100% found this document useful (3 votes)

526 views452 pages

Advanced Control Unleashed

Uploaded by

dancord19

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

100% found this document useful (3 votes)

526 views452 pages

Advanced Control Unleashed

Uploaded by

dancord19

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

Advanced Control Unleashed

Plant Performance Management

for Optimum Benefit

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Terrence L. Blevins
Gregory K. McMillan
Willy K. Wojsznis
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Michael W. Brown

Copyright International Society of Automation

The information presented in this publication is for the general education of the reader. Because
neither the author nor the publisher have any control over the use of the information by the reader,
both the author and the publisher disclaim any and all liability of any kind arising out of such use.
The reader is expected to exercise sound professional judgment in using any of the information
presented in a particular application.

Additionally, neither the author nor the publisher have investigated or considered the affect of any
patents on the ability of the reader to use any of the information in a particular application. The
reader is responsible for reviewing any possible patents that may affect any particular use of the
information presented.

Any references to commercial products in the work are cited as examples only. Neither the author
nor the publisher endorse any referenced commercial product. Any trademarks or tradenames
referenced belong to the respective owner of the mark or name. Neither the author nor the publisher
make any representation regarding the availability of any referenced commercial product at any
time. The manufacturer’s instructions on use of any commercial product must be followed at all
times, even if in conflict with the information in this publication.

Copyright © 2003 ISA – The Instrumentation, Systems, and Automation Society

All rights reserved.

Printed in the United States of America.

10 9 8 7 6 5 4 3 2

ISBN 1-55617-815-8

No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or
by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior writ-
ten permission of the publisher.

ISA
67 Alexander Drive
P.O. Box 12277
Research Triangle Park, NC 27709

Library of Congress Cataloging-in-Publication Data is in process.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Provided by IHS under license with ISA Licensee=PDVSA - Complejo Petroquimico Site 12/9986712018, User=Cordova, Daniel
No reproduction or networking permitted without license from IHS Not for Resale, 10/11/2016 [Link] MDT
Dedication
This book is dedicated to Karen Blevins, Cathy MacDonell Brown,
Carol McMillan, and Susan Wojsznis who have provided encouragement
and support throughout our careers.
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

ACKNOWLEDGEMENT xiii

ABOUT THE AUTHORS xv

FOREWORD xvii

Chapter 1 INTRODUCTION 1

Chapter 2 SETTING THE FOUNDATION 5

Practice, 5
Overview 5
Opportunity Assessment 12
Examples 15
Application, 20

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
General Procedure 20
Application Detail 26
Rules of Thumb 74
Theory, 76
Process Time Constants and Gains 76
Process Time Delay 79
Ultimate Gain and Period 80
Peak and Integrated Error 82
Feedforward Control 84
Dead Time from Valve Dead Band 84
Nomenclature, 85
References, 86

vii

Copyright International Society of Automation

Chapter 3 APC PATHWAYS 89

Practice, 89
Overview 89
Opportunity Assessment 94
Examples 103
Application, 106
General Procedure 106
Application Detail 108
Rules of Thumb 115
References, 116

Chapter 4 EVALUATING SYSTEM PERFORMANCE 119

Practice, 119
Overview 119
Opportunity Assessment 121
Examples 125
Application, 129
General Procedure 129
Application Details 131
Rules of Thumb 143
Guided Tour 144
Theory, 147
Using Statistics for Control Performance Evaluation 150
Extending the Concept to the Multi-variable Environment 153
Addressing Advanced Control 154
Diagnostic Tools 156
References, 160

Chapter 5 ABNORMAL SITUATION MANAGEMENT 163

Practice, 163
Overview 163
Opportunity Assessment 165
Examples 166
Application, 168
General Procedure 168
Application Details 169
Rules of Thumb 171
Guided Tour 173
Theory, 177
Introduction to Expert Systems 177
Rules 178
Inference Engine 180
Facts 181
References, 182

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Chapter 6 AUTOMATED TUNING 183

Practice, 183
Overview 183
Opportunity Assessment 185
Examples 187
Application, 197
General Procedure 197
Application Detail 200
Rules of Thumb 202
Guided Tour 206
Theory, 208
Introduction to Auto Tuners 208
Basics of Relay-Oscillation Tuning 210
Model Based Tuning 218
Robustness Based Tuning 221
Adaptive Control 225
References, 237

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Chapter 7 FUZZY LOGIC CONTROL 239
Practice, 239
Overview 239
Opportunity Assessment 240
Examples 240
Application, 241
General Procedure 241
Rules of Thumb 242
Guided Tour 242
Theory, 244
Introduction to Fuzzy Logic Control 244
Building a Fuzzy Logic Controller 247
Fuzzy Logic PID Controller 251
Fuzzy Logic Control Nonlinear PI Relationship 254
FPID and PID Relations 257
Automation of Fuzzy Logic Controller Commissioning 258
References, 259

Chapter 8 PROPERTIES ESTIMATION 261

Practice, 261
Overview 261
Opportunity Assessment 263
Example — Dynamic Linear Estimator 265
Examples – Neural Networks 269
Application, 274
General Procedure 274
Application Detail 279

Copyright International Society of Automation

Rules of Thumb 289

Guided Tour 289
Theory, 294
Dynamic Linear Estimator 294
Neural Networks 296
References, 305

Chapter 9 MODEL PREDICTIVE CONTROL 307

Practice, 307
Overview 307
Opportunity Assessment 310
Examples 316
Application, 337
General Procedure 337
Application Detail 339
Rules of Thumb 353
Guided Tour 355
Theory, 362
The Basics of Process Modeling 364
Identifying the Process Model 367
Unconstrained Model Predictive Control 369
Integrating Constraints Handling, Optimization and Model Predictive Control
373
References, 381

Chapter 10 VIRTUAL PLANT 383

Practice, 383
Overview 383
Opportunity Assessment 386
Examples 387
Application, 389
General Procedure 389
Online Adaptation 393
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Application Detail 395

Rules of Thumb 399
Guided Tour 400
Theory, 403
References, 408

Appendix A ADDITIONAL OPPORTUNITY ASSESSMENT QUESTIONS 409

Appendix B BATCH-TO-CONTINUOUS TRANSITION 415

Appendix C DEFINITIONS 419

Copyright International Society of Automation

Appendix D TOP 20 MISTAKES 425

INDEX 431

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

The authors wish to express their appreciation to Mark Nixon and Ron
Eddie from Emerson Process Management, for their enthusiastic support
and commitment of resources for this book, to Jim Hoffmaster, Bud Keyes,
Duncan Schleiss, John Berra, and Gil Pareja from Emerson Process Man-
agement for their inspiration and support in establishing the DeltaV
advanced control program, to Karl Astrom from Lund University, Tom
Edgar from the University of Texas at Austin, Dale Seborg from the
University of California, Santa Barbara and Tom McAvoy from the Univer-
sity of Maryland for their guidance in the pursuit of new technologies,
Mike Gray and Mark Mennen from Solutia Inc. for the initiation and
sustenance of advanced control applications and innovations, Ken
Schibler from Emerson Process Management for his help in setting the
direction of the book, Robert Cameron, Michael Mansy, Glenn Mertz, and
Gina Underwood from Solutia Inc. for their valuable comments, and
finally, Scott Weidemann from Washington University, and Jim Cahill,
Brenda Forsythe, and Cory Walton from Emerson Process Management
for their essential contributions to the videos and demos on the CD. The
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

authors extend their thanks to the developers of the advanced control

tools that were the inspiration for this book. This includes Vasiliki Tzovla,
Ron Ottenbacher, Dirk Thiele, Ashish Mehta, Yan Zhang, Peter Wojsznis,
John Gudaz, Ian Nadas and Mei Yang. Also, we would like to recognize
the valuable contribution of Tom Aneweer, Dennis Stevenson, Jay Colcla-
zier, Darrin Kuchle, Dick Seemann, Joe S. Qin, Steve Morrison, Mike Ott,
and Sai Ganesamoorthi.

The discipline of Process Control and Advanced Process Control is an

exciting, challenging and rewarding field of engineering. Some of us
moved into this discipline by chance, while others made a conscious deci-
sion to become Process Control Engineers. Regardless of our entry point or
xiii

Copyright International Society of Automation

Provided by IHS under license with ISA Licensee=PDVSA - Complejo Petroquimico Site 12/9986712018, User=Cordova, Daniel
No reproduction or networking permitted without license from IHS Not for Resale, 10/11/2016 [Link] MDT
motivation, we all appreciate the fact that engineers before us took the
time and effort to teach us the tools and techniques that allowed us to
achieve success. This book passes along the knowledge of many years and
many people and acknowledges the efforts of our engineering mentors.
We hope it will allow others not only to benefit from all the experience we
have benefited from so greatly but also to take the technology to the next
level.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Terrence L. Blevins has been actively involved in the application and

design of process control systems throughout his career. For over fifteen
years, he worked as a systems engineer and group manager in the design
and startup of advanced control solutions for the pulp and paper industry.
Terry was instrumental in the establishment of Emerson Process Manage-
ment’s Advanced Control Program. He is the Fieldbus Foundation team
leader for the Function Block Specification. In this capacity, Terry is
involved in the movement of Fieldbus Foundation function block work
into international standards. Terry is the US expert to the IEC TC65 WG6
and SC65C WG7 function block committees. He wrote the fieldbus section
included in the Process/Industrial Instrumentation and Controls handbook.
Terry has eight patents and has written over forty papers on process con-
trol system design and applications. He received a BEE from the Univer-
sity of Louisville in 1971 and a Master of Science degree in Electrical
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Engineering from Purdue University in 1973. Presently, Terry is a principal

technologist in DeltaV Product Engineering and the team leader for
DeltaV advanced control product development.
Phone: (512) 418-4628
E-mail: [Link]@[Link]

Gregory K. McMillan retired as a Senior Fellow after a 33 year career with

Monsanto and Solutia Inc, where he specialized in improving loop perfor-
mance, controller tuning, valve dynamics, opportunity assessments,
dynamic simulation, fermentor control, pH control, and reactor control.
Greg is the author of numerous articles and books, his most recent book
being: Good Tuning – A Pocket Guide. He has contributed to several hand-
books and is the editor of the Process/Industrial Instrumentation and Controls
handbook and the coauthor of a monthly column titled “Control Talk”.
xv

Copyright International Society of Automation

Greg is an ISA Fellow and received the ISA “Kermit Fischer Environmen-
tal” Award for pH control in 1991, the Control Magazine “Engineer of the
Year” Award for the Process Industry in 1994, and was one of the first
inductees into the Control Magazine “Process Automation Hall of Fame”
in 2001. He received a B.S. from Kansas University in 1969 in Engineering
Physics and a M.S. from University of Missouri – Rolla in 1976 in Control
Theory. Presently, Greg is an affiliate Professor at Washington University
in Saint Louis, Missouri and is a consultant through EDP Contract Services
in Austin, Texas.
Cell Phone: (314) 703-9981
E-mail: gkmcmi@[Link]

Willy K. Wojsznis has been involved in developing advanced control

products over the last twelve years focusing on model predictive control
and auto tuning. Over the previous nearly 25-year of his career he was
developing computer control systems and applications in cement, steel,
mining and paper industries. His professional work resulted in a number
of successful and innovative advanced control products, fourteen patents,
and above thirty technical papers. He received control engineering degree
(EE) from Kiev Technical University in1964, M.S. in Applied Mathematics
from Wroclaw University in 1972, and Ph.D. from Warsaw University of
Technology in 1973. Presently, Willy is a part of DeltaV advanced control
group. He conducts applied research in the areas of optimization, adaptive
control and model predictive control.
Phone: (512) 418-7475
E-mail: [Link]@[Link]

Michael W. Brown has spent his entire career in the application of

Advanced Process Control technologies in the continuous processing
industries. Over the previous 15 years of his career, he has served as an
Advanced Process Control Consultant, providing implementation exper-
tise and technology guidance for many operating companies. His exten-
sive knowledge and experience in the areas of model based predictive
control and real-time optimization have assisted many companies in cap-
turing the benefits of these performance improvement technologies. His
work has resulted in several technical papers, published in various control
journals. Michael is a Chemical Engineer, received a [Link]. from the Uni-
versity of Waterloo in 1987 and completed his Masters in Applied Science
in 1989. Presently, Michael is the Applications Business Manager, with
Matrikon Inc., where he continues to work with industry to bring
Advanced Control Technology to the next level.
Phone: (905) 282-9248
E-mail: [Link]@[Link]

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

There has been a dynamic development of control over the past 50 years.
Many new methods have appeared. The methods have traditionally been
presented in highly specialized books written for researchers or engineers
with advanced degrees in control theory. These books have been very use-

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
ful to advance the sate of the art. They are however difficult for an average
engineer. The reasons are that it is necessary to read many books to get a
good coverage of advanced control techniques and that the level of mathe-
matics used requires a substantial preparation. This is a dilemma because
several of the advanced control techniques have indeed been very benefi-
cial in industrial and more engineers should be aware of them. Even if
many details of the new methods are complicated the basic underlying
ideas are often quite simple. Many methods have also been packaged so
that they are relatively easy to use. It is thus highly desirable to present the
industrially proven control methods to ordinary engineers working in
industry. This book is a first attempt to do this. The book provides a basis
for assessing the benefits of advanced control. It covers auto-tuning,
model predictive control, optimization, estimators, neural networks, fuzzy
control, simulators, expert systems, diagnostics, and performance assess-
ment. The book is written by four seasoned practitioners of control, hav-
ing jointly more than 100 years of real industrial experience in the
development and use of advanced control. The book is well positioned to
provide the bridge over the infamous Gap between Theory and Practice in
control.

Karl J. Astrom

xvii

Copyright International Society of Automation

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
The advent of powerful and friendly integrated software has moved
advanced process control (APC) from the realm of consultants into the
arena of the average process control engineer. The obstacles of infrastruc-
ture and special skill requirements have started to disappear and we are
poised for an accelerated application of APC.

It is well known that APC seeks to discover, incorporate, and exploit

knowledge about raw materials, process, product, equipment, instrumen-
tation, and final elements. What is not often recognized is the significant
increase in the knowledge base of both plant and field operations that
occurs as the APC system is developed. In fact, an appreciable portion of
the benefits are achieved by improvements made in operating procedures,
set points, sensors, and control valves as a result of the methodical analy-
sis, testing, modeling, and prototyping that are part of the best practices
used in the implementation of APC systems.

Until recently, most of this knowledge ended up with consultants, and the
success of the application often deteriorated once they departed. There is
now an opportunity for the engineers closest to the process and daily
operations to take a much more active role in the development and sup-
port of APC applications. It is a win-win situation in that the cost of APC
can be reduced by using consultants primarily in a higher-value-added
role of conceptual design and optimization. Even more importantly,
greater understanding, support, and involvement of onsite engineers can
increase the success rate, the on-stream time, and the longevity of an APC
application. This decrease in the cost and increase in the benefits will in
turn lead to a larger number of successful APC installations and a greater
interest in APC as a method of improving process efficiency and capacity.

Copyright International Society of Automation

However, much of the purpose and use of APC has been clouded in the-
ory. The theory is scattered among many books written for graduate
school programs in advanced process control. Application papers typi-
cally concentrate on the benefits of specific APC projects and serve more
as advertisements for particular consulting or software firms than as
implementation guides. Little if anything has been written for the practic-
ing engineer on how to select, design, configure, commission, and tune
APC systems. The purpose of this book is to demystify APC and make it
more accessible. To that end, the book focuses on practice and applications
backed up by enough theory to insure a deeper understanding.

Each chapter is organized to provide concise practical information that a

user can readily explore and reference to start and complete a successful
implementation. Each chapter has three major sections, entitled Practice,
Application, and Theory.

The PRACTICE section starts with an Overview that provides a concise

explanation of the technology and its importance. It provides the motiva-
tion and basis for investing more time in learning and pursuing the tech-
nology. Next is an Opportunity Assessment subsection that offers a simple
approach to determine whether the technology is applicable to a particular
unit operation. It consists of a set of concepts and questions to start the
thought process and discussions to find potential applications. The Exam-
ples subsection rounds out the section. The sample of applications pre-
sented here help to instill a better practical understanding of the use of the
technology in the process industry.

The APPLICATION section starts with a General Procedure subsection that

presents a “go-to” checklist to introduce the user to the normal sequence
of events for a successful application. This list provides a good reference to
make sure all bases are covered and is useful for planning, scheduling,
estimating, and monitoring APC projects. Next there is an Application
Detail subsection that summarizes most of what a user needs to know. A
building-block approach is used wherever possible, starting with a basic
“bare-bones” application and adding successively more capabilities to end
up with a full complement of advanced features. Next is a series of Rules of
Thumb concisely phrased to be readily referenced and remembered. A
brief explanation with any notable exceptions follows each rule. The sec-
tion ends with a Guided Tour to give the reader a feel for how advances in
software have made implementation easy enough that the user can focus
on the opportunity offered by these APC tools to discover, incorporate,
and exploit plant knowledge.

The THEORY section presents the major facets of selected approaches to the
deployment of each APC technology as part of a state-of-the-art tool set.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

For brevity, the section does not survey all the possible methodologies and
techniques, but focuses on those that are innovative and simple enough to
be integrated into a distributed control system.

This book covers a great deal of ground. Each of the technologies dis-
cussed here could easily fill a book in itself. However, users today don’t
have the time or inclination to read a lot of material. Lists, hints, rules of
thumb, and concise explanations are employed to save the reader time and
to provide both a better perspective on the whole picture and an improved
ability to drill down to obtain specific implementation guidance. The book
concentrates on what is most important. Users can quickly get to the heart
of the matter without getting lost in the details associated with a specific
tool or suffering from information overload.

While a user can go directly to a given chapter to learn about a particular

technology, the authors recommend that Chapters 2 and 3 be read first.
They provide the necessary foundation on which to build an APC applica-
tion and the logic to select the most appropriate technology.

Included with the book is a compact disc that contains a set of examples of
the technologies discussed in the book. They demonstrate, by means of a
step-by-step procedure and a detailed dynamic process model, how to
configure, test, and run each APC application. Configuration and case files
use a virtual plant that has a complete scalable Distributed Control System
(DCS) with a suite of APC tools and a high-fidelity plant simulation.

A companion set of Power Point slides that illustrates all of the major Fig-
ures, equations, tables, lists and rules included in the book is on the CD.

These slides and the hands-on exercises make the book practical as a text-
book for courses on both basic and advanced process control. Chapters 2
and 6 receive the most extensive treatment because introductory courses
are most common. Also, students and users alike need to first concentrate
on getting the basic regulatory control system designed correctly and
tuned properly before moving on to more advanced topics. Most of the
material has been tested in an introductory course on process control for
junior and senior chemical engineers at Washington University in Saint
Louis. These students have demonstrated the ability to immediately apply
these APC tools to example problems after a brief tutorial, using their
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

computer skills and a powerful integrated Windows® environment. The

sound mathematical foundation of APC makes it easier to learn than basic
control, which is more heuristic.

Copyright International Society of Automation

The tutorials and presentations on the CD do not require any special soft-
ware or hardware beyond a PC with a media player, speakers, and a dis-
play with a screen area of at least 1024 by 768 pixels.

This book with its appendices and CDs should enable the average process
engineer to develop a good understanding of the representative principles
and techniques of APC. This knowledge will be helpful in setting objec-
tives, evaluating potential APC opportunities, and applying the most
appropriate APC technologies. Readers should feel free to contact the
authors at their e-mail addresses if they have any questions about the use
of the book, exercises, demos, slides, or APC tools described.

All royalties from this book will be given directly to universities, consortia,
and educational programs to promote and enhance the development and
use of advanced process control. A beneficiary of each year’s royalties will
be chosen by the authors.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Practice

Overview
The advanced control projects with the largest benefits usually have made
significant improvements in the basic regulatory control system. While
advanced process control (APC) techniques can partially compensate for
such limitations as missing measurements, excessive dead time, and poor
signal-to-noise ratios, a solid foundation will provide the lowest total cost,
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

greatest total benefit, and the longest lifecycle for the advanced control
system. Deficiencies in the measurement and the final element can
increase the time required for process testing and identification by a factor
of 5 or more and can significantly reduce the improvement in process
capacity and efficiency provided by APC.

The core of a solid foundation for advanced process control is good mea-
surements and final elements. The measurement is the window into the
process and must be able to provide an undistorted view of small changes
in the process. The final element is the means of affecting the process and
must be able to make small changes to the process. This overview pro-
vides a perspective of how these objectives are best met by reducing the
reproducibility error, noise, and interferences in the measurement and
decreasing the stick-slip and dead band in the control valve.

Measurement
Reproducibility is the closeness of agreement of an output for an input
approaching from either direction at the same operating conditions over a
period of time. Repeatability is the closeness of agreement of an output for

Copyright International Society of Automation

successive inputs approaching from the same direction at the same operat-
ing conditions. Reproducibility includes the repeatability as it deteriorates
over time plus drift, and is the better number for control. Another impor-
tant consideration is the interference from changes in process fluids and
operating conditions. Unfortunately, the specifications given by manufac-
turers for such measurements as accuracy, linearity, or rangeability are
extraneous if not misleading because they are either not as important as
reproducibility, drift, and interference or are generated under fixed labora-
tory conditions.

If the measurement is noisy or not reproducible, the controlled variable

will change when there is no change. If the change in the process variable
is less than the resolution limit of the sensor or its digital representation or
conversion, there is no change. Consequently, it is important to make sure
the measurement reproducibility error, resolution, and noise band total
less than 1/5 of the allowable control band, or 1/5 of the permissible vari-
ability about the set point. Measurement resolution without any sensor
limitations is 0.05% of span for a twelve bit digital conversion, so it is nor-
mally not an issue except when large spans associated with thermocouple
and resistance temperature detector (RTD) input cards are used in loops
that need to control the temperature within 0.5° Centigrade of set point. It
is important to make sure that measurement time delay and time con-
stants do not excessively slow down or attenuate the view of the actual
changes in the process variable. Of lesser concern is an offset and nonlin-
earity in a measurement, since these can be compensated for by a shift in
the set point or a simple zero adjustment in the transmitter and signal
characterization, respectively.

Changes in process or ambient conditions show up as drift or slow noise.

For example, changes in a fluid’s density will change the liquid head and
hence the level reading from a differential pressure measurement; changes
in a fluid’s kinematic viscosity will change the meter coefficient and hence
the flow reading from a vortex meter; and changes in the activity coeffi-
cient of the hydrogen ion will change the glass electrode potential and
hence the pH reading. In this text, drift will be considered to be in effect a
long-term reproducibility error.
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Final Element
The most common final element is the control valve. Controller outputs
also manipulate the speed of pumps and power to heaters. With final ele-
ments that are totally electronically set, there are no issues of stick and slip
as there are for control valves, and any dead band that exists is purposely
introduced and adjustable to reduce the response to noise. Also, the
response of the manipulated variable (flow for the pump and heat for the
heater) is linear with controller output. Variable-speed drives have essen-

Copyright International Society of Automation

tially no time delay or time constant and rate limiting is normally adjust-
able and not an issue except for surge control. Heaters are inherently slow,
but most temperature processes are also slow.

Usually, a control valve will not move on its own or when the controller
output is constant unless the actuator is undersized or the positioner is
unstable. Also, if the valve were to drift, the positioner and process con-
troller would correct for it. Thus, long-term reproducibility and noise are
not normally issues for control valves. While noise is not generated in the
valve stroke, noise in the process variable can be passed on as rapid
changes in the valve signal, which, if they exceed the resolution limit or
dead band of the control valve, can cause excessive wear and tear and pre-
mature failure of the packing.

If a control valve sticks, there is no change in flow until the accumulated

changes exceed the resolution of the valve. If the valve slips, the change in
flow is more discrete than continuous and is much larger than demanded
by the corresponding change in controller output [2.2] [2.3]. For dead band
or backlash, there is no change in flow when the controller reverses direc-
tion until the change exceeds the dead band as shown in Figure 2-1.

Stiction, a combination of the words “stick” and “friction,” adds time

delay to a loop for any change in the controller output; dead band adds
time delay only for reversals in direction; and slip causes the process vari-
able to pass right by the set point. The dead time from stiction and back-
lash will increase the error from load upsets due to the additional time
delay but doesn’t slow down the set point response as much if the changes
in set point are rather large. Studies that introduce white noise or set point
changes instead of load upsets will report a negligible increase in inte-
grated error from dead band [2.2]. Other studies that concentrate on the
effect of the additional time delay from dead band on load upsets show a
50% increase in the peak error and a 100% increase in the integrated error
from 10% dead band [2.4]. Oscillations in loops with dead band tend to
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

dissipate unless there are interactions or aggressive tuning. Slip, on the

other hand, will cause a limit cycle if there is any reset action, because the
process variable will not come to rest exactly at set point and when reset
works to eliminate the offset the eventual change in flow is too large. The
stroking time of the control valve is not as big an issue except for very fast
loops and very large actuators.

Thus, in the normal scheme of things, slip is worse than stick, and stick is
worse than dead band, and dead band is worse than stroking time. For
sliding stem valves, stick-slip and dead band go hand in hand since the
common cause is excessive packing friction. In fact, if the slip is equal to
the stick, it is effectively the same thing as the resolution limit. The resolu-

Copyright International Society of Automation

tion of sliding stem valves can be estimated as half of the dead band [2.4].
In other words, where you have excessive dead band, you tend also to
have excessive stick-slip. However, in rotary valves, there are different
sources of stick-slip and dead band. A rotary valve could have a large
dead band but little stick-slip.

Dead band is 5% - 50%

without a positioner !
Dead band

Pneumatic positioner
requires a negative Stroke
signal to close valve (%) Digital positioner
will force valve
shut at 0% signal
Stick-Slip

0 Signal
dead band (%)

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)

Figure 2-1. Definition of Dead Band and Stick-Slip

Until recent years, when you asked a control valve manufacturer to esti-
mate the dynamic response of a control valve, you were given the stroking
time of the actuator. Even now, if you ask for a response time that includes
the valve, it will be for a change of 10% in controller output at 50% posi-
tion so that the effect of stick-slip and dead band are largely removed [2.5].
In actual operation, the change in controller output per scan is typically
less than 0.5% and can occur at positions less than 20% where the friction
of the sealing surfaces increases the stick-slip. Tests done at these condi-
tions will unearth the real response problems. In valves, stick and slip go
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

together and can be identified while the loop is operating for fast measure-
ments, as shown in Figure 2-2. Here, stick is the amount of change in the
controller output where there is no change in the process variable and slip
is the rapid change in the process variable divided by the product of the
valve and process gain.

The effect of nonlinearity in the control valve is often misunderstood. In

order to analyze the effect on the control loop, you must look at the prod-
uct of the valve gain and the process gain. Since the process gain is
inversely proportional to flow for temperature and composition control
[2.5], the equal percentage characteristic, where the valve gain is propor-
tional to flow, helps to compensate for process-gain nonlinearity. For con-
trol loops where the process gain is not inversely proportional to flow or
where the installed characteristic deviates from the ideal equal percentage

Copyright International Society of Automation

59
58.5
3.25 Percent
58 Controller Output
Ball Travel Backlash + Stiction
57.5
57
56.5
Stroke 56 Dead band is
% slip peak to peak
55.5 amplitude
55
stick
54.5
54
53.5
53
0 100 200 300 400 500 600 700 800
Time ( Seconds )

Figure 2-2. Identification of Stick and Slip in a Closed-loop Response

characteristic, the nonlinearity of the valve characteristic becomes impor-

tant and there is a range of permissible valve gain, as shown in Figures 2-
3a through 2-3c. Also, the operating-point nonlinearity from the installed
valve characteristic has a greater detrimental effect than the operating-
point nonlinearity associated with a process variable, because the operat-
ing point of the valve changes with load per the demands of the controller
whereas the operating point of the process variable is driven back to the
set point. Correspondingly, process nonlinearity in secondary loops in a
cascade or advanced control system becomes more important because the
set point is driven to new operating points to meet the demands of the
master, supervisory, or model predictive controller. The secondary loop
should be 5 times faster than the primary loop so that the primary loop
isn’t affected by the nonlinearity inside the secondary loop.

Signal characterization of the controller output is a relatively quick fix for

unwanted valve nonlinearity because it is easy to implement in Fieldbus
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

functional blocks. What is not so easy to do is to calculate the actual shape

of the installed characteristic, unless the changes in the inlet and outlet
pressure to the control valve are measured. Also, the operator must be
made aware of the fact that the controller output now represents percent
desired flow instead of percent valve position. For maintenance and trou-
bleshooting, the characterizer output that represents the calculated valve
position target should be displayed along with any feedback of actual
valve position. In addition, signal characterization will increase and
decrease the effect of dead band or stiction for operating points on the
steep and flat portions of the installed characteristic, respectively. [2.5]
Thus, a greater improvement in loop performance is realized on the flat
tail of the installed characteristic curve associated with rotary valves in
Figures 2-3a and 2-3b. However, the user must remember that these curves
can become so flat above 60 degrees opening that there is essentially no

Copyright International Society of Automation

change in flow for a change in rotation and the valve gain approaches
zero.

To summarize, the numbers that traditionally have been cited by the man-
ufacturer for valve performance, such as leakage, stroking time, linearity,
and rangeability, do not provide the information needed to measure con-
trol loop performance. The user needs to know the stick-slip, dead band,
and sensitivity of the installed valve assembly at operating conditions.

Valve Travel (degrees) Too flat above 45 deg!

0 10 20 30 40 50 60 70 80 90
1000

800 Installed flow

characteristic
Flow Many older plants
600
(gpm) push valve to here
400
Linear valve Experimental Data
200 Flow Model
0
Can use signal characterization for flat part of curve
3

Gain too high Gain Model

2
Gain Linear valve
(%Flow/%Input) EnTech Gain
1
Gain is slope Gain too low Specification
of installed flow
characteristic
0
0 10 20 30 40 50 60 70 80 90

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Valve Travel (degrees)
Suggested throttle range is 25 to 45 degrees

Figure 2-3a. Installed Characteristic of a Butterfly Valve

Valve Travel (degrees)

0 10 20 30 40 50 60 70 80 90
1000
800
Flow
(gpm) 600
400 Experimental Data
Flow Model
200
0

3
Gain Model
2
Gain
(%Flow/%Input) EnTech Gain
1 Specification

0
0 10 20 30 40 50 60 70 80 90
Valve Travel (degrees)
Suggested throttle range is 10 to 60 degrees

Figure 2-3b. Installed Characteristic of a Ball Valve

Copyright International Society of Automation

Valve Travel (%)

0 10 20 30 40 50 60 70 80 90 100
1000
800
Flow
(gpm) 600
400
Experimental Data
200 Flow Model

3
Gain Model
2
Gain
(%Flow/%Input) EnTech Gain
1 Specification

0
0 10 20 30 40 50 60 70 80 90 100
Valve Travel (%)
Suggested throttle range is 5 to 75 %

Figure 2-3c. Installed Characteristic of a Sliding Stem Valve

Effect on APC
Advanced control tools such as feedforward control, online estimators,
and model predictive controllers can minimize to a significant degree the
effect of measurement deficiencies. Feedforward control can bypass the
irregularities and delay in the controlled variable but still must work
through the manipulated variable. Since the exact size of stick and dead
band is extremely variable, undercorrection is normal and the overall
improvement is minimal. Filters can reduce the effect of noise; and model
predictive control can reduce the adverse effect of noise, resolution, and
reproducibility by minimizing the error between a process vector created
from a model of the process and the set point vector [2.6]. However, its
model is based on the assumption that the control valve actually moved
for the recent past changes in the controller output. Thus, advanced con-

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
trol algorithms are more vulnerable to deficiencies in the control valve
than measurement. While a kicker algorithm can theoretically reduce the
effect of dead band and stiction, overcorrection will cause excessive move-
ment similar to slip [2.4]. There is no computational correction for valve
slip. The effect of slip is amplified by high valve sensitivity (valve gain)
and high process sensitivity (process gain). The only solution for slip, and
the best solution for stiction and dead band, is a change in the valve type,
assembly, and accessories or the use of a variable-speed drive.

To summarize, while it is desirable for the measurement to be linear and

accurate and the control valve’s effect on the process to be linear and pre-
cise, it is most important that the signal-to-noise ratio, reproducibility,
time delay, and time lag of the measurement, and the stick-slip and dead
band of the control valve, be scrutinized in case they prevent the advanced

Copyright International Society of Automation

control system from doing its job. This scrutiny involves an analysis of the
type, location, and installation of the instrument and final element. Both
the degree to which deficiencies in the measurement can be compensated
for by advanced control techniques, and the permissible amount of stick-
slip and dead band, must be part of a cost-benefit analysis.

This chapter offers enough theory to provide a deeper understanding of

the origin of the relationships and rules presented, including the funda-
mentals necessary for graduate studies and to develop new tools for
advanced control. The Theory section covers the setup and direct solution
of the differential equations for material and energy balances to show the
fundamental relationships between process parameters and process time
constants and process gains. It also covers Frequency Response in enough
detail to show the source of the equations for the ultimate period and gain
and the basis of the first-order-plus-dead-time-approximation method for
loop analysis. The theory behind the various advanced control tools is dis-
cussed in succeeding chapters.

Opportunity Assessment
In this section, some questions are offered that could form an OA to find
improvements in a basic control system. Question (1) deals with the ability
to overdrive the manipulated variable on startup or for a major set point
change in a batch operation to reduce the amount of time it takes for the
controlled variable to reach set point. This question is also important in
performance of an advanced control system to help reduce the time lag in
the manipulated variable for the model predictive controller. There is, of
course, a tradeoff between rise time and degree of permissible overshoot,
but in general, for temperature and composition control of volumes with
mixing and for the start of a continuous or batch process, the output
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

should initially be saturated high but backed off from this limit before the
controlled variable approaches set point. Various algorithms and tuning
methods are available to provide overdrive. The fraction that the startup
time or a batch cycle can be reduced is proportional to the ratio of the
missing area of overdrive to the total area of the manipulated variable dur-
ing the rise time.

Questions (2) through (4) determine whether a control loop is creating

variability instead of transferring the proper amount of variability from
controlled variables to manipulated variables. Questions (5) and (7)
involve looking at an application and installation and spotting some obvi-
ous flaws. The next three questions look for opportunities to apply cas-
cade and ratio control (also known as flow feedforward). These questions
generally apply to both batch and continuous unit operations. Questions
(11) through (14) look for opportunities to reduce batch cycle time. Appen-
dix B expands upon these questions to take a closer look at the methods to

Copyright International Society of Automation

reduce the time to reach a batch set point and eliminate operator attention
requests.

1. Could the valve position be initially driven beyond the resting

position when the controller settles out at set point, to reduce
continuous process startup time or batch process cycle time?

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
2. Is the variability less in the controlled variable of the loop when the
controller is in manual?
3. Is the variability less in other loops when the controller is in
manual?
4. Is the variability less in the process variable for an important
constraint if the controller gain or rate setting is decreased or
integral time is increased?
5. Would better reproducibility and less noise in measurement reduce
the variability in a process variable for an important constraint?
6. Have tight shutoff valves, high temperature packing, key lock
shafts, vane actuators, scotch yoke actuators, or valves without
digital positioners been used in control loops that affect important
constraints?
7. Have any of the top 20 mistakes been made in an important loop?
(See Appendix D for a list of the mistakes made every year for the
last forty years.)
8. Are there opportunities to linearize the manipulated variable for a
primary controller by creating a secondary loop that encloses the
nonlinearity?
9. Are there opportunities to attenuate a load upset to a primary loop
by creating a secondary loop that encloses the disturbance?
10. Are there flows that can be ratioed and used as a feedforward
signal to enforce a material balance for startup and to compensate
for changes in flow rate?
11. For batch operations, can phases be eliminated by going from
sequential to parallel actions, such as simultaneous heating, filling,
pressurization, and venting?
12. Can batch cycle time be reduced by a decrease in wait times, hold
periods, operator attention requests, manual actions, or lab sample
analysis time?
13. Can batch end points be automated by the use of a property
estimator, trajectory, or sustained rate of change?

