Paper ID #11184

From ’system modeling’ to ’controller hardware testing’ in three hours: a

robotic arm controller design lab using MATLAB Real Time Windows Tar-
get to reinforce classical control theory
Dr. Charles Birdsong, California Polytechnic State University
Charles Birdsong has expertise in automotive safety, vibrations, controls, signal processing, instrumen-
tation, real-time control, active noise control, and dynamic system modeling. He received his B.S.M.E.
at Cal Poly San Luis Obispo, and his M.S. and Ph.D. at Michigan State University, where he worked
on active noise control applications for the automotive industry. He has worked in the vibration test and
measurement industry helping to drive new technologies to market and working with industry to meet
their emerging needs. He is currently a Professor at California Polytechnic State University at San Luis
Obispo in the Department of Mechanical Engineering teaching dynamics, vibrations and controls. He is
involved in several undergraduate and master’s level multidisciplinary projects and interested in engineer-
ing education research.

Page 26.798.1

c American Society for Engineering Education, 2015

 From ‘system modeling’ to ‘controller hardware testing’ in three
hours: a robotic arm controller design lab using MATLAB Real
Time Windows Target to reinforce classical control theory

Abstract
A 3 hour hands-on lab experience was designed to give students the opportunity to model, design
and test a controller for a nonlinear electro-mechanical robot-arm system as a culminating pro-
ject in mechanical engineering controls course. Students used linearization to model a nonlinear
pendulum DC servo system using first principles. They used experimental frequency response
data to derive a numerical transfer function model. Next they used analytical techniques includ-
ing root locus and Simulink modeling to design a PID controller. Finally they tested their con-
troller design using a hardware-in-the-loop system with MATLAB’s real-time windows target
system to assess the performance of their controller. The entire process was started and complet-
ed in one three hour lab period. The goals of the exercise were to give students the chance to
complete an entire control system design cycle from modeling to hardware testing in one sitting,
incorporate as many of the course concepts as possible and give the students a practical under-
standing of the application of the theory. Assessment was conducted using pre and post online
quizzes testing conceptual understanding of the major topics such as linearization, frequency re-
sponse, and the effect of proportional, integral and derivative control. The assessment indicate a
significant improvement in understanding of the theory and positive attitudes regarding the expe-
rience.

Introduction
Our university’s philosophy is that students learn best through a combination of lecture and lab
experiences and industry feedback indicates that our undergraduates are unparalleled at hitting
the ground running and working with real world problems. While many have reported on the
effectiveness of including hands-on laboratory exercises to enhance learning [1], these labs are
expensive and there is pressure to eliminate them in the times of increasing budget pressure.
This study suggests that the cost is justified because learning outcomes are significantly im-
proved compared to a lecture only course.

Mechanical Controls is a four unit, required senior level course that consists of three-one hour
lectures and one-three hour lab per week for the ten week quarter. The course covers single-
input single-output linear system modeling, time domain analysis, transfer functions, root locus,
Page 26.798.2

frequency response methods, PID and lead lag controllers. The lab is taken concurrently with
the lecture and is designed to support the topics covered in lecture while also illustrating the real-
ities of real world systems, modeling and controller implementation. There are currently four
exercises: a two week Matlab/Simulink simulation exercise, a two week analog DC servo posi-
tion control experiment, a two week digital controller tank water level regulator experiment and
a two week digital controller hydraulic servo control experiment.

The digital control experiments use Matlab Real Time Windows Target (RTWT) with a National
Instruments data acquisition board (DAQ) interface to implement a simple PID controller in real-
time. This software interface is ideal for the learning environment and is use in many universi-
ties because the Simulink code is displayed as a block diagram that is identical to the theory that
is presented in the textbook [2] [3] [4]. This makes modifying and developing new controllers
simple for the student with minimal programming experience. This hardware was introduced in
this lab in 2008 in an experiment using a hydraulic control system [5].

This paper describes the DC servo experiment that is performed early in the term and also the
addition of a new one-week PID Design Project that is intended to include elements from many
of the lecture topics and end the term with a single creative and self-contained experiment. As-
sessment of the new project was conducted using a pre and post quiz and surveys.

