2011 Engineering Design Seminar

Advanced
d a ced Process
ocess Controls
Co t o s

Dr. Duane Morningred

Honeywell Process Solutions
Camarillo, CA

© 2011 Honeywell. All rights reserved.

 What is your highest level of
experience with Process Control?

1. No experience
p 20% 20% 20% 20% 20%
2. School courses
3. Tuned PID controllers
4. Designed and
implemented advanced
regulatory control
strategies
5. Implemented
multivariable model
predictive control
strategies

1 2 3 4 5
2 Honeywell - CONFIDENTIAL
 What is Advanced Process Control?

• Control
- Manipulating
p g something
g to get
g a desired effect to achieve
a specific purpose.
 Flow rates
 Temperatures
 Pressures
 Power
 Qualities
• Process
- Pertaining to the process industries
• Advanced
- Building on top of basic single-input, single-output control
- Using more than raw measurement feedback
- Handling
H dli interactions
i t ti
- Optimizing
3 Honeywell - CONFIDENTIAL
 Incentive for Advanced Control

• Reduce variability around setpoint

- aallows
o s you to ope
operate
ate c
closer
ose to a co
constraint
st a t
 less give-away

• More stable and safe operation

- Reduced
R d d problems
bl
• Increased plant flexibility
- Move from one operating
p g region
g to another
• Enables operators to move plant conditions closer
to optimum economic operation
• Reduction
R d ti in i maintenance
i t
- Reduce thermal, mechanical, and hydraulic stresses
- Extend catalyst
y life

4 Honeywell - CONFIDENTIAL
 Advanced Control – What’s It all About?

• Advanced control and optimization is key to running

process plants
p p
V i bl
Variable
- Safely Constraint Limit
- Reliably $
$ Advanced
- Efficiently Good Regulatory
Control

- Responsively Poor control

Control
Time

• Typically APC projects give payback time of < ½ year

• Wide range of methods a available

ailable
- Pick the right one for the job
 Cost vs benefit

5 Honeywell - CONFIDENTIAL
 Nature of Control Problem
Variables that vary but cannot be
manipulated
Also called feedforward, disturbance and
independent variables

Variables that Variables to be kept

can be
b adjusted
dj t d at a value or within
Also called
manipulated and
Process a range
Also called controlled
independent variable
and dependent variable

Control needed to:

1 Protect constraints
1.
2. Minimize impact of disturbances
3. Move
3 o e the
t e Process
ocess to new
e ope
operating
at g points
po ts
for safety and greater profitability
6 Honeywell - CONFIDENTIAL
 Topics

• Advanced Control Overview

• Basic Regulatory Control Troubleshooting (Optional)
• Advanced Regulatory Control
• Multivariable Predictive Control

7 Honeywell - CONFIDENTIAL
Advanced Process Control
Overview

© 2011 Honeywell. All rights reserved.

 Levels of Process Control

• Basic regulatory
• Advanced regulatory (intermediate regulatory)
• Multivariable, model based
• Constrained economic optimization
p
• Multi-unit constrained economic optimization
• First principles economic optimization (RTO)
• Planning, Scheduling, Supply chain optimization
Supply
Chain
Plant-wide
Optimization

Multi-Area Optimization

Soft Model Predictive Control Early Event

Se so s
Sensors Detection
Advanced Regulatory Control, Safety Systems

Distributed Control System, Basic Controls

9 Honeywell - CONFIDENTIAL
 Optimization Local LP or QP Process Unit Simulation based
Optimization Optimization Site wide Optimization

Model Based
Predictive
P di ti Multivariable
M lti i bl Engineering
E i i
Control Control Predictive models with
Control predictors

Advanced
Feedforward Dynamic On-stream Constraint
Regulatory
Control
control decoupling Analyzers Control

Regulatory
Single Cascade
Control
PID

Different technologies available at each level

10 Honeywell - CONFIDENTIAL
 Basic Regulatory

• Single sensing element – single final control element

- Single Input Single Output (SISO)
• Examples
- Proportion-Integral-Derivative (PID) controllers
- Profit Loop (Honeywell model-based controller)
• Also, Simple cascades, Ratio control
• Primary goal:
- Operational stability
 Disturbance rejection
 Provides
P id operators
t with
ith a h
handle
dl FC
to control the process

11 Honeywell - CONFIDENTIAL
 Advanced Regulatory

• Improve over performance achieved w/ Basic

Regulatory
g y Controllers
- Normally still limited to DCS tools
• Pressure or temperature compensation
• Feed-forward
F df d based
b d 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

12 Honeywell - CONFIDENTIAL
 Advanced Regulatory Example

• Coil Outlet Temperature Control

13 Honeywell - CONFIDENTIAL
 Multivariable, Model Based

• Simultaneous consideration of:

- many controlled variables (CVs), e.g. process qualities
- by adjusting many manipulated variables (MVs), e.g.
flowrates, temperatures, pressures
- while incorporating feedforward or disturbance variables
(DVs)
• Internal model of process behavior
- predicts future paths of controlled variables
• Multiple Input Multiple Output (MIMO)

14 Honeywell - CONFIDENTIAL
 Distillation Control Example

PC
FC
H2 & CH4
C3 Refrig.
TI

LC CVs
FC

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

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

C3 refrig. pres.

