Advances in Industrial Control

Series Editor
Michael J. Grimble, Industrial Control Centre, University of Strathclyde, Glasgow,
UK

Editorial Board
Graham Goodwin, School of Electrical Engineering and Computing, University of
Newcastle, Callaghan, NSW, Australia
Thomas J. Harris, Department of Chemical Engineering, Queen’s University,
Kingston, ON, Canada
Tong Heng Lee , Department of Electrical and Computer Engineering, National
University of Singapore, Singapore, Singapore
Om P. Malik, Schulich School of Engineering, University of Calgary, Calgary, AB,
Canada
Kim-Fung Man, City University Hong Kong, Kowloon, Hong Kong
Gustaf Olsson, Department of Industrial Electrical Engineering and Automation,
Lund Institute of Technology, Lund, Sweden
Asok Ray, Department of Mechanical Engineering, Pennsylvania State University,
University Park, PA, USA
Sebastian Engell, Lehrstuhl für Systemdynamik und Prozessführung, Technische
Universität Dortmund, Dortmund, Germany
Ikuo Yamamoto, Graduate School of Engineering, University of Nagasaki,
Nagasaki, Japan
Advances in Industrial Control is a series of monographs and contributed titles focusing on
the applications of advanced and novel control methods within applied settings. This series
has worldwide distribution to engineers, researchers and libraries.
The series promotes the exchange of information between academia and industry, to
which end the books all demonstrate some theoretical aspect of an advanced or new control
method and show how it can be applied either in a pilot plant or in some real industrial
situation. The books are distinguished by the combination of the type of theory used and the
type of application exemplified. Note that “industrial” here has a very broad interpretation; it
applies not merely to the processes employed in industrial plants but to systems such as
avionics and automotive brakes and drivetrain. This series complements the theoretical and
more mathematical approach of Communications and Control Engineering.
Indexed by SCOPUS and Engineering Index.
Proposals for this series, composed of a proposal form (please ask the in-house editor below),
a draft Contents, at least two sample chapters and an author cv (with a synopsis of the whole
project, if possible) can be submitted to either of the:

Series Editor
Professor Michael J. Grimble
Department of Electronic and Electrical Engineering, Royal College Building, 204
George Street, Glasgow G1 1XW, United Kingdom
e-mail: [email protected]
or the
In-house Editor
Mr. Oliver Jackson
Springer London, 4 Crinan Street, London, N1 9XW, United Kingdom
e-mail: [email protected]
Proposals are peer-reviewed.
Publishing Ethics
Researchers should conduct their research from research proposal to publication in line with
best practices and codes of conduct of relevant professional bodies and/or national and
international regulatory bodies. For more details on individual ethics matters please
see: https://www.springer.com/gp/authors-editors/journal-author/journal-author-helpdesk/
publishing-ethics/14214

More information about this series at https://link.springer.com/bookseries/1412

Steve S. Niu · Deyun Xiao

Process Control
Engineering Analyses and Best Practices
Steve S. Niu Deyun Xiao
Houston, TX, USA Department of Automation
Tsinghua University
Beijing, China

ISSN 1430-9491 ISSN 2193-1577 (electronic)

Advances in Industrial Control
ISBN 978-3-030-97066-6 ISBN 978-3-030-97067-3 (eBook)
https://doi.org/10.1007/978-3-030-97067-3

MATLAB is a registered trademark of The MathWorks, Inc. See https://www.mathworks.com/trademarks

for a list of additional trademarks.

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my wife, Min, for her continuous support;
Also to my former colleagues, who inspired
me to write this book.
—Steve S. Niu
Series Editor’s Foreword

Control engineering is viewed rather differently by researchers and those that must
implement and maintain control systems. The former develop general algorithms
with a strong underlying mathematical basis, whilst the latter have more immediate
concerns over the limits of equipment, quality of control, safety and security, and
plant downtime. The series Advances in Industrial Control attempts to bridge this
divide and hopes to encourage the adoption of advanced control techniques when
they are likely to be beneficial.
The rapid development of new control theory and technology has an impact on
all areas of engineering and applications. This monograph series has a focus on
applications, which are now so important as the rate of new technological develop-
ment continues to increase as rapidly as it does. It is the challenges from industry
that often stimulate the development of new control paradigms. A focus on appli-
cations is also desirable if the different aspects of the “control design” problem are
to be explored with the same dedication that “control synthesis” problems already
receive. The series provides an opportunity for researchers to present new work on
industrial control and applications. It raises awareness of the substantial benefits that
advanced control can provide whilst tempering its enthusiasm with a discussion of
the challenges that can arise.
This monograph is concerned with engineering analyses and best practice in
process control systems. The authors’ applications and industrial experience results
in a text that is written in an entertaining style that is rather different from the
more usual modelling-and-design-methods-focused texts. The questions raised and
answered in Chap. 1, for example, will have been thought about by many engineers,
but clear answers are not often apparent. The authors’ industrial experience is evident
from even this first chapter.
Chapter 2 covers some standard ideas in control, but all the problems are consid-
ered from a process control viewpoint. Moreover, some of the things—like ratio
control and split-range control—that are common in process control but not used so
much elsewhere are described. Some of the main ideas in advanced control are also
mentioned briefly. The text has a useful role since topics like training simulators,
alarm handling and life cycles may not be covered in more theoretical texts.

vii
viii Series Editor’s Foreword

Chapter 3 on PID control, classed as “Essential Knowledge”, has some familiar

material but the explanations and diagrams make it very accessible. Moreover, the
practical aspects of the use of PID are explored including the uncertainties that affect
performance. An example of the practical utility of the text lies in the sections on
crippled operations and shut-down and start-up.
More sophisticated PID control is considered in Chap. 4, on complex PID control
loops. The multivariable nature of some control problems is explored and tools like
decoupling are discussed. The section on the handling of constraints using PID-based
control, for a rather special situation, is unusual and interesting. The applications
from process control illustrate the special nature of process problems, although some
problems like the need for bumpless transfer occur frequently in other application
sectors as well.
Chapter 5 is on advanced process control and not surprisingly Model Predictive
Control (MPC) is discussed. Before introducing the basic concept, the essential areas
of modelling, identification and estimation are covered from a rather practical and
insightful viewpoint. The need to explore the topic is partly the multivariable nature
of many process systems and the constraints that are often important for quality
control. Other topics, like testing and commissioning, that are valuable for engineers
working on process plants are covered.
Chapter 6 is the first chapter in the section concerned with analytic skills and
problem solving. Much of the material is very useful for plant process control and
instrumentation engineers and is not very accessible elsewhere. The methodology
of process analysis discussed in Chap. 6 involves a lot of common sense rather than
theory, but it is useful to have a formal presentation and summary. The chapter is
a reminder that working on a process plant involves a lot more than control loop
design, and engineers bear a heavy responsibility for safe and effective operation.
Chapter 7, on the methodology of process control design, covers topics like the
process hierarchy and plant-wide control. The discussion is very suitable for process
engineers, who are often faced with a new system designed by a major process
control supplier or automation vendor, and where it is not the detailed control design
questions but more how to understand and use the system provided that is of prime
importance.
Troubleshooting in Chap. 8 is one of the fun things to do if accompanied by an
experienced engineer with good instincts, but for the novice it is daunting when
the source of the problem is elusive and the cost of the loss of production is high.
Sound advice from the text is therefore valuable. For example, asking the process
plant operators for their thoughts can often be revealing. Even if their knowledge
of Nyquist stability conditions is non-existent, they will know if particular sensors
frequently malfunction.
Chapter 9, on control loop tuning and improvement, will be one of the most
popular, dealing with the ubiquitous PID controllers. The explanations cover some
standard tuning procedures. It is also evident from the examples that understanding
the physical process is important if good results are to be obtained.
Chapter 10, on common calculations, deals with some typical calculations that
are needed, covering for example the inferential nature of many process control
Series Editor’s Foreword ix

problems, or the need to determine the relationships between variables and to estab-
lish the quality of data. The companion Chap. 11 is on equipment control and is
more concerned with the control problems for common plant equipment, such as
compressors. Both chapters have very relevant process plant control examples.
The final Chapter, 12, involves one of the most important topics, namely, plant-
wide control and unit control. It is useful that both a bad design and a good design
are described. In the days of cheap energy and few concerns about the environment
any design that was safe and reliable might have been good enough. This is no longer
the case, and it is good to be reminded that simple changes can often make a big
difference if the process and control problem is well understood.
The authors have industrial experience and this text covers topics in process control
very much from a plant-control-engineering perspective. The monograph should
therefore be a help to engineers faced with real problems in the process industries
and is a welcome addition to the series.

Glasgow, UK Michael J. Grimble

September 2021
Foreword by Sirish L. Shah

Since the advent of real-time computing, the field of process control has advanced
significantly over the last several decades. Methods and techniques with much theo-
retical rigor are available to solve some of the most difficult problems. However, a
good working knowledge of the application of these methods requires good mathe-
matical background, typically taught only at post-graduate levels at most universities.
A plethora of books are available for researchers and engineers with advanced degrees
to learn about such methods.
An equally important aspect of a successful application of these advanced tech-
niques is process or domain knowledge that is only learned in the field in the industry.
Very few books are available that complement discussions of advanced control algo-
rithms with unit process operations to guide a control engineer to a successful appli-
cation. This book is an attempt to fill this synergistic gap between control theory and
process applications.
The book is divided into four parts, starting with parts I and II that discuss the
essentials of control theory in a very tutorial manner. Part III focuses on the method-
ology of process analysis and process control design. These sections are uniquely
articulated in a problem-solving setting. The last section (part IV) follows these
methodologies and discusses some typical control applications that require tempera-
ture, pressure, and level regulation, including the control of rotating machinery such
as centrifugal pumps and compressors. This section ends with a discussion of the
important topic of plant-wide control to achieve good operational performance.
The book is an outcome of two very skilled control theoreticians and practitioners
with many years of advanced industrial process control applications to their credit.
The book fills that infamous gap between the theory and practice of process control.

August, 2021 Sirish L. Shah, Ph.D., FCAE, FCIC, FIEEE

Emeritus Professor, University of Alberta
Toronto, ON, Canada

xi
Foreword by Michael W. Brown

I have been very fortunate in my career to have received a graduate degree in process
control. I was even more fortunate to have had the opportunity to benefit from this
education in practicing process control in many industries and companies around the
world. Although the academic side of process control provides a rich set of techniques
and tools to understand and optimize system performance, it is the application of
process control theory to the real-world applications that has helped me build the
analytical skills that I consider the greatest learning over my 35-year career. As we
apply these technologies to industry successfully, the skills we have built through
this practice also have a very broad impact on our understanding and analysis of data
and systems in general.
For those of us lucky enough to have enjoyed a long and exciting career in the field
of process control, I would suggest that there have been three critical factors in this
achievement: a solid education, a good mentoring and development program, and
an opportunity to practice and develop the required skills. This book contributes to
advancing the effort of good mentoring and development program. I was lucky to have
had some of the best mentors in the industry as the author had, and this book captures
most of what I also learned over the years in the field. Parts I and II of the book present
the core process control material in a simple and intuitive way to help all practitioners
with their development journey. The structured analytical methodologies and best
practices in Part III, which are seldom found in other textbooks, are crucial to building
the necessary competence and skills in practical control. The typical control examples
in Part IV demonstrate the structured thinking and work process for working with
real-world applications. These knowledge and skills require many years of practical
experience to capture and articulate, and are beneficial to both the practitioners and
researchers pursuing excellence in the area of process control.
I had the pleasure of working with Steve Niu early in his career, and his dedication
and commitment to process control are always inspiring. Now capturing these many
years of knowledge and experience and sharing it with others is a testament to their
continued dedication to the process control field, and to the young practitioners who
also aspire to develop an exciting career in process control.

xiii
xiv Foreword by Michael W. Brown

Process control has blessed me with a sort of industrial passport for over 35 years,
working in a dozen countries, 50+ companies, and several industrial sectors. It is a
career path I would not go back and change. My mentors and their knowledge sharing
were vital to this journey. We are fortunate that these authors have now captured their
knowledge to share in this book. I believe this book will help many others in their
exciting journey in the area of process control.

August 2021 Michael W. Brown, M.A.Sc., P.Eng.

Executive Vice-President
Technology & Innovation
SMS Equipment Inc.
Edmonton, AB, Canada
Preface

Process control is an engineering specialty vital to process safety and operational

efficiency in any modern operating plant. Process control relies heavily on tech-
nological knowledge and practical experience to excel, evident from the fact that
the process control team usually has the highest concentration of higher educa-
tion degrees. The process control achievements are primarily through improvements
in designs and innovations in operation and typically need little to no additional
hardware investment or extra energy consumption.
Process control is also full of challenges. First of all, process control is a branch of
automatic control. Most of the automatic control theories and techniques originated
from the aerospace and electromechanical engineering fields. There is an exten-
sive framework of control theories and a massive collection of control technologies
through its long track record of success. Unfortunately, only a (small) subset is
directly applicable to the process industry due to industrial processes’ unique and
distinct characteristics. The process control practitioners can easily be overwhelmed
by the rich selection of technologies and get lost on where to spend their effort.
Therefore, it is necessary to delineate the relevant part of the theory and technology
applicable to process control from the vast scope of automatic control. At the same
time, it is also essential to summarize the unique challenges and particular needs
of the process industry and highlight the limitations of the current technologies.
The purpose is to help the practitioners make the best use of the currently available
technologies and set the expectations for the new technologies.
Secondly, process control is about controlling a process. There is a well-known
saying: we cannot control something that we can not measure. More importantly,
we can not control something that we do not understand. It is crucial to understand
the underlying processes in order to control it. In practice, due to the vast number
and extreme diversity of industrial processes, a process control practitioner cannot
become intimately familiar with all the processes, nor is it possible to cover all of
them in any process control books. The more practical approach is developing analyt-
ical skills and effective methodologies for conducting structured process analysis,
extracting the necessary information for control, and developing the most appropriate

xv
xvi Preface

control strategy. The development and standardization of these skills and method-
ologies are thus crucial. Many of the techniques and methodologies can be extracted
and abstracted into best practices to be replicated.
Thirdly, there is a substantial gap between theory and application in addressing
process control’s unique needs. There has also been a big disconnect between what is
taught in school and what is needed in practice. “The gulf between the theory taught
in academia and the practice of process control is astounding. Academia is correct
to teach theory and basic principles, but in process control, the theory being taught
is largely irrelevant to the practice of process control.” (Smith 2009) In academic
research and teaching, people are more interested in how things normally work and
how to advance the theory; In practice, the real challenge is knowing when things
may fail and how to prevent the failure. Students are rarely taught when and how
things will fail, what to do to prevent them from going wrong, and how to respond
when things do go wrong. For example, to design a PID controller and make it work
is absurdly simple and straightforward. To make the PID controller reliably work in
a real-world environment, the designer needs to address many “what if ...” types of
practical considerations. The practical challenges are rarely taught or researched in
academic research, although many have excellent research values. In an operating
environment, this shift from “what it does ...” to “what if it does not ...” is the
primary difference between the thinking process of an inexperienced junior engineer
and that of an experienced senior engineer. This shift is also a major hurdle for most
practitioners that academia could have provided more help for them to pass.
Lastly, there are often confusions about the roles and responsibilities of process
control. Process control is often treated as a branch of automatic control in academic
teaching without giving adequate attention to the distinct requirements of process
control. In the field, process control’s roles and responsibilities are often assigned to
process engineers during the project design phase and then to control systems engi-
neers for implementation and maintenance. Unfortunately, most process engineers
do not know the control systems to produce a proper control design. On the other
hand, control systems engineers do not have enough process knowledge or opera-
tional experience to understand the design intent and operating requirements. This
disconnect has led to sub-optimal control designs and incorrect DCS implementa-
tion. It is the primary reason that as high as half of PID controllers can not stay in the
intended control mode in many operations, especially upstream. A process engineer
or a control system engineer must know enough about process control to perform
process control duties.
Most textbooks or reference books in process control focus on the theoretical
depth and completeness, with the practical applications as a complement to enhance
the understanding of the theories. This book takes an entirely different approach. It
first summarizes the essential knowledge and core technologies currently in use (and
in need) in the process industry. It then focuses on developing structured problem-
solving skills and best-practice methodologies. The goal is to guide the practitioners
and researchers on the thinking process and general procedures for working on a
new and unfamiliar process flow. The skills and methodologies should be helpful
to both process control engineers working in the field and academic researchers
Preface xvii

interested in the practical side of process control. For this purpose, the presentation
is deliberately made simple and intuitive by using illustrations and tables as much
as possible. Control theories are reduced to the minimum and are introduced only
when needed. As a result, we will limit the explanation to time-domain and avoid
the advanced topics such as stability analysis and state-space model since they have
minimal practical applicability due to the complexities and uncertainties of process
dynamics.
The book has a practical-oriented theme, and the presentation follows the general
path: overview ⇒ essential knowledge ⇒ analytical skills and best-practice method-
ologies ⇒ unit and plant-level control applications. This sequence coincides with the
learning and progression path from awareness, knowledge, skills to mastery. Along
these lines, the book’s content is organized into four parts.
Part I, The Big Picture, provides a general introduction to process control. This
part explains process control’s unique and irreplaceable role in an operating plant
and answers the general what/why/who/where/when types of questions about this
extraordinary cross-discipline engineering specialty (Chap. 1). It then summarizes
what core technologies are available and what they can do, thus defining the essential
knowledge framework for process control practitioners (Chap. 2).
Part II, Essential Knowledge, provides more detailed discussions on the key
concepts and core technologies of process control outlined in Chap. 2. It explains the
minimum and essential knowledge that a process control engineer should acquire,
from basic PID-based control loop (Chap. 3), complex PID control scheme (Chap. 4),
to advanced control solutions (Chap. 5). The focus is not only on how they normally
function but also on making them not fail in all practical scenarios.
A common challenge facing process control design is where to start with control
design or troubleshooting when faced with a new and unfamiliar process. Part III,
Analytical Skills and Problem-Solving Methodologies, introduces the best prac-
tice work processes to deal with the common process control tasks. These skills
and methodologies include a structured approach for conducting process analysis
(Chap. 6), general guidelines and procedures for process control design (Chap. 7),
best-practice methodologies for control problem troubleshooting (Chap. 8), and
control performance tuning and improvements (Chap. 9).
Some typical control applications are presented in Part IV, including common
control-related calculations (Chap. 10), control of typical equipment (Chap. 11), and
control of typical processes (Chap. 12). These practical examples demonstrate the
whole thinking process and complete analytical procedure for solving real-world
problems based on the skills and methodologies presented in Part III.
This book’s targeted audiences are the process control practitioners in the field and
students and academic researchers interested in the practical side of process control.
General audiences such as non-engineering personnel or operation management
can use Chap. 1 as an overview to understand what process control is all about.
Junior control engineers and undergraduate students can start with Part I to build their
knowledge framework in process control. Part I can also serve as an introduction to
process control technologies for other engineering teams to raise their awareness.
For example, process engineers, control system engineers, operators, and automation
xviii Preface

managers often need to work together with process control for their decision-making;
thus, it is beneficial to understand what process control can offer.
From Part I to Part IV, process control engineers and academic researchers can
appreciate the unique challenges that process control technologies face in a real-
world environment. It progressively builds the knowledge, skills, and eventually,
the process control “sense” or “instinct” to perform their duties more effectively.
Experienced process control engineers or academic researchers can benefit from
the process analysis and control design methodology in Part III and enhance them
through the complete real-world examples in Part IV.
Process control is cross-disciplinary and interacts with process engineering,
rotating equipment, operations, control systems (DCS), instrumentation, and infor-
mation technology. A process control engineer cannot be skillful in all these fields in
his short career. For example, there are countless processes in different industries. It
is impossible (and unnecessary) to be an expert on all the processes. That is the job
of process engineers and equipment engineers, who are the best source for in-depth
process knowledge. For a process control engineer, the skill and methodology to effi-
ciently extract and abstract the specific information required by process control are
more valuable and practical. Similarly, sensors, transmitters, valves, PID controllers,
DCS/PLC platforms, communication networks are essential tools and platforms for
process control. A process control practitioner needs to have a working knowledge
of these tools and infrastructure; however, it is unnecessary to become an expert on
all the fast-changing tools. That is the job of DCS and instrumentation engineers or
technicians, who are the go-to persons for support.
The roles and responsibilities of a process control engineer are to utilize the latest
available tools to achieve higher goals, i.e., overall operational excellence. Process
control engineers must have a clear picture of their knowledge structure, and know
what must know, what is good to know, and what should not be wasting time to know.
“To know what you know and what you do not know, that is true knowledge.”
(Confucius)

Houston, USA Steve S. Niu

Beijing, China Deyun Xiao
September, 2021

This book was triggered and motivated by my experience transitioning from academia
to industry. Personally, when I left academia to join the process industry as an
advanced process control engineer, I had a B.Sc., M.Sc., and Ph.D. degree under my
belt, all in process control, plus several years of research experience in top universities
in the area of advanced process control. So I felt confident and well-prepared to take
up the new challenges, but only to find that I just stepped into an unfamiliar world
as a complete rookie. The advanced control theories gained in school found little
relevance to the jobs to be performed. State-space models, Routh stability criteria,
Nyquist plot, Bode Plots, LQG control, H∞ could hardly see any use. On the other
hand, practical knowledge such as the impact of control mode changes, practical PID
tuning methods, and process analysis was desperately lacking.
Preface xix

However, I was fortunate to have a very competent and caring mentor who guided
me into the real world and helped fill up the initial gap between theory and practice
(of course, the academic background did help me pick up speed quickly). Later, when
I started training and mentoring other young engineers, I found they were all faced
with the same challenges as I did before. There was so much missing between what
was taught in school and what was needed in the field, most young engineers had to
learn it the hard way on their own. I felt a strong need for a book that summarized
and explained the essential knowledge and core skills in one place. Unfortunately,
very few people from the industry have the time and interest to write and share their
practical knowledge.
It is easy to bring a junior engineer up to speed to perform regular and routine
tasks; however, it takes many years of real-world experience to become a skilled
engineer. The difference is in handling new challenges, abnormal conditions, and
emergencies; These capabilities can only be learned from experiencing the abnormal
conditions and emergencies. This learning process can be greatly accelerated through
an experienced mentor. Without a mentor, the learning is by trial and error on the
actual plants, and the cost of learning can be very stiff. In this increasingly cost-
sensitive operating environment, a practical-oriented book like this one could be the
next best thing to having a great mentor or training program.
The content design of this book is based on my experience with academic research
and training/mentoring of countless young engineers. My 10-year academic research
and 25-year industrial experience covered almost all aspects of process control, from
research, development, to applications; from project management, product manage-
ment, to project execution; Much experience was accumulated from the design and
implementation of hundreds of control solutions covering base-layer control, inte-
grated compressor control, advanced process control, real-time optimization. These
projects range from small control improvement projects to multi-billion-dollar mega-
projects spanning oil & gas upstream production, refining, petrochemical, chemical,
and pulp & paper processes.
I hope this book can help make the typical process control tasks a little more
technical and less artistic.

Houston, USA Steve S. Niu

September 2021
Contents

Part I The Big Picture

1 Process Control Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Introduction to Process Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 What Is Process Control? . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Why Do We Need Process Control? . . . . . . . . . . . . . . . . . 5
1.1.3 When Do We Need Process Control? . . . . . . . . . . . . . . . . 6
1.1.4 What Is Process Control Not? . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Operating Objectives and Control Objectives . . . . . . . . . . . . . . . . . 9
1.2.1 Process Flow and Operating Objectives . . . . . . . . . . . . . . 9
1.2.2 Control Objectives and Control Strategy . . . . . . . . . . . . . 10
1.2.3 Process Control for Process Safety . . . . . . . . . . . . . . . . . . 12
1.2.4 Process Control for Production Efficiency . . . . . . . . . . . . 15
1.3 Process Control in Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 Process Automation Hierarchy . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Process Automation System (PAS) . . . . . . . . . . . . . . . . . . 17
1.3.3 Symbology and Identification . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.4 Control and Safeguarding . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.5 Operation by Setpoints and Limits . . . . . . . . . . . . . . . . . . 21
1.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2 Process Control Knowledge Framework . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1 Simple Control Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.1 Open-Loop Versus Closed-Loop Control . . . . . . . . . . . . . 25
2.1.2 Feedback Control and Its Essential Components . . . . . . . 28
2.1.3 Control Loop Component: Process Dynamics . . . . . . . . . 29
2.1.4 Control Loop Component: Process Measurements . . . . . 30
2.1.5 Control Loop Component: Final Control Elements . . . . . 34
2.1.6 Control Loop Component: Control Objectives . . . . . . . . 37
2.1.7 Control Loop Component: Control Algorithms . . . . . . . . 40
2.2 Complex Control Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

xxi
xxii Contents

2.2.1 Split-Range Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.2.2 Dual-Controller Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.3 Fan-Out Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.4 Cascade Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.2.5 Feedforward Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.2.6 Ratio Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.2.7 Override Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.2.8 Selective Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.3 Advanced Process Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.3.1 Limitations of PID Control . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.3.2 Process Models: Abstraction and Generalization . . . . . . 53
2.3.3 Inferential Property (Soft Sensor) . . . . . . . . . . . . . . . . . . . 54
2.3.4 Model Predictive Control . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.3.5 Self-Tuning PID Control and Adaptive Control . . . . . . . . 56
2.3.6 Control Performance Monitoring . . . . . . . . . . . . . . . . . . . . 57
2.3.7 Real-Time Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.3.8 Dynamic Simulation and Operator Training
Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.3.9 Alarm Rationalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.4 Process Control Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.4.1 Process Control Philosophy . . . . . . . . . . . . . . . . . . . . . . . . 59
2.4.2 Process Control Life Cycle . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Part II Essential Knowledge

3 Basic PID Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.1 PID Control Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.1.1 Introduction to Laplace Transform . . . . . . . . . . . . . . . . . . 67
3.1.2 PID Control Algorithms, Ideal Form . . . . . . . . . . . . . . . . . 69
3.1.3 PID Control Algorithms, Other Forms . . . . . . . . . . . . . . . 72
3.1.4 PID Equation Types (“PID Structure”) . . . . . . . . . . . . . . . 72
3.1.5 PID Dimension Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.1.6 Actual Output Versus Incremental Changes . . . . . . . . . . . 78
3.1.7 PID Discrete Implementation . . . . . . . . . . . . . . . . . . . . . . . 79
3.1.8 Positional Versus Velocity Forms . . . . . . . . . . . . . . . . . . . . 80
3.1.9 PID with Primitive Function Blocks . . . . . . . . . . . . . . . . . 82
3.2 PID Mode of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.2.1 PID Control Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.2.2 Variations in PID Implementation . . . . . . . . . . . . . . . . . . . 87
3.3 PID Loop Integrity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.3.1 Process Flow and Data Flow . . . . . . . . . . . . . . . . . . . . . . . . 90
3.3.2 Process Dynamics and Loop Dynamics . . . . . . . . . . . . . . 93
3.4 Tracking and Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Contents xxiii

3.4.1 Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.4.2 Back-Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.4.3 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.5 PID Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.5.1 Output Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.5.2 Direction of Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.5.3 Setpoint High/Low Limit . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.5.4 Execution Frequency and Execution Order . . . . . . . . . . . 106
3.5.5 PV Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.5.6 Initial Tuning Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.5.7 PV Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.5.8 Anti-Reset Windup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.5.9 Bumpless Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.6 PID Mode Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.6.1 PID Mode Change in Normal Operations . . . . . . . . . . . . . 109
3.6.2 PID Modes During Crippled Operations . . . . . . . . . . . . . . 111
3.6.3 PID Modes During Shutdown and Startup . . . . . . . . . . . . 112
3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4 Complex PID Control Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.1 Applications of Complex PID Control Schemes . . . . . . . . . . . . . . . 115
4.1.1 Main Challenges for Complex Control Loops . . . . . . . . . 115
4.1.2 Dealing with Multivariable Interactions . . . . . . . . . . . . . . 118
4.1.3 Handling Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.1.4 Optimizing Flow Control . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.1.5 Tackling Nonlinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.1.6 Rationalizing Control Scheme . . . . . . . . . . . . . . . . . . . . . . 135
4.2 Structure of Complex Control Loops . . . . . . . . . . . . . . . . . . . . . . . . 143
4.2.1 Common Function Blocks . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.2.2 Built-in Function Versus Custom Function . . . . . . . . . . . 150
4.3 Configurations of Complex PID Loops . . . . . . . . . . . . . . . . . . . . . . 151
4.3.1 Data Flow in Complex PID Loops . . . . . . . . . . . . . . . . . . . 151
4.3.2 Common Configuration Considerations . . . . . . . . . . . . . . 152
4.3.3 Anti-reset Windup in Complex Control Loops . . . . . . . . 154
4.3.4 Bumpless Transfer in Complex Control Loops . . . . . . . . 161
4.3.5 Bumpless Transfer Versus Anti-reset Windup . . . . . . . . . 164
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5 Advanced Process Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.1 Why Advanced Process Control? . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.1.1 Process Challenges and PID Control Limitations . . . . . . 170
5.1.2 Process Control Solutions Before MPC . . . . . . . . . . . . . . 170
5.1.3 Advanced Process Control Technologies . . . . . . . . . . . . . 171
5.2 Model-Based Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
xxiv Contents

5.2.1 Process Analytics and Process Models . . . . . . . . . . . . . . . 173

5.2.2 Data Analysis and Process Identification . . . . . . . . . . . . . 176
5.2.3 First-Order-Plus-Delay-Time (FOPDT) Model . . . . . . . . 177
5.3 Inferential Property Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.3.1 Static Inferential Property Estimation . . . . . . . . . . . . . . . . 181
5.3.2 Dynamics Inferential Property Estimation . . . . . . . . . . . . 183
5.4 Model Predictive Control—The Concept . . . . . . . . . . . . . . . . . . . . 188
5.4.1 MPC as Multivariable Control . . . . . . . . . . . . . . . . . . . . . . 189
5.4.2 MPC as Predictive Control . . . . . . . . . . . . . . . . . . . . . . . . . 193
5.4.3 MPC as Constrained Control . . . . . . . . . . . . . . . . . . . . . . . 194
5.4.4 MPC as Optimizing Control . . . . . . . . . . . . . . . . . . . . . . . . 196
5.5 Model Predictive Control—Practical Considerations . . . . . . . . . . . 197
5.5.1 Process Noises and Disturbances . . . . . . . . . . . . . . . . . . . . 197
5.5.2 Baselayer Control Versus Advanced Control . . . . . . . . . . 198
5.5.3 MPC Maintenance and Sustainability . . . . . . . . . . . . . . . . 200
5.5.4 Knowledge Management and Training . . . . . . . . . . . . . . . 201
5.5.5 Future Developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.6 Model Predictive Control—The Work Process . . . . . . . . . . . . . . . . 204
5.6.1 Project Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
5.6.2 Project Kickoff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
5.6.3 System Installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
5.6.4 Baselayer Review and Improvement . . . . . . . . . . . . . . . . . 206
5.6.5 Pre-testing Activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5.6.6 Plant Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
5.6.7 Process Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
5.6.8 Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
5.6.9 Model and Controller Review . . . . . . . . . . . . . . . . . . . . . . 211
5.6.10 Controller Commissioning . . . . . . . . . . . . . . . . . . . . . . . . . 212
5.6.11 Operator Training and Documentation . . . . . . . . . . . . . . . 213
5.6.12 Controller Handover to Operators . . . . . . . . . . . . . . . . . . . 213
5.6.13 Post-implementation Review (PIR) . . . . . . . . . . . . . . . . . . 214
5.6.14 Project Close-Out . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
5.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215

Part III Analytic Skills and Problem-Solving Methodologies

6 Methodology of Process Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
6.1 Operating Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
6.1.1 Operating Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
6.1.2 Product Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
6.1.3 Limits and Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
6.1.4 Mode of Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
6.2 Process Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.2.1 Process Flow and System Boundary . . . . . . . . . . . . . . . . . 223
Contents xxv

6.2.2 Key Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224

6.3 Supply and Demand Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
6.3.1 Material and Energy Balance . . . . . . . . . . . . . . . . . . . . . . . 225
6.3.2 Supply-Driven and Demand-Driven Operation . . . . . . . . 225
6.3.3 Supply-Driven and Demand-Driven Control . . . . . . . . . . 226
6.3.4 Swing Capacity and Swing Control . . . . . . . . . . . . . . . . . . 230
6.4 Cause and Effect Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
6.4.1 Controlled Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
6.4.2 Manipulated Variables and Disturbance Variables . . . . . 232
6.4.3 Cause and Effect Relationships . . . . . . . . . . . . . . . . . . . . . 233
6.5 Degree of Freedom Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
6.5.1 Degree of Freedom Requirements . . . . . . . . . . . . . . . . . . . 234
6.5.2 Pairing of Process Variables . . . . . . . . . . . . . . . . . . . . . . . . 235
6.6 Dynamics Response Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6.6.1 Process Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6.6.2 Dynamics Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
6.7 Process Analysis Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
7 Methodology of Process Control Design . . . . . . . . . . . . . . . . . . . . . . . . . 245
7.1 Process Hierarchy and Process Control Scope . . . . . . . . . . . . . . . . 245
7.1.1 Process Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
7.1.2 Plant-Wide Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
7.1.3 Unit-Level Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
7.1.4 Loop-Level Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
7.2 Control Hierarchy and Control Strategy . . . . . . . . . . . . . . . . . . . . . 253
7.2.1 Overall Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
7.2.2 Normal Regulatory Control . . . . . . . . . . . . . . . . . . . . . . . . 255
7.2.3 Protective Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
7.2.4 Sequential Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
7.2.5 Instrumented Safeguarding . . . . . . . . . . . . . . . . . . . . . . . . . 258
7.3 Process Control Functional Design . . . . . . . . . . . . . . . . . . . . . . . . . . 259
7.3.1 Variable Pairing and Degree of Freedom Validation . . . . 260
7.3.2 Control Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
7.4 Process Control Visualization and Documentation . . . . . . . . . . . . . 262
7.4.1 Process Control Narrative Document . . . . . . . . . . . . . . . . 263
7.4.2 Control Overview Diagram . . . . . . . . . . . . . . . . . . . . . . . . . 265
7.4.3 Control Loop Narratives . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
7.4.4 Loop Configuration Details . . . . . . . . . . . . . . . . . . . . . . . . 267
7.5 Building Control Overview Diagram—The Work Process . . . . . . 270
7.5.1 Collect Relevant Information . . . . . . . . . . . . . . . . . . . . . . . 271
7.5.2 Build Process Overview Diagram . . . . . . . . . . . . . . . . . . . 271
7.5.3 Add Regulatory Control Loops . . . . . . . . . . . . . . . . . . . . . 271
7.5.4 Add Overriding and Protective Control Loops . . . . . . . . . 272
xxvi Contents

7.5.5 Add Special Configuration Requirements . . . . . . . . . . . . . 273

7.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
8 Methodology of Control Problem Troubleshooting . . . . . . . . . . . . . . . 275
8.1 Common Control Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
8.2 General Troubleshooting Procedure . . . . . . . . . . . . . . . . . . . . . . . . . 277
8.2.1 Obtain a Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . 277
8.2.2 Identify Human Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
8.2.3 Isolate the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
8.2.4 Identify the Root Cause . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
8.2.5 Propose a Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
8.3 Troubleshooting of Complex Control Scheme . . . . . . . . . . . . . . . . 283
8.3.1 Overall Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
8.3.2 Dynamic Supply and Demand Balance . . . . . . . . . . . . . . . 284
8.3.3 Dynamic Cause and Effect Relationship . . . . . . . . . . . . . . 286
8.3.4 Degree of Freedom Review . . . . . . . . . . . . . . . . . . . . . . . . 288
8.4 Troubleshooting of PID Control Loop . . . . . . . . . . . . . . . . . . . . . . . 289
8.4.1 Process Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
8.4.2 Process Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
8.4.3 Final Control Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
8.4.4 Control Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
8.4.5 The Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
8.5 Troubleshooting Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
8.5.1 Data Trending and Scatter Plots . . . . . . . . . . . . . . . . . . . . . 297
8.5.2 Control Performance Monitoring Tools . . . . . . . . . . . . . . 300
8.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
9 Control Loop Tuning and Improvement . . . . . . . . . . . . . . . . . . . . . . . . . 305
9.1 PID Control Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
9.1.1 Performance Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
9.1.2 Practical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
9.2 Classical PID Controller Tuning Methods . . . . . . . . . . . . . . . . . . . . 308
9.2.1 Typical Tunings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
9.2.2 Tuning by Trial and Error . . . . . . . . . . . . . . . . . . . . . . . . . . 309
9.2.3 Ziegler–Nichols Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
9.3 Model-Based Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
9.3.1 Model-Based Tuning Concept . . . . . . . . . . . . . . . . . . . . . . 314
9.3.2 Step Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
9.3.3 Model Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
9.3.4 Tuning Parameter Calculation . . . . . . . . . . . . . . . . . . . . . . 319
9.3.5 Tuning Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
9.3.6 PID Self-Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
9.3.7 Best Practice of PID Tuning . . . . . . . . . . . . . . . . . . . . . . . . 322
9.4 Tuning Flow Control Loops with History Data . . . . . . . . . . . . . . . . 324
Contents xxvii

9.4.1 Flow Loop Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324

9.4.2 Flow Loop Tuning Procedure . . . . . . . . . . . . . . . . . . . . . . . 325
9.5 Tuning Level Control Loops with Vessel Sizing Data . . . . . . . . . . 329
9.5.1 Level Loop Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
9.5.2 Level Control Loop Tuning Procedure . . . . . . . . . . . . . . . 331
9.5.3 Characterization of Level Loop Dynamics . . . . . . . . . . . . 335
9.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338

Part IV Process Control Typicals

10 Control Typicals: Common Calculations . . . . . . . . . . . . . . . . . . . . . . . . 341
10.1 Signal Transmission and Transformation . . . . . . . . . . . . . . . . . . . . . 341
10.1.1 Application Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
10.1.2 Theory and Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
10.1.3 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 345
10.1.4 Application Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
10.1.5 Special Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
10.2 Data Quality and Bad Data Handling . . . . . . . . . . . . . . . . . . . . . . . . 350
10.2.1 Application Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
10.2.2 Theory and Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
10.2.3 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 352
10.3 Flow Compensation and Conversion . . . . . . . . . . . . . . . . . . . . . . . . 355
10.3.1 Application Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
10.3.2 Theory and Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
10.3.3 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 360
10.3.4 Application Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
10.4 Temperature Compensation with Pressure . . . . . . . . . . . . . . . . . . . . 365
10.4.1 Application Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
10.4.2 Theory and Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
10.4.3 Application Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
10.4.4 Special Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
10.5 Level Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
10.5.1 Application Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
10.5.2 Theory and Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
10.5.3 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 375
10.6 Internal Reflux Versus External Reflux . . . . . . . . . . . . . . . . . . . . . . 375
10.6.1 Application Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
10.6.2 Theory and Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
10.6.3 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 377
10.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
xxviii Contents

11 Control Typicals: Equipment Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 381

11.1 Ejector and Educer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
11.1.1 Process Description and Analysis . . . . . . . . . . . . . . . . . . . 382
11.1.2 Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
11.1.3 Functional Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
11.1.4 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 388
11.2 Centrifugal Pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
11.2.1 Process Description and Analysis . . . . . . . . . . . . . . . . . . . 390
11.2.2 Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
11.2.3 Functional Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
11.2.4 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 404
11.2.5 Practical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
11.3 Centrifugal Compressor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
11.3.1 Process Description and Analysis . . . . . . . . . . . . . . . . . . . 409
11.3.2 Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
11.3.3 Functional Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
11.3.4 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 428
11.3.5 Practical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
11.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
12 Control Typicals: Plant-Wide Control and Unit Control . . . . . . . . . . 433
12.1 Liquid Tank Blanketing Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
12.1.1 Process Description and Analysis . . . . . . . . . . . . . . . . . . . 434
12.1.2 Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
12.1.3 Functional Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
12.1.4 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 440
12.2 Boiler Drum Level Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
12.2.1 Process Description and Analysis . . . . . . . . . . . . . . . . . . . 441
12.2.2 Control Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
12.2.3 Practical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
12.3 Well Test Separator Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
12.3.1 Process Description and Analysis . . . . . . . . . . . . . . . . . . . 449
12.3.2 Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
12.3.3 Functional Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
12.3.4 Implementation and Operation . . . . . . . . . . . . . . . . . . . . . . 458
12.3.5 Practical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
12.4 Firewater Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
12.4.1 Process Description and Analysis . . . . . . . . . . . . . . . . . . . 460
12.4.2 Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
12.4.3 Functional Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
12.4.4 Implementations and Operation . . . . . . . . . . . . . . . . . . . . . 465
12.5 Unit Control: Acetylene Hydrogenation Process . . . . . . . . . . . . . . 466
12.5.1 Process Description and Operating Requirements . . . . . . 467
12.5.2 Supply and Demand Analysis . . . . . . . . . . . . . . . . . . . . . . 467
Contents xxix

12.5.3 Cause and Effect Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 468

12.5.4 Dynamic Response Analysis . . . . . . . . . . . . . . . . . . . . . . . 469
12.5.5 Control Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
12.5.6 Practical Consideration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
12.6 Plant-Wide Control: E&P Surface Production Process . . . . . . . . . 472
12.6.1 Process Description and Operating Requirements . . . . . . 472
12.6.2 Process Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
12.6.3 Supply and Demand Analysis . . . . . . . . . . . . . . . . . . . . . . 474
12.6.4 Cause and Effect Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 474
12.6.5 Degree-of-Freedom Analysis . . . . . . . . . . . . . . . . . . . . . . . 475
12.6.6 Control Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
12.6.7 Practical Consideration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
12.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479

Appendix: Special Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481

Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
Abbreviations

AI Analog In; Artificial Intelligence

ALARP As Low As Reasonably Practical
AO Analog Out
APC Advanced Process Control
ARM Alarm Rationalization and Management
ARWU Anti-Rest Windup
BEP Best Efficiency Point
BFW Boiler Feedwater
BHP Brake Horse Power
BPD Barrels per Day
BS&W Basic Sediments and Water
BTU British Thermal Unit
CAPEX Capital Expenditure
CCR Central Control Room
CET Cause and Effect Table
CG Chromatographs
COD Control Overview Diagram
CPI Controller Performance Index
CPM Control Performance Monitoring
CSTR Continuous Stirred Tank Reactor
CV Controlled Variable
DCS Distributed Control System
DMC Dynamic Matrix Control
DOF Degree of Freedom
DP Differential Pressure
DV Disturbance Variable
E&P Exploration and Production
ERP Enterprise Resource Planning
ESD Emergency Shutdown
FAT Factory Acceptance Test
FCE Final Control Elements

xxxi
xxxii Abbreviations

FCS Fieldbus Control System

FEED Front End Engineering Design
FGS Fire and Gas System
FIR Finite Impulse Response
FMEA Failure Mode and Effect Analysis
FOPTD First Order plus Time Delay
FSR Finite Step Response
FWKO Freewater Knockout
GLR Gas Liquid Ratio
GOR Gas Oil Ratio
GPC Generalized Predictive Control
HART Highway Addressable Remote Transducer
HAZOP Hazard and Operability
HMI Human Machine Interface
HRSG Heat-Recovery Steam Generators
HSE Health, Safety and Environmental
I/O Input/Output
I/P Current to Pressure Converter
IMC Internal Model Control
IPE Inferential Property Estimation/Estimator
ISA Instrument Society of America
ISO International Standard Organization
KPI Key Performance Index; Key Performance Indicator
KPV Key Process Variables
LDPE Low-Density Polyethylene
LOPC Loss of Primary Containment
LP Linear Programming
LTI Linear Time-Invariant
MCSF Minimum Continuous Stable Flow (of a pump)
MCTF Minimum Continuous Thermal Flow
MIMO Multi-Input Multi-Output
MISO Multi-Input Single-Output
MMSCMD Million Standard Cubic Meters Per Day
MOC Management of Change
MOP Maximum Operating Pressure
MOV Motor Operated Valve
MPC Model Predictive Control
MPFM Multi-Phase Flow Meter
MSV Multi-port Selection Valve
MV Manipulated Variable
NN Neural Netwrok
NPSH Net Positive Suction Head (of a pump)
NPSHA Net Positive Suction Head Available (of a pump)
NPSHR Net Positive Suction Head Required (of a pump)
NRV Non-Return Valve
Abbreviations xxxiii

OOS Out of Service

OPC Open Platform Commication; OLE for Process Control
OPEX Operating Expenditure
OTS Operator Training Simulator
P&ID Piping and Instrumentation Diagram
PAS Process Automation System
PCN Process Control Narrative
PFD Process Flow Diagram
PI Propotional-Integral algorithm
PID Proportional-Integral-Derivative
PIR Percentage In Range
PLC Programming Logic Control
POD Process Overview Diagram
PPE Personal Protection Equipment
PSD Plant Shutdown
PV Process Variable
QP Quadratic Programming
RGA Relative Gain Array; Relative Gain Analysis
RO Restriction Orifice
ROI Return On Investment
RTO Real Time Optimization
RTU Remote Terminal Unit
RVP Reid Vapor Pressure
SAT Site Acceptance Test
SCADA Supervisory Control and Data Acquisition
SIA Secure Instrument Air
SIL Safety Integrity Level
SIS Safety Instrument System
SISO Single Input Single Output
SOP Standard Operating Procedure
TTSS Time To Steady State
TVP True Vapor Pressure
VRU Vapor Recovery Unit
Symbols

English Letters

A Cross-section area
Cp Heat capacity at constant pressure
Cv Heat capacity at constant volume
Fm Mass flow rate
Fv Volumetric flow rate
Kc PID controller gain (min)
Kp Process gain
Kp Normalized Process gain
Hp Polytropic head (of a compressor)
M Gas molecular weight
N Speed (of a pump or compressor
Ps Suction pressure (pump, compressor)
Pd Discharge pressure (pump, compressor)
Ti PID controller integral time (min)
Td PID controller derivative time; Compressor discharge temperature
Ts Compressor suction temperature; Sampling time in process model
W Power (of a motor, pump, compressor)
Z Gas compressibility
d Time Delay (discrete-time)
n Compressor polytropic exponent
u Process Input
v Gas velocity
y Process Output
z Process Output with Noise

xxxv
xxxvi Symbols

Greek Letters

κ Ratio of specific heat = C p /Cv

λ First order filter constant in PV filtering; Closed loop performance indicator in
λ tuning
ρ Density of gas, vapor, or liquid
σ Compressor polytropic index
τ Time constant (lag)
θ Process time delay (dead time)

Constants

g Gravitational Acceleration, g = 9.80665 m/s2

P0 Atmospheric pressure, −101.35 kPa
R Gas constant, R = 8.31446 J · K−1 · mol−1
T0 Absolute zero temperature, −273.15 ◦ C

Functions

ln Base-e logarithm (natural)

log Base-10 logarithm

Fonts

italics Used for emphasis or for special terminologies

Typewriter Used for tag names, DCS variables, or programming codes
List of Figures

Fig. 1.1 Process control is cross-disciplinary . . . . . . . . . . . . . . . . . . . . . . 4

Fig. 1.2 Process control helps maximize profit . . . . . . . . . . . . . . . . . . . . . 5
Fig. 1.3 Process overview diagram for an E&P plant . . . . . . . . . . . . . . . . 6
Fig. 1.4 Project execution: desired versus delivered . . . . . . . . . . . . . . . . . 7
Fig. 1.5 Disconnect between process engineering and control
systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Fig. 1.6 Operating objectives: from design to operation . . . . . . . . . . . . . 10
Fig. 1.7 Overall control strategy for a general E&P process . . . . . . . . . . 11
Fig. 1.8 Layers of process safety protections . . . . . . . . . . . . . . . . . . . . . . 12
Fig. 1.9 More profitable production through stabilization
and optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Fig. 1.10 Plant automation hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Fig. 1.11 Process automation system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Fig. 1.12 Simplification of symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Fig. 1.13 Example of control and safeguarding . . . . . . . . . . . . . . . . . . . . . 20
Fig. 1.14 Operating setpoints and limits . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Fig. 2.1 Open-loop control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Fig. 2.2 Closed-loop control (feedback control) . . . . . . . . . . . . . . . . . . . . 27
Fig. 2.3 Naming of variables in a control loop . . . . . . . . . . . . . . . . . . . . . 27
Fig. 2.4 A process with noise and disturbance . . . . . . . . . . . . . . . . . . . . . 29
Fig. 2.5 Process dynamics from a step response test . . . . . . . . . . . . . . . . 29
Fig. 2.6 Process measurement components . . . . . . . . . . . . . . . . . . . . . . . . 31
Fig. 2.7 Control valves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Fig. 2.8 Valve characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Fig. 2.9 Control valve arrangement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Fig. 2.10 Control objectives and constraints . . . . . . . . . . . . . . . . . . . . . . . . 38
Fig. 2.11 “Goal Zero” for HSE incidents . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Fig. 2.12 On/off control and PID control . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Fig. 2.13 Deadband in on/off switching control . . . . . . . . . . . . . . . . . . . . . 41
Fig. 2.14 Structure of a PID control loop . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Fig. 2.15 A split-range control loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

xxxvii
xxxviii List of Figures

Fig. 2.16 Dual-pressure control in place of split-range control . . . . . . . . . 45

