Вы находитесь на странице: 1из 5

1 Introduction

Modern Industry is built on our ability to control physical process as best as we can, in fact it is so essential that we
have an entire niche hardware industry for building PLC (Programmable Logic Controller) so that these systems do not
fail. Most of the time while modeling a control process we use our understanding of the underlying physical process that
dictates the system to create our control algorithm however under a lot of circumstances it may not be possible to know
prior what those physical processes are or those physical processes require heavy computation for us to make predictions,
PID controllers are an eective and computationally cheap control algorithm which can be used under these circumstances.
This report shall discuss a particular case of using a PID controller to control the elevation angle of a quanser helicopter.
2 Background information
2.1 Closed loop SISO control
Figure 1: SISO Control System
In general, the objective of a control system is to ensure that a given system G produces an output response y(t), which
closely follows a desired output trajectory r(t). This is done by controlling the input, u(t)of the controller plant.
In particular, for the quanser experiment, the given system G is of SISO type: single-input, single-output. In this case
the output response y(t) is an actual elevation angle of the lever arm, measured in degrees.j The desired trajectory is the
demand elevation angle, provided by the quanser operator via Simulink. System input u(t) is the DC voltage of the motor
which drives two propellors. The task is to design a controller system which results in closest matching between y(t) and
r(t), i.e. achieving minimal overshoot and settling time.
2.2 Explanation of PID components
u(t) = K
P
e(t) + K
I

t
0
e(t)dt + K
D
de(t)
dt
(1)
where
u(t) is the input signal to the plant process which in our case is the power supplied to the helicopter.
e(t) is the error which is expressed as r(t) y(t) where r(t) is the set point elevation we want from our controller
and r(t)is the elevevation that the helicopter currently is in.
K
P
, K
I
and K
D
are the Coecients we are tuning.
(1) is the general equation for PID controller where K is the proportional term,T
I
is the integral term and T
D
is the
derivative term, s is the error input while D(s) is the output of the plant, for the PID controller to be useful their needs
to be adjustment made experimentally to set K, T
I
and T
D
.
1
Figure 2: SISO Control System with PID Controller
The helicopter laboratory is an investigation on 3-DOF quad-copter, the aim of the investigation is to implement a
control algorithm for eective control of quad-copter elevation angle measured from ground, in our case the algorithm we
are using is a PID controller and are to investigate the suitability of it and experimentally determine the proportional ,
derivative and integral constants required.
3 Optimised response
Trial and error approach has led to nding such a conguration of PID controller as to give small overshoot and settling
time. This is presented in the graphs below:
(a) Optimised response (b) Settling time calculation
The gures will be analysed shortly.
2
4 The eect of controller gains on response
While experimenting with the system, it has been conrmed that each of the three components of PID has a dierent
eect on system behaviour. These are visualised and analysed below, backed by theory:
1. Proportinal gain
(a) Expected result of increasing component:
Higher amplitude of oscillations and thus longer settling time, i.e. the oscillations are bigger and last longer.
This follows directly from the equation P = K
p
err(t). Higher K
p
results in bigger changes (oscillations)
in the output, for any given change in error.
As the consequence of bigger changes in output, the steady-state error is decreased with high proportional
controller due reduction in rise time towards the set target.
It is worth noting that a pure proportional controller is not capable of eliminating a steady-state error. In
that case increasing proportional component would bring the actual output closer, but never quite equal
to target output.
(b) Expected result of decreasing component:
Lower amplitude of oscillations and quicker settling time. This follows as the opposite of increasing K
p
from the above given equation. component.
Also, the output response after settlign is further from the desired set-point, as compared to high propor-
tional.
(c) Actual result of changing proportional gain component:
In graph d), one can observe that there are comparably higher oscillations for desired response equal to 5

,
than for the same set-point value as for graph c).
Also, for the same time duration of 7.5 seconds (half-period, i.e. how often the target-value changes), the
response in graph d) seems to reach an elevation angle which is closer to target, than the response in graph
c). This agrees that higher proportional gain allows for reaching a smaller steady-state error.
(c) Lower-than-optimal proportional gain (d) Higher-than-optimal proportional gain
2. Integral gain
(a) Expected result of increasing component:
As a result of increasing the integral component, there is to be an expected increase in oscillatory behaviour.
This is deduced from the equation I = K
i