Linear DC Servo Setup

The new PID Design project is conducted in the last week of the term. However, students test
and analyze the DC servo at the beginning of the quarter with only an inertia disk and gear sys-
tem as the mechanical load. The experimental apparatus shown in Figure 1 is a DC servo control
system based on the Motomatic Inc. hardware. The angular position is measured by an angular
potentiometer sensor. The sensor signal is measured by a DAQ interface to a Matlab/Simulink
RTWT PID controller. The controller generates an actuator signal via the DAQ analog output
channel which is connected to the motor amplifier to close the loop.

Potentiometer
Position Sensor

Motor Amplifi-
er

DC Motor

Page 26.798.3

Figure 1. Photograph of DC servo control apparatus showing DC motor, potentiometer position sensor and
motor amplifier.
The DAQ communicates with the MATLAB/Simulink RTWT software [6], [7]. The software is
a special block-set within the Simulink software that allows DAQ inputs and outputs to be used
as sources and sinks in a Simulink model. Before the model can be run in external real-time
mode it must be compiled into C code. This process is initiated by clicking on a button and is
then handled automatically by MATLAB. The student does not need to have any special pro-
gramming skills beyond Simulink programming.

Data can be logged and exported to the MATLAB workspace for controller evaluation. The
sampling rate of the system depends on the complexity of the model and the speed of the PC
processor. The PC processor is used to perform the real time calculations while the RTWT soft-
ware manages the Windows operating system and guarantees the control process receives the
maximum processor resources. Using a Pentium 1.9 GHz PC the PID Positioner model can be
run as fast as 10 kHz however the system was limited to 1 kHz which is adequate for the servo
control experiment. Faster sampling rates could probably be attained for other applications with
faster PC processors.

The DC servo motor model is simply a first order transfer function and produces acceptable
agreement with experimental results. A significant amount of static friction in the motor and
gears results in a nonlinear step response, where the shaft overshoots only once then sticks at a
final position instead of oscillating and decaying gradually as a linear system. A nonlinear mod-
el can be created by adding a dead-zone function between the power amplifier and the DC servo
blocks to introduce the static friction effect. Figure 2 shows a block diagram of the system in
closed-loop feedback.

Figure 2. Block diagram of DC Servo position control system.

Figure 3 shows a typical comparison of closed-loop step response of the model and experimental
results from a typical student report. The experimental response is the solid line, the non-linear
model including the dead-zone block is the thin solid line and the tuned non-linear model is the
dashed line. As usual, the model does not agree with the experiment perfectly because of exper-
imental error in approximating the system parameters plus an inexact model of the nonlinearities
of the system. The model and can be improved by tuning the system parameters. The tuned
model is the result of modifying the system parameters in an ad-hoc fashion to match the exper-
imental results, such as peak time, steady-state value, etc., as well as possible.
Page 26.798.4
 experimental
response

Figure 3. Comparison of closed-loop model (thin solid line), tuned model (dashed line) and experimental re-
sults (thick solid line) of closed-loop proportional controller with DC servo control and step input.

This experiment is conducted early in the quarter term at a time when the students may not have
enough theory from the lecture to be able to design a PID controller based on design specifica-
tions. Although the model can be used to try PID gains in a trial and error fashion to determine
acceptable controller gains. Students are asked to vary the controller gains and observe the
qualitative effects of the proportional gain on response measures such as settling time, percent
overshoot, damping, and steady state error [8]. The lab is very successful and gives the students
an opportunity to get a hands-on understanding of the effects of PID controller gains. Because
the lab is conducted before the theory for controller design is complete, it helps motivate the
concept of controller design as an improvement over the ad hoc method of tuning the controller
used in the lab.

