You are on page 1of 24

ROBT303 Linear Control Theory with Lab

Controller synthesis via root locus


Tohid Alizadeh

Academic year 2015-2016

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

1 / 39

Part I
Improving steady-state error:
PI and lag compensators

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

2 / 39

Denition of the PI controller


Imagine to use two components of the control action:

kp

the rst proportional to the error through a constant

the second proportional to the integral of the error through a constant

We get a

ki

Proportional-Integral (PI) controller


Z

u ( t ) = kp e ( t ) + ki

k
e ()d U (s ) = kp + i E (s )
s
0

The transfer function of this controller is

Gc (s ) =

kp s + ki
U (s )
s + zc
k
=
=K
, K = kp , zc = i
E (s )
s
s
kp

which has a pole at the origin, and a zero at


Tohid Alizadeh (NU)

ki / kp .

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

3 / 39

Denition of the PI controller


So, once

kp

is xed

ki zero close to the origin


ki zero far from the origin

low value of

high value of

Notice that, given an initial error

e (0),

the control variable

Z0

u (0) = kp e (0) + ki

u (0)

will be equal to

e ()d = kp e (0)

From time 0, the PI controller will start taking into account the integral of the
error: in practice, it will considerably increase the control amplitude if the error
does not decrease quickly enough
If no saturation is present on
an innite increase of

u (t ):

u (t )

(ideal case), a constant error

e (t )

will lead to

this is why the PI controller is not a BIBO stable

system
The integral action will never give up until the error is steered to zero
Now, let's see what are the benets of such a controller using the root locus

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

4 / 39

Adding an integrator improves steady-state error


If a proportional controller does not allow us to achieve a desired steady-state
error, we can insert a pole at the origin in

Gc (s ),

which compensates the

system dynamics
In this way, if

Gp ( s )

is Type-0, by adding one integrator will make

L(s )

Type-1,

with all the consequent steady-state improvements

9.3

A is:
e root
thout
sator;
n the
us with
sator
ded;
ontinues)
Typically, adding a pole will change the root locus, and decrease the maximum
value of the controller gain for which the closed-loop system is stable.
Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

5 / 39

How to preserve the transient behavior


To obtain a root locus similar to that of the original system, thus obtaining
analogous transient characteristics, we add a zero close to the origin

Figure 9.3
(continued)
c. approximately on the root locus with compensator
pole and zero added
Control Systems Engineering, Fourth Edition by Norman S. Nise

In this way, the angular contributions


of
zero
and
integrator
are approximately
Copyright 2004
by John
Wiley &
Sons. All
rights reserved.
canceled out, and the system type is increased
Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

6 / 39

PI controller: example
Consider the system

Gp ( s ) =

(s + 1)(s + 2)(s + 10)

Design a controller such that the closed-loop system has

e = 0

for a constant

reference, and a damping ratio for the dominant complex poles equal to 0.174
Procedure (let's do it with

rltool):

design a proportional controller to meet the transient requirements

add an integrator and a zero near to the origin

tune the position of the zero and the gain

to satisfy the transient

requirements with the new controller


Solution:

kp = 164.6 we obtain the desired


kp
Kp = lim L(s ) =
= 8.23 e = 0.108
Setting

s 0

Tohid Alizadeh (NU)

damping ratio, with

20

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

7 / 39

Figure 9.7

PI
controller:
example (cont'ed)
Ideal
integral compensated

system response and the


uncompensated system
Then, we
add the integrator
and a zero at 0.1.
response
of Example
9.1
Finally, we adjust the gain

K,

i.e.

K = 15.81,

will satisfy the requirements

Control Systems Engineering, Fourth Edition by Norman S. Nise


Copyright 2004 by John Wiley & Sons. All rights reserved.

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

8 / 39

uncompensated system
response of Example 9.1

PI controller: example (cont'ed)

Control Systems Engineering, Fourth Edition by Norman S. Nise


Copyright 2004 by John Wiley & Sons. All rights reserved.

t =3

Notice that the two step responses are similar until

s, after which the

eect of the integrator becomes apparent


The output

c (t ),

t = 3 s, converges very slowly (Ts ,2% = 17.5s ) to the


r (t ) = 1, while for uncompensated system Ts ,2% = 5.14s .

after

constant reference

This is due to the fact that the very slow pole of


the integrator is

almost

T (s )

which originates from

canceled by the zero of the PI controller: as a

consequence, the term of the step response associated to this pole will be:

very slowly decaying (large time constant)

with a small amplitude (small residue, due to the presence of the zero)

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

9 / 39

Lag compensator
In order to slightly reduce this eect, we can place the pole very near to the
imaginary axis (but not at the origin) and a zero with a slightly smaller time
constant

igure 9.9

. Type 1
ncompensated
ystem;
. Type 1
ompensated
ystem;
. compensator
ole-zero plot

