Académique Documents
Professionnel Documents
Culture Documents
Shaohua Sun
Laoning Information Vocational Technical College
Liaoyang 111000, China
Abstract—There are Z - N oscillation method and reaction curve controller output. Record the controller critical gain K P = K C
and so on in based methods classical approaches. In fuzzy PID
auto-tuner conventional gain scheduling involves using extra and the oscillation period of controller output, PC .
information from the plant, environment, or user to tune (via Adjust the controller parameters according to Table Ⅰ; there
“schedules”) the gains of a controller. We study how the two-link
is some controversy regarding the PID parameterization for
flexible robot operates under the “no control” situation. The
robot endpoint position shows a significant amount of endpoint
which the Z-N method was developed, but the version
oscillation. We can overcome a large overshoot and oscillations described here is applicable of best knowledge.
near its set-point of open-loop control to use fuzzy control system Example. Consider a plant with a model given by
with a supervisory level (PID controller). The experiment shows 1
that the fuzzy control system with supervisor level (PID
G0 ( s ) = (1)
( s + 1) 3
controller) gave better results in the case of large and loaded-tip
slews. Find the parameters of a PID controller by using the Z-N
oscillation method. Obtain a graph of the response to a unit step
Keywords- PID controller; fuzzy rules; feedback control system; input reference and to a unit step input disturbance. We first
design Methods;
compute the critical gain K C and critical frequency ωc . These
I. CLASSICAL PID CONTROL DESIGN METHODS values must satisfy
One of the traditional ways to design a PID controller was K C G0 ( jωC ) = −1 ⇔ K C = −( jωC + 1) 3 (2)
to empirical tuning rules based on measurements made on the
real plant. The following sections review the best-known of the From this equation we obtain K C = 8 and ωC = 3 .
classical tuning methods. Hence, the critical period is PC ≈ 3.63 . If we use the settings
in Table Ⅰ,and PID controller equation
A. Ziegler-Nichols (Z-N) Oscillation Method
1 T s
This procedure is valid only for open-loop stable plants, C PID ( s ) = K P (1 + + d ) (3)
and it is carried out by means of the following steps. Tr s τ D s + 1
Set the true plant under proportional control, with a very We obtain the following:
small gain.
K P = 0.6 * K C = 4.8; Tr = 0.5 * PC ≈ 1.81; Td
TABLE I. ZIEGLER-NICHOLS TUNING, USING THE OSCILLATION
METHOD = 0.125 * PC ≈ 0.45
284
For future reference, we note the following alternative A. objectives and open-loop control
representation of a PID controller.
We will first study how the robot operates under the “no
Lemma . Any controller of the form
control” situation. We simply apply v1 = ν 2 = 0.3615 volts at
n2 s 2 + n1 s + n0 t = 0 seconds and return v1 and ν 2 to zero voltages as soon
C (s) = (9 )
d 2 s 2 + d1 s as the links reach their set-points. Note that for this experiment
we monitor the movement of the links but so not use this
n1d1 − n0 d 2 n0 information as feedback for control.
KP = Ki =
d 12 d1 The results of the “no control” experiment are plotted in
Figure 5, where the endpoint position shows a significant
amount of endpoint oscillation. It is well known that the effect
of mass loading slewing flexible beam is to reduce the modal
n2 d12 − n1d1d 2 + n0 d 22 d2 frequencies and this is indeed the case for this experiment.
Kd = τD =
d13 d1 Indeed, when a 30-gram payload is attached to the robot
endpoint, the first model frequency of the second link
(endpoint) reduces significantly. This effect causes
This proof is ignored.
performance degradation in fixed, linear controller. In Figure 5,
as in all plots to follow, endpoint position refers to the position
III. APPLICATION OF FUZZY PID CONTROLLERS
of the elbow link endpoint. Note that the inset shown in Figure
The two-link flexible robot shown in Figure 4 consists of 5 depicts the robot slew employed. The two dashed lines
three main parts (1) the robot with its sensors, (2) the computer describe the initial position of the links. The arrows show the
and the interface to the robot, and (3) the camera with its direction of movement, and the solid line shows the final
computer and interface. The robot is made up of two very position of the links. Hence, for this open-loop experiment, we
flexible links constrained to operate in the horizontal plane. wanted 90 degrees of movement in each link. In the ideal case
The “shoulder link” is a counterbalanced aluminum strip that is the shaft should stop moving the instant the voltage signal to
driven by a DC direct-drive motor with an input voltage v1 . the motor amplifier is cut off. But the arm had been moving at
The “elbow link” which is mounted on the shoulder link a constant
endpoint is a smaller aluminum strip. The actuator for the
elbow link is a geared DC motor with an input voltage v2 . The
sensors on the robot are two optical encoders for the motor
shaft position Θ1 and Θ 2 , and two accelerometers mounted on
the link endpoints to measure the accelerations a1 and a2 .