Copyright International Society of Automation

14. Can batch cycle time be reduced by overdrive or an all-out run and
coast?
If the variability in a loop decreases when the loop is in manual, it indi-
cates that the loop was doing more harm than good, due to poor control
valve performance, inappropriate tuning, and/or interaction. If the valve
does not respond to small steps (e.g., 0.25% to 0.5%) in the controller out-
put, the oscillations are probably due to the control valve. If an increase in
the controller gain or a decrease in the integral time increases the variabil-
ity, it is mostly due to incorrect tuning. Lastly, if the variability in other
loops is less when a loop is put in manual, the variability is the result of
interaction.

If the variability in a loop increases when the loop is in manual, there are
load upsets that were being attenuated by the loop and it was doing some
good. If the variability stays the same, the fluctuations are mostly due to
noise or lack of measurement reproducibility.

There are also some obvious flaws that will stand out from some simple
tests. If there are significant non-uniform fluctuations in the measurement
regardless of the mode of the controller, then the selection and installation
of the transmitter are suspect. These problems are most often associated
with insufficient runs of straight pipe upstream or sensing line problems.
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

If the loop develops a sawtooth oscillation in the controller output when it

is moved out of manual, it is most likely due to excessive packing or seat-
ing friction, backlash in the linkage, shaft windup, and/or poor or missing
positioners. The selection and installation considerations that lead to these
problems will be discussed in the Application Detail section of this chap-
ter.

One of the last and most obvious opportunities is the use of cascade con-
trol and ratio control. The most common type of cascade control is a flow
loop that deals with the nonlinearity of the control characteristic and com-
pensates for pressure upsets so that the primary control loop can manipu-
late flow instead of valve position and not see the effect of pressure
swings. The next most common cascade control system uses a jacket, or
coil inlet or output temperature secondary loop, to insulate a primary
crystallizer or reactor temperature control loop from changes in coolant
temperature and the nonlinearity associated with the manipulation of
coolant makeup flow.

The largest and most frequent opportunities in basic control are summa-
rized in Table 2-1 and discussed in detail throughout the rest of Chapter 2.
Simple equations for the fundamental relationship between either the
standard deviation or the peak or integrated error for upsets can be used
for each type of improvement.

Copyright International Society of Automation

Table 2-1. Largest and Most Frequent Opportunities in Basic Control

• Decrease stick-slip and improve the sensitivity of the final element (Standard
deviation is the product of stick-slip, valve sensitivity (valve gain), and pro-
cess 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.
• If GraphoilTM packing must be used, aggressively tune/monitor the smart posi-
tioner.
• Improve valve type/sizing or add signal characterization to increase valve sen-
sitivity.
• Use variable speed drives where appropriate for the best precision and sensi-
tivity.
• Improve the short- and long-term reproducibility and reduce the interference
and noise in the measurement (Standard deviation is proportional to repro-
ducibility and noise):
• Use vortex, magnetic, and coriolis mass flow meters to eliminate sensing lines.
• Use smart transmitters to reduce process and ambient temperature/pressure
effects.
• Use RTDs and narrow-span temperature transmitters to decrease noise and
drift.
• Reduce loop dead time (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 time constants (sensor lag, dampening, and filter
time).
• Decrease discrete device delays (scan or update time).
• Tune the controllers (Integrated error is inversely proportional to the control-
ler 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 loop to the process time constant of the primary
loop).
• Add feedforward control (Standard deviation is proportional to the root mean
square (RMS) of the measurement, feedforward gain, and timing errors).

Examples

Neutralization Process
Figure 2-4a shows a two-stage neutralization process. The economic vari-
able is yield. The optimum yield is for pH between 6 and 8. A byproduct is
formed that is 1% of the total product when the pH goes above 8. Of
greater concern is the fact that the reaction time increases from 2 minutes
by a factor of ten for each pH unit below 6 pH. The first stage is a static
mixer with a residence time of 2 seconds and the second stage is a well
mixed vessel with a residence time of 20 minutes. The titration curve has a

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

particularly steep slope between 6 and 8 pH (1 ∆pH per 0.0001 ∆ratio) and
will greatly amplify a valve stick-slip limit cycle, as shown in Figure 2-4b.

The following improvements can be made to reduce pH variability and

improve the yield.

1. Change the reagent valves from ball valves without positioners to

sliding stem control valves with digital positioners; otherwise the
stick-slip multiplied by the valve gain will be a limit cycle that
causes a dramatic reduction in yield. Make the second-stage
reagent valve 10 times smaller than the first stage.
2. Close couple the second-stage reagent valve to the recirculation
line just before reentering the vessel to eliminate the huge reagent
delivery-time delay from a small reagent flow dribbling down a
dip tube backfilled with process fluid.
3. Add a secondary reagent flow control loop for the second stage to
create a pH-to-flow cascade control system that compensates for
upsets in reagent pressure and linearizes the manipulated variable
for the pH loop. Do not add a cascade control system to the first
stage because the secondary loop is not 5 times faster than the
primary loop and the equal percentage valve characteristic has a
valve gain that is proportional to flow that helps compensate for
the process gain that is inversely proportional to flow. The large
well mixed volume of the second stage has a time constant
inversely proportional to flow that cancels out the effect of the
process gain, which is also inversely proportional to flow (see
Equation 2-24); so the linearization by the addition of a flow loop is
desirable for the second stage.
4. Add input signal characterization to translate the controlled
variable from pH to % reagent demand to linearize the first stage
since it will operate over a wide portion of the pH titration curve.
Signal characterization is not necessary for the second stage since it
should be operating well within the control band of 7 to 9 pH.
Provide a faceplate for the operator that displays the actual pH
rather than the linearized controlled variable of reagent demand
for the first stage.
5. Add a feedforward signal to the second stage pH controller output
to handle changes in feed flow and startup. Use a feedforward
summer instead of a feedforward multiplier to the second-stage
pH controller so that a nonlinearity, which is proportional to flow,
is not introduced by multiplying the controller output by feed
flow. Set up the summer so that a controller output of 50% is no
correction and feedback corrections of -50% and +50% are

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

provided by the controller. Display the actual and desired ratio of

reagent flow to feed flow for the second stage and provide a
bumpless transfer to feedforward control.
6. Add a head start to the output of the first-stage pH controller to
help speed up the startup of the system. The head start is
calculated based on estimated ratio of reagent to feed flow
captured from the last run. The calculation must include the effect
of the nonlinear flow characteristic of the reagent valve. Since the
reagent flow is small, the pressure drop is relatively constant and
the inherent characteristic can be used as the installed
characteristic for the control valve. The head start is used to
initialize the controller output at startup. The pH controller
ultimate period should be less than 1 minute for the inline system.
7. Move the reagent flow meters upstream of the control valve for
both stages, to greatly improve the flow profile for the meter and to
reduce reagent delivery delay. If coriolis flow meters are not used,
make sure there are enough straight runs upstream and
downstream of the flow meter. Add an isolation valve close
coupled to the injection point that will automatically close just
before and reopen just after the reagent control valve closes and
opens, respectively, to minimize the backfill of process fluid into
the reagent piping.
8. Move the pH electrodes for the first stage 20 pipe diameters
downstream of the static mixer outlet to provide enough distance
for the streams divided by the mixer to recombine.
9. Move the pH electrodes for the second stage upstream of the heat
exchanger in the recirculation line to within 20 pipe diameters of
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

the discharge pump to minimize the transportation delay.

10. Use a middle-signal selector for each stage to choose the middle
reading from the three pH measurements, thus ignoring a single
signal failure of any type and reducing noise and spikes.
The benefits from the reduction in variability afforded by the listed
improvements will be estimated in Chapter 3. The improvements 1-9 are
illustrated in Figure 2-4c.

Distillation Process
Figure 2-5a shows a distillation column, feed tank, and a storage tank for
the distillate product. The series of plots in Figure 2-5b are indicative of the
nonlinear relationship between tray temperature and both the distillate-to-
feed ratio (Fd/Ff) and the impurity in the product. The process gain seen
by the temperature loop is the slope of the plot versus Fd/Ff. The inverse
of the slope of the plot of temperature versus impurity concentration

Copyright International Society of Automation

Reagent AC
Stage 2 2-1

FT
Reagent AC
2-1
Stage 1 1-1 AT
FC 2-1
1-2 FT
1-1 AT
1-1
FT
Static Mixer
1-2
Feed
2
pipe
diameters

Neutralizer

Discharge

Figure 2-4a. Basic Neutralizer Control System Before Improvements

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Reagent Flow
Influent Flow

Figure 2-4b. Nonlinearity and Sensitivity of pH

expresses the sensitivity of the impurity concentration to measurement

error.

Thermocouple cards with a 400o C span are used for the temperature mea-
surements. The distillate control valve has GraphoilTM packing and a
pneumatic positioner. The storage tank residence time is 4 hours and the

Copyright International Society of Automation

Feedforward
Summer RSP

FT FC
Σ 2-1 2-1
Signal
FT Characterizer
1-1
AC Reagent
f(x) Stage 2
1-1
FC Reagent
1-2 Stage 1 *1
AT
*1 1-1
FT
Static Mixer
1-2
Feed
20
pipe AC
diameters 2-1
Neutralizer
*1 - Isolation valve closes when control valve closes
AT
2-1

Discharge
20
pipe
Figure 2.4c Basic Neutralizer Control System diameters

Figure 2-4c. Basic Neutralizer Control System After Improvements

time delay in the temperature loop is 1 hour. The reflux-to-feed ratio is 10.
If concentration of impurities in the product in the storage tank exceeds
the spec by more than 0.1%, the product must be recycled. For every 0.1%
reduction in impurity the steam flow to the reboiler must be increased by
0.1%.

1. Add a digital positioner to the distillate control valve and tune it

aggressively to deal with the high friction from the Graphoil™
packing and reduce stick-slip. It is important to realize that settings
obtained from the instruction manual or from running a preset
automated step test are generally too sluggish.
2. Change the pairing of the loops at the top of the column so that the
temperature controller manipulates the distillate flow and the
distillate receiver level controller manipulates the reflux flow, since
the distillate flow is too small for level control.
3. Use tray 6 instead of tray 10 for temperature control because tray 6
has a larger and more symmetrical temperature response, per
Figure 2-5b, for a change in distillate flow. Also, the slope versus
impurity concentration is higher, which means the reproducibility
error and resolution of the measurement shows up as a smaller
concentration error as shown in Figure 2-5b. This is generally more
important than the increase in the time delay associated with the
lower tray.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

4. Move the location of the sensor down into the tray so it always is
immersed in the liquid rather than in the vapor, or even worse, a
splashing liquid.
5. Replace the thermocouple and its DCS input with a 3- or 4-wire
RTD with a smart temperature transmitter.
6. Tune the overhead distillate receiver level controller with a high
controller gain to insure that the effects of small changes in
distillate flow translate into changes in reflux flow and thus
changes in the column temperature.
7. Tune the feed tank level controller with a low controller gain to
smooth out the changes in feed to the column. Consider the use of
error squared control. For a batch-to-continuous transition in an
undersized feed tank, use an adapted velocity limited feedforward
per Appendix B for optimum smoothing.
8. Add signal characterization to the controlled variable to
compensate for the nonlinearity in the process variable depicted in
Figure 2-5b. Provide a faceplate for the operator that displays the
actual tray temperature rather than the linearized controlled
variable of distillate demand.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
9. Add a secondary flow controller to each column loop to
compensate for the nonlinearity associated with the control valve
and to prepare the column loop for feedforward control.
10. Add a flow feedforward signal to the temperature and level
controller outputs and display the actual and desired ratio of
distillate to feed. Make sure the feedforward action is active when
the temperature controller is in manual and the operator can easily
go to flow ratio control and adjust the ratio for the startup of the
column.
The benefits from the reduction in variability afforded by the listed
improvements will be estimated in Chapter 3. The improvements are illus-
trated in Figure 2-5c.

Application

General Procedure
1. Track down and correct the source of sustained oscillations. A
power spectrum analyzer may be required to find the loops with
the common period of oscillation. Beware of a slow scan time of
the I/O and controller that will cause a slower than actual period
and a smaller than actual amplitude from aliasing. For trends or
data obtained from data historians, make sure the data highway

Copyright International Society of Automation

PC
3-1

LT LC Vent
3-1 3-1
Feed Tank

Distillate
Receiver

PT
3-1

Reflux Overheads

FC
3-3 Thermocouple

TE TC
Tray 10 3-2
3-2
FT
Column
3-3
Feed
Storage Tank
FC LC
3-4 3-2

FT LT
3-4 3-2
Steam

Bottoms

Figure 2-5a. Basic Column Control System Before Improvements

Operating
Point

Temperature

Measurement Error Tray 6

Measurement Error Tray 10

Distillate Flow

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Feed Flow

% Impurity
Impurity Errors

Figure 2-5b. Nonlinearity and Sensitivity of Tray Temperatures

reporting and the time intervals between data points for historical
data are not too slow and the trigger for exception reporting and
compression is not set too high. Also, the data must be saved for at
least a month to catch different process conditions and modes of
operation.

Copyright International Society of Automation

PC
3-1 Feedforward summer

LC LT Vent
FT3-3 Σ 3-1 3-1
Feed Tank
RSP
FC RSP FC
Distillate
3-1 Receiver 3-2

PT
3-1 FT FT
3-1 3-2
Reflux Overheads
Feedforward summer
FC
3-3 FT3-3 Σ
Signal Characterizer
RTD
FT
Column TT TC
3-3 f(x)
Feed Tray 6
3-2 3-2
Storage Tank
FC LC
3-4 3-2 RSP

FC
3-5
FT LT
3-4 3-2
Steam
FT
3-5
Bottoms

Figure 2-5c. Basic Column Control System After Improvements

The correction most often involves improving the responses of

control valves and the tuning of controllers. Stick-slip, dead band,
and time delay from the final element can be eliminated by the use
of a variable-speed drive. Wave form recognition, step tests, and an
understanding of the relative effects of valve response, controller
tuning, noise, resonance, interactions, and nonlinearities can pin-
point the root cause.

a. The most likely source of a sustained oscillation is a control

valve with stick-slip from excessive friction in either the
packing or sealing surfaces. This occurs most often in valves
with high-temperature or environmental packing, particularly
if it has been tightened. The next-greatest source of stick-slip is
high-performance valves designed to provide tight shutoff. It
can also occur in almost any control valve so severely
oversized that it rides the seat where the valve alternately
closes and then breaks free. In some cases, an aggressively
tuned digital positioner may be sufficient. It is important to
realize that these oscillations are always present because all
valves have some stick-slip (a finite resolution). It is just a
matter of degree. If the stick-slip is small enough, it is
smoothed out by signal filtering and data compression or
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`

Copyright International Society of Automation

washed out by back-mixed volumes. It is not surprising that

hidden oscillations have been found in most of the loops of a
plant [2.2].
b. The second most likely source of oscillations is a level
controller that has a control valve with either stick-slip or dead
band. For integrating processes such as level, sustained
oscillations occur for valves with excessive dead band besides
stick-slip, which is the source of limit cycles in processes with a
self-regulating response. Dead band, also known as backlash,
can result from linkages and shaft connections that are not tight
and is thus more likely to occur in rotary valves because they
involve either a translation of vertical actuator motion to rotary
motion or the connection of a rotary actuator shaft to the shaft
of the ball or disc. The Application Detail section explains how
dead band increases the susceptibility of the level loop to a

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
reset cycle and how the integral time must be increased [2.3].
The best fix, outlined in the Application Detail section, can be
relatively expensive in that it requires a new control valve
designed to minimize backlash and friction or a variable-speed
drive. An alternative that can mitigate but not eliminate the
limit cycle is a level-to-flow cascade loop.
c. Controllers can create periodic upsets from noise if the reaction
to the noise causes the controller output to exceed the dead
band of the control valve from the gain or rate setting being too
large. This most often happens in level controllers, where
controller gains can be quite large.
d. Controllers can amplify periodic upsets whose period is close
to the natural period of the control loop. Resonance occurs
from the feedback action of the controller being in phase with
the disturbance oscillation. It most often occurs for control
loops in series that have similar loop time delays such as liquid
pressure and flow, and inline equipment in series (heat
exchangers, static mixers, and desuperheaters).
e. Interacting controllers can cause sustained oscillations. Here,
the output of a controller affects another controller and vice
versa. A steady state relative gain analysis (RGA) can reveal
the nature of the interaction. However, the dynamics must be
considered as well since the interaction is particularly severe if
the periods of oscillation of the loops are similar. The best
solution is a change in pairing of the control loops per the
RGA. If this is not feasible, model predictive control (MPC)
should be used.

Copyright International Society of Automation

However, beware of stiff or singular processes that may not even be

handled by MPC. A singular matrix occurs when process gains for one
control variable are a factor of the process gains for another controlled
variable, making it impossible to achieve independent set points for these
controlled variables (see Chapter 9 for more details). These processes are
becoming more common as plants push the capacity limits of their plants
without making the necessary modifications to the process equipment.
a. A measurement resolution that approaches the size of the

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
control band can cause sustained oscillations. With the
disappearance of mechanical sensors, this primarily occurs for
temperature control loops that use thermocouple input cards
instead of narrow-range smart transmitters.
b. Although less common, sustained oscillations can also occur
from controllers tuned so aggressively that they bang back and
forth between output limits, and from nonlinear loops that
have a very high central gain region surrounded by
exceptionally low gain regions. This can occur for control
valves when an insufficient fraction of the system drop has
been allocated as a valve drop, strong acid and base titration
curves, and the temperature response of some monomer and
water distillations. For process nonlinearities, the addition of
signal characterization of the controlled variable and rate
action can eliminate or mitigate the limit cycle. For valve drop
problems, the size of the piping and/or the pump impeller
may need to be increased.
2. Track down the source of long settling times. Here, the oscillations
eventually die out but take too long or cause too much variability.
The most common cause is inappropriate controller tuning, such as
the use of too much reset action (too small an integral time) in
evaporator, reactor, or column temperature or concentration
controllers, or other loops dominated by a large time constant; too
much gain or rate action in level controllers on surge tanks; and too
much gain or rate action in liquid pressure, flow, inline
concentration (blending), or sheet gauge or moisture controllers or
in loops dominated by a large time delay (dead time dominant).
3. Check the sensor selection, installation, and location for
opportunities, per best practices, to improve the reliability,
reproducibility, rangeability, and resolution, and to reduce noise
and decrease loop time delay. Orifice meters and chromatographs
are some of the least reliable measurements and are the biggest
sources of excessively fast and slow noise. Chromatographs are
also the largest source of measurement time delay from sample
transportation and analysis cycle time. Look for ways to eliminate

Copyright International Society of Automation

sensing lines and sample lines by the use of sensors that mount
directly in the pipeline or on the vessel [2.7].
4. Look for ways to reduce the time delay in control loops by changes
in the design of the equipment, piping, instrumentation, final
elements, and the pairing of controlled variables with manipulated
variables.
5. Tune the controllers for the best compromise between robustness
(stability), performance, and smoothness. It is important to realize
that the tuning rules change with the ratio of time delay to time
constant and that all loops will see both load upsets and set point
changes. Methods that focus on set point changes (servo control)
and noise introduced into the measurement are applicable to
aerospace, web, and parts manufacturing but not to processes for
the chemical, petroleum, food, and drug industries and
environmental control. Make sure the tuning method takes into
account the relative degree of dead time and provides the proper
capability for load rejection. Beware of any control loop analysis
that concentrates solely on set point response, and the introduction
of noise or upsets downstream of the process and directly into the
measurement [2.8]. These methods were developed from control
programs in system science or electrical engineering and tend to
ignore the effect of the process and equipment dynamics,
characteristics, and objectives.
6. Find opportunities to employ cascade control. Wherever there is a
reliable flow measurement and a primary loop whose time delay
and time constant are more than five times slower than a flow
loop, a secondary flow controller should be created. There are
some cases where the dynamics are not appropriate for cascade
control. Examples of undesirable choices would be inline pH-to-
reagent flow and liquid pressure–to-flow cascade control because
the primary and secondary loops have about the same time delay.
7. Look for opportunities to add feedforward control, especially flow
feedforward where manipulated flows are ratioed to a feed flow.
Make sure the feedforward signal does not arrive too soon and
cause inverse response.
8. For improvements that cannot be covered by the maintenance
budget, the benefits from the reduction in variability can be
estimated by the calculations in Chapter 3 to justify the project.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Application Detail
This section will take a closer look at the methods to improve the response
of valves and measurements, reduce the total loop time delay, tune con-
trollers, employ cascade control, and add feedforward control.

Valve Selection
Control valves are often selected based on the lowest cost valve that has
the required materials of construction. Often tight shutoff is sought.
Nowhere in the valve specification is there a requirement that the control
valve move or respond to a change in signal. Consequently, rotary valves
are chosen because they are the least expensive and offer models with low
leakage rates. They are also thought to offer the highest rangeability. In
reality, the rotary valve has the least usable rangeability because the
installed characteristic gets too flat for small and large controller outputs.
Figure 2-3a shows how the characteristic is too flat below 5 degrees and
above 45 degrees for a butterfly valve. Figure 2-3b shows how the charac-
teristic is too flat below 10 degrees and above 60 degrees for a ball valve. If
you further take into account that the stick-slip significantly increases
when these valves are less than 15 degrees open, the actual usable range-
ability of these valves is less than 20:1, instead of the 200:1 and 400:1 stated
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

in the literature.

By contrast, the sliding stem valve installed characteristic doesn’t get too
flat until it gets below 5% open or above 75% open as illustrated in Figure
2-3c. Also, its stick-slip is an order of magnitude or more less and usually
doesn’t increase dramatically until the valve is less than 5% open. Also,
unlike the rotary valve, the trim movement of a sliding stem valve closely
matches the actuator shaft movement so that a digital positioner, whose
feedback is typically actuator-stem position, can, by aggressive tuning,
actually compensate for high packing friction. As a result, the real range-
ability of sliding stem valves with digital positioners is 40:1.

Of course, valve manufacturers who offer only rotary control valves will
develop clever ways of diverting attention from these issues or even pitch
the opposite by the use of labels like “high performance” and “high range-
ability” that ignore flat installed characteristics and stick-slip. The user
must realize that “high performance” indicates the ability of the valve to
provide tight shutoff and to withstand high temperatures. These same fea-
tures translate into excessive friction and low performance in terms of con-
trol. The use of a digital positioner cannot correct for the inherent stick slip
problems of rotary valves and can essentially deceive the user into think-
ing it is doing a great job by fancy plots and statistics of the step response
of the actuator stem position. Unfortunately, the ball or disc position does
not track the actuator shaft position, because of gaps in linkages, toler-

Copyright International Society of Automation

ances of shaft connections, and shaft windup. A sensitive, low-noise flow

measurement is needed to test the response online [2.9]. Otherwise, the
valve has to be removed and a sensitive travel gauge mounted on the ball
or disc to detect actual movement for small steps or a slow ramp in the
valve signal.

New designs of sliding stem (globe) valves, such as that shown in Figure
2-6, reduce the amount of metal used in the body and pockets and crevices
where process material can stagnate and accumulate. This makes the valve
more competitive with the rotary valve for exotic materials, large line
sizes, and fouling or slurry service. Above 6 inches in line size, the cost of
sliding stem (globe) valves can become large enough to warrant further
investigation. If the reduction in stick-slip and loop variability offered by a
sliding stem valve doesn’t provide an acceptable rate of return on the
additional investment, the user should take a closer look at rotary valves,
but avoid any valves originally designed for isolation or interlocks. Sepa-
rate automated “high performance” or “on-off” valves should be used for
isolation or interlocks, and low friction valves for throttling service. Since
many of the rotary valves are flangeless (wafer bodies), the lifecycle cost
should include not only the cost of increased variability, but the increased
difficulty of proper installation and alignment and the increased risk of a
safety incident and reportable release of hazardous materials.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Integral flange

Stem
guided

Retainer
seat ring
Stream-lined passages

Figure 2-6. Sliding Stem Valve with Streamlined Passages and Less Metal

Rotary valves must also pass the checks on the maximum pressure drop.
The rotary valve must meet the maximum pressure drop rating at shutoff
and the maximum allowable pressure drop to avoid choked flow, flashing,
cavitation, and exceeding the noise limit. In general, sliding stem valves
offer higher pressure drop ratings, higher allowable pressure drops to pre-
vent cavitation, and noise reduction trim, and are thus the first choice for
high pressure, boiler-feed water, steam and condensate systems.

Copyright International Society of Automation

If a rotary valve is still the best choice, make sure the connection of the
actuator shaft to the ball or disc stem is a splined connection, as shown in
Figure 2-7, to minimize the tolerance and associated play in the connection
so that the backlash is less than 0.5%. Key lock connections can cause a
backlash of 8%. Also, the shaft diameter should be large and the shaft
length should be short so that shaft windup does not cause a stick-slip
greater than 0.5% [2.10].

Figure 2-7. Splined Shaft Connection

To summarize the performance differences, sliding stem valves with low

friction packing and digital positioners have a stick-slip that approaches
the analog-to-digital converter resolution of 0.05% and a rangeability that
approaches 50:1. Rotary valves exhibit a stick-slip and backlash from 0.5%
to 10% and a rangeability of 10:1 to 20:1, even with a digital positioner and

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
low friction packing, with the greatest deterioration found in designs that
employ keyed connections, long slender shafts, and high friction sealing
of surfaces for tight shutoff. To help avoid the many traps of creative
advertising, the user should keep in mind the popular myths listed in
Table 2-2.

Table 2-2. Popular Myths About Control Valves

1. Rotary valves have the greatest rangeability.
2. High Performance valves offer high control-loop performance.
3. A digital positioner will make any control valve perform well.
4. A fast actuator will make any valve fast.
5. A piston is the fastest and most reliable actuator.
6. Tight shutoff is desirable for a control valve.
7. The most cost-effective valve is a rotary valve.
8. Fast loops should use boosters instead of positioners.

In the old days of analog controllers, theoretical studies based on Nyquist

plots showed that, for fast loops such as flow and liquid pressure loops, a

Copyright International Society of Automation

booster would provide better control than a positioner. In real life,

unknown bench settings, excessive booster dead band, positive feedback
from high booster outlet port sensitivity, and the high breakaway torque of
tight shutoff valves makes the omission of a positioner risky business
[2.10]. Compressor surge valves have slammed shut due to booster posi-
tive feedback when a booster was used without a positioner [2.1]. In fact,
where large actuators need the high relay capacity of a booster, it should
be used on the outlet of a positioner and a bypass around the booster
adjusted for stability [2.1] [2.11]. In the chemical industry, positioners have
either improved, or at worst had a benign effect on, the performance of all
types of control loops [2.12]. The scan time and optional filter of a digital
process controller, and the tuning flexibility of a modern positioner, elimi-
nate the concern about the violation of the cascade rule where the second-
ary valve position loop is not sufficiently faster than the master process
loop. Actually, there never was much of a concern because flow loops
were tuned with mostly integral, rather than gain, action.
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Pneumatic and electro-pneumatic positioners went out of calibration

within 6 months, the linkages were difficult to adjust, and remote indica-
tion of position required the mounting of a separate valve position trans-
mitter. Digital positioners hold their calibration, offer an indication of
actual valve position (AVP), and perform a wide spectrum of diagnostics
and tests that can be used for predictive rather than reactive maintenance.
The AVP can be monitored and compared to the controller output to deter-
mine dead band, stick-slip, and dead time for sliding stem valves. For
rotary valves, a low-noise-flow measurement is needed since the actuator
stem position does not reflect the actual ball or disc position.

A piston actuator can reduce the stroking time of large valves once a valve
starts to move. However, the design of most pistons exhibits poor resolu-
tion and dead band that will cause an exceptionally slow response to small
changes in valve position that can get worse if the cylinders are not prop-
erly lubricated. For small changes in valve position, a diaphragm actuator
is generally faster and more precise. Also, a diaphragm actuator does not
require lubrication or as much maintenance unless its temperature rating
is exceeded.

For concentrated slurries, particularly those with clumps of material or

rocks, such as waste treatment lime slurry, rotary valves may be the only
choice. In these cases, pulse interval modulation has been used to provide
an extended rangeability when the downstream back-mixed volume was
sufficient to attenuate the pulses [2.1].

Finally, pharmaceutical and food applications may require sanitary control

valves. Angle plug, ball, and diaphragm valves are used. The sliding O-

Copyright International Society of Automation

ring stem seal of an angle valve creates a difficult-to-clean void between

bonnet and stem, and does not insure product media will not leak past the
O-ring and into the environment. The rotary ball valve also has a void
between seals where product can accumulate, and has excessive operating
friction. By far the most popular choice is the Saunders diaphragm valve,
but both it and the pinch valve have a very limited rangeability of flow
and a quick-opening type of flow characteristic that is undesirable for con-
trol. A crevice-free rolling metal diaphragm control valve like that shown
in Figure 2-8 has an almost friction-free mechanism and sliding stem actu-
ation that affords exceptional rangeability and sensitivity [2.13].

Figure 2-8. Crevice-free Sanitary Valve with High Rangeability and Sensitivity

In all valves, there is a valve prestroke time delay: the time it takes for
enough air to move into or out of the actuator to change the air pressure
enough to start to move the actuator stem. The stroking time is the time
required to complete its transition to the new stem position after the actua-
tor starts to move. The tests to document the prestroke time delay and the
stroking time typically consisted of 10% or larger steps, done with the
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

valve disconnected. The results depended solely on the type and size of
actuator and the type and flow capacity of the actuator connections and
accessories; they did not include the effect of valve dead band or stick slip.

Until recently, valve manufacturers reported the actuator prestroke dead

time and stroking time when you asked how fast a valve would respond.
These days, a valve manufacturer will perform actual step tests of the
complete valve assembly but will chose a step size of 10% and an initial
valve position of about 50%, in order to show the fastest response.

Copyright International Society of Automation

To get a realistic picture of valve response under actual loop conditions,

the smallest and largest step size at the minimum expected throttle posi-
tion should be used. For large valves used for compressor surge or pres-
sure control, step sizes of 10 to 50% are appropriate and the prestroke time
delay and stroking time dictate the speed of response. For most other con-
trol valves, step sizes of 0.5% or less should be used and the valve dead
band and stick-slip will dominate the response.

A ramp (a series of small steps held for the loop scan time) would better
duplicate the actual valve response for closed-loop control. The use of a
ramp is particularly important for pneumatic positioners and boosters
because they exhibit a drastic increase in response time as you approach
the resolution limit of the linkages and flapper nozzle assembly. To quan-
tify dead band, stick and slip, a series of steps, first for a reversal of posi-
tion and then in the same direction are used. Each step is held for the
prestroke dead time and stroking time identified from a 10% step. The
steps are continued until movement occurs. The actual valve movement in
excess of the size of the last step is indicative of the amount of slip.

Figure 2-9 shows how the response time changes as a function of the type
of control valve, shaft connections, actuator, and positioner. Diaphragm
actuators, sliding stem valves, and digital positioners have the fastest
response by far to a range of small step sizes, which is the goal of 99% of
all control valves. The combination of an electrical or hydraulic actuator
and a sliding stem valve can have an even better resolution. The only dead
band is what is introduced in the setup of the positioner to eliminate
dither. The prestroke dead time is essentially zero but the stroking time
increases proportionally to the step size and can become quite large for the
electrical actuator. Hydraulic actuators provide the fastest response for
small and large step sizes but are complex and expensive.

Valve Installation
The installation and location requirements for a control valve are generally
less than for a sensor. Ideally, control valves should have the same straight
run of pipe upstream and downstream as a differential head flow meter,
since both constitute a variable orifice [2.9]. Adherence to the straight-run
requirements rarely occurs in industry but is common in the flow test labs
used to establish flow characteristics and flow coefficients [2.9]. The repro-
ducibility error from an erratic flow profile is not as important for a final
element as it is for a measurement because the control loop will drive the
manipulated variable as necessary to reach set point. However, if the flow
is going to be computed through the control valve based on valve position
and pressure drop, the reproducibility of the resulting flow measurement,
and hence the straight-run requirements, become more important.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

400
3 4 5

40.0
Response 6
Time (sec)
1
4.0
2 7

0.4
0.1 1.0 10.0 100
Step Size (%) All valves look good
for about a 10% step

1 - variable speed drive with dead band adjustment set equal to zero
2 - sliding stem valve with diaphragm actuator and a digital positioner
3 - sliding stem valve with diaphragm actuator and pneumatic positioner
4 - rotary valve with piston actuator and digital positioner
5 - rotary valve (tight shutoff) with piston actuator and pneumatic positioner
6 - large valve or damper with any type of positioner
7 - small valve with any type of positioner

Figure 2-9. Response Times for Different Types of Final Elements (Valves,
Actuators and Positioners)

If the control valve is upstream of equipment with large pressure drops,

the minimum pressure in the valve is more likely to stay above the vapor
pressure that starts flashing. A location that offers a lower temperature,
besides a higher pressure, can prevent flashing. Cavitation can be pre-
vented by a staged pressure drop or a low-recovery control valve, such as
a sliding stem valve, so that the bubbles do not implode from the rise of
the discharge pressure above the vapor pressure. The staged pressure
drops can be done internally by special trim, two control valves in series,
or a restriction orifice [2.9]. If cavitation is unavoidable, the damage to pip-
ing can be avoided if the valve is mounted on a vessel nozzle so that the
bubbles implode inside the vessel vapor space [2.9].

A huge source of time delay occurs in concentration control loops when-

ever the volume between a control valve and the injection point into
equipment or piping is either partially filled or large compared to the flow
rate.

For partially filled lines, there is an excessive time delay even when the
valve stays open. A change in valve position causes a crest or valley in the
wave to travel down the pipe or channel. The transportation delay is the
distance divided by the velocity of the wave but the velocity is difficult to
estimate. For a very low flow down a vertical pipe, the velocity of a falling
film can be computed [2.14]. Whenever the control valve closes, manipu-
lated fluid in the downstream piping and injectors or dip tubes slowly
migrates into the equipment or destination and the process fluid backfills
the same volume. The result is a long delay between the closure of the
valve and the end of manipulated fluid flow and a similarly long delay
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

between the opening of the control valve and the start of the manipulated
fluid flow into the equipment or destination.

For a pressurized, completely full pipeline, dip tube, and injector, the time
delay can be estimated as the volume between the valve and the injection
point divided by flow of the manipulated fluid. For pH control systems
where the manipulated fluid is a reagent, the flow can be as low as one
gallon per hour and just one gallon of volume downstream of the valve
can result in one hour of time delay every time the reagent control valve
closes [2.14].

The control valve should have block and drain valves so that it can be
removed safely and easily. For large continuous processes, it is desirable to
have a manual bypass valve to keep the unit online while the valve is
tested or repaired. Also, the outlet isolation valve can be closed and the
bypass valve, shown in Figure 2-10, opened for inline testing and tuning
of the digital positioner at process pressures and temperatures. For slurry
service, rotary valves can be mounted in a vertical-flow up pipe to pro-
mote self-draining and prevent solids buildup. For sliding stem valves, a
vertical mounting will cause additional wear of the packing from the
weight of the actuator.

flush flush

FT
1-1

A B

drain drain

Bypass valves are important

to keep a continuous plant
running when meter or valve
has been removed for repair

Figure 2-10. Installation Requirements for a Flow Meter and Control Valve

The packing should be consistently tightened with a torque wrench to a

specification of less than 90 foot-lbs. A packing that is tightened exces-
sively, to supposedly prevent leakage, will cause a limit cycle and prema-
ture failure of the packing that can cause a release. For high temperatures
and highly hazardous materials, extension bonnets and bellows seals
should be considered instead of high-friction packing.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

The Variable-speed Drive Alternative

The cost of the variable-speed drive (VSD), also known as the variable-fre-
quency drive (VFD), has dropped to the point where it may now be an
attractive final element wherever there is a motor-driven pump or fan.
There is no stick-slip and the only dead band is that introduced by the user
in the electronics. In fact, there is also no measurable time delay, so that for
pressure and flow loops, the loop dead time is largely set by the controller
scan time [2.1]. For furnace pressure control, the use of a VSD on the
induced draft fan and a scan rate fast enough to be equivalent to an analog
controller can lead to much tighter pressure control and fewer pressure
trips [2.15].

There are common misconceptions, inadvertently promoted by a manu-

facturer, that the main advantage of a VSD is energy savings and that a
VSD is slow and lacks rangeability.