t
0
err(t) dt.. Increased K
i
will have the eect of enlarging the
sum of past errors and thus ampling the controllers action.
(b) Expected result of decreasing component:
Intuitively, a small integral gain will have the opposite eect of higher gain. Smaller accumulated sum
of errors will provide a slower correction to the measured position, thus taking longer time to approach
output where steady-state error is zero.
On the plus side, there should be less, if any, overshoot and the actual output will eventually reach desired
state.
3
(c) Actual result of changing integral gain component:
From graph e), where integral action is low, it is visible that within a given time of 7.5 seconds, the actual
output doesnt even reach the desired state: it can be seen approaching it slowly from below. This agrees
with above theory.
Graph f) displays small oscillation (while graph e) displayed none) in controller output. There is some
overshoot as well, and in this case the actual elevation angle doesnt settle within the half-period either.
(e) Lower-than-optimal integral gain (f) Higher-than-optimal integral gain
3. Dierential gain
(a) Expected result of increasing component:
The derivative gain constant K
d
multiplies the rate of change of the slope of the error, to calculate the
derivative term, i.e. D = K
d

d
dt
err(t). Higher K
d
results in smoother, more horizontal slope, thus
decreasing controller output.
Consequently, that increases the time it takes to reach the desired state.
(b) Expected result of decreasing component:
Inversly, low derivative gain component gives rise to higher slopes, and thus overshoot and often oscillation.
There is not expected any improvement in steady-state error, whether dierential gain is high or low, since
that is determined by integral term already.
(c) Actual result of changing dierential gain component:
Graph g) conrms the existance of higher oscillations with lower derivative term. In this case it also seems
that settling time is shorter than that of high K
d
.
On the other hand, thanks to smoother approach towards target value, there is no overshoot with higher
K
d
term, as seen in graph h).
(g) Lower-than-optimal dierential gain (h) Higher-than-optimal dierential gain
4
5 Result analysis and discussion
5.1 Evaluation of practical results
As mentioned earlier, the optimized response aimed at achieving minimal overshoot and settling time.
The overshoot visible in graph a) is equal 0.2 degrees, i.e. less than 1% of demand elevation angle. Since overshoot is
negligible and the shown response is a best possible solution for this system (assumption, evaluated at a later stage), the
underlying system can be considered a critically damped second order system.
Therefore, settling time for such system is known to be t
s
=
4
wn
. It is also known that y(
4
wn
) = 0.908. In the case
of a quanser, the steady state value is the desired elevation angle, 25

. But the quanser can be assumed to only begin


elevation from initial 5

. Taking their dierence into account, steady state value is calculated to be 0.908 20 +5(oset)
= 23.16

. The x-value corresponding to this is 31.58 seconds. But the oset is 30 seconds, thus the settling time for this
system is found to be 1.58 seconds.
Say about prioritizing between: 1) safety: minimal overshoot - would just need to increase/decrease X
2) rise time: might need to reach a certain level quickly, but can allow for higher oscillation later - would choose to
increase/decrease Y instead
3) settling time: dont know what to say here. Would probably need to be a mix of little overshoot and quick rise
time, i.e. allow some small overshoot for example, like in our case.
- say whether 1.58 seconds is a good time to reach 90.8 % of steady-state value?
- compare results and theory...?
5.2 Application of PID controllers
1. Heat Exchanger are used to control temperature in chemical and industrial processes where the temperature needs
to be kept constant for a long period of time and there may be rapid uctuation in temperature during reaction. In
this case y(t)would be heat while u(t) could be a multitude of things from valve switch, fan speed controlling ow ,
etc .
2. Satellite Antenna require precise setting to intercept signals sent from satellites orbiting in space, in this case we
have a similar plant to our helicopter as generally it would be a 3 DOF system. r(t) would the position of satellite
in space y(t) would be the polar co-ordinate reference of our antenna and u(t) would the angular velocity of the
antenne ( it would be a small value ).
5.3 Drawbacks of using PID Controller
PID controllers have many drawbacks, the most crutial one for this project would the inablity to use Ziegler-Nichol Tuning,
the reason being that to nd the ultimate gain we risk causing damage to the quanser as this involes increasing the
proportional term to really high values resulting in osillations with large amplitudes , we could tune it using a computer
simluation and nd out if the values provided by Ziergler-Nichol is better than the values we found.
6 References
1. Gene F.Franklin J.David Powell Abbas Emami-Naeini, Feedback control of dynamics systems (1994), third edition,
Chapter 4 (pages 185 - 186).
2. Fred White Principles of Control Engineering (1995)
3. Werner Leonhard 1976 Introduction to Control Engineering and Linear Control Systems Springer Intrernational
Student Edition (pages 147-178)
4. Joseph J. DiStefano(1995) Schaums Outline of Feedback and Control Systems, 2nd Edition
5

Вам также может понравиться