Download as pdf or txt
Download as pdf or txt
You are on page 1of 48

Honeywell Forge APC-4516

Process Controller Implementation


Advanced Control Overview

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted
1 September 26, 2022
Contents Honeywell
• Advanced Control Overview
– Levels of process control
– Benefits
• MVC Overview
• MVC Techniques
• Model Based Predictive Control
– Horizon strategy & prediction
– Model reconciliation
– Target CV trajectory & control
– Optimization

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 2 September 26, 2022
Levels of Process Control Honeywell
• Basic regulatory
• Advanced regulatory (intermediate regulatory)
• Multivariable, model based
• Constrained economic optimization

Increasing complexity and costs and


diminishing benefits

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 3 September 26, 2022
Basic Regulatory Honeywell
• Single sensing element - single final control element
– Single Input Single Output (SISO)
• Simple cascades, ratio control
• Primary goal:
– Operational stability

FC

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 4 September 26, 2022
Advanced Regulatory Honeywell
• Pressure or temperature compensation
• Feed-forward based on measurements
• Simple process calculations based on measurements
– Heat and material balance
– Simple inferential estimations, duty control
• Multiple Input Single Output (MISO)
• Primary Goals:
– Enhanced Stability
– Enhanced performance

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 5 September 26, 2022
Example of ARC: Coil Outlet Temperature Control Honeywell

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 6 September 26, 2022
Multivariable, Model based Honeywell
• Simultaneous consideration of:
– many controlled variables (CV)
– by adjusting many manipulated variables (MV)
– optionally observing feedforward variables (DV)
• Internal model of process behavior
– predicts future paths of controlled variables.
• Multiple Input Multiple Output (MIMO)

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 7 September 26, 2022
Example of Different Kinds of Control: Distillation Tower Honeywell

PC
FC
H2 & CH4
C3 Refrig.
TI

LC CVs
FC

FC FC Calculated CVs
RMPCT
CV:
FC AI Product Qualities Ethylene AI
C2 Feed AI Ethylene
Reflux Drum Level Quality Calc.
Compressor Limits T,F,P
Valve Positions
Tower Flooding Bottom Product AI
MV: Quality Calc.
DVs AI
Pressure
Reflux Flow
Top Product Flow Tower Flooding T,F,P,DP
Reboiler Duty Calculation
AI DV:
Feed Flow
Non-Linear Feed Quality
LC
Level Control

AI FC
MVs
Reboiler Ethane Recycle
FC
Duty Control
C3 refrig.

C3 refrig. pres.

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 8 September 26, 2022
Constrained Economic Optimization1 Honeywell
• Premise
– Some of controlled variables are not controlled to setpoints
• Required to be maintained between high and low ranges
• User specifies an objective function
– Assign costs or values to variables
– Translates into a direction to move the manipulated variables
• Optimizer finds best position (corner)
– within space constrained by controlled and manipulated variable limits

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 9 September 26, 2022
Constrained Economic Optimization2 Honeywell
• Optimizer is a “steady state controller”
– Calculates one set of MV moves to maximize or minimize objective function.
– Generally too aggressive to implement directly in closed loop
– Typically used in conjunction with a multivariable control engine to
implement solution

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 10 September 26, 2022
Levels of Process Control Honeywell

Optimization Local LP/QP Multi-unit LP/QP Rigorous model based Gate to Gate
Optimization Optimization Optimization Optimization

Model Based
E.g.
Smith Multivariable
Control Profit Controller or
predictors predictive
etc. control [DMC] +

Advanced Feedforward Dynamic On-stream Constraint


Regulatory control decoupling Analyzers Control
Control

Basic Single Cascade


Regulatory PID
Control

Different technologies available at each level

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 11 September 26, 2022
Benefits of Advanced Control Honeywell
• Advanced control can improve plant profitability in a number of ways
– Setpoint control
• Reduced variability allows closer operation to the constraint (i.e. less
give-away)
– Quicker grade changes
• Increased plant flexibility
– Basis for more sophisticated advanced controls
• Optimization

