Control Systems Lab (SC4070)

Laboratory Sessions

Overview AD
The laboratory sessions are an essential part of this course.
The purpose is to gain hands-on experience with the control y
of a real physical process. You will be working in groups Matlab experiment
Simulink
of two students. After you have chosen a partner to work >>

with and a laboratory setup, you should proceed along the

following lines:
computer u
DA
1. Implement a nonlinear, physical model of the setup
in Simulink as your home preparation. Prepare an Figure 1: General scheme of an experimental setup.
identification experiment to identify a black-box
model of the system (or its part) and test it with the
physical model. a proper sampling period h in the M ATLAB workspace. Ta-
ble 1 shows the signals u and y for the individual experi-
2. In the lab: calibrate the physical model to match
the process. Carry out the identification experiment, mental setups.
collect the data.
3. Analyze the results from the precious step, design Table 1: Inputs and outputs for the individual setups.
a controller for the simulation model as your home Experiment u y
preparation. Pendulum motor voltage cart pos., pend. angle
4. In the lab: test and fine-tune the controller on the Balance motor voltage cart pos., balance angle
process. Helicopter motor voltage arm angle, prop. speed
5. Present the results one presentation per group, one Rot. pend. motor voltage arm and pendulum angle
report per group.
More details about the above steps are given in the subse- In Simulink, these signals can be displayed in graphical
quent section of this text. Week 1 is the first week of the scopes and can also be exported to the Matlab environment
quarter. One lab session takes three to four hours. If the (using the scopes themselves or the To Workspace block)
schedule permits, you may, of course, spend more time in and saved. A digital controller will be implemented also
the lab. Let us stress that it makes little sense to come to in Simulink, typically using blocks from the libraries Dis-
the lab sessions unprepared. For this course it holds per- crete and Control System Toolbox.
haps more than for other ones that what you get out of it is The sensor readings y are expressed in their respective phys-
directly proportional to what you yourself put in. ical units (e.g., meters, radians, etc.). The actuator com-
mand u, however, is always normalized between -1 and +1
(or 0 and +1), where 1 corresponds to 100% of the available
Experimental Setups
range.
Each experimental setup consists of a hardware system (in-
verted pendulum, helicopter, inverted wedge or rotational
Design Procedure
pendulum), a computer running M ATLAB and Simulink,
and the interfacing hardware (amplifiers, A/D and D/A con-
Each experimental setup comes with a certain control task,
verters), see Figure 1.
e.g., balance the pendulum while making the cart follow a
Simulink interacts with the hardware in real time by reading specified reference trajectory. You may choose to design a
the measured systems output y and commanding the con- pole-placement controller (output feedback) controller or a
trol input u that drives the corresponding actuator (motor, PID controller. It is recommended to use output-feedback
pump, etc.). Both u and y can be multi-dimensional sig- control, as this approach is more systematic and works for
nals (vectors). The corresponding real-time input and out- all setups without using special schemes (e.g., cascade con-
put Simulink blocks are provided, you only need to define trol). For either choice of the controller, the design proce-
 .
dure is roughly the following: 1 x 1 x
y
s s
..
x
1. Implement a Simulink model and find the unknown
system parameters by using simple (but well thought algebraic
block f
of) experiments. The response of your simulation
u (+,-,*,sin)
model should reasonably fit the measured data. Store
some typical results for your presentation.
Figure 2: General model structure in Simulink.
2. Linearize this model around an operating point. Even
m
though our simple lab processes can easily be lin-
F
earized by hand, it is recommended to implement the d
linearization procedure as a M ATLAB script (using
functions trim and linmod), such that the controller Figure 3: Sliding mass with viscous friction.
can easily be re-designed for different operating points.
3. Choose design parameters (closed-loop natural fre- where x is the position of the mass. This equation is in
quency, damping, etc.) for the controller and observer the required form and can directly be implemented in a
(if used). Design the controller (using, e.g., place or Simulink model shown in Figure 4, where the position of
acker) and implement it in Simulink. the mass is the output of the system.
4. Check the controller with your models (nonlinear and .
linearized). If the response of the system is not satis- 1 x 1 x
y
s s
factory, change the controller parameters and try a new ..
x
controller. Store some typical results for your presen-
tation. +
1/m -d
5. Only after the simulated closed-loop response meets F
the requirements, test the controller on the actual sys-
tem. Because of some model-plant mismatch, you may
Figure 4: Simulink model for the sliding mass.
need to fine-tune the controller by changing the design
parameters. Therefore, the design procedure should be
implement as an easy-to-use M ATLAB script or func-
tion. Store some typical results for your presentation.
Parameter Identification
Compare the results with your expectations after each step,
Each experimental setup has certain parameters that need to
and if the results are not satisfactory, go back a step (or
be identified by means of experiments. These experiments
more). Since all the computations are done by M ATLAB,
are not prescribed, you should figure them out yourself.
it is easy to try a different setting and quickly compute the
corresponding controller to check what happens (rapid pro- One possible way to come up with a useful experiment is
totyping). the following. Analyze the model equations and try to elim-
inate most of the terms in the equation (e.g., by assuming
derivatives equal be zero). Then, devise an experiment that
Simulation Model satisfies these assumptions and collect data. Finally, com-
pute the value of the parameter using the reduced equations
For mechanical systems, the model equations are typically that apply to that particular experiment.
given as (pairs of) second-order nonlinear differential equa-
As an example, let us identify the friction coefficient for our
tions of the form
sliding mass. By assuming x = 0 (or x = const), equation
x = f (x, x, u)
(1) reduces to
where x is the generalized position (angle, horizontal posi- F = dx .
tion, etc.), u is the input, and f () is some static function
parameterized by several (physical) parameters. Equations Hence, when we apply a constant force, plot the resulting
of this type can easily be translated into a Simulink model position as a function of time, and look at the part of the
such as the one shown in Figure 2. plot where x = 0, the friction coefficient follows easily as
the ratio of the force and the resulting velocity.
As an example, consider a mass sliding across a surface with
viscous friction (coefficient d ), pushed by a certain (known) Each experiment should be repeated several times to see
input force F (Figure 3). whether it is reproducible and to obtain a reliable estimate
of the parameter value.
The model equation for this system is
Note: Always start experiments with small actuator values
1 
(e.g., 0.2) in order not to damage the setup.
x = F dx (1)
m