15 Honeywell - CONFIDENTIAL
 Constrained Economic Optimization

• Premise
- So
Some
eoof controlled
co t o ed variables
a ab es are
a e not
ot controlled
co t o ed to setpoints
setpo ts
 Required to be maintained between high and low ranges

• User specifies an objective function

- Assign
A i costs
t or values
l to
t variables
i bl
- Translates into a direction to move the manipulated variables
• Optimizer
p finds best position
p (corner)
( )
- Within space constrained by controlled and manipulated
variable limits

16 Honeywell - CONFIDENTIAL
 Constrained Economic Optimization2

• Optimizer is a “steady state controller”

- Ca
Calculates
cu ates o
one
e set o
of MV moves
o es to maximize
a eoor minimize
e
objective function
- Generally too aggressive to implement directly in closed
loop
- Typically used in conjunction with a multivariable control
engine to implement solution

17 Honeywell - CONFIDENTIAL
 Class Exercise

• Debutanizer & Depropanizer Units

- Define MV's, CV's, and DV's for one or two controllers
- Define LP costs for each variable using a Product Value
Optimization (PVO) approach

18 Honeywell - CONFIDENTIAL
 P
P
T
T C3s/C4s Propane

dPtop F Q
F Q

Reflux Reflux %iC4

%iC5
F T
T F

D b
Debutanizer
i Depropanizer
F
dPbottom F Steam Reboiler
Steam Reboiler
%C3
Feed %nC4 L
L Q
Q
T
T
C5s Butane

19 Honeywell - CONFIDENTIAL
 Gas Plant 1

• Feedstock
- C2 to C8
• Products
- C2 and lighter to fuel gas
- C3 (Propane)
- C4 (Butane)
- C5+ (Heavier cut for further processing)
• Product Quality Specifications
- Max. % nC4 in C5+
- Max. % iC5 in C4
- Max. % C3 in C4
- Max. % iC4 in C3

20 Honeywell - CONFIDENTIAL
 Gas Plant 2

• Objectives
- Maintain product qualities within specifications
- Minimize energy consumption (steam)
- Maximize feedrate
• Process
P C
Constraints
t i t
- Steam utilization
- Off-gas
g from debutanizer
- Flooding (both towers)
- Depropanizer reflux drum level control valve position

21 Honeywell - CONFIDENTIAL
 MVs DVs CVs

22 Honeywell - CONFIDENTIAL
 MVs DVs CVs

Unit feed Feed temperature C3/C4 % iC5 (Q)

C5+ % nC4 (Q)
DeC4 reflux
C3 iC4 (Q)
DeC3 reflux
iC4/nC4 % C3 (Q)
DeC4 steam
DeC4 P (flooding)
DeC3 steam DeC3 P (flooding)
DeC4 pressure DeC4 fuel gas valve %
DeC3 pressure DeC3 reflux drum valve %
C5 flow (PVO)
One possible design: C3 flow (PVO)
One controller iC4/nC4 flow (PVO)
DeC4 steam valve %
Missing instrumentation DeC3 steam valve %

23 Honeywell - CONFIDENTIAL
 Gasplant PVO Objective Function



-  Product flows  Product prices 

J    Feed flows  Feed costs 
 

  Energy/utility flows  Energy/Utility costs 


J = -C5 x $50/MBPD – C4 x $30/MBPD

-C3 x $10/MBPD + FEED x $40/MBPD
+(DEBUT STEAM) x $2/TPH
+(DEPROP STEAM) x $2/TPH

24 Honeywell - CONFIDENTIAL
 What do you think?

The value of a disturbance

57%
variable (DV) should be
minimized to reduce
disturbances to a process. 43%

1. True
2. False

1 2

25 Honeywell - CONFIDENTIAL
Basic Regulatory Control
Troubleshooting

© 2011 Honeywell. All rights reserved.

 Basic Controls (BC) Importance

• All Advanced Control depends on Basic Control

- BC must be able to attain and maintain requested setpoint
- BC must be able to adequately reject high frequency
disturbances
- BC must NOT introduce process variability
• DO NOT attempt to correct BC deficiencies with
changes to the Advanced Controls

27 Honeywell - CONFIDENTIAL
 Most Common Control Loop Problems

• Our experience shows...