Fig. 2.17 A fan-out control loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Fig. 2.18 A cascade control loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Fig. 2.19 A feedforward control loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Fig. 2.20 A ratio control loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Fig. 2.21 A protective control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Fig. 2.22 General architecture of advanced process control . . . . . . . . . . . . 52
Fig. 2.23 Process model and its applications . . . . . . . . . . . . . . . . . . . . . . . . 53
Fig. 2.24 An example of inferential property application . . . . . . . . . . . . . 54
Fig. 2.25 Model predictive control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Fig. 2.26 Generic architecture of adaptive control . . . . . . . . . . . . . . . . . . . 56
Fig. 3.1 A PID control loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Fig. 3.2 The P, I, and D response of a PID controller following
a setpoint change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Fig. 3.3 PID process value (PV), setpoint (SP), and output (OP) . . . . . . 73
Fig. 3.4 Dimensions of PID controller . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Fig. 3.5 PI controller with primitive building blocks . . . . . . . . . . . . . . . . 83
Fig. 3.6 I-only controller with primitive building blocks . . . . . . . . . . . . 83
Fig. 3.7 PI controller with primitive building blocks and valve
read-back . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Fig. 3.8 Variables in a basic PID controller (Left: Honeywell,
Right: Yokogawa) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Fig. 3.9 PID control function block in Yokogawa Centum
(simplified) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Fig. 3.10 Common modes of PID controllers . . . . . . . . . . . . . . . . . . . . . . 86
Fig. 3.11 Closed-loop control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Fig. 3.12 Process flow, data flow, and back-calculation . . . . . . . . . . . . . . 97
Fig. 3.13 PID back-calculation in a flow control loop . . . . . . . . . . . . . . . . 98
Fig. 3.14 Information flow in a cascade control loop . . . . . . . . . . . . . . . . 99
Fig. 3.15 A basic PID cascade control loop . . . . . . . . . . . . . . . . . . . . . . . . 105
Fig. 3.16 Handshake process for a PID mode change . . . . . . . . . . . . . . . . 110
Fig. 3.17 PID controller mode during shutdown . . . . . . . . . . . . . . . . . . . . 112
Fig. 4.1 A typical gas compression process flow . . . . . . . . . . . . . . . . . . . 116
Fig. 4.2 A complex control loop: integrated compressor control . . . . . . 117
Fig. 4.3 Block diagram of feedforward control . . . . . . . . . . . . . . . . . . . . 120
Fig. 4.4 Example of decoupling control . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Fig. 4.5 Example of protective control . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Fig. 4.6 A process with flow balancing requirements . . . . . . . . . . . . . . . 128
Fig. 4.7 Split point for flow balancing control . . . . . . . . . . . . . . . . . . . . . 129
Fig. 4.8 Split-range design for flow balancing control . . . . . . . . . . . . . . 130
Fig. 4.9 Valve position control example . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Fig. 4.10 Valve position control scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Fig. 4.11 Gain scheduling: gap control and quadratic gain . . . . . . . . . . . . 134
Fig. 4.12 Tank level control: simple cascade control . . . . . . . . . . . . . . . . . 136
Fig. 4.13 Tank level control: with protective control . . . . . . . . . . . . . . . . . 137
List of Figures xxxix

Fig. 4.14 Tank level control: nested control . . . . . . . . . . . . . . . . . . . . . . . . 137

Fig. 4.15 Split point in split-range control . . . . . . . . . . . . . . . . . . . . . . . . . 138
Fig. 4.16 Split point calculation in split-range control . . . . . . . . . . . . . . . 139
Fig. 4.17 Examples of split-range control loop . . . . . . . . . . . . . . . . . . . . . 140
Fig. 4.18 Examples of split-point calculation with gains . . . . . . . . . . . . . 141
Fig. 4.19 Dual controller versus split-range . . . . . . . . . . . . . . . . . . . . . . . . 142
Fig. 4.20 Function blocks: PID controller, control splitter,
and control selector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Fig. 4.21 Examples of override control . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Fig. 4.22 Override control without anti-reset windup enabled . . . . . . . . . 157
Fig. 4.23 Data flow in an override control loop . . . . . . . . . . . . . . . . . . . . . 158
Fig. 4.24 Override control with anti-reset windup enabled . . . . . . . . . . . . 159
Fig. 4.25 Override control with an alternative configuration . . . . . . . . . . 160
Fig. 4.26 A PID control loop requiring bumpless transfer . . . . . . . . . . . . 164
Fig. 4.27 A PID control loop for back-calculation . . . . . . . . . . . . . . . . . . 165
Fig. 4.28 Back-calculation in a PID control loop . . . . . . . . . . . . . . . . . . . 166
Fig. 5.1 Architecture of advanced process control . . . . . . . . . . . . . . . . . . 172
Fig. 5.2 Type of models for control applications . . . . . . . . . . . . . . . . . . . 174
Fig. 5.3 Step response and step response model. Left: stable
process, right: ramp process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Fig. 5.4 An example of process identification (modeling) software
tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Fig. 5.5 Common model types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Fig. 5.6 An orifice plate for DP measurement . . . . . . . . . . . . . . . . . . . . . 182
Fig. 5.7 IPE for feedback control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Fig. 5.8 A sample implementation of IPE . . . . . . . . . . . . . . . . . . . . . . . . . 186
Fig. 5.9 Inferred TVP values, with hourly feedback from analyzer . . . . 187
Fig. 5.10 Inferred TVP values, with daily feedback from analyzer . . . . . . 187
Fig. 5.11 A generic fractionator column with typical measurements . . . . 189
Fig. 5.12 Interactions in a MIMO process . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Fig. 5.13 Raw model matrix for MPC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Fig. 5.14 Final model matrix for MPC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Fig. 5.15 Example of constrained control . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Fig. 5.16 Constrained control with MPC . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Fig. 5.17 A fractionator column, baselayer control scheme . . . . . . . . . . . . 199
Fig. 5.18 MPC benefits forecast based on maintenance level . . . . . . . . . . 200
Fig. 6.1 A gas compression process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
Fig. 6.2 Control options for supply and demand operation . . . . . . . . . . . 227
Fig. 6.3 Supply-driven liquid processing operation . . . . . . . . . . . . . . . . . 228
Fig. 6.4 Demand-driven liquid processing operation . . . . . . . . . . . . . . . . 228
Fig. 6.5 Supply-driven gas processing operation . . . . . . . . . . . . . . . . . . . 229
Fig. 6.6 Demand-driven gas processing operation . . . . . . . . . . . . . . . . . . 229
Fig. 6.7 Supply and demand operation with override protection . . . . . . . 229
Fig. 6.8 Supply-driven control with demand override . . . . . . . . . . . . . . . 230
Fig. 6.9 A gas compression process: key process variables . . . . . . . . . . . 232
xl List of Figures

Fig. 6.10 A gas compression process: with control valves . . . . . . . . . . . . . 233

Fig. 6.11 Step response of common process dynamics . . . . . . . . . . . . . . . 239
Fig. 6.12 Example of nonlinear process: pH process gain . . . . . . . . . . . . . 241
Fig. 6.13 Visualization tools for process analyses . . . . . . . . . . . . . . . . . . . 242
Fig. 6.14 Scope and checklist for process analysis . . . . . . . . . . . . . . . . . . . 243
Fig. 7.1 From process analysis to control design . . . . . . . . . . . . . . . . . . . 246
Fig. 7.2 Overall control strategy for E&P process, supply-driven . . . . . 249
Fig. 7.3 Overall control strategy for E&P process, demand-driven . . . . 249
Fig. 7.4 An acetylene hydrogenation process . . . . . . . . . . . . . . . . . . . . . 252
Fig. 7.5 A gas compression process: control strategy . . . . . . . . . . . . . . . 261
Fig. 7.6 Scope and checklist for process control design . . . . . . . . . . . . . 274
Fig. 8.1 PID loop troubleshooting procedure . . . . . . . . . . . . . . . . . . . . . . 278
Fig. 8.2 Bad process design: gas compression . . . . . . . . . . . . . . . . . . . . . 285
Fig. 8.3 Bad process design: supply/demand imbalance . . . . . . . . . . . . . 286
Fig. 8.4 Bad process design: unreliable cause and effect . . . . . . . . . . . . 287
Fig. 8.5 Bad control design: over-specified pressure control . . . . . . . . . 288
Fig. 8.6 Bad control design: over-specified flow control . . . . . . . . . . . . . 289
Fig. 8.7 Process dynamics verses loop dynamics . . . . . . . . . . . . . . . . . . 290
Fig. 8.8 Amount of flaring before and after PID tuning . . . . . . . . . . . . . 296
Fig. 8.9 Scatter plot of PV verses OP for a normal PID loop . . . . . . . . . 297
Fig. 8.10 PID data trending: sticky valve . . . . . . . . . . . . . . . . . . . . . . . . . . 298
Fig. 8.11 PID data trending: severe stiction . . . . . . . . . . . . . . . . . . . . . . . . 299
Fig. 8.12 PID data trending: poor performance . . . . . . . . . . . . . . . . . . . . . 299
Fig. 8.13 PID data trending: slippery valve . . . . . . . . . . . . . . . . . . . . . . . . 300
Fig. 9.1 Closed-loop transfer function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
Fig. 9.2 PID closed-loop response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Fig. 9.3 Oscillating loop due to too strong integrating action . . . . . . . . . 310
Fig. 9.4 Oscillating loop after tuning change . . . . . . . . . . . . . . . . . . . . . . 311
Fig. 9.5 Failure example of trial-and-error tuning . . . . . . . . . . . . . . . . . . 312
Fig. 9.6 Model-based PID tuning procedure . . . . . . . . . . . . . . . . . . . . . . . 315
Fig. 9.7 Step response of a stable process . . . . . . . . . . . . . . . . . . . . . . . . . 316
Fig. 9.8 Step testing with multiple step changes . . . . . . . . . . . . . . . . . . . . 317
Fig. 9.9 Process and process model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Fig. 9.10 Step response of a stable process . . . . . . . . . . . . . . . . . . . . . . . . . 318
Fig. 9.11 Step response of a ramp process . . . . . . . . . . . . . . . . . . . . . . . . . 319
Fig. 9.12 A three-phase separator with level-flow cascade loop . . . . . . . . 326
Fig. 9.13 Trending the flow controller data . . . . . . . . . . . . . . . . . . . . . . . . . 327
Fig. 9.14 PV-versus-OP plot for a flow loop . . . . . . . . . . . . . . . . . . . . . . . . 328
Fig. 9.15 Trending the flow controller data (incomplete) . . . . . . . . . . . . . . 328
Fig. 9.16 X-Y plot for a flow loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
Fig. 9.17 Level in a vessel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
Fig. 9.18 Vessel sizing information for level tuning . . . . . . . . . . . . . . . . . . 333
Fig. 9.19 Vessel cross-section area calculation . . . . . . . . . . . . . . . . . . . . . . 334
Fig. 9.20 Nonlinearity of conic shaped vessel . . . . . . . . . . . . . . . . . . . . . . . 336
Fig. 10.1 Transformation of control signals in a PID control loop . . . . . . 342
List of Figures xli

Fig. 10.2 Dual-range measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353

Fig. 10.3 Flow meter components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
Fig. 10.4 Nonlinearity in pressure compensation . . . . . . . . . . . . . . . . . . . . 371
Fig. 10.5 Boiler drum level indicator: gauge glass . . . . . . . . . . . . . . . . . . . 373
Fig. 10.6 Boiler drum level measurement: DP-based transmitter . . . . . . . 374
Fig. 11.1 A gas ejector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
Fig. 11.2 Instrumentation for an ejector . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
Fig. 11.3 Control design for an ejector: fixed capacity . . . . . . . . . . . . . . . 388
Fig. 11.4 Control design for an ejector: variable capacity . . . . . . . . . . . . . 389
Fig. 11.5 Control design for an ejector: fully rated for vacuum . . . . . . . . 389
Fig. 11.6 A typical pumping process, with difference in elevation . . . . . . 391
Fig. 11.7 Flows in pumping process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Fig. 11.8 A typical pump curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
Fig. 11.9 Capacity control for fixed-speed pumping . . . . . . . . . . . . . . . . . 397
Fig. 11.10 Capacity control for variable-speed pumping . . . . . . . . . . . . . . 398
Fig. 11.11 Process flow, recycle flow, and pump flow . . . . . . . . . . . . . . . . . 398
Fig. 11.12 Head–flow pump curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Fig. 11.13 Head–flow and head–flow-squared curve . . . . . . . . . . . . . . . . . . 401
Fig. 11.14 Protective flow control for centrifugal pump . . . . . . . . . . . . . . . 401
Fig. 11.15 Pump startup with a throttling valve . . . . . . . . . . . . . . . . . . . . . . 407
Fig. 11.16 Pump startup with double block valves . . . . . . . . . . . . . . . . . . . 407
Fig. 11.17 A single-stage gas compression unit . . . . . . . . . . . . . . . . . . . . . . 411
Fig. 11.18 The three variables that define compressor dynamics . . . . . . . . 413
Fig. 11.19 Compressor performance curves (Left: fixed speed;
Right: variable speed) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
Fig. 11.20 Compressor resistance curve(s) . . . . . . . . . . . . . . . . . . . . . . . . . . 414
Fig. 11.21 Compressor operating points . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Fig. 11.22 A two-speed hair dryer with an add-on concentrator . . . . . . . . . 415
Fig. 11.23 Performance curves of a two-speed hair dryer . . . . . . . . . . . . . . 416
Fig. 11.24 Compressor operating envelope . . . . . . . . . . . . . . . . . . . . . . . . . 416
Fig. 11.25 Surge and reverse flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Fig. 11.26 Operating requirements of gas compression process . . . . . . . . . 421
Fig. 11.27 Supply-driven control of a variable-speed compressor . . . . . . . 422
Fig. 11.28 Demand-driven control of a variable-speed compressor . . . . . . 422
Fig. 11.29 Supply-driven control of a variable-speed compressor,
with demand override . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
Fig. 11.30 Surge reference line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
Fig. 11.31 Basic anti-surge control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
Fig. 11.32 Anti-surge control with high discharge pressure override . . . . . 430
Fig. 11.33 A typical integrated compressor control scheme . . . . . . . . . . . . 431
Fig. 12.1 A free water knockout drum with gas blanketing . . . . . . . . . . . 435
Fig. 12.2 Tank gas blanketing control: a naive design . . . . . . . . . . . . . . . . 437
Fig. 12.3 Tank gas blanketing control: a improved design . . . . . . . . . . . . 438
Fig. 12.4 Tank gas blanketing control: split point . . . . . . . . . . . . . . . . . . . 438
Fig. 12.5 Tank gas blanketing control: a good design . . . . . . . . . . . . . . . . 439
xlii List of Figures

Fig. 12.6 Boiler drum level process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441

Fig. 12.7 Inverse response of boiler drum level . . . . . . . . . . . . . . . . . . . . . 443
Fig. 12.8 One-element control for boiler drum level . . . . . . . . . . . . . . . . . 445
Fig. 12.9 Two-element control for boiler drum level . . . . . . . . . . . . . . . . . 446
Fig. 12.10 Three-element control for boiler drum level . . . . . . . . . . . . . . . 447
Fig. 12.11 From wells to test separator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
Fig. 12.12 From wells to test separator, with multi-port selection
valve (MSV) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
Fig. 12.13 A three-phase test separator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
Fig. 12.14 Basic control scheme for a test separator . . . . . . . . . . . . . . . . . . 455
Fig. 12.15 Semi-dump control test separator level . . . . . . . . . . . . . . . . . . . . 456
Fig. 12.16 Level and flow during semi-dump control . . . . . . . . . . . . . . . . . 457
Fig. 12.17 A three-phase test separator, with double weirs . . . . . . . . . . . . . 459
Fig. 12.18 Process flow scheme for a firewater process . . . . . . . . . . . . . . . 462
Fig. 12.19 Pump curve in a firewater process . . . . . . . . . . . . . . . . . . . . . . . 463
Fig. 12.20 Firewater control scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
Fig. 12.21 Firewater control sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
Fig. 12.22 An acetylene hydrogenation process . . . . . . . . . . . . . . . . . . . . . 467
Fig. 12.23 Acetylene hydrogenation process control: design #1 . . . . . . . . 470
Fig. 12.24 Acetylene hydrogenation process control: design #2 . . . . . . . . 470
Fig. 12.25 Acetylene hydrogenation process control: design #3 . . . . . . . . 471
Fig. 12.26 Acetylene hydrogenation process control: design #4 . . . . . . . . 471
Fig. 12.27 Process overview diagram of an E&P surface facility . . . . . . . . 473
Fig. 12.28 Plant-wide control: key process variables . . . . . . . . . . . . . . . . . 475
Fig. 12.29 Plant-wide control: normal regulatory control . . . . . . . . . . . . . . 477
Fig. A.1 Volume of a cuboid vessel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Fig. A.2 Volume of a spherical vessel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Fig. A.3 Volume of a vertical cylindrical vessel . . . . . . . . . . . . . . . . . . . . 483
Fig. A.4 Volume of a conical vessel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
Fig. A.5 Volume of a horizontal cylindrical vessel . . . . . . . . . . . . . . . . . . 485
List of Tables

Table 1.1 Instrument identification letters . . . . . . . . . . . . . . . . . . . . . . . . . 19

Table 2.1 Type of controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Table 2.2 Typical process measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Table 2.3 Type of flow meters and their measuring principles . . . . . . . . . 32
Table 2.4 Comparison of flow measurement technologies
(G—gas, L—liquid, S—steam, M—mass flow,
V—volume flow) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Table 2.5 Types of valves and their pros and cons for control . . . . . . . . . 35
Table 2.6 Comparison of on/off control and PID control . . . . . . . . . . . . . 42
Table 2.7 Common complex control loops . . . . . . . . . . . . . . . . . . . . . . . . 43
Table 2.8 Process control deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Table 3.1 PID equation types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Table 3.2 PID discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Table 3.3 Pseudo-codes for PID implementation . . . . . . . . . . . . . . . . . . . 82
Table 3.4 Mode of operation for PID controller in Yokogawa
Centum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Table 3.5 Common operating modes of PID controllers . . . . . . . . . . . . . 87
Table 3.6 Different names for the PID parameters . . . . . . . . . . . . . . . . . 87
Table 3.7 Common PID attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Table 3.8 Operating modes of PID controllers by DCS vendors . . . . . . 89
Table 3.9 Typical configuration parameters of a PID controller . . . . . . . 103
Table 3.10 Control signal and valve fail-safe position . . . . . . . . . . . . . . . . 104
Table 3.11 Loop components and the sign of loop gain . . . . . . . . . . . . . . 104
Table 3.12 AO and controller valve should produce a positive gain . . . . . . 105
Table 3.13 Determining PID direction of action . . . . . . . . . . . . . . . . . . . . . 105
Table 4.1 Changing number of degree of freedom with complex
control loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Table 4.2 Flow balancing control logic . . . . . . . . . . . . . . . . . . . . . . . . . . 130
Table 4.3 Output mapping in a split-range control . . . . . . . . . . . . . . . . . 139
Table 4.4 PID-based complex control loops . . . . . . . . . . . . . . . . . . . . . . 144
Table 4.5 Function blocks in fieldbus control system . . . . . . . . . . . . . . . 145

xliii
xliv List of Tables

Table 4.6 Complex control loops and function blocks . . . . . . . . . . . . . . 149

Table 4.7 Common configurations for complex PID loops . . . . . . . . . . . 153
Table 4.8 Equation type for complex PID loops . . . . . . . . . . . . . . . . . . . 153
Table 4.9 Mode of operation for complex PID loops . . . . . . . . . . . . . . . 154
Table 5.1 Type of process dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Table 5.2 Multivariable interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Table 5.3 Examples of process disturbances . . . . . . . . . . . . . . . . . . . . . . 198
Table 6.1 PGC example: cause and effect table, preliminary . . . . . . . . . 234
Table 6.2 PGC example: cause and effect table, updated . . . . . . . . . . . . 237
Table 7.1 The PGC example: cause and effect table, with process
gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Table 7.2 PGC example: cause and effect table, final . . . . . . . . . . . . . . . 258
Table 7.3 Process control hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Table 7.4 PGC example: cause and effect table, updated . . . . . . . . . . . . 261
Table 7.5 The phased process control deliverables . . . . . . . . . . . . . . . . . 265
Table 7.6 Essentials elements of control narratives . . . . . . . . . . . . . . . . . 267
Table 7.7 Sample control narratives for compressor anti-surge
control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
Table 7.8 Sample control narratives for compressor capacity
control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
Table 7.9 Control loop configuration (sample) . . . . . . . . . . . . . . . . . . . . . 269
Table 8.1 Common control issues and possible causes . . . . . . . . . . . . . . 276
Table 9.1 Typical values for process gains and controller tunings . . . . . . 309
Table 9.2 Loop dynamics and controller tunings . . . . . . . . . . . . . . . . . . . 312
Table 9.3 Controller tuning versus controller response . . . . . . . . . . . . . . 312
Table 9.4 Ziegler–Nichols tuning method . . . . . . . . . . . . . . . . . . . . . . . . . 313
Table 9.5 Modified Cohen–Coon method for PID tuning . . . . . . . . . . . . 321
Table 9.6 The IMC tuning rules for stable process . . . . . . . . . . . . . . . . . . 321
Table 9.7 The IMC tuning rules for ramp process . . . . . . . . . . . . . . . . . . 321
Table 9.8 Level tuning parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
Table 10.1 Pressure friction loss versus flow rate . . . . . . . . . . . . . . . . . . . . 348
Table 10.2 Example of flow signal scaling and transformation . . . . . . . . . 350
Table 10.3 Flow compensation and conversion with density . . . . . . . . . . . 358
Table 10.4 Flow compensation/conversion
with pressure/temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
Table 10.5 Flow compensation verses flow conversion . . . . . . . . . . . . . . . 359
Table 10.6 Linear volumetric flow meter . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Table 10.7 Linear mass flow meter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
Table 10.8 DP-based volumetric flow meter . . . . . . . . . . . . . . . . . . . . . . . . 360
Table 10.9 DP-based mass flow meter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
Table 10.10 Compensation method lookup table . . . . . . . . . . . . . . . . . . . . . 362
Table 10.11 Antoine Coefficients for some compounds (data
from http://ddbonline.ddbst.com and Towler and Sinnott
(2012)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
Table 10.12 Bubble points: pressure and temperature in a debutanizer . . . . 369
List of Tables xlv

Table 10.13 Density of water and steam at different pressures . . . . . . . . . . 376

Table 11.1 Cause and effect table for ejector control . . . . . . . . . . . . . . . . . 384
Table 11.2 Cause and effect table for a centrifugal pump . . . . . . . . . . . . . 393
Table 11.3 Cause and effect table for pump capacity control . . . . . . . . . . 394
Table 11.4 Surge points from pump curves . . . . . . . . . . . . . . . . . . . . . . . . . 403
Table 11.5 Pump operating range by surge indicator . . . . . . . . . . . . . . . . . 404
Table 11.6 Loop configuration details for pump control . . . . . . . . . . . . . . 405
Table 11.7 Operating points of a two-speed hair dryer . . . . . . . . . . . . . . . . 415
Table 11.8 Cause and effect relationship for compressor operation . . . . . 417
Table 12.1 Cause and effect table for gas blanketing control . . . . . . . . . . . 435
Table 12.2 Loop configuration details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
Table 12.3 Cause and effect table for boiler drum level control . . . . . . . . 442
Table 12.4 Cause and effect table for test separator . . . . . . . . . . . . . . . . . . 453
Table 12.5 Cause and effect table for firewater process . . . . . . . . . . . . . . . 462
Table 12.6 Acetylene hydrogenation: preliminary cause and effect
table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
Table 12.7 Acetylene hydrogenation: cause and effect table
with cascade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
Table 12.8 Cause and effect table for the generic E&P process
in Fig. 12.28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
 Part I
The Big Picture

In a modern operating plant, once a plant is built and running, a core facility remain-
ing in the plant/field is the control room, which hosts the process control system and
personnel. Process control as an engineering specialty is indispensable to process
safety and production efficiency. Process control has its unique roles and responsi-
bilities, supported by distinct know-how and skills. It is thus irreplaceable by other
engineering specialties.
Process control works with ideas, algorithms, and software and is thus less visible
or transparent than other teams or specialties. As a result, although the process control
achievements are evident and substantial, how process control works is usually mys-
terious to most people and thus leads to many misunderstandings or under-valuing.
On the other hand, by the cross-disciplinary nature, process control must collaborate
closely with many other teams and specialties such as process engineering, rotating
equipment, control systems engineering, operations, instrumentation, and measure-
ment. It is crucial to raise the general awareness of process control among all the
related teams and specialties to unleash the full potential.
For this purpose, Chap. 1 of this book starts with a high-level overview of process
control and explains to the general audience what process control is, why it is needed,
and what it needed. More importantly, it clarifies the roles and responsibilities of
process control engineers and explains how they are different from others and why
close collaboration is necessary to achieve the best control design and optimal control
performance.
Process control evolved from general automatic control. Over the past century,
automatic control has developed a rich collection of theories and technologies. How-
ever, these theories and technologies are mainly targeted at aerospace or electrome-
chanical applications. Only a small subset is applicable to process control due to
industrial processes’ distinct characteristics and unique challenges. A process con-
trol practitioner, especially those new graduates joining the industry, may find it
discouraging: on the one hand, they are armed with sophisticated control theories
and techniques, from classical to modern; but on the other hand, they found little
relevance between what they have learned and what is actually needed. Lacking the
2 Part I: The Big Picture

necessary knowledge and skills to perform their assigned duties, they have to re-start
a long learning journey, and often need to learn the hard way.
To help junior engineers and people from other engineering backgrounds pick
up speed quickly, Chap. 2 in this part of the book summarizes the process control
technologies prevalent in practical use. This summary defines a general knowledge
framework based on practical needs, which may significantly differ from the knowl-
edge framework built from academic leanings. This summary can also serve as
an awareness education to non-process control people such as process and equip-
ment engineers, control system engineers and technicians, operators, and operation
management. It helps them understand what process control is capable of and the
irreplaceable roles in process safety and production efficiency.
Chapter 1
Process Control Overview

Process control is a vital part of process operation. On the one hand, it is indispensable
by function and irreplaceable by other engineering specialties. On the other hand,
process control is such a unique engineering specialty that it is not widely understood
or appreciated in an operating plant. Raising general awareness of process control
is a great way to help operation and production. This chapter provides a high-level
introduction to process control by starting in Sect. 1.1 with a brief explanation on
why and when process control is needed, where and how process control fits in. This
chapter also tries to clarify some confusion and misunderstandings on process control.
Section 1.2 describes the plant operation requirements and shows how operating
objectives and control objectives are linked and achieved. Finally, Sect. 1.3 puts
process control into the context of overall plant automation and explains the integral
role that process control plays in safe and efficient operation.

1.1 Introduction to Process Control

Process control is a unique engineering specialty often confused with other engineer-
ing specialties such as process engineering and control system engineering (instru-
mentation, control systems, etc.). In fact, there is a large gap between process engi-
neering and control systems engineering, and process control bridges this gap and
plays a critical role in a modern operating plant.

1.1.1 What Is Process Control?

A process stands for a series of unit operations to produce a material in large quan-
tities, either continuously or in batch. Typical industry processes include

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 3

S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_1
4 1 Process Control Overview

• oil and gas production and refining;

• petrochemical and chemical processing;
• biochemical and pharmaceutical;
• pulp and paper manufacturing;
• power generation plants;
• food and beverage.

Process control is a branch of automatic control that focuses on the theory and tech-
nology to operate the above industrial processes safely and efficiently. Process control
continuously monitors a process for deviations and abnormalities, takes immediate
action to correct those abnormalities, and brings the process back to its optimal state.
In a typical work process, process control starts with basic process engineering,
translates the engineering design and operating requirements into control objectives
and requirements, designs the proper process control strategy, specifies the func-
tional requirements, and eventually transforms it into the final control scheme on
the targeted control system infrastructure. Thus, process control’s primary role and
responsibility are designing, implementing, and maintaining a sound process control
solution to ensure safe and efficient operation.
To achieve this goal, process control works closely with process engineering
during process design to ensure that the engineering design delivers the required
controllability and stability. During the commissioning and operation phase, process
control works with control systems personnel to ensure the process design is imple-
mented and operated as designed. Process control is a very cost-effective solution for
improving process safety and production efficiency, typically with low investment
and high return (Liptak 2006).
Process control bridges the gap between process engineering, operation and main-
tenance, and control systems (Fig. 1.1). It overlaps and thus collaborates with several

Fig. 1.1 Process control is

cross-disciplinary
1.1 Introduction to Process Control 5

related engineering specialties. For this reason, process control engineers often have
a more coherent view of the entire process than other engineers.
The critical deliverable from process control is the control scheme in the control
systems and the up-to-date process control narrative (PCN) document.

1.1.2 Why Do We Need Process Control?

All commercial productions are for-profit, and thus maximizing profit is the ultimate
goal of the operation. In general, maximum profit is achieved through the highest
production rate of the right products at the best selling price and with the minimum
capital expenditure (CAPEX) and lowest operating expenditure (OPEX) (Fig. 1.2).
Full-capacity operation with the highest up-time ensures maximum production while
keeping the products consistently on-spec and meeting the high demand leads to the
best selling price.
In this highly volatile market, and with increasingly stringent requirements on
health, safety, and environmental (HSE), the operation must stay safe and efficient
to maximize profit. The key process variables must be maintained at their desired
values during normal operation and within a safe range during plant upsets. For
example, the vapor pressure of the crude oil from a stabilizer must be maintained
close to the design value to meet environmental regulations even when the feedrate
or the temperature/pressure condition fluctuates. Likewise, the temperature inside
a packed-bed reactor must be maintained at its design value to ensure the desired
conversion and prevent reaction run-away.
Automating operation is the proven way to improve safety and efficiency, and
process control is a crucial enabler of automated operation (Marlin 2015). Take
an upstream exploration and production (E&P) process as an example. As shown
with a process overview diagram (POD) in Fig. 1.3, the multi-phase fluid from the
production wells is collected and sent to the processing facility for separation into
gas, oil, and water, which are further processed and conditioned for sale, injection,

Fig. 1.2 Process control helps maximize profit

6 1 Process Control Overview

Fig. 1.3 Process overview diagram for an E&P plant

or disposal (Campbell 2004). The operating objective is to safely and efficiently

separate the well fluid, meeting the requirements on both quality and quantity for the
produced oil, gas, and water by controlling the process variables such as pressure
(P), flow (F), and quality (Q).
A sound process design (piping, equipment, line-up, material, and energy balance)
makes it possible to achieve this objective. However, in actual operation, constant
attention and intervention are needed to maintain the desired quality and quantity
due to inevitable disturbances such as continuous fluctuations in supply and demand,
frequent process upsets, and occasional failures. The attention and intervention are
provided by process control. Plant operators and the process control system work
together to ensure that the plant runs as desired. Therefore, process control consists
of manual process control (operator control) and automatic process control. Better
automatic control typically results in less operator control and vice versa.
Fully automated operation (“un-manned” operation) has long been an aspiration.
The automated operation is necessary for some process operations under harsh condi-
tions or with less stringent control requirements. However, fully automated operation
relies on credible data, reliable information, and optimal decisions, which are not
yet up to the expectation in complex operations. For mission-critical processes, the
fully automated operation is still not realistic due to abnormal operating conditions
that are beyond what machine intelligence can handle. Abnormal conditions include,
for instance, severe disturbances, the onset of a hazard, or emergencies.
Section 1.2 continues the discussion on operating objectives and control objec-
tives, while Chap. 6 is dedicated to the understanding and analysis of operating
needs.

1.1.3 When Do We Need Process Control?

A plant typically takes one to five years to design and build and is expected to operate
for 20 or more years. Experience shows that many operating difficulties can be traced
1.1 Introduction to Process Control 7

Fig. 1.4 Project execution: desired versus delivered

back to the process control design flaws. Operations have to live with or work around
these flaws for the plant’s entire life cycle. Even a tiny flaw in the design may result
in a significant accumulated loss over the many years of plant operation. Figure 1.4
illustrates the project life cycle and the impact of a bad design.
There has been a growing awareness of the criticality in integrating the process
design and the control design at the early stage in a new project (Sakislis et al.
2004). Process design should take process operability and controllability as critical
considerations. For this reason, process control should be involved as early as possible
in the project life cycle, preferably from the early conceptual design phase, to ensure a
sound overall control strategy. Early process control involvement can eliminate many
operating issues from the source and is thus much more cost-effective. Otherwise,
the design flaws may be carried over to operation phase to be fixed. Once the plant
is in operation, the troubleshooting effort is typically limited to local areas and ad
hoc improvements. As a result, there are limited opportunities to re-consider and
improve the control design’s overall integrity and optimality.

1.1.4 What Is Process Control Not?

Process control has a significant overlap with process engineering and control system
engineering (Fig. 1.1). However, process control is a very different specialty (King
2016; Lee and Weekman 1976), and the difference is not widely recognized. There
is a significant disconnect between process and control (Fig. 1.5) in many practical
operations, contributing to many day-to-day operational challenges.
8 1 Process Control Overview

Fig. 1.5 Disconnect

between process engineering
and control systems

1.1.4.1 Process Control Is Not Process Engineering

Process engineering is responsible for the design of a production process that can
potentially run. However, process engineers typically do not (nor are they required to)
know the available process control technologies or the control systems infrastructure.
It is the responsibility of a process control engineer to produce a process control
solution that meets the operating requirements with the available control technology
and infrastructure.
For this reason, a process control engineer views the process differently than the
process engineering people:
1. Static versus dynamic: Process engineers are more focused on the static behavior
of the process, such as steady-state material and energy balance. However, for
process control engineers, above and beyond the steady-state relationship, their
primary interest is in the dynamic behavior of the process (process dynamics),
including the dynamic material balance, dynamics cause and effect relationships,
and dynamic response characteristics of the process variables of interest.
2. Feedforward versus feedback: Process engineers are more customized to rely on
first-principle calculations to meet the material or energy balance requirements,
while process control engineers count more on feedback control to achieve the
same. For example, to control a temperature, a process engineer is more inclined
to propose elaborate calculations based on heat duty, heat transfer, and energy
balance to determine the required energy input. On the other hand, a process
control engineer is more likely to request an online temperature measurement and
use a simple feedback controller to adjust the fuel flow to achieve the temperature
target.
3. Absolute values versus incremental changes: Another difference between the two
professions is that process control engineers are more interested in the “delta
changes” than the absolute steady-state values that the process engineers are
interested in. For example, a temperature controller compares the temperature
measurement (process value) with the temperature setpoint (desired value) and
calculates the amount of heat duty to be removed or added (the incremental
changes in control action) to bring the temperature back to its setpoint. As long
as the temperature measurement is not the same as the desired temperature, the
controller will keep adding or removing heat! After all, the controller does not
care about the absolute value of the heat duty. The downside of this powerful
mechanism is that it can often drive the heat duty to saturation if it is not informed
of the heat duty limit.
1.1 Introduction to Process Control 9

This difference in thinking significantly influences the decision on how the plant
should be designed and controlled.

1.1.4.2 Process Control Is Not Control System Engineering

Control systems engineering is responsible primarily for the control system infras-
tructure, including the sensors, transmitters, final control elements, the distributed
control system infrastructure, safeguarding system, and the controllers in isolation.
As a result, they rarely (nor are they required to) care about the process design phi-
losophy or overall control strategy. Process control engineers develop the functional
specification based on the control strategy and pass them on to the control systems
engineers for implementation. With better knowledge of the available control sys-
tem infrastructure and capability, the control system engineers implement the control
solution in the targeted control platform. The focus of a process control engineer and
a control system engineer is thus quite different:
1. Software versus hardware: While a process control engineer is responsible for
providing the control strategy and the functional specification, a control system
engineer’s mission is to implement the functional specification with the available
hardware/software in the targeted control systems infrastructure.
2. Holistic view versus localized view: The process control engineer must have a
holistic view of the control solutions that connect with the process design, the
control systems, and the operation; while a control systems engineer is charged to
precisely achieve each specified control function on the given platform. They do
not necessarily care about the purposes of the control functions and if/how they
all work together.
Control systems engineers talk about devices, while process control engineers work
on ideas (Åström and Kumar 2014). Ideas are less visible than devices, and therefore,
there is a tendency of under-valuing process control.

1.2 Operating Objectives and Control Objectives

The ultimate goal of production is to maximize profitability, which demands safe

and efficient operation. Therefore, the control objectives must serve the operating
objectives to make the operation safe and efficient.

1.2.1 Process Flow and Operating Objectives

A process is the continuous or batch processing of certain materials. Feeds go into

the process, and products come out of the process. Process engineering design puts
10 1 Process Control Overview

Fig. 1.6 Operating objectives: from design to operation

together the proper piping and equipment to ensure the material’s movement and
quality meet the business requirements (Fig. 1.3). The ultimate objectives are to make
the plant run, and run it safely and efficiently (Fig. 1.6). A sound process design is
a prerequisite for a solid process control solution. By design, a process must be
inherently operable. In other words, a plant must be able to run manually (start, run,
and shutdown) before it can be operated automatically.1 The process must be able
to continuously operate to produce the desired material up to specifications if the
designed conditions are provided.
Chapter 6 provides a detailed discussion on how to dissect and understand a given
process in order to control it.

1.2.2 Control Objectives and Control Strategy

Process control makes an operable process operative, i.e., run as per desired perfor-
mance requirements. Process control, consisting of operator control and automatic
control, provides the required operating condition for the process to continuously
operate to produce the on-spec product in the desired quantity and quality.
To serve the operating objectives, a process control solution must meet the fol-
lowing operating requirements:
1. During normal operation, deliver the right product with the right quality and
quantity.
2. During process upsets, temporarily deviate from the operating target but stay
inside the operating envelope.
3. During operation change, ensure a safe and quick transition from one condition
to another.

1Operators always have the privilege to override automatic control in case of an emergency. One
of the design flaws in the Maneuvering Characteristics Augmentation System (MCAS) in Boeing
737 Max that caused the two catastrophic crashes in 2019 was that the MCAS system is given the
privilege of overriding the pilot’s control. Consequently, one of the improvements is “... MCAS
will never override the pilot’s ability to control the airplane using the control column alone.” (from
https://www.boeing.com/737-max-updates/mcas).
1.2 Operating Objectives and Control Objectives 11

The general control strategy typically consists of the following three levels of
control actions:
1. Regulatory control: Maintain the operating point at (or move it to) the desired
quantity and quality target.
2. Protective control: Keep the operating point within the operating envelope during
process upsets.
3. Operator control: When automatic control fails to react correctly or timely, the
operator intervenes and makes manual corrections.
The basis for control design is the process flow, supply and demand model, cause
and effect relationship, and dynamic response behavior of the concerned process
variables. Therefore, it is essential to follow a systematic approach to understand
the underlying process and ensure the control strategy is adequate at all levels of the
plant operation (Stephanopoulos 1984). See Chap. 7 for the in-depth discussion on
control design.
The control strategy is typically designed by process control engineers and imple-
mented by control system engineers. The implementation involves the following
activities:
1. Place the right sensors and transmitter at the right location to provide and necessary
measurements. Process measurements provide information on the current status
of the process conditions.
2. Position the proper final control elements such as valves or motors at the right
location. The final control elements are the means to influence the operation.
3. Perform the required calculations of the corrective moves based on the measure-
ments and adjust the final control elements to maintain the operation at the desired
targets, specified either by the operators or higher-level application in the control
hierarchy.
Figure 1.7 shows the locations of some process measurements such as pressure,
flow, and quality, along with the locations of the final control elements such as control
valves and electric motors for pumps and compressors.

Fig. 1.7 Overall control strategy for a general E&P process

12 1 Process Control Overview

1.2.3 Process Control for Process Safety

Operating safety is of paramount importance in the design and operation of a plant.

Unsafe operation is the most significant risk to profitability since one fatal accident
can easily wipe out a whole year’s profit or even force the plant to close down
permanently.
Operating safety is achieved through multiple layers of safety measures, as shown
in Fig. 1.8, where process control is the first layer of protection against unsafe oper-
ations! This layered setup is sometimes nicknamed the “safety onion.”
Operating safety typically consists of two parts:
1. Process safety: Process safety concerns the prevention of fires, explosions, and
accidental releases of hydrocarbons and chemicals in an operating facility dealing
with hazardous materials. Process safety is related to design and engineering and
has not received adequate attention compared to human safety. From the safety
onion in Fig. 1.8, inherently safe process design is the fundamental requirement
by process safety, while process control is the first line of defense against unsafe
operations.

Fig. 1.8 Layers of process safety protections

1.2 Operating Objectives and Control Objectives 13

2. Human safety: If both the process control and safeguarding layers fail, incidents
may happen, and human safety may be endangered. Proper personal protection
equipment (PPE) and standard operating procedures (SOP) are the primary tools to
protect human safety. However, as shown in Fig. 1.8, improving the control and
safeguarding effectiveness is more efficient than passive protection for human
safety.
In this layered protection design, each layer is a vast topic by itself. Only a very
brief description is provided here:
1. Process design: The first and utmost assurance for safe and profitable plant opera-
tion is through an engineering design that is inherently safe and stable. That means,
during normal operation, the safety requirements such as material balance, energy
balance, equipment sizing, the material of construction, and pressure rating are all
adequately fulfilled. When operation conditions change, the plant can typically
settle by itself to a new steady-state, which may not be the desired condition but
is still safe.
Most processes are inherently stable. For example, even without control, a heat
exchanger’s outlet temperature cannot increase without bound due to the limit in
heat duty. However, there are also many unstable processes such as vessel levels
and most exothermic chemical reactions. These unstable processes require much
more stringent control and safeguards to keep the operation within the safe limits.
2. Process control: All plants are designed to operate within a particular operating
region, defined by constraints and limits based on safety, operability, and operat-
ing efficiency. This operating region is called the operating envelope. There are
usually one or several optimal operating points within the operating envelope that
yield the best profitability for operation.
Process control ensures that the plant operates at or near the optimal operat-
ing points during normal operation and stays inside the operating envelope dur-
ing process upsets. Operators perform process control manually in the old days,
while automatic control has been the norm in a modern operating plant. However,
operator intervention is required during abnormal operating conditions such as
severe disturbance, extreme turndown, emergency, or startup/shutdown. Alarms
and alerts are the means to notify the operator of abnormal conditions.
In a modern control system, alarms are so easy to add to the extent of being abused.
Operators are often distracted or even overwhelmed by nuisance alarms. Alarms
are meant for the operator to take action. Matters that are not worthy of operator
attention should not be configured as an alarm. To ensure the effectiveness of
alarms, many operating plants have taken alarm rationalization as a mandatory
exercise.
3. Safeguarding system: The purpose of the safeguarding system is to protect against
human injuries, loss of assets, and pollution of the environment by proactively and
quickly shutting down the operation area at risk once an out-of-control situation
occurs. A shutdown of the operation incurs costs and losses by itself; however, it
is a proactive action to prevent more severe consequences.
14 1 Process Control Overview

A safeguarding system typically consists of a safety instrument system (SIS) and

a fire and gas mitigation system (FGS). SIS typically triggers a unit shutdown,
while a fire and gas incident usually forces a plant-wide shutdown. The standard
safeguarding functionalities include trips, interlocks, and emergency shutdowns:
a. Safety instrument system (SIS): In the case of a major process upset, the
process control layer may be too slow or insufficient in response to prevent
the operation from going outside the operating envelope. The “out of control”
operation can result in off-spec products or even unsafe operations.
The SIS layer takes automatic, independent, and fast action to prevent a haz-
ardous incident from occurring and protect personnel and plant equipment
against potentially serious harm. Once a severe failure is detected, the safe-
guarding layer proactively shuts down (trip) the operation to prevent the con-
dition from further deterioration.
b. Fire and gas system (FGS): If the SIS system also fails, a major incident may
follow, typically in the form of loss of primary containment (LOPC). A gas
leak or liquid spill is a direct result, leading to potential fires and explosions.
For certain operations with toxic gases (e.g., with high H2 S concentration), loss
of containment can be a direct fatal threat to human life in the surroundings.
The F&G system takes proactive action to reduce the hazardous event’s con-
sequences after it has occurred. Its purpose is to mitigate and minimize the
damages. Upon detecting a fire or gas incident, the FGS proactively shut down
the plant operation (emergency shutdown).
4. Mitigation by physical or mechanical protection: The safeguarding system typi-
cally relies on power and instrument air to stop the plant operation and bring it
to safety. In case of loss of power or instrument air, mechanical protection pro-
vides another layer of protection. The goal is to prevent the damage from further
escalation. Examples include pressure relief valves or containment barriers. For
example, in a high-pressure low-density polyethylene (LDPE) plant, the operating
pressure can be as high as over 2,000 atm. The hyper-compressor and the pipings
must be surrounded by a concrete protection wall (“blast wall”) up to 10 meters
high and 2 meters thick against a potential explosion.
5. Emergency response: The plant and community emergency responses, including
firefighting brigade and site/community evacuation procedures, are the last resort
to respond to a major incident.
In summary, sound process engineering provides the basis for safe and efficient
operation; process control is responsible for running the plant at the designed operat-
ing point and protects it from running out of the operating envelope during production
upsets.
1.2 Operating Objectives and Control Objectives 15

Fig. 1.9 More profitable production through stabilization and optimization

1.2.4 Process Control for Production Efficiency

Process control stabilizes operation and reduces uncertainty; therefore, it allows the
plant to operate closer to the operating limits for better profitability. See Fig. 1.9 for
an illustration of this stabilize and optimize philosophy.
Most advanced process control solutions such as model predictive control (MPC,
Chap. 5) have optimizing engines built-in to drive the operating point closer to the
most profitable operating region. More advanced production optimization can be per-
formed by dedicated applications such as real-time optimization (RTO, Sect. 2.3.7).
Chapters 6 and 7 provide in-depth discussions and general guidelines on analyzing
the process requirements and producing the proper process control solutions.

1.3 Process Control in Context

From the operating objectives (Fig. 1.6) and the layers of process safety protection
(Fig. 1.8), it is evident that process control is an essential and integral part of the
overall operation and safety establishment. They work together to achieve the desired
safe and efficient production.
16 1 Process Control Overview

1.3.1 Process Automation Hierarchy

Process control is implemented in a layered fashion, as shown in Fig. 1.10. This

layered layout is sometimes called the automation pyramid. Each layer relies on the
layers below it to function correctly. At the same time, each layer provides the basis
for the layers above it to perform the higher-level functions:

1. The physical plant: The safe and efficient operation of the plant is the ultimate
goal of process control and optimization.
2. The sensor and transmitter layer provides the necessary view into the plant opera-
tion variables (process measurements). The final control elements such as valves
and motors provide the means to influence the plant operation.
3. The baselayer controls, predominantly PID controllers implemented in a control
system environment such as distributed control system (DCS) or programmable
logic control (PLC) system, perform the real-time control functions to keep the
plant operation at the optimal operating points and within the operating envelope.
4. Advanced process control (APC) typically consists of complex PID control
schemes and model predictive control (MPC) applications, focusing on the stable
operation of a more extensive process area such as an entire distillation column
or a compressor unit.
5. Real-time optimization (RTO) uses a combination of detailed mathematical mod-
els, explicit economic objective functions, and real-time access to the control
systems to maximize the profitability of an operating unit or even the entire plant.

Fig. 1.10 Plant automation hierarchy

1.3 Process Control in Context 17

6. Planning and scheduling: Management information and decision system for mak-
ing mid-term planning and scheduling decisions on production, inventory, ship-
ping, and purchasing.
7. Enterprise resource planning (ERP): ERP is tied to corporate supply-chain man-
agement and sets the long-term operating targets for the control and optimization
applications.

The higher-level applications have a broader scope of concern and a lower fre-
quency of execution. They generate the operating directives for the next layer below.

1.3.2 Process Automation System (PAS)

A plant automation system (PAS) is typically a combination of distributed control

systems (DCS) and programmable logic control (PLC) based platforms. A typi-
cal PAS infrastructure consists of the following three subsystems, as illustrated in
Fig. 1.11:
1. Distributed control system (DCS): A DCS is the core platform and operator inter-
face for monitoring and control. The DCS system typically also acts as the human–
machine interface for the operators for all other subsystems.
2. Safety instrumented system (SIS): SIS is typically a PLC-based system running
condition-based logic and sequences. In case of emergency, it performs shutdown
of process and equipment, isolation of hydrocarbon inventories, switch off of elec-
tric systems, depressurization, blow-down, and emergency ventilation control.
3. Fire and gas system (FGS): The fire and gas system continuously monitors all
plant areas for abnormal conditions such as a fire or combustible/toxic gas release
via various fire and gas detection instruments. In the event of a hazardous situation,
the system alerts the operator of the hazardous situations. Depending upon the
hazard level, the fire and gas system will also proactively shut down the process
equipment and activate automatic fire fighting systems to prevent escalation of
the incident.
In addition to the DCS and PLC, there may be vendor-provided, dedicated control
and protection packages. Ideally, all vendor packages are interfaced with the DCS
to provide the operator with a single standard interface. Through PAS, operators and
engineers can remotely monitor and control the plant from a central control room
(CCR).
For maximum reliability and integrity and to minimize the chance of common
failures affecting both control and safeguarding, field instruments for each subsystem
are functionally and physically segregated from each other. However, all the displays
and alarms are typically routed to the DCS to be viewed and operated in the central
control room (CCR). Sufficient measurements and associated display functions such
as trends and recording functions, indications, and alarms are provided to help detect
18 1 Process Control Overview

Fig. 1.11 Process automation system

abnormal conditions in the plant and give the operator the possibility to observe the
development of imminent disturbances.
For legacy or practical reasons, full PAS integration may not be realistic, and
variable levels of integration exist from plant to plant.