The time from minimum to maximum speed is adjustable within the limits
imposed by the impeller inertia and the motor horsepower. The factory
setting is conservative. The speed response is a velocity-limited ramp rate
with no time delay or lag. Consequently, the response for small speed
changes is fast. For example, if it takes a drive 9 seconds to go from 10% to
100% speed, it will take only 0.1 second to change the speed by 1%.

The minimum speed is also adjustable. The factory setting is especially

conservative. Cogging is not a problem for today’s inverters, and over-
heating is not a problem for variable-torque loads typical for a pump or a
fan. Tests have shown that the rangeability of a VSD is greater than the
rangeability of a magnetic flow meter (25:1), instead of the 5:1 suggested
by the manufacturer. However, the minimum speed must provide a total
discharge pressure that is greater than the static pressure to prevent
reverse flow.

To summarize, the VSD provides a consistent and rapid response. In fact,

the response can be so fast that it scares operators because the flow can
reach a set point in less than a second. The controller output and set point
limits must be set to prevent excessively high or low flow rates and the
process variable filter must be set higher to prevent reaction to noise since
there is no dead band. Set point velocity limits may also be required. An
isolation valve may need to be added for shutoff and fail-safe action. The
inverter must be mounted in a climate-controlled room and be maintained
by technicians with special skills for troubleshooting complex electronics.
In addition, it may be necessary to increase the motor enclosure size to
meet the area classification because of the reduced cooling capacity of the
motor’s fans at low speeds. Most motor vendors will only rate the motor
for the area classification if they supply the motor and VSD as a package.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Finally, isolation transformers and best practices must be used to prevent

the introduction of noise from the inverter into the electrical ground sys-
tem. Consequently, the installed cost and the life-cycle cost of a VSD may
be more than a control valve if isolation valves must be added, the plant is
not set up for VSD maintenance, and the benefits from faster or more pre-
cise manipulation of flow are relatively small.

For strong acid and base pH systems, the requirement for precise adjust-
ment would best be met by a VSD, but the flow rates are often too small
for a centrifugal pump and the location of the pump on the ground creates
a huge reagent delivery transportation delay. Instead, an electronically set
metering pump is used with the piping designed to stay full.

Measurement Selection
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

The number of measurement devices is too large for a comprehensive

treatment. This chapter offers an important perspective, highlighting the
most desirable alternatives and the relative performance advantages and
limitations of the most common types for the most general types of appli-
cations. There are exceptions to every rule. The user should consult hand-
books, manuals, and the manufacturer before proceeding with an
implementation.

The objective is to select a measurement with the best reliability, reproduc-

ibility, rangeability and resolution, and the smallest amount of noise by
means of digital devices and the elimination of process lines, process and
ambient effects, and mechanical components. Smart transmitters should
always be used because they provide the best accuracy and rangeability,
compensate for extraneous effects, and offer online diagnostics. However,
the device with the inherently best principle of operation should be
favored over the one with the fanciest diagnostics that will constantly
remind you of a bad choice.

For flow measurements, inline meters such as coriolis mass flow meters,
magnetic flow meters, vortex shedding meters, and thermal mass flow
meters should be considered because these meters eliminate sensing lines,
external connections, and small holes that are the largest source of errors,
failures, leaks, and maintenance. To reduce the effect of flow profiles and
changes in process fluid, preferred order of selection is the order listed.

Coriolis mass flow meters require no straight runs, are not affected by Rey-
nolds Number or fluid properties, and have by far the best reliability,
reproducibility, rangeability, and resolution for the measurement of both
mass flow and density. The coriolis meter is the only true mass flow meter.
Magnetic and vortex flow meters are velocity volumetric devices and can
be used to compute mass flow only for a fixed and known composition by

Copyright International Society of Automation

the measurement of temperature and pressure. This is also true for pitot
tubes with differential pressure, pressure, and temperature transmitters
that are advertised as mass flow meters. The thermal mass flow meter is
not a volumetric meter but its reading will depend upon the heat capacity
and thus the composition of the fluid.

Coriolis flow meters above 2 inches get expensive, but still may be justi-
fied where the ability to measure and control a mass balance is important.
Often overlooked are the benefits of an accurate density measurement and
an approximate temperature measurement (the temperature sensor is on
the outside surface of the tube and is not in contact with the fluid) to create
online estimators of fluid composition. Unfortunately, the the fluid may be
too corrosive for the materials of construction offered, the fluid tempera-
ture or concentration of particles may be too high, or the pipe size too
large.

Coriolis flow meters are so accurate that they can be used to replace load
cells or weigh tanks for batch charges. Also, for flat titration curves and
constant composition feeds and reagents, simple mass ratio control with
coriolis flow meters has proven to be more accurate and more reliable than
a pH control. The hardware cost of a coriolis flow meter is high compared
to a differential head meter, but the installation cost may be less since there
are no straight-run requirements or additional measurements to compen-
sate for pressure and temperature. The project cost may be still higher, but
the life-cycle cost is often significantly better for the coriolis meter, as
shown in Figure 2-11, because the coriolis meter requires less maintenance
and accumulates benefits from tighter control.

Cost cost of maintenance

($) and loss in process
efficiency from drift,
noise, interference,
failures, and leaks

Lost opportunity
after 10 years
coriolis

orifice

design install operate

Time (yrs)

The higher purchase price to get the coriolis technology is partially offset
by lower installation costs but will still often lead to higher project costs
Benefits but can lead to lower Lifecycle costs from less maintenance and better
($) yields from more accurate mass balances and control of stoichiometry

Figure 2-11. Life-cycle Cost Comparison of a Coriolis and Orifice Meter

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Next to the coriolis mass flow meter, the magnetic flow meter has the fewest
interferences since it is not affected by either Reynolds Number or viscos-
ity and is relatively insensitive to flow profile. The main limitation is that
the fluid conductivity must be greater than 1 micromho/cm (0.1 for spe-
cial units). For erosive service, ceramic linings are offered.

The flow profile is a big factor for the vortex meter, particularly near the
low end of the meter’s range. If the velocity drops below 1 fps, or Rey-
nolds Number is less than 20,000, or the viscosity is greater than 30 centi-
poises, or the concentration of particles is above 2%, the vortices are not
shed uniformly and the vortex frequency deviates from a proportional
relationship to flow. At low velocities the reading can become very erratic.

An orifice meter may have a good short-term repeatability specification

under laboratory test conditions. However, the effects of Reynolds Num-
ber, flow profile, orifice wear, and sensing lines significantly deteriorates
the long-term reproducibility. Measurement noise and the square-root
relationship also lead to a rangeability of just 4:1. Smart transmitters can
double this if the straight run of pipe upstream is sufficient. If a large line
size requires the use of a differential head meter, a large-bore direct
mounted smart transmitter should be used as shown in Figure 2-12a. For
less sensitivity to Reynolds Number, flow profile, and process conditions,
use a purged averaging pitot tube or a large-slot Annubar with a direct
integral-mounted smart transmitter with pressure and temperature com-
pensation. The purge is used to prevent the plugging of the small sensing
holes of a pitot tube.

Table 2-3 summarizes the relative performance capability of each of the

most common types of flow measurement. The rangeability and accuracy
of the measurement can be improved generally by a factor of two by the
use of a smart transmitter. The orifice, vortex, and magnetic flow meters
are volumetric meters, so flow ranges in pounds per hour are based on an
assumed velocity profile and density or concentration if temperature-cor-
rected. The numbers in Table 2-3 are largely from Reference 2.16, except
for the reproducibility percentages. They are based on the author’s experi-
ence and reflect the effect of wear and the change in meter coefficient with
operating conditions such as Reynolds Number. The manufacturer should
be consulted for an estimated reproducibility for the actual range of oper-
ating conditions for the application. The drift from wear will be greatest
for slurries or corrosive service and high velocities.

The hardware and setup costs of the radar level measurement has decreased
and many of the calibration complexities have been automated to the
point where the lifecycle cost is often less than the differential pressure
(d/p) method of level measurement. It is the most accurate of the common

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Table 2-3. Comparison of Common Flow Measurements [2.16]

Piping
Type Sizes Range Interferences Reproducibility
A/B**
Coriolis ¼-6” 80:1 1/1 solids, alignment, vibration 0.1% of rate
Magmeter ¼-78” 25:1 5/1 conductivity, electrical noise 0.5% of rate
Vortex ½-12” 9:1* 10/5 profile, viscosity, hydraulics 1.0% of span
Orifice ¼-78” 4:1 10/5 profile, Reynolds Number 5.0% of span
* Assumes a minimum and maximum velocity of 1 and 9 fps, respectively.
** A is the number of upstream pipe diameters and B is the number of downstream
pipe diameters of straight run required (see Figure 2-10).
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

types of level measurement. The main limitation of radar is that the fluid
must have a dielectric constant greater than 2. For tall narrow vessels, the
minimum 8-degree angle of divergence of the beam may result in the
gauge not being able discern the bottom of vessel. The gauge must be pro-
grammed to ignore any obstruction, including dip tubes and agitator
blades, and typically requires an empty tank at some point during the cal-
ibration procedure. Pulsed lasers that are not adversely affected by dust or
vapor can potentially become an even more accurate method of level mea-
surement [2.17]. The angle of divergence is less than a degree and there is
no dielectric requirement. However, a relatively clean sight glass window
is required.

The differential pressure level measurement depends upon density of the fluid
and the condition of the sensing and equalization lines. A second d/p
with both connections below the minimum level can be added to provide
a representative measurement of the density if the vessel is well mixed,
although the accuracy is usually good to only two significant digits. The
sensing and equalization lines can be eliminated by the use of capillary
systems or transmitters mounted directly on vessel flanges for both the
bottom total pressure and top equalization pressure. However, the error
introduced by capillary systems can be significant if there are bubbles in
the fill, or differences in the temperature or length of the capillary. The
communication of signals for the computation of level from dual transmit-
ters is best done digitally to eliminate digital-to-analog (D/A) and analog-
to-digital (A/D) converter errors. Even so, the error increases as the vessel
pressure increases and can become unacceptable because the bottom d/p
measures both liquid head and vessel pressure relative to atmospheric
pressure.

The admittance-probe type of level measurement depends upon the dielectric

constant of the fluid. A second probe can be added to provide a represen-
tative measurement of the dielectric constant if the vessel is well mixed.
Since the dielectric constant is rarely known accurately, the vessel level

Copyright International Society of Automation

must be varied a known amount during the calibration. For hazardous

materials, there is usually no way to visually check the level and another
method of level measurement is needed.

The ultrasonic level measurement device beam is scattered by foam or dust

and is affected by anything that affects the speed of sound. Thus, readings
are affected by changes in the pressure, temperature, and composition of
the vapor space.

Nuclear level measurements are completely isolated from the process, but are
affected by density unless a second device is added. Strip sources are rec-
ommended to eliminate the need for compensation of the difference in
radiation path length from a point source to the strip detector. The license
procedure is considered a hassle and anything nuclear tends to scare peo-
ple even though the exposure is less than what they receive from the sun.

For temperature measurement, a 3- or 4-wire resistance temperature detector

(RTD) is preferred over a thermocouple because it has much less drift and
a much better repeatability and resolution, per Table 2-4. For very high
temperatures, however, a thermocouple may be the only option if an opti-
cal pyrometer is not feasible. A smart transmitter should be used for con-
trol loops instead of thermocouple or RTD input cards in the DCS. The
speed of response of an RTD is only slower than a thermocouple for a
bare-element installation of both because the thermowell and its air gap is
the largest source of measurement lag. Bare-element installations are rare
and even then the difference is only a few seconds, which is negligible
compared to the other thermal lags in most processes.

Table 2-4. Comparison of Common Temperature Measurements [2.19]

Range Drift Repeatability Resolution Power
Type
(°F) (°F/yr) (°F) (°F) (watts)
RTD -200 to 1600 0.02 to 0.2 0.05 to 1 0.002 4x10-2
Thermocouple -300 to 3100 2 to 40 2 to 15 0.1 2x10-7

For analytical measurements, inline meters and probes that do not require
sample systems will pay off by the elimination of sample transportation
delays and the reduction in the life-cycle cost of a sample system. If a den-
sity measurement is sufficient, the coriolis meter offers the fastest and
most accurate and reliable response. For simple water mixture and com-
plex general mixtures, inline meters such as microwave and nuclear mag-
netic resonance (NMR) should be investigated respectively. [Microwave
can be used for simple 2 or 3 component water mixtures, while NMR can
handle aqueous and non-aqueous complex mixtures]. For coal, oil and
mineral slurries, the prompt gamma neutron activation analyzer can

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

deliver rapid, sample-free, elemental analysis on large top-size, bulk mate-

rials moving at meters per second [2.18]. For covalent chemical bonds,
Raman scattering emission spectrophotometers can provide fast analysis
of multiple components in multiple streams by the use of fiber-optic
probes [2.23]. The hardware and setup costs of these analyzers have
decreased with advances in computational capability for complex spec-
trum analysis. Once the system is successfully calibrated and commis-
sioned, the maintainability and reliability, not to mention the speed, are
much better than what can be achieved with chromatographs. However,
the extensive experience base of analyzer specialists (largely chemists)
with chromatographs, and a lack of appreciation for the value of a tradeoff
of some accuracy for better availability and elimination of dead time, have
slowed down the introduction of new inline meter and probe technolo-
gies.

Unfortunately, nothing has proven to be reliable and accurate enough to

replace the glass measurement electrode. The accuracy of iridium oxide
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

sensors is too adversely affected by oxidation reduction potentials, and the

Field Effect Transistor (FET) has not yet been made rugged enough to be a
viable general-purpose replacement for the glass electrode. Significant
improvements have been made in the reference electrode by the use of a
solid polymer with a large area active and imbedded electrolyte, instead of
a tiny porous junction that can plug and internal liquid or gel fill that can
be contaminated with process material. However, the greatest accuracy
and speed of acclimation requires the use of a pressurized flowing refer-
ence junction [2.18].

A three-probe installation with the middle reading selected as the con-

trolled variable offers the greatest accuracy and availability. This has been
the best practice for important pH loops in Monsanto and Solutia since
1986. Middle selection will inherently ignore a single failure or erroneous
reading of any type, will provide online diagnostics, and will reduce the
noise band without the addition of a lag. Two probes are quite common
but introduce more questions than answers since the electrodes rarely
agree. However, if the probe life is short due to chemical attack of the
glass, the preferred solution is then a piston actuated retractable assembly
with an automated immersion, wash, calibration, and soak cycle [2.18].
The probes should be immersed only long enough (e.g., 2 minutes) to get a
valid reading to reduce exposure time and any drift as the reference seeks
to reach equilibrium with the process. The automated calibration check
can be done on a shift or batch basis to reduce the time between successive
insertions to less than 5 minutes.

Copyright International Society of Automation

Measurement Installation
Sensing lines should be eliminated wherever possible by direct mounting
a differential pressure (d/p) transmitter for flow measurement or pressure
measurement, as shown in Figure 2-12a and 2-12b, to the pipe connection.
Isolation, flush, drain, and equalization valves are necessary to minimize
the exposure to chemicals during the removal of the transmitter. An equal-
ization valve is used when both the high and low sides are connected to
the process to offer the opportunity for a zero adjustment and to protect
the d/p from “over-range” from just seeing the upstream pressure.

d/p
FT
Equalization valve H L
1-1

flush drain
Large bore connections

The flush, drain, and equalization valves may be part of

a manifold assembly that is bolted on to the d/p body

Figure 2-12a. A Direct-mounted d/p Transmitter for a Flow Measurement

d/p
PT
H L
1-1
flush drain

gas

Liquid filled connections

require freeze protection
flush
d/p
PT
liquid H L
1-1

drain

Figure 2-12b. A Direct-mounted d/p Transmitter for a Pressure Measurement

The flow meter should be mounted upstream of the control valve as

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

shown in Figure 2-10 to minimize the distortion of the flow profile and to
provide a more constant pressure. For liquid flow, the upstream location
helps prevent a partially filled meter, besides reducing exposure to flash-
ing and cavitation. Bubbles adversely affect the accuracy of all flow meters
and the implosion of bubbles can cause serious damage in addition to

Copyright International Society of Automation

erratic readings. Isolation, flush, and drain valves are again recommended
for safe removal of an inline meter, although some practices for high haz-
ardous materials seek to minimize the number of connections and leak
points. A bypass valve allows the plant to run on manual while the instru-
ment is repaired. The isolation valves upstream of the flow meter shown
must be wide open when the flow meter is in service for the same reasons
that a control valve is undesirable upstream of a flow meter. If there are
solids, the meter can be installed in vertical pipe with flow up to help
drain the piping. For coriolis meters, a single straight tube is desirable to
eliminate erosion at the bends of a U-tube and unequal distribution of sol-
ids in a dual tube. The meter size should be chosen to provide the opti-
mum velocity to minimize the effect of solids concentration on accuracy.

Figure 2-10 and Table 2-3 show the relative straight run requirements for
different types of flow meters (the A and B values are in Table 2-3). An ori-
fice with a large beta ratio (high orifice bore to inside pipe diameter ratio)
has the greatest upstream straight-run requirement, followed by the vor-
tex meter operating at a low fluid velocity. The upstream requirements
also increase for multiple fittings in different planes or valves upstream
that are not completely open or are partially plugged. The upstream
straight-run requirement can be dramatically reduced by the use of
straightening vanes. The manufacturer should be consulted for actual
requirements based on your piping details and process conditions. The
coriolis flow meter has no upstream or downstream straight-run require-
ments.

Sample lines should be eliminated wherever possible for all types of elec-
trodes by the use of insertion or injector assemblies. Injector electrode
holders with manual or automatic retraction are now offered. Even though
these assemblies have built-in isolation, flush, and drain capability, the
user may choose to have the piping set up as shown in Figure 2-13a and 2-
13b to provide additional protection for hazardous materials. For three
electrodes, a series arrangement, as shown in Figure 2-13b, is favored to
help keep the velocity and concentration the same for all three meters. The
electrodes must be pointed down at a 30- to 60-degree angle, as shown in
Figure 2-13c, to prevent a bubble from becoming lodged in the tip or at an
internal electrode. The first electrode should be at least 20 pipe diameters
from the discharge of a pump or static mixer to reduce pressure pulsation
and facilitate some mixing. The electrodes should also be separated by 10
pipe diameters to help establish a more uniform velocity and they should
be inserted far enough into the line or vessel to get a representative read-
ing. The mounting of electrodes in a pipeline with a 5 to 9 fps velocity is
preferred to a vessel because the higher velocity in the pipe makes the
electrodes respond faster and keeps them cleaner. The bulk fluid velocity
in even highly agitated vessels rarely exceeds 1 fps except near the agitator

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

blade tip. For solids and high caustic or temperature service, there is a
compromise to keep the velocity low to decrease erosion and chemical
attack and yet high enough to reduce coatings. The slot in the protective
shroud of the electrode tip should be oriented to shield the electrode from
abrasion and chemical attack but provide a sweeping action to decrease
fouling. Finally, the transportation delay between the vessel or the point of
reagent addition and the electrodes should not exceed 5 seconds.

AE
1-1
pressure drop for
each branch must throttle valve to
be equal to to keep adjust velocity
the velocities equal AE
1-2
flush

AE
1-3

20 pipe diameters drain

Figure 2-13a. Parallel Installation of pH Electrodes

Differences in velocity, concentration, and temperature are less for probes in series !

AE AE AE
1-1 1-1 1-1 throttle valve to
flush adjust velocity
10 pipe 10 pipe
diameters diameters

Injector electrode assemblies with

integral isolation valve
20 pipe diameters drain

Figure 2-13b. Series Installation of pH Electrodes

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

20 to 80 degrees
from horizontal

The bubble inside the glass bulb

can be lodged in tip of a probe
that is horizontal or pointed up or
caught at the internal electrode
of a probe that is vertically down

Figure 2-13c. Orientation and Insertion of pH Electrodes

For temperature sensors, insertion length should be maximized in the cen-

ter of the pipe so that the tip temperature is close to the actual process tem-
perature despite the conduction of heat between the tip and the sensor’s
external connection. This is particularly important for an RTD since the

Copyright International Society of Automation

resistance of the entire element responds to temperature, and for polymers

because of a greater temperature difference between the wall and the cen-
terline of the pipe and a lower convective heat transfer coefficient. How-
ever, an excessively long sensor may vibrate and fail. Computer programs
can calculate the conduction error and maximum length to prevent vibra-
tion for a given thermowell and sensor construction and process condi-
tions. The use of an elbow connection with the sensor or thermowell
facing into the flow as shown in Figure 2-14 provides the greatest accu-
racy. The least desirable choice is option 4, particularly for small diameter
pipes.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Advanced Temperature Measurement and Control, ISA, 1995, pp. 14, Figure 2.3

Figure 2-14. Orientation and Insertion of Thermowells

At the outlet of a static mixer, desuperheater, or multiple-pass heat

exchanger there should be about 20 pipe diameters between the equip-
ment outlet and the sensor location, as shown in Figure 2-15, for the
streams to recombine to provide a more representative temperature. Also,
it is desirable for the velocity to be between 5 and 11 fps to maximize the
convective heat transfer coefficient at the tip from a higher velocity and to
maintain a cleaner surface. The higher coefficient not only improves the
speed of response but also reduces the heat conduction error. For the static
mixer, the thermowell lag is the largest lag in the loop. For non-hazardous
and non-corrosive fluids at safe temperatures and pressures, a bare ele-
ment with a protective shroud could be considered for tighter temperature
control of blending operations by elimination of the thermowell lag.

For d/p measurements of level, either the sensing or equalization line

should be purged as shown in Figure 2-16a. Normally a liquid such as
water is used for the sensing line and a gas such as nitrogen for the equal-
ization line. If nitrogen is used for the bottom connection, a slotted bubbler
tube is used. To prevent drying out the tip of bubbler, which can lead to
solidification of material and plugging, water is sometimes purged along
with nitrogen. The purge flow rates are normally regulated and indicated

Copyright International Society of Automation

flush

elbow

Desuperheater TE
or static mixer 1-1

20 pipe
diameters
drain

Figure 2-15. Location of Temperature Sensor

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

the water purge and the nitrogen purge lines must each have a check valve
before being combined, and each line must have a check valve. The sens-
ing and equalization lines for low to moderate vessel pressures can be
eliminated by the use of separate d/ps for the total pressure and equaliza-
tion, direct mounted on the bottom and top nozzles as shown in Figure 2-
16b. A third d/p can be direct mounted at an intermediate nozzle to com-
pute fluid density, and a temperature sensor can be used to compensate
for the changes in the dimensions of the vessel, to provide a more accurate
level measurement. Flush connections or extended diaphragms are used
for the lower nozzles to help keep the diaphragm clean. The signals are
communicated digitally to a computer or a programmable electronic con-
trol system.

equalization line

flush

LT
H L Purge
1-1
d/p
drain drain

Figure 2-16a. Purged Sensing and Equalization Lines for Level Measurement

Open Loop Response

If a controller is put in manual, and a step change is made in the controller
output, the trend recording will show the open loop response. In indus-
trial processes, the open loop response is rarely oscillatory but follows a

Copyright International Society of Automation

flush
d/p
LT
H L
1-2

drain digital

LY LT
1-3 1-3

flush digital

LT
H L
1-1
d/p
drain

Figure 2-16b. Direct-mounted Transmitters for Level Measurement

path of a self-regulating, integrating, or a runaway process as shown in

Figure 2-17. If oscillations occur, the root cause is hunted down and cor-
rected. Sometimes oscillations occur in the open loop response of a pri-
mary controller of a cascade control system, but this is easily fixed by
retuning the secondary controller to provide a smoother response. Indus-
trial practice now is to eliminate oscillations even in the closed-loop
response (the response when the controller is in automatic). The quarter
amplitude decay for closed-loop control in the literature is a benchmark
for minimum peak error and is only seen as part of a closed-loop tuning
procedure that then seeks to suppress the oscillation as described in the
next section on controller tuning.

Thus, for purposes of this chapter, the open loop response will be a first-
order response that can be characterized by a total time delay, a negative
or positive feedback time constant, and a steady state gain. The time it
takes the process variable to get out of the noise band after a step change
in the controller output (CO), is the observed time delay (τd), or dead time.
It excludes any time delay due to valve dead band or stiction. The Theory
section shows how to estimate the additional time delay from valve dead
band. The time it takes after the time delay for the response of the process
variable (PV) to reach 63% of the final change for a self-regulating
response is the negative feedback open loop time constant (το), or time lag.

Alternatively, a tangent to the inflection point can be visualized on the

trend. The intersections of the tangent with the initial and final values of
the PV mark the end of the total time delay and the open loop time con-
stant, respectively. The final change in the process variable in percent
divided by the step change in the controller output in percent is the open
loop steady state (static) gain, more commonly known as the process gain
even though it is the product of the valve, process, and measurement static
gains as shown by Equation 2-1 and Figure 2-18. Included in the process

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

gain for temperature and composition control loops is a flow-ratio gain

that is the inverse of the feed flow as shown in Figure 2-18 for pH. All flow,
liquid pressure, most gas pressure, temperature, and composition loops
have a self-regulating response.

For an integrating or a runaway response, the process variable does not

line out at a new steady state, but continues to ramp or accelerate, respec-
tively, until a physical limit is hit. For an integrating response, there is an
integrator gain that is the ramp rate in percent per minute divided by the
step change in controller output in percent, per Equation 2-2. It can be
approximated as a large open loop time constant and gain. In fact, pro-
cesses with a large time constant look like a ramp in the control region and
are called pseudo-integrators. For a runaway process, there is a positive
feedback time constant that is the time, after the time delay and any nega-
tive feedback time constant, for the open loop response to reach 172% of
the self-regulating response as shown in Figure 2-17.

The open loop response of each major component of the plant (control
valve, process, and measurement) in Figure 2-19 has a first-order-plus-
dead-time approximation. The controller also contributes a time delay
from the scan time and time constants from the signal filters. Material and
energy balances are used in the Theory section to show origin of the pro-
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

cess time constants and gains and how they change with operating condi-
tions. Equations in the Theory section also estimate the dead time from
mixing, transportation delay, and valve dead band.

Process
Variable
0.72∗Eo
(%)

0 Acceleration

Ramp
1 ∆PV
Eo Load
Upset ∆t
curve 0 = Self-Regulating
2 curve 1 = Integrating
curve 2 = Runaway

Time
0 τd τ τ’ (min)

Intersection of the tangent line with the initial Positive Feedback

value of PV marks the end of the time delay Time Constant

Figure 2-17. First Order-plus-Dead Time Approximation for an Open Loop

Response

Copyright International Society of Automation

Controlled Ko ∆MV
Manipulated Kmv
Variable ∆CV Variable ∆CO
(%) (gpm)

Time Controller Output

(minutes) (%)

Steep slope (high process gain) makes loop

sensitive to valve resolution and stroke slip Finally something linear

∆PV
Process Kpv Controlled Kcv
∆CV
Variable Variable
(pH) ∆FR
(%) ∆PV

Reagent to Feed Flow Process Variable

Ratio (gpm/gpm) (pH)

Figure 2-18. Steady-State (Static) Gains for a pH Loop

Delay Lag Gain

τdL τL KL
∆DV Fi
Load Upset

Delay Lag Gain Delay Lag Delay Lag Gain

τdv τv Kmv τdp1 τp1 τdp2 τp2 Kpv
∆MV Fr ∆FR
Valve Process ∆PV

∆CO Local
Set Point
PID Kc Ti Td

∆CV
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Delay Lag Gain Lag Delay Lag

τc2 τdc τc1 Kcv τm2 τdm2 τm1 τdm1
Lag Delay
Controller Measurement

Figure 2-19. Block Diagram for a First Order-plus-Dead Time Approximation

Equations

∆%CV
K o = ------------------- = K mv * K pv * K cv (2-1)
∆%MV

∆%PV ⁄ ∆ t K
K i = --------------------------- = ------o (2-2)
∆%CO τo

Copyright International Society of Automation

where:

Ko = open loop gain (%/%)

Ki = integrator gain (%/minute/%)
Kmv = steady-state gain of the manipulated variable (valve gain)
Kpv = steady-state gain of the process variable (process gain)
Kcv = steady-state gain of controlled variable (measurement gain)
το = open loop time constant (minutes)

The total loop time delay (dead time) is the most important of the three
key variables that describe the open loop response of a control loop. It
delays the ability of a controller to see or react to a disturbance. The
minimum peak error is the maximum excursion of the process variable
during this time delay. The oscillation period is also proportional to the
time delay. Thus the integrated error is proportional to the time delay
squared for self-regulating processes with a large time constant that limits
the excursion within the time delay to less than the full change of the
process variable. The equations to approximate these relationships are
developed in the Theory section.

Perfect control is theoretically possible if the total time delay is zero and
there is just a single process time constant. However, in industrial pro-
cesses the total loop time delay is never negligible because even if the pro-
cess time delay is negligible, the addition of a measurement, valve, and
digital controller adds time delay. For flow, pressure, and level control,
most of the time delay in a control loop comes from the automation system
[2.1]. While the time delay cannot be zero, the objective, particularly for
loops with operating point nonlinearities, like the pH and column temper-
ature control examples, is to reduce the total time delay. This is because the
extent of the excursion on the titration curve or tray temperature curve,
and hence the effect of the nonlinearity, is decreased by a reduction in loop
dead time.

Pure dead times come from transportation delays (pipes, sample lines,
static mixers, coils, jackets, conveyors, sheet lines, and textile fiber lines),
valves (prestroke dead time, dead band and stiction), and anything digital
or with a cycle time (microprocessors and analyzers). Equivalent dead
time comes from time constants in series from instrumentation (sensor
time lags, thermowell time lags, and transmitter filter times and dampen-
ing adjustments), analog input cards, (analog filters), and process variable
filter times (digital filters). The exact values are not important, just the rel-
ative sizes. The engineer or technician should work on the largest, most
cost-effective sources of dead time [it is not just the job of the control engi-
neer but also the process engineer who is responsible for equipment
design and the selection of instruments for small automation projects and

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

the technician who often determines the sensor location and loop scan
times. Some plants don’t have a control engineer].

A time constant is beneficial, if it is located in the process after the entry

point of load upsets, because it slows down the excursion of the process
variable from the disturbance. If a time constant is between the controller
output and the entry point of the load upset, it slows down the correction
for the disturbance from the controller. If a time constant is between the
process output and the controller input, it affects the recognition of the
disturbance. In the block diagram in Figure 2-19, the only beneficial time
constant is the second process time constant (τp2).

The largest time constant does not have to be in the process. For processes
not dominated by time delay, an increase in the time constant, no matter
where it appears in the loop, will allow an increase in controller gain, even
though the final effect is not beneficial. For example, a large time constant
in the measurement, such as a large thermowell lag or process variable fil-
ter time setting, will allow a higher controller gain and give the illusion of
better control because the controlled variable is an attenuated version of
the real process variable. Equation 3-2 can be used to estimate the effect.

All time constants much smaller than the largest time constant are con-
verted to equivalent dead time. While the fraction converted to dead time
depends upon the relative size of the small to the large time constant, very
small time constants can be summed up as totally converted to dead time
because it is difficult to find and estimate all the small time constants and
sources of dead time. Thus, the total time delay for a control loop is the
sum of all the pure time delays and the small time constants. Dead time
compensators and model predictive controllers can account for the effect
of time delay on the response to changes in the controller output, but the
minimum peak error for unmeasured load upsets and the initial delay
before the start of the set point response is still fixed by the total time
delay.

The open loop time constant is approximately the largest of the time con-
stants plus the portion of all of the small time constants not converted to
time delay. If each of the small time constants is less than 10% of the larg-
est time constant, so that each is essentially converted to equivalent time
delay, the largest time constant can be considered to be the open loop time
constant. The purpose here is to show the relative sizes and sources of
time delay. There are too many unknowns to calculate an exact value. In
industry, the open loop time constant, total time delay, and open loop gain
can not be accurately calculated, except possibly for flow and level, and
must obtained by plant tests.
--`,```,,,```,`,````

Copyright International Society of Automation

The open loop gain (sensitivity) can be too low or too high. If the valve
gain is too low (valve sensitivity is too low), the controller has little effect
on the process and the controlled variable will wander and be at the mercy
of upsets [2.5]. If the process or measurement gain (sensitivity) is too low,
the controlled variable is not representative of the process performance. If
the valve gain (sensitivity) is too high, the effect of stick-slip is excessive
and just the act of putting a controller in automatic can cause unacceptable
oscillations. If the process or measurement gain (sensitivity) is too high,
the nearly full-scale oscillations will scare most people even if the actual
performance of the process is acceptable. The classic example of this prob-
lem is the pH control of a strong acid and base system with a static mixer.
Try explaining to the operator that the actual change in hydrogen ion con-
centration is tiny for a system that oscillates between 2 and 12 pH.
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

The total loop time delay, open loop time constant, and open loop gain are
rarely constant but rather a function of operating conditions. For example,
the process gain for composition and temperature control is inversely pro-
portional to feed rate, and the process time constant is inversely propor-
tional to feed flow for back-mixed volumes, whereas the time delay is
inversely proportional to feed flow for plug flow. The theory section
shows the source of these process nonlinearities.

PID Controller Tuning

Proportional-Integral-Derivative (PID) controllers in the basic process-
control system must be tuned for both servo response (set point changes)
and regulation (load rejection). While either of these may be more impor-
tant than the other to a given application, loop tuning methods that focus
on one to the exclusion of the other are not seeing the whole picture.

Increasingly, the control loop set point is changed by either a master loop
for cascade control or a model predictive controller for advanced control,
or by a unit operation for an automated startup sequence, product transi-
tion, and batch operation. Even for those loops whose set point is not
changed from a remote source, the local set point is a handle used for star-
tup, sweet spots, and to relieve boredom. Operators are notorious for
moving set points despite claims to the contrary. Plus, the operator has to
start up the unit, which may consist of a series of set point changes as he
walks the unit up to operating conditions.

On the other hand, if there were no process upsets, you wouldn’t need a
controller: You could find and manually set an output to a final element
that would be good indefinitely.

The traditional performance index has been integrated absolute error

(IAE), that is, the integral of the absolute error between the set point and

Copyright International Society of Automation

controlled variable. The integrated error (Ei) can be estimated from the
tuning settings and is equivalent to the IAE if the response is not oscilla-
tory. Of increasing importance is the settling time (Ts), which is the time it
takes for a loop to stay within a specified band around the set point after a
set point change or load upset to detect sustained oscillations. Since pro-
cesses and equipment have limits that can trigger interlocks, violate envi-
ronmental constraints, or initiate side reactions, the overshoot (E1) for set
point changes and the peak error for load upsets (Ex) are also important.
The decay ratio is the amplitude of the first peak (E1 or Ex) divided by the
second peak (E2). Finally, the rise time (Tr) (time it takes the controlled
variable to first reach a specified band around the set point) is important
for reducing batch cycle, startup, and transition time and decreasing the
open loop response time (T98) (time to reach 98% of the final value) for
master and model predictive controllers. Figures 2-20a and 2-20b show the
closed-loop performance indices for a set point change and a load upset,
respectively. In the Theory section, equations are developed to estimate Ei
and Ex for load upsets.

Overshoot = A
A B

Decay = B/A

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Rise
Time

Settling Time

Figure 2-20a. Closed Loop Performance Indices for a 10% Setpoint Change

The proportional mode provides a contribution to the controller output that

is the product of the controller gain (Kc) and the control error, which is the
difference between the set point and the controlled variable. In Figure 2-
21, the controller is not actually connected to a process so the controller
response can be isolated from the process response. Figure 2-21 shows that
the response of the proportional mode to a step change in the measure-

Copyright International Society of Automation

Peak Error = A
A
B

Decay = B/A

Settling Time

Figure 2-20b. Closed Loop Performance Indices for a 40% Load Upset

ment is a step change in the controller output. Consequently, any abrupt

change in the error will be passed on to the output. For a persistent control
error, the proportional mode does nothing more after the initial reaction.
The proportional mode promotes an immediate response, but will not
eliminate an offset. The size of the offset is inversely proportional to con-
troller gain, so for high gains the offset is small. However, digital displays
promote unrealistic expectations and some operators will get concerned
over insignificant persistent errors and ask for the controller to be tuned to
get rid of an offset. The direction of the change in controller output
depends upon the sign of the change in error, so the proportional mode
has some sense of direction of approach to set point.

