You are on page 1of 29

4 CONTROL DESIGN

The first step when starting a control design is to collect the required
specifications and translate them into a design point in the complex plane
representing the dominant poles that the closed loop system needs to have to
satisfy the requirements. Specifications are usually given in term of natural
frequency (e
0
) and damping factor (), or other performance factors (chap.
2) that can always be expressed in function of e
0
and .
The aim of control design is therefore to select gain, poles and zeros of G
c
(s)
so that the Root Locus (RL) goes through the desired closed loop roots
(poles), i.e. the selected dominant 2
nd
order poles according to the control
specifications requirements (design point).
Example:
Specifications: e
o
=2.5rad/s, ,=0.707
Dominant poles = o je
d
A general controller structure will be of the form:
Where k is the root locus parameter.
We choose z
i
and p
i
so that the RL goes through our design point.
We can have a simple proportional controller (k), an integral controller a
first order controller (like for example a PI controller), a second order
controller (like a PI plus pole/zero compensation) and so on.
We can gradually build up the number of zeros and poles of the controller to
suit our requirements.
RULE
G
c
(s) should have equal number of poles and zeros
In general zeros pull RL over to the left, therefore they make the system
more stable (need to be in the left hand side of the complex plane though!).
Design point
However zero can introduce noise and they can give problems with the
implementation.
4.1 Proportional Control
In this case the feedback control signal is made to be linearly proportional to
the system error. The control law is u=ke.
As k varies, the closed loop poles can be placed only on the root locus with
G
0
(s) = G(s)H(s).
Example
Let H(s)=1
The root locus is:
If the closed loop design point is represented by P, that specification cannot
be met.
Disadvantages:
1) The root locus is determined only by G(s) and so is the closed loop
response. Therefore we cannot meet a general closed loop specification
(no control!)
2) The steady state (SS) error is determined by the plant. If the system is
type 0, then will get a SS error to a step input; if the system is type 1, will
get a SS error to a ramp input, etc. This SS error gets smaller as the gain
is increased, but high values of k may cause instability.
For higher order systems, large values of proportional feedback gain will
almost always lead to instability and there is an upper limit to the size of the
gain if we are to achieve a well-dumped stable response. This limit may still
have an unacceptable steady state error.
The standard way to improve the SS accuracy of the control without adding
extremely high proportional gain is to introduce integral control.
This will place an open loop pole at the origin and the steady state error can
be improved since this action increases the system type by one.
In the case of an ideal integral controller:
EXAMPLE
Let
Adding the system becomes type 1 (zero steady state erro to a
step input), but the root locus is pulled to the right. The natural frequency e
0
decreases and the system is closer to instability. Furthermore the root locus
is again fully determined by G(s) and the integrator, so we cannot meet a
general closed loop specification.
Also if the plant itself has already a desirable transient response generated
by a closed loop pole, if we add only a pole at the origin to increase the
system type and improve the steady state error, the new root locus can no
longer give that specific pole.
4.2 Proportional + Integral control (PI)
A solution to these problems could be obtained by adding an integral term to
the proportional controller, resulting in the Proportional plus Integral
control.
The control law is:
For which:
This feedback structure has the primary virtue that in the steady state its
control output can be a non-zero constant value even if the error signal at its
input is zero.
This feature means that a constant disturbance can be cancelled by the
controller integrator`s output even while the system error is zero.
Can be written as
Where
In the design procedure we must choose the controller zero in a way that the
closed loop design point is on the root locus and then adjust the gain K
P
to
get the required closed loop dominant poles.
Note
In the case in which G(s) has already a good transient response generated by
a closed loop pole, we can choose the zero close to the pole at the origin.
Now the angular contributions of the compensator zero and compensator
pole cancel out, the desired point is still on the root locus and the system
type has been increased.
DESIGN EXAMPLES
4.2.1 1
st
order type 0 plant
Let e
0
~ 10rad/s
Use PI controller
STEP 1
Specifications e
0
= 20rad/s, =0.707
Closed loop design point: P=-j
d
P=-14.14j14.14
Need to choose a so that the root locus passes through P.
STEP 2
Angle criteria:
a = 21.8
(Alternatively get a protractor and calculate angles graphically!)
STEP 3
Gain Criteria
1 P
V = Distance between P(-o, e
d
) and the pole in zero (0, 0):
2 2 2 2
1
[ 0] ( 0) [ 14.14 0] (14.14 0) 20
P d
V o e = + = + =
Same concept for calculating
2
and
P z
V V
2 P
V = Distance between P(-o, e
d
) and the pole in (-10, 0):
2 2 2 2
2
[ ( 10)] ( 0) [ 14.14 10] (14.14 0) 14.75
P d
V o e = + = + + =
z
V = Distance between P(-o, e
d
) and the zero in (-21.8, 0):
2 2 2 2
[ ( 21.8)] ( 0) [ 14.14 21.8] (14.14 0) 16
z d
V o e = + = + + =
(Alternatively get a ruler and calculate the modules graphically!)
Notes on the above design
1) Simplified method valid only for first order plants
The closed loop system with the controller in the previous case would be:
Closed loop transfer function:
2
10 ( )
10 ( ) ( 10)
( ) =
10 ( )
10( 1) 10
1
( 10)
o
k s a
k s a s s
G s
k s a
s k s ka
s s
+
+ +
=
+
+ + +
+
+
Comparing the denominator with the general second order polynomial in
function of and
0
2 2
0 0
+2 s+ s e e
and remembering that it was =0.707;
0
=20rad/s, we have the following
two equations:
0
2
0
2 10( 1)
=10ka
k e
e
= +

