You are on page 1of 50

Control Loop Foundation

for
Batch and Continuous Control
GREGORY K MCMILLAN

use pure black and white option for printing copies

11/10/2008

Presenter

Greg is a retired Senior Fellow from Solutia Inc. During his 33 year career with
Monsanto Company and its spin off Solutia Inc, he specialized in modeling and
control. Greg received the ISA Kermit Fischer Environmental Award for pH
control in 1991, the Control Magazine Engineer of the Year Award for the
Process Industry in 1994, was inducted into the Control Process Automation
Hall of Fame in 2001, and honored by InTech Magazine in 2003 as one of the
most influential innovators in automation. Greg has written a book a year for the
last 20 years whether he needed to or not. About half are humorous (the ones with
cartoons and top ten lists). Presently Greg contracts via CDI Process and
Industrial as a principal consultant in DeltaV Applied R&D at Emerson Process
Management in Austin Texas. For more info visit:
http://ModelingandControl.com
http://www.easydeltav.com/controlinsights/index.asp (free E-books)
11/10/2008

See Chapter 2 for more info on Setting the Foundation

11/10/2008

See Chapters 1-7 for the practical considerations of improving tuning and valve dynamics

11/10/2008

See Appendix C for background of the unification of tuning methods and loop performance

11/10/2008

See Chapter 1 for the essential aspects of system design for pH applications

11/10/2008

Overview
)

This presentation covers highlights or low lights of current loop


performance and how to improve batch and continuous processes:

Pyramid of Technologies
Valve and Flow Meter Performance
Process Control Improvement Examples
Basic Control Opportunities Summary
Reactors and Column Loop Tuning
Facts of Life
Transfer of Variability for Batch
Sources of Disturbances
Transition from Basic to Advanced Regulatory Control of Batch
Online Data Analytics for Batch and Continuous Processes
Virtual Plant
Uses and Fidelities of Dynamic Process Models
What we Need
Columns and Articles in Control Magazine

Pyramid of Technologies
TS
APC is in any technology that
integrates process knowledge

RTO
LP/QP

The greatest success has been


Achieved when the technology
closed the loop (automatically
corrected the process without
operator intervention)

Ramper or Pusher
Model Predictive Control

Foundation must be large and


solid enough to support upper
levels. Effort and performance
of upper technologies is highly
dependent on the integrity and
scope of the foundation (type
and sensitivity of measurements
and valves and tuning of loops)

Property Estimators
Fuzzy Logic
Abnormal Situation Management System
Process Performance Monitoring System
Loop Performance Monitoring System
Auto Tuning (On-Demand and On-line Adaptive Loop Tuning)
Basic Process Control System
TS is tactical scheduler, RTO is real time optimizer, LP is linear program, QP is quadratic program

11/10/2008

Loops Behaving Badly


A poorly tuned loop will behave as badly as a loop
with lousy dynamics (e.g. excessive dead time)!

1
Ei = ------------ Ti Eo
Ko Kc

You may not want to minimize the integrated


error if the controller output upsets other loops.
For surge tank and column distillate receiver
level loops you want to minimize and maximize
the transfer of variability from level to the
manipulated flow, respectively.

where:
Ei = integrated error (% seconds)
Eo = open loop error from a load disturbance (%)
Kc = controller gain
Ko = open loop gain (also known as process gain) (%/%)
Ti = controller reset time (seconds)
(open loop means controller is in manual)
Tune the loops before, during, and after
any process control improvements
11/10/2008

Unification of Controller Tuning Settings


All of the major tuning methods (e.g. Ziegler-Nichols ultimate oscillation and reaction curve,
Simplified Internal Model Control, and Lambda) reduce to the following form for the maximum
useable controller gain

1
K c = 0.5 *
Ko max
Where:
Kc = controller gain
Ko = open loop gain (also known as process gain) (%/%)
1 = self-regulating process time constant (sec)
max = maximum total loop dead time (sec)

11/10/2008

10

Definition of Deadband and Stick-Slip


Dead band is 5% - 50%
without a positioner !
Deadband

Pneumatic positioner
requires a negative
signal to close valve

Stroke
(%)