An increase in the controller gain will greatly reduce the peak error, the
rise time, and the return time, but may increase the overshoot and the set-
tling time. High controller gains will amplify noise, increase interaction,
and pass on more variability from the controlled variability to the manipu-
lated variable. Figures 2-22a and 2-22b show the effect of the controller
gain setting on the response of an effectively proportional-only controller
to a load upset and a set point change, for a process with a time constant
that is 5 times larger than the time delay. An increase in controller gain
dramatically speeds up the initial rate of approach to set point by over-
driving the controller output. It will also start to back off the controller
output from the output limit as soon as the controlled variable comes
within the proportional band, which is the percent change in the control
error (difference between the measurement and set point) necessary to

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Signal Hump or spike

(%) Immediate change from rate*
from gain

∆%CO2 = ∆%CO1
Ramping or driving
∆%CO1 action from reset
seconds/repeat

∆%PV offset

set point
The offset is inversely proportional to gain but is
only completely eliminated by integral action

0 Time
(seconds)

•rate amplifies noise and the abrupt changes in dead time

dominant processes (so does gain but to a lesser extent)

Figure 2-21. Contribution of the Proportional, Integral, and Derivative Mo des

cause a full-scale change (100%) in the controller output. For a propor-

tional-only controller with zero bias and a 50% set point, the proportional
band is centered on the set point. High controller gains by themselves do

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
not cause a classic overshoot but instead an oscillatory approach to set
point. It is high controller gain combined with integral action that causes
overshoot.

Figure 2-22a. Response of a Proportional-only Controller to a 10% Set point

Change

Copyright International Society of Automation

Provided by IHS under license with ISA Licensee=PDVSA - Complejo Petroquimico Site 12/9986712018, User=Cordova, Daniel
No reproduction or networking permitted without license from IHS Not for Resale, 10/11/2016 [Link] MDT
Chapter 2 – Setting the Foundation 55
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Figure 2-22b. Response of a Proportional-only Controller to a 40% Load Upset

The integral mode’s contribution to controller output is the product of the

controller gain (Kc), the inverse of the integral time (Ti), and an integral of
the control error. Figure 2-21 shows that the response of the integral mode
to a step change in the measurement is a ramp. The time it takes to repeat
the contribution of the proportional mode is the integral time in minutes
or seconds per repeat. The integral mode is not satisfied with any offset. It
provides a gradual but driving action. The direction of the change in con-
troller output depends solely upon the sign of the error, so the integral
mode has no sense of direction or rate of approach to set point.

A decrease in integral (reset) time, which is an increase in reset action, has

a negligible effect on the peak error for step load upset and the initial
approach to set point, but will decrease the rise or return time. Unfortu-
nately, reset action increases the overshoot, the period of oscillation, and
settling time, unless the controller gain is reduced or derivative (rate)
action is added. Figures 2-23a and 2-23b show the effect of the integral
time setting on the response of an effectively integral-only controller to a
load upset and a set point change, for a process with a time constant that is
5 times larger than the time delay. The controller output gradually
approaches the output limit and does not reverse direction until well after
the controlled variable has crossed the set point despite anti-reset windup
protection. The integral mode does not amplify noise or increase interac-
tion but will increase the response time of a master or model predictive
controller.

Copyright International Society of Automation

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Figure 2-23a. Response of an Integral-only Controller to a 10% Set point
Change

Figure 2-23b. Response of an Integral-only Controller to a 40% Load Upset

The derivative mode’s contribution to controller output is the product of the

controller gain (Kc), the derivative (rate) time (Td), and the rate of change
of the controlled variable. Figure 2-21 shows that the response of the deriv-
ative mode to a step change in the measurement is a bump instead of a

Copyright International Society of Automation

spike because of a built-in filter whose time constant is typically 1/8 of the
derivative time. The contribution will decay to zero since the offset has a
slope of zero. The direction of the change in controller output depends
upon the sign of the change or the acceleration of the error, so the propor-
tional mode has a definite sense of direction and rate of approach to set
point.

The PID algorithm can have the derivative action on either the control
error or on the controlled variable. The latter method was developed to
reduce the bumps to the controller output from rapid set point changes
made by the operator. Unfortunately, for set point changes made by
sequences, cascade, and advanced control systems, the derivative mode
works against the change requested because it only knows that the mea-
surement is starting to move and that any movement is undesirable. For
this reason, derivative action on control error is preferred so that deriva-
tive action is beneficial if the loop is dominated by a time constant and set
point velocity limits are readily available. An increase in derivative time
will decrease the peak error and reduce overshoot and the period of oscil-
lation. Too much derivative action can increase the rise, or return, time and
the settling time. The derivative mode can respectively speed up or slow
down the initial approach for a set point change by acting on the control
error or the controlled variable.

Figures 2-24a and 2-24b show the effect of the derivative time setting on
the response of a proportional-plus-derivative controller to a load upset
and a set point change, for a process with a time constant that is 4 times
larger than the time delay. The derivative mode is even more likely than
the gain mode to amplify noise, increase interaction, and transfer variabil-
ity to the manipulated variable. It can decrease or increase the response
time of master or model predictive controllers depending on how it is
used. It should not be used on dead time–dominant systems or any system
with abrupt or erratic changes. It works best on processes with large time
constants, low noise, and good measurement repeatability and resolution
so that the response of the controlled variable is smooth.

Flow and liquid pressure loops normally use proportional-integral (PI)

controllers because the loops response is too fast and abrupt for derivative
action. Level and gas pressure loops can use a high controller gain for tight
control. In most cases, the upper limit on controller gain depends upon the
degree of measurement noise and on how much variability should be
passed on to the manipulated flow. Derivative action is undesirable except
for those unusual level loops with large time lags because the combination
of a high gain and derivative action greatly amplifies noise. For level con-
trol of surge volumes, derivative action would be counterproductive as it
would pass on more variability from the level to the manipulated flow.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Figure 2-24a. Response of a Proportional-plus-Derivative Controller to a 10%

Set point Change

Figure 2-24b. Response of a Proportional-plus-Derivative Controller to a 40%

Load Upset

Temperature and composition loops are the prime candidates for PID con-
trollers.
--`,```,,,```,`,````,``,`

Copyright International Society of Automation

Nearly all temperature controllers should use PID controllers because the
derivative mode provides a phase lead that compensates for the phase lag
of the thermal lags in the thermowell and process. Wherever the process
variable can accelerate, whether due to positive feedback or nonlinearity,
derivative is helpful since it reacts to a change in the rate change. In the
distillation-column example, the control point is on the knee of a plot of
tray temperature versus distillate-to-feed ratio. An increase in feed can
cause the drop in temperature to accelerate on the steep slope. However, if
thermocouple or RTD input cards are used instead of a smart transmitter,
the steps from hitting the resolution limit seriously reduce the amount of
derivative action that can be used. A temporary fix is to add a filter that
smooths out the steps. If there is an inverse response, where the initial
response is opposite to the final response, the derivative mode cannot be
used. This can occur in furnace temperature control where the controller
output is the firing demand that works within a cross limit to make air
lead fuel on a load increase.

Composition control loops on back-mixed volumes should use PID con-

trollers because the derivative mode provides anticipatory action impor-
tant for the slow response. In the pH control example, derivative action
should be used in the pH loop on the neutralizer but not on the static
mixer. However, many analyzers have an erratic response and analyzers
with an analysis cycle time, such as chromatographs, have staircase
responses that preclude the use of derivative action.

For tight control of processes with a time constant much larger than the
dead time, there is benefit in aggressive preemptive and anticipatory
action. Thus, these processes should maximize the gain and derivative set-
ting and overdrive the output to reduce the rise and return time. The inte-
gral time is increased (reset action decreased) since it has no sense of
direction and increases overshoot. If the controller gain is larger than 5,
there is enough muscle from the proportional mode, and the derivative
setting can be small or zero. Derivative mode is a necessity regardless of
gain setting for a process whose control variable can significantly acceler-
ate, whether due to a nonlinearity (pseudo-runaway) or positive feedback
(real runaway), such as a polymerization reactor or a fermentor in the
exponential growth phase. For these processes, the integral time setting is
increased to about 10 times the ultimate period so that gain and rate action
dominate the response.

Conversely, for tight control of processes with a time delay much larger
than the largest time constant, gain and rate action must be minimized
and integral (reset) action maximized, which means the integral (reset)
time must be minimized. The integral mode adds smoothness not inherent

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

in the process. The integral time factor is decreased and can be as small as
1/8 of the ultimate period for a pure dead time process.

Figures 2-25a through 2-25f show the effect of the three mode settings on a
PID controller for a process with a time lag 4 times larger than the loop
time delay. Note that a gain setting too large increases overshoot due to
the presence of reset action. An integral (reset) time setting that is too
small causes a greater overshoot and increases the period of oscillation.
Too much derivative action causes an oscillatory response and a shorter
period but no real overshoot. Figures 2-26a through 2-26d show the effect
of two mode settings on a PID controller for a process with a time lag 4
times smaller than the loop time delay. The doubling of the gain setting is
much more disruptive.
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Base Case Gain Halved

Gain Doubled

Figure 2-25a. Effect of Gain on Set point Response of PID for a Large Time
Lag–to–Time Delay Ratio

A starting point is needed on tuning settings, particularly for new plants

or existing plants with upgraded instrumentation and control valves. In
Table 2-5, the first number outside the parentheses provides a tuning, scan
time, and filter time setting suitable for a download. Within the parenthe-
ses is a range of typical settings [2.5]. An auto tuner should be run as soon
as the plant is up to normal operating conditions. If the calculated settings
are outside the range noted in Table 2-5, the auto tuner test should be run
again. If the results are still outside of the range, it means either that it is an
unusual loop or that there is a problem with the measurement or control

Copyright International Society of Automation

Gain Halved

Base Case

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Gain Doubled

Figure 2-25b. Effect of Gain on 40% Load Upset to PID for a Large Time Lag–
to–Time Delay Ratio

Integral Time (Ti) Doubled

Base Case Integral Time (Ti) Halved

Figure 2-25c. Effect of Reset on Set point Response of PID for a Large Time
Lag–to–Time Delay Ratio

valve. Whenever a loop is commissioned or tuned, it should be closely

watched for several days for a variety of operating conditions to make
sure the loop is stable and the performance is acceptable.

Copyright International Society of Automation

Integral Time (Ti) Doubled

Base Case Integral Time (Ti) Halved

Figure 2-25d. Effect of Reset on 40% Load Upset to PID for a Large Time Lag–
to–Time Delay Ratio
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Derivative Time (Td) Halved

Base Case

Derivative Time (Td) Doubled

Figure 2-25e. Effect of Rate on Set point Response of PID for a Large Time
Lag–to–Time Delay Ratio

Some loops will fail during an auto tuner pretest because the loop
response is too small or too large within the allowable time frame of the
pretest, or because the valve has too much stick-slip. Also, fast runaway

Copyright International Society of Automation

Derivative Time (Td) Doubled

Base Case

Derivative Time (Td) Halved

Figure 2-25f. Effect of Rate on 40% Load Upset to PID for a Large Time Lag–to-
Time Delay Ratio

Base Case

Gain Doubled
Gain Halved

Figure 2-26a. Effect of Gain on Set point Response of PID for a Small Time
Lag–to–Time Delay Ratio

and integrating loops cannot be safely taken out of the auto mode. For
these loops, a closed-loop tuning method is best because it is the fastest
method for a large time constant, it keeps the controller in service with

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Gain Halved

Base Case

Gain Doubled

Figure 2-26b. Effect of Gain on a 20% Load Upset to PID for a Small Time Lag–
to–Time Delay Ratio
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Base Case

Integral Time (Ti) Halved

Integral Time (Ti) Doubled

Figure 2-26c. Effect of Reset on Set point Response of PID for a Small Time
Lag–to–Time Delay Ratio

maximum gain, (safest for processes that can get into trouble quickly or
have a non-self-regulating response), and it includes the effect of poor
valve response in the tuning [2.29].

Copyright International Society of Automation

Integral Time (Ti) Doubled

Base Case Integral Time (Ti) Halved

Figure 2-26d. Effect of Reset on a 20% Load Upset to PID for a Small Time
Lag–to–Time Delay Ratio

Table 2-5. Typical Tuning Settings [2.5]

Application Type Scan Gain Reset Rate Method

(seconds) (seconds) (seconds)
Liquid Flow/Press 1 (0.2-2) 0.3 (0.2-0.8) 6 (1-12) 0 (0-2) λ

Tight Liquid Level 5 (1.0-30) 5.0 (0.5-25)* 600 (120-6000) 0 (0-60) CLM

Gas Pressure (psig) 0.2 (0.02-1) 5.0 (0.5-20) 300 (60-600) 3 (0-30) CLM

Reactor pH 2 (1.0-5) 1.0 (0.001-50) 120 (60-600) 30 ( 6-30) SCM

Neutralizer pH 2 (1.0-5) 0.1 (0.001-10) 300 (60-600) 70 (6-120) SCM

Inline pH 1 (0.2-2) 0.2 (0.1-0.3) 30 (15-60) 0 (0-3) λ

Reactor Temperature 5 (2.0-15) 5.0 (1.0-15) 300 (300-3000) 70 (30-300) CLM

Inline Temperature 2 (1.0-5) 0.5 (0.2-2.0) 60 (12-120) 12 (12-60) λ

Column Temperature 10 (2.0-30) 0.5 (0.1-10) 300 (300-3000) 70 (30-600) SCM

* An error square algorithm or gain scheduling should be used for gains < 5
Methods: λ - Lambda, CLM - Closed-loop Method, SCM - Shortcut Method

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Table 2-6. Closed-loop Tuning Method

1. Put the controller in automatic at its normal set point. If it is important not to make
big changes in the manipulated variable (analog output), narrow the controller out-
put limits.
2. Decrease reset action (increase reset time) by a factor of 10 if possible and trend
record the process variable (PV) and controller output (CO).
3. Add a PV filter to keep output fluctuations from noise within the dead band of the
control valve. Set the scan time per Table 2-5 based on the loop type.
4. Bump the set point in both directions and increase controller gain if necessary to
get a slight oscillation. Note the time delay between the bump and first excursion
out of the noise band.
5. Stop when loop has about a quarter amplitude oscillation or the gain has
reached your comfort limit, and note the period. For gain settings greater
than one, the oscillation will be more recognizable in the controller output
(CO). Make sure CO stays on scale within the valve’s good throttle range.
6. Reduce the gain until the oscillation just disappears so that recovery is as smooth
as desired and the overshoot and settling time are acceptable.
7. For a temperature loop with a smooth response (no chatter,
inverse, square wave, or interaction) , use rate action. If the gain is larger than 10,
reset and rate action are not needed. If the manipulated flow will upset other loops,
decrease the gain or use error squared control. If a high gain is used, add set point
velocity limits, and configure the set point to track the PV in manual and remote out-
put (ROUT) so the loop can restart.
8. If rate is used, add set point velocity limits, set the rate equal to 1/10 of period, and
make sure noise in the output does not cause the valve to dither. Make another set
point change and adjust the gain to get a smooth response.
9. Use Equations 2-3 and 2-4 to calculate the integral (reset) time setting from the
noted time delay and period of oscillation for quarter amplitude decay. After entering
the integral time, make another set point change and make sure the rise time, over-
shoot, and settling time are acceptable.
10. If gains less than 5 are used for level, make sure the integral (reset) time is greater
than 10 minutes, and add error squared control and/or a velocity limited feedfor-
ward as defined in Appendix B.
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

11. Return the output limits to their proper values if narrowed for testing.

Set point changes are used because they are more likely to cause an oscilla-
tion than a change in the controller output: A step change in a set point is a
step change in the error seen by the controller, whereas a step change in
the controller output is smoothed out by the time constants in the loop.
After the new tuning settings are entered, the loop should be checked for a
step change in the controller output and the load rejection capability of the
new settings monitored and compared to historical data for the old set-
tings.

Copyright International Society of Automation

Equations

Tu = 0.7*To (2-3)

Ti = To / [0.1+2∗{(4 ∗ τd)/Tu}2] (2-4)

where:

Ti = integral (reset) time

To = oscillation period for quarter amplitude decay
Tu = ultimate period
τd = time delay

Equation 2-4 extends the utility of the closed loop by compensating for
large and small time delay–to–time constant ratios. It estimates an integral
(reset) time to be respectively 1/8 and 1/2 times the oscillation period for
a self-regulating process with a pure time delay and with a large time
constant respectively. [1/8 factor is for a pure time delay and ½ factor is a
large time constant] For integrating and runaway processes, Equation 2-4
yields an integral (reset) time that approaches 10 times the ultimate period
as the negative-feedback time constant becomes large compared to the
time delay or positive-feedback time constant. For pure time delays, the
Lambda tuning method provides a more accurate calculation of the tuning
settings for a desired closed-loop time constant. For very slow loops, the
shortcut tuning method, whereby the user only needs to see the time delay
and the initial rate of change of the PV, can be used to save time, as
detailed in Reference 2.4.

The Theory section shows that a self-regulating process with a pure time
delay and a large time constant will oscillate at 2 and 4 times the time
delay, respectively. See Reference 2.1 for equations to estimate how the
ultimate period increases from 4 times the time delay for integrating and
runaway processes.

Cascade Control
A cascade control system consists of a primary controller that manipulates
the set point of a secondary controller that in turn manipulates a final ele-
ment. The secondary controller can be operated in the automatic mode
with a local set point or in the cascade mode with a remote set point.
Industrial systems are designed to make sure the remote set point of the
secondary controller (output of the primary controller) is equal to the local
set point when the secondary controller is switched from the automatic to
cascade mode so that there is a bumpless transfer.

For optimum performance, the time delay and time constant for the sec-
ondary controller should be 5 times faster than the respective values for

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

the open loop response of the primary controller. If this is not possible and
the periods start to approach each other in value, the loops will fight and
may resonate. The interaction can be reduced by artificially slowing down
the primary controller by an increase in its process variable filter and scan
time or by a decrease in its gain. Unfortunately, this also slows down the
ability of the primary controller to react to load upsets that originate in the
primary loop. It may still be worthwhile to go to cascade control, however,
because most of the upsets are in the secondary loop and/or the lineariza-
tion of the primary loop is beneficial.

If the time delay of the secondary (inner) loop is small, its ultimate period
is small and any oscillation is effectively attenuated, per Equation 3-2, by
the primary (outer) loop time constant. If the inner open loop time con-
stant is large compared to its time delay, the secondary controller gain can
be relatively high, which gives this controller muscle to rapidly correct for
inner loop upsets. Also, the maximum excursion (peak error) for the inner
loop is reduced. Furthermore, by going to cascade control, one of the two
largest time constants that created equivalent time delay and was detri-
mental to performance of the single loop is now a beneficial term for the
inner loop [2.30]. In this scenario, the inner loop can correct for inner loop
upsets before they are even seen by the outer loop, so it doesn’t matter that
the primary controller may need to be detuned if the inner loop time con-
stant is not much smaller than the outer loop time constant. The plot of
simulation results in Figure 2-27 shows how the ratio of the peak error for
a cascade loop to the peak error for a single loop decreases as the inner-to-
outer loop ratio of time delays decreases and the ratio of time constants
increases [2.30]. The results shown in the figure are for self-regulating
inner and outer loops. The improvement is greater for integrating and
runaway loops.

One of the most beneficial uses of cascade control is in reactor control, as

shown in Figure 2-28, where the reactor temperature primary controller
output is the remote set point of a secondary coolant temperature control-
ler. The secondary loop can correct for changes in coolant temperature or
pressure or changes in the heat-transfer coefficient of the coils or jacket.
Also, the steady-state gains for the primary loop are now constant and the
secondary loop takes responsibility for the nonlinearity of the process and
control valve gains for the coolant system.

An important benefit of cascade control is the linearization of the manipu-

lated variable, which facilitates the addition of flow feedforward. The tem-
perature and composition control of distillation columns has been greatly
improved by ratioing reflux, distillate, or steam flow to the feed flow.
Often the column is started up on ratio control until the column reaches
operating conditions and the temperature sensors are completely

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Ratio of Inner to Outer Loop Time Lags

(τi / τo)
Ratio of Inner to Outer Loop Time Delays
Curve 1 : τdi / τdo = 0.6
Curve 2 : τdi / τdo = 1.0

Tuning and Control Loop Performance: A Practitioner’s Guide, 3rd Edition, ISA, 1994, pp. 241, Figure 11.2

Figure 2-27. Reduction in Peak Error for Inner Loop Upsets by Cascade Con-
trol

Secondary (Inner) Primary (Outer)

Loop Loop
TC RSP TC
2-1 2-2

TT
TT
2-2
2-1

Cooling Tower
Cooling Tower
Makeup
Reactor Return

Product

Figure 2-28. Reactor Temperature to Coolant Temperature Cascade Control

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

immersed and representative of tray composition. At this point, the tem-

perature controller goes on auto and corrects the feedforward. If the tem-
perature-to-flow cascade were not used and the feedforward signal went
directly to a control valve, the nonlinearity of the control valve would
need to be included in the ratio or feedforward gain by a calculation of
flow versus valve position. It is difficult to compute the actual flow
through the control valve because the pressure drop available to the valve

Copyright International Society of Automation

changes and the lack of straight runs results in poor repeatability of any
calculation. While it is true that the nonlinearity of an equal per cent flow
characteristic compensates for the inverse relationship between process
gain and feed flow for a temperature loop, this compensation is far from
exact and can be better done by output signal characterization or a feed-
forward multiplier instead of a summer.

Cascade control can do more harm than good if the signal-to-noise ratio,
rangeability, or reliability of the secondary measurement is poor or the sec-
ondary loop is actually slower than the primary loop. This is the case for a
static mixer pH-to-reagent flow cascade loop, particularly when the
reagent flow measurement is an orifice meter. If the reagent flow loop
used a coriolis flow meter, and its scan time was small, the stiction and
dead band of the valve was small, and the digital positioner and flow con-
troller were tuned for a fast response, it would be a different story; cascade
control might be beneficial and facilitate reagent-to-feed ratio control.

The secondary controller should be tuned for fast response. If the inner
loop is much faster than the outer loop, it is permissible for the closed-
loop response of the secondary controller to be quite oscillatory because
the amplitude of the oscillations is effectively attenuated by the time con-
stant of the outer loop. Also, an offset in the inner loop is theoretically of
little consequence to the outer loop since the inner loop only exists for the
purpose of the outer loop. This implies that the secondary controller
should use mostly gain and rate action. This is true for valve positioners.
These are secondary controllers whose remote set point (desired valve
position) is the output of the primary process controller. Pneumatic posi-
tioners were high-gain proportional-only controllers. Digital positioners
often use some form of proportional-plus-rate algorithm. Reset is never
used in a positioner because it would cause overshoot and the offset from
proportional-only control is small due to the high gain action. The second-
ary coolant temperature controller in Figure 2-28 should be tuned with
mostly gain and rate action. However, in many situations reset is used in
the secondary controller because operators get concerned about offsets,
secondary set point limits may need to be enforced, and the ratioing of
flows needs to be exact, especially for inline blending. Reset action is used
in flow controllers to not only eliminate offsets but to also help make the
set point response of the flow loops match up for better timing.

Feedforward Control
In feedforward control, a controller output is calculated to compensate for
a measured disturbance. It provides a preemptive action to enforce a mate-
rial or energy balance. The block diagrams in Figures 2-29a and 2-29b
show the use of feedforward, with and without cascade control. The feed-
forward signal must arrive at the same point in the process simulta-

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

neously with the load upset and must be equal but opposite in sign to the
load upset. Feedforward control is never perfect and should be corrected
by feedback control if a suitable feedback measurement or estimator is
available. The total error can be approximated as the root mean square of
the errors in the feedforward measurement, feedforward gain, and feed-
forward timing. In general, the feedforward measurement goes through a
dead time block and a lead-lag block for proper timing and is finally
biased by a process controller output for feedback correction. The feedfor-
ward gain can be the gain in the lead-lag block or a separate gain. If the
controller output goes directly to a control valve, the feedforward calcula-
tion must also go through a signal characterizer that computes the valve
position for a desired flow from the installed valve characteristic. If the
valve nonlinearity is beneficial for feedback control, the characterization is
done before the bias from the feedback controller is added to the feedfor-
ward signal; otherwise it is done after the feedback correction as shown in
Figure 2-29a. It is vastly preferred that the controller output be the remote
set point of a flow controller, as depicted in Figure 2-29b, so that valve
nonlinearity, dead band, and pressure upsets are not issues.

Feed
Forward Delay Lag Gain
Delay τdff τdL τL KL
∆DV Fi
Load Upset

Delay Lag Gain Delay Lag Delay Lag Gain

Lead
Lag τLL τdv τv Kmv τdp1 τp1 τdp2 τp2 Kpv
∆MV Fr ∆FR
Valve Process ∆PV
Filter τff

∆FF
Gain Kff Σ

∆CO Local
Set Point
PID Kc Ti Td

∆CV
Delay Lag Gain Lag Delay Lag
τc2 τdc τc1 Kcv τm2 τdm2 τm1 τdm1
Lag Delay
Controller Measurement
Figure 2-29a. Feedforward Control Block Diagram

The most common form of feedforward control is flow feedforward, where

the load upset is a feed flow. The feedforward gain is simply the ratio of
the manipulated flow to the feed flow. Sometimes, the feed set point is
used instead of the feed-flow measurement to eliminate noise. The ratio is
found from steady-state simulations or calculations. In the Theory section,
the ratio of coolant flow to feed flow and bypass flow to feed flow are

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Feed
Forward Delay Lag Gain

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Delay τdff τdL τL KL
∆DV Fi
Load Upset

Delay Lag Gain Delay Lag Delay Lag Gain

Lead
Lag τLL τdv τv Kmv τdp1 τp1 τdp2 τp2 Kpv
∆MV Fr ∆FR
Valve ∆PVi ∆PVo
Process
Filter τff Delay τdmi
Remote
∆FF Set Point
Gain Kff Σ Lag τmi Inner Loop

∆COo Local ∆COi

Set Point
Kc Ti Td Kcvi
PID Kc Ti Td
∆CVi
∆CVo
PID Gain
Delay Lag Gain Lag Delay Lag
τc2 τdc τc1 Kcv τm2 τdm2 τm1 τdm1
Lag Delay
Controller Measurement
Figure 2-29b. Feedforward and Cascade Control Block Diagram

computed from an energy balance for a heat exchanger. Usually, the ratio
is entered by the operator but there is an opportunity to put the energy or
material balance calculation online if there are reliable temperature or
composition measurements.

A common feedforward control system for sheets, webs, films, and con-
veyors is speed ratio control, where the roll speed is ratioed to another roll
speed or an extruder speed. The ratio is corrected by a controller of aver-
age gauge thickness. The timing requirement is very tight: Speed ratio
control of one roll to another must be done within milliseconds.

In model predictive control, extensive online testing is done to establish

the dynamics of the feedforward signal and the correct timing is built into
the algorithm solution. When feedforward control is used in the basic con-
trol system, it is up to the user to adjust the dead time and lead-lag blocks.
The consequences of poor timing are illustrated in Figure 2-30. If the feed-
forward correction arrives too soon, feedforward creates inverse response
and does more harm than good. If it arrives too late, it creates a secondary
disturbance. If the lead-lag gain is too high, the whole response goes in the
opposite direction of the effect of the load upset.

Normally, the feedforward time delay and lag are adjusted to make sure
the feedforward doesn’t arrive too early due to a time delay or time lag in
the path of the load upset. The lead is adjusted to compensate for a time
lag in the path of the manipulated variable. The lag is increased as neces-

Copyright International Society of Automation

PV PV

perfect!
none!
uncorrected load upset
feedforward gain and timing just right

Time Time

PV PV

too late! not enough gain!

prolonged upset attenuated response

Time Time
PV PV

too much gain!

inverted response
too early!
inverse response

Time Time

Figure 2-30. Effect of Feedforward Timing and Gain

sary to make sure that noise from the feedforward measurement doesn’t
show up as dither in the final element.

The greatest benefits of feedforward control occur in loops with large time
delays and large time constants because the integrated error and the tim-
ing window are large. Distillation columns are the prime candidates, fol-
lowed by reactors, crystallizers, and evaporators. About the toughest
application in which to get the timing rate is liquid pressure control,
because the process time constant is so small. Feedforward control is
essential for relatively fast periodic upsets, although the better solution is
to eliminate the root cause, which is typically another control loop. If the
period of the disturbance is less than twice the ultimate period of the loop,
the feedback controller cannot correct for the upset within the settling time
and may amplify the upset and do more harm than good.

When feedforward control is used, there are some important best practices
to consider that are outlined in Table 2-7.

There are opportunities to adapt the feedforward signal to make it

smarter. Appendix B details the adaptation of a velocity limit for a batch-
to-continuous transition for optimum smoothing of surge tank level con-
trol. Since the feedforward gain or ratio is based on parameters that
change with time and on measurements that have errors, there is an
opportunity to home in and make corrections using an integral-only con-
troller whose controlled variable is the feedback correction and whose out-
put is a trim of the feedforward gain or ratio. The integral-only controller

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Table 2-7. Best Practices for Feedforward Control

1. Display the feedforward signal, feedforward gain, and feedback correction. For flow
feedforward, display the desired ratio and actual ratio of flows.
2. Make the feedforward gain adjustable by the operator. For flow or speed feedfor-
ward, allow the operator to set the desired ratio.
3. Allow the feedforward signal to be active in manual operation. For flow or speed
feedforward, allow the operator to start up on flow or speed ratio control or to go to

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
flow or speed ratio control for feedback measurement problems and maintenance.
4. Display any feedforward timing errors, particularly when the feedforward time delay
is too short, and automatically calculate and correct for any known transportation
delays for sheet lines and pipelines.
5. Make sure the transfer from feedforward to feedback control or vice versa is bump-
less.
6. For dead time–dominant processes, such as sheet lines and pipelines, make sure
the feedforward timing is accurate. For flow or speed ratio control, make sure the
manipulated variable response is in unison with the disturbance variable response.

can slowly optimize the feedforward gain to account for drift and
unknown parameters.

If there are unmeasured load upsets, the feedforward can ask for the
manipulated variable to go in the wrong direction. For example, if the col-
umn temperature makes a sharp turn downward below the set point due
to an unmeasured load upset, an increase in feed that would increase
reflux flow by flow feedforward would make the unmeasured upset
worse. Experienced operators will decrease the reflux flow. Just one of
these situations is enough for the operator to lose confidence in the feed-
forward. An adaptive strategy could compensate for this feedforward mis-
take by looking at trajectories of feedback and feedforward measurements.
It would be easier to do this with model predictive control because a better
feedforward trajectory is available, although a trajectory of the bias correc-
tion would need to be generated to indicate the path of the unmeasured
load upset.

Rules of Thumb
Rule 2.1. — The largest opportunity for final elements is to eliminate stick-slip
and dead band. The effect of slip is worse than stick and stick is worse than
dead band.

Rule 2.2. — The control valve with the best response is a sliding stem control
valve with a diaphragm actuator and a digital positioner. The dead band and
rangeability limits in the variable-speed drive must be relaxed. If the slid-
ing stem valve has high temperature or environment packing, the digital
positioner must be aggressively tuned. If a rotary valve must be used,

Copyright International Society of Automation

make sure it has a splined connection between the disc and actuator shaft
and short large-diameter shaft. For some extremely fast critical applica-
tions, such as polymer pipeline and incinerator pressure control, a vari-
able-speed drive is essential.

Rule 2.3. — The largest opportunity for measurements is the selection and instal-
lation of a sensor for better reproducibility, less noise, and minimal interference.
The reproducibility can be estimated as the repeatability and drift. The
need for reliability and resolution is a given and less of a problem today
than reproducibility and noise.

Rule 2.4. — The best flow measurement is the coriolis flow meter. The main lim-
itations are that the coriolis meter may not be offered in suitable materials
or size, or may not have the temperature rating needed.

Rule 2.5. — The best temperature measurement is a smart transmitter connected

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
to a 3-or 4-wire RTD sensor installed in a piping elbow. The main limitation is
that the RTD may not be suitable for very high temperatures.

Rule 2.6. — The best level measurement is a radar gauge. The main limitations
are a fluid with a dielectric constant of less than 2.0 and a tall, narrow ves-
sel.

Rule 2.7. — Check the life-cycle cost, including the cost of variability, before
choosing a less expensive control valve or measurement. The hardware cost is
generally a small part of the life-cycle cost.

Rule 2.8. — Use smart transmitters. The improved accuracy and diagnostics
are well worth the extra cost.

Rule 2.9. — Use Fieldbus for major upgrades and new installations. The
reduced cost of commissioning and wiring, the expanded diagnostics, and
improved accuracy from the elimination of A/D error are significant.

Rule 2.10. — Use a closed-loop method if an auto tuner pretest fails or is not safe.
The closed-loop method keeps the loop in auto and includes the effects of
valve stick-slip and dead band.

Rule 2.11. — For a process with a large time constant, use more gain and rate
action to overdrive the manipulated variable, to decrease rise time, peak error, and
return time. If the measurement is smooth, you can use rate action to
reduce overshoot.

Copyright International Society of Automation

Rule 2.12. — For dead time–dominant loops, significantly decrease the integral
(reset) time setting. It can be as small as 1/4 of the time delay or 1/8 of the
ultimate period for a pure dead time process.

Rule 2.13. — For integrating and non-self-regulating processes, greatly increase

the integral (reset) time. It may need to be as much as 10 times the ultimate
period for slow integrators and runaway reactors.

Rule 2.14. — Go for the largest and least expensive ways to reduce loop dead
time. The automation system is the largest source of time delay for flow,
level, and pressure loops.

Rule 2.15. — Use cascade control to correct for secondary loop disturbances
before they affect a primary loop, or to linearize the manipulated variable for feed-
forward or ratio control. If the secondary loop is not 5 times faster than the
primary loop, the scan time or filter time must be increased or the gain
decreased for the primary controller to slow down the primary loop.
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Rule 2.16. — Use feedforward control for loops with a large time delay, time lag,
or periodic disturbance, to provide a preemptive correction for load upsets. The
timing and gain must both be right and the feedforward signal must not
arrive too soon.

Theory

Process Time Constants and Gains

If the differential equations for a material or energy balance are set up in
the following form, the process gain (Kp) and time constant (τp) terms, and
the relative effects of process design and operating conditions, can be
identified for an input Xi, which is a manipulated or disturbance variable,
and an output (Y), which is the process variable.

For a negative-feedback or self-regulating process, the differential equa-

tion is: [2.1]

τp∗dY/dt = Kp∗Xi − Y (negative feedback) (2-5a)

For a positive feedback or runaway process, the differential equation is:
[2.1]

τp’∗dY/dt = Kp∗Xi + Y (positive feedback) (2-5b)

If there is no process feedback, we have an integrator gain Ki . [2.1]

Copyright International Society of Automation

dY/dt = Ki∗Xi (2-6)

The general equation for energy or material balance with generation and
consumption terms for a reaction is:

rate of accumulation = input – output + generation – consumption (2-7)

To illustrate the methodology, the differential equations will be set up and

converted to the proper form for a reactor and its coolant system. The rate
of accumulation of energy in a reactor is the sensible heat of the flow in
minus the flow out of the reactor plus the heat of reaction and minus the
heat loss to the jacket [2.1] [2.20].

Mr∗Cr∗dTr/dt = Ff ∗Cf∗Tf − Fo∗Cr∗Tr + (dQr /dTr)∗Tr − U∗A∗(Tr −Tj) (2-8)

If we combine terms, we can isolate the coefficient of the process feedback.

Mr∗Cr∗dTr/dt = Ff ∗Cf∗Tf − ( Fo∗Cr + U∗A − dQr /dTr)∗Tr + U∗A∗Tj (2-9)

If we divide through by the coefficient of Tr we get the process time

constant and gain for reactor temperature control.

τp = (Mr∗Cr) / ( Fo∗Cr + U∗A − dQr /dTr) (2-10a)

If dQr /dTr < Fo∗Cr + U∗A, and dQr /dTr > Fo∗Cr + U∗A, then we have a
negative-feedback time constant and positive-feedback time constant,
respectively. The time constant is not constant but is proportional to
reaction mass and inversely proportional to the outlet flow and the
product of the heat transfer coefficient and area for the jacket.