1.3.3 Symbology and Identification

A process variable may need to be measured, monitored, controlled, and safeguarded.

A standard representation, including naming and symbols, facilitates communica-
tion, documentation, and operation. The process control schematics typically follows
the symbol and identification standards of ISO 3511, or ISA S5.1 and S5.3, which
include the standard naming convention as follows:

nXY-nZ Example: 32FICA-101A, 12AI-201.

The first part (n-) is a two to three-digit number indicating the process unit. The two
or more letters following the unit number designate the type of measurement (X) and
its function (Y); The dash “-” is optional but preferred for clarity. The number after
the letters is a two to four-digit serial number, and the optional suffix (Z) can make
the tag name uniquely identifiable.
For example, tag name 32FICA-101A indicates that this is a process variable in
unit #32. It is a flow measurement (F-) with an indication in DCS (-I-). It performs a
control function (-C-) with alarm(s) (-A-) configured. The tag serial number is 101
with a suffix (-A), indicating that there may be other tags with a similar tag name,
e.g., 32FICA-101B. See Table 1.1 for the commonly encountered identification
letters. For a complete list, please consult the ISO-3511 or ISA-S5 standards.
1.3 Process Control in Context 19

Table 1.1 Instrument identification letters

Letter As measurement code As function code Examples
A Analysis Alarm 21AI-102 analyzer indicator
E Voltage Sensor 21FE-101 flow sensor
D Differential 21PDI-101 differential pressure indicator
F Flow Ratio 12FICA-101 flow control
32FFC-211 flow ratio control
H Hand 21HIC-101 hand control
I Current Indicator 21AI-102 analyzer indicator
J Power 21JI-101 power indicator
K Time, schedule 21KS-121 on/off switching control
L Level 21LIC-101 level control
P Pressure 21PC-101 pressure control
Q Quantity Totalize 21FQI-101 flow totalization
S Speed, frequency Switch 21FS-101 flow switching control
T Temperature Transmitter 21PT-101 pressure transmitter
U Multivariable Multi-function 21UC-102 MPC control or complex control
21UZ-102 a SIL trip function
V Valve 21FCV-101 flow control valve
21FZV-101 flow shutdown valve
21KSV-101 on/off switching valve
X Unclassified 23XC-101 load controller
Y Event, State Calculation 23FYI-101 flow compensation calculation
Z Safeguarding 23FZA-101 flow tripping with alarm

The full tag names are unique within the same operating plant, while the serial
number is unique in each group of instruments sharing the same measurement and
function codes. On the other hand, the different functions for the same process vari-
able share the same serial number. For instance, the sensor, transmitter, indicator, con-
trol, and safeguarding for a flow variable can be named 12FE-101A, 12FT-101A,
12FI-101A, 12FC-101A, 12FZ-101A, respectively, all with the same serial
number. The indication and alarm function may be added to the control function to
become 12FICA-101A. A redundant flow measurement is named 12FICA-101B.
Although it is necessary to specify the full tag name on P&ID for proper construc-
tion and configuration, the tag names are often simplified for clarity where space is
of concern. For example, on a DCS schematic, the flow tag 23FICA-101A may be
reduced to 13FC101A without sacrificing clarity. In addition, the instrument symbol
23FT-101A is often omitted without loss of clarity (see Fig. 1.12).
Some of the typical instrument symbols for indicators, controllers, or safeguarding
units used in this book are illustrated in Fig. 1.13, and more details can consult the
ISO and ISA standards. For clarity, the unit number of all the tag names is omitted
for the clarity of the graphics.
20 1 Process Control Overview

Fig. 1.12 Simplification of symbols

Fig. 1.13 Example of control and safeguarding

1.3.4 Control and Safeguarding

The process safety protection setup in Fig. 1.8 shows that the safety instrument sys-
tem (SIS) is the next layer of defense against unsafe operation after process control.
Actions from the safeguarding layer are typically unit shutdown or plant-wide shut-
down and are thus very disruptive. For instance, the emergency shutdown of large
equipment such as a gas compressor can potentially be damaging. Close integration
between process control and safeguarding is thus crucial in an integrated design.

Example 1.1 Control and safeguarding of a gas/liquid separator. Figure 1.13 is an

example of a typical two-phase separator in an upstream oil and gas production
facility. A separator receives the multi-phase fluid from wells and separates it into
gas and liquid under the proper pressure and interface level.
1.3 Process Control in Context 21

There are four layers of control and protection in this process configuration:
1. Normal regulatory control: The level controller LC-101 and the pressure con-
troller PC-101A provide continuous regulatory control during normal operation.
The stable level and pressure ensure satisfactory separation of gas and liquid in
the separator.
2. Abnormal protective control: If the regulatory pressure control PC-101A fails
to maintain the desired pressure, the pressure may rise above the high limit set by
the protective controller PC-101B, which will open the valve to flare. Temporary
flaring will stop the pressure from going dangerously high while waiting for the
operation to come back to normal.
3. Instrumented safeguarding: If the pressure continues to go up even with the
protective controller active (or failing), then the safeguarding pressure trip
PZ-101 proactively triggers a process shutdown through UZ-101 to close all the
inlet/outlet isolation values to ensure positive containment of the hydrocarbons.
Similarly, PZ-101A and PZ-101B safeguard the vessel against high or low
levels by proactively triggering a shutdown and close the inlet or outlet isolation
valves.
4. Pressure relief valves: In the rare case of power or instrument air failure, the
safeguarding layer fails to act, and the pressure goes up to a dangerous level; the
relief valve PSV-101 will pop open to stop the pressure from bursting the vessel.

1.3.5 Operation by Setpoints and Limits

The different layers of control and protection are achieved by operating at different
setpoints and limits in a staggered fashion. Figure 1.14 provides a list of all the poten-
tial setpoints and limits, with SP being the desired setpoint, SPH/SPL the setpoint
high/low limit, H/L the high/low alarm limits, and HH/LL the high/low trip limits.
The setpoints and limits are carefully specified based on process design and oper-
ating requirements. Most of the time, only a subset of the limits and setpoints is
required. For example, an operation may be high constrained (Fig. 1.14 left), low
constrained, or both (Fig. 1.14 right).
In Fig. 1.13, the normal operating pressure inside the vessel is controlled by the
pressure controller PC-101A. The protective controller PC-101B is assigned with
a setpoint higher than that of PC-101A. During normal operation, the protective
controller PC-101B is inactive, and the flare valve PCV-101B remains closed.
Only when the normal regulatory pressure controller PC-101A fails to maintain the
pressure at the normal setpoint and the vessel pressure rises above the PC-101B
setpoint will this protective controller PC-101B becomes active. The protective
controller opens the flare valve to let go of the vessel’s excessive gas to flare (a
waste!).
If both the normal controller PC-101A and protective controller PC-101B fail to
maintain the vessel pressure and the pressure surpasses the high alarm limit, an alarm
22 1 Process Control Overview

Fig. 1.14 Operating setpoints and limits

will sound to call for operator attention and intervention. If the operator intervention
is unavailable or unsuccessful, and the pressure continues to rise and reaches the trip
limit, the safeguarding element PZ-101 will kick in and proactively shut down the
separator operation by closing all the inlet and outlet isolation valves.
In summary, a process is designed to operate under different setpoints and lim-
its, including normal control setpoint, protective control setpoint, alarm limits, and
tripping limit. If automatic control fails, an alarm alerts the operator for manual inter-
vention. If operator control fails, instrumented safeguarding such as SIS kicks in to
shut down the operation. Therefore, the setpoints and alarm limits must be properly
aligned and spaced to allow ample time for the next defense line to react.

1.4 Summary

The key messages conveyed by this chapter include the following:

1. Process control is essential for process safety and production efficiency.
2. Process control has its unique roles and responsibilities, supported by distinct
know-how and skills. As a result, it is indispensable in a modern operating plant
and is irreplaceable by other engineering specialties.
3. Process control achieves its objectives by improving control strategy and modi-
fying software implementation. Therefore it is cost-effective with low investment
and high return.
1.4 Summary 23

4. Process control is cross-disciplinary and very much application-oriented. The

collaboration and interactions with other teams, especially process engineering,
control systems, and operations, are instrumental.
5. Process control has been one of those most undervalued or misunderstood engi-
neering specialties, and there are tremendous practical benefits in raising aware-
ness of the roles and contributions of process control.

References

Åström KJ, Kumar PR (2014) Control: a perspective. Automatica 50(1):3–43

Campbell J (2004) Gas conditioning and processing—the equipment modules, vol 2, 8th edn. John
M, Campbell and Company
King M (2016) Process control: a practical approach, 2nd edn. Wiley
Lee W, Weekman VW (1976) Advanced control practice in the chemical industry. AICHE J 22(27)
Liptak BG (2006) Process control and optimization, instrument engineers’ handbook, vol II, 4th
edn. Taylor and Francis
Marlin TE (2015) Process control—designing processes and control systems for dynamic perfor-
mance, 2nd edn. McGraw-Hill
Sakislis V, Perkins JD, Pistikopoulos N (2004) Recent advances in optimization-based simultaneous
process and control design. Comput Chem Eng 28(10):2069–2086
Stephanopoulos G (1984) Chemical process control—an introduction to theory and practice.
Prentice-Hall, New York
Chapter 2
Process Control Knowledge Framework

Process control evolves from automatic control. For historical reasons, most of the
automatic control theories and techniques originated from the aerospace and elec-
trical engineering fields, of which only a (small) subset is directly applicable to the
process industry due to the unique characteristics and distinct requirements. On the
other hand, some new technologies have been developed over the years to address
the particular requirements of the process industries. The scope has become so large
that it is necessary to delineate the relevant part of the knowledge for process control
from the general scope of automatic control and complement it with the relevant
new technologies developed over the recent years, such as those related to process
engineering and process modeling. This chapter presents a knowledge framework
that summarizes the core process control technologies currently in use (or in need) in
the process industries, from simple PID controllers (Sect. 2.1), complex PID control
schemes (Sect. 2.2), to advanced process control (Sect. 2.3) solutions. The control
design philosophy (Sect. 2.4) dictates how these technologies are utilized.

2.1 Simple Control Loops

Most industrial control loops are simple PID control loops (Åström and Kumar 2014;
Lee and Weekman 1976; Shinskey 2001; Smith 2010). A simple PID control loop is
also called a standalone PID control loop. It typically consists of only one variable
to control (control target) and one variable to manipulate (control handle).

2.1.1 Open-Loop Versus Closed-Loop Control

Control can be divided into the four broad categories as shown in Table 2.1:

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 25

S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_2
26 2 Process Control Knowledge Framework

Table 2.1 Type of controls Manual operation Automated operation

Open-loop control No control Sequential control
Closed-loop control Operator control Automatic control

1. Open-loop control: A block diagram of open-loop control is given in Fig. 2.1.

In an open-loop control system, the process (➀) to be controlled is usually well
understood and well behaved with no significant uncertainties such as noises
and disturbances, or the effect of the uncertainties on the controlled parameter
is within the acceptable range. The control action (➂) is generated by a fixed
logic or sequence (➄) following a pre-defined control target (➃), “start it and
trust it”! There is no need to check whether the controlled parameter responds as
desired (➁), and thus the next control action is independent of the control result
of the current step. Some open-loop control examples include firing a bullet,
switching traffic lights, and toasting bread with a toaster. Sequential control
such as automated plant startup and shutdown, the transition from one operating
mode to another, and batch operation are primarily open-loop control with little
or no feedback.
2. Closed-loop control: Figure 2.2 is the block diagram of a typical closed-loop
control. Most industrial processes have significant noises and disturbances (➀)
with unpredictable impacts on the controlled variables. The manipulated variable
(➂) must be adjusted continuously to correct the deviations caused by noises and
disturbances and maintain the controlled variable at their target (➃). The action
is based on the feedback (➁) from the control result of the previous step.
Feedback control is one of the most commonly used mechanisms for closed-loop
control. The basic principle of feedback control is

Measure → Compare → Correct.

Both operator control and automatic control are closed-loop controls. They com-
plement each other to achieve the operating requirements. The controller (➄) in
Figs. 2.1 and 2.2 can be the operator, the control software, or both:

Fig. 2.1 Open-loop control

2.1 Simple Control Loops 27

Fig. 2.2 Closed-loop control (feedback control)

a. Operator control: Manual control by the operator is the primitive form of

closed-loop control. The operator monitors the value of the controlled vari-
able, compares it with the desired value, and adjusts the manipulated variable
(e.g., control valves) to keep the controlled variable at or around its target
value. This checking and adjusting keep on as frequently as necessary.
Driving a car is an example of manual closed-loop control. The driver (oper-
ator) periodically checks the speedometer for the driving speed, compares it
with the posted speed limit, and makes the necessary adjustment to the gas
pedal to stay below (but close to) the speed limit.
b. Automatic control: The measure-compare-correct process is highly
repetitive and tedious for an operator (e.g., the driver). For continuous opera-
tion, this type of manual operation is very prone to human errors. An automatic
control system replaces the operator to perform this repetitive work more effi-
ciently and reliably, becoming automatic closed-loop control. For example,
cruise control in a modern car provides drivers with relief from fatigue on a
long-haul trip.
Although automatic control is intended for normal operation, operator manual
control often needs to take over when the situation is too complicated for
automatic control to handle.

The variables in a control loop may have different names in a different context,
see Fig. 2.3 for some examples. These names are sometimes used interchangeably in
the book.

Fig. 2.3 Naming of variables in a control loop

28 2 Process Control Knowledge Framework

2.1.2 Feedback Control and Its Essential Components

The concept of feedback control is embedded in most control solutions, from single-
loop simple PID control to unit and plant-wide complex control solutions. A brief
history of feedback control can be found in Lewis (1992). A feedback control loop
is comprised of five essential components, as illustrated in Fig. 2.2, namely the pro-
cess to be controlled (➀), measurements and feedback mechanism (➁), final control
element (➂), the control objective (➃), and the control engine (➄):
1. Process: A process refers to the physical plant or equipment to be controlled.
The critical information required on the process is the dynamic cause and effect
relationships among all the variables and the dynamic response characteristic of
each relationship, including the noises and disturbances that affect the process
operation (see Sects. 2.1.3 and 6.4 for details).
2. The measurement, monitoring, and feedback mechanism: The measurement is
the window into the process and must provide an undistorted view of changes
in the process variables. We cannot control what we cannot measure.
A process measurement typically includes a sensor and a transmitter, with the
former providing the physical measurement and the latter transmitting the mea-
surement value to the controller as feedback (see Sect. 2.1.4). The availability
and credibility of process measurements are essential requirements for process
control. A wrong measurement is sometimes worse than no measurement due to
the false sense of security.
3. Final control elements (FCEs): The final control element is a manipulable device
that can influence the process and cause the process output to change in a pre-
dictable fashion. The most common final control element is a control valve. A
control valve is sometimes called a throttling valve to distinguish it from an
on/off valve. Other control elements such as electric current, motor speed, or
switches are also commonly encountered (see Sect. 2.1.5 for more discussions).
4. Control objectives: The control objectives serve the operating objectives and
dictate the control strategy and functionality of the control solution. For a simple
loop, the control objective is to keep the controlled variable at or around its
setpoint. In the context of control design, the control objectives determine what
variable to control and how to set the target values (see Sect. 2.1.6 and Chap. 7
for detailed discussions).
5. Control algorithm or logic: The control algorithm or logic calculates the con-
trol action to eliminate the control error (see Sect. 2.1.7). The technique most
frequently used for feedback control is the on/off control and PID control for sim-
ple processes (Chap. 3) and complex PID-loops (Chap. 4) or advanced process
control (Chap. 5) for more complex multivariable processes.
2.1 Simple Control Loops 29

2.1.3 Control Loop Component: Process Dynamics

Process dynamics is the transient behavior of a process variable responding to a

change in another variable. Response to a change is typically not instantaneous but
takes some time to re-settle. Some processes (e.g., level) may never settle, and some
processes may even diverge. In addition, there are always uncertainties in the process
value, such as the effect of measurement noises or process disturbances. For example,
a load change1 is a common disturbance that affects the process response behavior.
Figure 2.4 shows a block diagram for an abstract process. The input u(t) is typically
a control handle such as a valve; y(t) is the noise-free process output. The actual
available process output is the measured variable z(t), which includes the artifacts
of noise and disturbances.
Figure 2.5 is an example of an output response z(t) to a step-change in input u(t)
and is thus often called step response.
With a step change in input u(t), the process dynamics are about the following
transient behaviors in the process output y(t):

Fig. 2.4 A process with noise and disturbance

Fig. 2.5 Process dynamics from a step response test

1 Many times load is used interchangeably with disturbance.

30 2 Process Control Knowledge Framework

1. Gain (K p ): How much is the eventual change in output y(t)? For a stable process,
the process output will eventually settle to a new steady state. While for integrat-
ing or unstable process, the new steady state may not exist, and the definition of
gain is different.
2. Lag (τ ): How fast is the change? The process lag is usually related to the material
or energy capacitance. The different types of processes differ from each other,
mainly in the lag behavior or pattern.
3. Delay (θ ): When does the output y(t) start to change? Delay is also called time
delay or dead time and is typically caused by transport delay.
Note that the true output response y(t) is not known due to noises and disturbances.
Instead, only the noise-contaminated process value z(t) is available. The true process
dynamics u(t) → y(t) must be deduced from the apparent dynamics u(t) → z(t).
The impact of noises and disturbances is one of the main challenges facing process
modeling in model-based applications (Chap. 5).
Most industrial processes can be described with the three parameters above; how-
ever, some can have more complex dynamics and require more sophisticated mathe-
matical representations. Sections 6.6 and 9.1 provide more discussions on this topic.
A good understanding of process flow and dynamics is crucial for advanced control
applications such as model-based applications. Process analysis is the core content
of this book and the entire Chap. 6 is dedicated to the skills and methodology of
process analysis.

2.1.4 Control Loop Component: Process Measurements

Process measurement provides a window of view into the process operation and is
essential for all operations, including monitoring, control, optimization, and safe-
guarding. Process measurement typically includes a primary device such as a sensor
and a transmitter. The sensor senses the changes in the process variables, while the
transmitter converts the sensor output to an electric signal suitable for transmission to
other devices or control systems. Sometimes a secondary device is needed to further
process the data, such as a multi-phase flow meter (a flow computer) (Fig. 2.6).
The process measurement is typically an analog signal such as pressure, flow,
level, and temperature. The measurement needs to be transmitted to the control
systems for monitoring and control. Inside the modern control systems, the signal
is all digital. Signals for transmission from the process in the field to the control
system in the control room are typically in a 4–20 mA current or a 1–5 V voltage
signal, which is converted to digital signals inside the control systems. The latest
Fieldbus technology, which provides a digital, two-way, multi-drop communication
link among intelligent field devices and automation systems, allows the digital signal
to be extended to the field.
The control action is calculated based on credible and reliable process measure-
ments. In an operating plant, metering and measurement is an engineering specialty
2.1 Simple Control Loops 31

Fig. 2.6 Process

measurement components

by itself. For example, custody meters and online analyzers are big topics and require
dedicated teams to take care. However, process control practitioners must know what
is available and how to use them.
The typical process measurements and the major types of technologies for each
measurement type are summarized in Table 2.2:
1. The four conventional process variables are flow, pressure, level, and tempera-
ture. A flow can be a liquid, gas, or even solid flow (e.g., polymer granules). A
pressure can be gas pressure, steam pressure, or liquid pressure.
2. Multi-phase flow meter: In upstream oil and gas production, there is a need to
measure the flow rate at the wellhead. However, the challenge is that the flow
is multi-phase (oil, gas, water, and other impurities), and accurate phase flow
measurement is not practical with currently available technology. Much progress
is being made on this front, including multi-phase flow meters (MPFM). MPFM
is a small computer installed at the wellhead to analyze the fluid and infer the
individual phase flow rate.2
3. Quality: With more stringent requirements on operation and thanks to the tech-
nological advances, quality measurement via online analyzers is becoming more

Table 2.2 Typical process measurements

Measurement Instrument type
Flow DP-based, vortex, ultrasonic, Coriolis, electromagnetic
Level DP-based, RF induction, capacitance, radiation
Pressure Capacitance, piezoelectric, differential pressure
Temperature RTD, thermal couples, infra-red, optical
Multi-phase flow Multi-phase flow meter
Quality Online analyzer, lab sampling
Inferential property Soft sensor

2A related application is the well production measurement discussed in Sect. 12.3, where a process
control solution must be provided to work around the limitation of multi-phase flow measurement.
32 2 Process Control Knowledge Framework

widely available and affordable and has become another common process mea-
surement. For example, gas chromatographs (CG) are one of the most common
types of online analyzers.
4. Inferential properties: For advanced control applications, inferential properties
(soft sensors) are widely used for those properties that are not directly measur-
able or cannot be measured reliably or fast enough by online analyzers. Infer-
ential property is a highly viable and cost-effective way of complementing the
measurement needs for process control and real-time monitoring. More creative
and widespread use is encouraged and expected.
The fundamental requirements for a good measurement are repeatability and
accuracy. For process control, measurement repeatability is typically more important
than accuracy except for custody transfer and constrained control since feedback
control can automatically compensate for the inaccuracies as long as the measurement
is consistently repeatable.
Flow measurement is fundamental because the physical movement of material and
energy is a crucial operating requirement. The different flow meters, measurement
principles, and measurement results are summarized in Table 2.3 for a quick look-up
(Mulholland 2016).
The essential properties of a flow meter that is important to process control include
the following:
1. Service type: Certain flow meters are best for gas flow measurement, while others
are more suitable for liquid. Steam measurement poses some unique challenges
and may have different requirements on flow meters. There are also flow meters
such as DP-based that can work equally well on both gas and liquid.
2. Mass flow versus volumetric flow: A flow meter can produce either a mass flow
or volumetric flow. Certain flow meters such as DP-based can be configured to
produce either.
3. Measurement range: Measurement instruments all have an effective range. The
normal operating values should be typically within 10–80% of the measurement
range. Oversized or undersized meters may impact the measurement accuracy.

Table 2.3 Type of flow meters and their measuring principles

Flow meter type Measurement principle Indicated flow
Orifice plate Differential pressure → Raw mass flow
Orifice plate Differential pressure → Raw volume flow
Orifice plate Differential pressure → Standard volume flow
Vortex Vortex frequency → Velocity → Actual volume flow
Ultrasonic Radar signal transit-time → velocity → Actual volume flow
Ultrasonic Doppler frequency shift → velocity → Actual volume flow
Electromagnetic Induction voltage → velocity → Actual volume flow
Coriolis Motion → density → Actual mass flow
2.1 Simple Control Loops 33

4. Turndown ratio: The measurement typically loses its accuracy when the value
falls below the lower range. For a DP-based flow meter such as an orifice, the
effective range is typically 10% of the full range in DP measurement. Since the
flow is proportional to the square-root of the√pressure drop, the effective flow
measurement has a turn-ratio of about 30% ( 10 : 1).
5. Cutoff threshold: While all measurement accuracy decreases when the measure-
ment value approaches zero, certain types of flow meters, notably vortex meters,
will completely lose measurement readings when the flow rate is below a certain
threshold. This threshold is called the low cutoff flow. For critical variables, spe-
cial measures should be taken to take care of the non-availability measurement
below the cutoff limit.
6. Pressure loss: Some flow meters such as an orifice and a Venturi tube are based
on pressure drop across a measurement device. The pressure loss is usually
relatively small compared with the operating pressure. However, if the nominal
operating pressure is already low, the pressure loss across the flow meter may
become a concern.
7. Response time: All measurements have delays caused by factors such as the loca-
tion or response time of the measurement devices. For specific applications such
as compressor control (see Sect. 11.3), the response time of the measurement
becomes vital.
Certain flow meters require square-rooting on the measured value to arrive at
the correct flow, while others do not. Those requiring square-rooting are also called
the square-root types, while those that do not are called the linear types. Table 2.4
provides a summary of the typical applications of common flow meters:
1. Differential pressure-based (DP) meters are significant due to their low cost and
high reliability. However, the accuracy of DP-based measurement is related to the
operating condition. Most of the time, compensation of the measurement results
is needed to get the correct measurement. Flow compensation and conversion
are discussed in detail in Sect. 10.3.

Table 2.4 Comparison of flow measurement technologies (G—gas, L—liquid, S—steam, M—

mass flow, V—volume flow)
Service Mass or Square Turndown Pressure
Flow meter Type Volume Root? Accuracy Ratio Loss Cost
Orifice plate G/L/S M/V Y M ∼3:1 M L
Venturi tube G/L/S M/V Y L ∼3:1 L L
Ultrasonic G/L V – H >10:1 L H
Vortex G/L/S V – L ∼10:1 M M
Electromagnetic L V – H >10:1 L H
Coriolis mass G/L M – H >10:1 M H
Multi-phase MF V – L H
34 2 Process Control Knowledge Framework

2. Coriolis flow meter: A Coriolis flow meter (CM) has a high turndown ratio
and accuracy and is typically used for smaller lines where a high-pressure
drop is available. Coriolis mass flow meters can work in uni-directional and
bi-directional mass flow measurements with large viscosity and density varia-
tions.
3. Ultrasonic flow meters: Ultrasonic (US) flow meters are used for gas measure-
ment for larger line sizes. It usually has a high turndown ratio (20:1), with almost
no pressure loss.
4. Electromagnetic flow meter: Electromagnetic (EM) flow meters are typically
used for the services with a minimum conductivity above the manufacturer’s
requirement.
5. Vortex meter: Vortex meter (VM) is typically used for gas measurement. A
significant drawback is the value cutoff. When the flow is lower than a threshold,
the vortex meter will simply stop producing a value.

2.1.5 Control Loop Component: Final Control Elements

Final control elements (FCEs) can be control valves, motors, and switches, with
control valves being the most common (Fig. 2.7).
The final control element receives a control signal in the range of 4–20 mA. The
opening or closing of the control valve is usually done automatically by electrical,
hydraulic, or pneumatic actuators. Most control valves are still pneumatic, driven
by a pressure signal of 3–15 psi (0.2–1.0 bar). So the control signal has to be con-
verted from the 4–20 mA signal to 3–15 psi by a current to pressure (I/P) converter.
A positioner is usually used to control the opening or closing of the valve more
accurately.
The final control element is another overlapping area that process control needs
to know just enough about what is available and where to go for more information.
For this reason, detailed discussions on valves are not provided here. Instead, a list
of common valve types, along with their advantages and disadvantages for process
control, is provided in Table 2.5.
The most commonly used control valves are globe valves and butterfly valves.
Ball valves are typically used for on/off control or process isolation. A check valve,
also called a non-return-valve (NRV), is used to prevent reverse flow. A solenoid

Fig. 2.7 Control valves

2.1 Simple Control Loops 35

Table 2.5 Types of valves and their pros and cons for control
Valve type Throttling Isolation Tight shutoff Pressure loss
Globe valve *** * * H
Butterfly valve ** * * L
Choke valve ** *** *** H
Needle valve ** *** H
Gate valve * ** *** L
Ball valve *** *** L
Plug valve * ** L
Check valve * * L
Solenoid valve ***

valve is an electromechanically operated valve typically used for responding to a

trip signal (or loss of signal). When the solenoid valve is de-energized, it vents the
instrument air in the valve, thus causes the valve to go to the fail-safe position.
Key valve properties of concern to process control include the following:
1. Valve characteristic: Valve characteristics refer to the dynamic response behav-
ior. It is crucial to process control because the valve dynamics are an integral part
of the loop dynamics, affecting the control performance. Commonly available
characteristics include linear, equal-percentage, and fast-open. See Fig. 2.8.
The valve flow coefficient Cv is defined as the volume of water at 60 ◦ F in US
gallons that will flow through a valve per minute with a pressure drop of 1.0
psi across the valve at a fully open position. Therefore, the actual flow rate is a
function of the valve opening and the differential pressure. Valve characteris-
tics provided by vendors are the flow rate versus valve opening under constant
pressure across the valve. However, the pressure across the valve typically does

Fig. 2.8 Valve

characteristics
36 2 Process Control Knowledge Framework

not remain constant in practice but decreases as the valve is opened up. The
actual valve characteristic is thus different than in the manufacturer’s specifica-
tion. The actual behavior is called the installed characteristics as opposed to
the manufacturer’s characteristics or inherent characteristics. For example, an
equal percentage valve by specification behaves more like a linear valve when
installed. For this reason, control valves in practical applications are often of
equal percentage characteristics.
2. Valve sizing: The size of the valve determines the rangeability, which affects
the controllability of the control valve. Valve is inherently nonlinear in charac-
teristics (see Fig. 2.8), especially when approaching the two ends of the range.
The typical effective range for a valve is between 10 and 80%. The valve sizing
should match the operating range so that the valve operates at the mid-range
(more linear) during normal operation. It is a more common mistake to oversize
a control valve than undersizing it because of the false sense of additional safety.
Another consideration is valve leaking (also called passing). Due to wear and
tear over time, a valve may no longer be able to close tightly. The leaked flow may
become a problem for measurement or control. See Sect. 12.3 for an example.
An oversized valve may aggravate the leaking problem.
3. Fail-safe position: The position of a failed valve can significantly impact the
associated process, equipment, and control design. “Fail” refers to the loss of
“power,” the medium that moves the valve actuator. The most common “power”
medium is the instrument air, which is compressed air at a pressure of approx-
imately 7 to 8 bar(g). As per ISO-5208 standard , control valves may fail in
various positions, including the following:
a. Fail-open (FO): Upon loss of instrument air, the valve jumps to a fully open
position under the force of the spring.
b. Fail-closed (FC): In the case of instrument air failure, the valve quickly goes
to a fully closed position.
c. Fail-last/fail-locked (FL): If instrument air is lost, the valve position stays
where it was. A motor-operated valve (MOV) is a typical example.
d. Fail-indeterminate (FI): The fail position is uncertain or indifferent.
For a fail-close (air-to-open) valve, a control signal of 20 mA drives the valve to
fully open, while a 4 mA will send the valve to fully close. The opposite is valid
for a fail-open (air-to-close) valve. The fail-safe position is dictated by process
safety considerations and is typically identified on the piping and instrument
diagram (P&ID) (see the arrow direction in Fig. 2.7).
Since a stable feedback control loop must have an overall negative gain to reduce
the control error and the valve dynamics are part of the loop dynamics, the fail-
safe position affects the direction of control action of the controller.
The actual control valve in the field is typically an assembly consisting of a
control valve, two block valves, and a bypass valve, as shown in Fig. 2.9. The
block valve and bypass valve are all manual valves. When the control valve needs
to be taken out for service, the two block valves are closed, and the control valve
is removed. The flow is then manually adjusted by the bypass valve.
2.1 Simple Control Loops 37

Fig. 2.9 Control valve

arrangement

4. Pressure loss: A control valve is a constriction for flow and will always result
in some permanent pressure loss. In some applications, the pressure loss may
be too significant to be acceptable. The flow-balancing control in Sect. 4.1.4 is
an example of unacceptable pressure loss caused by the control valves, and a
unique flow-balancing control scheme is designed to minimize the loss.
5. Positioning accuracy: Out of the five components in a feedback control loop
(Fig. 2.2), the control valve causes the most control problems. The control valve
problems are typically associated with the valve being unable to move to the
position as requested. Common issues include stiction, slip, hysteresis, passing
(leaking), and deadband. The critical requirements on valve positioning include
the following:
a. Valve range: For an air-to-open valve, a 100% output from the controller,
converted to a 20 mA electric current, should put the valve in a fully open
position, while 0% output should produce a 4 mA current to close the valve
fully.
b. Dead angle: All air-to-open valves need to have more than 4 mA current to lift
the valve from its seat (for a globe valve) or start turning (for a butterfly valve)
from the fully closed position. However, if the valve requires a considerably
higher current than 4 mA to start moving, the control performance may be
degraded. Similarly, an air-to-close valve should not need too much less than
20 mA to cause the valve to start closing.
c. Transient response: Upon signal change, the valve should move quickly and
smoothly to its new position without much delay or overshoot.
See Sect. 8.4 for more details on valve problems and their troubleshooting.

2.1.6 Control Loop Component: Control Objectives

Control objectives serve the operating objectives. Generally speaking, from the con-
trol point of view, the control objective is to drive the control error to zero. From an
operation point of view, the control objectives are to maintain the desired operating
point at the target value during normal operation and keep the operating point within
38 2 Process Control Knowledge Framework

Fig. 2.10 Control objectives and constraints

the operating envelope during a process upset (Fig. 2.10). The control objectives
typically include one or more of the following:
1. Quality control: Ensure that production meets all the product specifications.
2. Capacity control: Maintain the dynamic supply and demand balance (in both
material and energy).
3. Protective control: Keep the operating point within the operating envelope.
4. Sequential control: Ensure quick and stable transition between different operat-
ing conditions.
The control objective dictates what process variable is to control and at what
value it should be controlled. For a simple control loop such as a flow controller,
this is usually a trivial decision. For a complex control scheme consisting of multiple
interacting control loops, however, determining the control objective for each con-
trol loop can be the most important and most challenging part of the control design.
It requires a good understanding of the process design and operating requirement
and a holistic view of all the control loops. Unfortunately, the importance of control
objective analysis is often underestimated when multi-input multi-output (MIMO)
variables are involved with multiple levels of control. Many controllability and oper-
ability issues are related to poor objective setting, which results in a flawed control
strategy.
2.1 Simple Control Loops 39

For example, it is unrealistic to expect the home A/C system to maintain a tem-
perature below 10 ◦ C on a hot summer day even if the targeted temperature at the
thermostat is set to 0 ◦ C. Another typical example is surge tank level control. The
purpose of the surge volume in the tank is to absorb the flow fluctuations. The level
should be allowed to fluctuate within a range. Keeping a tight control on the level
simply passes the fluctuations downstream and defines the design purpose of the
surge tank. A more subtle example is given below.

Example 2.1 “Goal-zero” vision. With the increased awareness of the health safety
and environmental (HSE) impact, many companies run HSE “goal-zero” campaigns
on HSE incidents.
Is “goal zero” achievable? The answer is no, at least not economically! From a
feedback control perspective, once goal zero is reached, the measurement is lost since
the number of incidents cannot be negative. See illustration in Fig. 2.11. There is no
more indication of the difference (the control error) between the actual number of
incidents (measurement) and the targeted number of incidents (setpoint) since both
are zeros now. We do not know how safe we are from the next incident until the
next incident actually occurs. However, it is precisely the next incident that we try
to prevent. A catch-22 situation!
The same or more effort needs to be continuously invested in maintaining the zero
incident status, but with no confidence in the result. Therefore, “goal zero” is a good
vision but not a realistic control objective or setpoint.
A near miss is a precursor to an incident. Many companies implement the number
of near misses as another performance indicator with the belief that the number of
near misses can predict incidents. Reducing the number of near misses can effectively

Fig. 2.11 “Goal Zero” for HSE incidents

40 2 Process Control Knowledge Framework

reduce the probability of incidents. Since the number of near misses is a non-zero
target, it can serve as a viable control objective.
Chapter 7 provides in-depth discussions on setting the control objectives and achiev-
ing the objectives with the proper control strategy.

2.1.7 Control Loop Component: Control Algorithms

At the center of the control loop is the controller. The controller compares the process
value with the setpoint and calculates the next control move.
The controller can be a simple standalone PID controller with one control han-
dle and one control target or a complex control scheme consisting of multiple PID
controllers, supporting function blocks, and other elements. The selection of control
algorithms depends on many factors, and the critical considerations include perfor-
mance, reliability, and the life cycle ownership cost:
1. On/off control: On/off control is the simplest controller and thus is also the least
expensive in life cycle ownership. The manipulated variable can assume just two
values, e.g., open/close for a valve, on/off for a switch, or run/stop for a pump. For
this reason, on/off control is sometimes called snap-acting or switching control.
In some level control applications, on/off control is also called dump control.
Figure 2.12a shows an example of a simple on/off switching control, represented
by LS, where S stands for switching, with an on/off valve LSV-101.
On/off control is a very reliable and cost-effective option for control problems
where sloppy control performance is acceptable, such as with large capacity
temperature or level, large storage tanks, room heating, and hot-water supply
tanks. A simple switching logic is needed to open and close the valve.
A crucial consideration for the on/off switching logic is that a gap (or dead-
band) must be provided between opening and closing to avoid excessive chatter-
ing/hunting. Some basic understanding of the process dynamics (delay and lag
on overshoot/undershoot) is needed to set an adequate deadband. See Fig. 2.13
for an illustration of the deadband.
A typical application of on/off control is the level control of a large vessel
(Fig. 2.12a). When the level rises above a high limit, the on/off controller opens

Fig. 2.12 On/off control and PID control

2.1 Simple Control Loops 41

Fig. 2.13 Deadband in on/off switching control

the valve to “dump” the liquid. Conversely, when the level falls below a specific
low limit, the valve is closed, and the level will start to rebuild. A ball valve is
typically used for on/off control.
Home air-conditioning control is another example where on/off control is used
for simplicity and reliability. In some utility plants, the supply of nitrogen or
instrument air is controlled by simply starting and stopping the compressor.
The major downside of an on/off control scheme is that it cannot provide the
required granularity in control moves to maintain a tight setpoint.
2. PID control: The majority of industrial process control is by PID control loops.
A PID controller offers three types of control actions, as shown in Fig. 2.14:
a. Proportional action is a linear function of the control error and provides a
quick initial response to a change: For example, for temperature control, the
control action for a 20 ◦ C error is twice that for a 10 ◦ C error. However, since
proportional action is proportional to the control error, it requires a non-zero
control error to sustain the control action. This non-zero control error results in
a steady-state offset, which is a significant drawback of proportional control.
b. Integral action eliminates steady-state errors: The control action is propor-
tional to the time integral of the control error. For example, for the temperature
control, as long as there is a difference between the desired temperature and

Fig. 2.14 Structure of a PID control loop

42 2 Process Control Knowledge Framework

Table 2.6 Comparison of on/off control and PID control

Control algorithm Key advantages Key disadvantages
On/off control Simple, reliable, inexpensive Inaccurate, constant cycling
Proportional control Simple, fast Steady-state offset
Integral control Eliminates offset Extra time lag
Derivative control More responsive Sensitive to noise

actual temperature, the integral action will continue to change the heat input
in the same direction.
c. Derivative action provides anticipatory action for better disturbance rejection
by sensing the rate of change in the controlled variable. The control action
is proportional to the rate of change of the control error. However, noises
and disturbances are prevalent in the process variables. The derivative action
amplifies noises and may result in too aggressive control actions. For this rea-
son, most controllers in the process industry are P or PI controllers. D action
is rarely used.

A PID controller LC-101 (Fig. 2.12b) can replace the on/off switching control
LS-101 (Fig. 2.12a) for better control performance. Table 2.6 lists the pros and cons
of on/off control and PID control schemes. More detailed discussions on PID control
are provided in Chap. 3.

2.2 Complex Control Loops

The simple control loops in Sect. 2.1 are comprised of one controlled variable and
one manipulated variable. Although simple controllers are dominant in practice,
many applications involve multiple controlled variables and manipulated variables
and have to be considered together due to the interactions among the variables.
One approach is to have multiple simple controllers work together to address the
complexities. Table 2.7 provides a list of process control loops by the input/output
structure, with detailed explanations in the following subsections, as indicated by the
section numbers in the table.
There are three types of process control structures prevalent in practice: simple
PID control, complex PID control, and model-based control.
1. Simple control: Single-input and single-output (SISO), with one simple con-
troller. These are the simple control loops discussed in Sect. 2.1.
2. Complex control: Multi-input and multi-output (MIMO), with multiple simple
controllers. Complex control loops are built upon a group of simple control
loops.
2.2 Complex Control Loops 43

Table 2.7 Common complex control loops

Single output Multiple outputs
Single input On/off control (Sect. 2.1.7) Cascade control (Sect. 2.2.4)
Simple PID control (Sect. 2.1.7) Ratio control (Sect. 2.2.6)
Protective control (Sect. 4.1.3) Override control (Sect. 2.2.7)
Multiple inputs Split-range control (Sect. 2.2.1) Selective control (Sect. 2.2.8)
Dual controller control (Sect. 2.2.2) Decoupling control (Sect. 4.1.2)
Fan-out control (Sect. 2.2.3) Model predictive control (Sect. 2.3.4)
Feedforward control (Sect. 2.2.5) Advanced process control (Sect. 5.2)

3. Model-based control: Multi-input and multi-output, with one complex controller

performing sophisticated calculation and control logic. Model predictive control
is a successful example of model-based control.
PID-based control is sometimes called baselayer control as they are implemented
in the baselayer DCS/PLC platform; model predictive control is called advanced
process control and is typically implemented in dedicated software and hardware
separate from DCS/PLC. Adding to the confusion, complex PID control and model
predictive control are sometimes collectively called advanced process control as
opposed to simple PID control.
This section briefly introduces the common complex PID control schemes in
widespread use and explains what they are and when/where they should be used.
Discussions of advanced features and practical considerations associated with each
complex loop are deferred to Chap. 4.

2.2.1 Split-Range Control

Split-range control is a popular complex control scheme for the following application
scenarios:
1. Extending the control range by manipulating more than one control handles.
2. Providing preferential control if there is a difference in the priority of meeting
the control targets.
Figure 2.15 is an example of split-range control, where the process flow valve
PCV-101A is used to maintain the vessel pressure during normal operation. In case
the pressure goes abnormally high, the flare valve PCV-101B will start to open.
With split-range controller PC-101, the controller output between 0 and 70% will
cause pressure valve PCV-101A to open from 0 to 100%. When the controller output
increases to more than 70%, the main control valve PCV-101A will remain at 100%,
and the flare valve PCV-101B will open from 0 to 100%, corresponding to 70–100%
44 2 Process Control Knowledge Framework

Fig. 2.15 A split-range control loop

of the controller output. The value at which the signal is split is called the split point,
which is 70% in this example.
This split-range control arrangement provides both preferential control and
extended range. During normal operation, the preferred flow direction is through
PCV-101A to downstream. In case of excessive gas causing high back pressure,
the capacity of the main valve PCV-101A is not sufficient, and thus the control is
extended to the flare valve PCV-101B for additional capacity.
Some essential considerations or implementation challenges of split-range control
include the following:
1. Proper configuration of the split point: Many split-range control applications
use the default 50% split point for convenience or out of negligence. The default
50% is rarely the optimal split point. The correct split point should be calculated
based on the dynamics of the two ranges. In the above example, the split point
is 70%, and the two split ranges are 0–70 and 70–100% since the dynamics for
flaring are much faster.
2. Initialization during mode change: When the controller has switched from auto-
matic mode to manual mode and later back to automatic mode again, the con-
troller output may no longer be in the same range. Proper re-initialization of the
controller is required; otherwise, the transition may potentially be bumpy.
See Sects. 4.1.5 and 4.3.4 for discussions on how to address the challenges.
2.2 Complex Control Loops 45

2.2.2 Dual-Controller Control

Split-range control is widely used in practice; however, it is often misused or abused

due to the negligence of the unique challenges associated with it. In many applica-
tions, dual-controller control is a cleaner and more reliable scheme than split-range
control.
With dual-controller control, the split-range controller PC-101 is replaced with
two independent controllers (PC-101A and PC-101B, respectively). They share
the same measurement, but each has its own (but different) setpoint and manipulates
a separate valve (see Fig. 2.16).
Compared with split-range control, the configuration, tuning, and troubleshoot-
ing are overall much simpler. The only downside is that the two controllers must
have sufficiently different setpoints to meet the degree-of-freedom requirement (see
Sect. 4.1.2 for more discussions).

2.2.3 Fan-Out Control

Fan-out control is a special case of split-range control where the same output is sent
to two or more control elements simultaneously. That is, the manipulated variables
are “linked in parallel”. Figure 2.17 is an example of a fan-out control, where the
controller output is “fanned out” to more than two valves. This configuration is
common in upstream oil and gas operations, where it is not unusual to see a flow
controller output fanned out to dozens of injection wells.

Fig. 2.16 Dual-pressure control in place of split-range control

46 2 Process Control Knowledge Framework

Fig. 2.17 A fan-out control loop

Although fan-out control is straightforward in concept, there are some practical

challenges with the implementation and operation:
1. Controller tuning: The number of manipulated variables that the controller output
is fanned out to frequently change in response to operating needs. Consequently,
the same tuning may not be adequate for all the scenarios. Some kind of gain-
scheduling mechanism may be needed.
2. Back initialization and bumpless transfer: When some manipulated variables are
taken offline and put back online later, proper initialization is essential but may
not be automatically guaranteed.
Section 4.3.3 provides discussions on how to handle these challenges.

2.2.4 Cascade Control

Cascade control is probably the most popular complex PID control loop (Marlin
2015). It is widely used to improve control performance by reducing the impact of
nonlinearity, stiction, or hysteresis in the final control element.
A cascade control loop has one manipulated variable (handles) and two con-
trolled variables (targets). Figure 2.18 shows a simple level control loop LC-101
that directly manipulates the flow control valve FCV-101. However, due to the slow
and complex dynamics of the pump, the control action with the valve is not fast
enough to maintain a smooth level. Fluctuations in the flow affect both the flow and
level at the same time.
For improved performance, an intermediate controller FC-101 is added to the
control loop to “reject” the fluctuations locally and minimize their impact on the
level. The two PID controllers are linked in series, with one controlling the setpoint
2.2 Complex Control Loops 47

Fig. 2.18 A cascade control loop

of the other. The former is called the master controller, the primary controller, or
the outer control loop, while the latter is called the slave controller, the secondary
controller, or the inner control loop.
A valve positioner is an example of cascade control. A valve positioner is a simple
feedback control loop (P-only controller) to ensure that the actual valve position
tracks the desired setpoint. The position setpoint is received from the upstream control
actions, and the valve positioner serves as the secondary control loop. The valve
positioner suppresses the uncertainties such as stiction or hysteresis imperfection
and “hides” them from the PID controller, and ensures that the actual valve position
is what the controller requests.
Cascade control is very popular in practice. However, some requirements must
be met, and some practical challenges must be addressed to use cascade control
effectively:
1. There must be a definite cause and effect relationship from the final control
element to the secondary variable and from the secondary controlled variable to
the primary controlled variable. In this example, the cause and effect relationships
between the valve, flow, and level are self-evident.
2. Cascade control only adds value to the overall control performance if the sec-
ondary variable responds to the disturbance sooner and quicker than the primary
variable. The secondary loop should typically be five times faster than the pri-
mary loop.
3. Tuning: The tuning of cascade control loops should start from the secondary
loop with the primary loop in manual mode. When tuning the primary loop, the
secondary loop should be in cascade mode.
To cascade is to delegate. Cascade control is like running a business between
headquarter and branch offices. The local branch office is charged with making timely
decisions on local matters without escalating to the main office. The headquarter
only sets the long-term goals and address high-level issues. See Sect. 4.1 for more
discussions on the application scenarios and practical considerations.
48 2 Process Control Knowledge Framework

2.2.5 Feedforward Control

In Fig. 2.18, it is seen that the disturbance to the flow is “suppressed” by the flow
controller before it affects the outer level loop. However, if the disturbance is in the
feed flow to the tank, it will directly affect the level. The flow controller will react
only after the level is already disturbed.
Suppose the dynamics from the disturbance to the controlled variable is known.
In that case, the control scheme can try to “intercept” the disturbance by adjusting the
manipulated value with the correct magnitude and at the right time, in anticipation of
“canceling out” the effect of the disturbance before it reaches the controlled variable.
This “intercept” and “cancel” approach is the principle of feedforward control.
Figure 2.19 shows a boiler heater process where the fuel oil is adjusted following
the changes in feedwater flow to maintain the desired temperature (or steam quality)
in the outlet flow. Here the feedwater flow rate FI-101 is the primary flow, and
its fluctuations directly affect the temperature TC-102. Due to the pure feedback
nature, the cascade feedback control loop TC-102/FC-102/FCV-102 does not
take any corrective action before the controlled variable TC-102 is already affected.
The feedwater flow FI-101 can be added as a feedforward variable to the existing
temperature feedback control loop for improved control performance. The water
flow rate is measurable, and the dynamics from the flow FI-101 to the temperature
TC-102 are easily known and relatively stable.
Feedforward control is an open-loop control loop added to a feedback control
loop. The feedforward control action is calculated based on a good understanding
of the dynamics between the disturbance and controlled variable and must meet the
following conditions to function correctly:

Fig. 2.19 A feedforward control loop

2.2 Complex Control Loops 49

1. The dynamic relationship between the feedforward and controlled variable

(mainly the gain and delay) must be known and remain relatively constant over
time.
2. The time to reach the controlled variable TC-102 from the feedforward variable
FI-101 must be longer than that from the manipulated variable FC-102 for
best performance (“complete cancellation”).
The feedforward compensation to feedback control can be either dynamic or
static. A static compensation only requires the gain compensation for the dynamics,
while a proper dynamic compensation requires knowledge on the full dynamics,
including gain, delay, and lag. Section 4.1.2 provides discussions on some practical
considerations for the implementation and operation of feedforward control.

