Professional Documents
Culture Documents
7 Analytical methods
The previous methods showed quite clearly the significant overhead in manpower costs and
potential process disruption through manual tuning A more effective means of controller
tuning i s to use system knowledge transfer f unctions,and formulae f or time responses to
reference and disturbance inputs
■ Problem
Use the ultimate-cycle method to obtain the gains for (a) P action, (b) PI action, and (c) PID
action. Analyze the unit-step response of each design. The plant is
10
G p (s) =
2s 3 + 12s 2 + 22s + 12
■ Solution
The ultimate-cycle method starts by using proportional action only. The transfer function for P
action with this plant is
C(s) 10K P
= 3
R(s) 2s + 12s 2 + 22s + 12 + 10K P
The characteristic equation is
To apply the ultimate-cycle method, we must first find the ultimate period Pu and the associated
gain K Pu . To do this analytically, we note that sustained oscillations occur when a pair of roots
is purely imaginary and the rest of the roots have negative real parts. To find when this occurs,
set s = jωu , where ωu is the as-yet-unknown frequency of oscillation. Then s 2 = −ωu2 and
s 3 = − jωu3 . Substituting these into the characteristic equation we obtain
To satisfy this equation, both the real and imaginary terms must be zero. So we obtain two
equations to solve for the two unknowns ωu and K Pu . These are
The latter equation yields the nonoscillatory solution ωu = 0 as well as the one of interest—
namely,
√
ωu = ± 11 = ±3.317
Substituting this into equation (2), we see that the ultimate gain is K Pu = 12. The ultimate
period is
2π 2π
Pu = = √ = 1.89
ωu 11
For sustained oscillations to occur, as required by the ultimate-cycle method, the third root
must be negative. The first two roots are s1 = jωu and s2 = − jωu . The third root can be
determined from guide 10 of the root locus guides. Dividing equation (1) by 2 gives
s 3 + 6s 2 + 11s + 6 + 5K P = 0
Guide 10 says that the sum of the three roots must be −6. Thus,
+ jωu − jωu + r3 = −6
which gives r3 = −6. Since r3 < 0, the assumption of neutral stability is justified. Thus the
solutions for ωu and K Pu are valid and can be used with Tables to compute the PID gains.
We obtain the following results. For P action, K P = 0.5K Pu = 6. For PI action,
KP
K P = 0.45K Pu = 5.4 TI = 0.83Pu = 1.57 KI = = 3.44
TI
For PID action,
KP
K P = 0.6K Pu = 7.2 TI = 0.5Pu = 0.947 KI = = 7.62
TI
For PI action,
C(s) 54s + 34.4
= 4
R(s) 2s + 12s 3 + 22s 2 + 66s + 34.4
The unit-step responses are shown in Figure 11.4.2. Note the improvement as the control
action is changed from P to PI to PID. First, the steady-state error is eliminated, and then the
oscillations and settling time are reduced.
Note that this analysis uses only proportional control, but the gains for other control
laws can be computed with these results. The power of the method is shown by the
fact that we did not have to solve for the roots of a cubic or quartic equation. Note also
that I action applied to this system would result in a fourth-order system to analyze—a
difficult task—were it not for the Ziegler-Nichols method.
The Ziegler-Nichols methods are used to obtain gain values that are “in the ball-
park.” The engineer can use these values as a starting point for tuning the gains to obtain
improved performance.
Figure 1 Unit-step responses of the
controllers obtained in Example 1.
(a) P action. (b) PI action. (c) PID action.
2 2
1.8 1.8
1.6 1.6
1.4 1.4
1.2 1.2
c(t)
c(t)
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 5 10 15 0 5 10 15
t t
(a) (b)
2
1.8
1.6
1.4
1.2
c (t )
1
0.8
0.6
0.4
0.2
0
0 5 10 15
t
(c)
■ Problem
Consider the PID controller designed in Example 1. Use the root locus plot to adjust the
gains to reduce the overshoot.
■ Solution
The closed-loop transfer function for PID control of this plant is
C(s) 10K D s 2 + 10K P s + 10K I
= 4
R(s) 2s + 12s 3 + (22 + 10K D )s 2 + (12 + 10K P )s + 10K I
When the controller transfer function is expressed in the form
1
G(s) = K P 1+ + TD s
TI s
we can see that changing the proportional gain K P changes the proportional, integral, and
derivative gains by the same factor. For PID action, the Ziegler-Nichols settings give the following
values of K D and K I in terms of K P .