For a batch reactor (no outlet flow) with a negligible heat release (dQr /dTr
= 0), we have the general form of the equation to approximate the thermal
time lag of a closed volume. It can also be used for a thermowell by substi-
tution of the proper parameter values.

τp = (Mr∗Cr) / (U∗A) (2-10b)

The process gain depends upon the input under consideration. If the
manipulated or disturbance variable is feed flow:
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Kp = (Cf∗Tf ) / ( Fo∗Cr + U∗A − dQr /dTr) (2-11)

If the manipulated or disturbance variable is jacket temperature:

Kp = (U∗A) / ( Fo∗Cr + U∗A − dQr /dTr) (2-12)

Copyright International Society of Automation

In either case, we see that the process gain is not constant and is inversely
proportional to the outlet flow and the product of the heat transfer
coefficient and area for the jacket.

For a jacket coolant system with recirculation where there is equal volume
displacement of coolant return flow by coolant makeup flow, so that jacket
flow is constant, we have the following equation for the temperature of a
mixture of makeup and recirculation flow:

Tm = To − (To − Tc) ∗ (Fc / Fj) (2-13)

The process gain (Kp) for the control of the jacket inlet temperature (Ti) by
manipulation of coolant makeup flow (Fc) in a secondary loop is the
partial derivative dTm / dFc:

dTm / dFc = (To − Tc) / Fj (2-14)

We see that the process gain is proportional to the temperature difference

between the jacket outlet and the makeup coolant and is inversely
proportional to the jacket flow. The process time constant for this
secondary loop is negligible and the largest time constant becomes the
thermal time lag of the thermowell.

If we were to choose the outlet jacket temperature as the controlled vari-

able of the secondary loop, and the jacket is sufficiently mixed to do an
energy balance on the entire coolant mass in the jacket, we would end up
with:

Mj∗Cj∗dTj/dt = Fj∗Cj∗Ti − Fj∗Cj∗Tj + U∗A∗(Tr −Tj) (2-15)

If we combine terms, and divide through by the coefficient of the process

feedback, we can identify the process time constant and gain for jacket
temperature control.

τp = (Mj∗Cj ) / [(Fj∗Cj + U∗A) (2-16)

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Kp = [(Fj∗Cj ) / (Fj∗Cj + U∗A)] ∗ dTi / dFc (2-17)

If we substitute in Equation 2-14, we end up with the process gain

proportional to the temperature difference and inversely proportional to
the jacket flow.

Kp = [(To − Tc)∗Cj ] / (Fj∗Cj + U∗A) (2-18)

Copyright International Society of Automation

For a total mass balance, the rate of accumulation of mass is equal to the
mass flow in minus the mass flow out:

ρ∗Ar∗dL/dt = Fi - Fo (2-19)
Here we see that the integrator gain (Ki) for the level response is inversely
proportional to the product of the fluid density (ρ) and the cross sectional
area of the reactor (Ar).

To derive the material balance for component A for a first-order reaction,

the rate of accumulation of the mass of component A is equal to the mass
flow in minus the mass flow out of component A minus the consumption
of the mass of component A by the reaction. The concentration term (CA) is
the mass fraction of component A.

Mr∗dCAo/dt = CAf∗Ff − CAo∗Fo − Mr∗k∗CAo (2-20)

If we combine terms, and divide through by the coefficient of the process

feedback, we can identify the process time constant and gain for reactor
concentration control.

τp = Mr / (Fo + Mr∗k) (2-21)

Kp = CAf / (Fo + Mr∗k) (2-22)
Similar to what we saw for reactor temperature control, both the process
time constant and gain are inversely proportional to the outlet flow and
the reaction rate.

Process Time Delay

For sheet lines, webs, fibers, and conveyors, the time delay is simply the
distance between the manipulated variable and the process variable
divided by the speed. For example, the transportation delay for average
thickness control of a sheet is the distance between the die bolts and the
thickness sensor divided by the sheet speed.

For plug flow, the entire residence time is a transportation delay. The time
delay is the volume divided by the flow. The flow in pipelines, sample
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

lines, static mixers, coils, and heat exchanger tubes can be considered to be
essentially plug flow.

For perfect mixing, the entire residence time is a process time constant. In
well-mixed volumes of proper geometry and with baffles, the portion of
the residence time that shows up as time delay can be estimated as half the

Copyright International Society of Automation

turnover time, as shown in Equations 2-23 and 2-24, and most of the resi-
dence time becomes a process time constant [2.14].

τd = (0.5∗ρ∗V) / (Fo + Fa) (2-23)

τp = [(ρ∗V) / Fo] − τd (2-24)

Lastly, time constants in series create time delay. When the flow of
material or energy can reverse direction depending upon the sign of the
driving force, the time constants are interactive. Conductive heat transfer,
gas flow in pipelines, and the tray response in columns all have interactive
time constants. As the number of equal interactive and non-interactive
time constants in series increases, the time delay increases, from 0.02 to
0.16 and 0.14 to 0.88 times the sum of the time constants respectively. The
portion of the sum not converted to time delay is the process time constant
for a first order–plus–dead time approximation. Thus, interactive time
constants do not create much time delay and the time delay–to–time
constant ratio is always rather nice. A large number of non-interactive
time constants creates an extremely difficult to control dead-time-
dominant system.

Ultimate Gain and Period

The ultimate gain (Ku) is the controller gain that causes equal amplitude
oscillations. It is equal to the inverse of the product of the open loop
(static) gain (Ko) and the amplitude ratio at a phase shift of -180 degrees
(AR-180) [2.20].

1
K u = -----------------------------
- (2-25)
K o * AR –180

The amplitude ratio for a sine wave applied to a self-regulating process

with an open loop negative feedback time constant (το) is: [2.20]

1
AR –180 = ----------------------------------------------
2
- (2-26)
[ 1 + ( τo * ωn ) ]1 ⁄ 2

If Equation 2-26 is substituted into Equation 2-25, we have the ultimate

gain as function of the open loop negative feedback time constant (το) and
the natural frequency (ωn).

2
[ 1 + ( τo * ωn ) ]1 ⁄ 2
K u = ----------------------------------------------- (2-27)
Ko

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Since the natural frequency in radians per minute (ωn) is 2π divided by the
ultimate period (Tu), we can express the ultimate gain (Ku) as a function of
the ultimate period.

2
[ 1 + { ( τo * 2 * π ) ⁄ Tu } ]1 ⁄ 2
K u = --------------------------------------------------------------------- (2-28)
Ko

For a time constant much larger than the time delay (τ >> τd), the ultimate
gain is:

2*π*τ
K u = ------------------------o (2-29)
Ko * Tu

Since for this case the ultimate period is about 4 times the time delay (Tu ≅
4 ∗ τd), the ultimate gain can be simplified to a ratio of the time constant to
time delay.

τo
K u = 1.6 * ----------------- (2-30)
Ko * τd

Since the controller gain is a factor of the ultimate gain (Kc = 0.25∗Ku), the
controller gain is proportional to the time constant and inversely
proportional to the time delay and the open loop gain.

τo
K c = 0.4 * ----------------- (2-31)
Ko * τd

If the time delay is much larger than the time constant (τd >>το), it can be
shown that Equation 2-27 reduces to the ultimate gain being the inverse of
the open loop gain. This relationship can also be realized from the
amplitude ratio being 1 for a pure time delay.

1
K c = 0.25 * ------ (2-32)
Ko

The phase shift (f) from a time delay is: [2.20]

φ = -360 ∗ fn∗ τd (2-33)

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

If we substitute in the relationship between natural frequency in cycles per

minute and the ultimate period (fn = 1/Tu), we have:

Tu = (-360/φ) ∗ τd (2-34)

For a time constant much larger than the time delay there is a -90 phase
shift from the time constant, which leaves only -90 phase shift (φ) needed
from the time delay to reach the -180 total phase shift; the ultimate period
becomes simply 4 times the time delay.

For το >> τd:

Tu = 4 ∗ τd (2-35)

If, on the other hand, the time constant is so much smaller than the time
delay that essentially all -180 phase shift (φ) comes from the time delay, the
ultimate period approaches 2 times the time delay.

For τd >>το:
Tu = 2 ∗ τd (2-36)

The following curve fit shows how the ultimate period changes from a
multiple of 2 to 4 of the time delay and as a function of the relative sizes of
the time constant and time delay.

0.65
τo
T u = 2 * 1 + ---------------- * τd (2-37)
τo + τd

Peak and Integrated Error

Since a controller can neither sense nor compensate for an upset until after
one dead time, the minimum peak error (Ex) is the exponential response
after one dead time.

τ /τ )
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Ex = [1 – e(− d ο ]∗Eo (2-38)

If the time delay is less than the time constant, we can simplify the
relationship.

For τd < το:

τd
E x = ---------------- * E o (2-39)
τo + τd

Copyright International Society of Automation

If the time delay is much less than the time constant, we end up with the
ratio of the time delay to the time constant.

For τd < < το:

τd
E x = ----- * E o (2-40)
τo

The minimum integrated error (Ei) is approximately the peak error (Ex)
multiplied by the time delay and is thus proportional to the time delay
squared.

For τd < < τo:

τd
E i = ----- * τ d * E o (2-41)
τo

If we use Equation 2-31, we see the peak error (Ex) is inversely

proportional to the controller gain (Kc):

For τd < < το:

1
E x = ------ * E o (2-42)
Kc

If we further realize that the integral time (Ti) is a factor of the ultimate
period that is a multiple of the time delay, we have a relationship where
the integrated error (Ei) is proportional to the integral time and inversely
proportional to the controller gain.

For τd < < το:

1
E i = ------ * T i * E o (2-43)
Kc

For dead time–dominant loops, the peak error approaches the open loop
error (Eo), and the integrated error approaches the product of the open
loop error and the integral time.

For τd >>το:

Ei = Ti∗Eo (2-44)

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Feedforward Control
The equations for feedforward control are derived from the steady-state
material or energy balance at the point of control. For a heat exchanger
where the bypass is throttled for temperature control, the feedforward
equation for the hot bypass flow (Fhb) is: [2.19]

Fhb = [ (Tsp − Thi ) / (Thi − Tho ) ] ∗ (Fhb + Fho ) (2-45)

Similarly, a steady state energy balance can be used to develop the

following feedforward equation for the cold inlet flow (Fci) for a heat
exchanger without a bypass. [2.19]

Fci = [ Fho ∗ Ch ∗ (Thi − Tsp ) ] / [Cc * (Tci − Tco )] (2-46)

Dead Time from Valve Dead Band

The time delay (τdv) from a controller output that must transverse the
dead band after a change in direction is simply the valve dead band (DB)
divided by the rate of change of the controller output (∆CO / ∆t): [2.19]

DB
τ dv = ----------------------- (2-47)
∆CO ⁄ ∆t

The rate of change of controller output depends upon the controller tuning
and the error. If we consider the effect of just controller gain (Kc) for a loop
dominated by a large time lag so that the amount of reset action used is
small [2.19]:

∆CO / ∆t = Kc ∗ (∆CV / ∆t ) (2-48)

If we use Equation 2-31 for the controller gain with a detuning factor (Kx)
and realize that the rate of change of the controlled variable (∆CV / ∆t ) is
simply the pseudo-integrator gain (Ki = Ko / τo) for the large open loop
time lag (τo) multiplied by the change in the actual valve position (∆AVP),
we have: [2.19]

Kx
∆CO ⁄ ∆t = ------------------- * K i * ∆AVP (2-49)
K i * τ do
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

If we cancel out the integrator gains and approximate the change in actual
valve position on the average as the controller output minus one half of
the dead band (∆AVP = ∆CO – DB), we end up with an expression to
estimate the valve time delay from the valve dead band and the observed
loop time delay for a step change in controller output. [2.19]

Copyright International Society of Automation

DB
τ dv = ---------------------------------------------------- * τ do (2-50)
K x * ( ∆CO – DB ⁄ 2 )

Nomenclature
A = heat transfer area (ft2)
Ar = cross sectional area of reactor (ft2)
AVP = actual valve position (%)
AR180 = amplitude ratio at a phase shift of -180 degrees
CAo = mass fraction of component A in reactor outlet
CAf = mass fraction of component A in reactor feed
Cc = heat capacity of cold fluid (btu/lb*oF)
Cf = heat capacity of feed to reactor (btu/lb*oF)
Ch = heat capacity of hot fluid (btu/lb*oF)
Cr = heat capacity of liquid in reactor (btu/lb*oF)
Cj = heat capacity of liquid in jacket (btu/lb*oF)
CO = controller output (%)
CV = controlled variable (%)
DB = dead band (%)
Ei = integrated error (e.u.)
Eo = open loop error (e.u.)
Ex = peak error (e.u.)
fn = natural frequency (cycles/hr)
Fa = agitator pumping rate (lb/hr)
Fc = coolant makeup flow (lb/hr)
Fci = cold fluid inlet flow to exchanger (lb/hr)
Ff = feed flow to reactor (lb/hr)
Fhi = hot fluid inlet flow to exchanger (lb/hr)
Fho = hot fluid outlet flow from exchanger (lb/hr)
Fhb = hot fluid bypass flow around exchanger (lb/hr)
Fj = jacket coolant flow (lb/hr)
Fo = outlet flow from reactor (lb/hr)
k = reaction rate constant (btu/lb*hr)
Kc = controller gain
Kp = process gain (e.u./e.u.)
Ki = integrator gain (e.u./hr)
Ko = open loop static gain
Ku = ultimate gain
L = liquid level in reactor (ft)
Mr = mass of liquid in reactor (lbs)

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Mj = mass of liquid in jacket (lbs)

MV = manipulated variable (e.u.)
PV = process variable (e.u.)
ρ = liquid density in reactor (lb/ ft3)
φ = phase shift from time delay (degrees)
τd = time delay (hr)
τdm = measurement time delay (hr)
τdo = observed time delay from a step test (hr)
τdp = process time delay (hr)
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

τdv = valve time delay (hr)

τp = process time constant (hr)
τo = open loop time constant (hr)
ωn = natural frequency (radians/hr)
Tr = temperature of liquid in reactor (oF)
Tc = temperature of coolant supply (oF)
Tci = temperature of cold fluid at inlet to exchanger (oF)
Tco = temperature of cold fluid at outlet of exchanger (oF)
Tf = temperature of feed to reactor (oF)
Thi = temperature of hot fluid at inlet to exchanger (oF)
Tho = temperature of hot fluid at outlet of exchanger (oF)
Tj = temperature of liquid in jacket (oF)
Ti = integral time (hr/repeat)
Tm = temperature of coolant mixture at jacket inlet (oF)
To = temperature of coolant at jacket outlet (oF)
Tsp = temperature set point of hot fluid from the exchanger (oF)
Tu = ultimate period (hr)
Qr = heat release from reaction (btu/hr)
U = heat-transfer coefficient (btu/ oF*ft2)
V = volume of reactor liquid (ft3)
Xi = process input i (manipulated or disturbance variable) (e.u.)
Y = process output (process variable) (e.u.)

References
1. McMillan, Gregory K., Tuning and Control Loop Performance, 3rd Edition, 1994,
ISA.
2. Ruel, Michael, “Stiction: The Hidden Menace,” Control, November 2000, pp.
71-76.
3. Ruel, Michael, “Control Valve Health Certificate,” Chemical Engineering,
November 2001, pp 62-65.

Copyright International Society of Automation

4. McMillan, Gregory, “Valve Response Control, Improvement,” Encyclopedia of

Chemical Processing and Design, Vol. 61, 1997, pp 99-113.
5. McMillan, Gregory, Good Tuning: A Pocket Guide, ISA, 2000.
6. McMillan, Gregory, “A New Era for Model Predictive Control,” Control, May
2000, pp 45-50.
7. McMillan, Gregory, Mertz, Glenn E., and Trevathan, Vernon L., “Trouble Free
Instrumentation,” Chemical Engineering, November 1998, pp. 82-94.
8. Shinskey, Francis G., “Process Control: As Taught versus As Practiced,”
Advances in Instrumentation and Control, November 2001.
9. Coughran, Mark, T., “Valves: Testing for Peak Performance” InTech,
November 1994, pp 58-61.
10. McMillan, Gregory K. and Weiner, Stanley, et. al., How to Become an Instrument
Engineer – Part 1.523, ISA, 1994.
11. Mamzic, C.L., “Improving the Dynamics of Pneumatic Positioners,” ISA
Journal, 1958, pp 38-43.
12. Arant, James B., “Control Valve Notes and Positioner Myths,” Control,
November 1993 pp 34-40.
13. Ciancarelli, Joseph D., “Sanitary Control Valves for the Biotech Industry,”
InTech, 1991, May.
14. McMillan, Gregory K., pH Measurement and Control, 2nd Edition, ISA, 1994.
15. McMillan, Gregory K., “Pressure Control; Without Dead Time I might be Out
of a Job,” A Funny Thing Happened on the Way to the Control Room, ISA, 1989.
16. Battikha, N.E., The Condensed Handbook of Measurement, ISA, 1997, pp 27-31.
17. Vujicic, Michael, and Ortengren, Mike, “Laser on Level,” InTech, March 2000,
pp 56-58.
18. McMillan, Gregory K., editor, Process/Industrial Instruments and Controls
Handbook, 5th edition, McGraw Hill, 1999, pp 6.23-6.55 and 10.107-10.134.
19. McMillan, Gregory K., Advanced Temperature Measurement and Control, ISA,
1999.
20. Hariott, Peter, Process Control, McGraw Hill, 1964.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Practice

Overview
The objectives for this chapter are to help the user identify advanced pro-
cess control (APC) opportunities, estimate benefits, select the best technol-
ogy, sustain the solution, and track the benefits. Industry is driven by cost
and benefits analysis and is generally not interested in a great technology
looking for an application. This chapter combines the methods that have
been used extensively to identify opportunities for APC in the chemical
and the pulp and paper industries [3.1] [3.2] [3.3] [3.4] [3.5].

There is an interaction between steady-state design and controllability. As

the thermodynamic efficiency is increased, the controllability is decreased
[3.6]. The driving forces, such as pressure drops and temperature differ-
ences, available for manipulation by the control system are minimized.
The result is an increased variability and an optimum that is more difficult
to hold. Also, the existing plants in the petroleum and chemical industry
are pushed to run at twice the original plant nameplate capacity. The result
is shifting bottlenecks and unforeseen equipment constraints and operat-
ing conditions.

Yet, process design as taught and practiced focuses on steady-state opera-

tion. Engineering companies design exclusively for the steady state and
have essentially no experience with variability or dynamics. The emphasis
on minimizing changes in design to minimize risk and maintain the bud-
get and schedule often results in a plant built just like the last one. Not sur-
prisingly, plant variability audits show that most plants have the same
problems. On top of this, people in the plants are resistant to change [3.5].
89

Copyright International Society of Automation

The strongest advocates of maintaining the status quo are often the most
experienced people in operations. For batch operations this corresponds to
fixed feed rates, hold times, and valve positions.

Process control deals with change. If there were no variations in raw mate-
rials, utilities, process and ambient conditions, equipment performance,
production rates, or desired operating points, there would be little need
for control loops. In actual plant operations nothing is completely con-
stant. Thus, most of the opportunities for process control deal with
change.

A control loop moves variability in a process variable and a set point to a

manipulated variable. Advanced control also initiates change to optimize
plant operation. The knowledge of the root cause and route of undesirable
change and the ability to transfer and generate change is essential to get
the greatest benefits from advanced control.

Capital expenditures for most plants have come under increased scrutiny.
In order to make the improvements that provide the biggest bang for the
buck, the engineer needs to know what is important despite an over-
whelming number of choices of control technologies and an explosion of
information and data. This chapter provides a perspective on how to
choose the most appropriate technology.

The time frame, pattern, and sequence of changes are important for track-
ing down the sources and assessing the impact of variability. The faster the
change, the more difficult it is for a controller to correct. However, changes
with a short period are more effectively attenuated by process variable fil-
ters and back-mixed volumes. Table 3-1 summarizes the most common
sources of change, and their relative speeds.

Changes in raw materials, and in recycle, catalyst, and equipment perfor-

mance tend to result in a shift in actual and desired operating conditions
that occurs over a period of hours to days. Changes in ambient conditions
tend to be periodic and can be moderately slow (day to night), fast (rain
storms and fronts) and very slow (season to season). Sequences, batch
operations, and parallel trains starting up and shutting down tend to
cause step changes but with a long period, whereas valve stick-slip and
on-off control cause square waves that range from fast to moderately slow.
Split range points are discontinuous operating points as one control valve
starts to open as another closes or goes wide open. The stick-slip and
change in valve gain as a controller output crosses a split range point often
initiate a limit cycle back and forth across the split range point even for the
best of control valves.
--`,```,,,```,`,````,``,`,,`,`-`-

Copyright International Society of Automation

Noise and repeatability errors in a measurement are fast and are passed on
to the manipulated variable. Analyzers are a notable exception as they are
a source of slow frequency variability from a long sample transportation
delay and processing (cycle) time.

Controllers with inappropriate tuning or feedforward timing and interac-

tions will cause oscillations with a period proportional to the loop time
delay. If the period of an upset approaches the period of a control loop, the
controller will amplify the oscillation and the control loop does more harm
than good. One of the largest and most frequent sources of variability is a
level controller on a surge volume or feed tank that passes on too much
variability to a manipulated flow instead of allowing the level to swing.

Market changes are relatively slow, but the implementation is fast. Once a
rate change is decided upon, operators tend to make large step changes in
feed set points that are not simultaneous or coordinated. Transitions are
accomplished even more quickly to minimize the cross contamination of
products. As inventories decrease, market volatility increases, and the
demand for more flexible manufacturing increases, the frequency of rate
and product changes will increase to the point where steady-state opera-
tion may be a distant memory.

Operators tend to have favorite operating points or “sweet spots.” Shift

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
changes are often a major point of disruption. Operators also tend to get
bored and actually love being in charge of inventory control where they
have to changes flow set points to keep woefully undersized surge or feed
tank volumes from overflowing or running dry. Unfortunately, the set
point changes made by operators are fast, inconsistent, impatient, and do
not take into account the effect of time delays, time lags, and interactions.

Figure 3-1 shows the pyramid of technologies for advanced control. The
base is the solid foundation of basic process control discussed in Chapter
2. The next layers of loop and process performance monitoring provide
the tools to quantify the opportunities, sustain the performance, and track
the benefits of APC. Loop monitors can identify when and where an auto
tuner needs to be run again. Rules can be added on top of the
performance-monitoring layers to provide better explanations and
automatic corrections. Expert systems can be developed to deal with
abnormal situation management, such as equipment and automation
system failures and degradation.

Next, a layer of property estimators should be added to provide faster and

more reliable measurements than online or lab analysis. A plant should
take full advantage of opportunities for estimators before the implementa-
tion of higher levels of advanced control so that the impact on composition

Copyright International Society of Automation

Table 3-1. Sources and Speed of Plant Upsets

1. Process
a. Raw Materials (slow)
b. Recycle (slow)
c. Catalyst (slow)
d. Steam (fast)
e. Coolant (fast)
2. Equipment
a. Fouling (slow)
b. Failures (fast)
3. Environmental
a. Day to night (slow)
b. Rain storms and fronts (fast)
c. Season to season (slow)
4. Market
a. Rate changes (fast)
b. Product changes (fast)
5. Operators
a. Manual operation (fast)
b. Sweet spots (fast)
c. Inventory control (fast)
6. Discrete
a. On-off control (fast)
b. Sequences (fast)
c. Batch operations (fast)
d. Startup and shutdown (fast)
e. Interlocks (fast)
7. Valves
a. Stick-slip (fast)
b. Split range (fast)
c. Failures (fast)
8. Measurements
a. Noise (fast)
b. Reproducibility (fast)
c. Failures (fast)
9. Controllers
a. Tuning (fast)
b. (b) Timing (fast)
c. Interaction (fast)
d. Failures (fast)
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

and quality can be identified. Similarly, a layer of model predictive control

should be running before the layers of optimization are added so that
moves to operating points consider the projected effects on constraints.
Constraint pushers can easily be added to perform a simple optimization
of single variables, such as the maximization of feed flow or the minimiza-
tion of a reflux or steam flow. Linear programs can be set up to find the
optimum intersection of constraints for a more general optimum.

Finally, real-time optimization models must be in place and reconciled or

adapted to match the plant to find complex optimums that are highly

Copyright International Society of Automation

dependent upon process relationships and to make sure a business solu-

tion is feasible before tactical schedulers are put online to reduce invento-
ries and automatically adjust to market demands. For maximum
sustainable benefits, the pyramid should be built from the bottom up.

RTO

LP/QP

Ramper or Pusher

Model Predictive Control

Property Estimators
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Fuzzy Logic

Auto Tuner

Abnormal Situation Management System

Process Performance Monitoring System

Loop Performance Monitoring System

Basic Process Control System

TS is tactical scheduler, RTO is real time optimizer, LP is linear program, QP is quadratic program

Figure 3-1. Pyramid of Advanced Control Technologies

The advanced control technology with the best track record to date for
increasing plant efficiency and capacity is the model predictive controller
[3.7]. The built-in knowledge of the process dynamics and constraints
from extensive plant testing, the ability to readily control the amount of
variability transferred to the manipulated variables by move suppression,
plus the ability to readily add some optimization, are the features respon-
sible for its success. The objective of any advanced control program should
be to get at least to the level of model predictive control with a constraint
pusher for a simple optimization of a single variable. It is expected that as
process modeling tools become easier and less expensive to use and main-
tain, real-time optimization will start to deliver more consistent benefits
and applications that vie for the top of the pyramid will be more common
[3.8].

Real-time optimization (RTO) technology has proven itself on a technical

basis and can deliver benefits often greater than the MPC layer. However,
many RTO applications are not delivering the promised benefits, mostly
because of the requirements for highly skilled and significant mainte-

Copyright International Society of Automation

nance, a very real challenge given today’s levels of resources in manufac-

turing plants. No matter the promise of a technology, if it cannot be
supported and maintained within the resource limitations of a plant, it
will fall into disuse and eventually be shelved. Understanding and accept-
ing the resource requirements of any technology is a key element in APC
technology selection.

Opportunity Assessment
Figures 3-2a and 3-2b show how reducing the variability in a process vari-
able associated with a constraint translates to the opportunity to move the
variable closer to the constraint. The traditional way of depicting the
improvement is to show the current peak at the optimum location for the
given degree of variability, as illustrated by Figure 3-2a. The benefits are

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
then based on how much the peak can be moved if the standard deviation
is reduced and still provide the same small number of violations of the
constraint. In practice, the peak is typically set by the operator much more
conservatively so that even if the variability is not reduced, just a better
understanding and automation of the positioning of the peak relative to
the constraint affords an additional opportunity, as shown in Figure 3-2b.
Manual set points are chosen based on opinions and war stories rather
than data. The operator is usually the largest and most active constraint to
optimum operation.

Set Point Upper Limit

PV distribution for
original control

2-Sigma 2-Sigma

Set Point

value

PV distribution for
improved control

2-Sigma 2-Sigma

Figure 3-2a. Effect of Variability on Allowable Proximity of Process to Con-

straint

The variability in a process variable must be translated to an economic

variability, such as product profit, utility cost, or raw material cost. This
can be done by calculation or by a cross-correlation analysis to identify the
change in the economic variable for a change in the process variable. The

Copyright International Society of Automation

LOCAL
Set Point Upper Limit
PV distribution for
original control

2-Sigma 2-Sigma
RCAS
Set Point

value Extra margin when

operator manually
PV distribution for adjusts set points
improved control

2-Sigma 2-Sigma

Figure 3-2b. Effect of Operator Conservatism on Proximity of Process to Con-

straint

tools and procedures for identification of this economic gain are similar to
those used for identification of the process gain for property estimators
(Chapter 8). The process must start with an online measurement or calcu-
lation of capacity, yield, and utilities to provide the economic variables. In
some cases, missing or inaccurate measurements will be found in the pro-
cess of putting the economic variables online. It also provides a head start
for true performance monitoring (Chapter 4) and real-time optimization
(Chapter 10).

Any opportunity assessment (OA) must start with an opportunity sizing,

to determine the gaps between average and peak performance for the last
most representative year of operation based on discussions with opera-
tions and a detailed review of the cost sheets, data historian, and simula-
tion results. The peak performance can be the best batch or period of
operation from the cost sheet or data historian or the best operating condi-
tions or cycle time found by simulation or Multivariate Principal Compo-
nent Analysis (PCA), as discussed in Chapter 4. The advantage of a high-
fidelity process simulation is that the model can go where the plant has
not traditionally operated to find and explore new process opportunities.
Of course, the simulation must be validated against present and proposed
plant operation. The process simulation should be put online to compare
the predicted versus the measured values of key operating variables.
Chapter 10 will show how to automatically adapt the parameters of an
online model to match plant operation.

The gap can also be quantified by a statistical approach. The fractional

reduction in the standard deviation can be approximated based on the cur-

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

rent standard deviation and the minimum possible standard deviation

from the process control improvement. This reduction in the standard
deviation is multiplied by amplitude ratios and process gain factors and
then translated to a shift in the mean to a more optimum operating point,
or a decrease in the fraction of data points that exceed constraints or occur
at less optimum conditions. The shift in the mean or the decrease in the
fraction of deviant data points is multiplied by a production rate and a
conversion and economic gain factor to establish a theoretical gap for the
opportunity sizing.

After confirmation of the opportunity sizing, a detailed review of the pro-

cess and control systems is done over a period of two days with all ques-
tions oriented toward improvements that could eliminate part of the
identified gaps. The list of improvements generated from the questions
and ensuing discussions are compiled. The OA concludes with the assign-
ment of a percentage of gap elimination to each improvement and a work
plan.

Some questions are offered to help an OA to find improvements in a basic

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
control system. Questions (1) through (20) ascertain how important it is to
reduce variability in an operating limit (constraint) and what process vari-
ables are involved. The questions generally apply to both batch and con-
tinuous unit operations. Appendix A presents the more extensive list of
questions that were used by Monsanto and Solutia for opportunity assess-
ments in the last decade in all of the major processes, OAs that led to pro-
cess control improvements worth $60 million a year in benefits. ICI has
cited 2% to 6% of operating costs as the benefits achieved through
improved process control. Besides the cost savings, ICI is convinced it has
increased the safety and reduced the environmental impact of its plants
[3.9].

1. Could the yield in a reactor be increased by operation closer to a

minimum constraint for excess reactants?
2. Could steam flow to an evaporator or a column be decreased by
operation closer to a product quality constraint?
3. Could steam or fuel flow to a dryer be decreased by operation
closer to a high moisture constraint or low differential temperature
constraint?
4. Could fuel flow to a calciner or kiln be reduced by operation closer
to a high pressure, low oxygen, or low differential temperature
constraint?

Copyright International Society of Automation

5. Could feed rates or batch sizes for reactors, evaporators, and

crystallizers be increased by operation closer to high level and
pressure constraints for vent system loading and entrainment?
6. Could feed rates or batch sizes for fermentors be increased by
operation closer to the high level constraint for entrainment and
CO2 stripping rate?
7. Could feed rates or batch cycle times for fermentors be reduced by
operation closer to the high agitation or pressure limit associated
with excessive cell fracture and CO2 solubility, respectively?
8. Could feed rates be increased (residence time decreased) or batch
cycle times reduced by operation closer to the minimum
conversion time constraint?
9. Could feed rates to hydroclones be increased by an operation
closer to the low outlet pressure limits?
10. Could feed rates or batch sizes to centrifuges be increased by
operation closer to the high basket loading limit?
11. Could feed rates to batch and continuous reactors, evaporators,
and crystallizers be increased by operation closer to a utility or
heat transfer surface constraint?
12. Could reaction rates be increased by operation closer to a high
temperature or pressure magnitude or rate-of-change constraint
for quality or safety?
13. Could purge rates and yield loss be reduced by operation closer to
the high limit on impurities or inerts?
14. Could feed rates, run times, or number of batches between the
cleanout of crystallizers be increased by operation closer to the
mixing and high differential temperature constraints for the rate of
formation of fines and coating of heat-transfer surfaces?
15. Could the reagent use to a neutralizer be reduced by operation
closer to a low or high pH constraint?
16. Could fuel be reduced to a boiler or furnace by operation closer to
a low oxygen constraint?
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

17. Could energy be reduced to compressors, refrigeration units, and

heaters by the operation of feed or temperature control valves
closer to the high position constraint associated with the loss of
valve sensitivity and range of operation?
18. Could the waste of reagent from the cross neutralization of acids
and bases, or the waste of energy from alternating use of coolants

Copyright International Society of Automation

and steam, be reduced by the minimization of the crossing of the

split range point?

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
19. Could the amount of scrap be reduced in a sheet line by operation
closer to the cross and machine direction constraints on gauge?
20. Could the roll speed be increased on a sheet line by operation
closer to the sheet moisture and extruder speed constraints?
21. Could the amount of sheet giveaway to the customer be reduced
by tighter control of average thickness and profile near the edges?
22. Can the production rates of unit operations be coordinated to take
maximum advantage of surge volumes to increase plant capacity?
23. Could a reflux flow be reduced to decrease the steam flow to a
reboiler?
24. Could a recycle flow be reduced to decrease utility use and
pressure drops and to increase residence time and surge volume
capacity?
Most reactors operate with an excess of one or more reactants because the
downside of a deficiency due to less than ideal mixing or variability in the
feeds or the composition measurement is low reaction rates and the forma-
tion of byproducts that can lead to waste treatment problems and hazard-
ous operation, besides a loss in production. The amount of excess reactant
is often based on operating practice and how many times an operator has
had actual or perceived problems from operating too close to the con-
straint.

In order to estimate the benefits from reducing feed variability, the reduc-
tion in variability in the excess component of interest must be computed
from an online analyzer, estimator, or attenuation calculation. The online
estimator must include the time constant associated with mixed volume to
show the smoothing effect. It does not need to have the time delay associ-
ated with an analyzer. The cycle time of chromatographs and lab analysis
is too slow and will alias the relatively fast reactor concentration oscilla-
tions from feed variability.

Alternatively, an attenuation calculation can be done. The instantaneous

excess at the reactor inlet based on feed rates can be calculated and the
inlet amplitude (Ai) and period (To) of the oscillation noted. The outlet
amplitude (Ao) of the oscillation can be estimated from the equation for
the amplitude ratio from the frequency response for a first-order lag. The
time constant can be estimated as the residence time of the portion of the
volume that is well mixed minus the turnover time. This calculation or an
estimator can be put online to help monitor the sustainability of any
improvements made to the basic control system.

Copyright International Society of Automation

A
AR = ------o (3-1)
Ai

1
AR = ------------------------------------- (3-2)
2 0.5
[ 1 + ( τ*ω ) ]
For τ∗ω > 1:

1 -
AR = ----------------------- (3-3)
2*π*f 0 *τ

Since fo = 1/To:

To
AR = ---------------
- (3-4)
2*π*τ
where:
Ai = amplitude of the oscillations at the inlet to the volume
(e.u.)
Ao = amplitude of the oscillations at the outlet of the volume
(e.u.)
AR = amplitude ratio
fo = frequency of oscillation (cycles/minute)
ω = frequency of oscillation (radians/minute)
To = period of oscillation (minutes)
τ = time constant for back-mixed volume (minutes)

The attenuation of variability in utilities from the volume of a unit

operation can also be estimated from Equation 3-4, although it may
require an energy balance to compute the effect of the utility on process
temperature. For processing with negligible back mixing, such as static
mixers, desuperheaters, plug flow reactors, heat exchangers, extruders,
fiber spin lines, and sheet or web lines, there is no appreciable attenuation
of variability. If the product is a solid that is not chopped or cut into small
pieces, such as a sheet or roll of yarn, variability that enters anywhere into
the process is captured in the product. For liquids, gases, and chopped
solids, the product can be blended in storage vessels. Equation 3-4 can be
used to estimate attenuation from the blending. For storage tanks, there
may be sufficient mixing from eductors or the entrance of streams that a
portion of the volume can be considered to be back mixed. If the retention
time is long enough, the contents will be mixed by diffusion (movement of
molecules from high to low concentrations and vice versa). Thus, large
storage tanks provide a significant degree of blending and attenuation of
oscillations despite the absence of an agitator.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Variability is reduced by the use of feedback and feedforward control.