Stick-Slip
0
Deadband

Signal
(%)
Digital positioner
will force valve
shut at 0% signal

The effect of slip is worse than stick, stick is worse than dead band,
and dead band is worse than stroking time (except for surge control)
Stick-slip causes a limit cycle for self-regulating processes. Deadband causes a limit cycle in
level loops and cascade loops with integral (reset) action. If the cycle is small enough it can
get lost in the disturbances, screened out by exception reporting, or attenuated by volumes
11/10/2008

11

Saw Tooth Flow Controller Output Limit Cycle


from Stick-Slip

Controlled Flow (kpph)


Square Wave Oscillation

Controller Output (%)


Saw Tooth Oscillation

11/10/2008

12

Rounded Level Controller Output Limit Cycle


from Deadband

Controlled Level (%)


Saw Tooth Oscillation

Controller Output (%)


Rounded Oscillation

Manipulated Flow (kpph)


Clipped Oscillation

11/10/2008

13

Identification of Stick and Slip in


a Closed Loop Response
59
58.5
58

Controller Output

Flow

57.5

3.25 Percent
Backlash + Stiction

57
56.5
Stroke
%

56

stick

55.5
55
54.5

Dead band is
peak to peak
amplitude for
signal reversal

slip

54
53.5
53
0

100

200

300

400
500
Time ( Seconds )

600

700

800

The limit cycle may not be discernable due to frequent disturbances and noise

11/10/2008

14

Response Time of Various Positioners


(small actuators so slewing rate is not limiting)

Response time increase dramatically for steps less than 1%

11/10/2008

15

Control Valve Facts of Life


)
)
)
)

)
)
)
)

Pneumatic positioners are almost always out of calibration


Most tests by valve manufacturers for stick-slip are at 50% with loosely
tightened stem packing to minimize seating, sealing, and packing friction
Without a representative position feedback in the control room, it is anybodys
guess what the valve is doing unless there is a low noise sensitive flow sensor
Not all positioners are equal. Pneumatic positioners, especially the spool or
single amplification stage low gain ones will increase the valve response time
by an order of magnitude (4 -> 40 sec) for small changes in controller output
All valves look good when checking positions for 0, 25, 75, and 100% signals
Valve specs do not generally require that the control valve actually move
The tighter the shutoff, the greater the stick-slip for positions less than 20%
Smart positioner diagnostics and position read back are lies for actuator shaft
position feedback of rotary type isolation valves posing as throttling valves
particularly for pinned rather than splined shaft connections due to twisting of
the shaft. Field tests show stick-slip of 85 in actual ball or disc movement
despite diagnostics and read back indicating a valve resolution of 0.5%
The official definition of valve rangeability is bogus because it doesnt take into
account stick-slip near the seat. Equal percentage valves with minimal stickslip (excellent resolution and sensitivity) generally offer the best rangeability
11/10/2008

16

Top Ten Signs of a Valve Problem


(10) The pipe fitters are complaining about trying to fit a 1 inch
valve into a 10 inch pipe.
(9) You bought the valve suppliers monthly special.
(8) A butterfly disc wont open because the ID of the lined pipe is
smaller than the OD of the disc.
(7) The maintenance department personally put the valve on your
desk.
(6) A red slide ruler was used to size a green valve.
(5) Your latest valve catalog is dated 1976.
(4) The maintenance department said they dont want a double
seat A body.
(3) The valve was specified to have 0% leakage for all conditions
including all signals.
(2) The fluid field in the sizing program was left as water.
(1) The valve is bigger than the pipe.

Flow Meter Performance

)
)
)
)
)

Type
Coriolis
Magmeter
Vortex
Orifice

Sizes
-8
-78
-12
-78

Range
100:1
25:1
9:1*
4:1

Piping
1/1
5/1
10/5
10/5

Interferences
solids, alignment, vibration
conductivity, electrical noise
profile, viscosity, hydraulics
profile, Reynolds Number

Reproducibility
0.1% of rate
0.5% of rate
1.0% of span
5.0% of span

* - assumes a minimum and maximum velocity of about 1 and 9 fps, respectively

Coriolis flow meters via their accurate density measurement offer


direct concentration measurements for 2 component mixtures and
inferential measurements for complex mixtures.