From which we can design the values of k and a of the PI controller


K=1.828; a=21.88 (in agreement with the general method explained in
the previous paragraph).
2) Fast as possible design
Get a good design for the zero a with the highest possible bandwidth,
=0.707;
0
=?
If we choose a high value of a, for example a=450, the PI controller
would be:
And the root locus:
If we also choose as closed loop roots the largest interception of the root locus
with the line at constant =0.707 (P), the resulting
0
is around 620rad/s. Using
the gain criteria, we can also calculate the gain of the controller needed to obtain
this closed loop pole which is around k=80. Therefore:
This design will result in K
P
=80 and K
I
=36000 and in the following
implementation:
Two things will stop us from doing this:
a) Noise
Note the large numbers in the controller G
c
(s). Just a small quantity of
noise will be amplified a lot and it will result in the output of the
controller u going between limits:
b) High frequency poles and zeros
The model of the plant is very often only an approximation for
frequencies of interest.
For example let`s consider a 1
st
order plant of a water heater:
We are only interested in closed loop natural frequencies from 0.1 to 0.5,
comparable with the intrinsic characteristics of the system.
If for example we get
0
=100, then there are other poles we should
consider in our model of the plant.
Therefore if w0 is too high, the model expressed by G(s) may not be valid
anymore.
3) Effect of zero placement on closed loop response
Consider the design in 4.2.1 with some approximations:
The denominator gives the
closed loop poles at P
Note that we have a zero in the CLTF.
In fact all zeros in the forward path will be in the CLTF, while zeros in the
feedback path will not be.
The zero will contribute to an overshoot to the closed loop step response:
- If the zero is far to the left of P, the overshoot will be small;
- If the zero ~ to the real part of P or to the right of it, the overshoot can be
large.
Note: zeros will not give any overshoot in ramp response it does not matter
for tracking control.
If the overshoot is of concern for our design, then we can get rid of it in two
ways:
i) By PRE-FILTERING
Place a filter with a pole equal to the controller zero on the reference signal. The
CLTF will be now:
No zeros OK!
ii) Put the zero in the feedback path
In the steady state we want c = c, so we have to divide the zero by 20. At the
same time to keep the loop gain the same, we have to multiply by 20 the
forward integrator. So we have:
This is called rate feedback and it is very common in servo-systems.
4.2.2 2
nd
order plants with PI control
Consider
8
( )
( 2)( 4)
G s
s s
=
+ +
(-2 is the dominant pole open loop e
0
~2rad/s)
Design a PI controller to obtain the following closed loop specifications:
e
0
~4rad/s =0.707
( )
( )
c
k s a
G s
s
+
=
The design point will be:
P=-2.83j2.83
Angle criteria:
1 2 3
+ - 180
p p p z
0 0 0 0 + =
1
1
2.83
=180 -tan 135
2.83
p
0