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 12 September 26, 2022
Incentives for Advanced Control (Constraint Control) Honeywell

Variable
Constraint Limit

$ Advanced
Control

Good Regulatory
Control

Poor control
Time

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 13 September 26, 2022
Incentives for Advanced Control (Quicker Grade Changes) Honeywell

Quality

Quality Transition

Minimise transition time


Time

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 14 September 26, 2022
Incentives for Advanced Control (Optimization Platform) Honeywell

MV 1 High
MV 1 Low
MV 2 High

‘best’ operating region


Constrained Optimum

MV 2 Low

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 15 September 26, 2022
Multivariable Control Overview

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted
16 September 26, 2022
Multivariable Control Definition Honeywell
• Simultaneous control of more than one controlled variable (CV) by the
co-ordinated adjustment of more than one manipulated variable (MV)
• Controller may use additional feedforward or disturbance variable (DV)
information to improve control performance

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 17 September 26, 2022
Why Use MVC Techniques Honeywell
• When the process is interactive
– CV’s are affected ‘significantly’* by multiple MV’s.
– Metric for interactive process: RGA (Bristol)
• MV Rangeability
• Common CV constraints
– competing control sub-strategies
• Ease of use / configuration
– MV / CV relationships are similar in gain and dynamics

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 18 September 26, 2022
Debutanizer Column1 Honeywell

Interaction: Cooling water

CV1
Increase in reflux (MV1)
LC
causes: TI

↓ in top temperature FC MV1 FC


CV3
↓ in distillate
DV1
CV4 Distillate
↓ in base temperature Feed

↑ in base product
MV2
FC

Steam
CV5
LC
CV Controlled variable Base
DV Disturbance (FF) variable product
MV Manipulated variable CV2 TI

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 19 September 26, 2022
Debutanizer Column2 Honeywell

Interaction: Cooling water

Increase in steam (MV2) CV1

causes: LC
TI
Reflux
↑ in top temperature FC MV1 FC
CV3
↑ in distillate
DV1
CV4 Distillate
↑ in bottom temperature Feed

↓ in base product FC MV2

Both steam and CV5


Steam

reflux affect LC
Base
most column CV2
TI
product

variables

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 20 September 26, 2022
Dynamic Interaction Honeywell

2 CV1

1
CVs
0 CV2

MV2
1.5
1 MVs
0.5
MV1
0

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 21 September 26, 2022
Debutanizer Column3 Honeywell

Cooling water

MV Rangeability CV1

TI LC
With explicit pairing: Reflux

CV1 paired to MV1 (in Manual) FC MV1 FC


CV2 paired to MV2 CV3

CV4 Distillate
Is it possible to control DV1

both CV1 and CV2 ? Feed

FC MV2

Steam
CV5

LC
Base
TI product
CV2

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 22 September 26, 2022
Debutanizer Column4 Honeywell

Cooling water

MV Rangeability CV1

LC
TI
Without explicit pairing: Reflux
Is it possible to control FC MV1 FC
CV1 and CV2 ? CV3

DV1
CV4 Distillate
Probably not to two
Feed
setpoints
Probably to within two
ranges FC MV2

Steam
CV5

LC
Base
product
CV2
TI

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 23 September 26, 2022
Debutanizer Column5 Honeywell

Cooling water

Common Constraints
CV1

LC
TI
CV4 and CV5 constraints Reflux
to maximizing FC MV1 FC
fractionation. CV3

CV4 and CV5 also MV3


CV4 Distillate
constraints to maximizing Feed
feedrate

MVC (with an optimizer) FC MV2

allows objectives to be
prioritized. Steam
CV5

LC
Base
product
CV2
TI

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 24 September 26, 2022
Model Based Predictive Control1 Honeywell