11/10/2008

18

Neutralizer Control Before

Reagent
Stage 2
Reagent
Stage 1
FC
1-2

FT
1-2

FT
2-1

AC
1-1

FT
1-1

AC
2-1

AT
2-1

AT
1-1
Static Mixer

Feed

2
pipe
diameters
Neutralizer

Discharge

11/10/2008

19

Nonlinearity and Sensitivity of pH


pH

8
6

Good valve resolution or fluid mixing does not look


that much better than poor resolution or mixing due
amplification of X axis (concentration) fluctuations

11/10/2008

Reagent Flow
Influent Flow
or
Reagent Charge
Process Volume

20

Neutralizer Control After


Feedforward
Summer

RSP

FT
1-1

Signal
Characterizer
AC
1-1

FC
1-2

FC
2-1

FT
2-1

Reagent
Stage 2

f(x)

Reagent
Stage 1

*1
*1

FT
1-2

AT
1-1

Static Mixer
Feed
20
pipe
diameters

*1 - Isolation valve closes when control valve closes

AC
2-1
Neutralizer
AT
2-1

Discharge

11/10/2008

21

Distillation Column Control Before


PC
3-1
LT
3-1

LC
3-1

Vent

Feed Tank
Distillate
Receiver
PT
3-1
Overheads

Reflux
FC
3-3

Thermocouple
Tray 10

FT
3-3

TE
3-2

TC
3-2

Column
Feed

FC
3-4

LC
3-2

FT
3-4

LT
3-2

Storage Tank

Steam
Bottoms

11/10/2008

22

Nonlinearity and Sensitivity of Tray Temperature


Operating
Point
Temperature

Measurement Error

Tray 6

Measurement Error

Tray 10

Distillate Flow
Feed Flow
% Impurity
Impurity Errors

11/10/2008

23

Distillation Column Control After


PC
3-1

Feedforward summer
LC
3-1

FT3-3
Feed Tank

LT
3-1

Vent

RSP
FC
3-1
PT
3-1

Distillate
Receiver

RSP

FT
3-1

FT
3-2

Overheads
Feedforward summer

Reflux
FC
3-3

FT
3-3

RTD
Feed

Column
Tray 6

TT
3-2

FT3-3
Signal Characterizer

LT
3-2

FT
3-4
Steam

TC
3-2

f(x)

Storage Tank

LC
3-2

FC
3-4

FC
3-2

RSP
FC
3-5

FT
3-5
Bottoms

11/10/2008

24

When Process Knowledge is Missing in Action


PV distribution for
original control

2-Sigma

LOCAL
Set Point

Upper Limit

2-Sigma

RCAS
Set Point

Extra margin when


war stories or
mythology rules

value
PV distribution for
improved control

Good engineers can draw straight lines


Great engineers can move straight lines
2-Sigma

2-Sigma

Benefits are not realized until the set point is moved!


(may get benefits by better set point based on process
knowledge even if variability has not been reduced)
11/10/2008

25

Top Ten Ways to Impress Your Management


with the Trends of a Control System
(10) Make large set point changes that will zip past valve dead
band and local nonlinearities
(9) Change the set point to operate on the flat part of the titration
curve
(8) Select the tray with minimum process sensitivity for column
temperature control
(7) Pick periods when the unit was down
(6) Decrease the time span so that just a couple data points are
trended
(5) Increase the reporting interval so that just a couple data points
are trended
(4) Use really thick line sizes
(3) Add huge signal filters
(2) Increase the process variable scale span so it is at least ten times
the control region of interest
(1) Increase the historians data compression so that most changes
are screened out as insignificant

Basic Opportunities in Process Control


)

Decrease stick-slip and improve the sensitivity of the final element


(Standard Deviation is the product of stick-slip, valve gain, and process
gain)

Use properly tuned smart positioners, short shafts with tight connections,
and low friction packing and seating surfaces to decrease valve slip-stick and
dead band (do not use isolation valves for throttling valves)
If high friction packing must be used, aggressively tune the smart positioner
Improve valve type and sizing and add signal characterization to increase
valve sensitivity
Use variable speed drives where appropriate for the best sensitivity

Improve the short and long term reproducibility and reduce the
interference and noise in the measurement (Standard Deviation is
proportional to reproducibility and noise)