New PID Controller Design Project

Near the end of the term, after 4 weeks working on two other systems, students returned to the
DC Servo for the new PID Design Lab. The DC servo system was modified by the addition of a
cantilever beam with a mass at the end fastened to the shaft. The arm creates a moment that de-
pends on the angular position of the shaft. This results in a nonlinear model where the step re-
sponse is different at each steady state angular position. The challenge for the students is to pre-
dict the system model based on a description of the system, and to use the model to design the
controller. While the nonlinear system has many possible advanced control schemes [9] [10]
[11], for this introductory course, students were limited to linearization and PID control about a
Page 26.798.5

steady-state operating point which is covered in the lecture. This can be done by linearizing the
nonlinear equations of motion then using classical root-locus techniques.
 Cantilever beam
added to motor shaft

Figure 4. Photograph of DC motor with cantilever beam attached to the shaft to create a nonlinear system.

Page 26.798.6
 Figure 5. Closed-Loop PID Controller Interface in Matlab/Simulink RTWT

Figure 5 shows the closed-loop PID controller model that is used to implement closed-loop posi-
tion control of the system. This interface has the advantage that it looks like the closed-loop
feedback block diagram that the students study in the lecture. This helps the students understand
and visualize the program very quickly. A special pulse reference function was used to force the
students to consider several different effects. The desired input repeating sequence is a combina-
tion of ramp up, ramp down, dwell and return to start. A constant is added by the “SS Position”
block to move the pendulum arm from the neutral position to a steady-state position before the
repeating sequence is started. This results in a steady state angle of about 20 degrees from the
bottom, though this value can be changed to any value desired.

Students predict the linearlized transfer function model based on a description of the system as a
homework assignment. Then during lab, rather than perform system identification, students
were shown a demonstration of an open-loop swept sine frequency response test using an LDS-
Dactron Focus real-time signal analyzer and given the experimentally measured frequency re-
sponse (bode) data. They were asked to use the bode data to estimate the system transfer func-
tion, and then use the transfer function to perform analytical PID controller design. To empha-
size the importance of modeling and design versus trial and error controller tuning, they were
then given only two chances to test their design with the real system. Students filled out a form
with the controller transfer function and PID gains and submitted it to the instructor who entered
the gains on the test system and ran a test.

Figure 6 shows the results of a typical PID controller performance evaluation. The top graph
shows the controller output in volts. The middle plot shows the desired reference signal and the
measured position under the closed-loop PID control. The ramp from 2 to 3 seconds was includ-
ed to force the students to consider adding integral control to reduce the steady state error in this
region. The bottom plot shows the error function which was computed from the absolute value
of the difference between the reference and measured signals and summed for each data point.
The sum of the error values was displayed as the “Error Function” value and used to measure the
performance of the PID controller design. The Error Function value of 83 shown in the figure
was among the best performing controllers in the class. The evaluation and the figure was auto-
matically generated after the test by implementing an m-file in the Simulink callback function.
Students were given the plot and could use the data to refine their design for a second test.

A significant amount of electrical noise was present from the potentiometer position sensor,
which can be seen in the error signal. The overall Error Function is affected by this noise. Fur-
thermore, the high noise level limits the use of derivative control since derivative control ampli-
fies the high frequency noise signal and corrupts the actuator signal. A higher quality sensor with
less noise would be a good improvement in the system and is under consideration for the future.
On the other hand, the noise forces the students to face the common problem with noise and de-
rivative control.
Page 26.798.7
 Figure 6. PID controller design evaluation results with controller output (top), measured and reference sig-
nal (middle) and error function (bottom).

The figures below show some other student results. Figure 7 shows a controller that produced an
acceptable response, though the Error Function value is worse than the previous example due to
more steady-state error during the ramp; a higher integral gain would improve the performance.
Figure 8 shows a controller that produces poor response where the output does not track the input
well; the proportional gain was too low. Figure 9 shows a controller with acceptable response
until the step change at 3.5 seconds after which the nonlinearity of the system causes instability;
the proportional gain was too high in this case.

Figure 7. PID feedback control response with acceptable results.

Page 26.798.8
 Figure 8. PID feedback control response with poor response.

Figure 9. PID feedback control with good response until the step change; after which the nonlinearity results
in instability.

Students were asked to document their design methods and turn in a report showing the design
steps and any supporting calculations used in the process. Students were encouraged to model
the closed-loop response with their PID control design before trying their controller design on
the real system. Students were told that the report grade would be based mostly on documenta-
tion of the design procedure, not the performance of the controller, though, the top three designs
would receive extra credit. This seemed to help motivate the students in a somewhat competitive
environment and make the three hour project more fun.