- More
o e than
t a half
a o of co
control
t o valves
a es aare
e incorrectly
co ect y spec
specified
ed
 nonlinear installed characteristic
 incorrectly sized for application
- More than 50% of all control valves have mechanical
problems
 Hysteresis 26%
70
 Stiction 27% 60 68
Suboptimal tuning
- Almost 20% of control strategies 50
51 Incorrect valve
40
can be improved upon design
Hysteresis
30
- More than two
two-thirds
thirds of PID 20 27 26
19 Sti ti
Stiction
controllers have suboptimal 10
Poor control
tuning settings 0
% Loops with particular
strategy design

deficiency

Many published studies

support these numbers

28 Honeywell - CONFIDENTIAL
 How well are our control loops performing?

• Report Card
- 10% - 20% of industrial control loops run
in manual
- About one third of controllers produce
more variabilityy in auto than in manual
- Roughly 75% of control loops are not
very effective in
 setpoint following
 disturbance rejection

29 Honeywell - CONFIDENTIAL
 Why is this the case?

- Controllers are tuned during Plant Commissioning or 1st

Startup
- PI / PID parameters are set based on experience
- Problem loops are optimized through trial and error
- These settings
g normallyy remain for g
good
- Inadequate Control Valve Maintenance or analysis
- Any problems in future are addressed through trial and error
tuning
- Downsizing: lack of people close to the process responsible
for control performance

30 Honeywell - CONFIDENTIAL
 Increasing Efficiency Through
Variability Reduction

- Process constraint: 160 F

- Before optimization, Standard Deviation was 10
 Setpoint = 130
- After Optimization, Standard Deviation was 5
 Setpoint could be increased to 142
- 12 Deg increase in mean temperature == Increase in efficiency
0.1

=5
You will find 0.08

Increasing Process Efficiency

opportunities

Cons
0.06

for

straint
0.04  = 10
improvement! Operations
Comfort
0.02
Zones

 = Standard Deviation 0
0 20 40 60 80 100 120 140 160 180 200

31 Honeywell - CONFIDENTIAL
 Layers Supporting Loop Performance

• Optimizing process control

performance involves much
p
more than the trial-and- Loop
error tuning of controllers!! Performance

Controller
Most common mistake Settings
is to start with tuning.
Instrumentation &
Q. Why is that the Final Control Elements
most common
mistake? Control Strategy Design

A. Tuning is usually
the easiest to modify Process Design

32 Honeywell - CONFIDENTIAL
 Addressing Control Loop Problems

• Identify the loop causing the control problem

Problem Experienced:
Poor
oo Co
Composition
pos t o Co
Control
to

Solution Focus:
Composition Control Loop Tuning

Actual Source of Problem:

Sticky Cooling Water Control Valve

Optimization requires a more in-depth approach than simply tuning a controller

33 Honeywell - CONFIDENTIAL
 Process Design Issues

• Problems affecting control

• Designers lack process
control knowledge
• Process design choices lead
t a marginally
to i ll controllable
t ll bl
or uncontrollable process
• Control loops
p include
unnecessary lags and dead
times
• Control valves are oversized
• Control capability is under
designed

34 Honeywell - CONFIDENTIAL
 Control Strategy Problems

• Not using feedforward control

• Not using / misusing cascade control
• Not using controller gain scheduling
• p g interacting
Not decoupling g control loops
p
• Nonlinear ratio control loops
• Using overly complex designs (very easy to do in
a DCS).
DCS) Oft
Often will
ill iinitially
iti ll workk but
b t are nott robust
b t
over time.

35 Honeywell - CONFIDENTIAL
 Measurement Problems

• Transmitter span excessively wide

• Transmitter span too narrow
• Sensors positioned incorrectly
• gg g / fouling
Plugging g

36 Honeywell - CONFIDENTIAL
 Control Valve Deficiencies

• Incorrect Valve Sizing

- Full flow desired at 75% - 85%. Often IMPORTANT
the process changes over time but Always remember that,
the valves do not. just because a controller
output is moving, the
• Hysteresis valve may NOT be
• Stiction moving at all, or to the
right place!
• Nonlinearity of response
• Positioner Overshoot (defect)
• Asymmetric Response
• Butterfly valves

• Remedy: Find them and fix them

37 Honeywell - CONFIDENTIAL
 Identifying Hysteresis

Hysteresis: The valve actually moves to a different

position given the same controller output, depending
on the direction of valve movement.
movement
Same Output,
Different PV =
different valve
position

Output
decrease
to 40.2

Output
i
increase
to 40.2
38 Honeywell - CONFIDENTIAL
 Hysteresis – Causes, Concerns, Cures