11/10/2008

Use magnetic and Coriolis mass flow meters to eliminate sensing lines,
improve rangeability, and reduce effect of Reynolds Number and piping
Use smart transmitters to reduce process and ambient effects
Use RTDs and digital transmitters to decrease temperature noise and drift

27

Basic Opportunities in Process Control


)

Reduce loop dead time (Minimum Integrated Error is proportional to


the dead time squared)

)
)
)
)
)
)

Decrease valve dead time (stick and dead band)


Decrease transport (plug flow volume) and mixing delay (turnover time)
Decrease measurement lags (sensor lag, dampening, and filter time)
Decrease discrete device delays (scan or update time)
Decrease analyzer sample transport and cycle time

Tune the controllers (Integrated Error is inversely proportional to the


controller gain and directly proportional to the controller integral time)
Add cascade control (Standard Deviation is proportional to the ratio of the
period of the secondary to the process time constant of the primary loop)
Add feed forward control (Standard Deviation is proportional to the root
mean square of the measurement, feed forward gain, and timing errors)
Eliminate or slow down disturbances (track down source and speed)
Add inline analyzers (probes) and at-line analyzers with automated
sampling since ultimately what you want to control is a composition
Optimize set points (based on process knowledge and variability)

To realize the benefit of reduced variability, often need to change a set point

11/10/2008

28

Reset Gives Them What They Want


TC-101
Reactor Temperature
Out

PV

SP

steam
valve
opens

temperature

Reset wont open the water valve


Until the error changes sign, PV
goes above the set point. Reset
has no sense of direction.
set point (SP)

50%

PV
water
valve
opens

Should the steam or


water valve be open?

44

52

Reset action integrates the numeric


difference between the PV and SP
seen by operator on a loop faceplate
Reset works to open the steam valve

11/10/2008

time

Proportional and rate action see


the trajectory visible in a trend!
Both would work to open the
water valve to prevent overshoot.

29

Reactor and Column Loop Tuning

Most reactor and column composition, gas pressure, and temperature


loops have too much integral action (reset time too small), not enough
proportional action (gain too small), and not enough derivative action
(rate time too small).
Rate time should be 0.1x process time constant or 0.1x reset time with a
minimum value of sensor lag time.
Rate action is essential for exothermic reactors that can runaway

Often these loops are near integrators due to a large process time
constant . Batch processes often have true integrators because of a
lack of self-regulation (no steady state). Whether near integrators or
true integrators, these loops require much more gain action to impose
self-regulation and provide pre-emptive action. There is a window of
allowable gains where too low of a controller gain will result in slow
rolling oscillations from reset.
(controller gain) * (controller reset time) > 4 / (integrating process gain)

11/10/2008

30

Modeling and Control Facts of Life

Timing is Everything
In life, business, and process control (especially feedforward)

Without Dead Time I would be Out of Job


If the dead time was zero, the only limit to how high you can set the controller
gain or how tight you can control is measurement noise
Unlike aerospace, the process industry has large and variable time delays and
time lags from batch cycle times, vessel mixing times, volume residence
times, transportation delays, resolution limits, dead band, and measurements
Total dead time is sum of time delays and all time lags smaller than largest
Best possible integrated absolute error is proportional to dead time squared

11/10/2008

31

Modeling and Control Facts of Life


)

Models (experimental or theoretical) allow you to take the blindfold off


Models convey process knowledge and provide insight on what has changed and
what should be improved (e.g. largest source of dead time)
War stories rule where there are no models
Mythology rules where there are no models
Benefits are hearsay where there are no models

Nonlinearity is a reason to build models rather than avoid models


Unless you want job security for constantly retuning controllers. Also, implied in
most techniques is some model (e.g. reaction curve method)
Tight control greatly reduces the operating point nonlinearity (e.g. pH) and
secondary flow loops eliminate the valve nonlinearity for higher level loops
Signal characterization on the controller output (based on a model of the
installed valve characteristic) greatly reduces the valve nonlinearity

11/10/2008

32

Speed of Various Sources of Disturbances


(Speed Kills)
)

Process

A loop can catch up to a slow


disturbance. Liquid pressure
Is the fastest upset (travels at
the speed of sound in liquid).