| |
=
|
\ .
1
2
2.83
=180 -tan 106
0.83
p
0

| |
=
|
\ .
1
3
2.83
=tan =67.5
4 2.83
p
0

| |

|

\ .
135 +106 67.5 - 180
z
0 + = 128.5
z
0 =
1
1
180 tan 128.5
( )
2.83
tan =51.5
2.83
d
z
a
a
e
0
o

| |
= =
|
|

\ .
| |

|

\ .
a = 0.58
Gain Criteria
1 2 3
8
1
z
p p p
kV
V V V
=
1 P
V = Distance between P(-2.83, 2.83) and the pole in zero (0, 0):
2 2
1
( 2.83 0) (2.83 0) 4
p
V = + =
V
p2
= Distance between P(-2.83, 2.83) and the pole in (-2, 0):
2 2
2
[ 2.83 ( 2)] (2.83 0) 2.95
p
V = + =
V
p3
= Distance between P(-2.83, 2.83) and the pole in (-4, 0):
2 2
3
[ 2.83 ( 4)] (2.83 0) 3.06
p
V = + =
V
z
= Distance between P(-2.83, 2.83) and the zero in (-0.58, 0):
2 2
[ 2.83 ( 0.58)] (2.83 0) 3.61
z
V = + =
So we have
8 3.61
1 k=1.25
4 2.95 3.06
k
=