2.2.6 Ratio Control

Ratio control is a special case of feedforward control. A ratio control scheme has two
inputs, the primary flow and the secondary flow. The primary flow is also called the
wild stream. The ratio controller manipulates the secondary flow stream to achieve a
specified ratio with a primary flow stream. The ratio controller’s setpoint is thus the
ratio between the secondary flow and primary flow:

[secondary flow rate]

ratio = .
[primary flow rate]

This output of the ratio controller is sent down to a PID flow controller to manipulate
the secondary flow.
Figure 2.20 shows an example of a mixer with a primary stream FI-101 and
a secondary stream FC-102. The control objective is to maintain the desired ratio
between the primary and secondary flows to meet the quality requirement AI-101
on the mixed flow.

Fig. 2.20 A ratio control loop

50 2 Process Control Knowledge Framework

Ratio control is similar to cascade control, except that the ratio control loop does
not need to be five times slower than the flow control loop. Besides, the ratio typically
is not the ultimate goal of control. In this example, the quality variable after the mixer
AI-101 is the true objective.
A special consideration for ratio control is that ratio calculation is subject to noise
and failures, explicit division calculation should be avoided, and the DCS built-in
ratio control blocks should be used instead.

2.2.7 Override Control

Override control is used to maintain one controlled variable on target without

violating constraints on other controlled variables. In an override control, there are
two or more controlled variables (targets) sharing one manipulated variable (handle),
typically through a high selector (for high override) or low selector (for low override).
At any time, only one controller is active, i.e., only one output is selected by the
selector and sent down to the next level of control elements such as a valve. The
control action requested by the most demanding controller is selected and accepted.
An example in our daily life is the so-called adaptive cruise control in newer
cars. The regular cruise control maintains a speed target by manipulating the fuel.
Adaptive cruise adds another control target to the requirement: the distance with the
car in front. When following too close, the adaptive cruise control overrides the speed
control and automatically slows down to ensure a safe distance.
An override control can be an overriding protective control such as the pressure
controller PC-103 in Fig. 2.21, where the primary controller LC-101 is the regu-
latory control during normal operations, and the protective controller PC-103 only
kicks in during abnormal operating condition that cause the pressure to go higher
than its limits.

Fig. 2.21 A protective control

2.2 Complex Control Loops 51

An override control can also be an overriding selective control where all the
controllers have an equal chance to be selected by the selector. There is no distinction
between normal operating conditions and abnormal conditions in their roles. For
example, two tanks share one pump for level control. A selector can be used to select
which level controllers should be in control. Whether a high selector or low selector
should be used depends on whether the two tanks’ maximum or minimum level is
maintained.
The principle stays the same for both protective and selective override control:
each controller in the override control scheme is an independent PID control loop.
The output of the most demanding controller is automatically selected as the control
action, while the outputs of the other controllers are simply ignored.
There are more control targets than control handles. How to handle the inactive
controllers whose outputs are ignored is the challenge for override control. For those
controllers, which are effectively open-loop, the integrating action of the PID con-
troller may drive the output to saturation. When the controller is required to become
active, the output needs time to come out of saturation before being selected. This
delay is typically unacceptable for a protective control.
This saturated condition is called reset windup or integral windup. Prevention of
reset windup is crucial when implementing override control. Section 4.3.3 provides
detailed discussions on the prevention of reset windup.

2.2.8 Selective Control

When multiple handles are available to control multiple targets, and at any given
time, only one pair of target and handle needs to be connected, then a switch, either
manual or automatic, can be used.
The practical consideration for selective control is protecting those control han-
dles or control targets that are not currently selected. In the case of multiple targets,
it is similar to the scenario of overriding control (Sect. 2.2.7) and requires anti-reset
windup protection, while in the case of multiple control handles, the same bumpless
transfer behavior as in fan-out control (Sect. 2.2.3) needs to be considered.

2.3 Advanced Process Control

PID is powerful but has its limitations! Advanced process control (APC) technologies
are developed to overcome the limitations of PID control. Model predictive control
(MPC), adaptive control, and nonlinear control are examples of advanced process
control.
52 2 Process Control Knowledge Framework

2.3.1 Limitations of PID Control

While PID controllers can produce an acceptable performance for most applications,
they can perform poorly in some. The fundamental limitation of PID control is that
it is based on a feedback mechanism with a simple structure. Besides, the tuning
parameters are not based on direct knowledge of the underlying process. They remain
constant even when process dynamics have changed. As a result, many industrial
processes are beyond the capability of PID controls due to the following challenges:
1. Multiple inputs and multiple outputs with strong interactions.
2. Long time delay and very slow dynamics.
3. Nonlinear time-varying dynamics.
4. Design limits and operating constraints.
Advanced process control solutions are developed to overcome the limitations.
It is built on top of baselayer controls, deals with a more extensive scope of prob-
lems, and aims at higher level performance targets such as quality, efficiency, and
even profitability. The relationship between baselayer control and advanced process
control is illustrated in Fig. 2.22.

Fig. 2.22 General architecture of advanced process control

2.3 Advanced Process Control 53

2.3.2 Process Models: Abstraction and Generalization

There are countless industrial processes of various types, and each requires dedicated
expertise. It is unrealistic for a process control person to have intimate knowledge
of all processes. Instead, process control relies on a generalization and abstraction
of the processes to do control. This generalization and abstraction are via process
models.
A process model is a mathematical representation of the external cause and effect
relationship between the process inputs and outputs. The process model is derived
from process input/output data, which are typically obtained through deliberately
disturbing the process variable to reveal the causal relationship. This activity is called
plant test, step test, or bump test.
Model is at the center of most advanced process applications. Once the process
model is available, it can be used for many purposes, including simulation, control,
and fault detection. See Fig. 2.23 for an illustration:
1. Simulation (Fig. 2.23b): With a known process model, we can observe the process
output response to the specified changes in the process input.
2. Control (Fig. 2.23c): With the process model known, we can back-calculate the
process inputs to achieve the desired process outputs.
3. Fault detection (Fig. 2.23d): With the process model known, and with both pro-
cess input and output data available, trying to determine if the process has devi-
ated from what it was when the model was built, i.e., to detect if a model/plant
mismatch has developed.
There are many types of process models, from heuristic, empirical, to first-
principle-based. Advanced process control applications are all based on a partic-
ular type of process model. The more advanced the control is, the more reliance is
on process understanding and process models. The technique for obtaining process

Fig. 2.23 Process model and its applications

54 2 Process Control Knowledge Framework

models from input/output data is called process identification and is a core com-
petence for advanced process control (APC) practitioners. See Sect. 5.2.1 for more
detailed discussions on process models and process modeling.

2.3.3 Inferential Property (Soft Sensor)

Reliable and accurate measurement is a prerequisite for control. As process control

moves from controlling a specific process variable to controlling product quality,
online analyzers and lab sampling become increasingly important. However, despite
rapid technological advances, most online quality analyzers are still not sufficiently
reliable, fast, or accurate for closed-loop control.
Inferential property estimator, commonly known as a soft sensor, is a predictive
analytic that uses faster and more reliable process measurements to infer or predict
product quality. Online analyzer measurement or lab result provides the feedback to
correct the prediction periodically.
Figure 2.24 shows an example of a crude oil stabilizer, where the true vapor
pressure (TVP) is measured downstream of the stabilizer. The TVP value is a quality
target for control. However, the TVP online analyzer is neither reliable nor fast
enough for a PID controller to operate in closed-loop confidently.
An inferential property estimator (IPE) is built based on a process model between
the pressure/temperature in the stabilizer and the TVP value. This model runs online
to predict the TVP value using the fast and reliable pressure and temperature mea-
surements. A PID controller can then be used to control the predicted TVP value
instead of the unreliable measurement from the analyzer.

Fig. 2.24 An example of inferential property application

2.3 Advanced Process Control 55

The predicted TVP value is periodically corrected with feedback from the online
analyzer AI-101 when it deems that the analyzer reading can be trusted. If the
analyzer is unavailable (e.g., offline for calibration) or is believed to be untrustable,
then the TVP controller AC-101 can continue to run on the predicted value (within
a limited time).
There are many practical challenges for successful IPE implementations, such as
the following:
1. Availability of accurate predictive models, including both the model structure
and model parameters.
2. Quality assessment of the online analyzers for deciding if the analyzer value can
be trusted to correct the IPE value.
3. Timestamp matching between the predicted values and the analyzer values.
See Sect. 5.3 for more detailed discussions on inferential property estimators.

2.3.4 Model Predictive Control

Model predictive control (MPC) is an advanced process control paradigm that has
gained widespread applications in the process industry. Figure 2.25 illustrates the
control architecture. It is evident from a comparison with Fig. 2.2 that MPC is a
natural extension of the basic feedback control.
At the center of the MPC technology is the process model. The process model
captures the multivariable interactions in the process dynamics and provides the
convenience for more sophisticated calculations such as constraint handling and
optimization. Compared to PID-based baselayer control, MPC offers the following
capabilities beyond the reach of a PID controller:
1. Multivariable: Multiple controlled variables (CV) are controlled with multiple
manipulated variables (MV). The cause and effect relationship and the dynamic
response characteristics are elegantly captured in the multivariable process mod-
els.

Fig. 2.25 Model predictive control

56 2 Process Control Knowledge Framework

2. Predictive: The process model provides the capability of predicting the process
outputs into the future. Therefore, MPC controls the process based not only on
where the unit is currently running but also on where it is predicted to run in the
next few minutes or even hours.
3. Constrained: MPC monitors and honors the engineering limits and operating
constraints when computing the control actions. The sophisticated MPC calcu-
lation allows constrained optimization to produce the next moves that will not
drive the operating points outside the various constraints.
4. Optimizing: MPC pushes the operating points toward the most profitable oper-
ating areas, typically near the constraints.
The primary challenge for MPC is in obtaining high-quality process models.

2.3.5 Self-Tuning PID Control and Adaptive Control

Another important but very challenging direction for process control is adaptive
control. Due to the nonlinear and time-varying nature of the process dynamics, the
performance of a controller with a fixed structure and constant parameters will typ-
ically deteriorate over time. Adaptive control modifies the controller (structure and
parameter) online following the changes in process dynamics or operating condi-
tions. Figure 2.26 illustrates the concept and the adaptation mechanism for general
adaptive control.
A fully adaptive MPC is highly desirable, but the current technology is still not
sufficiently robust to handle abnormal conditions. The challenge is in the credibility
of the process model identified online, dictated by the data quality.
A small step in the direction of adaptive control is self-tuning PID, which performs
automated PID tuning by introducing step changes, model identification, tuning
improvement, all online. However, because of the same concern on data quality,
self-tuning is yet to gain more trust and acceptance by field engineers.

Fig. 2.26 Generic architecture of adaptive control

2.3 Advanced Process Control 57

2.3.6 Control Performance Monitoring

Performance monitoring can involve all levels and aspects of the operation and pro-
duction, including data quality, analyzer health, control loop performance, equipment
status, unit-level summary, and plant-level overview. Monitoring the performance of
the control loops is an essential component of performance monitoring.
Control performance monitoring relies on well-defined and well-understood KPIs.
Big data and predictive analytic are playing an active role in performance monitoring
in recent years. Artificial intelligence (AI) is a hot topic for achieving operational
excellence. However, machine intelligence has not (yet) reached the level of human
intelligence. The most effective monitoring and control are still based on sound
engineering insight into the physical working of the process. Process models, from
mathematical to first-principle, are still preferred whenever available. Reliability is
the biggest hurdle from passive monitoring to real-time closed-loop control.

2.3.7 Real-Time Optimization

Real-time optimization (RTO) is an extension of closed-loop process control. RTO

optimizes the process performance on a larger scale (Fig. 1.10). Unlike traditional
process controllers or model predictive control, the scope of RTO typically covers an
entire unit or even a plant. RTO is built upon large-scale steady-state first-principle
models and relies on a sophisticated optimization engine to optimize the process
performance, usually measured in terms of profitability, by taking into account the
real-time operating data, market data, inventory data, operating constraints, etc. The
results of the optimized operation condition are sent down to model predictive control
(MPC) or baselayer control as control setpoint or operating limits.
The fundamental nature of the models allows physically meaningful parameters
to be adjusted to reflect changes in plant equipment performance, such as catalyst
deactivation or heat exchanger fouling, or compressor blade erosion. The model
parameters are adjusted to make the calculated values match the actual measurements
(data reconciliation) before each optimization calculation. RTO continuously checks
and validates the measurement data to ensure that the process is at steady state before
using the measurements to fit the parameters.
The main challenge of RTO is in the quality and quantity of the equations. A
modestly complex process unit may require hundreds or even thousands of equations
and is usually a very time-consuming activity to build. Besides, the solving of the
equations is computationally intensive for online minute-to-minute execution.
58 2 Process Control Knowledge Framework

2.3.8 Dynamic Simulation and Operator Training Simulator

Dynamic simulation and operator training simulator (OTS) system are valuable tools
for many purposes:
1. Operator training for plant startup: An OTS system has been most successful
in training operators to start a new plant well before the plant is ready to start.
Experience shows that a well-utilized OTS can reduce the startup time by more
than one-third.
2. Operator training for knowledge retainment: With the successful implementation
of MPC, the operators gradually lose their operating skills, especially in handling
plant emergencies. So far, OTS is the only viable training method to help the
operators retain and refresh their operating skills.
3. MPC modeling with dynamic simulation data: Developing MPC models with
dynamic simulation or OTS data even before a new plant is built is a promising
and cost-efficient way of process modeling. The high-fidelity dynamic models
built for dynamic simulation have sufficient accuracy to generate the input/output
data required by MPC modeling.
4. “What if” scenarios analysis: Dynamic simulation during the design phase of the
plant can often identify many operability issues. The author had the experience
in a recent large project where a dynamic simulation study before the start of
detailed design revealed that the designed operating points are too close to the
pressure trip limits and affect the operating capacity. A large part of the process
design was subsequently modified to correct the problem in time.
Process control is usually the advocate for OTS and typically leads the effort in the
initial system development. The main problem, however, is with the ownership of the
system after delivery. Dynamic simulation and OTS need model updates following
process changes. Process control rarely has the staffing and budget to update the
models. It is more appropriate for the ultimate user, operations, to take over and
assume the responsibility, but they rarely do. Most OTS systems end up losing sync
with the actual plants and eventually are abandoned.

2.3.9 Alarm Rationalization

Modern control systems have made alarm configuration so easy that alarm flood-
ing has become a common problem across the industry. A large percentage of the
alarms are nuisance alarms that distract the operators instead of helping them. Alarm
rationalization and management (ARM) are becoming mandatory in many operating
plants. Industrial standards such as EEMUA-191 and ANSI/ISA88 are in place to
guide the alarm management effort.
After a new plant is built, many alarm configurations need to be reviewed, adjusted,
and a large part of them need to be disabled or removed based on the initial operation
2.3 Advanced Process Control 59

experience. Process control engineers are typically invited to be part of the alarm
rationalization effort.
Alarm rationalization is typically based on a careful evaluation of the HSE impact
and urgency on each alarm. The HSE impact is measured by the potential loss in
terms of health, safety, or environment impact if the alarm event was not addressed.
The urgency is the estimated response time for the operator to address the alarm
event. Based on the HSE impact and urgency, the alarms are classified into different
types and priorities and assigned appropriate limits.
The alarm configuration is dependent on the operating mode. The alarm limits are
typically different for startup mode, normal operating mode, scheduled shutdown
mode, emergency shutdown mode, and turndown mode. In a sophisticated alarm
management system, the new alarm limits and priority can be automatically down-
loaded to the control systems during mode change. For example, during startup and
shutdown, the flow rate and pressure can go down to zero, and thus many low flow
and pressure alarms may go off. For a particular mode of operation, many alarms
may have to be disabled to avoid nuisance alarms. This mode-based alarm suppres-
sion is called dynamic alarm suppression. In addition, many process variables have
a cause and effect relationship between them. An alarm on one variable may trigger
the alarm on another. In this case, only the alarm on the first variable is of concern.
Alarm configuration is included in the control design as a critical step. When
designing a complex PID control scheme, it is crucial to validate the control setpoints
against alarm limits, protective control limits, and SIS trip limits to ensure sufficient
spacing (thus response time) between the limits (Fig. 1.14).

2.4 Process Control Design

Process control is required during the design of a new plant or troubleshooting of

an existing plant. Process control design requires solid process control knowledge
and a high level of problem-solving skills, which can be possessed only by senior
process control staff.

2.4.1 Process Control Philosophy

Process design dictates what kind of process control strategy is needed, and the
process control design validates the process design for feasibility and operability.
This interaction ensures that process design has process controllability in mind in
order to achieve high operability, while process control design truly understands the
intent of process design to address the operating needs effectively. Below are a few
questions at the philosophy level that need to be answered early in a process design:
60 2 Process Control Knowledge Framework

1. The big picture versus the details: It is critically important to start with a “full-
view” picture in mind when designing new (or troubleshooting existing) control
solutions. The interaction and inter-operation between processing units and even
plants dictate the high-level control objectives and overall control strategy, which
in turn determine how the next level of details should be laid out. A top-down
approach for process control design is recommended following the different
phases of project execution.
2. Level of automation: The degree of sophistication of the control solution can vary
drastically from totally manual control to fully automated control. The level of
automation to achieve is always a trade-off between desire and practicality. It is a
compromise of many factors, including HSE impact and economics. In general,
the same ALARP (as low as reasonably practical) principle (NOPSEMA 2015)
should be followed.
The car driving is a good example of this compromise between desire and prac-
ticality and the choice is a compromise of many factors including financial
situation and risk tolerance:
a. Normal driving: Manual closed-loop control. The driver
• measure speed and direction;
• compare with posted speed and lane position;
• correct with fuel, steering wheel, and brake.
b. Cruise control: Automatic closed-loop control on speed. The cruise control
system
• measure speed;
• compare with the posted speed;
• correct with fuel.
The driver takes care of direction and brake.
c. Adaptive cruise control: Automatic closed-loop control on speed and brake.
The cruise control system
• measure speed and following distance;
• compare with the posted speed and the pre-set following distance;
• correct with fuel and brake.
The driver takes care of direction only.
d. Self-driving car: Fully automated closed-loop control. The self-driving sys-
tem
• measure speed and following distance;
• compare with the posted speed, the pre-set following distance, and lane
position;
• correct with fuel, brake, and steering wheel.
The driver is hands-free.
2.4 Process Control Design 61

3. Simplicity versus sophistication: The simplest is often the most reliable. For
process control, reliability is often more important than sophistication. The most
advanced may not necessarily be the most appropriate! Most of the time, the best
solution is the “fit-for-purpose” solution, which has the best value on return in
terms of “total cost of ownership.”
Various techniques may be used to improve the control of a process. However,
as the complexity of the control system increases, so does the cost for operator
training and maintenance. The reliability of a complex control system also tends
to decrease. It is common to see sophisticated control loops taken offline by
operators because they are too “sophisticated!” Therefore, it is compulsory to
strike the right balance between manual and automated process control and
between the complexity of the control system and the benefits provided.
4. First-principle versus feedback: For process control design, the governing prin-
ciple is the material and energy balance and process dynamics; therefore, it
is imperative to always go with the physical laws rather than fighting them.
However, process engineers rely heavily on first-principle-based calculations to
enforce the material and energy balance, which a reliable and straightforward
feedback control can readily achieve. In a project team comprising both process
engineers and process control engineers, this difference in thinking is often a
clash point in design philosophy.
5. Role of the operators: The level of operator intervention is another crucial con-
sideration in process control design. Some sites rely heavily on panel operators
to make most operation decisions, while others pose many restrictions on what
the operator can do. For example, some operations may forbid the operator to
make setpoint changes to the protective controllers, while others may allow the
operator even to change PID tuning parameters at will.
This operation decision can have a significant impact on the process control
strategy and performance. For instance, a protective controller is often locked
by the operator, and no changes are allowed even to the controller setpoint and
mode. However, a bad value in the measurement may shed the controller to
manual mode. Since the controller has been locked from the operator, including
mode change, back-end logic needs to be added to unlock the controller to allow
the operator to switch the controller back to normal mode (but not any other
mode).
6. Maintainability and sustainability: The control performance will inevitably dete-
riorate over time. A poorly performed control scheme not only reduces produc-
tion profits but may also become a liability. Sustainable performance and easy
maintainability should be critical considerations in process control design.

2.4.2 Process Control Life Cycle

Many operating challenges originate from poor overall control design or incorrect
implementation. Flaws in the overall control strategy are very challenging to diagnose
62 2 Process Control Knowledge Framework

Table 2.8 Process control deliverables

Project phase Process control deliverables Check point
Initial process control narratives (PCN)
Conceptual design + Process flow analysis Design review
+ Overall control strategy
+ Control overview diagrams
Front-end design + Key process control loops Design review + HAZOP review
(FEED) + Process control schematics
Detailed design + Detailed function specifications Design review + HAZOP review
+ Loop configuration details
Implementation + DCS control modules and sequences Factory acceptance test (FAT)
+ Logic flow diagrams
Commissioning + “As-Built” updates Site acceptance test (SAT)
Maintenance + “As-Built” updates MOC approval
Improvement + “As-Built” updates MOC approval

and costly to correct after the plant is built and operating. Therefore, process control
inputs at the early stage of process design are critical to eliminating many operational
problems from their source.
Large projects are typically carried out in phases such as appraise, select, define,
execute, and operate. The project design goes through conceptual design, front-
end engineering design, and detailed design. Table 2.8 is a simplified list of the
project phases and the process control deliverables at each phase. The process con-
trol narrative (PCN) document, the key project deliverable from process control,
is progressively developed through the different phases of a project to become the
authoritative document on process control implementation and operation. Even after
the plant starts, the PCN document remains a live document and is updated by incor-
porating all the ongoing changes, following proper management of change (MOC)
procedures.

2.5 Summary

Process control plays an indispensable and irreplaceable role in process design and
operation. This chapter puts together a knowledge framework summarizing the core
process control technologies currently in use (or in need) in the process industry and
how they have helped production and operation:
1. Feedback control is ubiquitous in process operations. On/off controllers and
standalone PID controllers account for most controllers in operation. PID control
is the most successful feedback control and consists of five essential components,
which should always be viewed together when discussing a PID control loop.
2.5 Summary 63

2. Complex PID control: Most practical problems consist of multiple control targets
and control handles that interact with each other. Several PID controllers and
supporting function blocks work together as a complex control scheme to address
the process complexity and become the complex PID control scheme.
3. Advanced process control: Anything that is above and beyond complex PID
control is considered advanced process control. The most popular advanced
process control is model predictive control (MPC), while other advanced control
technologies in use (or in need) include nonlinear control and adaptive control.
As PID control is standard for simple control problems, MPC has become the
de facto standard for complex control problems.
4. Process control design: Process control plays a critical role in the entire project
life cycle. Process control narrative is the crucial deliverable by process control
and is a progressively developed document valuable for many purposes.

References

Åström KJ, Kumar PR (2014) Control: a perspective. Automatica 50(1):3–43

Lee W, Weekman VW (1976) Advanced control practice in the chemical industry. AICHE J 22(27)
Lewis FL (1992) Applied optimal control and estimation. Prentic-Hall, Chapter 1. A Brief History
of Feedback Control
Marlin TE (2015) Process control—designing processes and control systems for dynamic perfor-
mance, 2nd edn. McGraw-Hill
Mulholland M (2016) Applied process control—essential methods. Wiley-VCH, Weinheim, Ger-
many
NOPSEMA (2015) ALARP Guidance Note. https://en.wikipedia.org/wiki/ALARP
Shinskey F (2001) Process control: as taught versus as practiced. Ind Eng Chem Res 41
Smith CL (2010) Advanced process control–beyond single-loop control. Wiley
 Part II
Essential Knowledge

Process control is a branch of automatic control. On one hand, automatic control has
been a vibrant field for over a century with a rich collection of theories and tech-
nologies, from classical control to modern control, from time-domain techniques to
frequency-domain approaches, from differential equations to state-space representa-
tion, from continuous-time techniques to discrete-time, and from PID control to H∞
control. On the other hand, industrial processes have many unique and distinct char-
acteristics, such as multivariable, nonlinear, time-varying process dynamics. The
complexities in the control structure, unpredictable load disturbances, unreliable
measurements, deteriorating valve characteristics, are among the critical challenges
in control design and operation, resulting in the most significant uncertainties to con-
trol performance. As a result, only a (small) subset of the automatic control theories
is directly applicable to the process industry. The control technologies that are in
prevalent use in the process industry include the following:
1. Simple feedback control (Chap. 3): Single variable on/off control and PID con-
trol are the dominant control solutions for the process industry, thanks to their
simplicity and reliability. Although building a PID control loop is absurdly easy,
making a PID controller work reliably under all operating scenarios is far from
trivial. Many practical enhancements and protections are required.
2. Complex PID control schemes (Chap. 4): A complex PID scheme is a natural
extension of the basic feedback control loop to handle complex control problems,
especially multivariable processes. A complex PID scheme consists of multi-
ple PID controllers and other supporting function blocks. In a complex control
scheme, the control signal can travel a complex route to reach the final control
elements, and the route may change automatically as the operating conditions
change. The change in signal path implies a change in the control structure that
can have a significant impact on the control performance. Therefore, protecting
the control loop integrity and ensuring a smooth transition during control structure
change is critical.
3. Advanced process control (Chap. 5): Complex PID control solutions can address
some complicated control problems but are not effective for many others. Besides,
66 Part II: Essential Knowledge

complex PID control solutions can become too complex for implementation. Loop
integrity is a primary concern. Advanced process control offers more elegant and
robust solutions for complex control problems and has caused a mindset change
in control design. The power of advanced control technologies comes from the
process models, which are the basis for almost all advanced control technologies
and applications, including real-time optimization, operator training simulator,
and equipment health monitoring. Model predictive control and inferential prop-
erty estimator are among the most successful model-based technologies for the
process industry.
Ironically, but not surprisingly, most of the above technologies prevalent in the
process industries originated from practice rather than academia. Their vast suc-
cess in practice slowly caught the attention of academia, which provided theoretical
proof and improvements to establish their soundness and helped proliferate to other
industries.
PID control was invented over a century ago, while model predictive control
(MPC) has been around for four decades. They have been very successful so far, but
there is an urgent need for more tools and technologies to address the process indus-
try’s increasingly stringent demands. Chapter 5 provides a summary of the process
industry’s unique challenges and particular needs and highlights the limitations of
the current technologies (Sect. 5.1) to help the practitioners make the best use of the
currently available technologies and set the expectation for the new technologies.
Process control is a practical-oriented engineering specialty. The technologies
that are suitable are those that are useful.
Chapter 3
Basic PID Control

PID is a simple three-parameter control algorithm. It is deceptively simple yet highly

competent to meet the needs of most applications. The majority of the control loops in
the process industry are of PID type. Although the success of PID control primarily
attributes to the simplicity and robustness of the algorithm, the many functional
enhancements in practical implementations play a critical role in its versatility and
reliability. The different forms of the PID algorithm (Sect. 3.1), the various operating
modes (Sects. 3.2 and 3.6), the loop integrity protection (Sects. 3.3 and 3.4), and the
rich configuration options (Sect. 3.5) are all crucial in making the PID control not
only work but also not fail. A good understanding of the operating requirements and
the control design is the key to the proper operation of the PID control algorithm.

3.1 PID Control Algorithms

The first practical application and theoretical analysis of the PID algorithm can be
traced back to the early 1920s (Minorsky 1922; Samad et al. 2020). Over the last
century, the PID algorithm has seen many generations of evolution in practical imple-
mentations, from the early mechanical, pneumatic, electronic, digital, to the latest
Fieldbus (Åström and Hägglund 2006; Marlin 2015; Visioli 2001). Nevertheless, the
PID algorithm remains a simple three-parameter equation.
A brief history of PID can be found in Åström and Hägglund (1995) and Åström
and Kumar (2014).

3.1.1 Introduction to Laplace Transform

Classical control analysis and design are based on linear control system theory and
limited to single-input single-output (SISO) processes. The underlying processes
are typically described with (high-order) linear differential equations that are often
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 67
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_3
68 3 Basic PID Control

cumbersome to handle. A transfer function representation allows algebraic manipula-

tions of the linear systems rather than working directly with the differential equations.
Laplace transform is a critical notion for creating transfer functions.
As a practical-oriented book, we strive to limit the mathematics and theoreti-
cal derivations to the minimum possible. However, Laplace transform (and the Z-
transform mentioned later) are concise and convenient representations that we find
hard to avoid. Therefore, we will assume our readers have a basic familiarity with
the Laplace transform and Z-transform for representing transfer functions. On the
other hand, even though state-space models as an important development in modern
control theory are more elegant in describing multivariable systems and are being
explored by process control applications such as model predictive control, we choose
to avoid them so that we can focus our discussion to the core and essential concepts
only.
The Laplace transform of a time-domain function f (t) is given by L( f (t)) as
 ∞
L ( f (t)) = F(s) = f (τ )e−sτ dτ (3.1)
0

which is a linear transformation,

L [C ( f 1 (t) + f 2 (t))] = C F1 (s) + C F2 (s)

 ∞  ∞
=C f 1 (τ )e dτ + C
sτ
f 2 (τ )esτ dτ. (3.2)
0 0

For example, a second-order differential equation of the following form

d 2 y(t) dy(t) du(t)

+2 + 10 y(t) = + 0.5 u(t) (3.3)
dt 2 dt dt
leads to a Laplace transform of

s 2 Y (s) + 2 s Y (s) + 10 Y (s) = s U (s) + 0.5 U (s). (3.4)

The transfer function is thus given by

Y (s) s + 0.5 s + 0.5

G(s) = = 2 = . (3.5)
U (s) s + 2s + 10 (s + 1)2 + 32

There are numerous references and resources on Laplace transformation, and the
reader is advised to review the basic concept and manipulation as needed.
3.1 PID Control Algorithms 69

3.1.2 PID Control Algorithms, Ideal Form

The basic PID controller provides a combination of three types of control actions

[Output] = [Proportional Action] + [Integral Action] + [Derivative Action] (3.6)

or in equation format
  
1 t d e(t)
u(t) = K c e(t) + e(τ ) dτ + Td (3.7)
Ti 0 dt

where

– u(t) control signal or control action (controller output OP),

– e(t) control error e(t) = ysp (t) − y(t),
– y(t) controlled variable (the process value or PV),
– ysp (t) setpoint (SP) of the controlled variable,
– K c controller gain (proportional action),
– Ti integral time, the parameter that scales the integral action, and
– Td derivative time, the parameter that scales the derivative action.

The three terms are illustrated in Fig. 3.1.

Assume a constant control error after a step change; the three types of actions are
illustrated in Fig. 3.2. The proportional action produces an initial “kick” proportional
to the control error and remains constant as long as the control error remains constant.
The integral term acts on the control error and accumulates over time at a fixed rate
if the control error remains constant. The derivative action produces another initial
“kick” proportional to the rate of change in the control error and then vanishes. As
a linear controller, the three actions are additive.
The Laplace form of the PID control algorithm is given by
 
1
U (s) = K c 1 + + Td s E(s). (3.8)
Ti s

Fig. 3.1 A PID control loop

70 3 Basic PID Control

Fig. 3.2 The P, I, and D response of a PID controller following a setpoint change

It is well known that the derivative action amplifies process noises and can cause
instability. Therefore, in actual implementation, a variation of Eq. 3.8 is sometimes
used to smoothen the derivative action:
 
1 Td s
U (s) = K c 1 + + Td E(s) (3.9)
Ti s N
s+1

where N is a constant and generally assumes a value between 1 and 33 (Visioli 2006).
In practice, the derivative term (Td s) is rarely used, except in some special appli-
cations. The selection of PID structure is based on the operating requirements and
process dynamics, i.e., how the controlled variable reacts to a change in the manip-
ulated variable. Although the three-term PID algorithm in Eq. 3.8 is useful for some
special temperature or pressure control loops, the vast majority of PID applications
are running the PI algorithm and occasionally I-only or P-only algorithms. For exam-
ple,
1. P-only: If the process is best represented at an integrator level, then a P-only
controller provides simplicity and reliability.
2. I-only: When the response of the controlled variable to a change in the manipulated
variable is instantaneous, i.e., the process is a pure gain, the I-only control typically
is a good choice.
3. PI: A PI controller is typically sufficient if the process dynamics can be adequately
represented as a first-order plus lag. The majority of the industrial processes fall
into this category.
4. PID: The full-form PID algorithm should be considered if the process is best
represented as a second-order system and the controlled variable contains little
noise.
3.1 PID Control Algorithms 71

The full-form of PID has a second-order transfer function, the PI controller has a
first-order transfer function, while the P-only algorithm has a scalar gain only:
⎧
⎪
⎪ K c E(s), P-only
⎪
⎪
⎨ Ti s + 1
U (s) = K c Ti s E(s), PI (3.10)
⎪
⎪ Td Ti s + Ti s + 1
⎪
⎪
2
⎩Kc E(s), PID.
Ti s

Some general rules of thumb for choosing control algorithms based on process
type are provided below (Brannan 2018; Smith 2009):
1. Flow loops: Flow response is typically fast and the flow measurement is usu-
ally noisy due to turbulent flow. A PI control algorithm is generally used with a
moderate gain and strong integral action.
2. Pressure loops: Depending on the process flow, the dynamics of a pressure loop can
be fast or slow. The speed of response is also different for steam, vapor, or liquid.
In some processes, such as a distillation column, the pressure response behaves
more like a ramp process within the operating range of interest. PI controller is
typically used for pressure control with a significant gain value.
3. Temperature loops: The temperature response is moderately slow because of heat
transfer lags and sensor response delay. Both PI and PID controllers can be used.
A significant gain is usually needed for tight control, and the integral time is in
the same order of magnitude as the process time constant. If the derivative term
is used, its value should generally be no larger than one-fifth of the integral time.
A PV filter should be considered if PV noise is significant.
4. Level loops: Most level controls are for inventory control with good surge capacity
available, and the level is only required to be maintained within a safe range.
There are also scenarios where tight level control is desired. For example, a
continuous stirred tank reactor (CSTR) level must be tightly controlled to maintain
the expected residence time. The level in an oil/gas separator requires the level to
be stable for phase separation. A PI controller is typically used for level control,
although in theory P-only controller can be used for level control without worrying
about steady-state offset. The gain and integral time are correlated for pure level
control, so a fixed control gain of about 1.0 is typically used, and the integral time
is the only tuning parameter to adjust.
5. Quality loops: Quality control has become essential as plant operation shifts from
process variable control to product quality control. Quality measurements from
online analyzers or lab sampling are typically slow and unreliable. For example, a
gas chromatograph (CG) may take a few minutes to an hour to produce the required
results. Quality control is thus the most important but also most challenging
control. A direct PID control is typically not feasible except for some quality
measurements that are fast and reliable such as oxygen analyzers. Advanced
control supplemented with inferential property estimation (Chap. 5) is generally
recommended. If a PID control is acceptable, a PI controller with a small gain
and considerable integral value is recommended.
72 3 Basic PID Control

3.1.3 PID Control Algorithms, Other Forms

Although the latest PID implementations have converged to a standard format, many
other forms of the PID algorithm are still around, either on legacy systems or for
reasons of back compatibility. The three popular PID equation forms are shown
below:
⎧  
⎪
⎪ 1
⎪
⎪ Kc 1 + + Td s , Ideal (Non-Interacting) Form
⎪
⎪ Ti s
⎪
⎪
⎪
⎨  
G c (s) = K 1 + 1 (3.11)
⎪
⎪ c (1 + Td s) , Serial (Interacting) Form
⎪
⎪ Ti s
⎪
⎪
⎪
⎪
⎪
⎩ K c + K i + K d s, Parallel Form.
s

If the derivative value Td is set to zero, the interactive and non-interactive forms
become identical. In practical implementation, there are many more variations based
on these three forms.
The difference in the PID forms impacts the tuning of the controllers. The popular
tuning rules such as Ziegler–Nichols (Sect. 9.2.3), Cohen–Coon, and IMC-based
methods (Sect. 9.3.4) all assume the controller has the ideal non-interactive form. It
is not difficult to convert the PID parameters from one form to another. For example,
the parallel form is related to the standard form as follows:
   
Ki Ki Kd 1
G c (s) = K c + + Kd s = Kc 1 + + s = Kc 1 + + Td s
s Kc s Kc Ti s
(3.12)
Kc Kd
→ Ti = , Td = . (3.13)
Ki Kc

In addition, a particular form may be more suitable for a specific application. For
example, the parallel form has separate gains for P, I, and D terms and is better suited
for online gain adjustments (e.g., gain scheduling in Sect. 4.1.5).
We will assume the standard PID form in all our discussions in this book for
clarity and convenience, unless otherwise mentioned.

3.1.4 PID Equation Types (“PID Structure”)

PID control aims to bring the process value (PV) to its setpoint (SP) by eliminating
the control error. See Fig. 3.3 for an illustration of the three variables in the PID
calculation. The control error is defined as
3.1 PID Control Algorithms 73

Fig. 3.3 PID process value

(PV), setpoint (SP), and
output (OP)

[Control Error] = [Setpoint] − [Process Value] = SP − PV. (3.14)

The control error can be caused by changes in either the SP or the PV. Setpoint
changes are typically initiated by the operator or another controller, while PV changes
are caused by load disturbances, measurement noises, and other changes to the PID
control loop. Based on the error source, the PID control serves two purposes:
1. Setpoint tracking: When an SP change occurs, the PID controller moves the PV
value to follow the SP.
2. Disturbance rejection: When the PV moves away from the SP due to process noise
or load disturbance, the PID controller brings the process value back to the SP.
In aerospace and robotic control, setpoint tracking is the primary objective, while
quick and smooth rejection of disturbance is critical in the process industry.
Due to the simplicity of the PID algorithm, it is often difficult to achieve optimal
performance for both setpoint tracking and disturbance rejection at the same time. As
the proportional (P), integral (I), and derivative (D) actions have a different response
on the control error (Fig. 3.2), the PID control implementations by commercial con-
trol systems all provide configuration options to allow the prioritization or trade-off
on the two objectives. The choice is typically between
1. balanced performance on both setpoint tracking and disturbance rejection and
2. prioritized performance on disturbance rejection only.
Table 3.1 provides a list of commonly available configuration options for a commer-
cial PID controller.

Table 3.1 PID equation types

Equation type Description
PID P, I, and D act on control errors (both SP and PV changes)
PI-D P and I act on control errors, D acts only on PV changes
I-PD I acts on control errors, P and D act on PV changes only
P-Only P acts on control errors
I-Only I acts on control errors
PI PI acts on control errors
PD PD acts on control errors
74 3 Basic PID Control

For example, equation type I-PD indicates that the integral action is on control
error (i.e., for both SP and PV changes), while the proportional and derivative terms
respond to PV changes only (i.e., disturbance rejection). This option eliminates the
“kick” action on SP change by the P and D terms (see Fig. 3.2) and provides the desired
trade-off between quick action on PV changes (because of proportional action) and
smooth response to SP changes (integral action only).
Commercial control systems all have their variation in the implementation of the
PID equations. For example, in the Yokogawa Centum system, the following five
equation types are available:
1. Basic type (PID): This is the standard PID above, and all three terms act on
control error.
2. PV proportional and derivative type (I-PD), where the P and D actions only apply
to the PV changes, not to SP changes.
3. PV derivative type (PI-D): The derivative action applies to the PV changes but
not to SP changes.
4. Automatic determination (default): The equation is I-PD in AUT mode, but auto-
matically changes to PI-D if the controller mode is switched to CAS or RCAS
mode (see next section for control modes).
5. Automatic determination type 2: Equation I-PD is used in AUT and RCAS mode.
The equation automatically switches to PI-D if the controller is switched to CAS
mode.
One interesting detail to note is that in Yokogawa Centum, the equation type may
change automatically following a controller mode change. For example, the default
control equation type is Automatic Determination Type 2, with which
the equation type is I-PD in AUTO and RCAS mode and automatically changes
to PI-D when mode changes to CAS. This default configuration is inappropriate
if the setpoint is provided by an external program such as model predictive con-
trol since proportional action is required to track the setpoint change. As part of
MPC commissioning, the PID equation type needs to be changed to automatic
determination type I, where the PID controller takes PI-D when the mode
changes to RCAS.
For comparison, in the Honeywell Experion system, the following four equation
types are usually provided:
1. Equation A (PID): All three terms act on the control error.
2. Equation B (PI-D): The proportional and integral terms act on the control error,
while the derivative term only acts on the PV change.
3. Equation C (I-PD): The integral term acts on the control error, while the propor-
tional and derivative terms only act on the PV change.
4. Equation D (I-Only): This equation provides integral action only.
3.1 PID Control Algorithms 75

The differences between PID, I-PD, and PI-D equations are as follows:
 
1
PID: U (s) = K c 1 + + Td s E(s) (3.15)
Ti s
 
1
I-PD: U (s) = K c PV(s) + E(s) + Td s PV(s) (3.16)
Ti s
 
1
PI-D: U (s) = K c E(s) + E(s) + Td s PV(s) (3.17)
Ti s

with E(s) = SP(s) − PV(s) being the control error. Note that the integral action, if
present, always acts on the control error since PID relies on the integral action to
eliminate the steady-state offset between setpoint and process value. On the contrary,
the derivative term D should never apply to the setpoint changes because a setpoint
change is typically a step change, and the derivative of the step change may be an
unacceptably large value.1
The derivative term produces anticipatory action and is sensitive to noisy data,
i.e., “amplifying” the process noise. It may produce a “kick” action (Fig. 3.2) that
is rarely desirable for setpoint changes. For this reason, the full-form PID equation
is usually not recommended. If the derivative action is needed, the I-PD or PI-D
form is preferred, and PV filtering should be considered.
Because disturbance rejection (on PV changes) is typically more critical than
setpoint tracking (error changes) in the process industry, a general recommendation is
to configure all PID loops as I-PD by default and use PI-D or PID on an exception
basis. For example, all standalone and protective control should be configured as
I-PD, while the secondary controller in a cascade loop or MPC scheme shall be
PI-D since fast setpoint tracking is desired.

3.1.5 PID Dimension Analysis

One crucial detail to notice is that in the PID calculation, all variables are dimen-
sionless. The control error as the input to PID calculation is 0–1 or 0–100%, so is
the controller output. The internal PID algorithm does not know or care about the
engineering units.
The basic PID calculation in Eqs. 3.7 or 3.8 can be rewritten as

[Output] = [Gain] · [Lead/Lag Compensation] · [Control Error]. (3.18)

All the variables must be normalized to dimensionless by dividing their value with
their measurement range before entering the equation:

1 An alternative to disabling derivative action is to use a setpoint filter, where a step change in
setpoint is dampened via a first-order filter.
76 3 Basic PID Control

1. The controller gain K c is dimensionless (or, with unit %/%) in the PID calculation.
The control gain K c is calculated based on the process gain K p . The process gain
is typically provided in the engineering unit and must be normalized as follows:

PV2 − PV1
PV2 − PV1 [PV Range]
Kp = , Kp = · 100% (3.19)
OP2 − OP1 OP2 − OP1
[OP Range]

where K p is the process gain in the engineering unit, and K p is the normalized
gain with no dimension.
From Eq. 3.19, it is also evident that the process gain is a dynamic gain that is the
change of process output (= controller PV) over the change of process input (=
controller OP).
Since the gain is normalized against the instrument range, re-ranging the instru-
ment also changes the process gain and inadvertently affects the controller gain.

2. The control error must be in %, even though the PV and SP are all in the engi-
neering units. That is

SP − PV
[Control Error] = · 100%. (3.20)
[PV Range]

The PV and SP must have the same engineer unit and the same measurement
range.
3. Controller output: The result of the internal PID calculation is always in percent-
age (0–100%). However, the controller output OP, which is the control action to
be sent to the final control element downstream, may be provided as 0–100% or in
the engineering unit, depending on the type of control system. For example, Hon-
eywell TDC or Experion system allows the output to be in percentage 0–100%
only, Schneider Foxboro allows engineering unit only, but Yokogawa Centum and
Emerson DeltaV allow both. Nevertheless, the 0–100% signal must be converted
to engineering unit to drive a secondary control element (see Sect. 10.1 for signal
transformation).
Figure 3.4 is an example of a simple level-to-flow cascade control loop. Assume
that the implementation is in a Foxboro DCS, and the values, units, and measurement
ranges are given in Fig. 3.4.
The level controller LC-101 and the pressure controller PC-101 are standard
overriding control through a low selector. The level controller is the active primary
controller, with an output value of 750 m3 /h.
If a change of 80 m3 /h in the flow rate (FC-101) causes the pressure (PC-101)
to change by 2 bar, then the dynamic process gain and the normalized process gain
are given, respectively, by
3.1 PID Control Algorithms 77

Fig. 3.4 Dimensions of PID controller

2
K p = 15 − 0 = 2 (%/%).
2
Kp = = 0.025 (bar/m3 /h), (3.21)
80 80
1200 − 0

The flow FC-101 has a measurement range of 0–1200 m3 /h, while the pressure
PC-101 is 0–15 bar. The implication is that a flow change between 0 and 1200 m3 /h
corresponds to a pressure change in the range of 0–15 bar, leading to a steady-state
gain from flow to pressure as (15 − 0)/(1200 − 0) = 0.0125 (bar/m3 /h). This steady-
state gain is not far off from the dynamic gain of 0.025 (bar/m3 /h), provided that the
selection of the measurement ranges is in line with the physical laws. However, these
two gains are different. In addition, the gain K c of the PID controller is based on the
normalized dynamic gain of 2%/% in Eq. 3.21, which has contributions from both
the dynamic gain and steady-state gain.
The pressure controller PC-101 is an override controller, and its output is not
selected by the low selector during normal operation. As shown in Fig. 3.4, the
pressure control error is 11 − 8.0 = 3.0 bar. If the control algorithm is proportional
only (P-only), the controller output would be given by

SP − PV 11.00 − 8.0
[Control Error] = = = 20% (3.22)
[PV Range] 15.0 − 0.0
u(t) = [Controller Gain] · [Control Error]
= 0.5 (%/%) · 20% = 10%
= 10% · (1, 200 − 0) m3 /h/% = 120 m3 /h. (3.23)
78 3 Basic PID Control

This 10% of u(t) is the incremental change in the controller output for this 20%
control error, with no regard to the actual output value. This 10% change, or 120 m3 /h
in the engineering unit, would be added to the current control output value to become
the new control output.
If the controller algorithm is PI or PID, the integral action will continue integrating
on the control error. The PID controller will continue generating incremental changes
in control action until the error reaches zero.

3.1.6 Actual Output Versus Incremental Changes

As we have already mentioned, the control action produced by the PID algorithm
is the incremental change to correct the error, not the actual value for the control
action to be sent to the next control element. The control calculation does not involve
the steady-state values of the process variables at all. For the example in Fig. 3.4,
the PID algorithm produces a 10% change in the OP in response to a 20% control
error in the PV, whether the 20% control error is built on a 5 bar pressure or a 10 bar
pressure is of no concern to the controller. Similarly, the 10% change in the OP is
purely driven by this 20% control error. The PID calculation does not know or care
whether the current OP is 40% or is already saturated at 100%.
Similarly, all the variables in Fig. 3.1, including the process input u(t) and process
output y(t) and z(t), are incremental changes rather than absolute values, to be
technically correct. This concept makes perfect sense in explaining the PID control
theory. However, in reality, the controller output will have to be converted to an
absolute value to drive the final control element, such as a valve. The valve expects
to receive an absolute value, e.g., a specific valve opening like 40%. Besides, the
measurement of the process variable is in absolute values, e.g., 8 bars. Even the
controller setpoint must be supplied in absolute values.
In other words, the process inputs and outputs are the actual values, while the
internal PID control calculation is with changes. The absolute values from the process
measurements must be converted to incremental values to perform PID calculation.
The PID calculation results, which are incremental changes, must then be converted
back to absolute values to drive the final control elements, e.g., valves. For this reason,
the full-form PID control algorithm in Eq. 3.7 should have been given as
  
1 t d e(t)
OP = K c e(t) + e(τ ) dτ + Td + bias (3.24)
Ti 0 dt

where OP is the absolute value of the controller output, and bias is an offset term
to account for the steady-state values of the control output signal. This bias term is
tricky and is a source of significant confusion in interpretation and implementation.
3.1 PID Control Algorithms 79

The conversion of the incremental change in controller output to the absolute

valve opening is typically done by adding the incremental change to the current
value of the final control element, such as the current valve opening or motor speed,
to become the full value for the next move. In Fig. 3.4, the next control output from
the pressure controller PC-101 is given by

OPnew = OPprev + u(t) · [OP Range]

= 750 (m3 /h) + 10% · 1200 (m3 /h) = 870 m3 /h.