• Causes:
- Play in mechanical linkages
- Excessive friction at valve stem seal
- Undersized actuator
- Sticky positioner
• Concerns:
- It decreases control loop performance
- Adds dead time to control loop
- Increases variability
- Hysteresis results in cycling on:
 On an integrating process
 On self-regulating process tuned to overshoot

• Cures TUNING CANNOT

- Overhaul or replace actuator linkages CURE
- Replace valve stem seal
HYSTERESIS!
- Use a larger actuator
- Install a positioner
- Overhaul an existing positioner
39 Honeywell - CONFIDENTIAL
 Stiction

• Stiction is short for Static Friction. With stiction more

force is required
q to induce movement than to sustain
movement. Like sliding a heavy box on a concrete
floor. You will see Output stair-step changes with no
PV changes.
g

Output Output
increases i
increases
3 times 4 more
before times
valve before
finally valve
moves finally
and pv moves
increases again
and pv
increases
40 Honeywell - CONFIDENTIAL
 Stiction: Causes, Concerns, Cures

• Causes:
- Over-tightened
g or sticky
y valve stem seal
- An undersized actuator
- A sticky positioner
• Concerns:
- It switches the process gain from zero to
excessively high
- Introduces (stick-slip)
(stick slip) cycling into the control loop
- Increases variability
• Cures:
- Replace or lubricate the valve stem seal TUNING CANNOT
- Use a larger actuator CURE STICTION!
- Install a p
positioner
- Overhaul an existing positioner
41 Honeywell - CONFIDENTIAL
 Asymmetric Response

• Process characteristics are direction dependent.

Slow Fast
response response
to output to output
decrease increase
of same
size

42 Honeywell - CONFIDENTIAL
 Asymmetric Response

• Causes
- Faulty positioner
- Air Leaks in positioner / diaphragm / piping
• Concerns
- Poor Control
• Cures
- Repair valve / positioner

• Tune for worst case in the interim

TUNING CANNOT
CURE
ASYMMETRIC
RESPONSE!

43 Honeywell - CONFIDENTIAL
 Decreasing Loop Performance:
Decreasing Benefits

• Reduced Profit—Running Further From Constraints

• Increased Maintenance Costs—Equipment and Valves Cycling
• Increased
I d Operator
O t Loading—Alarms,
L di Al IIntervention
t ti Required
R i d
Limit of Operability
Alarm Limit

Lowering of
Reduction Setpoint to
in Benefits Maintain Safety
Margin

Variability
Increases
Over Time

After Optimization Performance Decreases 2-3 Years Later

Over Time
44 Honeywell - CONFIDENTIAL
 What do you think?

The figure below illustrates

“hysteresis”.
73%
1. True
2 False
2.

27%

1 2
45 Honeywell - CONFIDENTIAL
Advanced Regulatory Control

© 2011 Honeywell. All rights reserved.

 Advanced Regulatory Control

• Cascade Control
• Feed-forward Control
• Ratio Control
• Inferential Control
- Simple process calculations – Heat and material balance
- Simple inferential estimations, duty control
• Pressure or temperature compensation
• Non Linear Control
• Multiple Input Single Output (MISO)
• Primary Goals:
- Enhanced Stability
- Enhanced performance
47 Honeywell - CONFIDENTIAL
 Cascade Control

Distinguishing features:
• Two FB controllers but only a single control valve
• The output of the primary controller becomes the set point of the
secondary controller
• Two FB control loops are "nested" with the "slave" (or "secondary")
inside the "master"
master (or "primary")
primary ) control loop
• Used for Disturbance rejection
Requirements:
• Secondary loop process dynamics must be at least four times
as fast as primary loop process dynamics
• Secondary loop must have influence over the primary loop
• Secondary loop must be measured and controllable
• Secondary loop should reject disturbances quickly
• Commonly encountered in industries
- Secondary
S d lloop: Fl
Flow
- Primary loop: Level, Temp, Pressure or Composition
48 Honeywell - CONFIDENTIAL
 Cascade Control: Distillation Columns

Tray 10 Temperature Control

TT TC
Tray 10

SP

FC

FT

49 Honeywell - CONFIDENTIAL
 Cascade Control: Furnace

Furnace Outlet Temperature Control

50 Honeywell - CONFIDENTIAL
 Feed Forward Control

Distinguishing Features
• Measures the Disturbance before its effects appear
pp in the
process measurements and computes appropriate control
action
• Veryy effective for fast load changes,
g , such as flow loops
p
• Widely used in Distillation columns
Requirements
• Disturbance must be measurable
• Sufficiently accurate model of the process needed
• Accurate measurements
Drawback of Pure Feedforward Control
• Process model mismatch results in steady-state
steady state offset from set point
• Measurement error can be more detrimental
51 Honeywell - CONFIDENTIAL
 Feed Forward with Feedback Trim

