of a neural network tracking
control for a single flexible link: Comparison with PD and PID controllers. IEEE Trans
Ind Electron 1998; 45: 307–318.
7. Subudhi B and Morris AS. Dynamic modelling, simulation and control of a manipula-
tor with flexible links and joints. Robot Autonom Syst 2002; 41: 257–270.
8. Slotine JJE and Li W. Adaptive manipulator control: A case study. IEEE Trans Autom
Control 1988; 33: 995–1003.
9. Quanser Consulting Inc. 2-DOF serial flexible link robot, Reference Manual, Doc. No.
763, Rev. 1, 2008.
10. Pradhan SK and Subudhi B. Real-time adaptive control of a flexible manipulator using
reinforcement learning. IEEE Trans Autom Sci Eng 2012; 9: 237–249.
11. Pradhan SK and Subudhi B. Nonlinear adaptive model predictive controller for a flex-
ible manipulator: An experimental study. IEEE Trans Control Syst Technol 2014; 22:
1754–1768.

Вам также может понравиться