The accumulation of incremental changes is a powerful feature of PID control

and is critical for eliminating the control error. On the other hand, the conversion
between absolute and incremental values introduces many challenges for practical
implementations. For example, if the control algorithm for PC-101 is PI or PID,
the integral action will keep producing incremental control actions added to the
controller output. Since the output of the pressure control PC-101 is not selected
by the selector, its control action would not affect the pressure, and the control error
will not go away. The continuous accumulation of the incremental changes will
eventually cause the control output to wind up to its high limit (100% or 1200 m3 /h).
This output saturation is called reset windup and is discussed in Sect. 4.3.3 in more
detail.
Thinking in dynamic changes is a crucial difference between the mindset of a
process control engineer and engineers in most other engineering specialties.

3.1.7 PID Discrete Implementation

The PID equations in the above discussions are all in continuous time (e.g., Eq. 3.7)
or Laplace form (e.g., Eq. 3.8). Because the practical implementation of PID control
is almost all in digital form nowadays, the PID equations must be converted from
continuous form (differential equations or Laplace form) to discrete form (differential
equation or Z -transform).
The discretization is through digital sampling, where a crucial parameter is the
sampling time Ts . In a digital implementation, the sampling time is determined by the
DCS scan-time or scan rate, which in turn is determined by the execution frequency:
⎧
⎪
⎨ K c e(k) + Ti e(i) + e(k) , Ideal Form
Ts Td
⎪ i Ts
u(k) = K c e(k) + Ts e(i) e(k) + TTds e(k) , Serial Form (3.25)
⎪
⎪ Ti i
⎩
K c e(k) + K i Ts i e(i) + KTsd e(k) , Parallel Form

where e(k) = e(k) − e(k − 1) is the change in control error between the two scans.
80 3 Basic PID Control

3.1.8 Positional Versus Velocity Forms

The PID algorithm in Eq. 3.7 is also called the positional form as it directly cal-
culates the control action (the incremental change). In some practical applications,
the so-called velocity form is sometimes preferred. The velocity form calculates
the difference between the last control action and the current one and is given by
differentiating both sides of Eq. 3.7:
 
d u(t) d e(t) 1 d 2 e(t)
= Kc + e(t) + Td . (3.26)
dt dt Ti dt 2

The two forms of the PID equation are mathematically equivalent. However, the
effect of the three types of control actions is more amenable to the understanding
of the velocity form. For example, the rate of change in control action is directly
proportional to the control error (the proportional action). As long as the control
error exists, there is always a change in control action (due to the integral action).
The accumulated integrated value in the position form is contained in the previous
output in the velocity form.
The discrete-time equations for the velocity form of the three PID equations in
Eq. 3.25 are given by
⎧  
⎪ Ts Td 2
⎪
⎪ K c e(k) + e(k) +  e(k) Ideal Form
⎪
⎪ Ti   Ts
⎨  
Ts Td 2
u(k) = K c e(k) + e(k) e(k) +  e(k) Serial Form
⎪
⎪ Ti Ts
⎪
⎪
⎪
⎩ K c e(k) + Ts e(k) + Td 2 e(k). Parallel Form
Ti Ts
(3.27)

where

e(k) = e(k) − e(k − 1)

2 e(k) = e(k) − e(k − 1) = e(k) − 2 e(k − 1) + e(k − 2).

The control output is then given by

u(k) = u(k − 1) + u(k). (3.28)

Note that the control action u(k) is the incremental change added to the previous
control output (see Eq. 3.24). u(k) is, in a sense, an incremental change to the
incremental change u(k), and Eq. 3.28 is only a variation of the equation in the internal
computation. In this velocity form, u(k) is proportional to the control error via the
integral terms. If the error is constant for a particular reason, the incremental control
3.1 PID Control Algorithms 81

Table 3.2 PID discretization

Positional form Velocity form
PID term Continuous-time Discrete-time Continuous-time Discrete-time
d u(t)
Output u(t) u(k) dt u(k)
P-action K c e(t) K c e(k) K c de(t) K c e(k)
Kc  t
dt
K c Ts k Kc K c Ts
I-action Ti 0 e(τ )dτ Ti i=1 e(i)Ts Ti e(t) Ti e(k)
2
D-action K c Td d de(t)
t
K c Td
Ts e(k) K c Td d de(t)
2t
K c Td
Ts 2 e(k)

action becomes constant as well, and there is no integration on control error. The
discrete form of the three terms in the PID algorithm (Eq. 3.8) is given in Table 3.2.
The following explicit equation, obtained by expanding Eqs. 3.27 and 3.28, is
suitable for computer implementation:

OPnew = OPprev + u(k)

= OPprev + u(k − 1) + u(k)
= OPprev + u(k − 1)
    
Ts Td 2 Td Td
+ Kc 1 + + e(k) − 1 + e(k − 1) + e(k − 2) .
Ti Ts Ts Ts
(3.29)

One advantage of using the velocity form is that the implementation of gain
scheduling (Sect. 4.1.5) can be more straightforward, combined with the parallel
PID equation (third equation in Eq. 3.27). Gain scheduling refers to the technique of
changing the PID gain online in a running controller. This online gain adjustment
is often desirable since a process may have very different dynamics under different
operating conditions. With a positional form PID algorithm, swapping in the new
proportional gain K c may incur a bump in the control action since the proportional
action works directly on the control error. The proportional “kick” does not exist
with the velocity form since the rate of change in control error is zero before and
after the proportional gain swapping.
The pseudo-codes for a bare-bone implementation are given in Table 3.3. The
control error is calculated from the controller setpoint and process value, and the
PID algorithm calculates the control action as the controller output. The calculation
waits for Ts seconds until the next sampling time.
The two variables in the pseudo-codes in Table 3.3, setpoint and
process_value, must first be normalized to 0–100% before entering the calcula-
tion. The calculation result, output, is also in percentage. Following the discussion
in Sect 3.1.6, the calculated PID output is the incremental change required to elim-
inate the control error. It needs to add to the actual valve position to produce the
new valve position.
A practical implementation of the same PID algorithm is substantially more com-
plicated than what the pseudo-codes have shown because various operating modes
82 3 Basic PID Control

Table 3.3 Pseudo-codes for PID implementation

initialization:
output_k1 = 0
error_k1 = 0
error_k2 = 0

loop:
error = setpoint - process_value
output = output_k1 + Kc*(1+Ts/Ti+Td/Ts) *error
- Kc*(1+2*Td/Ts)*error_k1
+ Kc*Td/Ts*error_k2
output_k1 = output
error_k2 = error_k1
error_k1 = error
wait(Ts)
goto loop

and failure scenarios must be taken into consideration. The practical considerations
are discussed in the loop integrity section later in this chapter.

3.1.9 PID with Primitive Function Blocks

PID controllers can be built with primitive function blocks (see Sect. 4.2.1) if a
PID control block is unavailable or the available PID block does not have the desired
functionality. For example, the PID functionality in most remote terminal units (RTU)
is very limited. It typically does not have built-in support for anti-reset windup in an
override control configuration. In this case, building the PID with primitive function
blocks is a last resort to work around the limitation.
Figure 3.5 shows a PI controller built with the gain and lead/lag function blocks
(in a Foxboro DCS system). It is easy to verify that the transfer function is given by
⎛ ⎞
 
OP(s) 1 ⎜ 1 ⎟ 1
G c (s) = = Kc = Kc ⎜
⎝
⎟ = Kc 1+
E(s) 1 Ti s ⎠ Ti s
1−
Ti s + 1 Ti s + 1
(3.30)
where OP(s) is the output and E(s) = SP(s) − PV(s) is the control error.
3.1 PID Control Algorithms 83

Fig. 3.5 PI controller with

primitive building blocks

Similarly, an I-only controller can be constructed with primitive blocks as in

Fig. 3.6, and the transfer function is easily verified to be

1
Ti s + 1 Kc
G c (s) = K c = . (3.31)
1 Ti s
1−
Ti + 1

From Figs. 3.5 and 3.6, it is seen that the controller feedback comes from the
output that it produces, and the controller does not know if the output reaches the
final control element. As long as there is a control error, the integrating action would
carry on, and the controller output would continue to change and may eventually
wind up or down to saturation.
Consider a scenario where a smart valve is used, and the actual valve position is
available as a read-back. The above controllers can take the valve read-back as its
feedback (FBK) instead of the output it generates, then the windup problem caused
by valve malfunction is conveniently avoided. See Fig. 3.7.

Fig. 3.6 I-only controller

with primitive building
blocks
84 3 Basic PID Control

Fig. 3.7 PI controller with

primitive building blocks and
valve read-back

3.2 PID Mode of Operation

Although straightforward in theory, the real-world implementation of the PID con-

troller is quite complicated since it has to accommodate all types of application
scenarios. One of the requirements is the different operating modes. For example,
the process may need to be controlled manually by the operator, automatically by
the PID controller, or programmatically by external software/logic through the PID
controller. For this reason, a practical implementation of the PID controller always
provides the various operating modes to meet the different application requirements.
The transition between the controller modes poses a significant challenge in imple-
mentation, and many practical considerations are required to ensure a reliable and
smooth transition.

3.2.1 PID Control Mode

A basic PID function block has two inputs and one output. The process value (PV or
CV2 ) is typically pulled from the process measurement, and the setpoint (SP or SV)
is provided by the operator or another program. The controller output (OP or MV) is
the corrective action sent down to the final control elements, e.g., valves. See Fig. 3.8
for an illustration.
In actual implementation, a PID controller is represented as a tag such as FC-101.
The tag attributes describe the property and behavior of the controller. The essential
PID tag attributes include the controller mode (MODE), the setpoint (SP or SV), the
process value (PV), and the controller output (OP or MV). They can be referenced as
PC101.MODE, PC101.SP, PC101.PV and PC101.OP.
Figure 3.9 is a simplified illustration of the basic PID controller implementation
in Yokogawa Centum.

2The nomenclature in different control system implementation is mentioned in Sect. 3.2.2, and for
now, we focus on the common concept and generic implementation only.
3.2 PID Mode of Operation 85

Fig. 3.8 Variables in a basic

PID controller (Left:
Honeywell, Right:
Yokogawa)

Fig. 3.9 PID control function block in Yokogawa Centum (simplified)

The MODE attribute determines how the controller produces the control action.
The most basic controller modes include manual (MAN ➀), automatic (AUT ➁),
and cascade (CAS ➂), as shown in Fig. 3.9. The remote cascade mode (RCAS ➃)
and remote output mode (ROUT ➄) are typically provided for advanced control
applications. These values are assigned to the MODE attribute to change the controller
behavior, e.g.,
FC101.MODE = CAS.

The operating requirement determines the mode of operation. When the operating
requirement changes with the operating condition, the control mode may need to be
changed. For example,
1. Manual control: The operator directly sets the valve position through the so-called
face-plate (the human–machine interface HMI) of the PID controller.
2. Automatic control: The valve position is automatically calculated and set by a PID
controller, whose setpoint is provided by the operator through the PID face-plate.
3. Supervisory control or custom control: The valve position is calculated and set
directly or indirectly through a PID controller by a non-PID calculation program
such as a calculation block or a piece of control logic.
In other words, the controller output can be supplied directly by the operator (MAN
mode), by the PID controller, or by another program (ROUT mode) through the PID
86 3 Basic PID Control

controller. If the PID controller calculates the output, it compares the active process
value (PV) against the setpoint (SP) and determines the control action. The controller
output is influenced by the setpoint SP, which can be set by the operator (AUTO), by
another controller (CAS), or by another program (RCAS) through the PID controller.
See Fig. 3.10 for an illustration of the different PID control modes.
Table 3.4 summarizes the modes of operation in Yokogawa Centum and Fieldbus
control system (FCS), showing where the setpoint is obtained and how the controller
output is calculated in different controller modes. Note that when the controller mode
is in MAN or ROUT, its SP is ignored. What value SP should be assumed depends
on many factors and is one of the critical considerations for PID configuration (see
Sect. 3.5).
A calculation block, including model predictive control (MPC, Chap. 5), is treated
as another program. The PID controller must be set to RCAS or ROUT mode to
receive a setpoint from MPC. Other controller modes are provided to address more
operating scenarios, such as transition needs and abnormal operating conditions.
Table 3.5 provides a complete list of the typical PID control modes.
Controller mode has priority and criticality. A higher priority mode is customarily
called a higher mode, and a lower priority mode is a lower mode. A higher mode
has a lower criticality and vice versa. Typically the controller is expected to run
in the highest mode designated by design, e.g., RCAS mode or CAS mode. This
design mode is also called the NORMAL mode in some control systems. In abnormal
operating conditions, the controller mode can or should “shed” to a lower mode for

Fig. 3.10 Common modes of PID controllers

Table 3.4 Mode of operation for PID controller in Yokogawa Centum

PID mode Description Setpoint, set by Output, provided by
MAN Manual mode – Operator
AUTO Automatic mode Operator PID
CAS Cascade mode Upstream PID PID
RCAS Remote cascade mode Upstream program PID
ROUT Remote output mode – Upstream program
3.2 PID Mode of Operation 87

Table 3.5 Common operating modes of PID controllers

MODE Controller mode Source of setpoint (SP) Source of output (OP)
OOS Out-of-service N/A System
LO Local override Master controller Controller
IMAN Initialization System System
MAN Manual Operator or tracking Operator
AUTO Automatic Operator Controller
CAS Cascade Master controller Controller
ROUT Remote output External program Controller
RCAS Remote cascade External program Controller

safety, e.g., from CAS to AUTO or even to MAN. This mode change can be performed
by the operator or initiated by a back-end logic. For example, in a cascade control
loop (Sect. 2.2.4), the setpoint for the secondary controller is automatically calculated
by the primary controller. If the operator does not trust the primary controller, he/she
can switch the control mode of the secondary control from CAS mode to AUTO and
manually assign the setpoint.
Typically, a PID controller running in lower mode than design would result in
less satisfactory control performance. A key performance indicator (KPI) for the
PID controller is the percentage of time the controller operates in its NORMAL mode.
This KPI is also called controller uptime.

3.2.2 Variations in PID Implementation

Although the PID concept is the same, PID implementation in different control
systems can be pretty different for legacy reasons, not only in the functionality but
also in the naming convention.
First of all, the three parameters in the PID formula have different names in
different control systems. Some examples are listed in Table 3.6.
The proportional band is defined as the percentage change at the controller input
required to produce 100% controller output, and thus

Table 3.6 Different names for the PID parameters

DCS vendor DCS system Control gain Integral action Derivative action
Honeywell TDC/experion Gain Integral Derivative
Yokogawa Centum Proportional band Integral Derivative
Emerson DeltaV Gain Reset Rate
Schneider Foxboro Proportional band Integral Derivatives
Foundation fieldbus FCS Gain Reset Rate
88 3 Basic PID Control

100 100
[Proportional Band] = , Gain = . (3.32)
Gain [Proportional Band]

A larger controller gain, or smaller proportional band, results in stronger proportional

action.
The reset is also called the integral or I-gain. It determines how much to change
the output over time due to the error. Reset or I-gain implies that a larger number
will have more effect. Integral implies the opposite:

1 1
Reset = , Integral = . (3.33)
Integral Reset

The reset is typically in reset/min, and the integral is in minutes. In specific control
systems such as DeltaV, the unit for integral time is in seconds, and correspondingly,
the reset should be in reset/sec. Similarly, the unit for the derivative parameter may
be in minutes or seconds, depending on the control system.
The most critical PID attributes, such as MODE, SP, PV, and OP, also have different
names in different control systems. See Table 3.7 for some examples.
Note that the Foxboro system uses multiple bit flags, including .RL for remote/
local and .MA for manual/auto, to define the controller mode. This difference can be
a special challenge for some applications such as control performance monitoring.
Although all major control systems support the common controller modes such as
MAN, AUTO, CAS, support for additional modes of operations varies from one control
system to another. However, the control modes provided by Foundation Fieldbus and
Emerson DeltaV have offered some uniformity that all can appreciate. The typical
control modes include the following:
1. OOS for out of service: The controller freezes its output at the last calculated
value, and the controller status changes to “Bad.”
2. LO for local override: The controller output is fixed at its last calculated value due
to a detected fault within the device. This mode is typically used by controller self-
tuning. It is also known as output tracking. The TRK mode in Centum provides a
similar function to LO.
3. PRD: In Yokogawa Centum, there is a primary direct (PRD) mode. When the PID
controller is put in PRD mode, the PID calculation is bypassed, and the upstream
control block directly sets the controller outputs. The output of the primary con-

Table 3.7 Common PID attributes

DCS vendor DCS system Mode Setpoint Process value Output
Honeywell TDC/experion .MOD .SP .PV .OP
Yokogawa Centum .MD .SV .PV .MV
Emerson DeltaV .MODE .SP .PV .OP
Schneider Foxboro .RL + .MA .SP .MEAS .OUT
Foundation fieldbus FCS .MODE .SP .PV .OP
3.2 PID Mode of Operation 89

troller goes directly to the next control element downstream of the failed controller.
The necessary unit conversion and range scaling are automatically performed as
needed.
4. IMAN or INIT for initialization manual or simply initialization: IMAN or INIT
mode indicates that the forward path to a physical output is broken, and the
output is tracking the downstream block. The PID output is typically fixed at its
last calculated value because the output signal path is broken. The PID output is
back-calculated to provide bumpless transfer when the loop is re-connected.
5. MAN for manual: The controller output is fixed at a value determined by the
operator. The controller status is “Good.”
6. AUTO for automatic: The PID controller calculates the output with the selected
PID equation. The operator sets the setpoint to the controller.
7. CAS for cascade: The PID controller calculates the output with the selected PID
equation. Another controller sets the setpoint to the PID controller.
8. RCAS for remote cascade: This mode is also called supervisory mode in older con-
trol systems. The controller calculates the output with the selected PID equation,
while a calculation block or external program produces the setpoint.
9. ROUT for remote output: This mode is also called direct digital control in some
older control systems. A sequence or an external program sets the controller output
and the PID calculation is bypassed.
See Table 3.8 for the different modes of operation supplied by the four major DCS
vendors and the Foundation Fieldbus controller standards.
Some control systems may provide additional modes of operations in their PID
implementation. For example, in the Fieldbus control system or Emerson DeltaV sys-
tem, the controller modes can have different statuses, such as the target mode,
actual mode, permitted mode, and normal mode. Honeywell DCS has
additional modes called NORMAL and NONE. These modes are mainly for conve-
nience or for facilitating the handshake during mode change.
There are many other differences in the actual implementation, and it is outside the
process control scope to discuss in further detail. A process control engineer should

Table 3.8 Operating modes of PID controllers by DCS vendors

Controller mode Fieldbus control DeltaV Centum TDC/experion Foxboro
Out of service OOS OOS O/S
Local override LO LO TRK
Primary direct PRD
Initialization IMAN IMAN IMAN INIT
Manual MAN MAN MAN MAN LOCAL+MAN
Automatic AUTO AUTO AUT AUTO LOCAL+AUTO
Cascade CAS CAS CAS CAS REMOTE+AUTO
Remote cascade RCAS RCAS RCAS BCAS
Remote manual ROUT ROUT ROUT
90 3 Basic PID Control

be aware of these differences and know when to seek help from control system
engineers if more in-depth information is needed.

3.3 PID Loop Integrity

The working principle of a feedback control loop, including PID control, is that the
controller sees a control error between the controlled variable and its setpoint. It
then calculates the corrective action (as the controller output) to send down to the
manipulated variable, anticipating that the process output, which is the controlled
variable, would change in the direction, magnitude, and speed as desired.
The controller is designed and configured based on a good understanding of the
loop dynamics. If the loop dynamics significantly deviate from the baseline condition
for any reason, the control performance would be affected, from minor degradation in
performance to complete breakdown. For this reason, monitoring and safeguarding
the loop integrity is the most important practical consideration in a real-world PID
implementation and is the primary difference between a PID control in theory and
PID control in practice.
Control loop stability is a significant interest of research in the academic world.
Routh stability criterion (Routh 2016) was established over a hundred years ago and
has been the foundation for stability analysis. However, as shown in this section, the
many uncertainties in the control loop, especially in the loop data flow, have reduced
the value of stability analysis.

3.3.1 Process Flow and Data Flow

Figure 3.11 illustrates how the feedback control in Fig. 2.2 is implemented in a real-
world environment. It shows the five essential components in the control loop and
the communication between the field and the control system.
Figure 2.2 is the purest feedback control loop that we are familiar with in text-
books, where all the implementation details are swept into two or three abstract
blocks (the process, the measurement, and the controller), not even considering sam-
pling and signal conversion that are essential for digital implementation. Figure 3.11
provides a more realistic but still highly simplified view of the feedback control
loop. These two drawings illustrate the fundamental difference between a laboratory
simulation and real-world implementation of the PID control loop.
Each block in Fig. 3.11 is called a function block or control element. When dis-
cussing a specific function block, we call it the primary control element or primary
element. The block downstream of it is called the secondary control element, sec-
ondary element, or secondary block. A primary control element can manipulate a
secondary control element. A valve is a final control element since it is the ultimate
control element to drive the process.
3.3 PID Loop Integrity 91

Fig. 3.11 Closed-loop control

For engineering analysis, two types of flows are essential for process control: the
process flow (the black/solid lines in Fig. 3.11) and the data flow (blue/dashed lines
in Fig. 3.11).
1. Process flow: Process operation is the continuous movement of material and
energy facilitated by piping, vessels, and equipment. A process can refer to a
simple device such as the flow through a control valve or a complex unit operation
such as a distillation column or an entire plant. We will call this physical flow of
material and energy the process flow.
As illustrated in Fig. 3.11, the process flow starts from the process input, through
the process, to the process output. Figure 1.3 is a process overview diagram (POD)
that shows the process flow of a typical oil and gas processing facility.
The process flow is the result of process design and dictates the operability of
the process. The process flow design is documented with process flow diagram
(PFD), which contains the critical information of all the equipment, the process
flow streams, the utility streams, and the critical control loops (Turton et al. 2018).
2. Data flow: The purpose of process control is to regulate the material and energy
flow as per operating requirements. Process control is based on process measure-
ments and control signals. We will call the movement of measurement data and
control signals as the data flow.
For a feedback control loop, the data flow is a closed loop that starts from the
controller as the controller output OP, travels to the actuator, works on the process,
and returns to the controller as the process value PV. The data flow represents the
control strategy and is part of process engineering and process control design.
The design and implementation of the data flow are reflected in the piping and
instrumentation diagrams (P&ID, see P&ID standards ISA-S5 and ISO-3511)
92 3 Basic PID Control

as dotted or dashed lines.3 However, advanced control schemes such as model

predictive control (MPC, Chap. 5) cannot be adequately shown in P&ID, and
there is no consensus or standard on how to document advanced control schemes.
The data flow is analog (continuous) primarily in the field but is digital inside
the control system. The analog input (AI) and analog output (AO) function blocks,
which at the center are analog/digital (A/D) and digital/analog (D/A) converters,
perform the conversion between the analog and digital signals. The data transmission
between the field and control room is typically in an electric current of 4–20 mA.
The 4–20mA signal typically follows the NAMUR NE 43 recommendation. The
most crucial notion is the “live zero,” i.e., a zero value is transmitted as 4 mA instead
of 0 mA. This standard enables the detection of signal loss and sensor failure. See
Sect. 10.1 for more details on the signal transmission and transformation along the
data flow path.
With the new Fieldbus technology, the communication between the control system
and the field instruments can be all digital. The conversion between analog and digital
occurs inside field equipment instead of the AI and AO modules in the control system.
The actual design and working of control data flow are complicated and involve
substantially more details than shown in Fig. 3.11, and are not part of core knowl-
edge for process control engineers. Nevertheless, process control engineers should
understand how the process flow and data flow affect process control design and
functioning.
The primary concern is the integrity of the data flow with the following three
common abnormal scenarios:
1. Bad data: The data along the data flow path may become unavailable or unusable.
For example, communication between control elements may be lost; the controller
may receive a Bad PV value that it cannot act upon; a secondary element on the
data flow path such as the AO block may be out of service (OOS) and does not
propagate the control signal further.
A common cause of Bad PV is the loss of signal, or the PV value is out of
the measurement range. In some control systems, there is an option to clamp
out-of-range values. A clamped value is not reported as a bad value, but it stays
unchanged not tracking the actual value.
2. Disconnected data flow: The data flow involves multiple components and can
become disconnected for many reasons, most commonly by operating needs such
as mode changes. A broken connection causes the loop to open and thus puts the
loop out of control. For example,
• A secondary element such as a controller or AO block may no longer be in
cascade mode or becomes inactive and thus breaks the loop.
• A high-selector may have selected the control output from another controller.
• If the secondary element is in the initialization mode, the loop is also broken.

3Some process engineers tease the process control contribution as “only the dotted lines on P&ID,”
which grossly undervalues process control engineers’ role in a project.
3.3 PID Loop Integrity 93

3. Constrained data flow: When the control output is sent down to the final control
elements, it may travel a complicated path (see complex PID control loops in
Chap. 4). The control signal may exceed the high or low limits set by the secondary
elements and becomes constrained. For example,
• A valve has a limit of 0–100% in its opening, and a motor has a speed or power
limit.
• A control block has a high and low limit on its setpoint when in cascade mode.
When the control signal is limited, the effectiveness of the control action will
be compromised. The primary controller must be aware of the problem and take
mitigating actions as needed.

3.3.2 Process Dynamics and Loop Dynamics

Process control aims to control the process, and the process dynamics are the basis
for control. In Fig. 3.11, the process dynamics is the time response characteristics
from the process input to the process output, i.e., the dynamics of the process flow.
For example, in Fig. 3.4, the process dynamics for the level controller LC-101 is
from the flow FI-101 to the level LI-101. In Fig. 1.7, the process dynamics for the
compressor suction pressure controller is from the compressor speed to the pressure.
However, from Fig. 3.11, it is evident that the process is only a part of the con-
trol loop. A complete control loop comprises many essential components, including
the sensor/transmitter, the valve, the valve positioner, analog/digital converters, and
the current/pressure converter. They all have dynamics of their own, and the loop
dynamics are the combined dynamics of all the components along the control loop,
i.e., the dynamics of the data flow. That is,

G loop (s) = G AI (s) · G v (s) · G p (s) · G Tx (s) · G AO (s) · G f (s) · · · (3.34)

For example, the dynamics of the AI and AO function blocks, G AI (s), G AO (s), the
sensor and transmitter G Tx (s), the control valve G v (s), and the PV filter G f (s) are
all part of the loop dynamics. As a result, the loop gain and the process gain can be
significantly different.
Controller design and operation are based on the loop dynamics rather than the
process dynamics alone. Although the process dynamics is the most critical compo-
nent in the loop dynamics, there are applications where the dynamics of the other
components may be significant. For example, the compressor surge phenomenon
(see Sect. 11.3) is extremely fast. Therefore, the compressor anti-surge control is
required to execute every 100 ms or faster. The speed of transmitter response is at
the same order of magnitude as the controller, but the valve dynamics are two order-
of-magnitude slower. In this case, the valve dynamics become dominant in the loop
dynamics. Besides, some discrete events inside the control loop, such as the discon-
nection of the data flow or limiting of the signal, may have a more detrimental effect
94 3 Basic PID Control

on the overall loop dynamics. The detection and mitigation of these discrete events
is a critical consideration in a real-world PID implementation.
PID control loop design is based on an assumed cause and effect dynamic rela-
tionship between the manipulated variable (controller output OP) and the controlled
variable (the controller PV), as shown in Fig. 3.11. It also assumes that this relation-
ship remains (relatively) unchanged during operation. In practice, however, these
assumptions often become invalid due to changes to the data flow. Therefore, the
integrity of the data flow is the primary challenge to a control loop’s reliable opera-
tion. A mismatch between the PID tuning parameters and the changed loop dynamics
can deteriorate control performance, while broken loop dynamics can result in the
PID loop going completely off control.
Below are the potential consequences from loop dynamics issues:
1. Bad data: Data is not available due to Bad Data or control elements out-of-
service. In this case, the cause and effect relationship established by the loop
dynamics ceases to exist (G loop in Eq. 3.34 becomes meaningless), and the con-
troller is effectively off control.
2. Data flow is broken: In this case, the controller may still receive PV updates, but the
change in process output is not the result of the controller action. The established
cause and effect relationship is no longer valid since the control action will not
reach the final control elements and thus would not affect the controlled variable.
3. Limited data: If data value becomes limited (e.g., saturated) somewhere on the
data flow, the cause and effect relationship exists but is no more accurate. A change
in the control action will not cause the expected change in the PV.
4. Drifting in loop dynamics: The gradual deviation of the loop dynamics from
the design condition may be caused by various reasons such as a fouled heat
exchanger, a worn-out valve, or a flow meter out-of-calibration. The changed
loop dynamics result in a gradual deterioration in the control performance. These
abnormal conditions are more difficult to detect reliably, and remains a significant
challenge in both theory and practice.
The first three scenarios mentioned above are partially addressed by proper control
design and implementation. The fourth problem is related to performance monitoring,
and there is no standard solution yet. See Sect. 10.2 for more detailed discussions.

3.4 Tracking and Initialization

In a real-world environment, as shown in Fig. 3.11, the control action produced by

a PID controller may not always reach the final control element that it intends to
due to problems in the data flow. Therefore, it is critical to track the data integrity,
detect abnormal conditions, and mitigate. Tracking and initialization is a common
approach taken by most control system implementations.
3.4 Tracking and Initialization 95

3.4.1 Tracking

It is evident from the internal architecture of a PID controller (Fig. 3.9) that the control
modes and data processing options can potentially alter the data flow in a control
loop. For instance, switching the PID controller from cascade mode to automatic
mode interrupts the data flow in a cascade control loop.
The first step in protecting the loop integrity is tracking the data flow for dis-
connections or constraints. In most control systems, a control element can inform
the primary control element of its operating mode, error condition, and constraint
status. If all the control elements in a control scheme support status monitoring and
broadcasting, then the control loop’s overall integrity can be tracked. In case of any
issues as listed in Sect. 3.3.1, the controller can take appropriate action to mitigate.
Both the process and the controller have many operating modes other than the nor-
mal mode; some are legitimate normal operating modes, while others may be crippled
modes due to failures somewhere else in the loop. Mode changes, either intentional
or unintentional, are frequent events in operation. A mode change typically involves
the following three steps:
1. Transition from the normal mode to a different mode: For a mode change, a crucial
requirement is that the transition must be bumpless. One example is the transition
from the normal operating mode to a crippled mode after a failure is detected.
The transition may require changes to the controller mode, the setpoint, or the
output following pre-defined operating procedures. It is critical to minimize the
disturbance to the process operation.
2. Abnormal mode of operation: In case of abnormal mode operation such as crippled
process operation or shutdown, the controller mode, setpoint, and output settings
require careful considerations. For example, should the controller be switched to
manual mode and the output be set to fully closed?
3. Transition back to the normal mode: When the abnormal condition is resolved, and
the controller needs to resume normal control, the condition may have significantly
changed. Simply re-connecting the data flow may cause a bumpy transition. For
example, if the controller output is 50% and the AO has been in manual mode
with a 30% output, closing the loop would bump the AO setpoint (thus the AO
output) to 50% in one shot. Since the AO output is typically connected to a control
valve, this sudden change in AO output may be an unacceptable disturbance to
the process operation.
For a smooth transition, it is critical to track and monitor the loop dynamics for
changes continuously. The current practice includes the following:
1. Tracking the integrity of the data flow: The control systems tracks and monitors
the data flow for status and value changes such as disconnection or limitation in
the data flow. If the operating mode or data status poses a danger to the data flow
integrity, the PID controllers are informed, and appropriate mitigating actions
would be taken.
96 3 Basic PID Control

Due to technological limitations, the data flow tracking is usually only imple-
mented from the primary controller down to the AO block. Currently, the tracking
is limited to discrete events such as a change in the controller’s operating mode
or a change of selection in the control selector (see Fig. 3.11). If smart devices
are utilized, the tracking may extend to the final control elements but cannot go
further with the analog process data.
2. Monitoring data health: Control implementation continuously monitors and pro-
cesses the I/O data for anomalies. The data may be categorized into GOOD (nor-
mal), BAD, or UNCERTAIN to help the controller make appropriate decisions.
For example, an out-of-range data value is labeled BAD by default. However, if
the clamping option is enabled, the out-of-range value will be clamped to the
at-range value and be marked as UNCERTAIN.
The data health monitoring is part of the I/O processing in most control function
blocks but is still primitive. The health checking is typically limited to out-of-
range verification only. More practical and reliable monitoring technologies are
needed to protect the loop integrity and improve the control loop reliability.
For reliable PID control, tracking and monitoring must cover the data flow of
the entire control loop. However, many events that significantly impact the loop
dynamics do not have standard or mature theories or techniques to detect reliably
due to technological and practical reasons. For example,
1. Gradual changes in process dynamics that are caused by operating condition
changes, equipment fouling, or seasonal temperature changes.
2. Changes in valve characteristics, such as increased stiction, deadband, or hystere-
sis are caused by wear and tear.
3. Process measurements have drifted away, e.g., due to lack of calibration.

3.4.2 Back-Calculation

When the normal data flow is re-established after disconnection, all the control ele-
ments on the data flow path are expected to quickly and smoothly restore the normal
operating conditions without introducing significant disturbances to the controller
output. The resumption of control should not cause the final control elements to
change if the operating conditions have not changed. Even if the operating condition
has changed, the controller should not generate abrupt changes in the final con-
trol elements. This smooth transition is called a bumpless transfer and is typically
achieved through proper initialization (or re-initialization) of the control elements.
Because the goal is to eliminate or minimize the transition bumps in the final control
elements, the initialization starts from the final control elements and goes backward
along the data flow. This type of initialization is thus called back-initialization.
Back-initialization is typically facilitated by an additional information flow called
back-calculation data flow, also called secondary data flow, that complements the
primary data flow. The primary data flow propagates forward, carrying the control
3.4 Tracking and Initialization 97

Fig. 3.12 Process flow, data flow, and back-calculation

data (the calculated control action) from the controller to the final control elements,
while the back-calculation flow propagates backward, carrying the control elements’
values and statuses back to the controllers. See the dashed lines in Fig. 3.12 for the
data flow and the dotted lines for the back-calculation information flow.
The back-calculation flow of a simple flow control loop is shown in Fig. 3.13,
where the normal control loop is shown on the top, and the connection of the function
blocks is shown at the bottom. If supported, the back-calculation flow starts from the
control valve, propagates back to the AO block, and then to the PID block.
Back-calculation flow is provided to support back-initialization. It complements
the data flow and is part of the control information flow. The relationship between
the process flow, the data flow, and the back-calculation is illustrated by Eq. 3.35:
⎧ 
⎪
⎪ Physical Material Flow
⎪
⎪Process Flow
Closed-Loop ⎨ 
Physical Energy Flow
(3.35)
Control ⎪ ⎪ Process Data + Monitoring
⎪
⎪
⎩Data Flow Control Data + Back-Calculation.

The back-calculation information flow is primarily system-related, so it is beyond

this book’s scope to provide in-depth coverage. Only the critical information essential
to the control loop’s functional integrity is briefly discussed here.
Every control function block has an input connection and an output connection for
back-calculation. Different control systems may have different names for the back-
calculation signals. For example, in the DeltaV system, the back-calculation output
is called BKCALO, and the back-calculation input is called BKCALI. In the Foxboro
98 3 Basic PID Control

Fig. 3.13 PID back-calculation in a flow control loop

system, the back-calculation input and output are called BCALCI and BCALCO,
respectively. They need to be manually and explicitly connected in the control loop
configuration. In Honeywell Experion and Yokogawa Centum systems, the con-
nections between the primary function block and the secondary function block are
automatic and implicit, hidden to the user.
Without loss of generality, let us call the two connections as BKCALI and
BKCALO. Combined with other general-purpose connections that carry the con-
trol element’s status, they provide the value and status of all the control elements for
communication between the control elements. The value and status typically include,
among others:
1. Control modes: These are primarily the control modes of the function blocks.
For example, a function block in manual mode indicates that the data flow is
disconnected at this function block.
2. Connection status: If selectors and switches are used, the primary controllers must
know which controller’s output is selected.
3. Constraint status: The constraint status refers to whether the data flow at this
function block is high-limited, low-limited, or both.
4. Actual value: The process value (PV) of the function block. The purpose is to
inform the upstream function block of the current function block’s process value
(PV) or output value (OP).
5. Handshake information: The handshake information is communicated between
function blocks during mode changes.
3.4 Tracking and Initialization 99

The BKCALO connection is for the current controller to report its value to the
upstream control blocks. The current control element also receives similar back-
calculation information as BKCALI from the control elements downstream. The
controller responds to the secondary function block’s status changes and performs
appropriate initialization accordingly.
In the simple flow control loop in Fig. 3.13, the AO block sends its mode status and
output value to the PID controller via the connection from BKCALO to BKCALI. The
PID controller propagates its status and value to the upstream function block if exists.4
This daisy chain connection of BKCALO to BKCALI constitutes the back-calculation
information flow. This back-propagation is possible only if all the function blocks on
the data flow path fully support back-calculation, which is not always true in practice.
For example, in Fig. 3.13, the AO block may not receive the on/off status or position
from the downstream valve FCV-101. Besides, many arithmetic calculation blocks
do not support back-calculation by design.
The information flow, including the data flow and back-calculation, can become
overwhelmingly complicated for a complex control scheme. Its impact on control
performance is often underestimated.
Example 3.1 Back-calculation in a cascade control loop. The data flow of the simple
cascade control loop in Fig. 2.18 is shown as Fig. 3.14.
The control action from level controller LC-101 is sent down to the secondary
controller FC-101 and further down to the control valve FCV-101. The change in
the valve opening causes the flow FT-101 and level LT-101 to respond to maintain
the flow and level at their respective targets. However, this data flow can be disrupted
by many events, including the following:
1. The flow controller FC-101 is switched to manual (MAN) mode by the operator,
and thus the controller is disconnected from the primary controller LC-101. The
data flow path from LC-101 to the valve is broken.

Fig. 3.14 Information flow in a cascade control loop

4 There may be a limit on how many levels of upstream blocks the back-calculation and can propagate
in one scan cycle.
100 3 Basic PID Control

2. The flow controller automatically sheds to manual mode upon detecting bad flow
measurements (BAD PV) in FT-101.
3. The valve FCV-101 is stuck and does not follow the control output.
4. The analog output (AO) module is switched to manual mode and no longer prop-
agates the control output to the controller valve FCV-101.
5. The level measurement LT-101 is frozen or saturated. It still produces a value,
but the value remains constant.
6. Finally, even the control valve FCV-101 can be switched to manual and bypass
the control action in some control systems and with Fieldbus devices.
The feedback control loop that we have been familiar with refers to the data flow.
However, the data flow has no information on whether the change in the process value
is a result of the control action that it sends down to the valve. The back-calculation
flow (dotted blue line in Fig. 3.14) tracks the integrity of the data flow and “informs”
the controllers of the status and value of the downstream control elements. In this
sense, the back-calculation flow serves as the feedback for the data flow. In other
words, the data flow makes the control loop work, and the back-calculation flow tries
to make the control loop not fail:
1. At the secondary controller (FC-101, ➀), the mode of the flow controller is part
of the information fed back to the primary controller. Once the level controller is
informed that the flow control is no longer in cascade mode, the level controller
automatically switches to initialization mode (INIT or IMAN, depending on
control system type). The level controller’s output (LC101.OP) would typically
track the flow controller PV or setpoint (FC101.PV) provided via the BKCALO
to BKCALI connection and remain ready for bumpless transfer when the flow
controller is switched back cascade.
2. At the analog output block(AO), the AO output is fed back to the flow controller as
BKCALO through the back-calculation flow. The downstream valve information,
if available, is received at the BKCALI connection.
3. At the valve (FCV-101 ➂), it is expected that the actual valve position is available
to be used by the AO block as the BKCALO value. However, the actual valve
position is typically unavailable (disconnection at ➁) for older valves. If AO does
not receive the value position valve at its BKCALI, it would simply send its own
output as the back-calculation value to the primary controller. This “tail in the
mouth” data connection results in a floating value for the back-calculation and is
a common cause of controller windup or degraded performance.
Smart valves with Fieldbus configurations can now provide the actual valve posi-
tion as read-back, which the AO can use as the back-calculation value for the
BKCALI input. This value is then propagated back to the controllers.

Compared with the data flow, which is a closed-loop from the controller to the
process and back to the controller, the back-calculation information flow is only a
partial loop, from the final control element back to the controller at best. The status
and health of the rest of the data flow are missing. Tracking and monitoring the entire
data flow is necessary but is still a challenge.
3.4 Tracking and Initialization 101

3.4.3 Initialization

Bumpless transfer during transition is a crucial requirement for process control.

When the data flow is disconnected, the controller output will not reach the final
control element. The output of the primary element becomes out of synchronization
with the setpoint of the secondary element. The values must be back in-sync again
before closing the loop to avoid the so-called transfer bump, which is the purpose
of controller initialization. Several common scenarios may trigger an initialization,
including the following:
1. The control element is started or activated the first time.
2. A request from an external user program or logic.
3. The control output was floating (indisposable), and now one or more output con-
nections have become disposable.
Many practical considerations in PID configuration options are to ensure bumpless
transfer. The typical approach is back-initialization, which requires several pieces
of information to make the right decisions:
1. When to perform the back-initialization? When the loop is re-connected via mode
change or selection change, there is a complex procedure of status tracking and
broadcasting within the control system to inform the relevant control elements of
the status change. Most of the time, this internal tracking and communication are
sufficient to initiate the initialization procedure.
2. Is back-calculation possible? A simple controller shown in Fig. 3.12 has one sec-
ondary control element, so the back-initialization is straightforward. For a more
complex control scheme where the control output is shared by multiple secondary
control elements such as a split-range control, fan-out control, or override con-
trol, back-initialization may not be possible as the controller output is already
connected to some secondary elements. In this case, alternative actions must be
taken by the controller to achieve bumpless transfer. In extreme cases, extra func-
tion blocks may need to be provided in the control design to ensure bumpless
transfer. See Sect. 4.3 for discussions.
3. What value should the controller output be back-initialized to? In the simplest case
shown in Fig. 3.11, the PID controller is expected to align its output to match the
AO block’s setpoint before re-connection. However, how would the PID controller
be informed about the setpoint value of the AO function block?
Initialization of the function blocks starts from the final control elements and goes
backward along the data flow. When a disconnected data flow is re-connected (e.g.,
a secondary element is switched back to cascade mode), the secondary element will
make an initialization request to its primary element. For example, if the AO output in
Fig. 3.14 is at 30%, and the flow controller FC-101 output is at 50%, the controller
would set (re-initialize) its output to 30% before re-connecting with the AO block.
The PID controller FC-101 initializes itself to match the AO output, not the other
way around. The back-initialization propagates backward until it reaches the topmost
controller LC-101 or reaches a control element that cannot be back-initialized.
102 3 Basic PID Control

In most cases, the primary element can simply adjust its output accordingly to
match the secondary element. However, there are many scenarios that the back-
initialization may not be possible. For example, for a fan-out control (see Sect. 2.2.3),
the controller output may already be connected to one or more control elements. If
one additional control element (e.g., a valve) comes back online, the controller cannot
back-initialize to setpoint value of the newly connected element. As a result, an initial
gap exists between the primary element’s output and the newly connected element’s
setpoint, and a transfer bump may ensue at the time of re-connection. In this case,
it becomes the secondary element’s responsibility to initialize itself to the primary
controller. In a fan-out control scheme, assume the controller output is at 50% and
is connected with two valves; if a third control valve having a valve opening of 30%
is switched back to cascade, the controller cannot back-initialize its output to 30%
since it is already connected to the two valves at 50%. This 20% gap will cause a
bump in the valve opening upon re-connection.
A built-in bumpless transfer feature exists in many control function blocks, includ-
ing the control splitter used for fan-out control. The bumpless transfer function grad-
ually decays this initial gap to achieve a smooth transition. See Sect. 4.3.4 for a
detailed discussion. Process control engineers must be aware of the potential trans-
fer bump in the control scheme and ensure a bumpless transfer feature is available
in the design. If not, a dedicated function block may need to be inserted to achieve
bumpless transfer. The manual loader (or AUTOMAN block) is the simplest function
block in DCS that can be used for this purpose.
Another challenging scenario is with bad PV. If a PID controller’s PV goes bad
and subsequently comes back to normal, a large gap between the setpoint and PV
may already have developed. Restoring the PV may lead to a bump in the control
error, thus the controller output. If the bumpy behavior is unacceptable, then operator
intervention is expected to align the setpoint with the process value first and gradually
bring the setpoint to the desired value.
In summary, tracking and initialization are automatically achieved for most of
the commonly encountered control schemes consisting of standard control elements.
However, some control schemes with complex data flow do not have the full support
for back-calculation. These complex scenarios would require the process control
engineer to either change the control design or use back-end logic to ensure the
continuity of the back-calculation data flow. Therefore, it is crucial to know when
the control system can automatically take care of the back-calculation and when extra
help should be provided in the control design.

3.5 PID Configuration

PID configuration is an overlapping area between process control and control sys-
tems. Although primarily the responsibility of control systems engineers/technicians,
some configuration options are closely related to the process design and control strat-
egy and only the process control engineers are qualified to specify.
3.5 PID Configuration 103

Table 3.9 Typical configuration parameters of a PID controller

Configuration Description, options, or examples
Algorithm type P, PI, PID etc.
Equation type (structure) I-PD, PI-D, I-only, etc.
Normal mode AUTO, CAS, RCAS, etc.
Tuning parameters System default values typically are NOT acceptable!
Direction of action DIRECT/REVERSE
Setpoint high/low limit The valid value range for setpoint
Fastest scan rate/sequence 50 ms, 100 ms, . . ., 1 s, . . .
Output format Positional/Velocity
Output scaling 0% = Fully closed, 100% = Fully open
Fail-safe position Fail-open, Fail-close, Fail-stay
PV filtering Noisy signal can cause poor performance
PV tracking Setpoint tracks PV when MODE in MAN
Anti-reset windup Integral windup can cause delayed response
Bumpless transfer No “bump” during MODE change

As discussed in previous sections, implementing a PID controller is much more

complex than just solving a three-parameter equation. Many things can go wrong and
probably will go wrong if not correctly configured or prevented. Table 3.9 provides
a list of the crucial PID configurations that are common in all control systems. Their
incorrect configurations may have a significant impact on the control performance.
These configurations require a good understanding of the control objectives and
control strategy and are thus best determined by process control engineers as part
of process control design. Note that the control systems have default values for all
the configurations, but many of the default values are not acceptable and must be
supplied with the correct values!

3.5.1 Output Scaling

Controller output, typically in the range of 0–100%, is converted to a 4–20 mA

analog signal by the analog output (AO) module in the control system and then sent
to the valve’s actuator in the field. See Fig. 3.11 for an illustration. For fail-close
(air-to-open ) valves, a 4 mA signal drives the valve to the fully closed position
(fail-safe position ), and a 20 mA signal drives the valve to the fully-open position.
However, for a fail-open (air-to-close ) valve, a 4 mA signal corresponds to a fully
open position (fail-safe position), and a 20 mA signal would drive the valve to a fully
closed position.
104 3 Basic PID Control

Table 3.10 Control signal and valve fail-safe position

Controller output (%) Intended valve position For fail-close valve (mA) For fail-open valve (mA)
0 Fully closed 4 20
100 Fully open 20 4

With fail-open valves, a signal inversion has to be performed along the way from
the controller to the valve to ensure negative feedback. This inversion can be done
either in the controller, where 100% is defined as a closed valve, or in the analog
output (AO) module, where a 100% signal is mapped to a 4 mA instead of 20 mA.
Both configurations exist in practice, and which one to choose has been mostly a
personal preference. This decision, however, does affect the configuration of PID
direction of action and is also a source of confusion.
In modern process control design, to avoid confusion, it is customary to follow a
convention that an output of 0% from the PID controller always represents a fully
closed valve and 100% a fully open valve. The signal inversion, when required, is
performed in the AO module. See Table 3.10 for an illustration.

