Professional Documents
Culture Documents
become unstable. Thus there is a limit on the controller gain that can be used for a process with
time delay.
A process with inverse response violates the second condition for normal dynamic behavior.
The response, even though its step response eventually ends up heading in the direction of the
new steady state, it starts out initially heading in the opposite direction, away from the new
steady state, changing direction somewhere during the course of time.
Processes with inverse response will initially move in the wrong direction as they respond to
control action. Thus if the controller is tuned too tightly (high controller gain) it will attempt to
correct for the movement in the wrong direction and overcompensate. Again there is a limit on
the controller gain that can be used for a process having inverse response.
A process for which the step response is unbounded, i.e., the output increases (or decreases)
indefinitely with time, is said to be open loop unstable. An open loop unstable process violates
the third condition noted above; its output fails to settle to a new steady-state value in response
to a step change in the input.
Processes that are open-loop unstable will "run away" without control, so most of the controller
tuning procedures cannot be applied. In addition, open loop unstable processes can be unstable
for various reasons so that simple PI control may not be enough to stabilize them.
These difficult dynamics translate into unusual phase behavior as non-minimum-phase system.
with
AR =| g ( jω ) |= Re(ω ) 2 + Im(ω ) 2
and
⎡ Im(ω ) ⎤
φ = arg[ g ( jω )] = tan −1 ⎢ ⎥
⎣ Re(ω ) ⎦
For a class of processes (g1(s), g2(s), ..., gn(s)) having same amplitude ratio characteristics (i.e.,
AR1 = AR2 = ... = ARn), and phase angles (φ1, φ2, … φn) of which the minimum is designated
φmin. if φi ≠ φmin then the systems are referred to as non-minimum phase systems.
• Time delay system: For a normal system g1(s):
g 2 ( s ) = g 1 ( s ) e −α s
We have
AR1 = AR2
but
φ2=φ1-αω
• Inverse response system: the two processes having transfer functions:
g1(s) = g0(s)(l + ηs)
and
g2(s) = g0(s)(1- ηs)
We have
AR1 = AR2
but
φ2=φ1-1800
• Open loop unstable system: Consider the first order process
K
g1 ( s ) =
τs + 1
and
K
g 2 (s) =
τs − 1
We have
AR1 = AR2
but
φ2=-1800+φ1
In general, a system which contains any non-minimum phase element: a RHP pole, a RHP zero,
or a time delay, is a NMP system.
a)
K
g1 ( s ) =
τs + 1
A normal first-order system, the phase angle asymptotically approaches a limiting value of -900
Thus, the normal system can never be unstable under proportional feedback control since the
phase angle can never attain the critical value of -1800.
b)
K −αs
g1 ( s ) = e
τs + 1
A first-order system with time delay: the phase angle decreases monotonically with frequency, a
limiting value of proportional controller gain (phase angle crosses 1800) at which the system
becomes unstable. It is proportional to the value of the time delay.
PI Control
PID Control
Smith Predictor
Introduce a minor feedback loop around the conventional controller and a model with subscript
m
ym ( s) = g m* ( s)e −α ms u ( s)
Define:
y * ( s ) = g * ( s )u ( s )
and
y m* ( s ) = g m* ( s )u ( s )
Since
y ( s ) = g * ( s ) e −α s u ( s )
y*(s) is the output of un-delayed process output y(s).
Assuming that there are no model errors g*m(s) = g*(s) and αm = α, the signal reaching the
controller is a "corrected" error signal.
Then:
εc = yd − y(s) − ( y*(s) − y(s))
or
ε c = yd − y * (s)
The equivalent block diagram for the closed loop system is shown as:
The net result of the introduction of the minor loop is therefore to eliminate the time-delay factor
from the feedback loop - where it causes stability problems - and "move" it outside of the loop,
where it has no effect on closed loop system stability.
The characteristic equation of the equivalent system is
1 + g c ( s) g * ( s) = 0
which no longer contains the time-delay element and therefore allows the use of higher
controller gains without placing the closed-loop stability in jeopardy.
To establish the characteristic equation directly, assuming that there are no model errors.
Considering
u = g c*ε
where
gc
g c* =
1 + g c g (1 − e −αs )
*
g * g c e − αs
g * ( s )e −αs g c* =
1 + g c g * − g c g *e −αs
and
−αs 1 + g * gc
1 + g ( s )e
*
g =
*
1 + g c g * − g c g *e −αs
c
Design Procedure
1. Design the Smith Predictor (the minor loop)
The design of the minor loop involves setting up a means by which y* and y are
produced from the process model, y is obtained directly from the process model, and
y* is obtained from the undelayed version of the process model.