KP KP
K D = TD K P = 0.236K P KI = = = 1.056K P
TI 0.947
Thus, we can express the characteristic equation as
2s 4 + 12s 3 + (22 + 2.36K P )s 2 + (12 + 10K P )s + 10.56K P = 0
and the root locus equation in terms of the parameter K P is
s 2 + 4.24s + 4.47
1 + 1.18K P =0
s(s 3 + 6s 2 + 11s + 6)
Thus the root locus parameter is K = 1.18K P .
The root locus plot is shown in Figure 2. When K P has the value recommended by the
Ziegler-Nichols method, K P = 7.2, the roots are s = −0.77 ± 2.67 j, s = −1.98, and s =
−2.48. The dominant root is marked on the root locus, and has a time constant of τ = 1/0.77
= 1.3 and a small damping ratio of ζ = 0.28. The root locus shows that decreasing K P from 7.2
increases the time constant slightly to τ = 1/0.523 = 1.9 but increases the damping ratio quite
a bit. Thus the system will respond somewhat more slowly, but the overshoot should be greatly
reduced to perhaps none at all.
The root locus shows that we can improve the response by decreasing K P until the dominant
root has a damping ratio of 1. This occurs at the breakaway point (s = −0.523) and requires that
K P = 0.306, K D = TD K P = 0.236(0.306) = 0.070, K I = K P /TI = K P /0.947 = 0.323.
The improved step response, marked “Tuned,” is shown in Figure 4, along with the response
obtained with the Ziegler-Nichols settings, marked “Original.”
2
Imaginary Axis
–1
–2
–3
–5 –4 –3 –2 –1 0 1 2 3 4
Real Axis
Figure 4 Unit-step 1.5
responses for Example 2
Original
1
c (t )
Tuned
0.5
0
0 2 4 6 8 10 12
t
Review
olved Proble� I
_, �f'Ch/,..,, 1J1. krrnh,, i/J"' .rsna•
O:f � ,IQ Mar �" �
it ,._.�}nQJfy: .tluJ, 4n• t,-,,,t M� fr, &y
"( S,u��n1,� ! 11��//4/;'4.,,_,.
I I
I I
J : /g
•
t ►
j
� . 2.$ ,
�
i'
•
.ao ,
I
Tiu. sysk,,,,,, i:J $J:o,J,lc. Y .20J ... I., ls le "> o r;-r < /J.S.
I
O;.n«. Jc> O'
1 '1 • • ,. I "'"
EXAMPLE 3 Ultimate-Cycle Method with Routh analysi creterion
Example
Analytically find the parameters of the PID controller based on Ziegler-Nicholas con-
tinuous cycling Method for the control system below
Solution:
Based on limit of stability, ultimate sensitivity method.
Tuning Table:
P: K p = 0.5 K u
Pu
PD : K p = 0.45K u , TI =
1.2
Pu P
PID : K p = 0.6 K u ,TI = , TD = u
2 8
- ::f7 CJ � �J ru-q� C-ks_,vc::__ � u t-c
l,< , f-&-
J
- 9�01-..
Qs) - �
"-!�1
-
r2.{s; S3 + ..2s ')._ +- s -f- r
I + K
� >+2-.s '.)_ +s+1
(
I =---
5 ')....
S +2s +_s+r +k
�
J�s /k_ c./4__0vc_�,-f h-c
Tl �c__�'-- i:,�_J
_'.)
+s + (/+k)
)._
� -f-)_� D �
s 3
j_ 1 0
3 I
0
2--
__s
<J__
)._ I .-,-IL 0 5' ?- t-Mc 0
S' r i- {L;_
0
I
-s
:;;...- � f-1.<:)
-s
I 1.. t-k
s (1 -KL) s
€)
0
I
').. - ( 1--t-1CJ
=-s-
�
--s
fuv 'VlA.� vr-L.._,A
/-It(
I D
-r
_;:) :LS
}-
--r ;i_ - 0
• �) s ==- :: FT .=:
� l� ryq�� �
1A.) ==- i yJ (s,
�ol p� of -fk-:::::..