Chapter 4 details the equations to calculate the best possible standard
deviation for feedback control (Sfbc) from the process capability (Scap). The
standard way of computing the process capability involves successive dif-
ferences separated by the scan time. If the scan time is small compared to
the loop dead time, this index provides an indication of uncontrollable
noise. Since feedback control cannot make a correction in less than the
total loop dead time, a more accurate indication of the process capability
would include the effect of dead time. The calculation interval could be set
equal to the time delay [3.4].

Unfortunately, the user does not usually know the time delay and it is
rarely if ever constant. The time delay can be measured online as the time
it takes for the controlled variable to get beyond the noise band whenever
there is a change in the set point or manual output. Alternatively, the dead
time can be captured whenever an auto tuner is run on the loop. For flow
and liquid pressure, the time delay can be estimated as one half of the loop
scan time and the valve time delay, which is the time between a change in
the controller output and the actual valve position as measured by a smart
positioner for a sliding stem valve. In practice to date, accurate estimates
and identification of the time delay have not normally been available. The
key point is that the calculation interval should be slowed down for the
process capability calculation to be more representative of the actual capa-
bility of feedback control.

Once the standard deviation for feedback control is calculated, Equation 3-

5 can be used to compute the shift in the mean value as equal to the frac-
tional improvement in the standard deviation multiplied by the distance

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
between the limit and the mean value. This shift provides the same distri-
bution of data points beyond the limit. In most cases, the mean value is
more than two standard deviations away from the limit so that the actual
number of data points beyond the limit is quite small.

The improvement from feedforward control for a large measured distur-

bance is usually greater than the improvement from feedback control
alone. The minimum possible standard deviation for feedforward control
(Sffc) can be estimated as the standard deviation observed when the mea-
sured disturbance is constant. It can also be estimated via Equation 3-6 as
the root mean square of the standard deviation of the feedforward mea-
surement, gain, and timing errors if the measured disturbance is by far the
largest upset. Since feedback control will work on the error left over from
feedforward control, the effects are multiplicative as shown in Equation
3-7.

Copyright International Society of Automation

The result is a much more representative standard deviation for advanced

process control (Sapc) that includes the combined improvement from both
feedback and feedforward control. Equation 3-8 computes the shift in the
mean value that is possible from advanced process control. Figure 3-3a
shows how this new set point translates to benefits as seen on a trend
recording. If the limit is associated with a manipulated variable, the
expression can be multiplied by a process gain to convert to the shift in the
controlled variable as shown in Equation 3-9. Figure 3-3b illustrates how
operation closer to a limit for a manipulated variable, such as feed rate,
that is a process input translates to benefits in production above what is
set by the operator.

Equations 3-10 and 3-11 express the mean and shift in terms of an old and
new set point. It is important to realize that these are estimates for the low-
est possible variability in the controlled variable. In some cases, it may be
undesirable to transfer all this variability from the controlled to the manip-
ulated variable. Model predictive control excels at determining how much
variability is transferred and getting the most out of both feedback and
feedforward control. Thus, the full benefit offered by these calculations is
only approached by the application of model predictive control built on a
solid foundation of good valves and measurements.

$$Savings

Limit or
Spec Target
Setpoint

Before Advanced Control

Figure 3-3a. Benefits from Operation Closer to a Controlled Variable Unit

Equations

∆CV M = [ 1 – ( S fbc ⁄ S tot ) ] * CV L – CV M (3-5)

2 2 2 1⁄2
S ffc = [ ( S ffm ) + ( S ffg ) + ( S ffd ) ] (3-6)

S apc = ( S ffc ⁄ S tot ) * S fbc (3-7)

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Production
Increase $$
Throughput at
Throughput limit I.e. maximum
possible
Operator setpoint

Value of limiting
input or constraint
Limiting
Condition Process input of
constraint limit

Time

Figure 3-3b. Benefits from Operation Closer to a Manipulated Variable Limit

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

∆CV M = [ 1 – ( S apc ⁄ S tot ) ] * CV L – CV M (3-8)

∆CV M = K p * [ 1 – ( S apc ⁄ S tot ) ] * MV L – MV M (3-9)

SP old = CV M (3-10)

SP new = ∆CV M + SP old (3-11)

where:
Kp = process gain (∆CV/∆MV)
SPnew = new set point
SPold = old set point
Sapc = standard deviation possible for advanced process control
Sfbc = standard deviation possible for feedback control
Sffc = standard deviation for feedforward control
Sffd = standard deviation of feedforward dynamics error
Sffg = standard deviation of feedforward gain error
Sffm = standard deviation of feedforward measurement error
Stot = total standard deviation
∆CVm = shift in mean value of controlled variable
CVM = mean value of controlled variable
CVL = limit (constraint) for controlled variable
MVM = mean value of manipulated variable
MVL = limit (constraint) for manipulated variable

Copyright International Society of Automation

beneficial operation closer to limits, may not be easy or feasible. For

example, for fermentors, the growth rate of the cells depends on the
availability of oxygen and the removal of carbon dioxide. An increase in
pressure can increase the partial pressure of oxygen and the availability of
oxygen for the cells, but it increases the solubility of carbon dioxide, which
also decreases the pH of the broth. An increase in mixing can reduce areas
of stratification of cells and improve liquid and gas contact, but it fractures
the cells. Thus there are many interactions between dissolved oxygen
control, pH control, and dissolved carbon dioxide control, and the
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

variability in one loop can cause excessive variability in other loops.

Normally the cells are very sensitive to their local environment and the
mixing is minimal, so yields can easily be adversely affected by variability.

The questions listed do not cover any reduction in down time or penalties
associated with exceeding equipment and environmental limits. For exam-
ple, the life of glass linings and the rupture disks depend upon the number
of temperature and pressure cycles. Rupture discs can cause reportable
emissions and a brief excursion in an effluent stream below 2 pH or above
12 pH can classify a whole volume as hazardous waste.

In some cases, the opportunities are not so obvious or directly calculated.

Consider the case of a solvent recovery distillation tower at the back end
of a process. It performs a final cleanup of a recycled solvent and often
exhibits flooding conditions. When the tower floods, it consumes a signifi-
cant amount of the operator’s attention and it can take several hours to
unload the tower. Efforts were often made at justifying an APC applica-
tion on the tower, but the economics associated with simply the variability
in the recycled solvent would not merit the expenditure. A second look at
the data across the whole plant, as opposed to only the solvent tower,
clearly showed that when the tower was flooded, the operator was
focused on the flooding problems and not maximizing the feed rate and
reactor yield to the plant. Since the plant was in a sold-out position, free-
ing up the operator to maximize the primary economic drivers delivered
significant benefits and easily justified the APC application on the solvent
tower, with a payback on the order of a few weeks. When computing the
benefits of an application, consider the impact of the application on other
parts of the plant—good or bad.

Examples

Global Production Maximum

Many examples of a process parameter that has a global production maxi-
mum can often be found in a plant. For example, in a plant that is produc-
ing ammonia, the inlet temperature to the synthesis converter has a major
impact on plant production for a given feedstock input. As illustrated in

Copyright International Society of Automation

Figure 3-4, any variation in this temperature from best operating point will
result in less production at a given feedstock input. The slope is the pro-
cess gain used to calculate the benefits as you approach the optimum
operating point.

3500

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
3400 Mean Variation Production
(DegC) (DegC) Loss(%)

Ammonia 415 0 0
Production 3300 415 25 0.103
(mol/hr) 415 50 0.415
460 25 1.102
460 50 1.412
3200

3100
350 400 450 500

Bed #1 Inlet
Temperature (Deg C)

Figure 3-4. Impact of Synthesis Bed Temperature on Ammonia Production

Production Limited by Operating Constraint

Continuing with the ammonia plant example, there is a maximum pres-
sure at which the synthesis process may be operated based on the physical
construction of the process. The pressure is maintained through the adjust-
ment of purge gas flow. If the control of synthesis pressure is poor, then
the set point must be maintained far from the maximum pressure. By
decreasing the variation in the pressure control, it is possible to increase
production for a given feedstock rate, as illustrated in Figure 3-5.

Shifting Bottlenecks
Stressed-out plants with old equipment, difficult processes, and recycle
streams often have a shifting and confusing bottleneck. The proper APC
pathway can solve and demystify the problem. Often overlooked are the
significant side benefits of process knowledge gained from the implemen-
tation and day-to-day operation of an APC system.

Figure 3-6 shows the major unit operations for the production of a nylon
intermediate chemical. Since the solids content of the streams is high, the
plant is over 40 years old, and the plant is sold out and running at more
than four times the original nameplate capacity, one or more pipelines,
reactors, evaporators, crystallizers, and centrifuges are always shut down
for washout or maintenance. Centrifuges periodically trip on vibration or
spill over solids (slop) into the recycle. The equipment and operators are

Copyright International Society of Automation

Unmeasured
Wash Water Crystallizers (Cx)
Reactors (Rx) Evaporators (Ev) and Centrifuges (Cs)
Recycle Hydroclones (Hc)
Tk

Feed
Rx Rx Ev Cx Hc Cs

Feed Rx Rx Tk Ev Tk Cx Hc Tk Cs Tk

Feed Rx
Rx Ev Cx Hc Cs

Purge
Reactors, evaporators, and crystallizer heat transfer surfaces get coated Centrifuges slop
with solids and must be periodically shutdown and manually defrosted Recycle solids into recycle
Tk
All surge and recycle tanks (Tk) are undersized Operator sets crystallizer feed rate
based on visual inspection of solids
Purge in sample of hydro-clone overflow Unmeasured
Wash Water

Figure 3-5. Impact of Synthesis Loop Pressure on Ammonia Production

so stressed that the control room is always in crisis mode with incessant
alarms and flipping of screens. Most of the controller outputs are satu-
rated high. War stories rule. Engineer burnout and turnover are high. The
young engineers thrown into the fray defer to operator opinion rather
than applying chemical engineering principles. Management is afraid to
do anything because capacity has actually decreased after each debottle-
necking project.

A large amount of water is introduced into the process from washout con-
nections but is largely not metered. The equipment must be cleaned and
started up manually, so the operator is naturally reluctant to push capac-
ity. The surge volumes that are used as feed and recycle tanks are seriously
undersized for the present production rates, the additional water load,
and centrifuge slopping. The operators continually adjust feed rates or
add water to keep tanks from running dry. Sometimes the recycle tanks
overflow from the additional water load and product is lost to the sewer.

The feed rate to each of the final-stage crystallizers is set by the operator
based on a visual inspection of a sample from the recirculation line for the
amount of solids. The interpretation of the concentration and particle size
is subjective and the main goal of the operator is to reduce how often the
crystallizer must be taken down and the crystal buildup on the walls and
heat-transfer surfaces removed. A supervisory control system is periodi-
cally turned on that finds the lowest production rate and drags the whole

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

plant down to match it. Since the feed to the last stage of crystallizers is
typically set low from the margin of operator error and lack of confidence,
the tail wags the dog, and the plant capacity decreases each year despite
capital projects to replace and add equipment to reduce bottlenecks.

The solution is to remove concentration, production rate, and inventory

control from the operator and automate the washout and startup of the
unit operations. This gets operations and technical support out of the fire-
fighting mode and into a higher and more value-added supervisory and
advisory role.

Some pipeline and pump sizes are increased so control valves are not wide
open, and some on-off valves and magnetic flow meters must be installed
to measure and control water addition. Coriolis mass flow meters are
installed not only to provide more constant solids loading of the feed to
the unit operations but also to provide estimators of solids concentration
and crystal buildup on walls. Nuclear gauges are added to each centrifuge
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

to measure the cake mass in the basket. The combination of mass feed flow
and basket solids load control improves the production rate and on-stream
time of the centrifuges. The scheduling and the sequencing of valves for
automated cleaning and startup makes the performance of the equipment
more reproducible and predictable. The use of variable speed drives to
throttle the feed to the hydroclones increases the pressure drop available
for separation of solids in the hydroclones and eliminates significant slip-
stick since sliding stem valves are not suitable for the high solids concen-
tration. Model predictive control is used to control the solids concentration
in each reactor, evaporator, and crystallizer and to manage the overall
inventory control and purge rate. Rampers and pushers are used to maxi-
mize feed rates without a projected violation of a level or process con-
straint. Finally, real-time optimization is used to track bottlenecks,
coordinate feeds, optimize recycle and purge flows, reach an understand-
ing of the root causes, separate fact from fiction, and provide data to lead
to successful and viable projects.

Application

General Procedure
1. Install a control loop and process performance monitoring
systems.
2. Ask marketing to identify the key business drivers.
3. Develop management sponsors, onsite advocates, and resources.
4. Determine the percentage of time sold out, scheduled and
unscheduled downtime, profit per pound, variable costs per

Copyright International Society of Automation

Unmeasured
Wash Water Crystallizers (Cx)
Reactors (Rx) Evaporators (Ev) and Centrifuges (Cs)
Recycle Hydroclones (Hc)
Tk

Feed
Rx Rx Ev Cx Hc Cs

Feed Rx Rx Tk Ev Tk Cx Hc Tk Cs Tk

Feed
Rx Rx Ev Cx Hc Cs

Figure 3-6. Stressed-out Plant with Shifting Bottlenecks

pound (raw materials, utilities, waste treatment), fixed costs, and

product with less or no revenue (scrap, recycle, and giveaway).
5. Use best demonstrated periods of operation, statistical analysis
and/or high-fidelity dynamic simulations to determine target
values and average for the year or season to estimate actual values
of process efficiency and capacity.
6. Compute the gap between target and actual values and multiply
by the lost revenue or cost factors to complete an opportunity
sizing.
7. Review trend plots and performance of continuous and batch
operations to determine fact and fiction.
8. Meet with technical support and operations and discuss possible
advanced control solutions for each of the gaps.
9. Prioritize the solutions with a weighted rating based on benefits,
capital requirements, speed of implementation, ease of
implementation, and risk.
10. Benchmark the existing operation.
11. Select the top technology and supporting technologies.
12. Train technical support and operations in application of
technologies.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

13. Implement the technologies.

14. Measure the results online.
15. Benchmark the results.
16. Look for new improvements.

Application Detail

Neutralizer
The benefits from the reduction in variability afforded by the improve-
ments to the basic control system can be estimated for the pH example in
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Chapter 2. The largest sources of variability in the existing system are the
ball valves with excessive stick-slip caused by the excessive capacity and
friction and missing positioner. The limit cycle from the reagent valves in
terms of a reagent-to-feed ratio is translated to a pH limit cycle as illus-
trated in Figure 2-4b. However, the limit cycle from the first-stage valve is
first multiplied by the amplitude ratio based on the period of oscillation of
the first stage and the residence time of the second stage as shown in
Equation 3-12.

The loop period for quarter-amplitude oscillations is approximately 6

times the total loop time delay for the first stage, and the time constant of
the second stage is approximately the residence time. Since almost all of
the reagent is added to the first stage, the increase in the amount of unre-
acted feed from the increase in conversion time is estimated for the limit
cycle in the first valve as it propagates to the second stage. Note that the
residence time in the first stage is so short that there is little time for con-
version and most of the effects from pH exceeding the desired control
band occur in the second stage.

Equation 3-13b is a curve fit for the distribution of data points in neutral-
izer pH for a limit cycle. The standard deviation of the limit cycle for the
new precise control valves is estimated and the calculations are repeated.
To see the improvement in yield by other improvements, the decrease in
the fractional integrated error (∆Xp) can be estimated for a typical set of
feed rate changes. For the benefits in terms of less reagent use for a more
optimum pH set point, Equations 3-15 and 3-16 can be used to estimate the
shift in the mean (∆Xm) from reduced variability. Whether it is best to not
move the pH set point and reduce the fraction of product below the low
constraint, or to lower the pH set point to the point where there is no
change in the fraction of product below the low constraint, depends upon
the values of ∆Bp and ∆Bm.

S ni = ( S mi ⁄ 100% ) * ( F r ⁄ F f ) * P n * AR n (3-12)

Copyright International Society of Automation

z i = ( X L – X M ) ⁄ S ni (3-13a)

3 2
φ i = – 1.4743 * z i + 14.488 * z i – 46.888 * z i + 50 (3-13b)

∆X p = ( φ i – φ j ) ⁄ 100% (3-13c)

∆B p = ∆X p * ∆Y p * C t * F f * E f (3-14)

∆X m = ∆S ni (3-15)

∆B m = ( ∆X m ⁄ P n ) * C t * F f * E r (3-16)

where:
ARn = amplitude ratio for neutralizer
∆Bp = delta benefits from less product below low constraint ($/
hr)
∆Bm = delta benefits from shift of mean closer to low constraint
($/hr)
∆Yp = yield loss from distribution below low constraint
Ct = conversion factor for time (24hr/day)
Ef = economic factor for cost of feed ($/lb)
Er = economic factor for cost of reagent ($/lb)
Ff = feed flow (lb/hr)

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Fr = raw material (reagent) flow (lb/hr)
φi = percent of data points below low constraint for limit cycle i
(%)
Pn = process gain in neutralizer from titration curve slope
(∆pH/∆ratio)
Smi = standard deviation in mixer reagent valve position for limit
cycle i (%)
Sni = standard deviation in neutralizer pH for limit cycle i (pH)
∆Xp = delta fractional product below low limit
∆Xm = shift in mean (pH)
XM = mean (pH)
XL = low constraint (pH)
zi = number of standard deviations from the mean to the flow
constraint (see eq. 3-6a)

By the use of a sliding stem valve with a digital positioner, the standard
deviation of a limit cycle in the first reagent valve is reduced from 10%
(Sm1) to 0.1% (Sm2). The process gain is 10,000 pH per flow ratio (Pn), the
flow ratio at set point is 0.1 (Fr/Ff), the set point is currently 7 pH (XM), the

Copyright International Society of Automation

feed flow is 100 kpph (Ff), the limit cycle period is 1 minute (To), the
residence time in the well mixed neutralizer is 20 minutes (τ), the cost of
feed and reagent are both $0.1/lb (Ef and Er), and the conversion decreases
linearly from 98% at 6 pH (XL) to 88% at 5 pH.

The benefits in reduced feed costs from a better conversion for a 7 pH set
point can be estimated as follows with Equations 3-17 through 3-25:

For Sm1 = 10%:

1
AR n = ------------------------
- = 0.008 (3-17)
2 * π * 20

S n1 = ( 10% ⁄ 100% ) * ( 0.1ratio ) * ( 10000∆pH ⁄ ∆ratio ) * ( 0.008 ) = 0.8 pH (3-18)

z 1 = ( 6.0ph – 7.0ph ) ⁄ 0.8pH = 1.25 (3-19)

3 2
φ 1 = – 1.4743 * ( 1.25 ) + 14.488 * ( 1.25 ) – 46.888 * 1.25 + 50 = 11% (3-20)

For Sm2 = 0.1%:

S n2 = ( 0.1% ⁄ 100% ) * ( 0.1ratio ) * ( 10000∆pH ⁄ ∆ratio ) * ( 0.008 ) = 0.008 pH (3-21)

z 2 = ( 6.0ph – 7.0ph ) ⁄ 0.008pH = 125 (3-22)

φ2 = 0% (3-23)

∆X p = ( 11% – 0% ) ⁄ 100% = 0.11 (3-24)

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
∆B p = 0.11 * 0.04 * 24hr/day * 100000pph * 0.1$/lb = $1,056/day (3-25)

The benefits in reduced reagent costs from a shift of the set point toward
the low constraint can be estimated as follows with Equations 3-26 and
3-27:

∆Xm = 0.8 pH (3-26)

∆Bm = (0.8 pΗ/10,000∆pH/∆ratio) ∗24 hr/day∗100000pph∗0.1$/lb = $20/day (3-27)

From the above analysis, it is clear that it is better to leave the set point at
7 pH and reduce the fraction of product that is below 6 pH.

Distillation Column
The benefits from the reduction in variability afforded by the improve-
ments to the basic control system can be estimated for the distillation
example in Chapter 2 with equations similar to those given for the pH

Copyright International Society of Automation

example. The limit cycle before and after the addition of the aggressively
tuned digital positioner is multiplied by an amplitude ratio to determine
the attenuation from the storage tank volume. The period of the limit cycle
is about 6 times the temperature loop time delay, and the time constant of
the storage tank with just an eductor for mixing is about half the residence
time. The attenuated limit cycles are next translated from an oscillation in
the distillate-to-feed ratio to temperature oscillation as illustrated in Fig-
ure 3-5c. The amplitudes of the temperature oscillations are then multi-
plied by the process gain to translate from temperature to impurity
concentration and finally by the economic gain to translate from impurity
concentration to the cost of extra recycle and steam.

The benefits from better measurement resolution and reproducibility

afforded by the use of an RTD and a smart transmitter at a better control
point, and the reduction in noise and loop time delay from the better sen-
sor location, can be estimated by use of the resolution and half of the peak-
to-peak noise as the amplitude of a temperature oscillation. The oscillation
amplitude is multiplied by the amplitude ratio, process gain, and eco-
nomic gain to find the value of the improvements. The use of high control-
ler gain in the distillate receiver reduces the loop time delay. The time
delay introduced by the level loop is approximately the reflux valve dead
band divided by the level ramp rate (integrator gain) multiplied by the
level controller gain. For a horizontal receiver, the improvement is signifi-
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

cant. The change in total temperature loop dead time affects the loop
period and hence the amplitude ratio.

The value of flow feedforward and better feed tank level controller tuning
can be estimated by the reduction in fractional integrated error (∆Xp) in
product impurity concentration for a typical set of flow upsets. Case stud-
ies have shown a reduction of 2:1 from better control valves and a 5:1
reduction from an improved control scheme in product variability for a
distillation process [3.10].

The value of reduced valve stick-slip, although real and significant, is one
of the most difficult improvements to quantify. The estimation of benefits
from advanced control is usually much simpler because higher value-
added variables are more directly affected or optimized.

Paper Machine
The benefits from the use of model predictive control for the basis weight
of a paper machine shown in Figure 3-7a can be rather easily estimated by
Equations 3-28 and 3-29. Figure 3-7b illustrates how a decrease in the 2
sigma standard deviation can result in a shift of the mean much closer to
the optimum.

Copyright International Society of Automation

Long transport delay

AT
1-1
Basis Weight

Basis Weight Valve

Figure 3-7a. Basis Weight Control in a Paper Machine

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Number of
Samples

Basis Weight
Product Spec
Minimum Wt.
Shift in aveage
(mean) value

Figure 3-7b. Shift of Basis Weight Closer to Optimum

∆Xm = ∆S2 / Xm (3-28)

∆Bm = ∆Xm∗Ss∗Cw∗Cr∗Ct∗Es (3-29)
where:
∆Bm = delta benefits from shift of mean closer to low constraint
($/day)
Cr = conversion factor for ream (ream/3000ft2)
Ct = conversion factor for time (1440 min/day)
Cw = conversion factor for width (ft)
Es = economic factor for cost of stock ($/ream)
Ss = sheet speed (ft/min)
∆S2 = delta reduction in two standard deviation (lbs)
∆Xm = fractional shift in basis weight mean
XM = basis weight mean (lbs)

Copyright International Society of Automation

Through the implementation of model predictive control, the 2S standard

deviation in basis weight was reduced from 0.8 to 0.4 for the production of
30-pound paper. The sheet speed is 2400ft/min and the sheet width is 30
feet. If the cost of stock is $2.50 per ream, the benefit from a reduction in
excess basis weight is $1,125/day by the use of Equations 3-30 and 3-31 as
follows:

∆Xm = 0.4/30.8 = 0.013 (3-30)

∆Bm = 0.013∗(2400ft/min)*(30ft)*(ream/3000ft2)*(1440min/day)*($2.50stock/ream) (3-31)
∆Bm = $1,125/day (3-32)

The implications of variability in the kraft, pulp, newsprint, fine paper,

and linerboard sectors of the industry have been quantified in audits of
digesters, bleach plants, stock preparation areas, and paper machines. A
new index has been developed that classifies variability by frequency
decades to help predict end-use problems of pulp bales and paper rolls
[3.11].

Reactor
A key operation in a plant is the catalytic reaction system. In this example,
the catalyst concentration is critical to maximize the conversion of raw
material to product, to improve yield and to minimize costs. Every pound
of the product is sold so it is also important to avoid operating conditions
that could lead to downtime and lost production [3.12].

The yield is correlated to catalyst concentration (Xc) by Equation 3-33

where yield (Yp) is defined as the pounds of raw material converted to
product per pound of raw material fed to the reactor times 100% [3.12].

Yp = 80% + 9∗Cc (3-33)

The business driver is to maximize catalyst concentration subject to a con-
straint (XH) where 0.55% or higher concentrations cause excessive plug-
ging and downtime. The average (mean) catalyst concentration is
presently 0.42% (XM), the total standard deviation is 0.12% (Stot), and the
capability standard deviation that includes the effect of dead time is 0.06%
(Scap). The average raw material flow rate is 100 kpph. The average uptime
is 8000 hr/yr and the raw material cost is $0.2/lb [3.12]. The benefits from
advanced control can be estimated by calculating the minimum standard
deviation possible from feedback control (Sfbc) and then the percent
improvement from a shift of the average (mean) concentration closer to
the high constraint using Equations 3-34 through 3-38, documented in
Chapter 4 for performance monitoring.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Sfbc = Scap ∗ [2 – (Scap / Stot)2]0.5 (3-34)

Sfbc = 0.06% ∗ [2 – (0.06%/0.12%) 2]0.5 = 0.08%

∆Stot = [1 – (Sfbc/ Stot)] (3-35)
∆Stot = [1 – (0.08%/0.12%)] = 33%
∆XM = [1 – (Sapc / Stot)] ∗ [XH −XM] (3-36)
∆XM = [1 – (0.08%/0.12%)] ∗ [0.55% − 0.42%] = 0.043%
∆YM = 9∗∆XM (3-37)
∆YM = 9∗0.043% = 0.387%
∆BM = (∆YM/100%)∗Ff∗Ct∗Ef (3-38)
∆BM = (0.387%/100%)∗100000pph∗24hr/day∗0.2$/lb = $1858/day
where:
∆Bm = delta benefits from shift of mean closer to high constraint
($/day)
∆YM = yield improvement from higher mean concentration of
catalyst (%)
Ct = conversion factor for time (24hr/day)
Ef = economic factor for cost of feed ($/lb)
Ff = feed flow (lb/hr)
Stot = standard deviation in catalyst concentration (%)
∆Xm = shift in mean of catalyst concentration (%)
XM = mean catalyst concentration (%)
Sfbc = standard deviation for feedback control

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
XL = low constraint for catalyst concentration (%)

Popular Excuses
The following is a compilation of popular excuses used by people who
want to maintain the status quo and stay in their comfort zones. Most
reflect a process-design and steady-state viewpoint and a lack of under-
standing of dynamics and the advanced control goal of building and
incorporating process knowledge by extensive process monitoring, test-
ing, and modeling. In case these excuses are used as a means to delay an
APC opportunity, the APC counterpoint is given.

1. It is an equipment or process problem, not a control problem.

If there were no equipment or process limitations, there would be
no need for advanced control.

Copyright International Society of Automation

2. The bottleneck keeps shifting.

Advanced control can find it.

3. We don’t know which constraint is more important.

4. Advanced control can determine relative importance.
5. We don’t know the relationships for the constraints.
6. Advanced control will identify and model them.
7. Advanced control covers up equipment problems.
8. Advanced control will quantify how much of the time and to what
degree each constraint actively limits production .
9. Process monitoring, testing, and modeling will expose the
problems.
10. With APC, my operators will forget how to run the process.
11. With a properly designed APC interface, they will become better
operators.
12. You can’t optimize this process; there are too many constraints.
13. The greater the number of constraints, the greater the economic
opportunity, and the greater the need for APC.

Rules of Thumb
Rule 3.1. — Measurements of present variability and estimates of reduced vari-
ability, attenuation factors for back-mixed volumes, conversion factors, and an
economic gain factor can be used to provide a more accurate estimate of the bene-
fits from advanced control. The gains and factors must trace the path from
the controlled variable to the process variable in the product that leaves
the plant. Dynamic property estimators may be useful to find process
gains and put the calculations online.

Rule 3.2. — Find the best practical production rate or yield from the best periods
of operation or batches from cost sheets and the best theoretical rate or yield from
steady-state and dynamic simulation of new operating conditions. The opportu-
nity assessment questions should all be oriented to find how to eliminate
the gap between the actual and practical or theoretical performance.

Rule 3.3. — Loop and process performance monitoring systems and online plant
economic performance calculations are essential to maximize and sustain benefits.
Without these calculations in place, the benefits will get lost in the noise or
attributed to other activities.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Rule 3.4. — If the benefits are not documented and reported, advanced process
control will not get the recognition needed to insure resource availability and com-
mitment. After any initial enthusiasm, the effort will rapidly fade. Due to
downsizing and information explosion, the average user is faced with an
overwhelming number of initiatives and supposedly neat ideas.

Rule 3.5. — If online analyzers have long sample-transportation delays, long

processing delays, long cycle times, poor signal-to-noise ratio, or an insufficient
on-stream time, develop and install dynamic estimators before process testing.
Model predictive controllers should use these estimators.

Rule 3.6. — Insure that at least a model predictive control (MPC) with some
degree of optimization is implemented. MPC has the best track record for ben-
efits.

Rule 3.7. — APC technologies must be employed in closed loop control to achieve
most of the benefits. Advanced control improvements that stay in an advi-
sory mode achieve a small fraction of the benefits possible, because of
operator inattention and bias.

Rule 3.8. — A simple maximization or minimization of feed, steam, reflux, or

recycle flow can be done with a ramper or pusher. A linear program is often
back-engineered to achieve what can be better tuned and maintained with
a ramper or pusher. In some cases, a pusher may be included in the MPC
capability.

Rule 3.9. — A linear combination of economic objectives with a linear relation-

ship to constraints is best optimized by a linear program (LP). The LP will find
the optimum intersection of constraints.

Rule 3.10. — Plants with changing economic objectives, complex recycle effects,
shifting bottlenecks, and complex nonlinear relationships in several unit opera-
tions need a real-time optimization. High-fidelity steady-state models are
used for the real-time optimization of continuous constant conversion pro-
cesses, whereas dynamic models are needed for batch operations and for
processes where reaction and polymerization rates are important for opti-
mization.

References
1. Bialkowski, William L., “Process Control Audits Have Major Impact on
Uniformity,” American Papermaker, September 1990, pp. 50-57.
2. McMillan, Gregory K., “Benchmarking Studies in Process Control,” InTech,
November 1992, pp. 44-46.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

3. Shunta, Joseph P., “Use Statistics to Identify Process Control Opportunities,

Chemical Engineering Progress, October 1996, pp. 47-49.
4. Shunta, Joseph P., Achieving World Class Manufacturing Through Process

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Control, Prentice Hall, Inc., 1995.
5. Bialkowski, William L., “Process Analysis and Diagnostics,” Fisher-
Rosemount Systems Users Group Meeting, November 1996.
6. Luyben , Michael L. and Luyben, William L., Essentials of Process Control,
McGraw Hill Chemical Engineering Series, 1997.
7. Chia, T.L., and Lefkowitz I, “Add Efficiency with Multivariable Control,”
InTech, September 1997, pp. 85-88.
8. Mansy, Michael M., McMillan, Gregory K., and Sowell, Mark S., “Step Into the
Virtual Plant,” Chemical Engineering Progress (CEP), February 2002, pp. 56-61.
9. Tinham, Brian, “Control in the Chemical Industry,” Control and
Instrumentation, January 1993, pp 34-35.
10. Beal, James F., “Process Control Analysis, Improvements and Results,” ISA
Expo/Conference, Houston, September 10-13, 2002.
11. Bialkowski, William L., “Process Control Related Variability and the Link to
End Use Performance,” Control Systems Conference, Helsinki, September 17-
20, 1990.
12. Shunta, Joseph P., “Assessing & Implementing Control Improvement
Opportunities,” ISA Short Course SC05(Du), 1996, Instructor’s Notes.

Copyright International Society of Automation

Practice

Overview
Target product quality and production levels can be maintained by contin-
uously evaluating the performance of the process and the control system.
The maximum throughput and operating efficiency of a plant are ulti-
mately determined by the process design and equipment selection. How-
ever, in many cases a plant's operation is far from achieving the ultimate
capability inherent in the plant design and equipment. For example,
numerous studies done in the pulp and paper industry show that loop uti-
lization ranged from 55% to 76%, depending on production area.

The reasons for process variation and poor control utilization can be attrib-
uted to one or more of the following:

• Field measurement or final control element: analyzer and process

measurement failures, noise, and repeatability errors; incorrect
ranges and scan times; improper sensor and valve locations;
damper or valve stick-slip; nonlinear installed damper and valve
characteristics; and incorrect valve sizes
• Process or control design: insufficient capacity, process changes,
undersized units, and lack of lab feedback
• Process interaction and mechanical problems
• Process, lab, and raw material problems (more so for chemical
than pulp and paper)

119
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,

Copyright International Society of Automation

One of the main reasons such conditions exist is that the downsizing of
support services in many plants has resulted in plants operating with a
minimal staff for process control and instrumentation maintenance. Often
there is only enough manpower available to fix the critical problems found
today that are limiting production and affecting product quality. Under
these circumstances, there is little time to study the process operation to
determine if an abnormal condition exists that could soon be a major
source of process disturbance if not addressed. A problem may not
become visible until the situation has deteriorated to the point where it
affects product quality or production. Operating in this “firefighting”
mode may lead to variations in operation that result in less than maximum
production in a sold-out market, or to operating at less than best efficiency,
or to wider variation in product quality, regardless of production rate.

The benefits from early detection of control system performance decay

have been well documented by Bill Bialkowski [4.1] and others [4.2], [4.3].
The early detection of control system performance decay, of changes in
process dynamics and the detection of new sources of variability, is abso-
lutely necessary. This is an ongoing process that should focus on detection
through continuous analysis and correction through diagnostics to mini-
mize impact on plant operation.

One of the key factors is that the performance of control loops decays with
time as a result of the wearing of control valves, loss of calibration of trans-
mitters, or changes in the operation of the process. Some plants have real-
ized the impact that control and field devices are having on their
operation.

Efforts to achieve best plant performance must address both the areas of
analysis and diagnostics.

• Process Analysis is the examination of a process, its elements and

their relationships.
• Process Diagnostics is the investigation of the cause or nature of a
process condition or problem.

The relationship between process analysis and process diagnostics is

illustrated in Figure 4-1.

In some cases, additional staff has been added and a performance moni-
toring tool has been purchased to periodically evaluate control loop and
field device performance. However, this solution is costly and often hard
to justify in the short term. Tools layered on top of a traditional Distributed
Control System (DCS) to detect abnormal operation have had limited suc-
cess in the detection of problems in instrumentation and fast processes

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Measure
Process Identify Determine Eliminate
Variation, Areas for Root Cause Source of
Control Improvement of Variation Variation
Utilization

Analysis
Work Order - Diagnostics
Operation Problem

Figure 4-1. The Reduction of Variability Is an Ongoing Process

because of limitations in the speed with which information may be

accessed — limitations that are inherent in the design of most DCS and
Programmable Logic Controllers (PLC) systems in use today.

However, some manufacturers have introduced a whole new generation

of field-based, scalable process control systems and Fieldbus field devices.
The core technology used in some of these systems provides a solid foun-
dation that allows the continuous monitoring and detection of abnormal
control and field device operation.

In this chapter, the basis for automatic detection of abnormal operating

conditions using this new technology is examined. The chapter covers per-
formance monitoring tools for online analysis and monitoring of control
loops. It also introduces the user to the use of multivariate Principal Com-
ponent Analysis (PCA) and Partial Least Squares (PLS) for the online anal-
ysis and monitoring of a unit operation in batch or continuous processes.
A multivariate regression technique, such as PCA and PLS, can be
designed to handle cases with a large number of correlated inputs and can
extract contributions of individual inputs and identify breakdown in input
correlation structure

Opportunity Assessment
The justification for investment in tools and people for process and control
analysis and diagnostics is the reduction in process variability and the
associated improvement in plant profitability. A process parameter’s effect
on a particular plant’s production depends greatly on that plant’s limita-
tions and operating conditions. If the answer is yes to some of following
questions, there is enough of an economic incentive to implement a perfor-
mance monitoring system in a plant.