SPs

OP1
PV1
OP2
PV2 MBPC OP3
PNn OPm

Optimizer

Prediction Controller Process


MV’s CV’s

+
Model -

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 25 September 26, 2022
Model Based Predictive Control2 Honeywell

CV Targets
Optimizer

MVs
Prediction Controller Process
CVs

Model
-

Prediction Model Error

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 26 September 26, 2022
Model Based Predictive Control3 Honeywell
• Control structure is flexible, hence can be used for control problems that
are
– square (#MVs = #CVs)
– over-specified (#MVs > #CVs) (Fat control problem)
– under-specified (#MVs < #CVs)
• Applicable to many types of process control problems
• Not too difficult to implement, even for relatively large problems
• Robust technology insensitive to model error

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 27 September 26, 2022
Prediction (future) Honeywell

CV Targets
Optimizer

MVs
Prediction Controller Process
CVs

Model
-

Prediction Model Error

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 28 September 26, 2022
Horizon Strategy Honeywell

12
Now
10

Unforced Response
8

MV
Value

6 CV

2
History Settling Time
0
0 5 10 15 20 25 30 35
Time Intervals

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 29 September 26, 2022
(SISO) Prediction Models1 (Parametric models) Honeywell

Laplace Domain model:

(Τ1S + 1)e − sDT


Y( s ) = U (s)
(Τ2 S + 1)(Τ3S + 1)

Discrete Time model:

Y( z ) =
(b1Z −1 + b2 Z −2 ....bn Z −n )Z −d
U ( z)
1 + f1Z −1 + f 2 Z −2 .... f n Z −n

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 30 September 26, 2022
(SISO) Prediction Models2 Honeywell

2.5

1.97 1.99 1.9


2 1.95
1.90 1.98 1.999 2.00
1.81 1.93 1.96
1.86
1.73
1.62 Finite step response model
1.5 1.47 MV
CV
Value

1.26

1 0.97

Deadtime =2
0.57
0.5 Lag =3
Unit MV Change at t = 1 Gain =2
0
0 1 5 10 Time Intervals15 20 25

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 31 September 26, 2022
(SISO) Prediction Models3 (MV stepped one unit at t = 1) Honeywell

A × ∆U = Y

 0.00   0.00 
 0.00   0.00 
   
0.57  0.57 
   
 0 . 97   0 . 97 
1.26  1.26 
   
1 . 47  1 . 47 
1.62  × [1]t =1 = 1.62 
   
1.73  1.73 
1.81  1.81 
   
1.86  1.86 
   
 .......   ....... 
 2.00 
   2.00 
 

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 32 September 26, 2022
(SISO) Prediction Models4 Honeywell

A × ∆U = Y

0.00  0.00  − 0.00  − 0.00 


       
0.00 0.00  0 . 00   − 0 . 00   − 0 . 00 
0.57 0.00 0.57   − 0.00 + 0.57 
       
0.97 0.57   0 . 97   − 1 . 14   − 0 . 16 
1.26 0.97  1.26   − 1.95   − 0.68 
       
1.47 1.26   1  1 . 47   − 2 . 53   − 1 . 06 
1.62 ×   = + =
1.47  − 2
1.62   − 2.95  − 1.32 
       
1.73 1.62  1.73   − 3.24   − 1.52 
       
1.81 1.73   1 . 81   − 3 . 46   − 1 . 65 
1.86 1.81  1.86   − 3.61  − 1.75 
       
 ...... .......  .......   .......   ....... 
2.00 2.00 2.00 − 4.00  − 2.00 
      

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 33 September 26, 2022
(SISO) Prediction Models5 Honeywell

12
Now
10 Unforced Response

8
Value

MV
CV
4

2
History Settling Time
0
0 5 10 15 20 25 30 35
Time Intervals

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 34 September 26, 2022
Prediction Models6 Honeywell

MV 1 MV 2 MV 3 DV 1

CV 1

CV 2

CV 3

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 35 September 26, 2022
Model Reconciliation1 Honeywell