1
.J_ ..--
.__.. j_ re-c (
-
y
_;)-
� D<f,�l � J--
- Or6
----- I
=- o, > - c>,r�s
fr-
Analytical tuning based on Root locus
Review:
Magnitude criterion
Definition 3 .5.5. The root locus must everywhere satisfy the criterion that
|KG(s)H(s)| = 1
Phase criterion
Definition 3.5.6. The root locus must everywhere satisfy the criterion that
These criteria are always satisfied and so they will guide the understanding and sketching root
locus and designing with the root locus.
(a) Consider a unity feedback control system shown in Fig. 2. The plant transfer function is
10
Gp (s) =
s(s + 2)(s + 8)
i. State with reason if a proportional controller can achieve closed loop poles at −2 ±
√
j2 3. (Show all your working), Also explain how we can control the system with
closed loop poles on the mentioned poles.
ii. Design a proportional derivative controller for the system to achieve closed loop
√
poles at −2 ± j2 3.
Example 2 i Solution
pb1cs �� � J �
Jw
J .:L"3,
-g
fo ec=-TT �
('
� tt:;_ d �s-, J
y e- c,) �c1_ r StSp r57 4
-2 t.J2V3
- � rlt> tt5
r�
fcl�t-h]'J � p\J;�
r
vve_ �-e..{Cj) �
-- � +J 2 .f:i" (.e_ - L� fa 7 �-f C1._cP_ L.c-/
�+
U"h �
t1i a.are... r-
J Sr�/� G-o lk C -z l
<
'
9i� -
I ?...o -jo -5 o - (?_ � +-< j I <f-o-o
IJ- � ::::::.- D
�
ifz_ -- .2--4--v
0
� I �o
u
-> e2 6 D
0
L. 6 0 -
----
:;i.13,
�e,Q_ -TC,,,....
z-.2
()
-)..
-8
19
EXAMPLE
First-order system gain is K and time constant is τ . For now let the controller have only
proportional control, i.e. K(s) = Kp
Applying the final value theorem gives the steady state responses to reference and distur-
bance inputs
20
KKp r KKp
θo−ref −ss (s) = lim s = r
s→0 τ s + 1 + KKp s 1 + KKp
τs + 1 d 1
θo−dist−ss (s) = lim s = d
s→0 τ s + 1 + KKp s 1 + KKp
We can now select a value of Kp which satisfies time constant or steady state response
specifications
Example
It is decided that a controller is required for a process in a chemical works. The open-loop
step response is a 50 C output temperature change for a 100 C input change and the open-
loop time constant is 10 minutes. The only design specification is: Effect closed-loop time
constant of about 3 minutes
Solution
Desired closed-loop time constant,τCL , is 3 minutes. Uncontrolled system time constant τ
is 10 minutes and uncontrolled system gain K is 0.5.
τ 10
τCL = = =3 Therefore Kp = 4.67
(1 + KKp ) (1 + 0.5Kp )
Setting the value of Kp equal to 4.67 fixes the level of the steady-state responses:
KKp 0.5 × 4.67
θo−ref −ss = r= r = 0.7r
1 + KKp 1 + (0.5 × 4.67)
1 1
θo−dist−ss = d= d = 0.3d
1 + KKp 1 + (0.5 × 4.67)
Therefore the steady state errors will be
EXAMPLE
First-order system gain is K and time constant is τ . For now let the controller have only
proportional control, i.e. K(s) = Kp + Ksi
√ −πζ 2
OS(%) = 100 exp (1−ζ )
For 0<ζ<1
If there is a specified overshoot we first calculate
OS(%)
LOS = ln )
100
The required damping ratio is then given by
LOS
ζ=p
(π 2 + L2OS )
There is a link between settling time, damping ratio and undamped natural frequency;
3 3
ts (5%) ' therefore ζωn '
ζωn ts (5%)
22
4 4
ts (2%) ' therefore ζωn '
ζωn ts (2%)
So a specification of overshoot and / or settling time can be converted into damping and
natural frequency requirements.