• Since perfect models of disturbance characteristics

are rare, Feed Forward will not p
provide perfect
p
disturbance rejection

• Use
U aFFeedback
db k Trim
Ti

• Tasks of the Feedback Trim:

- Rejects that portion of the measured disturbance that make
it past the feed forward element (model mismatch)
- Works to reject unmeasured disturbances
- Provides set point tracking capabilities as needed

52 Honeywell - CONFIDENTIAL
 Feed Forward Control with Feedback Trim

Feedback control eliminates any error due to model mismatch

53 Honeywell - CONFIDENTIAL
 Ratio Control

• Type of a Feed-forward control

• Two streams must be adjusted in a ratio to each other
• Generally used for controlling the flow rates of two
streams
• Two
T configurations:
fi ti
- Calculate Ratio
- Calculate Set point
• Wild Stream & Controllable Stream
• Common applications:
- Gasoline Blending
- Air-Fuel ratio in furnace
- Hydrocarbon / H2 ratio in Reforming
- Purge / Recycle ratio in Reactor systems
- Stoichiometry control
54 Honeywell - CONFIDENTIAL
 Ratio Control – Example

N2 / H2 Ratio Control in Ammonia Synthesis Reactor

55 Honeywell - CONFIDENTIAL
 Ratio & Cascade Control

× Ratio
TC

FT TT

Flue
SP Gas

FC
FT
Process
Fluid
Fuel

Furnace Coil Outlet Temperature Control

56 Honeywell - CONFIDENTIAL
 Feed Forward & Cascade Control

+
SP

TC PC Steam

PT
FF TT

TT
Feed
Condensate

Heat Exchanger Outlet Temperature Control

57 Honeywell - CONFIDENTIAL
 Cascade, Feed Forward & Ratio Control

• Coil Outlet Temperature Control

58 Honeywell - CONFIDENTIAL
 Inferential Control

• Problem: Controlled variable cannot be measured

or has large
g sampling
p g period.
p

• Possible solutions:
- Control a related variable (e.g., temperature instead of
composition).
- Inferential control: Control is based on an estimate of the
controlled
t ll d variable.
i bl
 The estimate is based on available measurements
• Examples: empirical relation, Kalman filter
 Modern
M d tterm: soft
ft sensor

59 Honeywell - CONFIDENTIAL
 Inferential Control Scheme

• Approaches
- First principle model (usually simplified)
- Regression of process data
- ((Empirical
p models))
• Linear
- Ordinary least squares
 For simple
p cases
- Partial least squares
 When input data is correlated
• Non-linear
- Non-linear partial least squares
- Neural nets
60 Honeywell - CONFIDENTIAL
 Inferential Modelling

• Building empirical models

- Collect data (e.g. process data
plus lab results)
- Validate data
- Understand which are most
influential variables on variable
to be inferred
 e.g.
g temperature
p and p
pressure
predict distillation column qualities
- Appropriately time-shift input
Composition prediction
data
- Regress data to develop model
- Test model against other data
or on plant
- Use in control when a good
inference is developed
61 Honeywell - CONFIDENTIAL
 Inferential Control: Reboiler Duty

Computed Qspec
Reboiler Duty
Control

AC

TT SP

FC
TT FT
AT
Quench Water

Qspec
Fsp 
C p (Tin  Tout )

62 Honeywell - CONFIDENTIAL
 Inferential Control: Internal Reflux Control

TT

C.W.
C p Fex (Toh  Tr )  Fint H vap
 C p (Toh  Tr ) 
Fint  Fex 1  
  H vap 
TT
Fintspec
Fex 
C p (Toh  Tr )
FT AT 1
H vap
FC

SP AC
Internal
Fex Fint
Reflux
Controller

63 Honeywell - CONFIDENTIAL
 Flow Compensation

FY

PV
Flow
Compensation
Calculation
SG FI FC

Description
• Calculation to adjust the raw measurement signal to account for
differences between flowing conditions and design of flow orifice

Issues
• Gas flows are compensated
p for temperature
p and p
pressure
• Liquid flows can be compensated for temperature or density
• Critical for modeling
64 Honeywell - CONFIDENTIAL
 Non Linear Control

• When controlled variables are highly non-linear function of

process parameters (e.g. Viscosity, FCC Reactor Yield, etc.)
• Lack of expertise or calculation ability, many plants use poor
control
• Solution : Non Linear Control – Many approaches
- Gain Scheduling (Variable Gain)
- Nonlinear Transformation
- First Principle models
- Gray-box models
• Examples
- Nonlinear Level Control
- Fuel Oil Blending: Viscosity control
- FCC Reactor: Severity Control
- Ethylene
y Cracker control
- Profit®NLC