The design project will be used in the future and has the benefit that the system can be modified
to change the dynamics by modifying the mass and arm length and steady-state angle. Also dif-
ferent reference functions can be used to change the best controller design results. Hopefully this
will eliminate the possibility that students will get results from previous years and get a short cut
to a good design. The challenge is in formulating the design constraints in a way that allows
classical linear control theory and PID controller design methods to result in good solutions
while at the same time illustrating the real world effects such as saturation, noise and other non-
linear phenomena.

Implementation Issues
While this exercise was conducted using the DC servo hardware that is already installed in the
controls lab, it should be possible for others to reproduce a similar exercise with a limited budg-
et. This exercise can be conducted with a single experimental setup to minimize the cost of mul-
tiple stations: students use analytical methods to design a controller and then the controller is
Page 26.798.9

tested on one unit by the instructor. The hardware consists of a typical DC motor, a potentiome-
ter position sensor, and a power amplifier. Data acquisition hardware that is compatible with the
Matlab RTWT is available below $400. The most expensive item is the Matlab/Simulink with
the Real Time Windows Target license, though many universities purchase a site license that in-
cludes all the required software. Matlab is the key to the experiment because it allows the in-
structor to input controller gains and run the system in few seconds so that many students can
test their designs in a short amount of time. On the other hand, it would be possible to imple-
ment a real-time PID controller using a low cost microcontroller such as the Arduino [12] and
achieve the same results without the Matlab license. Hopefully the reader agrees that giving the
student the opportunity to complete the entire system modeling, controller design and testing ex-
perience in one three hour setting is worth investing in the hardware and software.

Assessment
The PID Design Lab exercise was assessed to measure how it improved students understanding
of the main concepts and their own confidence in the concepts embedded in the experiment. It
should be noted that all of the concepts that were assessed had been covered in the lecture por-
tion of the course in the preceding weeks. Also students were told to complete the homework
assignment that involved deriving equations of motion, linearizing the nonlinear terms and deriv-
ing a linearized transfer function model from a description of the system before the assessment.
Therefore, the students should have had significant knowledge from the course and the home-
work assignment to answer the questions. A pre and post assessment was given using an online
quiz and survey instrument. Eight technical questions were given in an online quiz the night be-
fore the experiment asking students to predict how the system will respond to different PID con-
troller configurations. The next day the students conducted the experiment in the lab. They were
then told to complete the post assessment online within twenty-four hours so that as little time as
possible and other confounding events would not affect their responses. In the post assessment,
they were asked the exact same technical questions to determine if their understanding had im-
proved. Figure 10 illustrates the results. The results only include students that completed both
pre and post assessments. The students were told they would get full credit worth one homework
assignment for completing the assessments regardless of their performance.

In most cases the students performed much better after the lab experience as indicated by the
significant positive change between pre and post. The text of the eight questions is presented in
the appendix.

Page 26.798.10
 Compare Pre and Post Assessment Question Performance
100%

90%

80%

70%

60%

50%

40%

30%

20%

10%

0%
Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8

Pre Post delta

Figure 10. Pre and post assessment showing percent of students selecting the correct answer to eight technical
questions, and the change (delta), n=22.

In addition, students were asked to rate their own understanding of four main concepts (Ques-
tions 11-14 are listed below): linearization, proportional, integral and derivative control gains,
before and after the experiment. Students answered using a Likert scale [12] [13]. Student re-
sponses changed significantly towards indicating a better understanding after the exercise.

Q11: I understand the difference between a linear and a non-linear system and how to
model both.

Q12: I understand the effects that a proportional controller has on a closed-loop feedback
system

Q13: I understand the effects that an integral controller has on a closed-loop feedback
system

Q14: I understand the effects that a derivative controller has on a closed-loop feedback
system. Page 26.798.11
 Pre-Lab Assessment Post-Lab Survey
1.00 1.00

0.80 0.80

0.60 0.60

0.40 0.40
0.20
0.20
0.00
0.00
Q11 Q12 Q13 Q14
Q11 Q12 Q13 Q14
Strongly disagree Strongly disagree
Somewhat disagree Somewhat disagree
Neither agree nor disagree Neither agree nor disagree
Somewhat agree Somewhat agree
Strongly agree Strongly agree

Figure 11. Comparison of pre and post survey questions 11 through 14, n=22.