Once we know the required values of ζ and ωn we can calculate the necessary values of Kp
and Ki
The standard second -order system characteristic equation is compared to the closed-loop
system characteristic equation.
s2 + 2ζωn s + ωn2 = 0
1 + KKp KKi
s2 + s+ =0
τ τ
2
2ζωn τ −1 τ ωn
to give Kp = K
and Ki = K
EXAMPLE
Example: PI control
It is decided that a controller is required for a process in a chemical works. The open-loop
step response is a 50 C output temperature change for a 100 C input change and the open-loop
time constant is 10 minutes. The design specifications are:
Solution: PI control
Specification of zero-steady state error means that PI control must be used
3 3
ζωn ' = = 0.15
ts (5%) 30
23
For little overshoot, choose high damping:
0.15
Fix ζ = 0.9 to give ωn = = 0.169
0.9
Calculate Kp and Ki from ζ and ωn
τ ωn2 10 × 0.1672
Ki = = = 0.56
K 0.5
EXAMPLE
Example;
The figure below shows a PD controller used for a control system. Determine the value of
τd so that the system will be critically damped. Calculate settling time.
Solution;
(1 + sτd )4
G(s) = and H(s) = 1
s(s + 1.6)
24
C(s) (1 + sτd )4
= 2
R(s) s + (1.65 + 4τd )s + 4
Comparing denominator with standard form
ωn2 = 4 Thus ωn = 2
4 4
Settling Time Ts = = = 2 Sec
ζωn 2 × 1
EXAMPLE
(a) A system with Proportional Derivative controller is shown in Fig. 1. Find the values of
Kp and Kd so that the velocity error constant is 500, and the damping ration is 0.5.
1. Integration Windup: A significant problem with integral action is that when the
error signal is large for a significant period of time. This can occur every time when
there is large change in set point. If there is a sudden large change in set point, the
error will be large and the integrator output in a P-I-D control will build up with time.
As a result, the controller output may exceed the saturation limit of the actuator. This
windup, unless prevented may cause continuous oscillation of the process that is not
desirable.
The actuator in the linear range is unity. As a result, when the actuator is operating
in the linear range the error eA is zero, and the controller acts as a PI controller. But
when the actuator is in saturation mode, the error eA is negative for a positive e. This
will reduce the integral action in the overall control loop.
Where ep = ap r(t) − y(t), ei (t) = r(t) − y(t) and ed (t) = ad r(t) − y(t)
Where constants ap and ad are as yet undetermined. Each term has a different ”error”
associated with it. Note that when ap = ad = 1 then we the original PID design. Note
also that when r(t) is a piecewise constant signal (only step changes), then for all time
except at actual step locations ṙ(t) = 0 and thus
ded (t) d
= (ad r(t) − y(t)) = −ẏ(t)
dt dt
As we have seen, changing ad doesnt change the overall design. Changing ap , however,
may change the design. The rationale behind ap is that if, for example, ap = 0.5, then
a large step change in r(t) doesnt create such a large control magnitude. However, in
general, ep does not go to zero when y = r. Thus, there is a persistent control applied
even when it is not necessary. The use of this, then is of questionable value. Setting
ap = 1, however, brings us back to the original error.
1. Pneumatic Controller:
It has been already mentioned that the early days PID controllers were all pneumatic
type. The advantage of pneumatic controllers is its ruggedness, while its major limi-
tation is its slow response. Besides it requires clean and constant pressure air supply.
The major components of a pneumatic controller are bellows, flapper nozzle amplifier,
air relay and restrictors (valves). The integral and derivative actions are generated by
controlling the passage of air flow through restrictors to the bellows.
A simple scheme for implementation of a pneumatic PI controller is shown in Figures
below.
29
Proportional action
Fig: P Control
The action of this particular controller is direct, since an increase in process variable
signal (pressure) results in an increase in output signal (pressure). Increasing process
variable (PV) pressure attempts to push the right-hand end of the beam up, causing
the baffle to approach the nozzle. This blockage of the nozzle causes the nozzles
pneumatic backpressure to increase, thus increasing the amount of force applied by the
output feedback bellows on the left-hand end of the beam and returning the flapper
(very nearly) to its original position. If we wished to reverse the controllers action,
all we would need to do is swap the pneumatic signal connections between the input
bellows, so that the PV pressure was applied to the upper bellows and the SP pressure
to the lower bellows. Any factor influencing the ratio of input pressure(s) to output
pressure may be exploited as a gain (proportional band) adjustment in this mechanism.