1. Are loops that affect plant capacity or efficiency always in their

highest mode?

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

2. Do loops that affect plant capacity or efficiency have too much

variability?
3. Is equipment and instrument maintenance more reactive than
predictive?
4. Are there batch operations where the quality or yield is affected by
the variability or limits in feeds, process variables, or manipulated
variables?
5. Are there batch operations where the cycle time is affected by the
variability or limits in feeds, process variables, or manipulated
variables?
If there is a large variability in variables that adversely affect production or
product quality then by identifying these areas of variation and address-
ing the source of variation, it is often possible to reduce variability and
thus increase production, increase operating efficiency, or improve prod-
uct quality.

The built-in diagnostic and analysis tools of modern scalable control sys-
tems provide the unprecedented capability to automatically identify pro-
cess variation caused by under-performing loops. This is done by
continuously calculating the improvement in control that is possible and
comparing this to the baseline loop performance. This approach to perfor-
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

mance monitoring automatically examines input-output and control

blocks in the controller for abnormal operating conditions that may indi-
cate maintenance is required.

By automating the task of monitoring measurements, actuators and con-

trol performance, it is possible to virtually eliminate the need for plant
audits with portable PC-based tools and thus allow plant maintenance
resources to focus on resolving the problems that affect plant operation
most. Since these systems utilize the function-block architecture and
model established for Fieldbus, they can use the input and output parame-
ter status and block mode to determine abnormal behavior. In particular,
this information is used to automatically determine the percent time a
block was not in its designed mode of operation, or an input status was
“bad,” “uncertain” or “limited,” or a downstream condition limited con-
trol action. Blocks having an abnormal status or mode condition may be
listed based on the percent time the condition existed. Plant control is
summarized by the Performance and Utilization values. Such metrics can
be used to judge overall control performance and whether control is being
fully utilized.

Once a problem is detected, finding the solution to the problem may be

straightforward for certain conditions. For example, a high variability

Copyright International Society of Automation

indication may in some cases only require that the PID control block be re-
tuned to correct a change in operating condition. An uncertain indication
on a transmitter may sometimes be quickly resolved using the online diag-
nostics provided by the manufacturer of the device.

Unfortunately, in some cases the source of the problem may not be appar-
ent and further analysis is required. In particular, resolving the source of
high variability may not be as easy as just re-tuning the loop. Even though
the loop may be correctly tuned, valve stick slip may cause the loop to
continually oscillate, as discussed in Chapter 2 and illustrated in Figure
4- 2.

Setpoint (SP)

Controlled Process Variable (PV)

Implied Valve Position (IVP)

Stem Position

Figure 4-2. Limit Cycle from Valve Stick-Slip

Tools for high-speed trend, histograms, and power-spectrum and cross-

correlation analysis allow quick diagnosis of such problems. In a modern
control system, such tools are integrated into the system design and allow
such analysis to be done automatically.

Manufacturing companies are judged by the quality, cost, and availability

of the products they make and sell. To achieve world-class performance in
these areas, the manufacturing operation must develop specific metrics to
assess how well they are performing day to day and guide them towards
improvement. In order to improve control, the data must be taken fre-
quently enough to show inertial effects. Also, statistical metrics should be
established to evaluate process and control system performance. With
such online tools in place, it is relatively easy to detect the presence of a
new point source of variability, as individual signals will start to exhibit
markedly different behavior.

Process monitoring can take many forms, from a purely mechanistic

approach guided by first principles to a purely statistical approach. In

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

today’s world, data drives the decision-making process and the ability to
make use of the vast amounts of process data is essential. In many plants,
the rate at which we are accumulating this data surpasses our ability to
extract knowledge from data and use it for better decision making. Multi-
variate statistical analysis is a technique that allows us to analyze plant
data in order to extract underlying themes in the behavior of the data, and
to then use these themes to monitor the state of the process.

Traditionally, the task of monitoring plant data has fallen into the statisti-
cal process control (SPC) or the univariate statistical analysis world. The
manufacturing industries have made great progress by focusing on key
quality variables and monitoring these variables with univariate analysis
techniques such as SPC. But are these techniques still relevant now that
the volumes of data being extracted by our automation systems have
increased by orders of magnitude? The answer is, Not always! In previous
years we had 10 engineers per quality parameter being monitored; today a
single engineer is asked to monitor 10 quality parameters. The drive to
understand quality coupled with the vast amounts of data can present a
formidable task, one well suited to the application of multivariate tech-
niques.

The multivariate techniques that are showing success today are called
Principal Component Analysis (PCA) and Partial Least Squares or Projec-
tion to Latent Structures (PLS). Multivariate statistical analysis is not a
new technology, but the application of PCA and PLS has gained attention
due mainly to the data explosion. The manufacturing and process indus-
tries have invested heavily in real-time data acquisition systems or Process
Information Management Systems (PIMS) and there is now a strong desire
by manufacturing directors to see this accumulation of data put to use to
improve plant operation; hence the strong interest in multivariate moni-
toring techniques. Other multivariate monitoring techniques include Fac-
tor Analysis, Eigen-vector Analysis and Singular Value Decomposition.

The field that uses multivariate statistical analysis tools for real-time anal-
ysis of process data is called Multivariate Statistical Process Control
(MSPC). The attraction of MSPC is in the ability to rapidly develop behav-
ioral models of your process, akin to fingerprinting normal process behav-
ior, and then to continuously compare current plant behavior to the
normal fingerprint. If a deviation from normal plant behavior is identified,
MSPC will allow you to identify which plant variables are the major con-
tributors to the cause of the deviation.

The ability to monitor, identify and diagnose the cause of process variabil-
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

ity is a task every engineer will recognize as important and one that will
improve the performance of our manufacturing facilities. The technologies

Copyright International Society of Automation

of today are focusing on these monitoring challenges, while the monitor-

ing technologies of tomorrow will be targeted more at relating variability
to the bottom-line profitability of our plants—an important link still not
well defined in most manufacturing environments.

Examples

Control Utilization and Its Impact

Control utilization is an indicator that can be used to quickly determine if
control and measurement problems exist within a control system. It is rela-
tively easy, using most control systems, to take a snapshot of control oper-
ation and to then calculate and document the control utilization. For
example, using the documentation features of most control systems, it is
possible to print the operation screens and later to compare the mode of
the control blocks to the design mode for the control. In some cases, a low
value of control utilization can be attributed to insufficient operator train-
ing on the control. However, surveys of many plants indicate that the pri-
mary reasons for control not being fully utilized fall into two areas:

• Field measurement or control element

• Process or control design

An immediate improvement in control utilization will be achieved by

addressing these problems. To achieve full utilization, improved
communications between maintenance and operations is important.

The automatic collection of control-utilization statistics is a major benefit

in identifying problems in measurement, control or process design. In
plants that do not have a control system that supports this capability,
problems may go unrecognized that have a direct impact on the plant
operation and efficiency of production. For example, at one pulp and
paper plant a snapshot of the control utilization was collected to allow the
state of the process control to be quantified. This survey revealed the fol-
lowing:

Total Control Loops not in

Plant Area Percent Utilization
Loops Normal Mode
Bleach Plant 78 60 76%
Power House 185 130 70%
Pulp Mill 174 116 66%
Paper Mill 236 134 56%

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

When the utilization was presented to the head of maintenance, he was

surprised to find it so low. To address this problem, a number of things
were put in place at this plant.

Automatic calculation of control utilization was implemented using fea-

tures of their historian.

An instrumentation team was formed to investigate loops that were not

running in their normal design mode. This team was responsible for mak-
ing sure measurement, control valve, and process problems were
addressed in a timely fashion.

The reduction in variability provided by these two steps led to significant

improvements in plant throughput and product quality. Long after the
team was formed, it continues to provide significant, quantifiable results
to the plant through constant evaluation and improvement of control per-
formance.

Control Variability
The response of a control loop in its designed mode of operation may not
be adequate to compensate for process disturbance and changes in set
point. The root cause of poor control performance may be traced to the
control design, tuning, measurement or actuator performance. Many mod-
ern control systems provide a means to automatically quantify the varia-
tion seen while on automatic control. Using this embedded feature, it is
possible to quickly identify control loops that require maintenance. If such
capability is not available in the control system, it may be possible to uti-
lize the statistical calculations included in a plant historian or to add loop
performance-monitoring packages that connect directly to the DCS or to
the plant historian through OPC connectivity. Both methods will monitor
control loop variability and provide insight into changes in control perfor-
mance.

By comparing the current variation over a shift or day to that previously

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
seen for the same operating condition and timeframe, it is possible to iden-
tify loops whose performance is degraded. Through the investigation of
problems that cause increased variability, it is often possible to find the
root cause and significantly improve plant operation. As discussed in
Chapter 3, this reduction in variation often makes it possible to shift oper-
ating points closer to operating constraints or to maintain the best operat-
ing set point, and thus provide greater throughput or improved quality.

Copyright International Society of Automation

Quality Variable Monitoring

We can use an SPC univariate chart to easily detect where the process is
going out of control. Consider the analysis of the two variables X1 and X2
in Figure 4-3. Although neither variable appears to demonstrate any sig-
nificant deviation from the norm, a scatter plot of the data clearly points to
an outlier. This irregularity in the operation of the process could only be
detected by revealing how this point differs from the normal correlation
structure of the process data. X1 and X2 could not independently pinpoint
the fault, but the relationship (correlation) of X1 to X2 clearly identifies a
change in the process. Thus, multivariate statistical analysis models the
correlation structure of the process data in time and detects abnormal
behavior by pinpointing deviations from the normal correlation structure.
The simple example of monitoring the correlation’s structure in the two
variables can easily be extended to monitoring hundreds of process vari-
ables.

Outlier

Figure 4-3. Scatter Plot Reveals an Outlier

Caster Monitoring
The next sample industrial process we will consider for MSPC is a caster
monitoring application in a steel mill. The process is a continuous, straight
mold, Demag slab caster, shown in Figure 4-4. Liquid steel enters the
caster process where it is partially solidified in a water-cooled copper
mould. The steel exits the mould, where the steel is contained within a
thin solidified shell or skin, and cools as it moves along the process line
towards the run-out. Upon leaving the mould, there is potential for the
loss of containment of the molten steel. This loss of containment and
release of molten liquid steel is a called a breakout. A casting breakout is
an extremely hazardous occurrence that causes equipment damage and
loss of production. An MSPC monitoring system was installed to provide
real-time monitoring of the casting process variables in order to predict
potential breakout conditions and provide operators with information to
help diagnose casting process problems.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Ladle - 300 tonne

Tundish - 60 tonne

Mould

Bending

Unbending

Runout

Figure 4-4. Slab Caster Process

The MSPC casting application has a few multivariate SPC control charts
representing the stable mould operation and alarms on detection of any
significant anomalies in the behavior of the process variables. At any time,
the operator can view how the measured plant values are contributing to
these alarms, to help diagnose process problems. The final PCA model of
the process contained over 240 inputs, collected selectively over several
months of operation in order to cover many different operating scenarios.
The PCA models were further summarized into control charts, using
Hotelling’s T2 statistic, to be monitored along with the prediction-error
control chart. T2 statistics provide an indication of variability. The T2 is
named for Harold Hotelling, a pioneer in multivariate statistical analysis.

These statistical control charts were again summarized in a composite set

of alarms for operator notification. This multivariate SPC information is
displayed on the Caster Main control screen, along with other pertinent
mould data, as shown in Figure 4-5.

These three charts form the basis of the real-time MSPC system, providing
continuous information to the operator on the stability of the mould oper-
ation. This example of an MSPC system, implemented in a steel plant,
delivered benefits in the areas of increased production, improved safety
and much improved process understanding—the ultimate goal of any
monitoring system [4.12].
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Figure 4-5. Display Interface for Slab Caster Example (Courtesy of Dofasco,
Inc.) [4.12]
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Application

General Procedure
1. Decide on the objectives of performance monitoring in terms of the
relative importance of diagnosing and predicting the following:
a. Measurement problems
b. Rotating equipment problems
c. Hydraulic resonance and shock waves
d. Pressure regulator and steam trap problems
e. Interlock and alarm system sequence of events
f. Control valve problems
g. Improper controller tuning
h. Improper controller modes
i. Interaction between control loops
j. Feedforward and disturbance analysis
k. Weeping and flooding of column trays
l. Distribution and mixing of feeds, components, phases,
polymers, and particles in fluidized beds, columns, fermentors,
reactors, and crystallizers
m. Analyzer and sample system problems

Copyright International Society of Automation

n. Lab sample and analysis process problems

o. Ambient temperature and humidity effects
p. Raw material problems
q. Fouling and coating (heat-transfer surfaces, polymer pipelines
and reactors, crystallizers, sensors, electrodes, valve trim,
column trays, particle filters, ion exchange beds, catalyst)
r. Activity evaluation (beds, catalyst, initiators)
s. Batch sequence and profile problems
2. Select the proper scale, exception reporting, filter time, and scan
time so that information is not lost due to poor resolution or
missing data points.
3. Activate process monitoring embedded in your control system or,
if not available, install a performance monitoring application that
meets your objectives.
4. Establish a best period of operation for each plant area or piece of
process equipment and use this as the baseline for relative
performance indices.
5. Turn off and flag the performance analysis of loops that are
intentionally out of service due to the shutdown or sequencing of a
unit operation.
6. Compute the economic performance of your system on line
(efficiency and capacity, preferably in terms of cost of goods and
revenue).
7. Set the deviation alert and alarm limits from the baseline wide
enough to avoid an initial barrage of abnormal conditions when
the system is first turned on, but narrow enough to flag important
issues for the more critical process variables and control valves.
8. Work on down the list of alerts or alarms, prioritized by economic
importance and relative performance, to fix automation and
process deficiencies.
9. Reset the baseline and narrow the deviation limits as you progress
and the performance improves.
10. Establish procedures and best practices for continual
improvement.
11. For constrained model predictive control systems:
a. Display the constraints that are currently governing, the
direction of the change in the affected controlled variables
demanded by these constraints, the percent of time each

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

constraint has been active, and the percent of time that each
constraint has been violated.
b. Provide trending and statistics of the bias correction to each
trajectory.
c. Provide trending and statistics of predicted values for each
trajectory.
d. Provide trending and statistics of optimization set points (set
points from rampers and pushers).

Application Details

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
With a modern control system, all control loops are automatically moni-
tored on a continuous basis and any degradation in loop performance or
the detection of an abnormal condition in a measurement, actuator, or con-
trol block is automatically flagged. These systems may thus identify prob-
lem areas sooner than can audits done with portable PC-based tools. By
using automated system performance monitoring and built-in diagnostic
tools, the typically limited resources of plant maintenance can be used to
advantage to resolve measurement, actuator and control problems. As a
result, maintenance costs may be reduced or a higher overall level of sys-
tem performance may be maintained and process variability reduced. Any
reduction in variability can lead directly to greater plant throughput,
greater operating efficiency and/or improved product quality.

Continuous Control
In evaluating the performance of a continuous process, control systems
may calculate indices that quantify loop utilization, measurements with a
“bad,” “uncertain,” or “limited” status, limitations in control action, and
process variability. In addition, for control loops, the systems show the
potential improvement possible in control loop performance. The follow-
ing are a few practical scenarios that show how such information might be
used to determine an operating problem.

Loop Not in Normal Mode

A critical process input flow loop is designed to normally run in cascade
mode with its set point determined by a composition loop. However, the
midnight shift operator has to change this flow loop to Auto because of
wide fluctuations in the set point. The next morning, the process engineer
for the plant sees that the system flagged the actual mode as being incor-
rect for more than 1% if the time. This information allows the engineer to
quickly spot that there was a problem in the primary loop tuning and thus
minimize any off-spec product that resulted from manual adjustment of
this key flow.

Copyright International Society of Automation

Limited Output
An instrument technician for the power plant notices that the system
shows the oil-flow control loop is limited in operation. Having been
alerted to this problem, the technician determines that the set point for the
oil header pressure control has been lowered below the design pressure,
forcing the oil valve to go fully open under heavy load conditions. When
the pressure control is readjusted to its designed target, the oil valve can
meet its set point without going fully open.

Bad I/O
A key temperature measurement is flagged by the system as having been
“bad” more than 1% of the time over the last day. Having been alerted to
the fact, the instrument technician re-examines the transmitter calibration
and finds that the device has been calibrated for a temperature range that
is too low. Re-calibrating the transmitter restores the accuracy of the mea-
surement and improves the operation of the process.

High Variability
During normal operation of the plant, the plant engineer sets all variabil-
ity limits to the current value plus 5 percent. After a few weeks, he notices
that the system has flagged a critical flow loop as having excessive vari-
ability. Upon further investigation, he discovers that the valve positioner
connection to the valve stem is loose, causing the control loop to cycle
severely. Fixing the valve positioner returns the variability index to its nor-
mal value.

All these cases are quite typical. Individually they may seem to have mini-
mal impact on plant operation. However, the net effect of these problems
could be significant if they were not addressed in a timely manner. An
automated analysis system to evaluate performance can detect abnormal
situations as they happen and thus play a key role in preventing produc-

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
tion losses and product quality problems.

Built-in diagnostic tools may be used to quickly determine the source of a

problem. Typical diagnostic tool sets will provide support for the follow-
ing:

1. Trends
A high-speed trend provides the ability to trace any parameter
associated with measurement or control actions at the same rate at
which these values change. A trend resolution of 100 msec is usu-
ally sufficient to analyze most measurement and control problems.

Copyright International Society of Automation

A high-speed trend requires over-sampling at the input card and

access at the function-block level.

Experienced control specialists can readily spot a control valve that

has excessive slip-stick from a trend recording of the process vari-
able and controller output. Until waveform recognition systems
are developed to reliably provide this analysis online, there are
some general patterns to look for to help track down the causes of
oscillations.

For example, a rounded sine wave and sawtooth oscillation with a

period much larger than the integral (reset) time are symptomatic
of a reset-induced cycle for a loop dominated by a time constant
and time delay, respectively. If the PV oscillation has a flatter top
(not as rounded for the sine wave or as sharp for the sawtooth), it
is and indication that the control valve has excessive dead band
(backlash).

If the PV oscillation becomes more like a square wave, then the

control valve has appreciable stick-slip. The CO oscillation will
resemble a sawtooth. For stick-slip, the oscillation is a limit cycle
rather than a reset cycle. Any reset action at all will initiate the
cycle and the cycle period will become proportional to the integral
time. A trend of a low-noise flow measurement is the best way to
see these oscillations. For sliding stem valves, the actual valve
position from a smart positioner can be trended. In both instances,
the unfiltered measurements are fast and the patterns for a dead
time–dominant response should be observed on the trend.

2. Histograms
A histogram allows the distribution of the variation in a measure-
ment value or actuator position to be analyzed. A bell shaped dis-
tribution indicates that the source of variation is random in nature.

3. Power Spectrums
A power spectrum is a frequency distribution of the components
that make up a measurement or actuator signal over a selected
period of time. Such information may be helpful in determining
the magnitude and frequency of process noise or the frequency of
disruptions caused by loop interaction or upstream processes.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

4. Cross Correlations
The influence of other control loops and upstream conditions on
the variability of a control loop may be determined through cross
correlation

Diagnostic tools for process and control analysis are available as embed-
ded features in some control systems. These tools may be used to analyze
both fast and slow processes. If an integrated tool set is not available, diag-
nostic applications can be layered on the control system for diagnostics,
although they are often limited to analysis of slow processes. Figure 4-6
illustrates the diagnostic information that is presented to the user by a typ-
ical performance monitoring system.

FT101/PID1
Analysis
+ANAL_FT101PID020602
ANAL_FT101PID020802
PowerSpectrum
Histogram
Autocorrelation
Trend
Trend Auto correlation
Cross Correlation
CORR_FT101PID020802
PI103/PID1/OUT
AI321/AI/PV
+CORR_FT101PID020902
+CORR_FT101PID021002

Power Spectrum Histogram

Figure 4-6. Process and Control Diagnostic

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Batch Control
The application of online performance monitoring is most often associated
with continuous processes. However, the percentage improvement from
its application to batch control could be larger because the state of the con-
trol loops typically receives less attention than the sequence of the batch
operation. The emphasis to date in batch control has been on event sched-
uling rather than on process control. The benefits of monitoring can be
especially significant when you consider the higher profit margin and
value-added opportunities for specialty chemical and pharmaceutical
products, which are almost always batch processes. In such cases, the
answers to Opportunity Assessment questions (4) or (5) are frequently
Yes.

Copyright International Society of Automation

Figure 4-7 shows a batch reactor for a sold-out high profit–margin spe-
cialty chemical plant. One of the feeds and a byproduct can be recovered
from the vent system if there is sufficient condensing to reflux unreacted
feed back to the reactor and there is no liquid carryover from high level or
foaming. Periodically there are interruptions due to high reactor pressure,
level, and exchanger temperature. The cooling water pressure is some-
times too low and its temperature too high. The feed profile is scheduled
to minimize the activation of interlocks and excessive riding of the con-
denser, vent system, and exchanger limits. The efficiency and safety of the
operation depends upon a uniform quality, flow rate, and distribution of
the feeds within the batch. There is no online composition measurement of
the product. The yield varies significantly with each batch.

PC
1-3
TC
1-3 PT
1-3
FC TT
1-3 1-3 Vent System

Eductor
PC
FT 1-1
1-3
Anti-Foam Coolant

FC PT
1-2 1-1 TC TT
1-2 1-1

FT TC RSP
1-2 1-1
Feed B

FC TT
1-1 1-2

FT Coolant
1-1
Feed A LT LC
Batch Reactor 1
1-2 1-2

Discharge

Figure 4-7. Batch Reactor Control System

A performance monitoring system is installed that provides indices on

variability, the detection of limits, power spectrums, and cross correla-
tions. The results are as follows:

The standard deviation and variability indices are high and the control
output blocks are limited at times during the batch for the reactant B and
antifoam feed loops, the condenser and exchanger temperature loops, and
the reactor pressure loop.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

The low-limit flag for the output function block of the reactant B flow con-
troller is periodically activated throughout the batch. After some investi-
gation, it is discovered that the original sliding stem control valve was
replaced with a ball valve that has too much capacity and too much fric-
tion near the closed position. The reactant B flow controller ramps up its
output but there is no appreciable increase in flow above the leakage flow
until its output reaches 6 per cent. Then there is a burst of flow so high
above the set point that the controller must close the valve. This square

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
wave of flow from the stick-slip of the oversized control valve continues
through most of the batch.

The antifoam controller output during some batches ramps up to its high-
output limit. A correlation analysis shows that this coincides with the
simultaneous demand for antifoam by other batch operations and a dip in
supply pressure. An undersized antifoam pump is pinpointed as the cul-
prit. A further look at other variables shows there is a significant cross cor-
relation between high antifoam flow controller output and high reactor
pressure controller output and high condenser temperature controller out-
put. This indicates the possibility of carryover into the vent system that
can result in contamination of the byproduct and a corresponding loss in
yield of the main product.

A trend of the condenser temperature variability indices and limit flags

show that the condenser performance is often worse during hot summer
afternoons. There appears to be no correlation with the sequence of other
batch operations. A review of the cooling-water system shows that the
cooling-water valves for other unit operations are left wide open all the
time and that the reactor condenser is at the highest elevation in the plant.
Furthermore, the cooling tower water temperature on hot days reduces
the Log Mean temperature difference for the exchanger by more than 20%.
This, combined with the inability to increase cooling water flow due to the
low pressure at the condenser elevation, causes the condenser control
valve to go wide open and the condenser temperature to float above the
set point. The high condenser temperature decreases the reactor yield and
increases the variability in the composition of the byproduct and recycle
recovered by the vent system. The installation of a higher head cooling
tower water pump and the commissioning of temperature control loops
for the other batch processes is a much lower-cost solution than the addi-
tion of a new cooling tower as favored by operations.

A power spectrum analysis of the eductor pressure loop in the vacuum

system and the reactor pressure controller shows that there are large peaks
at the same frequency and that these peaks get larger during hot summer
months. Upon inspection, the integral time of the eductor pressure loop is
much less than the period of oscillation that corresponds to the common

Copyright International Society of Automation

frequency with significant power. This is a clue that there is a reset cycle
from excessive integral action from the eductor pressure loop. During hot
summer days the process gain and time constant are less, making the reset
cycle more severe.

The above analysis reveals the importance of extending the scope of per-
formance monitoring to other unit operations, utility systems, vent sys-
tems, and ambient conditions. It also shows the value of being able to
trend performance indices, status flags, events, and variables to look for
coincident events. It is important that the trend be fast enough to deter-
mine what happened first and that the oscillation period or waveform not
be distorted by aliasing. Finally, there is obviously a need for some basic
understanding of the system to determine the actual cause-and-effect rela-
tionships.

The knowledge gained from performance monitoring systems leads not

only to immediate improvements in valve and pump sizes and controller
tuning, but also to the implementation of advanced control techniques. In
the chemical plant example, the identification of the limits can be used to
provide constraints for model predictive control. In this case, the con-
straints are condenser, exchanger, and vent valve position; the variable to
be optimized is reactor feed. A model predictive controller could be set up
to maximize reactor feed without violating utility and vent system limits.

Once the basic system has been improved, the use of performance moni-
toring can be extended to include online multivariate principal component
analysis (PCA). A PCA worm plot can provide rapid recognition of
batches that start to trend away from normal operation to warn of abnor-
mal conditions before they cause significant problems of reduced yield or
capacity. It can also lead to Partial Least Squares (PLS), Neural Network,
and dynamic linear estimators of batch end time and product concentra-
tion based on various peaks during the batch cycle. For the above reactor,
it turned out that the amount of fresh feed added to reach the first peak in
temperature was an indication of the composition of the recycled feed and
a sustained low condenser and vent valve position could be used to pre-
dict the batch end point.

While the above analysis was for a specialty chemical, the opportunities
for fermentors could be even greater due to the need for tight dissolved-
oxygen and pH control despite interactions and interferences and the
advantages of having online estimators to detect deviations of batch con-
ditions and cell and product concentration. Similarly, the analysis should
be extended to substrate and nutrient feed systems, antifoam systems,
vent systems, and utility systems. Figure 4-8 shows the main control loops
for a batch fermentor.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

PC
FC 1-1
1-3 RSP
PT
1-1
Vent
FT
1-3 AC
Anti-Foam 2-2
Dissolved

Reagent
Oxygen
VSD
TC RSP TC
2-1 2-2

steam coolant
TT
TT 2-2
FC 2-1
1-2 pH
AT
AC AT 2-2
2-1 2-1
FT
1-2 Batch Fermentor
Substrate

tempered water
RSP FC
1-1

FT
1-1
Air
Figure 4-8. Batch Fermentor Control System C S
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Multivariate Analysis
Although the methods for monitoring plant data have traditionally been
limited to statistical process control (SPC) or univariate statistical analysis,
manufacturing industries have made good use of these methods to iden-
tify and monitor key quality variables. Today, the volume of data being
extracted by automation systems has increased by orders of magnitude.
The drive to understand quality coupled with the vast amounts of data
can present a formidable task.

What are the factors that make process data analysis a challenge?

• Correlation: Although we may be monitoring hundreds of

variables, there are not hundreds of independent events occurring.
The significant correlation in the variables, with respect to each
other, points to only a few underlying themes.
• Dimensionality: The average chemical plant now records and
stores thousands of variables every minute in a plant historian. This
volume of data is difficult to handle, forcing people to look at only a
small subset of variables, and thus only a fraction of the overall
information.

Copyright International Society of Automation

And what are the elements that make PCA and PLS good multivariate
techniques for modeling plant behavior from real-time operating data?

• Handle large volumes of data

• Deal with highly correlated variables by reducing the space to a set
of independent factors
• Robust to missing data and noisy data
• Ability to drill down and understand contribution of plant
variables
• Standard set of graphical tools to represent results

The flow of the MSPC analysis path is quite simple.

1. Build a correlation model of your process

Although a process may contain hundreds of variables that are
being monitored by plant information systems, there are never
hundreds of independent occurrences. Most of the data moves in a
correlated fashion, which is the basis of our model. MSPC will
form new variables, called latent variables, which are simply linear
combinations of the existing plant variables, but which are selected
to model the greatest directions of variability in the data. These
latent variables are generated to not only extract the maximum
degree of variability from the data, but to do so such that the new
latent variables are independent of (orthogonal to) each other.
Often, a very large portion of the variability in plant variables can
be summarized and compressed into a relatively small number of
latent variables.

2. Monitor the reduced set of latent variables

Reducing the order of the problem is the key to MSPC. If the vari-
ability in hundreds of plant variables can be captured and modeled
with several new latent variables, our task is much more manage-
able.

3. Identify deviation from normal plant behavior

As with univariate SPC, MSPC relies heavily on visualization and
graphical tools to monitor the reduced set of latent variables.
Although the new latent variables do not have any physical mean-
ing, tracking these variables in a control chart fashion can clearly
identify deviation from normal behavior.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

4. Identify which and how the original plant variables contribute to the
deviation
One of the principal strengths of the MSPC approach is the ability
to drill down and identify how the original plant variables contrib-
ute to the abnormal behavior in plant performance. Knowing how
the original plant variables contribute allows action to be taken to
bring the plant back within control.

In order to better understand the process described above we can graphi-

cally interpret MSPC applied to a simple process monitoring 3 process
variables: temperature, pressure and flow. Time-series plots of the vari-
ables are shown in Figure 4-9 with no apparent abnormal behavior. The
data can also be plotted in a 3-D space, as shown in this figure.

Temperature (X1)

Pressure (X2)

Flow (X3)

Figure 4-9. Plots of Process Variables

The first step in MSPC is to draw a line through the data in the direction of
maximum variability. This line is called the first principal component and
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

projecting the data points onto this component defines our first latent vari-
able. A second line can now be drawn through the data, orthogonal to the
first (linearly independent), and in the direction defining the second great-
est variability as shown in Figure 4-10. This is known as the second princi-
pal component, and again the data points can be projected onto this
principal component to generate latent variable 2.

These 2 new principal components now form a new plane in our data
space, which is commonly referred to as a scores plot. The scores plot for
our temperature, pressure and flow data is shown in Figure 4-11; it cap-
tures over 92% of the variability in the dataset. The scores plot has reduced
the order of our data from 3 dimensions to 2 dimensions. Monitoring how
the data points move on this plot is one of the graphical tools used in iden-
tifying abnormal behavior.

Copyright International Society of Automation

PC1

PC2
X1

Figure 4-10. First and Second Principal Components

5
4
3
2
1
PC-2
Scores 0
-1
-2
-3
-4
-5

-20 -15 -10 -5 0 5 10

PC-1 Scores

Figure 4-11. Scores Plot Shows Contribution of Principal Components and a

Strong Outlier

For example, the temperature, pressure and flow data show significant
deviation from model norm, behavior not easily detected from the time
series trends. Although we have limited our analysis to reducing 3 vari-
ables down to 2 principal components, we can often reduce hundreds of
variables into only several principal components and still capture a signif-
icant amount of the variability in the dataset.

In any statistical online monitoring application, graphical output and

visualization of the data are important tools for identifying abnormal
behavior. In MSPC the most common plots include the scores plot, the
model residual plot and contribution plots.

Scores Plot
The scores plot allows the observations to be viewed in the new co-ordi-
nate system of principal components. The new principal components will

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

form a sub-space, and projecting the observations down onto this sub-
space will allow us to visualize how our observations relate to our PCA
model of the process. A common technique for real-time monitoring is to
connect the observations in the scores plot, forming a worm of a preset
length. Every new observation adds a value to the head of the worm,
while an older observation is dropped from the tail. Following the direc-
tion of the worm can indicate when the process is trending towards abnor-
mal behavior.

Hotelling T 2
The Hotelling T2 is a statistical parameter indicating variability of the mul-
tivariable process. It can be used to classify when an observation in the
data is a strong outlier. A strong outlier in the data indicates abnormal
behavior. The Hotelling T2 statistic is most often represented as an ellipse
on the scores plot with 95% or 99% confidence intervals. Thus, any obser-
vation outside of these limits would indicate a strong outlier that does not
conform to the normal correlation model of the process data. Figure 4-11
illustrates the result of the Hotelling T2 classification for a process with
two principal components. Prediction Error or Model Residual Plot

The prediction error is an indication of a moderate outlier and is com-

puted as the residual for each observation. In MSPC, the prediction error is
usually plotted in time with a critical tolerance level.

Contribution Plot
One of the many advantages of the PCA approach is that the information
in the model is open for inspection. If a situation is detected the system can
present how the original variables are contributing to the detected fault.
These are most often presented as a contribution difference from model
center or contribution difference between two observations.

Although this treatment of the MSPC topic has been limited to the moni-
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

toring and analysis continuous systems, the techniques can also be

applied to batch modeling and monitoring. In the analysis of batch data,
many batch variables can be modeled using the techniques of PCA to gen-
erate a set of new latent variables. Analysis of known good batches can
then generate a series of latent variables that define the golden batch tra-
jectory and target path for the batch process. This is known as intra-batch
modeling and helps provide early detection of abnormal batch operation,
within a specific phase of the batch. Moreover, MSPC has also been effec-
tively applied to the monitoring of overall batch performance, with the
ability to condense the overall batch information such that it can be com-
pared to several other batches.

Copyright International Society of Automation

Rules of Thumb
Rule 4.1. — Make sure the scan time of the I/O, module, and performance moni-
toring system is faster than 1/5 of the dead time for fault diagnostics. This will
prevent misalignment of events and aliasing of loop oscillations because it
will insure at least 10 samples per oscillation for dead time–dominant
loops and 20 samples per oscillation for self-regulating processes with a
large process time constant.

Rule 4.2. — To analyze items 1a through 1e in the General Procedure, the scan
time should be as fast as the input card or field device to avoid improper identifica-
tion of periods of oscillation (aliasing) and the sequence of events. The speed and
richness of diagnostics from Fieldbus devices can be an important advan-
tage.

Rule 4.3. — To track down the cause of oscillating control loops, locate the loops
with significant power at the same frequency on a process flow diagram (PFD)
and trace the direction of manipulated flows between the loops. The loop from
which the manipulated flow originates is the source of the oscillations. It is
generally, but not necessarily, the furthest loop upstream.

Rule 4.4. — Watch out for periodic disturbances. Aggressive tuning of level
loops, valve stick-slip, too small a reset time on loops dominated by a time
lag (reactors, fermentors, evaporators, crystallizers, and columns), and too
high a controller gain for loops dominated by a time delay (webs, sheets,
and pipelines) are the most common causes of a periodic disturbance. If
the loop manipulates an inlet or outlet flow to the volume, the oscillating
loops will be upstream or downstream, respectively.

Level loops on surge and feed tanks tend to be tuned too tightly. Level
measurement noise is a frequent problem, particularly if a high gain or
any rate action is mistakenly used. Valve stick-slip often appears as a
square wave limit cycle. If the reset time is less than the oscillation period
for a loop dominated by a large time constant, the reset time is probably
too small. These loops need to overdrive the manipulated variable and
require more gain than reset action (see Chapter 2). Conversely, the pulp
and paper industry tends to have loops that are dominated by large time
delay, and consequently the use of a gain or reset time that is set too high
is more of an issue.

Rule 4.5. — If the oscillation period is less than 4 seconds, it probably originates
from rotating equipment, pressure regulators, actuator instability, burner insta-
bility, resonance, or measurement noise. Incipient surge and buzzing due to
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

dips in the characteristic curves of compressors and high-pressure pumps

are a principal cause of fast oscillations in gas and boiler feed water sys-
tems, respectively. Oversized field pressure regulators and inadequately

Copyright International Society of Automation

sized actuators and springs can cause large rapid fluctuations in flow.
Burner instability can cause rapid oscillations in fuel flow and furnace
pressure.

Rule 4.6. — Items 1a through 1i in the General Procedure can usually be handled
by performance monitoring systems that provide relative statistical measures of
variability, capability, utilization, sustained cycling, oscillation frequency and
power (Power Spectrums), saturated controller outputs, and limited process vari-
ables.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Rule 4.7. — The diagnosis and prediction of process equipment, operation,
sequence, allocation, and raw material problems (items 1j through 1s in the Gen-
eral Procedure) generally require cross correlation analysis and multivariate prin-
cipal component analysis (PCA).