2. Design gc
According to the Smith predictor scheme, the controller is designed for the undelayed
system. This permits the use of much higher controller gains than would otherwise be
allowable.
However, the Smith predictor requires a perfect model, and real models are never perfect, we
must be cautious in choosing the controller parameters. In practice, one would choose controller
parameters large enough to achieve much better performance than feedback control alone, but
not so large as to cause serious deterioration in performance resulting from inevitable
plant/model mismatch.
Assuming there is a variation in dead-time, td can only be approximated. The Smith predictor
cannot fully compensate the dead-time effects; there is dead-time residual in the system. The
uncompensated dead-time give rise to additional phase lag and leads eventually the reduction of
crossover frequency and ultimate gain margin. In this case, too big gain will lead unstable
system.
Another problem is the disturbance, Smith predictor has no real effect to disturbance rejection
improvement.
• Stability
• Disturbance rejection
• Transient characteristic
• Robustness
y(s) g *e −αs
=
d (s) g * g c e −αs
1+
1 + g * g f + g * g c (1 − e −αs )
[1 + g * g f + g * g c (1 − e −αs )]g *e −αs
=
1 + g * g f + g * gc
To reject disturbance, it should have
1 + g * g f + g * g c (1 − e −αs ) = 0
we obtain:
1 + g * g c (1 − e −αs )
gf =−
g*
Writing the closed loop transfer function
y ( s) g * g c e −αs
=
yd (s) 1 + g * g f + g * g c
Substitute gf(s) into above equation, we obtain
* −αs
y(s) g g c e
= =1
y d ( s ) g * g c e −αs
It means that the system can exactly following the reference signal and disturbance has been
totally rejected.
Consider a system
g(s) = g1(s)-g2(s)
g1(s): representing the "main mode";
g2(s): as the "opposition mode";
The second model g2(s) has negative sign, opposite to the first model g1(s).
K c (−3s + 1)
g (s) =
(1 + 2 s )(1 + 5s )
From this diagram, we obtain the following critical information required for the Ziegler-Nichols
design:
The crossover frequency: ωco= 0.55 radians/time
The magnitude ratio at this point: MR = AR/KKc = 0.5
The ultimate gain and period are given as: Ku =2
and
2π
Pu = = 11.42
ω co
By the following table
Controller K τi τD
P 0.5Ku
PI 0.4Ku 0.8Pu
PID 0.6Ku 0.5Pu 0.125Pu
The Ziegler-Nichols recommended values for the PID controller parameters are obtained as:
Kc=1.2; τI=5.7 τD=1.4
Inverse-Response Compensation
Objective: choose the quantity λ such that the signal reaching the controller appears to be from a
"normal" system.
Define the variable y'
y'(s) = g'(s)u(s)
generated by the minor loop. As a result of this minor loop, the signal reaching the controller is
given by:
εc = yd − y(s) − y'(s)
or
ε c = yd − [g ( s ) + g ' ( s )]u ( s )
Now, let:
g * ( s) = g ( s) + g ' ( s)
and
y * ( s ) = g * ( s )u ( s )
then it becomes:
εc = yd − y*(s)
Introducing g and g' into g*, we obtain:
g*(s) = g0(s)(I - ηs) + g0(s)λs
or
g*(s) = g0(s)[l + (λ-η)s]
choose λ such that:
λ>η
y* no longer contains a RHP zero. Thus the minor loop provides a corrective signal that
eliminates the inverse response from the feedback loop.
In case of plant-model mismatch, choosing λ>η (as opposed to λ=η). Usually select
λ=2η
Example: Design an inverse-response compensator for the inverse-response system
(−3s + 1)
g ( s) =
(1 + 2 s )(1 + 5s )
Solution: For the process
1
g 0 (s) =
(1 + 2 s )(1 + 5s )
and
λs
g ' ( s) =
(1 + 2 s )(1 + 5s )
For this system η = 3, a good value of A to be used is λ =6 so that:
6s
g ' (s) =
(1 + 2 s )(1 + 5s )
is the transfer function to use in the inverse-response compensator loop. The apparent process
transfer function is given by:
1 + 3s
g * (s) =
(1 + 2 s )(1 + 5s )
with no RHP zero.