Equipment

Flow (fast)
Gas pressure (fast)
Liquid Pressure (very fast)
Raw Materials (slow)
Recycle (very slow)
Temperature (slow)
Catalyst (slow)
Steam (fast)
Coolant (fast)
Fouling (slow)
Failures (fast)

Environmental

11/10/2008

Day to Night (slow)


Rain Storms and fronts (fast)
Season to Season (very slow)
33

Speed of Various Sources of Disturbances


(Speed Kills)
)

Valves

Measurements

Stick-slip (fast)
Split Range (fast)
Failures (very fast)
Noise (very fast)
Reproducibility (fast)
Failures (very fast)

Controllers

Feedback Tuning (fast) *


Feed forward Timing (fast)
Interaction (fast)
Failures (very fast)

* Most frequent culprit is an oscillating level loop primarily due to excessive reset action

11/10/2008

34

Speed of Various Sources of Disturbances


(Speed Kills)
)

Market*

Operators

Rate changes (fast)


Product transitions (fast)
Manual operation (fast)
Sweet spots (fast)
Inventory control (fast)

Discrete

On-off control (very fast)


Sequences (fast)
Batch operations (fast)
Startup and shutdown (very fast)
Interlocks (very fast)

*For minimized inventory, changes in market demand can result in


fast production rate changes and product grade or type transitions

11/10/2008

35

Batch Control
Variability Transfer from Feeds to
pH, and Reactant and Product Concentrations

Feeds

Concentrations
Optimum
Product

Reactant
Product
Reagent

Optimum pH
pH
Reactant
Optimum
Reactant

Most published cases of multivariate statistical process control (MSPC) use the process
variables and this case of variations in process variables induced by sequenced flows.
11/10/2008

36

PID Control
Variability Transfer from pH and Reactant
Concentration to Feeds

Concentrations

Feeds

Optimum
Product
Reactant
Product

Reactant
pH
Reagent
Optimum pH

Optimum
Reactant

The story is now in the controller outputs


(manipulated flows) yet MSPC still focuses
on the process variables for analysis
11/10/2008

37

Model Predictive Control


Variability Transfer from Product Concentration
to pH, reactant Concentration, and Feeds

Feeds

Concentrations
Reactant
Optimum
Product
Product

pH

Reagent

Optimum pH

Reactant
Optimum
Reactant

Time

Time

Model Predictive Control of product concentration batch profile uses slope for CV which makes
the integrating response self-regulating and enables negative besides positive corrections in CV
11/10/2008

38

Example of Basic PID Control


TC-3

CTW

vent
PC-1
condenser
FC-1

feed A
RC-1
ratio
control

CAS

feed B

TT
FT
PT
FC-2
TT

FT

TC-1 cascade control


CAS

coolant
makeup

TC-2

TT

Conventional Control

reactor
product

11/10/2008

39

Example of Advanced Regulatory Control


override control
maximum
production
rate

ZC-3

<

TC-3

CTW

CAS

ZC-4
vent

ZC-2

PC-1
CAS

<

feed A
RC-1
ratio

CAS

feed B

condenser
FC-1
TT
FT
PT
FC-2
CAS

TT

FT

ZC-1

CAS

coolant
makeup

Override Control
ZC-1, ZC-3, and ZC-4 work to keep their respective
control valves at a max throttle position with good
sensitivity and room for loop to maneuver. ZC-2
will raise TC-1 SP if FC-1 feed rate is maxed out
11/10/2008

TC-1

TC-2

TT
reactor
product

40

Function Blocks for Online Data Analytics

Function blocks
developed to support
on-line batch and
continuous analytics
PCA Block
PLS Block
Analyzer Block

11/10/2008

41

Analyzer Block for Online Data Analytics


History Collection of Lab and
Spectral Analyzer Data
Off-line Modeling

Operator
Station

Historian

Processing of Sample Data for


Use in Analytics
Lab Results

Module
Analyzer
Block

Other
Data
Controller

11/10/2008

42

Dynamic Time Warping for Online


Batch Data Analytics

Reference trajectory
Trajectory to be synchronized
Synchronized trajectory
11/10/2008