3.5.2 Direction of Action

The PID controller can be configured as either DIRECT or REVERSE acting. Along
the closed data flow path of the control loop, the process can have a positive or
negative gain, the control valve can be air-to-open or air-to-close, and the AO signal
can be direct or inverted. Therefore, the loop gain can be either positive or negative,
out of a combination of 16 possibilities (see Table 3.11). A stable control must be
a negative feedback control, which requires that the overall loop dynamics have a
negative gain. If the direction of action is wrong, nothing else will matter because
the controller output will quickly run off-scale in the wrong direction regardless.
The convention mentioned above in Sect. 3.5.1 requires that 0% of controller
output cause the valve to close fully, and 100% of controller output fully opens the
valve. That means the AO is configured so that the combined dynamics of the AO
block and the control valve always have a positive gain. The analog output (AO) needs
to perform a signal inversion to produce a positive gain for an air-to-close valve, as
illustrated in Table 3.12.
With this simplification, the AO and controller valve is taken out of the picture.
The sign of the controller gain (direction of action) is determined by the process
gain alone and should always be opposite to the sign of the process gain to ensure
negative feedback. See Table 3.13.
Table 3.11 Loop PID controller Analog output (AO) Control valve Process
components and the sign of
loop gain Direct acting Normal scale Air-to-open Positive gain
or or or or
Reverse acting Inverted scale Air-to-close Negative gain
3.5 PID Configuration 105

Table 3.12 AO and controller valve should produce a positive gain

Control valve AO block Combined gain
Air-to-open × Normal signal → Positive gain
Air-to-close × Signal inversion → Positive gain

Table 3.13 Determining PID direction of action

Process gain PID direction of action Example
Positive gain → Reverse acting Flow control FC-101 in Fig. 3.14
Negative gain → Direct acting Level control LC-101 in Fig. 3.14

In practice, the following rule of thumb can be followed to determine whether a

controller should be configured as DIRECT or REVERSE acting: when the process
value (PV) to be controlled is higher than the controller setpoint (SP), if the desired
action is to increase the controller output (OP), the controller action is defined as
DIRECT action, and vice versa.
For example, in Fig. 3.15, the level controller LC-101 should be DIRECT acting
because if the level is above the desired setpoint, the outlet flow should be increased.
Similarly, the flow controller FC-101 should be configured as REVERSE acting so
that if the flow were to increase suddenly, the valve opening would be decreased
to reduce the flow. This decision is completely decoupled from the decision on the
direction of action of the AO block and the fail-safe position of the valve.

3.5.3 Setpoint High/Low Limit

When the PID control is in AUTO mode, the setpoint is entered by the operator. The
setpoint high and low limits (see SPH and SPL in Sect. 1.3.5) should be configured
to protect against human entry errors. The decision on the range between SPL and
SPH is a joint decision between process engineering and operations.
In a cascade control loop, the setpoint high and low limits of a secondary PID
controller are part of the status tracking. When the limits are violated, the data flow

Fig. 3.15 A basic PID cascade control loop

106 3 Basic PID Control

is reported as limited, and the primary control automatically activates the anti-reset
windup action.

3.5.4 Execution Frequency and Execution Order

The controllers have a default execution frequency (scan time or scan rate), typically
every second. Any controller operating faster or slower than the standard execution
frequency shall be mentioned explicitly in the control design. For example, compres-
sor anti-surge controllers typically need to execute at a much higher frequency, such
as 100 ms per execution.
Another important consideration during implementation is the order of execution
of the functional blocks. In a control system, each control block along the control
path is assigned a number to specify the order of execution.5 For a complex control
scheme, the calculation may be delayed by several scan cycles if the order of execution
is incorrectly specified. For example, for the cascade control loop in Fig. 3.15, the
execution should start from the primary loop LC-101; the calculated output is sent
down to the secondary loop FC-101, where the control action is calculated and
the output sent down to the valve FCV-101. Following this execution order, all the
calculations in this complex scheme are completed with the latest measurements
and within the same execution cycle. If the order is reversed and the secondary loop
FC-101 is calculated first, its setpoint would be the result from the last execution of
controller LC-101, based on the previous level measurement. This cascade control
will take at least two scan cycles to complete, with level and flow measurements
from two different cycles.
For a complex control that includes a selector block (such as in override control)
in its data flow path, the order of execution is crucial. All the primary blocks must
run before the selector block so that the feedback is correct.

3.5.5 PV Filtering

On a modern control system, filtering is typically provided as a standard configura-

tion option. Adding filtering to a PV is just a matter of activating this function by
specifying a non-zero filter constant. However, the PV filter introduces an extra lag
to the loop dynamics, potentially affecting the controller tuning and slowing down
the control action.
The following recommendations should be followed when configuring the PV
filter:

5In Honeywell Experion, for example, if the execution order is not specified, the default order is
assigned which is based on the order that the function blocks are added to the system.
3.5 PID Configuration 107

1. The PV filter constant should be no larger than 1/5 of the controller integral time.
2. When the PV filter time is added or changed, its impact on the controller tuning
should be assessed to decide if the controller needs to be re-tuned.

3.5.6 Initial Tuning Values

Different control systems have different implementations of the PID control algo-
rithms. They all have a set of system default values for the PID tuning parameters.
For example, on the DeltaV system, the system default tuning values for a PID loop
has a gain of 0.25 and an integral of 10 s. Unfortunately, these default values are
typically not acceptable for most control loops. As an extreme example, all the level
control valves of the oil/gas separators in one new plant wore out (severe leaking
or passing) in just two years. Investigation revealed that all the PID tunings were
left at the system default at gain = 0.25 and integral = 10 s. The level controllers are
expected to have an integral time of about 20–30 min instead of 10 s!
Although the PID control parameters cannot be finalized until the plant is in
operation, it is essential to specify the initial PID tunings during the design phase
of a new project since the system default values are rarely appropriate. These initial
values are meant to set the controller on its foot for the startup. Some recommended
initial tuning parameters can be found in Table 9.1. During commissioning, the
parameters should be fine-tuned.
Tuning parameters for critical control loops, particularly those with long dynam-
ics, can sometimes be calculated based on dynamic simulation results if available.
As dynamic simulation becomes more popular, controller tunings can be calculated
based on simulation results even before the construction of the actual plant starts.
The high-fidelity process models from dynamic simulation can provide sufficiently
accurate data for PID tuning.

3.5.7 PV Tracking

When PV tracking is enabled, the PID controller setpoint will track the PV value
if the controller has switched away from cascade (e.g., to MAN), and the original
setpoint value is discarded. Enable PV tracking will force the control error to
zero so that when the PID controller is switched back to cascade mode again, there are
no transition bumps in the control action. However, some control applications require
that the controller setpoint retain its value during controller mode change, such as
protective and level controllers. It is thus a trade-off between retaining the setpoint
and ensuring bumpless transfer. The decision is thus controller-specific but should
be documented in the control narrative. Generally speaking, standalone controllers
and protective controllers should have PV tracking disabled, so the operator does not
need to remember and reset the setpoint values.
108 3 Basic PID Control

3.5.8 Anti-Reset Windup

Reset windup is a common concern with PID control. In modern control systems,
with the help of a back-calculation information flow, the anti-reset windup function
can be automatically achieved through proper connection and configuration of the
control blocks (controller, selectors, splitters). However, there are many cases in
complex control loops where back-calculation is not fully supported. In this case,
external feedback may need to be manually configured. Reset windup is an essential
consideration for a process control engineer during process control design. Detailed
discussion is provided in Sect. 4.3.3.

3.5.9 Bumpless Transfer

The data flow in a feedback control loop is round-trip. The controller’s output causes
the process to have the desired change, reflected by the process value fed back to
the controller. The path from the controller output to the final control elements must
remain connected. If the connection is interrupted anywhere along this path, e.g.,
the secondary controller in a cascade loop is switched away from the cascade mode
(CAS), and the output of the upstream controller (the master controller in the cascade
loop) is then “unconsumed” and left floating.
When the secondary controller is switched back to CAS mode, the output of the
master controller is expected to be the same as the value of the controller output
just before the switch. For this purpose, the primary controller should automatically
initialize itself to align with the secondary controller. Most controller initialization
requirements have been built into the control modules (via back-calculation) on a
modern control system; no special attention is required. However, under some special
circumstances (e.g., with a CALC function block), the controller initialization may
be more complex than the default functionality can support, and extra help is needed
from the process control engineer to complete the controller configuration during
control design. See Sect. 4.3.4 for detailed discussions.

3.6 PID Mode Changes

The different modes of the PID controller provide the flexibility of handling different
application requirements. On the other hand, the transition between different modes
also creates many challenges that require careful considerations during process con-
trol design and implementation.
3.6 PID Mode Changes 109

3.6.1 PID Mode Change in Normal Operations

A detailed discussion of the sequence of events during the mode change can be
quite involved and is outside the scope of a process control book. However, it is
essential to understand how the mode change affects the controller setpoint and
output. Some features are automatically taken care of by the built-in PID function
modules, and some features must be explicitly configured during implementation.
Incorrect configurations can lead to degraded performance, such as bumpy transition
and reset windup. Let us take the simple cascade control loop in Fig. 3.15 as an
example.
The primary controller LC-101 can be in either AUTO or MAN mode, while the
secondary controller FC-101 can assume MAN, AUTO, or CAS mode during normal
operation. The NORMAL operating mode for LC-101 is AUTO, while for FC-101
is CAS. The transition between the different modes triggers a chain of events:
1. MAN to AUTO: When the PID controller LC-101 is switched from MAN mode to
AUTO, the following occurs:
a. If PV Tracking (see Sect. 3.5) is enabled, the SP equals the PV value
when the controller is in MAN mode; thus, the control error is zero. The mode
switch from MAN to AUTO will not create a control error, and thus there will
be no change in output following this mode transition. The OP will remain
unchanged until a non-zero control error is developed. The switch is thus
bumpless. However, the original setpoint value is lost, and the operator needs
to bring the setpoint to the desired value.
b. If PV Tracking is not enabled, the SP retains the previous value when
the controller was in AUTO mode. This setpoint valve is most likely different
from the current process value (most PV values start from zero during startup),
and a significant control error may exist. The PID controller can produce a
sudden change in the output when the controller is switched to AUTO due to
the proportional kick, resulting in a bumpy action. On the other hand, with
PV Tracking disabled, the SP value is preserved during the mode switch,
and the operator does not need to remember and re-set the setpoint after the
mode change.
2. AUTO to MAN: When the PID controller is switched from AUTO mode to MAN, the
controller output (OP) will retain the last good value until the operator manually
adjusts it.
The mode change is more complex for the cascade control loop as two controllers
are involved:
1. AUTO to CAS: When the secondary controller (FC-101) is switched from AUTO
to CAS, the controller setpoint (the cascade setpoint, or CSV, in Fig. 3.9) will
take the value of the current setpoint, and the output of the primary controller
(LC-101) is automatically initialized to the setpoint value of the secondary con-
troller.
110 3 Basic PID Control

After completing a series of handshakes, the mode of the primary controller

changes from INIT (or IMAN on some other control systems) mode to AUTO
mode. Depending on whether PV Tracking is enabled, the primary controller
may have an existing control error between SP and PV, and the mode change may
immediately generate a significant change in the output and send it down to the
secondary controller as a setpoint change.
2. CAS to AUTO: When the secondary controller mode is switched (“shed”) from
CAS to AUTO, the secondary controller is bumpless because the SP will take the
value of the cascade setpoint (CSV); therefore, no transient control error.
The primary PID controller will change to INIT (or IMAN) mode, and the output
is left floating, and its value is dependent on the specific configuration.
The handshake process can be pretty complicated and is not guaranteed to succeed
if the mode change is not allowed or other permissions are not met. Figure 3.16 is a
simplified illustration of the handshake process for mode change from AUTO to CAS
in a DeltaV DCS that involves coordinating the current mode, requested
mode, target mode, and actual mode.
Although this handshake process is the scope of control systems instead of process
control, it is beneficial to have a basic appreciation of the complexities in the actual
mode switching. It is essential to understand how the mode change affects the control
setpoint and output and what initial values are assigned when the new control mode
is in effect.

Fig. 3.16 Handshake process for a PID mode change

3.6 PID Mode Changes 111

3.6.2 PID Modes During Crippled Operations

The crippled mode operation concerns the controller’s reaction to the sudden loss of
the process values (PV) due to, for example, measurement failures. Crippled mode
is a practical scenario that is often neglected in implementation. In case a measure-
ment failure is detected, the following decisions should be considered regarding PID
control response:
1. Controller mode: In the case of BAD PV, should the controller remain in control
(e.g., in AUTO or CAS) or be switched to manual operation (MAN mode)? Usually,
the controller should not be allowed to be in control if the measurement is lost or
cannot be trusted. The controller should be configured to automatically “shed” to
MAN mode when a BAD PV status is received.
2. After the controller sheds to MAN mode, what value should the controller output
assume? There are typically three options, depending on the criticality of the
control loop and the availability of the operator:
a. Retain the last good value: Keeping the last good value is the most common
option and is recommended for operations where operator attention is avail-
able reasonably quickly after alarm notification. It is a safety concern if the
valve position remains fixed for an extended duration.
b. Move to the fail-safe position: Set the valve to its fail-safe position in one
shot (fully close or fully open). This option may be the preferred action
for unmanned or safety-critical operations, although this could potentially
introduce a significant disturbance to operation.
c. Ramp to a safe position: In theory, a better solution is to slowly ramp the
output to its fail-safe position at a pre-defined rate while waiting for operator
intervention. However, most control systems do not support this option out-of-
box, and back-end logic needs to be implemented to handle the ramping of the
output. Therefore this option is not practical or worthwhile for most operations
due to the added complexity in the configuration, except for very critical
operations or equipment. Compressor anti-surge control (see Sect. 11.3) is
an example of this use case.
After the controller sheds to MAN mode, should the controller retain its setpoints
or make the setpoint track the process value? Enabling PV Tracking will help
achieve bumpless transfer when the controller is switched back to AUTO; however,
some control applications such as level control and protective control typically
prefer to retain the setpoint during mode changes. See PV Tracking in Sect. 3.5
for details of PV Tracking configuration.
3. Once the failure is cleared, should the controller be automatically put back to
NORMAL mode by back-end logic or manually by the operator? Automated oper-
ation is undoubtedly desirable, but a necessary condition is that the failure condi-
tion is genuinely resolved. For critical operations, it is still safer to count on the
operator to verify the health of the measurement and make the call.
112 3 Basic PID Control

3.6.3 PID Modes During Shutdown and Startup

During emergency shutdowns (ESD), the controller mode, output, and setpoint
should be set to the correct values or position to ensure a safe shutdown and the
readiness for re-start. Typical considerations include the following:
1. Controller mode: Should the controller be switched to MAN mode or remain in
AUTO?
2. Controller output: Should the controller output be set to the valve’s fail-safe
position, or just retain the last good value?
3. Controller setpoint: Should the controller setpoint be retained or be set to different
values, such as tracking the process value?
Conventionally it is assumed that all the controllers should be put to manual (MAN)
mode during shutdown and switch to normal mode by the operator during or after
startup. For a large plant with an increasingly large number of control loops and fewer
operators, managing the controller mode, output, and setpoint during shutdown and
startup becomes a burden and distraction.
Fortunately, many control loops are safe to remain in their NORMAL mode during
the shutdown and startup process. Consider the example in Fig. 3.17:
1. Flow control loops: Flow control loops (e.g., FC-101) are typically REVERSE
acting. During shutdown with no process flow, the controller would drive the
valve to the fully open position. If an open valve is acceptable during the startup,
then the flow loop can be left in CAS or AUTO mode all the time (except during
BAD PV).
2. Level control loops: Most level control loops (e.g., LC-101) are manipulating a
downstream flow or valve and are thus DIRECT acting. During shutdown with a

Fig. 3.17 PID controller mode during shutdown

3.6 PID Mode Changes 113

low level, the controller automatically drives the flow setpoint or valve position
to zero. So it is typically safe to keep the level control in AUTO mode during
shutdown and startup. For instance, the level-flow-valve cascade loop (LC-101
to FC-101) in Fig. 3.17 can be left in AUTO mode all the time (except with BAD
PV).
3. Protective controllers: For a high-pressure protective controller such as PC-101B
in Fig. 3.17, the SP is higher than PV during the normal operating mode. As a
DIRECT acting controller, the controller output is zero when PV < SP. The flare
valve is fully closed as expected. During shutdown and startup, the pressure is
lower than the normal value, the control valve will remain in the fully closed
position, so it is safe to have the controller in AUTO mode all the time.
However, a low-pressure protective controller must start from MAN mode during
startup and switch to AUTO only after the process value is above the setpoint.
4. Overriding Control: The protective controller in an override control loop follows
the same principle as a standalone protective controller. For high protection, it
is typically safe to remain in AUTO all the time, while for low protection, it is
typically not.
5. Normal regulatory controller: Controller for normal operating control such as
PC-101A in Fig. 3.17 will need to be slowly brought to normal setpoint by the
operator during startup, either in MAN or AUTO mode. So these control loops will
typically be put to MAN during a shutdown. They are the process variables that
require the most attention from the operators during startup, shutdown, and other
mode transitions.
During design, the process control engineer should review each control loop indi-
vidually and specify how the controller should behave during startup, shutdown, and
mode transition. If left unspecified, they will assume the default system behavior,
and the default is often unacceptable.

3.7 Summary

This chapter discusses the three essential requirements for a practical PID imple-
mentation:
1. Versatility to meet different application requirements: Different PID equations
have a direct impact on the control performance (Sect. 3.1). Various controller
modes provide flexibility to perform manual control, automatic control, and super-
visory control (Sect. 3.2).
2. Reliability to ensure safe operations: A vital difference between a simulation
study and actual implementation is in reliability. It is easy to make a PID controller
work, but it is overwhelmingly difficult to make it not fail, especially in a complex
control loop (Chap. 4). Back-calculation is a common approach to monitoring the
health of data flow in a control loop (Sect. 3.4).
114 3 Basic PID Control

3. Scalability to support complex control loops: A complex control loop is comprised

of multiple PID controllers and supporting function blocks. The communication
and inter-operation between the function blocks can become very complicated,
and the proper configuration is critical to operating a control loop (Sect. 3.5).
Many configuration options require in-depth knowledge of the control design,
and only process control can make the correct decision.
Although PID configuration is primarily the responsibility of control systems engi-
neers, many configuration options are determined by operating requirements and
dictated by the process control strategy. They depend on process control to provide
the right decisions or guidance.

References

Åström KJ, Hägglund T (1995) PID controllers: theory, design, and tunings, 2nd edn. Instrument
Society of America
Åström KJ, Hägglund T (2006) Advanced PID control. ISA
Åström KJ, Kumar PR (2014) Control: a perspective. Automatica 50(1):3–43
Brannan C (2018) Rules of thumb for chemical engineers—a manual of quick, accurate solutions
to everyday process engineering problems, 6th edn. Elsevier
Marlin TE (2015) Process control—designing processes and control systems for dynamic perfor-
mance, 2nd edn. McGraw-Hill
Minorsky N (1922) Directional stability and automatically steered bodies. J Am Soc Navig Eng
34:280
Routh EJ (2016) Dynamics of a system of rigid bodies. Hansebooks
Samad T, Bauer M, Bortoff S, Cairano SD, Fagiano L, Odgaard PF, Rhinehart RR, Sanchez-Pena R,
Serbezov A, Ankersen F, Goupil P, Grosman B, Heertjes M, Mareels I, Sosseh R (2020) Industry
engagement with control research: perspective and messages. Ann Rev Control 49:1–14. Open
access
Smith CL (2009) Practical process control—tuning and troubleshooting. Wiley
Turton R, Shaeiwitz JA, Bhattacharya D (2018) Analysis, synthesis, and design of chemical pro-
cesses. Prentice Hall
Visioli A (2001) Tuning of PID controllers with fuzzy logic. IEE Proc—Control Theory Appl
148(1):1–8
Visioli A (2006) Practical PID control. Advances in industrial control. Springer, London
Chapter 4
Complex PID Control Loops

Many practical problems are more complicated than a simple PID controller can han-
dle. Multivariable, constrained, or nonlinear problems require complex PID control
solutions comprising multiple PID controllers and other function blocks working
together in concert (Sect. 4.2). Although PID-based control schemes are surprisingly
versatile in addressing many complex control problems (Sect. 4.1), the design and
implementation can be surprisingly tricky. As there are many ways to achieve the
same control objective with these building blocks, a good understanding of the oper-
ating requirements and adequate knowledge of these function blocks are crucial in
choosing the best complex PID control design. Ensuring the loop integrity is one
critical requirement that relies on in-depth knowledge of the data flow and the con-
trol design (Sect. 4.3). For a complex control solution, simplicity and reliability is
always the top priority in the control design and the best solution is thus always the
one that is fit-for-purpose.

4.1 Applications of Complex PID Control Schemes

Section 2.2 has summarized the commonly available complex control loops and
explained what functionality they provide and when they should be used. These
loops all consist of one or more PID controllers, supported by some auxiliary function
blocks, having a fixed structure and performing a predictable task. We will call them
the standard complex loops. These standard complex loops serve as the building
blocks for more complex control solutions (Smith 2010).

4.1.1 Main Challenges for Complex Control Loops

A PID control loop is straightforward and does what it is supposed to do most of

the time. In practical applications, however, the main challenge is not how to make
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 115
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_4
116 4 Complex PID Control Loops

the PID loop work but how to make the PID loop not fail, especially in a complex
control scheme. From practical experience, more than 90% of the effort is spent on
less than 10% of the possibility of PID loops not working as per design.
We will start with a compressor control example to demonstrate how a complex
control scheme is built with the basic PID controllers and standard complex loops.
This example is used here solely to help the discussions in this chapter on complex
control loops. Thorough coverage of this example is in Sect. 11.3, so there is no need
to delve into the control strategy for now.
A typical process flow of a gas compression unit is shown in Fig. 4.1.
A compressor is a critical piece of equipment in many plants. Its failure often
results in the shutdown of the processing unit or even the entire plant and thus must
be efficiently controlled and reliably protected.
Some compressor variables have very fast dynamic responses, typically in tens to
hundreds of milliseconds, so computation-intensive control schemes such as model
predictive control (MPC) are out of the question. The only viable choice is a complex
PID control scheme. Figure 4.2 illustrates an integrated compressor control scheme
designed for implementation and operation in a standard DCS platform.
As shown in the control schematic, most of the control loops are the standard
complex PID loops discussed in Sect. 2.2. For example,
1. Standalone PID control: PC-004 is a simple PID controller responsible for capac-
ity control.
2. Cascade control: PC-002 to XC-101 to SC-101 is a three-level cascade control
loop.
3. Protective control: PC-001 is a protective controller with a higher setpoint than
PC-002 to protect the suction header against excessively high pressure.
4. Overriding control: Several override controls are implemented in this example,
including PC-002/PC-003, UC-111/PC-112, and JC-101/SC-101. Con-
trol selectors, both high selection and low selection, are extensively used to support
the overriding controls.
5. Split-range control: The output of PC-002 is sent to XC-101 and the two control
valves ASCV-111 and ASCV-121, via a split-range control block XY-101.

Fig. 4.1 A typical gas compression process flow

4.1 Applications of Complex PID Control Schemes 117

Fig. 4.2 A complex control loop: integrated compressor control

6. Fan-out control: The output of PC-002 is sent to the two recycle valves
ASCV-111 and ASCV-121 through a fan-out control scheme.
7. Inferential Properties: The process values (UY111.PV and UY121.PV) for the
two anti-surge controllers (UC-111 and UC-121) are inferential properties, cal-
culated with UY-111 and UY-121, respectively.
This example shows that a PID-based complex control solution can become very
sophisticated. As the complexity increases, selecting the best control scheme and
protecting the loop integrity become the primary challenges:
1. The standard complex PID control loops discussed in Sect. 2.2 are the building
blocks for complex control schemes. For complex control problems, there can be
more than one way to achieve the same control objective, e.g., a split-range control
versus a dual-controller control, a ratio control versus or a feedforward control.
Each scheme has its pros and cons and may be appropriate for one control problem
but not for others. A sound control solution requires a deep understanding of the
process dynamics, operating requirements, and the functionality of the control
function blocks. A good design and a bad design may all work (to some extent),
but the difference is in the control performance, with a last-lasting impact on the
operation.
2. The data flow in a simple PID control loop is relatively simple but still has many
practical complexities affecting the control performance (see Sect. 3.2). The
information flow can become overwhelmingly complicated in a complex con-
trol scheme comprising multiple PID controllers and supporting function blocks.
For example, with control selectors (override control) and splitters (cascade and
fan-out control), the possibility of reset windup and bumpy transition is signifi-
cantly higher. Protecting the integrity of the data flow in process control design
requires significant knowledge and experience. The complexity is also a consider-
118 4 Complex PID Control Loops

able hurdle for the operators and maintenance personnel and contributes to many
sub-optimal operation problems and even incidents.
Complex PID control schemes can be used to address multivariable interactions,
constraint handling, preferential control, optimizing control, and nonlinear control.
However, there is a significant overlap between the applicability of complex PID con-
trol schemes and model predictive control (MPC). If a control PID scheme becomes
overly complicated, MPC should be considered, especially if the process dynamics
also exhibit heavy interactions, extensive delays, and stringent constraints. Model
predictive control is discussed in Chap. 5.

4.1.2 Dealing with Multivariable Interactions

Complex control loops are designed to solve complex control problems that consist of
multiple control targets with multiple control handles. The control targets and control
handles may exhibit a strong correlation and interaction and must be considered
altogether.

Degree of Freedom Requirement

A control handle is one degree of freedom (DOF) and can control one target. All the
control targets and control handles must be paired up in a complex control scheme.
In other words, the number of control targets must match the number of control
handles to leave no DOF unconsumed. Unconsumed control handles will eventually
be driven to saturation and cause degradation in control performance.
It is common to find that the number of control targets does not agree with the
number of control handles in a complex control problem. In this case, the standard
complex control loops in Sect. 2.2 can be used to adjust the balance of DOFs by
“merging” or “splitting” the control targets or control handles. For example, a fan-
out control loop has a single target manipulated by two or more control handles. An
override control loop has one control handle but multiple control targets, although
only one target is controlled at any given time:
1. Reduce the number of controlled variables by correlating the targets: Override
control (Sect. 2.2.7) is the most popular way of “reducing” the number of control
targets. For example, many override control schemes are utilized in the com-
pressor control example in Fig. 4.2. The two controlled variables, SC-101 and
JC-101, share one control handle, the gearbox speed. The compressor suction
pressure PC-002 and discharge pressure PC-003 are “merged” via a selector
and consumes one DOF. Ratio control (Sect. 2.2.6) “ties” two targets together
and thus manipulates only one control handle. A selective control selects one
target at a time and thus is counted as only one controlled variable in terms of
DOF requirement.
4.1 Applications of Complex PID Control Schemes 119

Another option is to use inferred variables. For example, instead of controlling

both temperature and pressure, a pressure-compensated temperature reduces the
number of targets from two to one (see Sect. 10.4). In Fig. 4.2, compressor surge is
physically related to the suction pressure PT-111, discharge pressure PT-112,
and discharge flow FT-111. Instead of controlling all three variables, a sim-
ple calculation is performed in UY-111 to relate the three variables to a single
variable, the surge indicator. That is, one controlled variable is used instead of
three.
2. Reduce the number of manipulated variables: Techniques such as selective con-
trol (Sect. 2.2.8), fan-out control (Sect. 2.2.3), split-range control (Sect. 2.2.1),
and feedforward control (Sect. 2.2.5) can be utilized to change the DOF provided
by the manipulated variables. For example, the compressor suction pressure in
Fig. 4.2 can be influenced by the compressor speed SC-101 and the two recy-
cle valves (ASCV-111 and ASCV-121). A split-range control block XY-111
“links” the two types of final control elements and effectively treats them as one
DOF.
3. Increase the number of controlled variables by “splitting” or “duplicating” the
measurements: Cascade control uses one control handle to control two tar-
gets (Sect. 2.2.4). Dual-controller control duplicates the control loop but with
different setpoints (Sect. 2.2.2). For example, in Fig. 4.2, the compressor suc-
tion pressure is “split” into two variables and controlled with two controllers,
PC-002 and PC-001, and thus consumes two DOFs. The normal regulatory con-
troller PC-002 treats the suction pressure as a regulatory control target (capacity
control), manipulated by the control speed. The overriding protective controller
PC-001 treats the suction pressure as a constrained variable, controlled with the
flare valve. Similarly, the split-range control block XY-111 “splits” the selector
output into two signals, one goes to the load controller XC-101, and the other is
fanned out to the two anti-surge control valves (ASCV-111 and ASCV-121).
4. Increase the number of manipulated variables: The number of control handles is
usually fixed once the process engineering is complete. It is typically not possible
to increase the number of manipulated variables to meet the DOF requirement
without process change.
Table 4.1 summarizes the options of using complex PID control schemes to adjust
the number of DOFs.
Similarly, when troubleshooting an existing control scheme, it is essential to val-
idate that all the controlled variables and manipulated variables are paired up, and
thus, all the DOFs are consumed.

Interactions Among Process Variables

One of the key challenges for multivariable control problems is the interaction
between the process variables. One manipulated variable may affect multiple con-
trolled variables, and conversely, one control target may be impacted by more than
120 4 Complex PID Control Loops

Table 4.1 Changing number of degree of freedom with complex control loops
Complex PID loop To increase targets To reduce targets To reduce handles Reference
Split-range control  Section 2.2.1
Dual-controller control  Section 2.2.2
Fan-out control  Section 2.2.3
Cascade control  Section 2.2.4
Feedforward control  Section 2.2.5
Ratio control  Section 2.2.6
Override control  Section 2.2.7
Selective control   Section 2.2.8
Model predictive control    Section 2.3.4

one control handle or disturbances. Complex control loops such as feedforward

control, ratio control, and decoupling control provide some capabilities to address
multivariable interactions; However, they all require a good understanding of the
process dynamics:
1. Feedforward control: A controlled variable is controlled by a manipulated vari-
able, subject to noises and disturbances that can be measurable or unmeasurable.
Feedforward control is a common approach to deal with measurable disturbances.
Feedforward control is an open-loop control and must work along with a feedback
control loop (Stephanopoulos 1984) in practical applications. Figure 4.3 shows
the data flow of both the feedback and feedforward channels. Feedforward control
“intercepts” the change in the disturbance variable and adjusts the manipulated
variable in anticipation to “cancel out” the effect of the disturbance.
It is vital to know both the feedforward and feedback dynamics to achieve “per-
fect” cancellation. If we use G d (s) to represent the disturbance dynamics and
G p (s) for the process dynamics, the feedforward control G ff (s) is given by

Fig. 4.3 Block diagram of feedforward control

4.1 Applications of Complex PID Control Schemes 121

G d (s) Ud (s) + G p (s) G ff (s) Ud (s) = 0 → G ff (s) = −G d (s)/G p (s) (4.1)

where Ud (s) is the disturbance input.

In most cases, we can simplify both the process dynamics and disturbance dynam-
ics into gain, dead time, and time constant (i.e., a first-order process model). In
that case, feedforward control can be interpreted as follows:
a Gain: The process gain and the disturbance gain are related by K d + K ff · K p =
0, which gives K ff = −K d /K p , with K d being the gain of disturbance channel,
K p being the gain of process channel, and K ff being the gain for feedforward
compensation.
b Dead Time: A necessary condition for feedforward control is that the dead time
from the disturbance variable to the controlled variable must be greater than or
equal to that from the manipulated variable to the controlled variable. That is,
θd ≥ θ p , where θd is the delay of the disturbance channel, and θ p the delay of
the process channel. The feedforward control must have a time delay equal to
θff = θd − θ p .
c Time constant: For best performance, the feedforward control should provide
lead/lag compensation for the difference in the dynamics of the two channels.
The calculation of the feedforward can be either static or dynamic. Static compen-
sation involves only a gain adjustment to the disturbance when being added to the
controller output. While a dynamic compensation needs also to consider delay
and lead/lag compensation. For example, assume the process dynamics G p (s)
and disturbance dynamics G d (s) are given by

Kp Kd
G p (s) = e−θ p s , G d (s) = e−θd s (4.2)
τps + 1 τd s + 1

then the feedforward transfer function is given by

G d (s) K d τ p s + 1 θd −θ p
G ff (s) = − =− e (4.3)
G p (s) K p τd s + 1

with θd ≥ θ p . If the process dynamics or disturbance dynamics have more com-

plex characteristics than first-order-plus- dead-time, the feedforward would be
more complicated. Due to the time-varying nature of the dynamics, it is unreal-
istic to expect “perfect” dynamic compensation with a fixed lead-lag term. As a
result, most feedforward control in practice only implements the gain and dead-
time calculation.
Feedforward control is typically implemented in its incremental form for better
stability. The delta change in the feedforward variable is used instead of the abso-
lute value to ensure bumpless calculation in case of a bad value in the feedforward
variable. That is

u(k) = u(k − 1) + u(k) + K FF [yFF (k) − yFF (k − 1)] (4.4)

122 4 Complex PID Control Loops

where u(k) is the new output, u(k − 1) is the previous output, u(k) is the
feedback control output, and yFF (k) is the feedforward variable.
The feedforward output is typically added to the feedback control output (additive
feedforward), as shown in Eq. 4.4. There is also the option of multiplying the
feedforward value with the feedback output (multiplicative feedforward), but it
is rarely needed.
Feedforward control is based on the assumption that the dynamics of both the
feedback and feedforward channels are known and stay relatively constant over
time. This assumption is rarely valid. Inaccurate compensation can sometimes do
more harm than good, so feedforward control should be used with discretion.
2. Ratio control: Ratio control is another example that two process variables can
impact the controlled variable. Ratio control forces the secondary variable to
change proportionally with a primary variable (the wild variable).
Continuing with the example in Fig. 2.20, the internal ratio calculation is usually
given as follows:

FC102.SP = FFC103.OP
= FFC103.SP · FI101.PV. (4.5)

The primary variable (the wild variable) is the process value for the ratio controller,
and the desired ratio value is the setpoint. The output of the ratio controller is the
setpoint for the secondary controller (the PID controller). Direct ratio calculation
should not be used to avoid divide-by-zero errors.
Ratio control is similar to a static feedforward control with a fixed gain term to
relate the two process variables. They both are open-loop control and must be
used along with a feedback control loop. If there are significant differences in the
dynamics between the two variables entering the ratio calculation, then lead/lag
and dead-time compensation may need to be added for the primary measurement
(FI-101.PV in this case). For example, if the primary flow FI-101 in Fig. 2.20
has a two-minute transport delay to the mixer, while the manipulated variable
FC-103 has a one-minute delay, then flow FI-101 should be delayed by one
minute before entering the ratio controller FFC-102.
Ratio control may appear similar in configuration to cascade control. However,
the difference is that the ratio controller does not have its dynamics and thus no
tuning requirement. Internally, it is a simple scalar calculation as in Eq. 4.5.

Decoupling control

Decoupling control is an early effort to address the multivariable interactions among

input and output variables (Gordon 2005; Shinskey 1994). Figure 4.4 shows an exam-
ple of strong interactions between level (LC-101) and temperature (TC-101).
Adjusting temperature control valve TCV-101 affects both the temperature
(TC-101) and the flow FC-101 (thus the level LC-101). Similarly, controlling
4.1 Applications of Complex PID Control Schemes 123

Fig. 4.4 Example of decoupling control

the level LC-101 by manipulating the flow control valve FCV-101 would also
affect the temperature TC-101.
With decoupling control, the level and temperature can be controlled indepen-
dently of each other. When the flow or temperature needs to be adjusted, the decou-
pling control scheme simultaneously moves the flow and temperature control valves.
One controlled variable will change as desired, and the other controlled variable
remains unaffected. Consider the following two scenarios:
1. The flow rate needs to be adjusted to change the level, but the temperature must
remain constant. In this case, the two valves would move together so that the total
flow change will meet the level control requirement, but the ratio of the two flows
remains constant so that the temperature will remain unchanged.
2. The temperature needs to be adjusted, but the total flow rate must remain constant
to maintain the same level. For this to happen, the two valves will move in the
opposite direction such that the total flow remains the same, but the ratio between
the two flows will drive the temperature to the new setpoint.
For a control valve, the flow through the valve is proportional to the valve coeffi-
√ fluid density ρ, and the differential pressure across the valve P. That is,
cient Cv , the
FC ∝ Cv P ρ. If the two control valves are of identical design, i.e., Cv,1 = Cv,2 ,
and the pressure drop across the two valves are the same, then the two scenarios
discussed above can be described as

FCV + TCV
FC.OP = (4.6)
2
TCV
TC.OP = · 100%. (4.7)
FCV + TCV
124 4 Complex PID Control Loops

FCV and TCV can be easily solved to yield

FCV = 2 · FC.OP · (1-TC.OP/100) (4.8)

TCV = 2 · TC.OP · (FC.OP/100). (4.9)

The calculations can be interpreted as follows:

1. The normal control position is FC.OP = 50% and TC.OP = 50%, leading to FCV
= 50% and TCV = 50%.
2. Suppose that the temperature control needs to increase its output from TC.OP =
50 to 75%. To keep the flow rate unchanged at FC.OP = 50%, the two valves will
need to move in opposite directions, to FCV = 25% and TCV = 75%.
3. Similarly, assume that FC.OP = 50% and TC.OP = 50% and a temperature
setpoint change require that FC.OP to decrease to 25%, the control scheme will
move both valves, i.e., to FCV = 75% and TCV = 25%.
4. Assume the temperature is stable and a level setpoint change requires that the flow
controller output FC.OP to change from 50 to 75%. In this case, both valves need
to open at the same time and in proportion to keep the temperature unaffected.
The final valve positions will be FCV = 75% and TCV = 75%.
5. Similarly, assume that FC.OP = 50% and TC.OP = 50% and the level controller
requires a change of FC.OP = 50% → 25%. The decoupling control scheme
will move both valves simultaneously and in proportion to keep the temperature
unaffected. The final valve positions will be FCV = 25% and TCV = 25%.
In reality, the two valves are usually of different sizes. The above formulas need
to be expanded
√ to accommodate this general scenario. For simplicity, let us define
C = Cv P. The two scenarios described in Eqs. 4.8 and 4.9 become

C1 FCV + C2 TCV
FC.OP = (4.10)
C1 + C2
C2 TCV
TC.OP = · 100%. (4.11)
C1 FCV + C2 TCV

To control the temperature but maintain a constant flow, the total valve changes shall
remain the same. Similarly, to maintain the temperature constant, the relative ratio
of the temperature control valve versus the total valve should remain constant. For
convenience, define

C2 Cv,2 P2
K = = (4.12)
C1 Cv,1 P1

and solve for FCV and TCV,

4.1 Applications of Complex PID Control Schemes 125
 
TC.OP
FCV = FC.OP · (1 + K ) · 1 − (4.13)
100
 
1 FC.OP
TCV = TC.OP · 1 + · . (4.14)
K 100

Decoupling control is in essence feedforward control. The disturbance to be

rejected for one controlled variable is the normal process reaction to another con-
troller. Therefore, it is crucial to have the exact dynamic cause and effect relationships
between the valves and the controlled variables to decouple the interacting variables.
The example above is static decoupling control and already appears complex to
understand and implement. To dynamically decouple the multiple input and output
variables, the control design would become much more complicated and is simply
impractical for baselayer control. For this very reason, decoupling control has never
received widespread acceptance in practice. On the other hand, model predictive
control (MPC) has provided a more elegant solution.

4.1.3 Handling Constraints

Contrary to popular belief, PID-based baselayer control can also handle constraints
in the form of protective control. A normal regulatory control is responsible for
maintaining the operating point during normal operation. When the operating point
deviates from the target value by too much, the protective controller kicks in to keep
the operating point within the operating envelope (see Sect. 1.3.5). During normal
operating conditions, the protective control remains inactive.
There are three types of protective control implementations: standalone, dual-
controller, and overriding.
1. Standalone protective control: A standalone protective control is a standard PID
controller, except that the controller is designed to be inactive during normal
operation. This is achieved by setting the setpoint higher (for high protective
control) or lower (for low protective control) than the normal operating value (e.g.,
the design value). During normal operations, the controller output is saturated
to fully closed (high protective control) or fully open (low protective control)
position. Only when the normal regulatory control can not maintain the setpoint
and the process value deviates from the setpoint to reach the protective control
setpoint (the constraint limit) will the protective controller “wake up” and take
action.
Figure 4.5 shows a control scheme for a centrifugal pump. Centrifugal pump has
a minimum flow requirement, below which the pump may become unstable.
Flow controller FC-102 acts as a minimum flow recycle controller. Its purpose
is to open up the recycle valve to re-circulate a portion of the flow back to the
tank to maintain a minimum flow through the pump in case the pump flow drops
126 4 Complex PID Control Loops

Fig. 4.5 Example of protective control

below the minimum flow threshold. See Sect. 11.2 for a detailed discussion on
pump control.
2. Dual-controller protective control (Sect. 2.2.2): Each controller in the dual-
controller configuration is similar to a standalone protective control in that each
has its own control handle and consumes one DOF. However, unlike a standalone
protective control, dual controllers work in pairs on the same process variable:
one is the primary controller for normal regulatory control. The other is for con-
strained control to protect the process value from going over the constraint limit.
The two pressure controllers, PC-101 and PC-102 in Fig. 4.2, or PC-101A
and PC-101B in Fig. 2.16, are typical dual pressure controller schemes.
3. Overriding protective control (Sect. 2.2.7): An overriding protective controller
shares the control handle with the normal regulatory control with the control
selector. It has a more tolerant setpoint so that the controller remains dormant and
its output is not selected during normal operation.
The pressure controller PC-103 in Fig. 4.5 is an example of an overriding pro-
tective controller. It protects the downstream piping or equipment by restricting
the downstream pressure to a safe limit. If the pressure rises above the setpoint,
the pressure control will override the flow controller and force the control valve
LCV-101 to open less.
In the integrated compressor control example in Fig. 4.2, there are several protec-
tive control loops. For example,
1. PC-001 and PC-102 constitute a dual-controller protective control scheme.
The controlled variable is the compressor suction pressure. The normal regula-
tory control loop for the suction pressure is PC-002. The protective controller
PC-001 remains inactive if the suction pressure is within normal range. If the
pressure rises above the protective setpoint, the protective controller PC-001
opens the flare valve to release the excess gas to the flare.
2. PC-003 at the compressor discharge is an overriding protective control. When
the discharge pressure goes too high, it overrides the normal regulatory controller
PC-002 and slows down the compressor to reduce the discharge pressure.
4.1 Applications of Complex PID Control Schemes 127

3. UC-111 and UC-121 are two overriding controllers. They share the recycle valve
with another two PID controllers. Whichever controller reaches its constraint
limit, it would open the recycle valve through the high selector, allowing the
highest demanding request to pass through.
As demonstrated by Fig. 4.2, protective control can become highly complicated
and require careful design for the overall control scheme to function in concert. Some
crucial considerations include the following:
1. Rationalization of the normal control setpoint, the protective control setpoint, and
the safeguarding limits so that they are adequately spaced to allow sufficient time
for the next level of protection or the operator to respond.
2. Controller tuning: Protective controllers are typically tuned more aggressive for
quick response than the normal regulatory control loop.

4.1.4 Optimizing Flow Control

Baselayer PID control schemes can achieve a certain degree of profit maximiza-
tion through the creative use of complex PID schemes. Preferential control, flow
balancing, and limit pushing are several examples.

Preferential Flow Control

A commonly encountered control requirement is preferential flow. When a material

or energy stream feeds multiple consumers, there is often a requirement to prioritize
the flow in case the supply is insufficient. For example, in a water processing unit,
the produced water from the wells is preferably sent to water lifting wells first. Any
excess should then be sent to the injection wells. If still with excess, the remaining
should go to deep water disposal (DWD) wells.
To achieve this type of preferential flow arrangement, the first approach coming
to mind is a split-range control scheme (Sect. 2.2.1). The high-priority flow path is
put at the lowest leg of the split range, while the least-preferred flow path should be
in the upper leg. In theory, the split-range control can be 1 to n split, where n is 2 or
more, but it is rare to see more than two splits due to controllability.
In the example in Fig. 2.15, the process gas is expected to be sent to the downstream
compressor as the normal process flow. However, if excessive supply causes the
pressure to go higher than a limit, the excess gas will be sent to flare. Dual-controller
control can also achieve the same preferential control, see Fig. 2.16. The downside
is that the setpoints of the two controllers must be set sufficiently apart to avoid
interfering with the normal regulatory control.
Another application scenario is multiple flows differentiated with a bias term. For
example, in a fan-out control, the same output is sent to multiple control elements.
Although this equal split is acceptable for most applications, there are scenarios where
128 4 Complex PID Control Loops

the different control elements need to receive different flows. A common approach
is to utilize a gain/bias function block to either apply a different ratio (gain) or a
fixed offset (bias) to the output received by each control element. As an example,
a pressure control output is sent to three flow controllers at the wellhead of three
water injection wells. If one of the wells has a higher capacity than the others, an
offset (bias) can be added to the output to cause a differentiated flow for this well. In
other words, instead of 40, 40, 40%, the output is distributed as 35, 50, and 35% after
applying a 15% bias term to the second well. The feedback mechanism automatically
determines the output value received by each flow control, but the bias is maintained.
Note that the gain/bias function block typically does not support back calculation
and may need a calculation block to provide the back-calculation information to the
primary controller explicitly.

Flow Balancing Control

Related to preferential flow control is an application scenario of flow balancing

control. Two parallel streams may need to maintain a specific ratio in their flow rates,
which can generally be achieved through a simple ratio control scheme. However,
an additional requirement is that the pressure loss across the control valves shall
be as low as possible. Permanent pressure loss results whenever a piece of device,
equipment, or pipe is added to a flow system. This pressure loss makes the pump or
compressor work harder to generate the same flow in the established system. Every
bit of pressure loss equals extra energy used (electricity, steam, or natural gas) to
pump or compress the fluid, which translates into more money spent to maintain
operations.
Example 4.1 Flow balancing control in a water treatment process. In the simplified
flow diagram shown in Fig. 4.6, unprocessed water passes through two de-sanding
vessels to remove the entrained sands.

Fig. 4.6 A process with flow balancing requirements

4.1 Applications of Complex PID Control Schemes 129

The two de-sanders are of different capacity, and thus a flow ratio of 3:2 between
FI101A and FI101B is required to achieve the best efficacy and lifespan. At the
same time, the pressure losses across the flow control valves are expected to be as
low as possible to meet the pressure requirements by the de-sanders and downstream
units because a lower pressure loss means less demand on the pump.
A conventional design would lead to a ratio control with a split-range logic on the
valve opening, as shown in Fig. 4.7a (left). As the split-ratio controller output goes
from 0 to 100%, flow controller valve FCV101A closes from 100 to 0% while flow
controller FCV101B opens from 0 to 100%. The normal operating area is around
50% for both valves for the desired controllable range.
One obvious flaw with this design is that at 50% of openings, significant pressure
losses across the valves will incur, which is not acceptable due to the capacity limit
of the pump.
A different split-range control is recommended, as shown in Fig. 4.7b. With this
improved design, both valves have an opening near 100% at the design condition,
thus resulting in minimum pressure losses to the process. Note that the pipe sizing
is selected such that when both valves are fully open, the flow split-ratio is at the
vicinity of the desired value.
A control design is provided in Fig. 4.8. The ratio control block FFC101 calculates
the flow ratio. The setpoint of the ratio controller is the desired split ratio, and the
output of the ratio controller provides the setpoint for flow controller FC101, whose
process variable is the flow rate of stream A. The flow controller FC101 manipulates
the control valve FCV101A to achieve the desired flow ratio. The openings of control
valves FCV101A and FCV101B are calculated by FY101 based on the split logic
in Table 4.2.