65 Honeywell - CONFIDENTIAL
 Non Linear Level Control

• Level in a drum is controlled very well by manipulating the

outlet flow controller
• Drawback => Outlet flow varies wildly which upsets
downstream process
• Solution => Non-linear Level Control (Gain Scheduling)
• Uses drum capacity to average out the disturbance and
results in steady outlet flows – Surge drums

Level Control –
Cascade

66 Honeywell - CONFIDENTIAL
Non Linear Level Control: Gain Scheduling

• Exact level of a surge drum is not

important, the controller can be tuned
y loosely.
very y

• “Gain on error squared" controller:

- Gain proportional to (error)^2
- Small control action in the vicinity of the
setpoint and large control action if the
level changes far from the setpoint

• Gap controller:
- Gain divided into 2 or 3 segments
- Middle segment on either side of the
setpoint has a low gain to avoid
excessive action
- Outer segments have a higher gain for a Gain Scheduling options
rapid action
- User can set the breakpoints and gains
above & below the setpoint differently

• Implementation - P or PI but no D
67 Honeywell - CONFIDENTIAL
 Non Linear Transformation

• Non Linear relation

- Linearize
ea e tthe
e variables
a ab es app
appropriately
op ate y & co
control
to

• Example
- Fuel Oil Blending: Viscosity control
- Log Transformation of Viscosity vs. Temperature
- Ln (Ln(µ+k)) = A Ln T + B
Heavy Btms

Kerosene + SP µ
Ln (Ln(µ+k)) C
Controller
ll Blending
µset
+

Sensor

Ln (Ln(µ+k))

68 Honeywell - CONFIDENTIAL
 First Principle Model: FCC Severity Controller

• Reactor Yields – Non linear function of severity (temperature,

feed rate, pressure, feed properties, etc.)
• Lack of expertise or calculation ability, many plants only
control temperature
• Poor Control – pproduct yields
y vary
y considerably y
• Examples: FCC Units, Hydrocrackers, Ethylene Crackers, etc.

Solution : Severity Controller

• Use First Principle model
• A form of calculated variable control
• Honeywell
– FCCU Severity Controller – Profit MAX, FCCU Toolkit
– Delayed Coked control

69 Honeywell - CONFIDENTIAL
Multivariable Predictive
Control

© 2011 Honeywell. All rights reserved.

 Model Predictive Control Technology

• A well-established industrial control technology

that dates back over 30 years.
y
• A wide variety of industrial technology offerings
are available on the market.
• MPC ttechnology
h l h
has ~8000
8000 vendor
d ddocumentedt d
industrial applications.
• Refiningg and Petrochemical applications
pp are
typically dominant but MPC is being rapidly
adopted in other markets.

Adapted from the following literature source:

Qin, Joe S., and Badgewell, Thomas A.,
“A Survey of Industrial Model Predictive Control Technology (2001 Draft)”

71 Honeywell - CONFIDENTIAL
 Advanced Process Control

Steady State
State,
Multi-unit Coordination Dynamic, Non-linear
Non-linear Process & Global Optimization Process Models
Models

Multivariable Control
& Unit Optmization

Advanced Regulatory
Control

Base Level Regulatory Control

Measurement and Control Devices

The
TheHierarchical
HierarchicalLayers
Layersofof
Advanced
ProcessProcess
ControlControl

72 Honeywell - CONFIDENTIAL
 Basic Feedback Control

Disturbance
Disturbance
Variable (DV)

Process

Adjustment Measurement

Controller
Manipulated
M i l d Controlled
C t ll d
Variable (MV) Variable (CV)

Setpoint
73 Honeywell - CONFIDENTIAL
 Debutanizer Column
C li water
Cooling t

Interaction: CV1

L
Increase in reflux (MV1) TI
C
causes: MV1
F F
p temperature
in top p C C CV3

in distillate DV1

CV4 Distillate

Feed
 in base temperature
mp

 in base product F MV2

C
Steam
CV5

CV Controlled variable L
Base
DV Disturbance (FF) variable C product
p
MV Manipulated variable CV2
TI

74 Honeywell - CONFIDENTIAL
 Debutanizer Column
C li water
Cooling t

Interaction: CV1

L
Increase in steam (MV2) TI
Reflux C
causes: MV1
F F
p temperature
 in top p C C CV3

 in distillate DV1
CV4 Distillate

Feed
 in bottom
m temperature
mp
Both steam and reflux
 in base product MV2
F
affect most column C
variables CV5
Steam

L
Base
C product
p
CV2
TI

75 Honeywell - CONFIDENTIAL
 MPC Overview

Model Predictive Control (MPC) technologies

use p
process relationships
p to better control and
optimize complex industrial processes.