43

Virtual Plant Setup


Virtual Plant
Laptop or Desktop
or Control System Station

Advanced Control Modules

This is where I hang out

11/10/2008

Process Models
(first principal
and experimental)
44

Virtual Plant Integration


DCS batch and loop
configuration, displays,
and historian
Embedded
Modeling Tools

Dynamic
Process Model

Embedded
Advanced Control Tools

Virtual Plant
Laptop or Desktop
Personal Computer
Or
DCS Application
Station or Controller

Online
Data Analytics

Loop Monitoring
And Tuning

Model Predictive
Control

Process Knowledge
11/10/2008

45

Typical Uses and Fidelities of Process Models


(Fidelity Scale 0 - 10)
)

Process Development
Media or reactant optimization and identification of kinetics on the bench top - 10
Optimization of process conditions in pilot plant - 9
Agitation and mass transfer rates - 8*
Process scale-up 8
* - assumes computational fluid dynamics (CFD) program provides necessary inputs

Process Design
Innovative reactor designs or single use bioreactors (SUB) - 7
Vessel, feed, and jacket system size and performance - 6

Automation Design
Real Time Optimization (RTO) - 7
Model Predictive Control (MPC) - 6
Controller tuning (PID) - 5
Control strategy development and prototyping - 4
Batch sequence (e.g. timing of feed schedules and set point shifts) 3
11/10/2008

46

Typical Uses and Fidelities of Process Models


(Fidelity Scale 0 - 10)
)

Online Diagnostics
Root cause analysis - 5
Data analytics development and prototyping - 4

Operator Training Systems


Developing and maintaining troubleshooting skills - 4
Understanding process relationships - 3
Gaining familiarity with interface and functionality of automation system - 2

Configuration Checkout
Verifying configuration meets functional specification - 2
Verifying configuration has no incorrect or missing I/O, loops, or devices - 1

11/10/2008

47

What Do We Need?
)

Loops that are not islands of automation


Unit operation control for integrated objectives, performance, and diagnostics
High speed local control of pressure with ROUT, CAS, and RCAS signals

)
)

Engineer with process, configuration, control, measurement, and valve skills


Virtual plants with increasing Fidelity (3 -> 7 chemical, 3->10 biological)
Product development, process design, real time optimization, advanced control
prototyping and justification, process control improvement, diagnostics, training

Smart wireless integrated process and operations graphics


Online process, loop, and advanced control metrics for plants, trains, and shifts

Yield, on-stream time, production rate, utility cost, raw material cost, maintenance cost*
Variability, average % of max speed (Lambda), % time process variable or output is at
limits, % time in highest mode, % deadband, % resolution, number of oscillations
Process control improvement (PCI) benefits ($ of revenue and costs)

3-D, XY, future trajectories of process and performance metrics response, data
analytics, worm plots, and trends of automatically selected correlated variables
)

Coriolis flow meters, RTDs, and online and at-line analyzers everywhere
Real time analysis via probes or automated low maintenance sample systems
Automated time stamped entry of lab results into data historian
Online material, energy, and component balances

Control valves with < 0.25% resolution and < 0.5% dead band

11/10/2008

48

Key Points

)
)
)
)
)
)
)
)
)
)
)
)

Tune the loops


Use digital positioners and throttle valves to get resolution better than 0.5%
Use Coriolis and Magmeters to get accuracy better than 0.5% of rate
Tune the loops
Add cascade and feed forward control for disturbances
Model the process to dispel myths and build on process knowledge
Improve the set points
Add composition control
Reduce the size and speed of disturbances
Transfer variability from most important process outputs
Add online data analytics (multivariate statistical process control)
Add online metrics to spur competition, and to adjust, verify, and justify controls

11/10/2008

49

Control Magazine Columns and Articles

)
)
)
)
)
)
)
)
)

Control Talk column 2002-2008


Has Your Control Valve Responded Lately? 2003
Advanced Control Smorgasbord 2004
Fed-Batch Reactor Temperature Control 2005
A Fine Time to Break Away from Old Valve Problems 2005
Virtual Plant Reality 2005
Full Throttle Batch and Startup Responses 2006
Virtual Control of Real pH 2007
Unlocking the Secret Profiles of Batch Reactors 2008

11/10/2008

50