Вы находитесь на странице: 1из 49

Metodo de Sintonia de Lazos

• Metodo de Ziegler Nichols


• Metodo de Cohen Coon
• Metodo de Sintesis Directa
• Metodo IMC
• Metodo Profit PID
Preparation for Control Design

• Definitions of IAE, ISE and ITAE


• They can be used to define the control performance

IAE = 0 e(t) dt absolute error
2
 [ e(t) ] dt squared error
ISE = 
0

ITAE =  0 t e(t) dt time-weighted absolute error

• Comparison of IAE, ISE, and ITAE


• They are very similar
• ISE often leads to a mathematical convenience
PID

OUT = K*E + K/T1∫ E dt + K*T2 dE/dt + Bo


P I D

E = PV - SP
PID

OUT = K*E + Bo E = PV -SP

Error Permanente
PID

OUT = K*E + K/T1∫ E dt Bo

E = PV -SP
PID

OUT = K*E + K/T1∫ E dt + K*T2 dE/dt + Bo

E = PV -SP
PID

OUT = K*E + K/T1∫ E dt + K*T2 dE/dt + Bo

E = PV -SP
PID

Standard Profit PID (1)

1 + Ti(s) + Td(s) 1 r (s) – y (s)


U(s) = k
Ti(s) 1 + Tf(s)

No Interactive PID (Ideal) (2)


1 + Ti(s) + Td(s) r (s) – y (s)
U(s) = k
Ti(s)
PID

No Interactive PID (Ideal) (3) P & I

1 + Ti(s) 1 + Ti(s) + Td(s)


U(s) = k R (s) - y (s)
Ti(s) Ti(s)

No Interactive PID (Ideal) (4) I


1 1 + Ti(s) + Td(s)
U(s) = k R (s) - y (s)
Ti(s) Ti(s)
PID

Interactive PID (No Ideal) (5)

1 + Ti(s) 1 + Td(s)
U(s) = k R (s) R (s) – y(s)
Ti(s) 1 + Td(s)

No Interactive PID (No Ideal) (6) P & I


1 + Td(s)
1 + Ti(s)
U(s) = k 1 + Ti(s) R (s) - 1 +  Td(s) y (s)
Ti(s) Ti(s)
PID

Interactive PID (No Ideal) (7) I


1 + Td(s)
1 1 + Ti(s)
U(s) = k R (s) - 1 + Td(s) y (s)
Ti(s) Ti(s)

Ziegler-Nichols (Original & Modified)

• Based upon a fist-order-with-delay dynamics


• No model identification, but does require oscillations of
the closed-loop system.
Steps:
1. Oscillate the loop with a gain-only controller to determine:
a) ultimate gain = Kcu -and- b) ultimate period = Pu

2. Obtain PID settings from the following table:


PID Settings Kc T1 T2
Original (1/4 decay ratio,1942)0.6 Kcu Pu/2 Pu/8
Some overshoot 0.33 Kcu Pu/2 Pu/3
No overshoot 0.2 Kcu Pu/2 Pu/3
Ziegler-Nichols (graphic explanation)

• Continuous Cycling
- Eliminate integral and derivative action
- Increase the controller gain until process oscillates continuously
- Determine ultimate controller gain Kcu and ultimate period Pu.
Pu

• Tune the controller to reach 1/4 wave decay ratio

A
B=1
B
Sp A 4
CV

t
Cohen-Coon (1)

• Based on a first order plus dead-time model (1953):


Kp - s
e
s + 1
• Designed to provide a response with a 1/4 decay ratio:
P Controller: Kc = 1  ( 1 +  )
Kp  3

PI Controller: Kc = 1  ( 0.9 +  )
Kp  12 

 30 + 3 (/)
I (or T1) =
9 + 20 (/)
Cohen-Coon (2)

1  16  3
PID Controller: K c 
Kp   12 

32  6 /  
 I (orT1) 
13  8 /  
4
 D  orT2  
11  2  /  

• Advantage
- Can get tuning constants without oscillating the process.

• Disadvantages:
- 1/4 decay responses are usually judged oscillatory by operators
- Considers only two points of closed-loop response (first two peaks).
Direct Synthesis (1)

A typical one-degree-freedom FB control loop:


L
R
+ E U + y
Gc Gp +
-

Classical Feedback

GpGc 1
The response (output) y  G p U  L  G p G c E  L  R L
1  Gp Gc 1  G p Gc
Where:
• Gc represents the dynamics of the controller (could be a PID)
• Gp represents the dynamics of the process.
Direct Synthesis (2)

• Suppose we could command a closed-loop response for the


setpoint tracking, y/R: (assuming for now L= 0)
G p Gc
• And also y/R
1  Gp G c
• The controller can be solved and would, if realizable, give
the closed-loop response commanded.
- solving for Gc gives the form of the controller required to give a
user-specified response, y/R:

1 y/R
Gc  ( )
Gp 1 y / R
Direct Synthesis (3)

• Perfect Control (if possible)


- Defined as response of y/R = 1
- Impossible to achieve since Gc = > 
• Finite Settling Time (if realizable)
- Defined as response of y/R = 1
c s + 1
Where c is closed-loop time constant.