Valve Position Control

Valve position control (VPC) is a simple and effective optimization scheme that can
be achieved with standard PID controllers. The opening of a control valve can vary

Fig. 4.7 Split point for flow balancing control

130 4 Complex PID Control Loops

Fig. 4.8 Split-range design

for flow balancing control

Table 4.2 Flow balancing Controller output Valve openings

control logic
FC101.OP (%) FCV101A (%) FCV101B (%)
0–50 100 0–100
50 100 100
50–100 100–0 100

between 0 and 100%, typically driven by a controller. Except for full-bore valves,
all control valves will cause pressure losses even when the valve is at 100% open.
Besides, all valves have a preferred range of operation, such as 10–80% for best
controllability. There are thus incentives to limit the valve position to a range that
can either save energy or improve controllability.
Valve position control (Luyben and Lyuben 1997; Shinskey 1978) is a smart
approach to optimize the valve operation (subject to other constraints). The basic
idea of valve position control is quite simple. Assume a variable somewhere else is
available to vary or “float” within a range. Adjusting this variable can cause the valve
position to shift without affecting the main control performance. A PID control loop
can then be implemented to use this free variable to drive the valve to the desired
position. This move must be very slow and aims to affect the steady-state condition
only. In other words, the valve is primarily used for dynamic control; and the valve
position control tries to influence the steady-state valve position to reduce energy
loss or improve controllability.
Let us use an example to explain the concept. Figure 4.9 is a pumping example
where the pump discharge pressure provides the water injection header pressure for
multiple (two are shown) water injection wells. The header pressure is the controlled
variable. The manipulated variables include the two flow control loops that drive the
two flow control valves.
4.1 Applications of Complex PID Control Schemes 131

Fig. 4.9 Valve position

control example

The valve flow is determined by the valve opening and differential pressure across
the valve. A higher setpoint will result in a higher discharge pressure and lower valve
openings, which means more pressure loss will occur across the control valves. A
lower discharge pressure will cause the flow control valve to open more. In other
words, some of the energy that the pump consumes to raise the discharge pressure is
wasted at the control valve as pressure loss. It has some economic benefits to reduce
the energy loss at the valve by shifting the steady-state valve position toward the
fully open direction without affecting the controllability.
The setpoint of the pump discharge header pressure is an operator input and can
be regarded as a free variable or an extra DOF. The setpoint can vary slowly within
a range without negatively impacting the dynamic flow control performance. The
valve position control is illustrated in Fig. 4.10. The control objective is to maintain
the steady-state valve position (the larger of the two valves) at around 70%.
Another example is a control scheme with two parallel valves, a larger one for
coarse control and a smaller valve for fine control. The smaller valve is often found
in saturation due to the smaller control range. The smaller valve may remain at the
saturated position for a long time until the larger valve is adjusted (often manually)
or the process condition changes back. A saturated valve renders the variable out of
control and is thus undesirable. In this case, a simple PID controller can be added that

Fig. 4.10 Valve position

control scheme
132 4 Complex PID Control Loops

takes the position of the smaller valve as the controlled variable and the position of
the larger valve as the manipulated variable. For example, with a setpoint of 50%, the
fine control with the smaller valve is achieved by varying the valve position around
50%. Suppose the process condition drives the valve position away from 50%. In
that case, the valve position control slowly adjusts the larger valve and causes the
position of the smaller valve to return to the vicinity of 50% gradually.
In summary, when considering valve position control, the following characteristics
should be taken into account:
1. Valve position control concerns the steady-state or average position of the valve.
The primary purpose of the control valve is to control other variables, and thus
the valve is meant to move freely for that purpose. Value position control intends
to drive the steady-state or average valve position to the desired range. For this
reason, the valve position control action must be very slow and smooth to not
hamper or interfere with the normal dynamic control function of the valve. An
I-only PID control algorithm or a PI algorithm with very low gain are typically
used for valve position control.
2. A clear cause and effect relationship must exist. The basis for valve position
control is that a free variable that has a deterministic cause and effect relationship
with the valve position is available. This free variable is typically another DOF
reserved for the operators, such as the setpoint of another controller. It serves as
the manipulated variable of the valve position controller to influence the valve
position.
The valve position control provides some incremental benefits beyond stable and
efficient control. However, it should be used with discretion. If misapplied, it can
also do much harm.

Limit Pushing

Process control improves the production efficiency by pushing the operating point
closer to the operating limits, typically the more profitable operating region. Limit
pushing is one of the primary reasons that model predictive control (MPC) has been
so successful (see Sect. 2.3.4 and Chap. 5).
In fact, PID-based baselayer control can be designed to push limits, although not
as elegant and versatile as MPC. For instance, the gas compression process in Fig. 4.2
is a supply-driven operation; the suction pressure controller PC-002 is the capacity
control responsible for regulating the compressor throughput. A lower pressure set-
point pushes the compressor to run faster and thus deliver a higher throughput. From
process analysis, it is known that the motor power JC-101 sometimes limits the
compression capacity. To avoid tripping the motor, the operator often sets the pres-
sure setpoint sufficiently high to not exceed the maximum motor power limit. This
high setpoint makes the operation overly conservative. A simple remedy is to add a
protective controller, JC-101, on the motor power. The pressure control PC-002,
4.1 Applications of Complex PID Control Schemes 133

with a setpoint slightly lower than the normal value, forces the compressor to run at
the highest speed possible within the motor power limit.
For limit pushing, a vital requirement is that a reliable “stopper” must exist to limit
how far the pusher can push. In this compressor example, a lower setpoint makes the
pressure controller PC-002 a capacity “pusher”. At the same time, the motor power
JC-101 is a reliable “stopper” to prevent the pusher from pushing too far and too
hard. Limit pushing is a standard feature in MPC that is discussed in Chap. 5.

4.1.5 Tackling Nonlinearity

Nonlinear behavior can be caused by various operating conditions and can also be
caused by control structure changes such as by a switch or selector. PID controller
has a fixed set of tuning parameters for the control of fixed loop dynamics. In a
complex control scheme, changes in the data flow may cause drastic changes in the
loop dynamics and render the existing PID tunings inadequate. Online modification
to the controller is one way to adapt to the changed loop dynamics. The modification
can be as simple as gain scheduling or as sophisticated as control structure change.

Gain Scheduling

Gain scheduling is a common approach for handling some specific types of process
nonlinearity. Gain scheduling applies a different controller gain to the running con-
troller as the operation transitions from one condition to another. In this way, the
controller gain more closely aligns with the changed loop dynamics.
Fan-out control (see Fig. 2.17) is a typical application where gain scheduling may
drastically improve control performance. For example, a pressure controller on the
injection manifold in a water injection process distributes the high-pressure water to
a dozen injection wells. A fan-out control loop is used to manipulate a dozen injection
flow controllers, one for each injection well. Any injection wells can be freely taken
out of, or put back in, service at the operator’s discretion based on operation needs.
Although the process dynamics remain the same, the number of wells online can
drastically change the loop dynamics and renders the tuning inadequate.
The unpredictable change in the number of secondary elements poses the follow-
ing challenges:
1. How to tune the pressure controller?
2. How should the controller respond to the change in the number of flow controllers
in cascade?
The recommendation is to tune the controller assuming all the flow controllers are in
cascade control. Suppose the gain is K c and the number of flow controllers is N , and
the number of flow controllers in cascade is n, then the controller gain is given by
N /n · K c . This approach also assumes that the dynamics of each injection well are
134 4 Complex PID Control Loops

identical, and there is an online mechanism to count the number of flow controllers
(n) in the cascade.
The controllers that are taken out of cascade control may have a zero flow (out
of service) or a fixed flow (in MAN). Either way, the loop dynamics for the pressure
controller are not affected; only the steady-state output of the pressure control will
be shifted.
Gain scheduling for nonlinear process behavior such as asymmetric response
(different process gain in the opposite direction) is more complicated. It needs con-
siderably more accurate modeling of the process dynamics and the benefits may not
be justifiable for the incremental performance improvement.

Gap Control

The built-in control functionalities for nonlinear PID control typically include PID
gap control and quadratic gain. There are many variations in different control sys-
tems. Gap control adjusts the PID gain in a pre-defined fashion based on the magni-
tude of the control error, as illustrated in Fig. 4.11.
A gap control or quadratic gain function produces a smaller gain when the control
error is small (the process valve is near the setpoint). A larger controller gain is
produced when the error is significant and needs more aggressive action. That is, for
simple gap control

Kc outside the gap: err < −gap/2 or err > gap/2
K gap = (4.15)
k · K c inside the gap: − gap/2 ≤ err ≤ gap/2.

We can assume that the quadratic gain function has the following format:

K gap = a · err2 + b. (4.16)

Fig. 4.11 Gain scheduling: gap control and quadratic gain

4.1 Applications of Complex PID Control Schemes 135

Assume that the gain is k · K c when the error is zero, and the gain is the regular
gain K c outside the gap; it is easy to find that the quadratic function is related to gap
control as below:
⎧
⎨Kc outside the gap
K gap = 4 (4.17)
⎩(1 − k) K c · err + k · K c inside the gap.
2
gap2

The gain scheduling function is typically built in the PID module, with the gap
width and scaling factor k as the tuning parameters. The default value for the scaling
factor is typically 0.25. That is, the control action is four times less aggressive inside
the gap. The selection of the gap size depends on the nonlinear characteristics and the
control objective to achieve. For example, the objective of surge tank level control is
to maintain a stable outlet flow at the cost of fluctuations in the tank level. The level
fluctuation is not a concern as long as it stays inside a specific range. In this case,
a gap control can be implemented with a gap size of, for example, 40%. Assume
the level control setpoint is 50%; the controller has a smaller gain when the level is
between 30 and 70% with slow control actions. The gain automatically switches to
the regular value for more aggressive action if the level goes above 70% or below
30%.1
Gain scheduling may effectively improve the control performance, but the
approach is highly empirical and typically not based on rigorous process information;
and thus, the tuning parameters are always approximate.

4.1.6 Rationalizing Control Scheme

There is often more than one way to achieve the same control objective, but the control
performance may be drastically different. The difference between a good and not-so-
good control design is sometimes subtle. The impact may not be make-or-break but
rather between optimal and sub-optimal. A sub-optimal design may deliver sub-par
performance over its entire life cycle and is often a maintenance burden as well.

Nested Loop and Cascade Loop

We will use a nested control loop to demonstrate that the control design can be pretty
creative once the basic principle is well understood. It needs a decent process control
“sense” (mastery skills) to turn a mediocre (but acceptable) control solution into a
great solution.

1On the other hand, the level is inherently nonlinear with the flow for a horizontal cylindrical vessel.
The process gain is the smallest at 50% and becomes higher when the level moves more away from
50%. With a fixed PID gain, the control performance is equivalent to a nonlinear controller with
quadratic gain.
136 4 Complex PID Control Loops

Fig. 4.12 Tank level control: simple cascade control

Figure 4.12 shows a water processing process where the produced water from
wells is sent to the free water knockout (FWKO) tank and then flows to a gas/liquid
separator to remove the entrained gas in the liquid. The water is pumped out and sent
downstream for further processing. The production rate (the capacity) is regulated
by the flow controller at the pump discharge line.
The operating requirement is to maintain the two vessel levels by manipulating
the two flow control valves. Level and flow measurements are available as needed.
For the FWKO tank, the level is the controlled variable, and the outlet flow valve
LCV-101 is the manipulated variable. For the separator, due to the pump’s nonlinear
dynamics, a cascade level (LC-102) to flow (FC-101) control loop is utilized to
improve the level control. The original design is illustrated in Fig. 4.12.
However, a significant challenge is that the separator has a much smaller capacity
than the FWKO does. The fluctuations in the FWKO outlet flow can easily overwhelm
the separator level controller LC-102. During the first few years of operation, the
separator was frequently tripped on a high level, resulting in multiple shutdowns.
To prevent the high-level trips, a second level controller LC-103 off the same level
transmitter on the separator was implemented as an overriding protective control on
FWKO tank level controller LC-101. See Fig. 4.13.
In case the separator level goes too high, the override controller LC-103 would
override the FKWO level control output and reduce the inlet flow to the separator.
This design provides the desired protection against high levels in the separator and
prevented high-level trips.
Although improved over the original scheme, this control scheme still has serious
flaws. In addition to the added complexity introduced by the protective controller
LC-103 and the overriding mechanics through the selector, the scheme may poten-
tially cause a high FWKO tank level. Once overridden by LC-103, the FWKO tank
level controller LC-101 is out of control. Even though the capacity of the FWKO
tank is large to allow some grace time, there is no protection against the level going
overly high to trip the tank in the design. As will be explained in Chap. 6, the fun-
4.1 Applications of Complex PID Control Schemes 137

Fig. 4.13 Tank level control: with protective control

damental flaw in this design is that the process is supply-driven, and the original
control solution is based on a supply-driven operating model. Introducing the pro-
tective controller LC-103 can force the operation into a demand-driven operation
without a swing stream at the FWKO.
Since the separator has a much smaller capacity (and thus much faster dynam-
ics) than the FWKO, a nested control scheme was proposed to improve the control
reliability, as shown in Fig. 4.14.
The faster separator level control is the inner loop, while the slower FWKO level
control acts as the outer loop. The separator level control is with the inlet flow control
valve LCV-01, and the FWKO tank level is controlled by the control valve at the
pump discharge (LCV-02). This nested control scheme is a much cleaner and more
reliable design, delivering better overall performance.
A nested loop is similar to a cascade loop in that the inner loop is the one with
a faster response and thus has tight control to suppress disturbances. However, they

Fig. 4.14 Tank level control: nested control

138 4 Complex PID Control Loops

are fundamentally different in that the inner loop is not an integral part of the outer
loop in terms of the control structure. They are two independent control loops that
interact with each other indirectly through the process dynamics.
Like cascade control, PID tuning of the nested control scheme should start with
the inner loop and move on to the outer loop. The closed-loop time constants of the
two loops should be at least five times different.

Split-Range Control Versus Dual-Controller Control

In Sect. 2.2.2, it was mentioned that a split-range control scheme could be used
to extend the range of control by manipulating multiple control valves. Split-range
control can also achieve prioritized control for applications where a preferred order
of sequence exists among the multiple manipulated variables for controlling the same
target.
Figure 4.15 shows a pressure control scheme for a production separator which is
very common in an E&P production facility. The control scheme has a split point
of 70% where the first 0∼70% of controller output is mapped to 0∼100% for pres-
sure valve PCV-101A; pressure controller output from 70% to 100% is mapped to
0∼100% for control valve PCV-101B. This output mapping is illustrated in Table 4.3
and Fig. 4.16.
This standard split-range control simultaneously provides the following two func-
tions:

Fig. 4.15 Split point in split-range control

4.1 Applications of Complex PID Control Schemes 139

Table 4.3 Output mapping in PC-101 output (%) PCV-101 (%) PCV-102 (%)
a split-range control
0–70 0–100 0
70–100 100 0–100

Fig. 4.16 Split point

calculation in split-range
control

1. Extending the control range: The pressure control during normal operating con-
ditions is via pressure control valve PCV-101A. In case of a significant plant
upset, the capacity of PCV-101A is not sufficient. The pressure controller opens
up flaring valve PCV-101B to extend the control range.
2. Prioritizing flow control: With this scheme, the control valve PCV-101A is the
preferred control. The process flow goes to PCV-101A first. When the capacity of
PCV-101A is exceeded (PC-101.OP>70%), the controller opens PCV-101B.
Conversely, when both PCV-101A and PCV-101B are open, if the pressure
returns to normal and the pressure controller needs to reduce outlet flow, it
will reduce and close PCV-101B first. Valve PCV-101A is throttled only after
PCV-101B is fully closed.
The use of split-range control can be very flexible and creative. Figure 4.17 shows
several variations of split-range control. For example,
1. The direction of action of the two valves can be the same or different, e.g.,
Fig. 4.17a, b.
2. There can be a gap or an overlap at the split point, e.g., Fig. 4.17c, d.
3. The output of the controller can be sent to a valve or another control element.
4. The split-range control can use more than two valves if justifiable, although it is
strongly discouraged for the reasons mentioned below.
140 4 Complex PID Control Loops

Fig. 4.17 Examples of split-range control loop

Split-range has been in widespread use to the extent that it is sometimes overused
or even abused. Some inherent drawbacks require careful considerations in the control
design and implementation to ensure integrity and performance. The considerations
include the split-point calculation, controller tuning, and back initialization:
1. Split-point calculation: A split point at 50% is typically the default, although this
default value is often inappropriate. The PID controller in a split-range control
only has one set of tuning parameters. The 50% split point implicitly assumes
that the dynamics of the two legs are identical. In reality, the two dynamics
4.1 Applications of Complex PID Control Schemes 141

can be significantly different, e.g., for the split-range control arrangement in

Fig. 4.15. As a result, the same control tuning is not adequate for the full range
of control. If the controller is tuned to produce an optimal performance for the
suction pressure valve, the same tuning would be too slow for the flare valve to
react to pressure upset. The split point, or transition point, should be calculated
based on the actual dynamics of the two control ranges to ensure the controller
tuning is adequate for both legs in the full range.
Example 4.2 Split-point calculation. A split-range pressure controller directly
manipulates two control valves for two steam flow streams. The two valves are
sized as 50 and 150 kg/h, respectively. The desired split point is then given by

50
Split-Point = · 100% = 25%. (4.18)
150 + 50

The control range is 0–25% for the smaller valve and 25–100% for the larger
valve.
Another approach to determining the split point is via step tests. Apply a 1 or
2% change to each valve, one by one. Assume that the change in pressure is 1.5
bar and 0.5 bar, respectively, then the split point can be calculated as

K p,1 0.5
Split-Point = = · 100% = 25%. (4.19)
K p,1 + K p,2 0.5 + 1.5

See Fig. 4.18 for an illustration of the split-point calculation based on loop gains.
The practical challenge is that the calculation of the split point can be tricky and
inaccurate. It is often difficult to find a fixed set of PID tuning parameters to
achieve good performance for both control valves. For critical control loops, it
may be worthwhile considering a more complex gain scheduling scheme.

2. Bumpless transfer: When the secondary control elements in the control loop are
disconnected and subsequently re-instated, the controller output will initialize to
the one that switches to CAS first. This behavior can potentially cause a bumpy
transition. For the example in Fig. 4.15, suppose both valves, PCV-101A and
PCV-101B, were taken out of service and subsequently put back to CAS. If
both valves are at 0% opening, and now PCV-101B is switched back to CAS
first, it is expected that the pressure controller PC-101 be initialized to 70%

Fig. 4.18 Examples of

split-point calculation with
gains
142 4 Complex PID Control Loops

because valve PC-101B is in the 70–100% leg of the split range of PC-101.
When PCV-101A is later re-instated to CAS, the output to PCV-101A would
see a jump from 0 to 100%. For mission-critical applications, this type of bumpy
transition cannot be tolerated. It is critical to specify in the operating procedure
that PCV-101A should be put back to CAS before PCV-101B, even though the
bumpless transfer feature may already be built-in in the control module.
3. Slow action: Another issue with split-range control is that the pressure controller
output must transverse the range of one valve before acting on the other valve.
In emergency conditions, this slows down the speed of response. For example,
in the split-range control in Fig. 4.15, if the controller output is at 35%, the two
control valves, PCV-101A and PCV-101B, will have openings of 50 and 0%,
respectively. Suppose a significant process upset is experienced, and the pressure
abruptly increases; In that case, we expect the pressure control PCV-101B to
open as quickly as possible to stop the pressure from going higher. With the
split-range control scheme, the controller output would increase from 35 to 70
to cause the normal regulatory control valve PCV-101A to open 100% before
opening the flare valve PCV-101B. The slow reaction may be unacceptable.

A simple solution for the above problems is to replace the split-range controller
with two independent pressure controllers. This arrangement is the dual-controller
control scheme. Both controllers read from the same pressure transmitter, but each
independently manipulates one of the two valves. The two independent controllers
can have independent tunings and are thus straightforward to achieve optimal tuning
for each. This dual pressure control scheme is depicted in Fig. 4.19.
The vital requirement for a dual-controller control scheme to work is that the two
controllers must have sufficiently different setpoints. Otherwise, the two controllers
would “fight” each other to achieve a similar setpoint. The requirement of different
setpoints is the primary drawback of the dual-controller scheme. For most control
applications, this is not an issue. However, for applications that require a single
control setpoint, split-range control is still the preferred scheme.

Fig. 4.19 Dual controller

versus split-range
4.1 Applications of Complex PID Control Schemes 143

In general, a dual-controller scheme is preferred over split-range control due to

its simplicity in tunings and configuration. The dual-controller scheme can even be
extended to a multi-controller control scheme to achieve more layers in protective
control. However, this is not recommended for processes with significant noises and
fluctuations.

4.2 Structure of Complex Control Loops

A basic PID controller comprises one controlled variable and one manipulated vari-
able (see Fig. 2.2) and is sufficient for most real-world applications. However, there
are also many applications with multiple interacting variables and require more com-
plex control schemes than a single PID controller.
There are two approaches to deal with multi-input multi-output (MIMO) prob-
lems. One approach is to use multiple PID controllers and make them work together to
address the multivariable interactions. The result is the complex PID control schemes
discussed in this chapter. The other approach resorts to a single advanced controller
consisting of sophisticated calculations that implicitly address the multivariable inter-
actions through models and calculations. This approach is represented by model
predictive control (MPC) that will be discussed in Chap. 5. To the operator, a com-
plex PID loop is a “white-box” solution that is fully transparent, while MPC is a
“black-box” solution with the details hidden.
For clarity, we will use the terms controller, control loop, and control scheme to
describe the increasing level of complexity for PID-based control solutions, such as a
standalone PID controller, a standard complex PID loop, and a more complex control
scheme. For a complete control design, simple or complex, we will call it a process
control solution. These terminologies are solely for convenience and clarity and may
be used interchangeably in a particular context. For example, a model predictive
control is a complex control scheme but may be called an advanced control loop in
control performance monitoring.

4.2.1 Common Function Blocks

A PID-based complex control scheme consists of one or more PID controllers and
some supporting control function blocks. These function blocks are built-in modules
in modern control systems such as distributed control systems (DCS) and supervisory
control and data acquisition (SCADA) systems. They are the building blocks for
process control solutions, both simple and complex. Table 4.4 provides a list of the
commonly available function blocks.
For example, in the Foundation Fieldbus control system and Emerson DeltaV
system, 10 “basic” blocks, 19 “advanced” blocks, and five other complex function
blocks are specified, as listed in Table 4.5.
144 4 Complex PID Control Loops

Table 4.4 PID-based complex control loops

Control blocks Description Back-calculation support
PID controller PID control 
Ratio control Ratio control 
Manual loader Manual control 
Automatic switch Two-way or multi-way switch 
Control selector High/low selector 
Control splitter Split-range, fan-out 
AO/DO Analog output/digital output 
AI/DI Analog input/digital input
Gain/bias block Gain and bias
Lead/lag Lead/lag calculator
Dead-time Time delay block
Manual switch Two-way or multi-way switch
Signal selector High/low selector
Signal characterizer Conditioning, selecting, filtering
Calculation block General-purpose calculation

It is beyond the scope of this book to go into the details of all the function blocks.
They can be looked up in the specific control systems manuals when needed, and the
IEC standard (IEC-61804) defines the common requirements. However, a reasonable
familiarity with their general functionality and typical configuration is essential for
designing and maintaining complex control solutions.
For practical reasons, the function blocks offered by different control systems
vendors are not standardized, even in their naming. We will use generic names in
this description, and focus only on the general functionality.
One crucial aspect is the tracking and initialization (Sect. 3.4) of the PID con-
trollers in a complex control scheme. For complex control schemes with multiple
function blocks, the tracking and initialization become more critical and complicated.
In consideration of loop integrity, we divide the function blocks into two categories:
control blocks and non-control blocks. Control function blocks provide built-in sup-
port for tracking and initialization. Non-control function blocks, such as arithmetic
and logic blocks, do not typically support tracking and initialization.

Control Function Blocks

The control function blocks provide BKCALI and BKCALO parameters to propagate
the initialization values for the control element’s output, which constitutes the back-
calculation data flow (see Sect. 3.4.2):
4.2 Structure of Complex Control Loops 145

Table 4.5 Function blocks in Control blocks Description

fieldbus control system
AI Analog input
AO Analog output
B Bias/gain
CS Control selector
DI Digital input
DO Digital output
ML Manual loader
PD Proportional/derivative control
PID Proportional/integral/derivative control
RA Ratio station
Pulse input
Complex analog output
Complex discrete output
Step output PID
Device control
Setpoint ramp
Splitter
Input selector
Signal characterizer
Dead time
Calculate
Lead/lag
Arithmetic
Integrator (Totalizer)
Timer
Analog alarm
Discrete alarm
Analog human interface
Discrete human interface
Multiple analog input
Multiple analog output
Multiple digital input
Multiple digital output
Flexible function block
146 4 Complex PID Control Loops

1. PID Controller: A basic PID controller block has two inputs (PV and SP) and one
output (OP). The process value (PV) is typically pulled from another function
block, such as an analog input AI block or a calculation block. The setpoint (SP)
is typically from the operator (AUTO mode), another function block (CAS mode),
or a user program (RCAS mode). The basic PID control block provides such
standard features as input signal scaling and limiting, PID calculation, output
signaling processing, alarming, and output limiting. Some implementations may
have additional control functions built into the PID function block to perform
additional functions such as feedforward and nonlinear gain scheduling.
PID controller typically offers full support for back-calculation through
BKCALI and BKCALO propagation (see Figs. 3.13 and 4.20a).
However, suppose the secondary element does not propagate the windup status
or initialization data back to the primary block, explicit external feedback would
have to be provided in the control design to help the PID control. The feedback
value is typically the process value or setpoint of the secondary block, which is
used to initialize the output value of the primary block. That is, the controller
output is then given by

CV = EXTFBK + GAIN · ERROR (4.20)

where CV is the controller output, EXTFBK is the external feedback, and ERROR
is the control error. See Fig. 3.7 for an illustration of using the valve position as
external feedback.
2. Control Selector: A control selector receives two or more2 control signals (typ-
ically controller output) and selects one of them based on the following config-
urations:
• High selection: A high selector selects the largest value of the inputs.
• Low selection: A low selector selects the smallest value of the inputs.
• Mid-of-three: With three inputs, the median value of the three is selected.
• High median: The median value is selected if the number of inputs is odd,
or the larger median value if the number of inputs is even. A high median
selection is the same as a mid-of-three selection in the case of three inputs.
• Low median: The median value is selected if the number of inputs is odd, or
the lower median value is selected if the number of inputs is even.
• Average: The average selector calculates the average value of all the inputs.
Figure 4.20b is an illustration of a selector with two inputs. A control selector
(e.g., an override control block) typically supports the following back-calculation
features:
• One BKCALI connection for the input, for receiving back-calculation infor-
mation from the downstream control element.

2 DeltaV supports two inputs while Experion supports up to four.

4.2 Structure of Complex Control Loops 147

Fig. 4.20 Function blocks: PID controller, control splitter, and control selector

• Multiple BKCALO connections, one for each output, to send the back-
calculation information to the upstream control elements (e.g., controllers).
Contrary to a control selector, a signal selector is an arithmetic function block
performing the straightforward signal comparison and selection without support
for tracking and initialization.
3. Control splitter: A control splitter splits the control signal into multiple ranges
and sends them to multiple downstream control elements. The control splitter
is commonly used for split-range control and fan-out control. Depending on
the control system, the splitter may be offered as one-to-two, one-to-three, or
one-to-many split.
Figure 4.20c shows a one-to-two control splitter. A control splitter (e.g., a fan-out
block) typically supports one or more of the following back-calculation features:
• Multiple BKCALI connections, one for each input, to receive back-calculation
value from the downstream control elements.
• One BKCALO connection, to send back-calculation value to the upstream con-
trol element.
4. Switch control block: A switch block can be positioned to assign a different
primary element to a secondary element. The switching can be initiated by the
operator, a user program, or another function block. A switch block typically sup-
ports back-calculation, similar to a control selector block. The difference is that
a selector automatically makes the selection based on the internal comparison,
while a switch forces the selection externally.
5. Manual loader: A manual loader allows the operator to adjust the manipulated
parameter, i.e., for manual control. A manual loader is also called a hand-
indication control (HIC). Another common application of the manual loader
block is providing “bumpless” output following initialization or mode changes.
148 4 Complex PID Control Loops

It is the simplest function block with built-in bumpless transfer function. For
example, a manual loader can be inserted between two control elements to
achieve bumpless transition when the bumpless transfer can not be achieved
with standard function blocks.
6. Advanced control blocks: Many DCS systems such as DeltaV have model pre-
dictive control built-in as a standard control module, similar to PID controller.
Additional functionalities may be built into the control selectors and splitters, most
commonly the bumpless transfer function. For example, in Honeywell Experion, the
selector output is the selected input plus two bias terms. A floating bias is generated
for the output upon re-connection with the downstream control element. The floating
bias gradually decays to zero to avoid bumpy transition:

OP = SELECTED_INPUT + BIAS.FIX + BIAS.FLOAT (4.21)

where BIAS.FIX is a fixed bias term and BIAS.FLOAT is a floating bias term. A
gap between the primary element’s output and the secondary element’s setpoint
may exist when a disconnected control loop is re-connected. The floating bias
BIAS.FLOAT is initialized to this gap value and gradually ramps to zero:

BIAS.FLOAT = BIAS.FLOAT − DECAY_RATE. (4.22)

This floating bias is a straightforward yet effective approach to achieving bumpless

transfer (see Sect. 4.3.4).
Some control systems choose to use a first-order filter to decay the initial bias,
which serves the same purpose.

Non-control Function Blocks

Non-control function blocks, such as arithmetic or logic calculation blocks, do not

typically support tracking and initialization. Therefore, it is critical to double-check
the tracking and initialization capability if the control loop containing any non-control
function blocks:
1. Gain block: A gain block is typically implemented as a gain plus bias function:

G(s) = K c . (4.23)

2. Lead/lag block: A basic lead/lag block has a transfer function of the following
form:
1 τ2 s + 1
G(s) = , G(s) = . (4.24)
τ1 s + 1 τ1 s + 1
4.2 Structure of Complex Control Loops 149

Since the denominator has a phase lag and the numerator causes a phase lead, thus
the name of lead/lag. It is a primitive building block for more complex functions.
3. Time delay block: As the name suggests, a pure time delay block delays a signal
by the prescribed duration. The transfer function is given by

G(s) = e−θ s . (4.25)

4. Calculation block: The general-purpose calculation block (also called Charac-

terizer in some control systems) provides both logical functions and arithmetic
computational capability within one integrated environment. A calculation block
provides the ultimate flexibility in signal processing. However, it typically does
not support tracking and initialization, and thus should be used with caution.
The gain, lead/lag, and delay blocks are the primitive function blocks. A first-
order-plus-time-delay (FOPTD) transfer function can be readily built from the three
primitive function blocks:

1
G(s) = K c · · e−θ s . (4.26)
τ1 s + 1

The PI control and I-only controller shown in Sect. 3.1.9 can be built with these
primitive blocks.
Table 4.6 provides a summary of the complex control loops and the function blocks
they need.
The configuration support for tracking and initialization varies from one control
system to another. For example, in Honeywell Experion and Yokogawa Centum, the
back-calculation data flow is automatically established upon the connection of the
control blocks when the control scheme is built. In Schneider Foxboro and Emerson
DeltaV, an explicit configuration is required to connect BKCALO with BKCALI.
Explicit connection is an extra step in configuration but offers added flexibility.

Table 4.6 Complex control loops and function blocks

Complex control loop Function block Initialization concerns
Ratio control PID + Ratio
Override control PID + Control selector Reset-windup
Split-range control PID + Control splitter Back-initialization
Fan-out control PID + Control splitter Back-initialization
Dual-controller control PID + PID
Feedforward control PID + Lead/lag + Delay + Gain/bias Bump transition
Decoupling control PID + Calc
Balancing control with offset PID + Gain/bias Saturation
Model predictive control
150 4 Complex PID Control Loops

4.2.2 Built-in Function Versus Custom Function

Modern control systems provide a rich collection of function blocks to fulfill the
different control needs, from simple control loops to complex control schemes. Dif-
ferent designs for complex control solutions can achieve the same control objective,
but the resulting control performance can be substantially different. Some general
recommendations are given below to help achieve better control performance:
1. Use the built-in control blocks wherever possible: These built-in function blocks
are carefully designed, feature-rich, and field-proven. There is no need to re-invent
the wheel unless absolutely necessary.
For the same reason, use the function blocks that offer the full features for the
need. For example, a PID can be built from primitive control function blocks
such as a gain/bias, lead/lag, and delay block, as shown in Fig. 3.5 or Fig. 3.6.
However, there is no reason not to use a single standard PID function block if
available. Similarly, a ratio control can be constructed with a calculation block
and PID controller, but a standard ratio control block should be used instead, if
available.
2. Prefer control function block to non-control function blocks: Control function
blocks have built-in support for tracking and initialization to protect the loop
integrity under different operating scenarios. On the other hand, a function block
that does not support the back-calculation, e.g., a calculation block, interrupts
the data flow and sometimes may need extra help to bridge the “gap” in the data
flow. For example, a control selector supports tracking and initialization, and the
loop integrity is protected against a broken connection or limited values. A signal
selector may be used in place of the control selector. However, the back-calculation
data flow is broken at the signal selector since tracking and initialization are not
supported.
Consider a practical example. In a water injection application, a pressure controller
controls the pressure of the injection header pressure. The output of the pressure
controller is fanned out to dozens of flow controllers; each controls the flow rate
to an individual injection well. Control splitter has back-calculation support, and
using the built-in control splitter is the preferred approach. However, the control
splitter on some control systems is limited to a 1-to-2 splitting.3 If there are 32
injection wells connected to this manifold, it will take five layers and a total of 1 +
2 + 4 + 8 + 16 = 31 splitters to fan out the controller output to all the 32 wells.
The multiple layers of selectors seem to be a very messy implementation. Thus,
it may be tempting to utilize a single calculation block to distribute the process
controller output to the 32 flow controllers. In reality, as we discussed before, a
calculation block does not support tracking and initialization, therefore to ensure
proper back-propagation of the value and status of the secondary controllers to
the pressure controller, some other calculation and logic blocks are needed to

3In Honeywell Experion, a fan-out block can have up to eight secondary elements, and with back-
calculation support.
4.2 Structure of Complex Control Loops 151

mimic the back-calculation data flow and the handshake process. Designing and
configuring the calculation and logic blocks can be very involved and are also
prone to errors. In addition, the custom design and implementation also pose a
significant challenge for maintenance if not adequately documented. Therefore, it
is better to accept the superficial complexities and use the built-in control splitters
to achieve this 1-to-32 splitting of the control signal.
3. Custom control logic: Sometimes back-end control logic needs to be implemented
to meet special requirements not provided by the standard control function blocks.
However, experience shows that the back-end logic can be another challenge for
operation and troubleshooting since it can be easily overlooked due to its low
visibility. Treat back-end logic as a necessary evil, and use it with discretion.

4.3 Configurations of Complex PID Loops

Although PID configuration is the responsibility of control system engineers or tech-

nicians, experience shows that many configuration options are incorrectly configured
or simply left to the system defaults. Many configuration options are dictated by the
control strategy and should be specified by process control engineers.
The PID configurations that may significantly impact the control performance
are discussed in Sect. 3.5. However, for complex PID control loops comprising
multiple PID controllers and supporting function blocks, the loop configuration can
be dauntingly complex and demands many additional practical considerations.

4.3.1 Data Flow in Complex PID Loops

A complex control problem typically involves multiple control handles and multiple
control targets. The data flow from the controllers to the final control elements is no
longer a one-to-one fixed path as in simple PID control but can be many-to-many.
Suppose selectors or splitters are utilized in the control scheme, the control action
produced by a particular controller may travel a different data flow path and reaches
a different final control element under different operating condition. In addition,
the data flow can be disrupted or limited at many locations and for many reasons.
Similarly, the changes in the final control element are fed back to the controllers
through back-calculation, and the back-calculation channel can be disconnected or
disrupted.
As discussed in Sect. 3.3, the four abnormal scenarios with the data flow become
exacerbated for complex control schemes:
1. Bad data: The control data is bad or missing due to loss of communication, out
of service control devices, or bad measurements.
152 4 Complex PID Control Loops

2. Data flow is broken. The data flow is blocked, or the flow path disconnected, e.g.,
for the following reasons:
• A control element such as a controller or analog output block on the data flow
path has switched away from cascade mode.
• The data is re-routed by a selector, a switch, or a splitter.
3. Data value is limited, saturated, or constrained. For example,
• Controller output is high or low limited.
• Setpoint sent down has reached the setpoint high or low limit of the secondary
element.
4. The loop dynamics have changed. Many reasons can cause this to happen, such
as the following:
• Process dynamics change due to aging, fouling, or process modifications.
• Valve problems such as stiction, hysteresis, and leaking.
• Measurement errors such as lack of calibration.
Reliable tracking and detecting gradual dynamics change is difficult and remains
a challenge for both theory and practice.
The tracking and initialization of the PID control implementation is a critical con-
sideration to ensure loop integrity. With tracking and initialization, the controller can
take appropriate actions to initialize itself when an abnormal condition is detected.
All control elements on the data flow path must have the same support to make the
complex control loop work. Unfortunately, not all the function blocks (see Table 4.4)
fully supports tracking and initialization.

4.3.2 Common Configuration Considerations

DCS implementation of a PID controller provides many configuration options to

meet the various operating requirements. Although PID configuration is the scope
of work of control system engineers, several configuration options can significantly
impact control performance and need to be addressed by process control engineers.
Section 3.5 has provided a general introduction to PID loop configuration from
the perspective of the PID controller itself. However, for complex control schemes,
the interaction and inter-operation of the controllers and the supporting control func-
tion blocks deserve special attention. For example, an essential configuration item
for overriding control is anti-reset windup, while bumpless transfer is a common
consideration for a fan-out control. In addition, for all complex control schemes,
careful consideration should be given to crippled mode operation and PV tracking.
Table 4.7 lists the common configuration requirements that should be considered for
each type of standard complex PID control loop.
4.3 Configurations of Complex PID Loops 153

Table 4.7 Common configurations for complex PID loops

Crippled mode PV Anti-reset Bumpless Gain
Operation Tracking Windup Transfer Scheduling
Cascade control X X
Override control X X X X
Switching control X X X X X
Fan-out control X X X X
Split-range control X X X X
Custom calculation X X X

The selection of PID equation type depends on the specific control scheme. See
Table 4.8 for a list of the recommended equation types. As mentioned in Sect. 3.1.4,
most PID control in practice uses PI algorithm; thus, the difference between PID
and PI-D is negligible.
The mode of operations of a PID control can significantly impact the loop dynam-
ics (see Sect. 3.2). A controller in manual mode breaks the PID loop, and a cascade
controller in local override mode simply bypasses the controller. For PID controllers
in a complex control scheme, the impact of the controller mode on the loop dynamics
is even more significant, especially when other control elements such as selectors and
splitters are involved. It is thus critical to review the requirements on controller mode,
setpoint value, and controller output value during the different operation modes as
listed in Table 4.9. All the question marks in the table need to be considered during
design.
There are many beautiful theories and techniques for control stability analysis.
However, they are of limited value to practical process control because these theo-
ries and techniques assume the process dynamics are known and fixed. For process
control, the stability analysis must be based on the loop dynamics rather than the
process dynamics alone. In a complex PID control scheme, the more detrimental
factors affecting control stability are often other components in the loop, e.g., the
control valve with stiction or saturation, the splitter or selector in the control that
alters the control structure online.

Table 4.8 Equation type for Equation type SP tracking PV

complex PID loops
Standalone PID I-PD NO
Protective PID I-PD NO
Overriding PID I-PD NO
Cascade (primary) I-PD NO
Cascade (secondary) PI-D YES
Fan-out (primary) I-PD NO
Fan-out (secondary) PI-D YES
154 4 Complex PID Control Loops

Table 4.9 Mode of operation for complex PID loops

Controller mode Controller setpoint Controller output
During shutdown ? ? ?
During startup ? ? ?
During normal operation ? ? ?
Entering crippled mode ? ? ?
During crippled mode ? ? ?
Leaving crippled mode ? ? ?
Emergency shutdown ? ? ?

In summary, it is highly desirable to reach the highest level of automation and

minimize the need for operator intervention, especially during busy startups and
emergency shutdowns. For this purpose, all the potential abnormal conditions should
be considered, and the PID controllers and supporting functional blocks should be
configured to be “fool-proof” and “fail-safe” as much as possible. On the other hand,
the practical reality limits how far we can push in automating the operation. The best
decision is always a trade-off between desire and practicality.

4.3.3 Anti-reset Windup in Complex Control Loops

As a closed-loop feedback control technique, PID control assumes that moving the
controller output (OP) in the right direction would bring the process value (PV)
closer to the desired setpoint (SP) and eventually eliminate the control error. If
the control error does not go away, the controller’s integral term will continue to
change the output in the same direction, regardless of whether the feedback is lost
or compromised. For example, override occurs when another controller takes over
control of a particular loop (e.g., for safety reasons). If the controller is unaware that
its output has been overridden, it will continue to act on the control error it can no
longer influence. Eventually, the output will be driven to saturation.
This phenomenon is called reset windup or integral windup (Marlin 2015) and
has been briefly explained in Sect. 3.5. The risk of reset windup is that when the
overridden controller needs to take control, the output may take some time to come
out of windup before becoming effective again. Reset windup degrades control per-
formance and causes an acceptable delay in action.
Reset windup can be made clear by revisiting the PID algorithm in Eqs. 3.7
and 3.25. The PID controller produces the incremental change in control action u(k)
for eliminating the current control error e(k):
4.3 Configurations of Complex PID Loops 155

1 t d e(t)
u(t) = K c e(t) + e(τ )dτ + Td (4.27)
Ti 0 dt
Ts Td
u(k) = K c e(k) + e(i) + (e(k) − e(k − 1)) . (4.28)
Ti i
Ts

The control u(k) is added to the current process output and becomes the new control
output
Unew = Ucurr + u(k) = Ucurr + f (SP − PV) (4.29)

where Ucurr is the current control output, and Unew is the new control output. U is the
actual output that drives the final control element, and u(k) is the increment change
in output produced by PID to eliminate the current control error. In other words,
u(k) represents the delta change in the control action, and U is determined by the
steady-state condition, which is the bias term in Eq. 3.24.
The integrating action in the PID algorithm is a powerful feature of PID control for
eliminating control offset, but it also generates many practical issues. The integrating
action in the PID controller does not know, nor does it care, if the control error
is caused by its own control action or other reasons. In addition, the PID control
calculation produces a change in the output, and does not know, nor does it care,
what the actual value of the output should be. As long as there is a control error e(k),
the integral action will continue integrating on the error and producing a non-zero
incremental change u(k) added to the current control output.
In practice, there are built-in default limits for most of the variables in a controller.
A valve opening is limited to 0∼ 100%, and a controlled variable (controller PV)
is limited to its measurement range. If the control error persists, the output would
continue to change in the same direction until it reaches the limits.
The standard solution is to stop the integrating action on the error once the control
action is limited. The PID controller then becomes a simple P-only controller. That
is
SP(k) − PV(k)
u(k) = K c · e(k) = K c · · [OP Range]. (4.30)
[PV Range]

In a practical implementation, the back-calculation provides the actual output Ucurr

via the BKCALO to BKCALI propagation. For example, in Fig. 3.14, the setpoint for
the AO block is the controller output, and the back-calculation value BKCALI for
the PID controller is the BKCALO value of AO block output. The BKCALO value
for the AO block is usually set to the value of the AO output, which is the actual
control action propagated down from the PID controller. This BKCALI value serves
as the current controller output Ucurr in Eq. 4.29. Therefore, when the control data is
limited, the output of the PID control becomes
156 4 Complex PID Control Loops

Unow = Ucurr + u(k)

= BKCALI + u(k)
SP(k) − PV(k)
= BKCALI + K c · . (4.31)
[PV Range] · [OP Range]

Similarly, the BKCALO value of the PID block is set to its process value (PV) and
is sent back to its upstream control block as its BKCALI input.
If BKCALI is not available, the controller uses its own output from the last exe-
cution as Ucurr , not knowing whether the output has reached the next control element
on the data flow.
Example 4.2 Anti-Reset Windup. Consider the example in Fig. 4.21. The level
controller LC-101 is for normal regulatory control, while PC-101 is for the high-
pressure protective override.
Without anti-reset windup protection, the SP, PV, and OP trends of both con-
trollers, LC-101 and PC-101, are shown in Fig. 4.22.
In the beginning, when the level controller LC-101 is in control, its output is
selected by the low selector FY-101 and sent down to the flow controller FC-101.
The output of the pressure controller PC-101 is ignored (overridden). The pressure
controller PC-101, not knowing that its output is not selected, continues increasing
its output, attempting to minimize the control error. This effort is fruitless since the
loop is effectively open at the selector. The integrating action eventually drives the
output to 100% at time 10 and remains at 100%.
At time 20, the pressure starts to rise for some process reasons. At time 33, the
pressure rises above its setpoint (the high-pressure limit). High pressure is an unsafe
condition, and the pressure controller is expected to act immediately to override
the level controller and reduce the flow. However, the saturated pressure controller
will have to un-wind the saturation first. The plot shows that the pressure controller
output (PC101.OP) takes 22 seconds to come down and become lower than the level
controller output (i.e., to have PC101.OP ≤ LC101.OP). Only after that point (time
55) does the pressure controller output (PC101.OP) get selected by the low selector

Fig. 4.21 Examples of

override control
4.3 Configurations of Complex PID Loops 157

Fig. 4.22 Override control without anti-reset windup enabled

FY-101, and the flow starts to decrease. At the same time, the pressure continues to
rise and reaches 2 bars above the pressure limit. This delay in action is unacceptable
for a protective control!
Now let us look closely into the anti-reset windup configuration. With anti-reset
windup configured, the information flow of this level-pressure override control loop
is shown in Fig. 4.23.
Assume that the level controller is in control, then the pressure controller (an
overriding protective controller) is open-loop, and its output is not “consumed.” For
the information flow, note the following:
1. First, note that the back-calculation path is from the valve to the two top-level PID
controllers. If the actual valve opening is not available, then the back-calculation
input for the AO block will default to its output.
2. Secondly, at the selector FY-101, the two data flow streams (going forward) from
the two PID controllers merge into one. Naturally, the back-calculation splits into
two paths at the selector. The back-calculated value at the selector is the actual
output of the selector.
A controller in the override control must always remain ready to act even when not
selected. Anti-reset windup aims to prevent the controller from going into saturation
and is achieved through two actions:
1. When the pressure controller output is not selected, i.e., the loop is open, the pres-
sure controller tracks the downstream status and value through back-calculation.
The valve position is back-propagated to the AO block, the flow controller
158 4 Complex PID Control Loops

Fig. 4.23 Data flow in an override control loop

FC-101, and then to the selector block (FY-101). The selector “informs” both
the level controller LC-101 and the pressure controller PC-101 which con-
troller output is currently selected and the actual value of the selector output.
2. The controller algorithm is temporarily switched to a P-only algorithm to avoid
integrating action, which is

OP = [BKCALI] + GAIN · ERROR. (4.32)

Considering the dimensionless calculation in the PID algorithm discussed in

Fig. 3.4 of Sect. 3.1.5, the control output with proportional-only calculation is given
by

PV − SP
OP = [BKCALI] + GAIN · · [OP_SPAN] (4.33)
[PV_SPAN]
11 − 8
= 750 + 0.5 · · 1200 = 870 m 3 / h (4.34)
15 − 0

where 750 is the output that is sent down to the AO block by the level controller
LC-101.
4.3 Configurations of Complex PID Loops 159

Fig. 4.24 Override control with anti-reset windup enabled

Figure 4.24 shows the process response when anti-reset windup is enabled. Instead
of 22 s, the pressure control takes control almost immediately, one second (one scan
time) after the process value rises above the pressure setpoint. The overshoot in
pressure is only 0.2 bar!
Note that in Fig. 3.4, since the control error is the difference between the setpoint
and process value, it decreases to zero when the pressure goes up to the same value
as the setpoint. The PID calculated output becomes zero as well (Eq. 4.30). From the
above equation, it is clear that the output decreases from 870 m3 /h to 750 m3 /h and
equals the level control output! At this point, the control error is zero, the two outputs
are equal, and the calculated output change is zero. So the cross-over and transition
(the takeover) are completely bumpless! If the pressure value increases further and
becomes higher than its setpoint, the pressure controller output would become lower
than 750 and take over the control from the level controller.
Figure 4.24 also clearly shows that the level LC-101 starts to rise above its
setpoint because the pressure controller PC-101 takes over the level controller,
and the level is basically off control. Its output, however, will not wind up thanks to
the anti-reset windup configuration on the level controller.
The override control scheme in Fig. 4.21 can also be implemented between the
flow and pressure controllers, as shown in Fig. 4.25.
Compared to Fig. 4.21, the data flow will differ, but the same principle and analysis
apply.
160 4 Complex PID Control Loops

Fig. 4.25 Override control with an alternative configuration

As discussed in Sect. 4.2.1, modern control systems have the anti-reset windup
(ARWU) function built into their basic control modules. Most control function blocks
support tracking and initialization that can sense the changes in the data flow and
communicate it back to the controllers. However, function blocks such as arithmetic
calculation and logic processing do not support them. Besides, most final control
elements (values, switches) may not support position read-back to report its actual
position. In control design, it is crucial to pick the right function blocks to build
the control scheme and know where extra configuration work is needed to assist the
function blocks that do not have built-in back-calculation.
As explained in Sect. 4.3.3, reset windup occurs when the data flow is broken or
limited by such events as controller mode change, selective control, switching, valve
not responding, frozen signal with measurement. The output can be overridden by a
signal selector, a switch block, or be limited by high/low limits on the output itself,
the setpoint of the secondary element. As a general rule, anti-reset windup protection
must be configured for any PID controller if its output may be limited or overridden
by other applications.
In a complex control loop, the data flow can become overwhelmingly compli-
cated. Therefore, the probability of the data flow being disrupted or compromised
is exponentially higher. As a result, the anti-reset windup configuration becomes
nontrivial. The anti-reset windup requirement is due to the reset (integrating) action
in the PID controller. All PID controllers with I-action configured must consider
anti-reset windup protection. For a P-only controller, reset windup is not a concern.
ARWU is preferably done with DCS built-in function wherever possible. If not,
external feedback shall be provided based on calculated feedback values.
4.3 Configurations of Complex PID Loops 161

4.3.4 Bumpless Transfer in Complex Control Loops

Generally speaking, a bumpless transfer is required whenever this is a potential

interruption in the data flow. When the data flow is re-connected, abrupt changes
(“bump”) in the control output to the final control elements should not occur.
PID control tracks the data flow (status and value) for disconnection and re-
connection and tries to initialize the controller to avoid transfer bumps (Sect. 3.3).
The tracking may be trivial for a simple control loop but can become overwhelmingly
intricate when a complex PID control scheme is concerned.