Table 1. Pre-Lab Assessment Results

Q11 Q12 Q13 Q14
Strongly agree 9 13 8 6
Somewhat agree 12 8 14 14
Neither agree nor disagree 0 0 0 0
Somewhat disagree 0 1 0 2
Strongly disagree 1 0 0 0

Table 2. Post-Lab Assessment Results

Q11 Q12 Q13 Q14
Strongly agree 16 17 19 15
Somewhat agree 5 5 3 6
Neither agree nor disagree 0 0 0 1
Somewhat disagree 0 0 0 0
Strongly disagree 0 0 0 0

Finally, after the experiment, students were asked to compare the lab experience with solving
textbook problems with their response to the following, “Q15: The PID Design Lab helped me
understand the controller design process more than solving textbook homework problems.”
(Post-lab survey only).
Page 26.798.12
 Q15
0.70

0.60

0.50

0.40

0.30

0.20

0.10

0.00
Strongly agree Somewhat agree Neither agree Somewhat Strongly disagree
nor disagree disagree

Figure 12. Response to Question 15: “The PID Design Lab helped me understand the controller design pro-
cess more than solving textbook homework problems” (Post-Lab Survey Only), n=22

The assessment results show that even though the concepts had been covered in the lecture por-
tion of the class in previous weeks, student understanding was relatively low before the lab. Af-
ter the 3 hour exercise understanding was significantly improved. If the course did not have the
practical lab component it seems that students would be left with the weaker understanding of
the concepts illustrated in the pre-lab assessment. Students also strongly agreed that the lab
helped them to understand the controller design process more than solving textbook problems.

Conclusions
The MATLAB/Simulink RTWT software was introduced into the Mechanical Engineering Con-
trols lab at Cal Poly and applied in PID Controller Design project. The software is easy to pro-
gram and easy for the students to understand in the context of linear control theory. In addition
the ability to modify the controller with very little effort enabled a PID design project to be add-
ed to the curriculum. The project forces students to apply many of the theoretical topics covered
in the lecture in a single three hour lab setting. It forces students to apply analytical tools to de-
sign a controller and then to immediately see the actual performance of the physical system. As-
sessment shows that the new project improved student understanding and self confidence in the
concepts of linearization of nonlinear systems and controller design. Furthermore the results
suggest that the hands-on experience had a larger influence than lecture and solving textbook
problems on student understanding and self-confidence. This exercise also has the advantage
that it can be repeated at other universities with a single hardware setup, thus minimizing the to-
tal cost of maintaining multiple hardware stations.
Page 26.798.13
Bibliography

[1] L. Chen and C. Lin, "Combining Theory with Practice in Information Security Education," in Proceedings of
the 11th Colloquium for Information Systems Security Education, Boston University, 2007.
[2] A. Johnson and J. Rubaai, "DSP-Based Real-Time Control Systems Design, Analysis, and Implementation for
Reinforement of Controls Education," in ASEE Conference, 2006.
[3] A. Mohammadzadeh and S. Haidar, "Analysis and Design of Vehicle Suspension System Using MATLAB and
Simulink," in ASEE, 2006.
[4] R. Throne, "Developing Models and an Understanding of Their Limitations," in ASEE, 2006.
[5] C. Birdsong, "Developing A Matlab/Simulink Rtwt Based Hydraulic Servo Control Design Experiment," in
ASEE Annual Conference & Exposition, Honolulu, 2007.
[6] MathWorks, MATLAB, version 7, Natic, MA: Math Works, Inc., 2005.
[7] MathWorks, Real-time Windows Target User’s Guide Version 2, Natick, MA: Math Works, Inc., 2004.
[8] N. Nise, Control Systems Engineering, Wiley & Sons, 2004.
[9] F. Reys and R. Kelly, "Experimental evaluation of identification schemes on a direct drive robot," Robotica, 15
(1997), p. pp. 563–571, 1997.
[10] R. Fernando and R. Kelly, "Experimental evaluation of model-based controllers on a direct-drive robot arm,"
Mechatronics, Volume 11, Issue 3, pp. Pages 267-282, April 2001.
[11] A. Nikranjbar, "Adaptive PID controller design guidelines for a class of non-linear systems," International
Journal of Engineering Systems Modelling and Simulation, pp. Volume 6, pp 12-22, 2014.
[12] P. Shetti, "Design and implementation of GSM based and PID assisted speed control of DC motor.,"
International Journal of Advanced Research in Electrical Electronics and Instrumentation Engineering, vol. 3,
no. 4, 2014.
[13] T. Cross and K. Angelo, Classroom Assessment Techniques, 2nd ed, San Francisco: Jossey-Bass, 1993.
[14] T. Watson, Gudelines for Conductin Surveys, Brooks AFT, Texas: Air Force Materiel Command Human
Resources Directorate, 1997.