- Choice of c will give a closed-loop response to a setpoint change


which resembles a first order process with time constant equal to c.
Direct Synthesis (4)

• How to use model to design Gc as a PI or PID controller

Given: 5 y
Gp =
10 s + 1
and = 1
R 5s+1

Solving for Gc gives:


1
Gc = 10 s + 1 5 s + 1 = 2 ( 1 + 1 )
5 1- 1 5 5s
5s + 1

- Remembering the Laplace form of the PI equation:

Gc = Kc ( 1 + 1 )
T1 s
Direct Synthesis (5)

• The PI tuning constants can be directly obtained.


2 1
Kc = T1 =
5 5
• Advantages:
- Controller can be simply obtained from process model
- User can choose a closed-loop response
- Gives insight to relationship between process model and controller.
- For simple systems, the resulting controller can be fitted into a PID

• Disadvantages:
- y/R may not be realizable for a delayed or inverse-response system
- Control performance relies heavily on the accuracy of the model.
IMC: Internal Model Control (1)

• IMC - Controller design is ISE-optimal control based on an


internal model. It tries to minimize the Integral-Square-
Error (ISE) of the closed-loop system.

• Like direct synthesis, design method is based on an


assumed process model and relates controller to model
parameters.

• IMC approach has two additional advantages:


1. It is ISE-optimal, thus the user does not need to choose y/R trajectory.
2. It allows the designer to trade off controller performance for closed-loop
robust stability.
IMC vs Classical
(2)

Comparison of block diagrams of IMC and Classical Feedback:


L
R E U + y
+ Gc Gp +
-

Classical Feedback

L
R U + y
+ f ^
G Gp +
- c

~
y
~ +
Gp -
~
y-y

IMC Feedback
IMC Relation (3)

• Closed-loop transfer function for IMC:

f Gc Gp 1 - f Gc Gp
y= R+ L
1 + f Gc ( Gp - Gp) 1 + f Gc ( Gp - Gp)

• When process and model are identical, the FB error


equals disturbance and the control loop becomes
effectively open.
IMC to Classical
(4)

• IMC is related to classical controller:


Gc
* L
Gc
R ^ U + y
+ + f Gc Gp +
- +
~
y ~
Gp

Rearranged IMC Feedback

Where the conventional controller is: Gc = f Gc


1 - f GcGp
IMC: Design Procedures (5)
˜p G
G ˜ p G
˜ p
1. Factor model into two parts:
˜ is composed of all-pass non-minimum phase components
•G p

(a time delay or a right half plane zero with an opposite pole):


s  1
es , ,
s  1
˜ is what is left, thus minimum phase.
•G p

1
2. The controller is determined as: Gˆ c  G˜ p and then G* c  G˜1 f
p

1
• f is a low-pass r-th order filter with gain of one: f =
c s + 1) r
• c is a desired closed-loop response time constant
*
• r must be a positive integer such that Gc is realizable
(The order of numerator must be Š the order of denominator+1,
assuming an ideal derivative is allowed.)
IMC: Performance vs Robustness
(6)

• Tuning parameter c :

1. If c =  closed-loop system will always be stable.

2. If model exactly matches process, then the ISE-optimal


response is given when c = 0.

3. c should be between 0 and  to give a proper control


performance and robustness.
IMC: An Example (7)

Ke s
˜ p (s) 
G
s  1
˜ p (s)  G
G   
˜ p  e  s
˜ p G
K
s  1

f s  1 1 1 s  1
*

Gc G  
p K c s  1 K  c s 1

*
Converting Gc to classical feedback controller Gc gives:

*
Gc 1 s 1
Gc  
1  Gc G
*
˜p K  c s  1  e s
IMC: An Example (8)

• Note that the Gc controller obtained doesn't give a PID


controller.
1 s  1
Gc 
K c s  1 e s

To convert to a PID, approximate the delay with a first order


Pade's approximation:
  
  s  1  1  1
 2  G c  Kc  1   D s 
e s  and fit with  Is  Tc s  1
 
 s  1
2 
Therefore:
2     c 
Kc  I    D  Tc 
2K  c    2 2   2 c   
IMC- PID (9)

• IMC applied to a PID:


 I or T1  D or T2
Case Model K*Kc

K
A s  1 c  

K 1   2 1   2 1 2 


B 1s  12s  1 c 1   2

K 2 
2
C 2 2
 s  2s  1 c 2

( Rivera, D. E., etc. 1986)