CV Targets
Optimizer

MVs
Prediction Controller Process
CVs

Model
-

Prediction Model Error

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 36 September 26, 2022
Model Reconciliation2 Honeywell

12
Now
10
Bias update
8
Value

6
MV
Reconciled CV trajectory CV
4 CVrev

2
History Settling Time
0
0 5 10 15 20 25 30 35
Time Intervals

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 37 September 26, 2022
Control Honeywell

CV Targets
Optimizer

MVs
Prediction Controller Process
CVs

Model
-

Prediction Model Error

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 38 September 26, 2022
Desired CV Trajectory Honeywell

12
Now SP Reconciled CV trajectory
10

Error Vector MV
Now/SS
Value

6 CV
Target
Error

History Settling Time


0
0 5 10 15 20 25 30 35 40
Time Intervals

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 39 September 26, 2022
Move Calculation (Simplified) Honeywell

Prediction: A Prediction matrix


[Y ] = [ A]× [∆U ] YT Target CV vector
Y Unforced CV vector
E Error vector
Controller Error: ∆U MV history vector
[E ] = f ([YT ], [Y ]) ∆UT MV future vector

‘Least squares error’ control solution:


[∆UT ] = [A ] −1
T
×A × AT × [E ]

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 40 September 26, 2022
Horizon With Control (one MV move) Honeywell

12
Now SP Forced CV trajectory
10

Error Vector MV
Now/SS
Value

6 CV
Target
Error

4
Future MV move

History Settling Time


0
0 5 10 15 20 25 30 35 40
Time Intervals

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 41 September 26, 2022
Horizon With Control (seven MV moves) Honeywell

12
Now SP Forced CV trajectory
10

Error Vector MV
Now/SS
Value

6 CV
Target
Error

4
Future MV moves

History Settling Time


0
0 5 10 15 20 25 30 35 40
Time Intervals

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 42 September 26, 2022
Move Calculations Honeywell
• Planning multiple future MV moves allows
– Improved control performance
• Multivariable system (dynamic decoupling)
• Complex dynamics and dead-times
– Game plan of where MV’s will move
• Can defer MV movement
• Least squares method is very aggressive / unstable
– No mechanism to tune controller performance
– Alternatives
• Modify reference trajectory
• Use movement suppression for MV’s
• Use alternative solution strategy

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 43 September 26, 2022
Optimization1 Honeywell

CV Targets
Optimizer

MVs
Prediction Controller Process
CVs

Model
-

Prediction Model Error

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 44 September 26, 2022
Optimization2 Honeywell
• Starting point for optimizer
– Current values of the MVs
– Steady state (unforced) values for the CVs
• Based on MV and DV moves until current time
• Optimizer solves for a set of new MV steady state values (one step
controller) taking into account
– MV constraint limits
– CV constraint limits
– An objective function (direction of steepest decent)
• Based on economics for the process

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 45 September 26, 2022
Optimization3 Honeywell

MV 2 Contour of equal cost


MV 1 Low MV 1 High

MV 2 High

Current position

MV 2 Low

Direction: from economics

MV 1

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 46 September 26, 2022
Optimization4 Honeywell
• Alternative solution technologies
– QP (Quadratic)
– LP (Linear)
• Alternative implementation strategies
– SP targets downloaded to controller
– MV targets superimposed on control plan
• Alternative methods of adjusting speed
– Step sizes
– tunable time horizon

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 47 September 26, 2022
Example Objective Functions Honeywell

Linear objective function:

J = ∑a
i =1, # MVs
i × MVi SS

Quadratic objective function:

∑ ai × MVi SS +
i =1, # MVs

j =1, # CVs
b j × CV jSS +
J =
∑ (
ci2 × MVi SS − MViT ) 2
+ ∑ (
d i2 × CV jSS − CV jT )
2

i =1, # MVs j =1, # CVs

Honeywell Confidential
Copyright © 2012 Honeywell International Inc.
Honeywell Restricted 48 September 26, 2022

You might also like