Changing bellows area (either both the PV and SP bellows equally, or the output
bellows by itself) would influence this ratio, as would a change in output bellows
position (such that it pressed against the beam at some difference distance from the
fulcrum point). Moving the fulcrum left or right is also an option for gain control, and
in fact is usually the most convenient to engineer.
Integral action
Interestingly enough, derivative (rate) and integral (reset) control modes are relatively
easy to add to this pneumatic controller mechanism. To add derivative control action,
30
all we need to do is place a restrictor valve between the nozzle tube and the output
feedback bellows, causing the bellows to delay filling or emptying its air pressure over
time:
Fig: PD Control
If any sudden change occurs in PV or SP, the output pressure will saturate before the
output bellows has the opportunity to equalize in pressure with the output signal tube.
Thus, the output pressure spikes with any sudden step change in input: exactly what
we would expect with derivative control action.
If either the PV or the SP ramps over time, the output signal will ramp in direct
proportion (proportional action), but there will also be an added offset of pressure at
the output signal in order to keep air flowing either in or out of the output bellows at
a constant rate to generate the force necessary to balance the changing input signal.
Thus, derivative action causes the output pressure to shift either up or down (depending
on the direction of input change) more than it would with just proportional action alone
in response to a ramping input: exactly what we would expect from a controller with
both proportional and derivative control actions.
Integral action
Integral action requires the addition of a second bellows (a reset bellows, positioned
opposite the output feedback bellows) and another restrictor valve to the mechanism1:
31
This second bellows takes air pressure from the output line and translates it into force
that opposes the original feedback bellows. At first, this may seem counter-productive,
for it nullifies the ability of this mechanism to continuously balance the force generated
by the PV and SP bellows. Indeed, it would render the force-balance system completely
ineffectual if this new reset bellows were allowed to inflate and deflate with no time lag.
However, with a time lag provided by the restriction of the integral adjustment valve
and the volume of the bellows (a sort of pneumatic RC time constant), the nullifying
force of this bellows becomes delayed over time. As this bellows slowly fills (or empties)
with pressurized air from the nozzle, the change in force on the beam causes the regular
output bellows to have to stay ahead of the reset bellows action by constantly filling
(or emptying) at some rate over time.
32
2. Electronic PID Controllers:
Most analog electronic PID controllers utilized operational amplifiers in their designs.
It is relatively easy to construct circuits performing amplification (gain), integration,
differentiation, summation, and other useful control functions with just a few op-amps,
resistors, and capacitors.
The following schematic diagram shows a full PID controller implemented using eight
operational amplifiers, designed to input and output voltage signals representing PV,
SP, and Output:
This controller implements the parallel, or independent PID algorithm, since each
tuning adjustment (P, I, and D) act independently of each other:
The above equation can be discretised at small sampling interval T0 as shown in Fig. 6.
The above algorithm is known as Position algorithm. But the major problem here is
that the error values at all the time instants are to be stored (or at least the second term
of the r.h.s of Eqn. (2) at each instant have to be stored). An alternative approach
known as velocity algorithm can be obtained as follows.
Where
34
The above algorithm is known as Velocity algorithm. The major advantage of this
algorithm is that it is of recursive type. It calculates the incremental output at each
sample instant. As a result, it requires only to store three previous values: e(k), e(k-1)
and e(k-2). Besides it has got several other advantages also those are elaborated below:
Bumpless Transfer: During the transfer from manual to auto mode, it is desired
that the input command to the process should not change suddenly. In Position algo-
rithm, due to the difference between the set point and the output variable, it is always
P
possible that the existing error will wind up and the value of e(k) be large when
the switching from manual to auto mode takes place. This will cause a large change in
the input u(k) in the auto mode. But in the velocity algorithm, this will be prevented,
since it provides only incremental change in input [u(k) − u(k − 1)]. This will lead to
bumpless transfer.
with low pass characteristics is used to filter out the unwanted noise before it reaches
the controller input.
APPENDIX Chap 2
π 1.8
tp = tr ≅
ωn 1 − ζ 2
ωn
ζ
−π
1−ζ 2 4.6
Mp =e ts =
ζωn
Steady-state Tracking
K p = lim C(s)G(s) n=0 Position Constant
s →0
K v = lim sC(s)G(s) n =1 Velocity Constant
s →0
K a = lim s 2C(s)G(s) n = 2 Acceleration Constant
s →0