Rule 4.8. — Online property estimators should be created to enhance the repeat-
ability and reliability of online and lab analysis systems by the addition of Partial
Least Squares (PLS) analysis, dynamic linear estimators, or neural networks.
Time delays and time constants must be used to align the inputs with out-
puts. Some of the need for data alignment can be reduced by the use of
long scan times but this slows down the calculation for estimation and
fault detection to a point where it may be too late or unable to resolve
what occurred first. This is a more important issue for continuous opera-
tions than batch operations since batch outputs can be tied to a batch, step,
and phase identification number and time (see Chapter 8).

Rule 4.9. — Control loops that are intentionally not in service due to batch oper-
ations, product or grade produced, or trains of equipment that are shut down to be
maintained or cleaned, must be flagged as such and other diagnostics automati-
cally suppressed. Otherwise, true problems are camouflaged by false alerts
and alarms. Also, the “Normal Mode” of loops used in such batch opera-
tions should be automatically updated to reflect the needs of the batch
sequence.

Guided Tour
This tour illustrates the potential ease of use and convenience of an inte-
grated interface that is possible for a performance monitor embedded in
an industrial control system. The following areas are addressed:

• Summary of measurement, actuator and control performance

• Examination of detail state
• Information provided to an operator

Copyright International Society of Automation

This performance monitoring capability is accessed from any operator and

engineer stations within the control system. The primary interface is
illustrated in Figure 4-12.

Figure 4-12. System Performance Monitoring Interface

From the explorer view at the left of the interface, the user may select the
entire plant or an individual process area to examine. Based on this selec-

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
tion, a summary is provided that shows the number of modules that are
being utilized for control, monitoring, and calculation within the selected
area. Also, a summary is provided of the modules that have an I/O or con-
trol block with abnormal conditions. The individual modules that have an
abnormal condition are listed in the right portion of the view in the Sum-
mary tab. The types of problems that have been detected are shown using
a frown face.

When a module is selected in the summary list, the individual I/O and
control blocks in that module are listed in the bottom right portion of the
interface. The percent time that an abnormal condition existed is shown
for each block. If this time exceeds the defined limit, then this condition is
flagged as an abnormal condition in the interface; that is, a frown face is
shown for the module. Through the Filter selections, it is possible to view
information for the previous or current hour, shift or day. Also, the user
may select the type of blocks and whether all modules or just those with
abnormal conditions are displayed.

The following types of abnormal conditions are automatically detected:

• The actual block mode is not in normal (designed) mode of

operation
• Control action is limited by a downstream condition

Copyright International Society of Automation

• Input status indicates the value is limited or has a Bad or Uncertain

status
• The calculated variability index and standard deviation exceed
their limits

In many plants, it is common for a piece of equipment to be shut down at

the end of a batch or to take it out of service for maintenance. During these
periods of down time, the measurements on that equipment may indicate
a Bad status. To prevent this information from being included in the
performance monitoring view, the user may select an area and then
indicate that the associated equipment if off-line. Such areas are clearly
indicated in the view of the equipment and are moved from monitoring.
An example where the RSU unit has been removed from monitoring is
shown in Figure 4-13.

Figure 4-13. Disabling a Selected Area of Monitoring

The user may find out more information on an abnormal condition by

using the detail tabs in the interface. For example, by selecting the Mode
tab, further information is provided on the actual and design mode of the
blocks within the module, as shown in Figure 4-14.

Figure 4-14. Detail Information for an Abnormal Condition

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

By selecting the Print icon, the user may choose to print a module sum-
mary report or a detail report. Also, to allow the operator to view the per-
formance and utilization information from his displays, a standard
function block and dynamo are used to include this information in the
operator interface. Through this dynamo an operator may also enable and

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
disable the monitoring in the associated area. This dynamo appears in an
operator display as shown in Figure 4-15.

Figure 4-15. Operator Access to Performance Information

Theory
The ability to quickly inspect control and measurement loops has a pri-
mary importance in industrial applications. Both poorly tuned loops and
malfunctioning field devices jeopardize product quality and production.
In the last decade this problem has received significant attention from both
academia and practitioners. Much of this work has focused on an assess-
ment of control loop performance using minimum variance controller as a
reference; see Harris [4.1] and Desborough and Harris [4.2] and [4.3]. Bea-
verstock et al. used heuristic definitions of unit production performance
tailored to specific applications, rather than loop performance [4.4].
Numerous other researchers have advanced performance monitoring. In
particular, Rhinehard [4.5] explored simple ways of computing standard
deviation from measurement-to-measurement deviations and filtering.
Harris et al. [4.6] and Huang et al. [4.7] provided guidelines for perfor-
mance assessment of multivariable controllers. Qin reviewed recent works
on performance monitoring in [4.8]. Shunta gave an excellent practical
summary of the performance monitoring in the monograph [4.9].

Automatic detection of abnormal operating conditions is possible with

Fieldbus architecture and Function Blocks (FB), features available in mod-
ern control systems. In older control systems, a monitoring application

Copyright International Society of Automation

may often be added as an OPC client. Based on these features, it is possible

to quantify loop utilization, “bad” or “limited” measurements, limitations
in control, and process variability. In addition, for control loops the poten-
tial improvement in loop performance is calculated. This functionality is
possible due to the Function Block features, defined by the Fieldbus Foun-
dation. Every function block has specific parameters for current block
input or output parameter status, block mode and total and capability
standard deviation. These block parameters are used by process monitor-
ing applications and for creating summary reports for hour, shift and day.

Major manufacturers of process instrumentation in the process industry

have introduced a new generation of field devices based on the standards
established by the Fieldbus Foundation. To allow interoperability of
devices from different manufacturers, a common implementation is sup-
ported using a standard set of function blocks for measurement, control
and calculations. A key concept in allowing control and calculations to be
distributed between devices or between a field device and a controller is a
common implementation of block mode supported by each function
block. Also, each input and output of a function block is designed to pro-
vide a status attribute along with a value, as illustrated in Figure 4-16.

Function Blocks Support Mode

Function Block Inputs and Outputs

Provide Engineering Unit Value and Status

Figure 4-16. Block Parameter Status and Mode are Utilized in Performance
Monitoring --`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

The status associated with each function-block output gives a direct indi-
cation of the quality to the measurement of control signal. Quality is
defined by describing the measurement as suitable for control (Good),
questionable for use in control (Uncertain) or not suitable for use in con-
trol (Bad). In addition, the status provides an indication of whether a mea-
surement or control signal is high or low limited. For example, if a
measurement is operating above its calibration range, the quality of the
measurement may be shown as “Uncertain.” Downstream blocks may

Copyright International Society of Automation

provide a status of “Good Limited” to the upstream control block to indi-

cate that a valve is at its upper or lower end of range.

The modes of Fieldbus blocks consist of four attributes, which may be

very valuable in examining the operation of a control block. When a con-
trol block is initially configured, the “Normal” attribute of the mode
parameter is defined based on the design mode of the block. During nor-
mal operation, the operator sets the “target” mode attribute to enable or
disable automatic control and to indicate whether the operator or another
block or application has control of the block set point and block output.
The block reflects the mode of operation it can achieve in its “actual”
mode attribute based on the target mode and the status of inputs to the
control block. By comparing the actual mode to the normal mode attribute
it is possible to determine if a function block is operating in its designed
mode of operation.

By continuously monitoring the status and mode supported by Fieldbus

function blocks, it is possible to automatically determine the following
abnormal conditions in control and I/O function blocks:

• Bad I/O — the status of the block process variable (PV parameter) is
“bad,” “uncertain” or “limited.” A sensor failure, inaccurate
calibration, or measurement diagnostics have detected a condition
that requires attention by maintenance.
• Limited (control action) — a downstream condition exists that
limits the control action taken by the block. Such limitations may
prevent the loop from achieving or maintaining set point.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
• Mode not Normal — The actual mode of the block does not match
the normal mode configured for the block. An operator may change
the target mode from normal because of equipment malfunction.

The percent time that these conditions exist over an hour, a shift, and a day
is computed for every block and compared to a configured global limit for
each condition. When one of these limits is exceeded, the associated
module is displayed in the main summary display.

It is possible that the status of all inputs to a control block is normal and
the mode of the block is correct and yet the control provided is poor. No
indication is included in the standard block defined by Fieldbus that
directly indicates this problem. However, statistical techniques exist that
allow the quality of control to be determined in a reliable manner. Leading
companies in the process industry use such techniques to evaluate the per-
formance of control loops[4.9]. Based on a knowledge of total and capabil-
ity standard deviation of the control measure, illustrated in Figure 4-17, it
is possible to compute a variability index for control measurements that

Copyright International Society of Automation

compares current control performance to the best achievable for the pro-
cess dynamics.

Process
Value

Time

“Total”: All Data

“Capability” Only Random or
Including Short and
Short Term Variability
Long Term Variablity

Figure 4-17. Capability and Total Standard Deviation for Process Variable

However, to accurately determine total and capability standard deviation,

it is necessary that the measurement of fast processes be sampled at a rate
that can not be achieved by the communication networks of most DCS sys-
tems used in plants today. Thus, in the past, the application of this technol-
ogy has been limited to the analysis of slower control loops or to special
dedicated tools that support the sample rates required for faster processes.
It is now possible for the manufacturer of the Fieldbus device and modern
control systems to build these calculations into the control and I/O func-
tion blocks to support control analysis. By taking this approach, the con-
trol of even the fastest process may be accurately evaluated.

Using Statistics for Control Performance Evaluation

The valuable role that statistics can play in evaluating control performance
is well known. Standard deviation of the process measurement (Equation
4-1) is commonly used and is an easy-to-understand measure of process
variability
n
2
∑ ( Xi – X )
S tot = i=1
-------------------------------
- (4-1)
n–1
where n is the number of sample, typically n = 120.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

∑ Xi
i=1 -
X = -------------- (4-2)
n
When loop performance is of concern, then standard deviation alone may
not provide sufficient information to evaluate loop tuning. To gain an
objective judgement, a reference value for the process control performance
is required. The best performing feedback control theoretically is mini-
mum variance control, Sfbc. This value may be calculated [4.9] directly
based on the knowledge of the total, Stot, and capability, Scap, standard
deviation of the process measurement as shown in the Formulas 4-3 and
4-4.

S cap 2
S fbc = S cap 2 – ---------- (4-3)
S tot

Scap is calculated from process samples XI as :

∑ ( Xi – Xi – 1 )
2

S cap = i=2
---------------------------------------
- (4-4)
2(n – 1)
Based on a the value of total standard deviation and standard deviation
for minimum variance control, it is possible to define a variability index
for the control loop that reflects how close control performance comes to
minimum variance.

 S fbc + s 
Variability Index VI = 100  1 −  (4-5)
 Stot + s 
where s is the sensitivity factor.
Finally, for control blocks, overall control performance is as follows:

Performance Index (PI) = (100-VI) (4-6)

The overall control utilization is an indication of what percent of time
loops were in normal mode:

Utilization = Average Percent Time in Normal Mode

To support the VI and PI calculations, the I/O and control function blocks
in the scalable system and Fieldbus devices calculate Scap and Stot and
make them visible as block parameters. Intermediate calculations are done

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

each execution of the function blocks. The parameter values are then
updated per every n executions of the function block. For a typical imple-
mentation, the update is done after 120 executions of the function block.

To simplify calculations and memory requirements, the total standard

deviation, Stot,, can be calculated by mean absolute error, MAE, computa-
tion as follows:

S tot ≅ 1.25MAE (4-7)

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
where:
N
1
MAE =
N
∑ | ( y (t ) − y ) |
1
t (4-8)

The measurement value is used in I/O blocks to calculate the mean value.
In control blocks, either the working set point or the measurement value is
used depending on block mode.

The relation between standard deviation and a mean absolute error can be
verified by computing a mean absolute value for the normalized Gaussian
distribution, p(x):

∞ x2
1 −
p( x ) =
2π ∫e
−∞
2
(4-9)

 x2 0 ∞

1 
∞ x2 0 x2 ∞ x2  x2
1 − − − 1  −2 −  2
| x|= ∫ | x|e 2
= − xe 2 + xe 2  =
∫ ∫ e −e 2
 = π = .7978845 (4-10)
2π 2π  −∞ 
 2π  
−∞ 0
 −∞ 0 

Since for the normalized Gaussian distribution standard deviation σ=1

| x | = .7978845σ or σ = 1.2533141 | x | (4-11)

The capability standard deviation Scap , is calculated as follows:

MR
S cap = ------------- (4-12)
1.128
where:

1 N
MR = ∑ | ( y(t ) − y (t − 1)) |
N −1 2
(average moving range) (4-13)

Only the summing component associated with the MAE and MR is done
each execution. The division of the sum by N or N-1 is done as part of the

Copyright International Society of Automation

Stot, and Scap calculations only once every N executions (default N=120).
Capability standard deviation calculation requires that the sampling rate
be fast enough. The requirements for the sampling rate are similar to the
scan rate of a control loop. A practical estimate for selecting scan rate for
control loops is sampling five or more times per time constant. [4.9].

Extending the Concept to the Multi-variable Environment

In many products designed for control and measurement system perfor-
mance evaluation, the scope of detection is limited to traditional I/O and
traditional control capability. This capability has been extended to address
I/O and control blocks in Fieldbus devices and to provide interfaces and
analysis tools for high-speed trend collection, device diagnostics and tools
for commissioning control. Since MPC control and Neural Network capa-
bility are being embedded in modern control systems, it is a natural exten-
sion of this work to include performance evaluation support that includes
advanced control.

For the purpose of performance evaluation, advanced control can be

defined as “monitoring or control applications that use multiple process
measurements to produce one or more outputs that are used directly or
indirectly to improve the operation of a plant.” As illustrated in Figure 4-
18, this definition fits the advanced control applications that are currently
used in industry today: model predictive control, fuzzy logic control,
adaptive tuning, soft sensors based on neural networks, plant optimizers
(such as RTO+), multi-stream blending.

TT AT FT Adaptive
1-1 2-1 3-8
Tuning
Model
FT Neural FT Predictive
1-2 Network 2-9 Control
AT TT TT
Control
1-2 2-5 3-8

FT FT
FT
4-3 6-2
5-7
Multivariable Real Time Blending
FT AT FT
Fuzzy Logic Optimization
4-6 5-8 6-6
Control
LT FT FT
4-6 5-8 6-8

Figure 4-18. Example of an Advanced Control Applications

In spite of the diverse technologies associated with each of these applica-

tions, their requirements for correct operation are basically the same as
those for base-level measurement and control.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

• Valid measurements (good and not limited).

• Control action is not limited by downstream conditions.
• The operator and operating conditions allowed the algorithm to
operate as designed; i.e., the application is in the correct mode of
operation.
• Control or calculation objectives were met.

When compared to the measurement and control-blocks requirement

addressed in single loop control, the primary differences are (1) multiple
inputs (rather than one) are used for the control or calculation and to
indicate downstream limitations, and (2) the measurement of control
performance is based on multiple inputs and targets.

Thus, to extend system evaluation to address advanced control applica-

tions, the techniques used for single loop control to determine abnormal
input, control limited, incorrect mode, and control index exceeds limit
must be generalized to include multiple inputs and targets.

Addressing Advanced Control

The detection of abnormal conditions or poor performance of an advanced
control application should take into account the multiple inputs and tar-
gets used in the control or calculation. Also, in many cases multiple inputs
must be examined to determine if control has been limited by downstream
conditions. Thus, the four basic indications used by system performance
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

evaluation to determine abnormal or under-performing control must be

generalized to address the following:

1. Inputs used in the application are abnormal (“bad,” “uncertain,”

or “limited”).
2. Inputs associated with downstream conditions indicate control
action was limited.
3. The algorithm was not able to operate as designed; i.e., its mode of
operation was not normal.
4. The process output did not meet the target-operating objective.

In this section, the manner in which each of the calculations may be

generalized to include multiple inputs and targets is detailed.

Abnormal Inputs
For a single input control block such as the PID or FLC control block, the
status of the primary input (PV) of the block may be used to determine if
the input is abnormal. For example, if the PV status is uncertain or “bad”

Copyright International Society of Automation

or is “limited,” then the block will be flagged as having an abnormal

input. In advanced control applications in which multiple inputs may be
required for correct operation, any of these inputs that have a status of
uncertain, bad or limited may prevent the control from achieving its goal.
Thus, for advanced control applications, the indication of abnormal
input(s) will be based on all inputs that directly affect the control or calcu-
lation. This may be calculated as follows:

Abnormal Input Condition Advanced Control =

∀ Abnormal Condition on Input I or block
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

where:
∀ = a logical sum of I = 1, …, N Conditions

Control Limited
Where advanced control is providing multiple process inputs, the control
interface to the process will be done through I/O or control blocks. Both
provide an output that will be used as a back-calculation input to the
advanced control application. If the control action taken is limited down-
stream, then this is reflected in the status of the associated back-calculation
input. Since the control objective will not, in general, be met if any of the
control outputs becomes limited, then the indication that control is limited
must consider the back-calculation input status associated with all control
outputs. This may be calculated as follows:

Control Limited = ∀ Control Action (I) is downstream limited

where:
∀ = a logical sum of I = 1, …, N Conditions

Incorrect Mode
For single input/output blocks, the status of the primary input, cascade
input, back-calculation input, and target mode must be used in determin-
ing achievable mode of operation. This mode of operation is reflected in
the mode parameter as the actual mode attribute. When a block is config-
ured, the customer may indicate the normal mode that the block is
designed to operate in. By comparing the actual mode attribute to the nor-
mal mode attribute for the block, it is possible to determine whether the
block is operating in its designed mode. Advanced control applications
may be engineered to use the standard status and mode definition. To cal-
culate mode, the status of all inputs used in the control and the status of all
back-calculation inputs will be utilized. Each output provided by the
advanced control application will be designed to support handshaking,
bumpless transfer, and windup prevention based on the back-calculation

Copyright International Society of Automation

inputs. Thus, mode will be treated exactly the same as other control blocks
in the control system. Incorrect Mode may be determined as follows:

Incorrect Mode = ∀ Incorrect Mode ( I )

where:
∀ = a logical sum of I = 1,…N Conditions where the input or
output are required to continue control.

Control Index
For single loop PID and FLC control, the variability index is calculated
based on the total and capability standard deviations calculated in the
control block. In an advanced control application involving multiple
inputs and targets, the measure of control performance must consider
each controlled input compared to its target value. For advanced control
applications, such as MPC or multi-variable fuzzy logic, this concept may
be extended to calculate an average index or minimum value of the index:

L
CIA = 1
L∑
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

CI ( i )
1

CIM = min CI(i)

Control Index for Advanced Control Application is an average or a
weighted average of control indices for individual controlled variables. A
maximum value of control index may be used as well in some applica-
tions.

We compute CI for a particular input in a similar way as for a single loop

control.

Diagnostic Tools
Insight into the source of variation may be obtained using tools that sup-
port the calculation of power spectrum and cross correlation:

• Power spectrum shows how measurement variability is distributed

over different frequencies.
• Cross correlation reveals dependencies between two process
variables.

The power spectrum may be calculated from the Fourier series coefficients
as illustrated in Figure 4-19.

The calculation of cross correlation is illustrated in Figure 4-20.

Copyright International Society of Automation

Trend ( time series)

+ +
+
+
+ + + +
+ +
+
Fourier Series
+
+
n
X (t ) = ∑ ( Ai cos( wi t ) + Bi sin( wi t ))
+ + +

i =1

i 1
Where wi = 2π
N ∆T
N = Number of po int s collected
Power Spectrum N
n=
+
2

Power +
+
Amplitude Pi at frequency wi is :
Pi
+ Pi = Ai2 + Bi2
+

+ +
+

Frequency Wi

Figure 4-19. Illustration of Power Spectrum Calculation

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Y (t ) + +
+
+
+
+
+ +
+
+ Cross Correlation
+ +
+
+ + +

X (t ) +
+
+ +
+ + +
+ + + + +
+ + + +
N −k
1
Time
N
∑(X i − X )(Yi + k − Y )
C xy (k ) = i =1
σ xσ y
+
+ Where N = Number of samples
Cxy +

K = 0,1,2,..., N − 1
+
+ +

+ +

+
+ + +
+ +
+ +

Time shift K

Figure 4-20. Illustration of Cross Correlation Calculation

System Communication Model

Selected function-block parameters in the system controller are reported
on a change basis to the performance monitoring application. This appli-
cation may be structured as a server in the control system’s main worksta-
tion. The information resulting from the calculation in the server may be
accessed through the client applications provided on other system work-
stations, as illustrated in Figure 4-21.

The status and actual mode attributes used by the performance monitor-
ing system to calculate loop utilization, limited condition, and bad mea-
surement normally do not change in value. Thus, communication
requirements may be minimized by reporting parameters only on a
change in these attribute values [4.10], [4.11]. If this approach is taken in
the system design, then the communication load for reporting these
parameters will normally be close to zero. Performance statistics may be

Copyright International Society of Automation

calculated over a specified period of time, e.g., 120 executions of the con-
trol or measurement block, and then reported to the performance monitor-
ing application. Thus, these parameters are reported every 60 seconds for
a block with an execution rate of 0.5 seconds.

Scalable System
Main Workstation Workstation Workstation
Server Client Client
Application Application Application

PM - FB =
Parameters
PM - FB Process/Performance
PM - FB Reported by PM - FB
Parameter Monitoring
Parameter Exception Parameter
Function Block
Scalable Scalable Scalable
Controller Controller Controller

Figure 4-21. Process Monitoring Communication Diagram

When the server is first placed online, the current state of the required
attributes is reported once and subsequent updates are sent by exception
reporting.

The support of fast speed trend for diagnostics places a special require-
ment on the control system design. Inaccuracies may be introduced by jit-
ter and aliasing as a result of measurement sampling:

• Jitter — time between sampling varies; i.e., sampling is not done on

a precisely periodic basis.
• Aliasing — sampling rate is less than twice the highest-frequency
component of the process measurement.

Jitter and delays may be introduced into data collected by a diagnostic

application that polls for information in the controller through the control
system communication network. The information captured in the system
historian for analysis is assumed to be reported on a periodic basis. If the
information is delayed in communication to the PC, then jitter occurs, as
illustrated in Figure 4-22.

The control system communication speed and response to request have

often limited the rate at which information in the controller may be
accessed by a diagnostic application, causing aliasing. To avoid aliasing,
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

the sample rate must be at least twice as fast as the highest frequency in
the sampled signal. If this is not possible, then aliasing occurs, as illus-
trated in Figure 4-23.

Copyright International Society of Automation

x
Measurement
x
Value Plotted
Assuming
Uniform
Sampling

x
x
Variation
in time of
sampling x

x
Sample
Taken Time

Figure 4-22. Impact of Jitter on the Signal Used in Diagnostics

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
x x

Measurement

Value Plotted
Assuming
Uniform Sampling

Periodic
Samples Time

Figure 4-23. Aliasing Occurs by Sampling at a Slow Rate

Historically, the only way to collect high-speed data for diagnostics was
dedicated tools that attached at the terminal strip of the control system, as
illustrated in Figure 4-24.

Scalable System
Workstation (s)

Scalable
Controller
Temporary
Wiring to I/O
I/O File terminations

PC Based Diagnostic
FT Tool With I/O
9-1

Figure 4-24. Traditional Diagnostic Tools

Copyright International Society of Automation

However, many Fieldbus devices introduced in the last few years support
the collection of high-speed data for diagnostics. Also, some modern con-
trollers allow measurement and control parameters to be collected in the
controller for diagnostic support. This trend information collected in the
Fieldbus device and controllers may be accessed without aliasing or jitter
and thus used within the control system for diagnostic support, as illus-
trated in Figure 4-25.

Scalable System
Workstation (s)
Diagnostic
Application
Communication Network

Scalable
Controller

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---
Trend Blocks
Trend objects

I/O File

Smart or Fieldbus
traditional
4-20ma TT
FT 1-1
9-1

Figure 4-25. Capture of Trends in Field Devices and Controllers

References
1. Harris, T., “Assessment of Control Loop Performance,” Can. J. Chem. Eng.,
1989, 67(10):856-861.
2. Desborough, L., and Harris, T.J, “Performance Assessment Measures for
Univariate Feedback Control,” Can. J. Chem. Eng., 1992, 70:1186.
3. Desborough, L., and Harris, T.J, “Performance Assessment Measures for
Univariate Feedforward/Feedback Control,” Can. J. Chem. Eng., 1993, 71:605.
4. Beaverstock, C. Malcolm, and Martin, Peter G, “Performance Control
Apparatus and Method in a Processing Plant,” US Patent Number 5,134,574,
July 28, 1992.
5. Rhinehart, R. Russell, “A Cusum type on-line filter,” Process Control and
Quality, 2 (1992) 169-179.
6. Harris, T., Boudreau, F., and Macgregor, J. F., “Performance Assessment of
Multivariable Feedback Controllers,” Automatica, 1996, 32(11):1505-1518.
7. Huang, B., Shah, S.L., and Kwok, K.Y., “Good, Bad or Optimal? Performance
Assessment of MIMO Processes,” Automatica, 1997, 33(6): 1175-1183.
8. Qin, S.J., “Control Performance Monitoring – A Review and Assessment,”
NSF/NIST Workshop, New Orleans, March 6-8, 1998.

Copyright International Society of Automation

9. Shunta, Joseph, “Achieving World Class Manufacturing Through Process

Control,” 1995, Prentice Hall PTR (ISBN 0-13-309030-2).
10. Blevins, T., Wojsznis, W., Nixon, M., Riley, K., and Thiele, D., “Process
Monitoring Benefits From Fieldbus Architecture,” ISA Conference, 1999.
11. Blevins, T., and OChoa, D., “Statistical Analysis Rides the Bus”, Intech,
March, 2000, pp.48-50.
12. Dudzic, Michael, Miletic, Ivan, Quinn, Shannon, and Vaculik, Vit, “The
Application of Advanced Statistical Technologies (Multivariate Statistics) in
On-Line Monitoring and Control,” Process Automation Technology, Dofasco,
Hamilton, Ontario, Canada.

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Practice

Overview
In modern control systems, expert system technology is playing an ever-
increasing role in assisting the operator in the detection and management
of abnormal situations in a plant. With the introduction of distributed con-
trol systems in the late 1970’s, the basic control systems of many process
plants went through major changes in organization and operation. In
many cases, the introduction of distributed control allowed control func-
tions to be concentrated into a few control rooms. The traditional control
panels for operator interface to the process were replaced with keyboards
and monitors. There were few limits on the amount of information that
could be accessed and displayed at these operator stations. These systems
allowed an operator to make changes and see the process alarms associ-
ated with his area of responsibility. In some cases, the system was
designed to allow all information about the plant to be accessed from any
terminal within the system [5.1]. As a result of this technology change, and
the increasing pressure on companies to increase productivity, the scope of
control that an operator was responsible for changed dramatically.

In one pulp and paper mill the introduction of a distributed control system
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

allowed three control rooms to be consolidated into one and for one oper-
ator to do the job formerly done by three [5.2]. In some process areas an
operator is responsible for as many as thousands of measurements and
hundreds of motors and control loops in addition to various subsystems in
a process area. Thus, it has become increasingly difficult for an operator to
be aware of all conditions in the plant. During normal operations, there is
insufficient time for an operator to examine all measurements in his area
163

Copyright International Society of Automation

to determine if an abnormal condition exists. He depends to a great extent

on process alarms to alert him to abnormal conditions. Through the early
detection of abnormal conditions using expert systems, it is possible to
improve plant operation or to prevent a failure that could cause a process
shutdown. Also, during shutdown and startup conditions, expert systems
can be used to draw the operator to abnormal situations that might affect
plant operation.

Some of the areas of abnormal situation management that an expert sys-

tem can effectively addressed are the following:

• Alarm screening—presentation of the root source of an alarm

condition so that the operator is not distracted from the critical
events that are associated with an equipment failure.
• Detection of process change—Degradation of equipment
performance because of improper equipment setup and wear or
buildup in pipes and heat-transfer surfaces may alter the behavior
of equipment. The automatic detection of significant process
changes helps prevent equipment damage and associated
production loss.
• Detection of abnormal conditions—The automatic detection of
patterns in plant operation that indicate an abnormal condition
may bring this information to an operator’s attention sooner that
process alarms.

Expert systems may also be used to automatically initiate corrective action

on detection of equipment failure. For example, expert systems are used to
prevent a condition known as alarm flooding. Process value alarms have
traditionally been used in a control system to alert the operator when a
critical measurement has deviated from its normal operating range. Also,
deviation alarming is commonly used in control applications to alert the
operator when a controlled parameter has deviated from set point. In
addition, alarms often accompany the detection of an interlock condition
or malfunctions such as high vibration or interlock active in motor control.
Under normal plant operations, the operator may easily address the
occurrence of individual process alarms. However, the failure of a major
piece of equipment that is critical to an operation, such as a motor, may
result in hundreds of process alarms being activated at once.

With the flood of alarm information, it is difficult for the operator to

quickly determine the source of the alarms if, for example, a motor
tripped. Also, in this flood of alarms, it is difficult to determine which
alarms indicate conditions that need to be immediately addressed to
insure safety or to prevent damage to process equipment. If an expert sys-
tem is monitoring individual events that indicate equipment failure, it is

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

possible for the expert system to automatically suppress other process

alarms that were triggered by the failure. This allows the operator to focus
on the fault and more effectively address the source of the problem.

A combination of individual events or process conditions may indicate a

problem. If an operator misses this relationship, this condition may go
undetected. For example, the detection of high discharge pressure on a
variable-speed pump in combination with low flow indication may indi-
cate a restriction in the downstream line, e.g., a blocking valve is closed. If
an expert system is used to detect this combination or pattern, it can auto-
matically alert the operator that such a condition exists. Through this early
detection, it is possible to avoid equipment damage or lost production.

Such proactive detection of abnormal conditions is especially helpful dur-

ing the startup or shut down of a complex process. Under these condi-
tions, the operator has to take a number of manual actions. His job is
complicated by the fact that it can be months or years between process
shutdowns and startups and thus he must often quickly deal with opera-
tion conditions that he seldom sees during normal operation of the plant.

Opportunity Assessment
Expert systems have been successfully applied in a variety of applications.
Within the process industry, there is significant benefit in using this tech-
nology for abnormal situation management. In assessing the potential
benefits of this technology, the following questions should be asked:

1. Are there areas of the plant operation where unscheduled

shutdowns commonly occur? Could these shutdowns have been
prevented if the operator had taken action on the available
information?
2. Is the process characterized by high maintenance of
measurements, control valves, or equipment? Would continuous
monitoring for abnormal conditions help reduce loss of production
or avoid unsafe processing conditions
3. Has the operator’s area of responsibility been expanded to the
point where he does not have time to proactively look at process
conditions? Are there cases where operations could have been
improved if the operator had taken advantage of process
conditions to increase throughput or improve operating efficiency?
4. During the startup of a process area, does it take longer to get to
normal operating conditions that would be expected? Would
detection of abnormal conditions during startup help the operator
reduce the time to complete startup?

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

5. When a piece of equipment fails, does the operator have difficulty

determining the source of the problem? Would automatic
screening of alarms help the operator react faster to these
situations?
6. Are batch cycles put on hold or new batch cycles delayed because
operations has difficulty determining the phase end points or the
cause and validity of failure expressions in the batch sequence?

The cost of expert-system software has dramatically decreased since the

initial introduction of commercial expert systems in the early 1980’s. In
most cases expert-system software can execute on a PC that uses a
common Microsoft operating system. However, interfacing these software
packages to the control system to allow the use of real-time data in
abnormal condition monitoring is often a time-consuming and costly
adventure. Also, there are wide differences in the ease with which the
knowledge associated with an application is entered into the expert
system. Often, more effort goes into the design and implementation of the
user interface than into the expert system itself. Many of these
implementation issues can be addressed by embedding the expert system
within the control system.

Examples

Alarm Screening
Expert systems are used in areas of the refining industry for abnormal sit-
uation management. One such use is in the prevention of alarm flooding.
For example, the regeneration unit of the hydrocracking process is vital to
plant operation and production. The interactive nature of the large num-
ber of measurements and control loops associated with this unit means
that a failure of one piece of equipment may result in the operator being
presented with many alarms: the original failure plus the alarms associ-
ated with measurements that the equipment affects. Under these condi-
tions, it is of great help to the operator if the alarm associated with the
failure is clearly presented and other alarms resulting from this event are
only logged, not presented to the operator. An expert system is used to
look for specific equipment failures and to automatically suppress other
alarms triggered by the failure. Under normal operating conditions, all
alarms would be active; the expert system only suppresses alarms when
specific operating conditions are detected.

As part of the development of an expert system, an expert on the regener-

ation process needs to define what types of failures on that process will
benefit from alarm suppression. For example, one condition that might be
identified is the tripping of the blower motor. When the blower motor

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

trips, a number of other upstream and downstream alarms are also gener-
ated that would make it difficult for the operator to quickly identify the
problem areas, as illustrated in Figure 5-1.

Figure 5-1. Alarms Following a Regenerator Blower Trip

The regenerator measurements and control loops that support alarming

are defined as facts that are available in the expert system. The fact associ-
ated with the blower motor includes a slot that reflects the status of the
blower motor. This fact is dynamically updated through the interface to
the control system. Only two rules are required to automatically change all
measurement and control-loop alarm priorities in the control system so
they are suppressed at the operator interface when a blower trip is
detected. Using the ability to define variables in these expert rules, it is
possible to implement rules for alarm suppression independent of the
number of measurements or control loops associated with the regenerator
unit. Based on this design, alarming of associated measurements and
loops alarms caused by the blower trip are automatically suppressed
when a blower trip is detected, as illustrated in Figure 5-2.

Fault Detection
An oil field may contain hundreds of wells. The early detection of an
abnormal condition such as blocked flow in the wellhead may avoid dam-
age to the associated pump. Because such conditions are often indicated
by a combination of measurement values, the traditional value or devia-
tion alarming cannot be used to alert the operator of them. However, by

--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Copyright International Society of Automation

Figure 5.2. Alarm Screening Allows the Source of Failure to be Identified

using an expert system to monitor the wells, the conditions that indicate
abnormal operation are detected and brought to the operator’s attention.

To implement the expert system, facts would be defined for the measure-
ments that are included in a production control system. To detect blocked
flow, one rule would be written that examines the conditions that indicate
blocked flow; e.g., oil flow and the pump pressure. By using the variable
definitions in both the left and right portions of the expert rule, it is possi-
ble to monitor all the wells with one rule. As wells are added to or
removed from the system, the only change required would be to update
the facts; the rule would not have to be modified. When a blocked-flow
condition is detected, the rule is designed to write to parameters in the
control system that cause the detected condition to be alarmed and dis-
played at the operator interface. An example of how this would be dis-
played to the operator is shown in Figure 5-3.

Application

General Procedure
1. Identify the areas to be addressed by the expert system:
a. Select applications that have significant impact on plant
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

operations.

b. Quantify potential benefits based on past operations.

Copyright International Society of Automation

Provided by IHS under license with ISA Licensee=PDVSA - Complejo Petroquimico Site 12/9986712018, User=Cordova, Daniel
No reproduction or networking permitted without license from IHS Not for Resale, 10/11/2016 [Link] MDT
Chapter 5 – Abnormal Situation Management 169
--`,```,,,```,`,````,``,`,,`,`-`-`,,`,,`,`,,`---

Figure 5.3. Operator Alerted to Condition Detected by Expert System

2. Develop a cost estimate for an expert system solution

a. Cost of equipment and software
b. Manpower and expertise required to execute the project
3. Implement, Test, and Maintain the Solution
4. Focus on the objectives established for the project
a. Test initial design concepts before doing full implementation
b. Test full prototype with detailed noisy dynamic simulation
c. Run full prototype online in computer or configuration room
d. Eliminate false diagnostics before installing system in the
control room
e. Establish procedures to monitor and maintain the system

Application Details
The following details were developed for the application of expert sys-
tems for abnormal situation management. To achieve the best results, the
user should adhere to the guidelines specified in this section.

Copyright International Society of Automation

Selection of the Problem

In selecting a problem to be addressed, the follow