Finally
1.25( 0.58)
( )
c
s
G s
s
+
=
Note on the above design
The closed loop transfer function of the combined controller plant system is:
1.25( 0.58) 8 10( 0.58)
( 2)( 4) ( 2)( 4)
( ) =
1.25( 0.58) 8 ( 2)( 4) 10( 0.58)
1
( 2)( 4) ( 2)( 4)
o
s s
s s s s s s
G s
s s s s s
s s s s s s
+ +
+ + + +
=
+ + + + +
+
+ + + +
3 2
10( 0.58) 10( 0.58)
( )
( 2)( 4) 10( 0.58) 6 18 5.8
o
s s
G s
s s s s s s s
+ +
= =
+ + + + + + +
The system closed loop poles, obtained from the denominator of the previous transfer
function imposing:
3 2
6 18 5.8 0 s s s + + + =
are: -2.83j2.83 and -0.36.
The design procedure was successful in creating a closed loop systems with
complex poles in 2.83j2.83, but the dominant pole of the system is effectively
the one in -0.36 and not those required by the specifications!
If we analyse the time domain response of the closed loop system to a step input
it results like the blue trace in the following figure, much slower than expected.
It is therefore concluded that the designed PI controller is not optimal and does
not verify the requirements.
If the closed loop pole and zero cancelled or nearly so, then the design would be
ok. Therefore the best response is obtained using the controller:
( 2)
( )
c
k s
G s
s
+
=
where the zero of the PI controller has been chosen to cancel the dominant open
loop pole of the plant. The open loop system becomes:
( 2) 8 8
( ) =
( 2)( 4) ( 4)
OL
k s k
G s
s s s s s
+
=
+ + +
The resulting root locus would be:
Where for =0.707,
0
=2 2rad/s 2.8rad/s e ~
4.2.3 Type 1 plants with PI control
Example:
10
( )
( 2)( 8)
G s
s s s
=
+ +
Let`s try to use a PI compensator
( )
( )
c
k s a
G s
s
+
=
Trying to place the zero a of the PI at -2 to cancel the dominant plant pole.
The OL transfer function will become:
2
( 2) 10 10
( ) =
( 2)( 8) ( 8)
OL
k s k
G s
s s s s s s
+
=
+ + +
The system will be type 2 and the root locus will be:
and the system is UNSTABLE!!
If we want to insist on using a PI, then we can apply an empyrical rule, the so
called rule of thumb.
Put
0
( ) 2
= - 0.4
5 5
open loop
a
e
= =
e
0
(open loop) ~ the dominant open loop pole (-2 in this case). Now the root
locus will be the following and the closed loop system much slower:
Conclusions on PI
Good for 1
st
order type 0 systems
Not good for 2
nd
order type 0 systems with real poles if the 2 poles are
close. If the second plant pole is large (eg.
1
( 2)( 20) s s + +
) then it is
OK. In fact the system will behave more like the 1
st
order system
1
( 2) s +
Not good if the open loop 2
nd
order type 0 system has complex poles
Poor for type 1 systems
4.2.4 Implementation
Use of Inverting operational amplifiers:
Difference between reference and feedback signal to generate the error signal:
The PI controller implementation is:
Where
2
2
1
1
=- with and T=R
P
out
P
e
K s
V R T
K C
V s R
| |
+
|
\ .
=
If 1/T is small, T can be large R2 1MO
4.3 PI+Lead control
For cases in which the PI control is poor, we can use a second order controller obtained
adding a further zero/pole element to the PI structure:
PI LEAD
The lead zero is to the RIGHT of the lead pole in the s plane.
Why it is called LEAD?
Note that the dc gain (s) is equal to 1 (or zero dB)
The frequency response of this element is:
GAIN:
PHASE:
The lead element gives a phase lead in the region:
T T
o
e s s
1
c
c s
b s
s
a s k
s G
c
s
+
+ +
= b re whe
) (
) ( ) (
) (
T
s
T
s
G
o
o
o o
+
|
.
|

\
|
+
=
> =
1
: as written be ore can theref part lead The
) 1 (with
b
c
: define Let`s
T
j
T
j
G
o
e
e o
+
+
=
1
e
1/T o/T
Gain
e
1/T o/T
Phase
+90
|
.
|

\
|
+ Z |
.
|

\
|
+ Z = Z
T
j
T
j G
o
e e
1
Note that the high frequency gain is greater than the low frequency one. So if the
feedback signal is noisy, we should not let o get too high, otherwise too much noise will
be amplified by the controller.
4.3.1 Second order type zero systems
Let
) 4 )( 2 (
8
) (
+ +
=
s s
s G
Design a PI+LEAD controller to get
0
=4 rad/s; =0.707 in the closed loop system.
This corresponds to a design point P= XpYp = -2.83j2.83.
STEP 1: Choose a (the PI zero) to cancel the dominant open loop pole: a=2
Step 2: ANGLE CRITERIA

180
2 1
= + +
z p
u u u u
Xp=-2.83
Yp=2.83

5 . 67
4
tan
1
1
=
|
|
.
|

\
|

=

P
P
X
Y
u

45 tan
1
=
|
|
.
|

\
|
=

P
P
X
Y
u

135 180
2
= = u u

5 . 22 =
z p
u u
By geometry | u u + =
p z
| u u =
z p

5 . 22 = |
Step 3: Choose b to the left of the real part of the design point P if possible
Eg. b=-5.5
Then:

5 . 46
5 . 5
tan
1
=
|
|
.
|

\
|

=

P
P
z
X
Y
u

24 5 . 22 5 . 46 = = = | u u
z p

24 tan
1
=
|
|
.
|

\
|

=

P
P
p
X c
Y
u c=-9.5
Step 4: GAIN CRITERIA:
1
8
4 3 1
2
=
V V V
kV
=
1
V distance between P(Xp, Yp) and the pole in c (-c, 0):
2 2 2 2
1
) 83 . 2 ( )] 83 . 2 ( 5 . 9 [ ) 0 ( )] ( [ + = + = Yp Xp c V
Same concept for calculating
4 3 2
V V V
In the end k=2.7, and the PI+LEAD controller is:
) 5 . 9 (
) 5 . 5 )( 2 ( 7 . 2
) (
+
+ +
=
s s
s s
s G
C
What is o? o is the ratio between the pole and the zero of the lead part of the compensator. In
this case
9.5
1.7
5.5
o = ~
The high frequency noise () will be amplified by o compared to the one at very low
frequency (=0). In this case the high frequency noise will be 1.7 times the low frequency
one, but during transients the closed loop will respond with ~
0
= 4rad/s.
Another measure of the controller effect on noise is the ratio between the controller gain at
high frequencies K and the one at the design frequency K
0.
In the previous example we have:
0
0
4
( )
2.7
1.35
( ) 1.99
Gc j
K
K Gc j
e
e e
e
e e

= =

= = =
4.3.2 Type 1 systems
Example:
Consider again:
10
( )
( 2)( 8)
G s
s s s
=
+ +
In 4.2.3 using a PI controller we could only get to a closed loop
0
~0.4rad/s.
Let`s try now for a modest increase in natural frequency using a PI+LEAD controller and
asking for the following closed loop specifications:
e
0
=2.5rad/s =0.707
PI+LEAD:
( ) ( )
( ) <
( )
c
k s a s b
G s b c
s s c
+ +
=
+
The design point will be:
P=-1.761.76
STEP 1: Choose the PI zero a to cancel the plant dominant pole: a=2.
STEP 2 Angle Criteria
1 2 3
+ + - 180
p p p p z
0 0 0 0 0 + =
1
1 2
1.76
= =180 -tan 135
1.76
p p
0 0

| |
=
|
\ .
1
3
1.76
=tan =16
8 1.76
p
0

| |

|

\ .
135 +135 16 - 180
p z
0 0 + + = 106
p z
0 0 = 106 | =
We cannot choose b to the left of the real part of the design point P!!
At this point we have different possible choices. The limit choices are:
i) b very small (eg. b=0.1)
ii) b as large as possible considering the value of the angle | (eg. b=1.22).
i) If b=0.1
1 1
1
1.76 1.76
106 tan 180 tan 106
1.76 1.76 0.1
1.76
tan =27.3
1.76
p z
c
c
0 0

(
| | | |
= =
| | (

\ . \ .

| |

|

\ .
c ~ 5.2
Gain Criteria
1 2 3
10
1
z
p p p p
kV
V V V V
=
1 P
V = V
p2
=Distance between P(-1.76, 1.76) and the double pole in zero (0, 0):
2 2
1 1
( 1.76 0) (1.76 0) 2.5
p p
V V = = + =
V
p3
= Distance between P(-1.76, 1.76) and the pole in (-8, 0):
2 2
3
[ 1.76 ( 8)] (1.76 0) 6.5
p
V = + =
V
p3
= Distance between P(-1.76, 1.76) and the pole in (-5.2, 0):
2 2
[ 1.76 ( 5.2)] (1.76 0) 3.9
p
V = + =
V
z
= Distance between P(-1.76, 1.76) and the zero in (-0.1, 0):
2 2
[ 1.76 ( 0.1)] (1.76 0) 2.4
z
V = + =
So we have
10 2.4
1 k=6.6
2.5 2.5 6.5 3.9
k
=