Page 26.798.14
Appendix
Below is the original text used in the pre and post assessments with the 8 technical questions fol-
lowed by 5 survey questions. Q15 was only given in the post-lab assessment.

Technical Questions Q4: For the next questions assume that the posi-
tion of the arm is measured with a sensor and a
Q1: If the system is first driven by a positive step PID feedback controller is used for position con-
input, then returned to the initial position, then trol. Assume that some initial setting is selected
an identical, but negative step input is applied to for the proportional, integral and derivative gains
the system, when comparing the positive and and step response is measured.
negative step inputs, which of the following will
be true? Which of the following will be true if
the proportional gain is increased but the in-
The direction will be opposite, but the magnitude tegral and derivative gains are held constant?
of the positive and negative step response will be Select one or more:
the same.

a. The steady-state error will be reduced

True
b. The settling time will get faster
False
c. The overshoot will be reduced
Q2: The settling times will be the same. d. The reset time, or time to get to zero
steady-state error will be reduced

True Q5: Which of the following will be true if

the integral gain is increased but the propor-
False
tional and derivative gains are held constant?
Select one or more:
Q3: The steady-state errors will be the same.

a. The steady-state error will be reduced

True
b. The overshoot will be reduced
False
c. The settling time will get faster

d. The reset time, or time to get to zero

steady-state error will be reduced

Page 26.798.15
Q6: Which of the following will be true if Q12: I understand the effects that a proportional
the derivative gain is increased but the propor- controller has on a closed-loop feedback system.
tional and integral gains are held constant? Select
one or more:
a. Strongly agree

a. The steady-state error will be reduced b. Somewhat agree

b. The settling time will get faster c. Neither agree nor disagree

c. The overshoot will be reduced d. Somewhat disagree

d. The reset time, or time to get to zero e. Strongly disagree

steady-state error will be reduced
Q13: I understand the effects that an integral
Q7: Consider the system type (defined by the controller has on a closed-loop feedback system.
number of integrators in the open-loop transfer
function. What is the system type? Select one:
a. Strongly agree

a. Zero b. Somewhat agree

b. One c. Neither agree nor disagree

c. Two d. Somewhat disagree

e. Strongly disagree

Q8: With a proportional controller and a step Q14: I understand the effects that a derivative
input, the steady-state error will be zero. controller has on a closed-loop feedback system.

True a. Strongly agree

False b. Somewhat agree

c. Neither agree nor disagree

d. Somewhat disagree
Survey Questions
e. Strongly disagree
Rate your understanding of the following con-
cepts Q15: The PID Design Lab helped me understand
the controller design process more than solving
Q11: I understand the difference between a line- textbook homework problems. (Post-Lab Survey
ar and a non-linear system and how to model Only)
both.

a. Strongly agree
a. Strongly agree
b. Somewhat agree
b. Somewhat agree
c. Neither agree nor disagree
c. Neither agree nor disagree
d. Somewhat disagree
d. Somewhat disagree
Page 26.798.16

e. Strongly disagree
e. Strongly disagree