Control Systems Engineering, Fourth Edition by Norman S. Nise

For reasons that will become Copyright


clear later,
this
is called
2004 by John
Wileycontroller
& Sons. All rights reserved.
Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

lag compensator

Academic year 2015-2016

10 / 39

Lag compensator (cont'ed)


To understand the eect of the compensator, let us consider a Type-1 process
To make it Type-2, i.e. to make

Kv = ,

we should insert one integrator in

Gc (s )
Instead, we add to the proportional term

kp

Gc ( s ) = K
In this way, we don't make

Kv = ,

Analogously to the PI controller,

a pole and a zero, as

s + zc
s + pc

but we increase it by a factor of

u (0)

will be equal to

Ke (0),

zc /pc

and then

u (t )

will

start increasing, but not indenitely, since


Constant

e (t ) = e

lim

u (t ) = K

zc
e
pc

Placing the pair pole-zero not too near to the imaginary axis will reduce the slow
convergence eect, but given a previously-determined

zc /pc

their distance will

increase (= no cancellation)

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

11 / 39

Lag compensator: example


Consider again the system

Gp ( s ) =

(s + 1)(s + 2)(s + 10)

Design a controller such that

the closed-loop system has a damping ratio for the dominant poles equal to
0.174

for a step reference is improved about 10 times with respect to using a

proportional controller which achieves the same damping ratio


Procedure (let's do it with

rltool):

design a proportional controller to meet the transient requirements

zc /pc

add pole and zero with a given

tune the position of the zero and the gain

kp

to satisfy the transient

requirements with the new controller


Solution:

kp = 164.6
e = 0.108

Setting

Tohid Alizadeh (NU)

we get the desired damping ratio, with

ROBT303 Linear Control Theory with Lab

Kp = 8.23

and

Academic year 2015-2016

12 / 39

Lag compensator: example (cont'ed)


A tenfold improvement on

means

e =
Therefore

e =
The improvement in

1
1 + Kp

Kp

0.108
10

= 0.0108 Kp =

1 e

= 91.59

from the old (uncompensated) to the new

(compensated) system is equal to

zc /pc ,

Kpnew
zc
= old =
pc
Kp
If we arbitrarily choose

= 0.0108

pc = 0.01,

then

i.e.
91.95
8.23

= 11.13

zc = 11.13pc = 0.1113

As an alternative, we can choose

pc = 0.001,

Remember to adjust the value of

and then

zc = 11.13pc = 0.01113

(after you place the poles of the

compensator) in order to place the poles of the closed-loop system at the desired
position

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

13 / 39

Lag compensator: example (cont'ed)

of
and
d

Let us compare the step responses of the uncompensated system, and the two
compensated systems

Figure 9.14
Step responses of the
system for Example 9.2
using different lag
compensators

We can see here that if the pair pole-zero is placed closer to the imaginary axis,
then a longer transient response is obtained
Control Systems Engineering, Fourth Edition by Norman S. Nise
Copyright 2004 by John Wiley & Sons. All rights reserved.

Engineering,
Edition
Normanas
S. Nise
As a consequence, it is good practice to place Control
the Systems
pole
and Fourth
zero
asbyfar
Copyright 2004 by John Wiley & Sons. All rights reserved.

possible from the imaginary axis, but not too much

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

14 / 39

Part II
Improving transient response:
PD and lead compensators

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

15 / 39

Denition of the PD controller


Imagine to use two components of the control action:

kp

the rst proportional to the error through a constant

the second proportional to the derivative of the error through a constant

We get an ideal

kd

Proportional-Derivative (PD) controller



u (t ) = kp e (t ) + kd e (t ) U (s ) = kp + kd s E (s )

The transfer function of this controller is

Gc (s ) =
which has a zero at
Tohid Alizadeh (NU)

kp
U (s )
= kp + kd s = K (s + zc ), K = kd , zc =
E (s )
kd

zc = kp /kd .
ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

16 / 39

Denition of the PD controller


So, once

kp

is xed

kd zero very far from the origin


kd zero near to the origin

low value of

high value of

Notice that the derivative action gives an anticipative action: if the error is
growing fast, the controller tends to make it decrease immediately
This kind of controller has three big problems:
1

First, it is not realizable in practice, since any physical system cannot have
more zeros than poles

Second, it would generate spikes in

e (t ))
3

u (t )

every time

r (t )

(and therefore

has a discontinuity (e.g., step)

Third, it would tend to amplify high-frequency components (i.e., sensor


noise)

For the moment though, let's assume that we can implement it, and let's
analyze its benet using the root locus

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

17 / 39

Eect of PD controller
We have already seen that an additional zero has an anticipative action on the
step response
A zero in

Gc (s ),

and therefore in

Therefore, the step response of

L(s ),

T (s )

also appears in

T (s )

can be made faster

Mp ,

More precisely, we can decrease the settling time maintaing the same
can decrease

Mp ,

or we

keeping the same settling time (or even decreasing it)

In general, a zero on the negative real axis tends to move the root locus
branches to the left, so we can even stabilize an unstable system
Root Locus

Root Locus

1.5

1.5
1
Imaginary Axis

Imaginary Axis

1
0.5
0
0.5

0.5
0
0.5

1.5

1.5

2
4

Tohid Alizadeh (NU)

1
Real Axis

2
4

ROBT303 Linear Control Theory with Lab

1
Real Axis

Academic year 2015-2016

18 / 39

Eect of PD controller: example


Consider the system

Gp ( s ) =

(s + 1)(s + 2)(s + 5)

Let us consider rst a proportional controller to achieve

= 0.4,

i.e.