The terminology shown below will be defined

defined.

Setpoints, Ranges, Optimization Objectives

CV’s
DV’
DV’s MPC MV’s
MV s

P
Process Models
M d l

76 Honeywell - CONFIDENTIAL
 What is a CV?

CV's are Controlled Variables. The MPC

controller is designed
g to "control" these variables
(within a range or to a setpoint.) Representative
examples would be product qualities, valve
outputs,
p level %, etc.

Setpoints, Ranges, Optimization Objectives

Controlled Variables
(CV’s)
Manipulated Variables
MPC (
(MV’s)
)
Disturbance Variables
(DV’s)

Process Models

77 Honeywell - CONFIDENTIAL
 What is an MV?

MV's are Manipulated Variables. The MPC

controller can move these values ((within a range)
g )
to control the CV's. Representative examples
would be flow, pressure, and temperature
controller setpoints.
p

Setpoints, Ranges, Optimization Objectives

Controlled Variables
(CV’s)
Manipulated Variables
MPC (
(MV’s)
)
Disturbance Variables
(DV’s)

Process Models

78 Honeywell - CONFIDENTIAL
 What is a DV?

DV's are Disturbance Variables. These variables

impact
p the p
process but the controller is not allowed
to move the value. DV's are included so that
changes to these variables can be accounted for
byy the controller. A typical
yp example
p of a DV would
be a unit feedrate.

Setpoints, Ranges, Optimization Objectives

Controlled Variables
(CV’s)
Manipulated Variables
MPC (MV’s)
Disturbance Variables
(DV’s)

Process Models

79 Honeywell - CONFIDENTIAL
 What is a Process Model?

For MPC, a Process Model is the

mathematical representation
p of the
dynamic response of one process
variable to the change in another
process variable.
p

Setpoints, Ranges,
Optimization
Opt at o Object
Objectives
es

CV’s
DV’s MPC MV’
MV’s

Process Models

80 Honeywell - CONFIDENTIAL
 MPC Structure

Optimization Parameters
(Prices, Costs, Directions, Targets)

Controlled Variables
(CV's)
MPC Control Targets
(Setpoints/Ranges)
Disturbance Technology
Variables (DV's)

MV’s

FC FC TC PC PC

Manipulated
M i l t d
Variables (MV's)

81 Honeywell - CONFIDENTIAL
 Model Predictive Control
Cooling water

CV1

LC
TI

FC MV1
FC

MPC Application
DV1

Feed
MV1 MV2

MV2 CV1
FC

Steam

LC
Base CV2
product
d
CV2
TI
82 Honeywell - CONFIDENTIAL
 Objectives of MPC

• Using multiple manipulated variables (MVs) and

multiple
p controlled variables (CVs) ( )
- Prevent violations of MV and CV constraints.
- Drive certain CVs to their optimal values, while maintaining
other CVs within p
prescribed ranges.
g
- Prevent excessive movement of the MVs.
- Control as much of the process as possible in the presence
of unavailable sensors or actuators.

83 Honeywell - CONFIDENTIAL
 Why Models?

• The essence of model predictive control (MPC)

- Predict
ed ct tthe
eeexpected
pected be
behavior
a o o of important
po ta t p
process
ocess values
a ues
without control.
 Using dynamic models
- Compare
p the predicted
p behavior to the desired behavior.
- Calculate values specified to the manipulated variables
that minimize the deviation of the predicted behavior from
the desired behavior.
• Thus the model is the essential element of an MPC
controller

84 Honeywell - CONFIDENTIAL
 What is a process model?

Process models can take on many forms…

Adapted from “Advanced Process Control” by Mark J. Willis and Ming T. Tham, (1994)

85 Honeywell - CONFIDENTIAL
 Generalized Statement of Technique

• The aim is to find values for the manipulated

variables that force a model to follow a desired
reference system or trajectory.

86 Honeywell - CONFIDENTIAL
 Multivariable Control & Models

These predictions
Th di ti
are coming from
the model

Sampling Instant

87 Honeywell - CONFIDENTIAL
 Conventional Model Predictive Control

Setpoint

Known Values e1 e2 e 3 e4 e5

CV Predicted
Unforced
Response

MV
Assumed Values

Past t=0 Future

88 Honeywell - CONFIDENTIAL
 Conventional Model Predictive Control

Setpoint

Known Values e1 e2 e 3 e4 e5

CV Predicted
Unforced
Response

MV
Assumed Values

MV Moves
Past t=0 Future

89 Honeywell - CONFIDENTIAL
 MPC – A Novel Approach

Past Future

Optimal Response

Setpoint PPredicted
Predicted
P di di
t dt d

CV Unforced
Unforced
Response
Response
Control Funnel

MV Assumed Values

Minimum Effort Move

T=0

90 Honeywell - CONFIDENTIAL
 MPC Features Review

• Multi-Variable
- MPC helps coordinate and decouple the effects multiple
process variable interactions.
• Model-Predictive
- Dynamic models are used to predict process behavior
and predictions are used to proactively control the
process.
• Constraint Aware
- MPC monitors and maintains MV’s and CV’s limits while
it is controlling the process.
• O
Optimized
ti i d Control
C t l
- MPC has integrated optimization capabilities to drive
applications toward specified design objectives.