Example: Investigate the closed loop stability properties of the following system under
proportional-only feedback control, first without any compensation, and then with the inverse
response compensator
( −3s + 1)
g (s) =
(1 + 2 s )(1 + 5s )
Solution: Under conventional, proportional feedback control, the characteristic equation for the
closed loop system is:
K c (−3s + 1)
1+ =0
(1 + 2 s )(1 + 5s )
which rearranges to:
10s 2 + (7 − 3K c ) s + (1 + K c ) = 0
The condition for stability is:
Kc < 7 / 3
With the inverse-response compensator, using the minor loop
u = g c*ε
where
gc
g c* =
1 + gc g '
The overall closed loop transfer function is
gg c*
y= yd
1+ gg c*
and the characteristic equation is:
1 + gg c* = 0
the characteristic equation becomes:
10 s 2 + (7 + 3K c ) s + (1 + K c ) = 0
which is stable for all positive values of Kc
The inverse response compensation compared to conventional PID control, the response to a
unit set-point change with the inverse-response compensator combined with a PI controller (Kc
=10, τI=0.167) is shown in the Figure. Note that the inverse-response compensator produces the
smallest negative deviation and a rapid response without overshoot.
Design Procedure:
1. Design the inverse-response compensator loop, this involves obtaining the appropriate
transfer function g'(s) and find λ to use in the minor feedback loop.
2. Design gc: Once the compensator has been designed, it is simply design the controller for
g*(s) with absence of the RHP zero permits the use of higher controller.
However, because of process/model mismatch, one must be careful not to increase the controller
gains too much.
Example: Obtain the range of Kc values required to ensure that the closed-loop system
involving:
K
g (s) =
τs − 1
and a proportional controller is stable.
Solution: The characteristic equation for the closed loop system is:
1 + KKc/(τs-1) = 0
which is
τs - 1 + KKc = 0
the one root is located at
s = (1-KKc)/τ
The root will be negative, if
Kc > 1/K
which will stabilize the open-loop unstable system.
Place the closed loop system poles in pre-specified locations in the LHP.
Example: Design a PI controller for above Example with K=1/6 and τ =0.25 that will
stabilize the dosed-loop system with the closed-loop system poles located at s = -2 and at s = -
4.
Solution: PI Controller is given by
⎛ 1 ⎞
g c ( s ) = K c ⎜⎜ 1 + ⎟
⎝ τ I s ⎟⎠
The closed-loop characteristic equation
1+gc(s)g(s)=0
becomes:
ττIs2 + (KKc-1)τIs + KKc = 0
solving this equation, gives the required controller parameter values:
6τ + 1
Kc = Kc = 15.5
K
And
8τ
τI = τI = 0.775
KK c
However, not all open loop unstable processes can be stabilized by P or PI control.
K
G p (s) = e − Ls
Ts − 1
With P controller in the inner feedback loop, the internal closed-loop transfer function Gl (s)
can be obtained as
G p (s) Ke − Ls
Gl ( s ) = =
1 + K l G p (s) Ts − 1 + KK l e − Ls
Using Taylor series expansion
e − Ls ≅ 1 − Ls + 0.5L2 s 2
we obtain
Ke − Ls
G l ( s ) ≅ G 'p ( s ) =
0.5 KK l L2 s 2 + (T − KK l L) s + KK l − 1
'
Since the characteristic equation of G p ( s ) should have negative poles to be stable, the
following condition must be satisfied from the Routh-Hurwitz stability criterion
1 T
K min = < Kl < = K max
K LK
The above expression indicates that a condition T/L>1 for unstable processes should be
satisfied. That means that the proposed method is suitable for unstable processes with small time
delays.
For the optimum gain margin
1 T
K l = K min K max =
K L
which results
e − Ls
G 'p ( s) =
L 1 1 T
(0.5 TL ) s 2 + (T − TL ) s + ( − 1)
K K K L
As the integrating and unstable processes are stabilized with the P controller in the inner
feedback loop, we can design a PID controller for the stabilized processes which have second
order plus dead time process structure.
e − Ls
G 'p ( s ) =
as 2 + bs + c
Writing PID controller transfer function as
⎛ As 2 + Bs + C ⎞
Gc (s ) = k ⎜⎜ ⎟⎟
⎝ s ⎠
Kd K K
where A = , B = p and C = i . Let controller zeros to be equal to the poles of model
k k k
G′p (s ) , i.e. A = a , B = b , C = c . Hence
ke− sL
G′p (s )Gc (s ) =
s
where k is determined based on gain and phase-margin specifications. Typical values of gain
margin and phase margin range from 2 to 5, and 30o to 60o , respectively. If assign Am = 3 , then
Φ m = 60 o and
π π
k= =
2 Am L 6L
Hence PID settings for unstable processes are given as
π T T
Kp = ( − )
6K L L
π T
Ki = ( − 1)
6 KL L
π
Kd = TL
12 K
Once the model is obtained, we can ignore the inner feedback loop and directly design PID
controllers for the unstable time delay processes. The tuning formulae are very simple and
straightforward.