Finally
6.6( 2) ( 0.1)
( )
( 5.2)
c
s s
G s
s s
+ +
=
+
The closed loop transfer function will approximately be (tedious calculation by hand, do it
with matlab):
2
66( 0.1)
( 0.11)( 3.52 6.19)( 9.56)
s
CLTF
s s s s
+
=
+ + +
As we can see the dominant pole is in -0.11 and not as required by the specifications (which
are represented by the second order polynomial). However, since the zero at -0.1 and the pole
at -0.11 are so close, their contributions almost cancel out and the response is good (check it
with MATLAB!!!). There will still be, though, a small effect of the pole at -0.11.
ii) If b = 1.22
1 1
1
1.76 1.76
106 tan 180 tan 106
1.76 1.76 1.22
1.76
tan 1
1.76
p z
c
c
0 0

(
| | | |
= =
| | (

\ . \ .

| |
~
|

\ .
c ~ 100
Applying the Gain Criteria we obtain K~217, therefore the controller will be:
217( 2) ( 1.22)
( )
( 100)
c
s s
G s
s s
+ +
~
+
The noise gain of the lead component
100
82
1.22
o = ~
Therefore the controller can introduce some noise. (Also the proportional gain Kp=217 is very
high!).
Note that the zero is to the right of the real part of the design point and hence a large
overshoot can be expected (check with Matlab!).
4.3.3 Complex plants
Example: 2
1
( ) G s
s As B
~
+ +
The PI+LEAD controller is good for this case too. We can have two possible choices:
a) Open loop poles very oscillatory
Choose a complex nominator for the PI+LEAD controller to cancel (or very near) the open
loop poles:
2
( )
( )
( )
c
K s As B
G s
s s c
+ +
~
+
The root locus will be:
The closed loop poles will be at P1 (design point) and at P2, but this last one very close to the
zero. The effect of P2 on the response will therefore be small as it cancels with the effect of
the zero.
b) Open loop poles not oscillatory
Choose the classical form of the PI+LEAD controller with real zeros:
( ) ( )
( )
( )
c
k s a s b
G s
s s c
+ +
=
+
The root locus will go from the complex open loop poles to infinity; P2 will now be the
dominant pole. The following two figures show two possible types of root locus in this case:
4.3.4 NOTES
1) PID control
After the PI this is the most common controller. Its transfer function is:
2
K ( )
K K
( ) +K s =
P I
d
d d I
c P d
K K
s s
K
G s K
s s
+ +
= +
The numerator may have complex or real zeros
( )( )
( )
c
K s a s b
G s
s
+ +
=
It`s easy to design a PID controller. Let`s consider an example:
8
( )
( 2)( 4)
G s
s s
=
+ +
Choose a to cancel the dominant plant pole. In this case a=2. After that the design
follows exactly the one of a PI controller.
Note that the differential term is very noisy. Therefore we will often need some
filtering (usually in the feedback loop). If we place a first order filter in the feedback
loop like for example
( )
f
K
s c +
the new overall controller will result in:
( ) ( )
( )
( )
tot
c
k s a s b
G s
s s c
+ +
=
+
equivalent to a PI+LEAD controller. The presence of the filter will affect the closed
loop response, so we have to design PID and filter together.
2) If we don`t want to increase the system type, then there is no need for an integrator.
The design is much easier and we can simply use a lead controller (closer to a
derivative action):
( )
( ) <
( )
c
s b
G s b c
s c
+
=
+
Or a Lag controller (closer to an integral action):
( )
( ) <
( )
c
s d
G s e d
s e
+
=
+
Or a lead-lag controller
( ) ( )
( ) < <
( ) ( )
c
s b s d
G s e d b c
s c s e
+ +
=
+ +
This will work properly for both 1
st
and 2
nd
order plants
3) If the plant contains zeros eg.
3
( )
( 2)( 8)
s
G s
s s
+
=
+ +
they can help us in the
control design, but in this case the controller may have a number of zeros lower than
the number of poles.
4.3.5 Implementation of a Lead/Lag controller
3 1 2
1
1 2 2
1
k= = T=R
out
i
k s
V R R R T
C
V R R R
s
T
o
o
o
| |
+
|
+
\ .
=
+
+
Lead Controller: o>1; Lag controller o<1.

You might also like