91 Honeywell - CONFIDENTIAL
 So Where does MPC Apply?

Potential MPC applications can be identified against

the following
g criteria:
• Is the Process Highly Interactive?
• Does the Process have Defined Objectives
• Does the Process have More MV’s than CV’s
• Does the Process have More CV’s than MV’s
• Does the Process have Measurable DV’s
DV s

Example MPC Applications

- Refinery Processes - Offshore Oil Production
- Petrochemical Plants - Paper Machine Control
- Chemical Reactors - Pulp & Paper
- Dairy Production - Mineral Processing
- Supply & Distribution
92 Honeywell - CONFIDENTIAL
 Where does MPC (Generally) Not Apply?

• Processes With Fast Execution Requirements

- MPC applications are typically run at once per minute.
Five-second execution is occasionally done but most
technologies are not capable of <1s execution for large
control strategies.
• Compressor Anti-Surge Controls
- Requires executions on the order of milli-seconds.
• Simple Heater COT Controls
- Disturbances in this case are well defined against a single
MV. Traditional feedforward controls will generally perform
as well as MPC with potentially faster control execution
execution.
• Whenever conventional control is clearly sufficient
and much easier to implement than MPC
- Single constraint control problems.

93 Honeywell - CONFIDENTIAL
 Example: Distillation Control

PC
FC
H2 & CH4
C3 Refrig.
TI

LC CVs
FC

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

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

C3 refrig. pres.

94 Honeywell - CONFIDENTIAL
 MPC Implementation

• Develop APC Project Plan

• Assess Basic Regulatory Control Layer
• Determine inferential models and put on-line
• Perform qquick plant
p test to get
g approximate
pp dynamic
y
models
• Carry out full plant step tests
• Id tif dynamic
Identify d i model
d l
• Tune/test controller offline
• Put controller on-line and fine tune

95 Honeywell - CONFIDENTIAL
 Benefits of MPC

MPC Benefits are field proven and generate high

rates of return on investment.

Tangible Benefits Intangible Benefits

• Increased Throughput • Improved Process
• Improved Yields Monitoring
• Decreased Operating Costs • Improved Operator
Effectiveness
• Improved Quality
Consistency • Reduced Plant Variability
• Increased Operating • Better Process
Fl ibilit
Flexibility Information
• Improved Process Stability

Payback periods < 6 months are common

96 Honeywell - CONFIDENTIAL
 Benefits of MPC

The following are generally accepted ranges for APC and optimization benefits
Petrochemicals Benefits (/yr)
Ethylene 2-4% Increase in Production
VCM 3-5%
3 5% Increased Capacity / 1-4%
1 4% Yield Improvement
Aromatics (50KBPD) 3.4M - 5.3M US$
Chemicals Benefits (/yr)
Ammonia 2-4% Increased Capacity / 2-5% Less Energy/Ton
Polyolefins 2-5% Increase in Production/Up to 30% faster grade transition
Benefits (/yr)
Oil & Gas
1-5% Increase in Production
Upstream production
Industrial Utilities Benefits (/yr)
Cogeneration/Power Systems 2-5% Decrease in Operating Costs
Pulping Benefits (/yr)
Bleaching 10-20% Reduction in Chemical Usage
TMP (Thermo Mechanical Pulping) $1M-$2M

Refining Benefits ((US$/yr)

y)
Crude Distillation (150 KBPD) 2.7M - 7.0M
Coking (40 KBPD) 2.2M - 4.8M
Hydrocracking (70 KBPD) 3.3M - 7.6M
Catalytic Cracking (50 KBPD) 2.4M - 5.4M
Reforming (50 KBPD) 1 8M - 4.7M
1.8M 4 7M
Alkylation (30 KBPD) 1.1M - 2.8M
Isomerization (30 KBPD) .3M - 1.8M

97 Honeywell - CONFIDENTIAL
 What do you think?

The greater an MV’s move

magnitude, the more that it
amplifies modeling error. 50% 50%

1. True
2. False

1 2
98 Honeywell - CONFIDENTIAL
99 Honeywell - CONFIDENTIAL