Gc (s ) = 23.72

Ts ,2% = 4.26 s, and can be obtained


quite precisely by second-order approximation

The settling time for a step response is

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

19 / 39

Eects of adding a zero


Let's say that we are happy with the damping ratio (i.e., with the consequent
value of

Mp = 25.4%),

but we want a shorter settling time. This cannot be done

just by adjusting the gain

kp ,

so we add a zero

The ideal PD controller is dened as

4, 3, 2,

Gc (s ) = K (s + zc ), where
K to get = 0.4

we move

zc

at

and then adjust the value of

9.15
ed)
Figure 9.15
ensator Using ideal derivative
3;
compensation:
ensatora. uncompensated;
b. compensator
4
zero at 2;
(figure continues)

Notice that the settling time decreases at 2.14 s, 1Control


.64 s,Systems
1.33 Engineering,
s, as the Fourth
zero Edition
is
by N
Control Systems Engineering, Fourth Edition by Norman S. Nise
Copyright 2004 by John Wiley & Sons. All r
moved towards the origin.
Copyright 2004 by John Wiley & Sons. All rights reserved.
Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

20 / 39

Lead compensator
Given the zero at
Condition

pc  zc

zc ,

we can insert a pole at

pc ,

with

pc  zc

is needed to avoid canceling the derivative action introduced

by the zero, at the same time adding a natural low-pass ltering of the control
action
In this way, we can practically realize the controller, at the same time avoiding
spikes and noise amplication
This practical implementation of the PD controller is also called

compensator,

lead

the transfer function of which is

Gc (s ) = K

s + zc
, K = kd , zc = kp /K
s + pc

Notice that, given any error signal

e (t )

with bounded

e (t ),

the control action

will remain bounded, which did not happen for the ideal PD controller
Now let's see how to design practical PD controllers (i.e., lead compensators)
using the root locus

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

21 / 39

Lead compensator: example


Consider the system

Gp ( s ) =

s (s + 4)(s + 6)

Design an ideal PD controller, and then a lead compensator, such that

the closed-loop system has

Ts ,2%

Mp

equal to 30%

for a step reference is improved twice (i.e. it's twice smaller) with

respect to using a proportional controller which achieves the same

Mp = 30%
Procedure for ideal PD controller (let's do it with

rltool):

design a proportional controller to meet the transient requirement

Mp = 30%

graphically compute Ts ,2% , and drag the new vertical line corresponding to
half of the previous value of Ts ,2%

add the zero of the ideal PD controller such that the root locus branches

intersect at a point with the required Mp and the new value of Ts ,2%
adjust the controller gain such that the poles are placed as required

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

22 / 39

Lead compensator: example (cont'ed)


rltool):
the zero closer

Procedure for lead compensator (let's do it with

starting from the previous design, move

to the imaginary

axis, since the additive pole will partially compensate the anticipative eect
of the zero

then, place the pole at high frequency, such that the root locus branches

intersect at a point with the required Mp and the new value of Ts ,2%
adjust the controller gain such that the poles are placed as required

Remember: after each design phase, check if the second-order approximation


holds, both looking at the position of the closed-loop poles, and at the
close-loop step response
If needed, slightly adjust the gain, or the position of the pole/zero of the
controller to meet the requirements
Solution:

Setting

kp = 63.2

we get the desired

Mp ,

which corresponded to

Ts ,2% = 3.97

s.

We move the vertical line corresponding to the settling time requisite to the left
until we reach the value equal to 3.97/2

Tohid Alizadeh (NU)

= 1.99

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

23 / 39

Lead compensator: example (cont'ed)


Solution (cont'ed):

To make the root locus branches pass at the desired points by adding a zero, we
see that this latter must be placed at

s = 6,

and then adjust the gain at about

31.2
To design the lead compensator, we can make dierent choices, in any case
moving the zero to values to the right of
(a) if
(b) if
(c) if

zc = 5,
zc = 4,
zc = 2,

then
then
then

pc = 42.96
pc = 20.09
pc = 8.97 (second

6:

order approximation not valid)

Figure 9.29
Uncompensated
system and lead
compensation
responses for
Example 9.4

Tohid Alizadeh (NU)

ROBT303 Linear Control Theory with Lab

Academic year 2015-2016

24 / 39

You might also like