Figure 4.Two-link flexible robot setup a constant velocity before the signal was cut off, and thus had a
momentum that dragged the shaft past the angle at which it was
A line scan camera is used to monitor the endpoint to stop. This movement depends on the speed at which the arm
position of the robot for plotting; this data is not used for was moving, which in turn depends on the voltage signal
feedback. The sampling period used for all sensors and control applied. System settling (elimination of residual vibrations in
updates is 15 milliseconds (ms). For comparative purposes, we endpoint position) within 2 seconds of motion initiation, and
use the camera data for robot movements that begin in some overshoot minimized to less than 5% deviation from the final
position and end in a fully extended position, to approximate desired position.
equal movements in each joint. When responses are plotted, the
final endpoint position is nominally indicated (on the plot) to B. Fuzzy control system with a supervisory level (PID
reflect (approximately) the total movement in degree, of the controller)
shoulder joint.
285
We can overcome a large overshoot and oscillations near only these two rules and since only one rule will be enabled at
its set-point of open-loop control to use fuzzy control system each time step, there is no need for the use of complex
with a supervisory level (PID controller). The simple fuzzy PID inference strategies in the fuzzy PID controller.
controller is shown in Figure 6, which monitors the position
Clearly, it would be possible to view the premises of the
error e1 (t ) to the shoulder motor controller and changes the above rules as linguistic statements then define membership
fuzzy sets and the rules-base of the shoulder motor controller. functions to quantify their meaning. Then, our supervisor
would be a fuzzy system that would gradually rather than
abruptly switch between the two conditions. Here, we did not
pursue this approach. Instead, we made sure that the rule-bases
were designed so that a smooth transition would occur as the
supervisor switched between the rules.
Rule-base construction
The membership function and the rules-base for the elbow
link controller are shown in Figure 7. In addition to the rule-
Figure 6. Fuzzy control system with a supervisory level (PID
controller) base in Table 3, another rule-base was added for the shoulder
motor controller. The PID controller therefore switches
The rule-base for the fuzzy PID controller consists of tow between these two rule-bases for “coarse control” and “fine
single-input multiple-output rules: control.” The membership function for the coarse controller
are the fuzzy controller case where the universe of discourse is
1) If e1 (t ) 〈 20 0 then use Rule-Base 1 (Table Ⅲ) and use [-250, +250] degree for the position error, [-2, +2] g for the
expanded universes of discourse. endpoint acceleration, and [-1.5, +1.5] volts for the out put
0
2) If e1 (t ) ≥ 20 Then use Rule-Base 2 (Table Ⅳ) and use voltage. The fine controller, with rule-base shown in Table 4,
compressed universes of discourse. uses the universe of discourse for the position error is [-25,
TABLE Ⅲ. RULE-BASE FOR COARSE CONTROL +25] degrees, and the universe of discourse is [-0.15, +0.15]
volts.
Notice that while going from the coarse to the fine control,
the widths of the universes of discourse for the position error
and the output of the shoulder link controller have been
reduced by a factor of ten, while the width of the universe of
discourse for the endpoint acceleration is reduced by a factor of
two. This choice was made after several experiments where it
was found that when the width of the universe if discourse for
the acceleration was reduced by a large factor, the controller
became too sensitive near the set-point.
The PID controller expands or compresses the universes of Table Ⅲ and Ⅳ show the rule-base used for the coarse
discourse by simply changing the scaling gains (explained and fine control, respectively. Notice that in row j = 0 for the
below). When the universe of discourse is expanded, a “coarse rule-base for fine control there are extra zeros as before to
control” is achieved, and when it is compressed, a“fine control” reduce the sensitivity of the controller became to a noisy
is achieves. It is a type of auto - tuning strategy. The form of acceleration signal. The rule-base for coarse control does not
the premises of the rules of the supervisor guarantees that one have these zeros as the offset voltage from the accelerometers
(and only one) of the rules will be enabled and used at each is of no consequence as long as the controller is operating in
time step. Since the control objectives can be achieved using this region. Also notice that while the patterns in the bodies of
286
the tables shown in Table Ⅲ and Ⅳ are similar, there are Experimental results obtained using this supervisory
differences included to reflect the best way to control the robot. scheme are shown in Figure 8. The requested slew is 900 for
Notice that the center values in the fine control rule-base both links as shown in the inset. The response is relatively fast
change more rapidly as we move away from the center of the with very little overshoot. Comparing this response to the
rule-base as compared to the coarse control rule-base. This obtained for the open-loop control (see Figure 5), we can see
causes a bigger change in the output of the controller for that the hump in the initial portion of the graph is almost
smaller changes in the input, resulting in better control over the eliminated. The response of the robot to a counterrelative slew
is better than the response when using the open-loop control.
From the results obtained for the supervisory control (PID
controller) technique, we see that the results are superior in the
cases tested. The supervisor gave better results in the case of
large, counterrelative, and loaded-tip slews.
Overall, we see that rule-base supervision can be used to
significant benefit in a practical application.
IV. CONCLUSION
287