SISO Example

PV
Reflux

Vapor

t
Drums MV
Coke

Coking Heater t
SISO Example

PV
Reflux

Vapor

t
Drums MV
Coke

Coking Heater t
SISO Example

PV
Reflux

Vapor

t
Drums MV
Coke

Coking Heater t
SISO Example

PV
Reflux

Vapor

t
Drums MV
Coke

Coking Heater t
Nominal Control Design

1.4

1.2

A fast controller:
1
2.5 times faster ( = )
0.8
c

0.6
A sluggish controller:
0.4
4 times slower (c = 48)
0.2

0
0 20 40 60 80 100 120 140 160 180 200

• Based on one model (the nominal model)


• Performance can be adjusted.
When Model Mismatches Process (1)

1.4

1.2

0.8

0.6
4 times slower (c = 48)
0.4

0.2

0
0 20 40 60 80 100 120 140 160 180 200

• When the sluggish nominal controller is applied to


different process dynamics.
When Model Mismatches Process (2)

1.4

1.2

0.8

2.4 times faster ( = )


0.6
c
0.4

0.2

0
0 20 40 60 80 100 120 140 160 180 200

• When the fast nominal controller is applied to different


process dynamics.
Sensitivity Checking is Desired

1.4

1.2
2.4 times faster (c = )
1

0.8

0.6
4 times slower (c = 48)
0.4

0.2

0
0 20 40 60 80 100 120 140 160 180 200
Robust Control Design (1)

1.4

1.2

0.8

0.6
4 times slower (c = 48)
0.4

0.2

0
0 20 40 60 80 100 120 140 160 180 200

• Based on a set of models (or a model with uncertainty)


• "Bandwidth" is minimized before performance is "pushed."
Robust Control Design (2)

1.2

2.4 times faster (c = 


0.8

0.6

0.4

0.2

0
0 20 40 60 80 100 120 140 160 180 200
Robust control looks for a family
of good performance.

1.4

1.2 2.4 times faster ( c= 


1

4 times slower (c = 48)


0.8

0.6

0.4

0.2

0
0 20 40 60 80 100 120 140 160 180 200
How to Design a Robust PID

Setpoint e u y
PID Controller K
Kc  I  D 2 s2 + 2 s +1

Controller Process

Integral Square Error (ISE), a control performance criterion,


is a function as follows:

ISE = f(Kc, I, D, K, , ) i.e., ISE = f


Model Uncertainty

u y
K
Process: 2 s2 + 2 s +1

K  K K
Parameter uncertainty     is introduced to
  

represent dynamics change or error in modelling.


Min-Max Principle

• Find a fixed PID controller that is insensitive to model


error and yet provides a good control performance.

• For any given PID controller, the worst control


performance (worst scenario) can be found:
Max ISE (Kc, I, D, K, , )
K, , 

• The PID controller with a best control performance


under a worst scenario is the desired solution:
Min Max ISE (Kc, I, D, K, , )
Kc, I, D K, , 
Understanding Min-Maximization

• Min-Max is not just "doing the best on the worst case."


• Min-Max improves the performance of the remaining
worst case, and the net effect is to improve the
performance of all cases.
• Min-Max is an integrated operator which obtains not only good
control performance but also less sensitivity to model error.
• The control performance of remaining worst case may be
better than that of a nominal controller.
• Min-Max reverts to nominal control under a perfect model.
Robust PID Example (1)

Process
 
0.5 K 1.5
G(s) = K with 10   30
2 s2 + 2 s +1
0.1   0.5
Controller
IMC PID Robust PID
with filter constant 5 with min-max principle
by using average model

Kc = 2.4, I = 12 and D = 100 Kc = 12, I = 30 and D = 30


3
Robust PID Example (2)

Setpoint e u y
PID Controller K
Kc  I  D 2 s2 + 2 s +1

Controller Process

Process 1: K = 1.0,  = 20 and  = 0.3


Test Process Dynamics: Process 2: K = 0.5,  = 30 and  = 0.5
Process 3: K = 0.5,  = 30 and  = 0.1
IMC PID Performance (3)

1.4

1.2 3
2
1
1

0.8

0.6

Process 1: Red
0.4
Process 2: Blue
0.2
Process 3: Green
0
0 20 40 60 80 100 120 140 160 180 200
Robust PID Performance (4)

1.4

1.2
3
1 1
2
0.8

0.6
Process 1: Red
0.4 Process 2: Blue

0.2 Process 3: Green

0
0 20 40 60 80 100 120 140 160 180 200
Robust PID Design

• Step 1: Obtain a model (doesn't have to be very good)


• Step 2: Estimate an uncertainty (say, 20~50%)
• Step 3: Compute the PID parameters (using Mac App.)

• Step 4: Apply on TDC system (or other systems).

Вам также может понравиться