Transfer Bump

As briefly discussed in Sect. 3.4, the data flow in a control loop may be interrupted
for many reasons such as mode change, override, selection, and switching. In a
cascaded connection where a PID control is the primary element and an AO block is
secondary, the primary element’s output is the secondary element’s setpoint. When
the connection is broken because the secondary element is switched away from the
cascade mode, a gap will develop between the primary element’s output and the
secondary element’s setpoint. When the data flow is re-connected, this gap can cause
a transfer “kick” or “bump” if proper initialization is not performed.
Standard complex control loops that may experience transfer bumps include cas-
cade control, fan-out control, ratio control, and split range controls (see Table 4.7 for
a summary):
1. Cascade control: In a cascade control loop (see Fig. 2.18), if the secondary con-
troller (inner loop) has switched away from cascade (CAS) mode, the primary
controller output is “unconsumed” and dangling. When the secondary controller
is switched back to cascade, its setpoint no longer matches the primary controller’s
output. The difference or gap may cause a “bump” in the secondary loop’s setpoint
if re-connected without adjustment of the gap.
2. Fan-out control: The primary controller’s output is fanned out to more than one
secondary control elements in a fan-out control scheme, as shown in Fig. 2.17. If
one of the control elements is disconnected (e.g., switched to manual) and then re-
connected (switch back to cascade) after some time, the setpoint of the secondary
element may no longer match the output of the primary controller. Again, this
gap can potentially cause a transfer bump if not correctly configured.
3. Split-range control: In a split-range control setup such as in Fig. 2.15, the controller
output is sent to two secondary control elements (valves). Assume that one leg in
the split-range control is broken by switching to manual control, and the other leg
of the split-range control is connected, the output of the primary control is still
consumed, and the loop is still closed. When the disconnected leg is re-connected,
the difference between the restored valve position and the controller output may
have already developed a gap, and the re-connection may cause a bump.
162 4 Complex PID Control Loops

There may be other scenarios where the disconnection and re-connection of the data
flow will cause a transfer bump. Some scenarios may be quite complex. For example,
consider the split-range control in Fig. 2.15 that has one controller and two valves,
with a split range of 0∼70% and 70%∼100%:
1. Assume that the controller output is 35%. Based on the 70% split point, the first
valve (in the lower leg) will be 50%4 open, and the second valve (in the upper
leg) is at 0%. Now assume that the first valve is switched to manual mode and a
persistent control error has driven the controller output to 73%.5 The 73% output
causes the second valve to open to 10%, and the first valve remains manual at
50%. If the first valve is switched back to cascade mode (re-connected), the valve
opening will jump from 50 to 100% because the controller output is higher than
70%. This sudden change in the valve is a “bump” that may be unacceptable for
some operations.
2. Similarly, assume that the controller output is at 85% and that both valves are
in cascade mode. With a 70% split point, the first valve should be 100% and the
second valve 50%. Suppose the second valve is switched to manual mode, and
the control error drives the controller output to 35% due to process condition
change. Now assume the second valve is now re-connected, the valve position
will experience a step-change from 50 to 0% since the controller output is now in
the 0∼70% range. This 50% bump in the valve opening may not be acceptable.
3. Now assume both valves are disconnected. In this case, which valve should the
controller track and which valve position should the controller output re-initialize
to? The controller will typically back-initialize itself to the first control element
switched to cascade, and this re-connection is thus bumpless. When the remain-
ing control elements are re-connected, the primary control can no longer back-
initialize. Each newly re-connected control element will have to initialize its
setpoint and take care of any gaps to ensure bumpless transfer.
Complex control schemes require a careful analysis of the bumpless transfer scenarios
during control design and implementation.

Bumpless Transfer

Bumpless transition is an essential control requirement. In modern control systems

such as DCS, most control function blocks have the bumpless transfer feature built-
in. For example, mode switching between manual, automatic, and cascade mode is
typically bumpless by design. Typically there are three options to respond to this
transfer bump:

4 The split-range of the first leg is 0%∼70% → 0%∼100%, therefore, 35% of controller output
35−0
drives the first valve to an opening of 70−0 × (100 − 0) = 50%.
5 Similarly, for the second leg with a mapping of 70%∼100%  → 0∼100%, a controller output of
73−70
73% is 100−70 × (100 − 0) = 10% in the opening of the second valve.
4.3 Configurations of Complex PID Loops 163

1. Do nothing: If the secondary element is a properly tuned PID controller, the bump
implies a step-change in the setpoint of the secondary controller at the time of
re-connection. The PID loop as a dynamic controller should be able to provide
the required damping in the response, which is acceptable by many applications.
In addition, the I-PD equation in the PID algorithm (Sect. 3.1.4) can be used to
eliminate the transition bump caused by setpoint changes.
2. The primary control loop adjusts its output to match the setpoint of the secondary
control elements, be it another controller or a valve: This type of initialization is
the so-called back-initialization, and has been the standard feature in most control
systems. It is automatically configured by default (see also PV Tracking in
Sect. 3.5). For example, a standard cascaded control loop relies on back initial-
ization to ensure bumpless transfer. With back-initialization, when the secondary
element is switched back to cascade, there will be no bump experienced since the
primary controller’s output has already lined up with the setpoint of the secondary
controller, resulting in a zero control error.
3. Secondary elements to decay the mismatch to avoid “bump”: Sometimes, the
primary control loop can not align its output to match the secondary elements’
setpoint because the primary controller’s output is already connected to other
control elements. In this case, each secondary element to be connected must ensure
bumpless transfer on its own if a bump is not acceptable. This type of initialization
is typically achieved with a built-in bumpless transfer function. With the bumpless
transfer function, the initial error at the time of re-connection is gradually reduced
and eventually eliminated over multiple scan cycles through a ramping function
or first-order filter.
For example, in a control scheme where the controller output is fanned out to two
valves, one valve is with 60% opening, and the other is offline and fully closed.
If the offline valve is put back to control, the valve opening will jump to 60% in
one shot, creating an unacceptable disturbance. An explicit function block such
as a manual loader (Sect. 4.2.1) can be inserted in front of the control valve to
provide a decaying function on the mismatch and achieve bumpless transfer. With
a bumpless transfer configuration, the 60% initial error will be gradually reduced.
See Fig. 4.26.
Several function blocks (Sect. 4.2.1) have the bumpless transfer function built-
in, including the manual loader block, control splitter, and the control selector.
Consult the DCS manual for accurate information.
Note that the bumpless transfer functionality is not the same as PV filtering. The
bumpless transfer function is activated only when the control path is re-connected
(e.g., mode change). The damping function applies only to the initial error at the time
of re-connecting the loop. Any errors developed after the re-connection will not be
subject to the damping function.
The bumpless transfer function is also different from controller back-initialization.
Back initialization is for the primary controller to align its output to the secondary
control elements and is the preferred approach since the ultimate goal is to eliminate
the transfer bump in the final control elements. A bumpless transfer function block
164 4 Complex PID Control Loops

Fig. 4.26 A PID control loop requiring bumpless transfer

is used when back-initialization can not be achieved, and the downstream control
element must take care of itself to prevent the transfer bump.
If none of the above standard options is applicable, an alternative control design
may be considered. For example, a dual-controller control is preferred over a split-
range control in practical applications for its simplicity and robustness in control
performance.

4.3.5 Bumpless Transfer Versus Anti-reset Windup

A transfer bump is caused by a mismatch in the data value between the primary and
secondary control elements due to the disconnection and re-connection of the loop.
On the other hand, reset windup is due to the data flow being limited or saturated at
some point in the control loop while the control loop is still closed.
Figure 4.27 is an example of an overriding control loop with a selector. The level
controller LC-101 is the primary controller for normal operation, while the pressure
controller PC-102 is an overriding protective controller to avoid high back-pressure
downstream the valve. The low selector selects the smaller of the two controller
outputs and sends it down to the flow controller FC-103. The flow controller output
is sent down to two control valves through a split-range control arrangement.
For loop integrity, a PID controller (LC-101, PC-102, or FC-103) in the com-
plex control loop not only has to send the controller output down to the valves; it also
must know whether its output reaches the valve and which one. Due to the branching
and merging of the control signal through selector and splitter, the output generated
from a controller may not be the one that is received by the target valve.
Based on the previous discussion, this control scheme is vulnerable to reset windup
due to the override control. It is also prone to transfer bumps because of the use of
4.3 Configurations of Complex PID Loops 165

Fig. 4.27 A PID control loop for back-calculation

split-range control. Figure 4.28 shows the simplified information flow, including both
the data flow and back-calculation. The back-calculation path propagates the valve
position information from the valves, through the splitter and selector, back to the
level controller LC-101, using BKCALO and BKCALI daisy-chain connections. As a
result, the level controller knows whether its output is “consumed” by the downstream
control elements. When the controller is “informed” by back-calculation that the
pressure controller has overridden its output, it simply stops the integrating action to
prevent output windup.
The function and operation of the back-calculation can become very perplexing
for complex control loops. Some DCS implementations, such as Honeywell TDC,
simply offer no back-calculation support for this specific control structure, i.e., a
splitter after a selector (Honeywell 2003).

4.4 Summary

A complex PID control scheme typically consists of multiple PID controllers and
other supporting function blocks. It can be surprisingly versatile in addressing com-
plicated practical problems such as multivariable interactions, constraint handling,
preferential flows, and limit pushing. However, effective use of complex PID con-
trol schemes require a good understanding of the operating requirements and also
intimate familiarity with the functionality of the building blocks.
Many application scenarios can be addressed by either complex PID schemes or
model predictive control, with different pros and cons. Complex PID schemes are
166 4 Complex PID Control Loops

Fig. 4.28 Back-calculation in a PID control loop

native DCS applications and are entirely transparent to operators and control system
engineers. However, the complexity in implementation can easily overwhelm the
operator and the maintenance personnel.
Feedback control relies on a known and stable data flow to close the loop. With
complex control loops, the data flow can become overly complicated. At each exe-
cution, the output of a controller in a complex control may travel a different path
to reach the final control elements. Many scenarios can cause the control signal to
be re-routed, disconnected, or limited along the data flow path. How to protect the
integrity of the data flow is crucial to the function of the control scheme. Modern DCS
implementation of the control scheme provides rich functionality in protecting the
4.4 Summary 167

loop integrity, with tracking and initialization being the standard features. However,
there are many scenarios for a complex control scheme that the built-in protection is
incomplete or insufficient. It remains a challenge to provide “fool-proof” protection
against loop integrity problems with DCS default functionality.
The loop integrity can be a determining factor in deciding the control strategy.
Process control engineers play a central role in ensuring that the control design has
considered all the possible scenarios. A process control engineer must know when
the built-in functions are sufficient to protect the loop integrity and when external
help must be provided to help the PID controllers.

References

Gordon L (2005) Advanced regulatory control: decoupling. Control Eng

Honeywell (2003) Advanced process manager. Control function and algorithm, Honeywell
Luyben WL, Lyuben ML (1997) Essentials of process control. McGraw-Hill
Marlin TE (2015) Process control—designing processes and control systems for dynamic perfor-
mance, 2nd edn. McGraw-Hill
Shinskey FG (1978) Control systems can save energy. Chem Eng Progress 74:43–46
Shinskey FG (1994) Feedback controllers in the process industries. McGraw-Hill Publishing, New
York
Smith CL (2010) Advanced process control—beyond single-loop control. Wiley
Stephanopoulos G (1984) Chemical process control—an introduction to theory and practice.
Prentice-Hall, New York
Chapter 5
Advanced Process Control

Most practical applications are more complex than simple PID controllers can han-
dle. Although complex PID control schemes are capable enough to address many
common needs, their inherent limitations and implementation challenges call for
more robust and elegant solutions (Sect. 5.1). Advanced process control technolo-
gies, such as inferential property estimation and model predictive control, result
from these practical needs. Most advanced control technologies are model-based
(Sect. 5.2), with inferential property estimation (IPE, Sect. 5.3) and model predictive
control (MPC, Sect. 5.4) being the most successful. MPC has also changed the land-
scape of practical process control and has become the most important alternative or
replacement of complex PID control solutions. MPC represents a mindset change
from localized control to large-area control. Its success depends on a good under-
standing of the process and a holistic view of the control problem, which is reflected
by an adequate control structure and high-quality process models. A well-developed
work process (Sect. 5.6) is critical to enforce this mindset change and ensure the
adequacy of the model quality and control structure.

5.1 Why Advanced Process Control?

As the dominating technology in process control for over 100 years, PID control has
been very successful. At the same time, being a simple feedback control algorithm,
PID also has severe limitations. There have been countless efforts to break these
limitations; Some enhance the existing PID algorithms, while others bank on entirely
different technologies. See Samad et al. (2020) for the result of a recent survey
conducted by the IFAC industry committee in 2018, where the impact of the current
and future control technologies are polled.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 169
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_5
170 5 Advanced Process Control

5.1.1 Process Challenges and PID Control Limitations

The process industry has the following significant challenges for process control that
are not common in most other industries:
Challenge #1: Strong multivariable interactions: Process control often involves
multiple inputs (control handles) and outputs (control targets) that heavily interact
and interfere with each other. Changes in one variable may cause more than one
variable to change; Similarly, there are multiple handles to use to effect a change
in one variable.
Challenge #2: Large time delay: Many processes exhibit long-time delays. Delay
dominant processes are particularly challenging for feedback control.
Challenge #3: Operating constraints: As process operation relentlessly strives for
efficiency and profitability, the operating points move closer and closer to their
limits and constraints, continuously raising the bar on the process control require-
ments.
Challenge #4: Noises and disturbances: Many sources of noises and disturbances
with unpredictable characteristics affect the process’s dynamic behavior. Many
complexities in the new control technologies are introduced just for dealing with
noises and disturbances.
Challenge #5: Unreliable or infrequent measurements such as quality measure-
ments based on online analyzers or lab samples: As process control is shifting
from isolated process variable control to higher-level quality control, the reliabil-
ity and response speed of most online quality measurements do not yet meet the
requirements for online closed-loop control.
Challenge #6: Nonlinear and time-varying dynamics: Most real-world processes
are inherently nonlinear. Besides, process behavior changes as the process oper-
ating point moves from one region to another or from time to time. Since most
controllers are designed for linear and time-invariant (LTI) processes, the control
performance may substantially degrade when the process behavior changes with
operating condition or time.
A PID control loop, with a simple algorithm and pure feedback mechanism, struggles
to face most of the above challenges. In the oil & gas industry, the downstream
operations have taken the lead in adopting advanced control technologies, mainly
due to process complexity (interactions), constraints, and optimization opportunities
from frequent changes in product demand, pricing, and economics. On the other
hand, the upstream exploration & production (E&P) have relatively simple process
flows but faces different challenges such as geographically dispersed assets, higher
variations in feed flow and quality, and measurement reliability.

5.1.2 Process Control Solutions Before MPC

The interaction and coordination among all the concerned process variables and con-
trollers are critical for controlling a wider operation area. A PID control loop controls
5.1 Why Advanced Process Control? 171

a specific variable in isolation and does not necessarily care about the interaction
and coordination with other controllers. Complex PID control schemes discussed in
Chap. 4 provide some capabilities to handle multivariable problems (Challenge #1
in Sect. 5.1.1). However, the inherent limitations of the PID feedback control leave
the rest of the above challenges unanswered. Besides, as the number of PID con-
trollers and supporting functional blocks increases, the control scheme’s complexity
increases exponentially. The implementation and maintenance become a significant
challenge (see Sect. 4.2 for complex loop structure).
Decoupling control (Sect. 4.1.2) (Gordon 2005; Shinskey 1994) is another
approach to address the interaction in multivariable control explicitly (challenge
#1). However, its complexity has limited its widespread implementation.
The Smith predictor (Smith 1957) is an early effort tackling the long time delay
problem (Challenge #2). The Smith predictor introduces an excellent control struc-
ture for compensating long dead time. However, due to the modeling requirement
and non-trivial tuning, the Smith predictor did not find widespread application in the
process industry (Åström and Hägglund 1995).
Nonlinear process dynamics is a more challenging problem, and there are not
yet mature theories or standard technologies for practical process control. Nonlinear
dynamics are usually addressed in one of the three ways:
1. Limiting the operating point to a local linear region.
2. Applying a nonlinear transformation on process variables.
3. Using a nonlinear control algorithm.
Most linear control solutions assume local linearity in the process dynamics. Signal
characterization (nonlinear transformation) is a popular technique for transforming
nonlinear variables to linear and is widely used. However, it is mainly on a case-
by-case basis. Gain scheduling (Sect. 4.1.5) has offered some help for managing
nonlinear process dynamics (Challenge #6), but it is for individual PID controllers
and is thus more or less ad hoc.

5.1.3 Advanced Process Control Technologies

Most process control technologies originate from the classical control theory. Over
the past decades, the unique challenges facing the process industry, as mentioned
in Sect. 5.1.1, forced the process control researchers and practitioners to select from
the rich pool of control technologies for those that can meet the process control
needs, or adapt them to address process control requirements. However, this effort
has not been easy due to the unique challenges of the process industry. The prevalent
process control techniques, from simple PID controller, complex PID schemes, to
model predictive control, all originated from the process control practice, developed
by process control practitioners, and propagate back to academia and then formalized
and improved.
172 5 Advanced Process Control

The success of the advanced process control technologies hinges on the tight
coupling between the control theory and practical needs. Model predictive control
(MPC) has been a significant breakthrough in process control technology (Cutler and
Ramaker 1980) that was developed by process control and for process control, with
a clear goal of addressing the specific process challenges (Sect. 5.1.1). The process
model at the center of MPC has provided the capability to handle multivariable
problems, compensate for long time delays, and apply optimization techniques to
honor constraints (Challenges #1, #2, and #3), all in a single controller. It is thus
a much more capable and elegant solution than complex PID schemes discussed in
Chap. 4.
The process control hierarchy (Fig. 1.10) shows that process control follows a
layered approach, with MPC as one layer above the baselayer control. MPC typically
runs at a lower frequency than baselayer control by a difference of one to two orders
of magnitude. Therefore, MPC is typically implemented on top of a baselayer control
rather than replacing baselayer control.
Figure 5.1 provides an outline of the process control architecture. At the bottom
is the PID-based baselayer control, which directly interacts with the process. At
the baselayer level, the control objective is primarily about maintaining individual
process variables such as temperature, pressure, flow, or level at their targets. MPC
is built on top of the baselayer and is concerned about variables or properties such as

Fig. 5.1 Architecture of advanced process control

5.1 Why Advanced Process Control? 173

quality, capacity, and yield. These higher-level control targets can only be achieved
by a coordinated effort of multiple baselayer control loops.
Many process variables cannot be directly, timely, or reliably measured using
currently available process measurement technology such as online analyzers or lab
tests. In this case, inferential property estimator (IPE) or soft sensors can provide the
missing measurements or improve the measurement quality. There is another type
of calculated value called key performance indicators (KPI) that provide the perfor-
mance indications for various process properties, including controllers for monitoring
purposes.
MPC has been widely used in the process industry and is currently the de facto
standard for advanced process control. However, MPC is not a panacea. Nonlinear
time-varying dynamics and unreliable measurements (Challenges #5 and #6) are still
open challenges in both theory and practice.

5.2 Model-Based Applications

For any process, be it industrial, economic, social, or human behavioral, there must
be a basic understanding of its behavior in order to control it. Usually, the more we
understand it, the better chance we can control it.
A process model is a popular approach to represent our understanding of the
process. The process model extracts the cause and effect relationships between the
process input and output and abstracts the behavior with a mathematical representa-
tion (see Fig. 2.23). Advanced calculations can then be applied to the mathematical
model rather than the actual process, thus opening up the opportunity for sophisticated
data processing including real-time control and optimization. It is not exaggerating
to say that all advanced process control applications will require a process model of
a particular type.

5.2.1 Process Analytics and Process Models

There are many types of models available for various purposes. See Fig. 5.2 for an
illustration of the different applications in advanced control and their model require-
ments. The model type can range from an implicit heuristic model for PID to rigorous
first-principle models for an operator training simulator (OTS). Typically the more
sophisticated the application is, the higher the model requirement. For example,
• PID controller tuning makes implicit assumptions on the process model. As dis-
cussed in Chap. 9, the trial-and-error tuning method requires a basic understand-
ing of the process dynamics in terms of high/low gain, slow/fast dynamics, sta-
ble/unstable response.1 On the other hand, model-based PID tuning relies on an

1 A fuzzy logic model is a more structured way of formulating this implicit process model practice.
174 5 Advanced Process Control

Fig. 5.2 Type of models for control applications

explicit process model to calculate the tuning parameters. The utilization of an

explicit model results in more confidence in the tuning parameters and leads to
more consistent performance than the empirical trial-and-error approach.
• Model predictive control (MPC) requires linear dynamic models, while real-time
optimization (RTO) is based on steady-state first-principle models.
• Dynamic simulation and operator training simulators (OTS) require the highest
rigorousness and fidelity in the models. Consequently, the dynamic simulation
model is also the most costly and time-consuming to develop and maintain.
See Table 5.1 for a list of the most common model types. As a general guideline,
the model type is application-dependent, and the best model type is the one that is
fit-for-purpose. A process model merely provides a simplified approximation to the
actual processes. The challenge is where and how to simplify it. “All models are
wrong, but some are useful” (Box 1976).
For advanced control applications, there are two major categories of models: static
and dynamic. A static model does not consider the time effect in the dynamics and

Table 5.1 Type of process dynamics

Linear vs. Nonlinear
Dynamic vs. Static
Deterministic vs. Stochastic
Parametric vs. Non-parametric
First-principle based vs. Empirical
Time-domain vs. Frequency domain
Continuous vs. Discrete
5.2 Model-Based Applications 175

focuses only on the steady-state behavior. The model output is related to the current
model input and does not consider any effect from the past values. For example, the
flow calculation from differential pressure is a simple static model:

Fv = C P/ρ (5.1)

where P is the measured pressure difference across an orifice, and ρ is the density
of the fluid. C is the parameter to be determined and specific to each orifice plate.2
A dynamic model has a memory of time. The model output is related to both the
current and the past input values. For example, a simple first-order linear dynamic
model is represented as

K p −θ s
G(s) = e (5.2)
τs + 1
 
y(t) = K p 1 − e− τ u(t − θ )
t−θ
(5.3)

where u is the process input, y(t) is the process output. The current output y(t) is
determined by u(t − θ ), which is θ minutes in the past.
A process model has two critical aspects: the model structure and the model
parameters. The model structure describes the type of (or lack of) cause and effect
relationships among the process variables; and the model parameters representing
the exact dynamic behavior for each relationship. For example, Eq. 5.1 defines a
square-root-type causal relationship between the input pressure P and the output
flow Fv , while Eq. 5.2 shows a linear dynamic relationship between the input u and
output y. The model parameters, C in Eq. 5.1 and K p , τ, θ in Eq. 5.2 define the exact
relationship.
When the past inputs and outputs are known, a dynamic model can predict the
future outputs.3 Conceptually, the future output is given by
⎧  
⎪
⎪ y(t + 1) = k p
t−θ +1
1 − e− τ u(t − θ + 1)
⎪
⎨
..
. (5.4)
⎪
⎪  
⎪
⎩ y(t + θ ) = k p 1 − e− τ u(t).
t

If we know the desired future output values, we can somehow “back-calculate”

the required inputs through matrix “inversion.” This back-calculation of the process
inputs from the desired future outputs is the basic principle of MPC (see Fig. 2.23).
Of course, the actual implementation is much more complicated.

2 With the recent development in data science, the static model can also be called process analytics
or descriptive analytics.
3 Dynamic models can also be called predictive analytics because of the capability of prediction.
176 5 Advanced Process Control

5.2.2 Data Analysis and Process Identification

As there are many types of process models, there are also many ways of obtain-
ing the process models. Static models are typically developed based on physical
knowledge and engineering insight, supplemented by trial-and-error experiments to
decide the model structure (i.e., the equation type). The equation’s coefficients are
typically determined by experimental data via data analysis techniques such as linear
regression, nonlinear equation solvers.
Dynamic models can be developed via theoretical modeling or process identifica-
tion. Theoretical modeling constructs the model based on physical knowledge of the
process behavior, while process identification builds the model from process data.
The former is also called white-box modeling, and the latter black-box modeling.
The dynamic model used by MPC is typically a linear dynamic model, and the
technique for obtaining the model is process identification. Process identification or
system identification is a very active research area with a vast scope (Ljung 1999;
Niu et al. 1992, 1994). In a nutshell, process identification is the methodology and
techniques to select, given the process input/output data and subject to a specific
criterion, the mathematical model of a given type that can best represent the actual
process (or rather, the process data). The three ingredients for process identification
are thus as follows:
1. Process data: The process data can be from the physical understanding of the
process, a well-designed plant test, or simply from the normal plant operation.
Since process identification is a data analysis technique, the data quality deter-
mines the final model quality. The goal of (black-box) process identification is
“data-in, model-out.” However, if the process data does not contain sufficiently
rich information, process identification can often become “garbage in, garbage
out.”
2. Process model: The process can be represented with various models, such as linear
versus nonlinear, time-variant versus time-invariant, deterministic versus stochas-
tic, discrete-time versus continuous-time, time-domain versus frequency domain,
lumped-parameters versus distributed-parameters, difference equation versus dif-
ferential equation, input–output versus state-space (Corriou 2018). However, the
most commonly used model type in process identification is the linear, time-
invariant (LTI) difference equations.
3. Criteria: Process identification aims to find the mathematical model to “best”
approximate the process input/output relationship. The criterion defines in what
sense the model is the “best.” The most commonly used criterion for process
identification is the least-squares criterion, a quadratic function of the estimation
errors.
In other words, process identification is a data-driven technique where the “black-
box” process is systemically disturbed, and the response data is gathered. The
dynamic cause and effect behavior is then abstracted to a mathematical model based
on the input/output response data under the assumption of a specific model structure
and particular criteria for the goodness-of-fit (Ljung 1999).
5.2 Model-Based Applications 177

Linear time-invariant dynamic empirical models provide a good trade-off between

complexity and fidelity and are preferred for advanced process control. Within this
category, there are still many types, such as the following:
1. Finite step response model (FSR) and finite-impulse-response model (FIR).
2. Transfer function model, including first-order-plus-time-delay model (FOPTD)
and higher-order models.
3. State-space models.
Except for the simplest model structure with clean data (e.g., Eq. 5.2), specialized
process identification software tools are needed to analyze the data and produce the
model, which consists of two major steps: structure determination and parameter
estimation. The quality of the data determines the quality of the model, which in
turn dictates the final control performance. Process identification is thus the vitally
important step in the MPC applications.

5.2.3 First-Order-Plus-Delay-Time (FOPDT) Model

For process control purposes, the first-order-plus-delay-time (FOPDT) model plays

a vital role because of its simplicity. Most real-world processes can be approximated
with this simple FOPTD model for most control applications. The FOPTD model
uses three parameters to describe when, how much, and how fast the output variable
y(t) responds to a change in the input variable u(t):
1. Process gain K p : By moving the input variable by one unit, how much change
(direction and magnitude) will be eventually caused on the output variable.
2. Time delay or dead time θ : After a change is made to the input (the manipulated
variable, or MV), when will the output variable (the controlled variable, or CV)
start to move?
3. Time constant or lag time τ: The amount of time that takes from the moment
that the output variable starts to respond until the output value reaches 63.2%
of the new steady-state value. The time constant describes how fast the process
dynamics are.
The transfer functions for the first-order-plus-dead-time (FOPDT) model, in both
continuous-time and discrete-time, are given in as follows:

K p −θ s  
→ y(t) = K p 1 − e− τ u(t − θ )
t−θ
G(s) = e (5.5)
τs + 1
b z −d−1
G(z −1 ) = → y(k) = a y(k − 1) + b u(k − d − 1) (5.6)
1 − az −1
with a = e−Ts /τ , b = K p /τ, d = θ/Ts (5.7)

where k denotes the current sampling time, y is process output, u is the process input,
and θ is the time delay. a, b, and d are the model parameters to be determined. Note
178 5 Advanced Process Control

that θ is the time delay in continuous time, and d = θ/Ts  is the same time delay
in discrete time, with Ts being the sampling time.
The FOPTD model can be conveniently and intuitively related to the step response.
When the process noise level is low, the three model parameters K p , τ, θ can often be
calculated from the step response with sufficient accuracy, as shown in Fig. 5.3. For
example, the process gain can be readily calculated as the change in output (y) over
the change in input (u) K P = y/u, the dead time is given by θ , and the time
constant by τ. The three parameters in continuous-time are related to discrete-time
by Eq. 5.7. Modeling from a clean step response is the most commonly used method
for obtaining a FOPDT model in practice. Note that the gain is calculated based on
changes in input and output and does not concern the steady-state values. In PID
control, the control output is calculated as the change in control action rather than
the actual value (see Sect. 3.1.6).
A second method is via solving a set of linear simultaneous equations. From
Eq. 5.6, there are three parameters to determine a, b and d. The time delay d is
nonlinear to the output y(t), so it must be determined first either visually or via
trial-and-error. Once the time delay d is known or assumed, the parameters a and b
can be calculated by solving the following equations:
⎧
⎪
⎪a y(k − 1) + b u(k − d − 1) = y(k)
⎪
⎪
⎪
⎨a y(k − 2) + b u(k − d − 2)
⎪ = y(k − 1)
a y(k − 3) + b u(k − d − 3) = y(k − 2) (5.8)
⎪
⎪ ..
⎪
⎪ .
⎪
⎪
⎩
a y(k − n) + b u(k − d − n) = y(k − n).

Fig. 5.3 Step response and step response model. Left: stable process, right: ramp process
5.2 Model-Based Applications 179

Fig. 5.4 An example of process identification (modeling) software tool

Equation 5.8 is a set of overdetermined simultaneous equations with two unknowns,

a and b. There are many tools, even spreadsheet software, for solving equations.
The generic approach for modeling is to use a dedicated process identification tool.
A process identification software tool can work on a wide variety of models, including
the first-order-plus-time-delay models and the finite-impulse-response model (FIR)
in Eq. 5.18. Figure 5.4 is an example of a generic process identification software.
For completeness, a first-order integrating model is typically used as follows:

Kp  
e−d s y(t) = K p t − 1 − e− τ
t−d
G(s) = → u(t − d). (5.9)
s (τ s + 1)

5.3 Inferential Property Estimation

Inferential property is one of the most successful model-based applications and has
been used for multiple purposes (Blevins et al. 2003; Marlin 2015), including the
following:
1. Improving process measurements: Poor process measurement is Challenge #5
in Sect. 5.1.1. Many desired measurements are either not available or unreliable.
Using inferential property based on some more reliable measurements is one
practical and effective approach to improve.
180 5 Advanced Process Control

2. Nonlinear transformations: Most advanced control techniques, including MPC,

are based on linear models. To directly control nonlinear process dynamics would
need an entirely different control technology. However, using nonlinear inferen-
tial property can isolate the nonlinear part of the process dynamics and results in
a more straightforward and cleaner solution. For example, controlling a pressure-
compensated temperature (PCT) is better than controlling both pressure and tem-
perature in a multivariable controller. The nonlinear correlation between temper-
ature and pressure can be addressed outside the linear controller.
Inferential property has been more popularly known as a soft sensor. The recent
upsurge of big data and the resurgence of artificial intelligence bring some new
terminology such as descriptive analytics, predictive analytics, and prescriptive ana-
lytics, which are different forms of inferential properties that have been in existence
in process control for many years. The fundamental idea is that for process variables
that can not be directly or reliably measured, we can use other commonly available
measurements to “infer” their values. The inferred value is thus called inferential
property.
Inferential property estimate (IPE) can be used for open-loop monitoring or
closed-loop control. Monitoring has a lower requirement on reliability than control,
and thus has found more widespread use for IPE. Many recent technologies such as
performance monitoring, abnormal situation management, equipment health moni-
toring, fault detection, data validation are wholly or partly dependent on inferential
property estimates (process analytics).
Inferential property can be broadly categorized into static and dynamic (see
Fig. 5.5). The static inferential property is usually based on spatial information where
several variables are used to infer another, while the dynamic inferential property
uses both spatial and causal information. For the causal information, both current
and past values are used to predict future values. The final form and technique of the
IPE are mainly dependent on the peculiarities of the specific application.

Fig. 5.5 Common model

types
5.3 Inferential Property Estimation 181

Models based on first principles relate vital physical features such as pressure and
temperature to the process dynamic behaviors. They are thus valuable in establishing
relationships between the process variables. Unfortunately, this approach has severe
limitations due to the complexities of physical processes. For example, a typical
distillation column may need several hundreds of differential equations and an over-
whelming number of parameters to characterize the thermodynamic relationships. It
is thus a significant effort to formulate the equations and solve them (Marlin 2015).
For many applications, it is not worth the effort and complexity to go with first-
principle models. In practical applications, many decisions are a result of a trade-off
between desire and practicality.

5.3.1 Static Inferential Property Estimation

Static IPE is typically based on a mathematical formula or static model. Most of the
time, the model is a nonlinear equation derived from physical understanding. For
this reason, the IPE can be highly flexible and powerful. This approach is also called
the analytical approach.
Static IPE has been in use for a long time under different names. Many examples
of static IPE exist even in this book. Below is a list of some IPE calculations in
various chapters:
1. Flow calculation (Sect. 10.3): Flow is never directly measurable. Instead, the flow
rate is inferred from other variables such as differential pressure, fluid velocity,
or density. See Table 2.3 for the measurement principles and Table 10.3 for the
inferential formulas. In fact, even the flow velocity is inferred from other directly
measurable variables such as frequency, elapsed time, or voltage. Most process
variables are not the direct measurement from the sensors. The process variable is
inferred from sensor output through a correlation governed by physical principles
(Marlin 2015).
2. Pressure-compensated temperature (PCT) (Sect. 10.4): Pressure and temperature
are highly correlated for gas and steam (Process challenge #1). The correlation
can be embedded into one new variable called pressure-compensated temperature
(PCT) (Eq. 10.20) and be used as one controlled variable instead of two. This PCT
is a typical static IPE.
3. Surge margin calculation (Sect. 11.3): The compressor surge margin can not
be measured and is typically inferred from pressure and flow (see Eqs. 11.36
and 11.37).
4. Signal characterization (Sect. 10.1): Signal characterization is a common approach
to transform a signal, typically from nonlinear behavior to linear. Signal charac-
terization is supported by most control systems, typically as a look-up table. With
the table look-up approach, the signal characterization can perform a generic
free-form transformation.
182 5 Advanced Process Control

5. Key Performance Indicator (KPI): Most KPIs are inferred from other measure-
ments or values.
Although in different shapes and forms, the static inferential properties have the
following characteristics in common:
• Based on engineering insight: The static inferential property estimators are typi-
cally backed by deep physical insight and engineering principles.
• Flexible in format: The estimator is highly flexible in format and can use any
mathematical formulas supported by the target system.
• Low cost: An inferential property estimator is merely a calculation in the control
system.
• Reliable: The inferential property estimator for basic measurements such as pres-
sure, temperature, flow, and level are regarded as direct measurements because
they are sufficiently accurate and reliable.
We here provide an example showing how a static IPE is developed from first-
principle engineering knowledge.
Example 5.1 Flow meter with an orifice plate. An orifice plate is a thin plate with
a hole in the middle (Fig. 5.6). The orifice is placed in a pipe in which fluid flows.
When the fluid reaches the orifice plate, the fluid is forced to converge to go
through the smaller hole; the point of maximum convergence occurs shortly down-
stream of the physical orifice at the so-called vena contracta point. As it does so, the
velocity and the pressure change. Beyond the vena contracta, the fluid expands, and
the velocity and pressure change once again. By measuring the difference in fluid
pressure between the normal pipe section and at the vena contracta, the volumetric
and mass flow rates can be inferred.
We can use Bernoulli’s equation to calculate the flow rate. Between point ➀ and
➁, we have

Fig. 5.6 An orifice plate for

DP measurement
5.3 Inferential Property Estimation 183

1 1
P1 + ρ1 v12 = P2 + ρ2 v22 . (5.10)
2 2
Assuming the change of fluid density across the orifice is negligible (ρ1 = ρ2 ),
and by continuity of mass, A1 v1 = Av2 = Fv , where A1 and A2 are the cross area
of the orifice and the vena contracta, and Fv is the volumetric flow, we then have
 
1 1 Fv2 F2 1 1 1
P1 − P2 = ρ v22 − v12 = ρ 2
− v2 = ρ Fv2 2
− 2 . (5.11)
2 2 A2 A1 2 A2 A1

Define P = P1 − P2 ,
  
A2 P π d2 2 P P
Fv =  2 =  = Cv (5.12)
1 − (A2 /A1 )2 ρ 4 1 − β4 ρ ρ

where d is the diameter of the orifice and D is the diameter of pipe, and with
2
1 1 A2 d d
A1 = π D 2 , A2 = π d 2 , = = β2, β = .
4 4 A1 D D

Cv is called the flow coefficient. Once an orifice is built, the flow coefficient remains
unchanged. The mass flow rate Fm is given by

Fm = ρ Fv = C P · ρ. (5.13)

The generic and precise equation for mass flow rate measurement is given by
ISO5167 (2003) standard4 as

Cd π 2
Fm =  ε d 2 P ρ (5.14)
1 − β4 4

where Cd is the discharge coefficient, ε is the expansion factor, d is the bore diam-
eter, and D is the pipe diameter. They are flow constants that are experimentally
determined after the orifice is built.

5.3.2 Dynamics Inferential Property Estimation

Inferential property has become popular with MPC. With MPC, the attention has
shifted from individual process variable control (flow, pressure, temperature, and

4Note that ISO 5167 (all parts) is applicable only to flow that remains subsonic throughout the
measuring section and where the fluid can be considered as single-phase. It does not apply to the
measurement of pulsating flow.
184 5 Advanced Process Control

level) to product quality control. However, the product quality is not as readily and
reliably available as the conventional process variables. Many properties are not
measurable, and those that can be measured via online analyzers and lab sampling
suffer from the problems of low availability and poor reliability, which are severe
challenges for online control:
1. Availability: The process variable is not directly measurable. Many process vari-
ables that need to be controlled are simply not measurable for technical or eco-
nomic reasons. In this case, it is often possible to use existing measurements to
infer the value of the desired process variable.
2. Reliability: The process measurement is not sufficiently reliable for real-time con-
trol. Conventional process measurements such as flow, pressure, temperature, and
level have been widely used for closed-loop control, although they occasionally
suffer from measurement failures. Quality measurements through online analyz-
ers are notorious for poor reliability and low up-time. Frequent maintenance such
as calibration is required but often not received.
3. Speed of response: The online process measurements are too slow for real-time
control. Many online analyzers, such as gas chromatography (GC), have a slow
response time producing the result. Lab sampling is even less responsive. The
long time delay is a significant challenge for feedback control.
With the advance in technology, a modern operating plant usually has abundant and
redundant process measurements. It is thus possible to utilize the more reliable and
more responsive measurements to infer other process variables.
Static inferential property estimation typically uses static, nonlinear, first-principle
models based on engineering insight of the process variables. Dynamic inferential
property estimation is typically data-driven and limited to linear dynamic empiri-
cal models. For example, the overhead composition in a distillation column can be
related to the temperature and pressure via a linear dynamic empirical model. On
the other hand, for a high purity distillation tower, the impurity level is typically a
nonlinear function of the inputs (e.g., temperature). A linear model is insufficient to
support the process operation within a wide range of variations in purity. A more
sophisticated nonlinear model such as a neural network or a linear model with a
nonlinear transformation may become necessary.
Building a dynamic inferential property estimator is relatively straightforward
among all model-based applications. Although data-driven, a good understanding of
the physical property and cause and effect relationship is essential. The data used
for parameter estimation can be historical data if sufficient excitation is present to
characterize the dynamic behavior. Otherwise, plant tests should be conducted to
obtain high-quality data.

Example 5.2 Inferential property estimator for a stabilizer column. For the stabilizer
column in Fig. 2.24, the total vapor pressure (TVP) is a control target, with an online
analyzer installed downstream of the column. However, the analyzer readings often
were not available or not credible due to calibration or maintenance issues.
5.3 Inferential Property Estimation 185

A PID controller was configured to maintain the TVP value on target. However,
the challenge was that the control target was the TVP value inside the column, but
the online analyzer was located a long distance downstream. The transport delay
and the analyzer response time added up to 40 minutes, which was impossible for a
feedback PID controller to handle. The analyzer’s long time delay and low up-time
had prevented the quality controller (AC-101) from running in closed-loop ever
since the plant startup.
An inferred TVP value from the pressure and temperature was developed later and
used as the PV for the AC-101 controller. The inferential property was estimated
from two process variables, MV1 (tray 9 temperature) and MV2 (column pressure),
with the transfer function as follows:

K p,1 e−θ1 s K p,2 e−θ2 s


PV(s) = MV1 (s) + MV2 (s) (5.15)
τ1 s + 1 τ2 s + 2

 was the estimated TVP value. MV1 and MV2 are the two process variables
where PV
used for the estimation. θ1 and θ2 were the time delays between the two inputs and
the measured property value. Keeping in mind the two issues we try to address with
online analyzers or lab samples are as follows:
1. Long time delay: The long delay time was a significant challenge for control.
In this example, it took about 40 minutes for the TVP analyzer to produce the
response, partly because of the long response time of the analyzer itself but mainly
because the analyzer was located a long distance downstream of the stabilizer
column for design reasons.
2. Low availability: The online analyzer had chronic problems with calibration and
maintenance. More than half of the time, the results were either not available or
could not be trusted.
One of the goals for the inferential property estimator was to “remove” the time
delay so that the dynamics between the TVP and the temperature/pressure become
more responsive. The simple approach was to remove the common delay θ0 =
min (θ1 , θ2 ) ≈ 36 min from the transfer function in Eq. 5.15:

K p,1 e−(θ1 −θ0 ) s K p,2 e−(θ2 −θ0 ) s


TVP(s) = MV1 (s) + 
MV2 (s) + offset. (5.16)
τ1 s + 1 τ2 s + 2

The IPE value with the common delay removed was then used as the controlled
variable for real-time control, either PID or MPC.
 in Eq. 5.16 is the static error that may develop between the pre-
The term offset
dicted value and the analyzer reading over time. The actual value from the analyzer
 The inclusion of
or lab sampling is fed back to the estimator to update the offset.
feedback is a practical yet effective approach to compensate for the discrepancy
between the process model and the actual plant. The main reasons for this model-
plant mismatch include the following:
186 5 Advanced Process Control

1. The process model is always over-simplified, especially in the presence of non-

linearity.
2. There are always unmeasured disturbances that can not be captured.
3. The process dynamics may drift away from the region defined by the data that
the model is derived from.
The vital step in taking the feedback from online analyzers is the timestamp align-
ment. That is, the offset is a deviation between the predicted value and the actual
value. Since the predicted value has removed the common delay θ0 from the model,
to compare the predicted value with the actual value, the predicted value needs to
add the delay back to align with the analyzer value in time. See Fig. 5.7 for an illus-
tration of IPE for feedback control, and compare it with standard feedback control
in Fig. 2.2.
A sample implementation of the inferential property estimator for TVP is given in
Fig. 5.8, where the gain, lag, and delay are primitive function blocks in standard DCS.
Depending on the target system for implementation, there may be more advanced
function blocks to lump the calculations together. For example, one function block
may be able to perform all three functions.
Figure 5.9 shows the inferred and the actual TVP values based on historical data,
with feedback correction based on the online TVP analyzer value once every hour.

Fig. 5.7 IPE for feedback control

Fig. 5.8 A sample implementation of IPE

5.3 Inferential Property Estimation 187

Fig. 5.9 Inferred TVP values, with hourly feedback from analyzer

Fig. 5.10 Inferred TVP values, with daily feedback from analyzer

Figure 5.10 shows the inferred TVP values with the same IPE but with feedback from
the TVP analyzer once every day. The IPE provides a decent estimate with analyzer
feedback once an hour. Even if the feedback input is reduced to once a day, the IPE
still offers valuable information for monitoring.

An open challenge for the inclusion of feedback in the IPE is the online quality
assessment on the feedback:
1. Is the analyzer feedback credible and reliable for correcting the predicted value?
2. If the feedback is from lab samples, how often is the lap sampling, and how to
reliably match the timestamp of the lab result with the actual time the sample is
taken?
188 5 Advanced Process Control

3. If the feedback or lab sample is not available, how long should the predicted value
be used for online applications such as control?
A minimum requirement for the data validation is that the feedback data, either from
an online analyzer or lab test, need to provide a quality indicator as good, uncertain,
or bad. Good data can be used for closed-loop control and uncertain data only for
monitoring.
The inferred value can provide sufficient accuracy and reliability for closed-loop
control with a high-quality model and periodic feedback from the online analyzer.
Even without feedback, the inferred property can often be good enough for real-time
control to remain closed-loop for several hours and thus continue to deliver better
performance than manual control.
Neural network (NN) has received much attention for inferring process variables
with a significant nonlinear relationship with the independent inputs. However, the
drawback of neural network models is that it typically requires a lot more “clean” data
to train the network than parsimony models do. For IPE applications where the “good”
measurement data from the online analyzer or lab are insufficient, the quality of the
NN model is typically the bottleneck. In the process industry, data is abundant, but
“good” data is scarce. Therefore, the NN model should be considered only if a simpler
linear model based on engineering insight is not available or not sufficient. It works
best for applications where the dynamic cause and effect relationship between the
specific inputs and outputs is known from fundamental knowledge and engineering
insight, with only the exact form of nonlinearity and the parameters to be estimated.
Modeling without understanding the physical principle of the cause and effect
relationship is usually counter-productive. This is especially true for multi-input
multi-output processes. For this reason, first-principle models are always preferred
over dynamic empirical models, which in turn is preferred over neural network
models. In other words, whenever possible, choose white-box models over gray-box
models. The last resort is black-box models.

5.4 Model Predictive Control—The Concept

Model predictive control (MPC) mainly refers to dynamic matrix control (DMC)
(Cutler and Ramaker 1980) and generalized predictive control (GPC) (Clarke et al.
1987). MPC is a vast topic deserving separate books for systematic coverage (Corriou
2018; Garcia et al. 1989; Lee 2011; Morari and Lee 1999; Qin and Badgwell 2003;
Rawlings et al. 2017). Here we will skip the theoretical derivations and focus only
on the fundamental concept and the best practice work process in implementation
and applications.
5.4 Model Predictive Control—The Concept 189

5.4.1 MPC as Multivariable Control

Model predictive control is inherently a multivariable control algorithm. It is capable

of addressing process challenge #1 in Sect. 5.1.1 elegantly as compared to a com-
plex PID control scheme in Sect. 4.1.2. Unlike complex PID schemes where the
multivariable interaction is addressed explicitly by separate function blocks (thus
a white-box approach), MPC relies on multi-input multi-output (MIMO) models to
implicitly handle the multivariable interactions (a black-box approach). The efficient
and reliable handling of multivariable interactions empowers APC to control a large
process area, such as an entire column or unit, with a single controller.
Figure 5.11 shows a generic distillation column, along with the key process vari-
ables. A distillation column is the most widely used unit operation for separating
mixtures. It typically consists of a tray section, a condenser, a reflux drum, and a
reboiler. The section above the feed tray is called the rectifying section (or enriching
section), and the part below is called the stripping section. In a distillation column, the
overhead temperatures and pressures are highly correlated, and all have a substantial
impact on the composition of the top product. The overhead conditions also heavily
interact with the bottom, affecting the bottom product quality. The feed temperature
and flow rate are significant disturbances to the entire column operation.
The interaction between the process inputs and outputs can be illustrated in
Fig. 5.12, where G 12 designates the transfer function between input #1 and output
#2. As the number of inputs and outputs increases, the number of transfer functions
can grow rapidly.

Fig. 5.11 A generic fractionator column with typical measurements

190 5 Advanced Process Control

Fig. 5.12 Interactions in a MIMO process

There are several critical aspects in the multivariable representations as follows:

1. Selecting the process variables: Selection of the process inputs (including manip-
ulated variables and disturbance variables) and the process outputs (the con-
trolled variables) is the first step and often goes back and forth before finalization.
MPC typically includes three types of variables in the controller: manipulated,
controlled, and disturbance variables. Some MPC implementations include con-
strained variables as the fourth type of variable. However, the constrained variable
can be treated as a controlled variable with a setpoint range.
2. Determining the model structure: The most important aspect of the multivariable
model is the model structure, which includes the cause and effect relationships
among the input and output variables. From Fig. 5.12, there can be as many as
m × n potential cause and effect relationships for a multivariable process with m
inputs and n outputs. As reported in a survey paper by Qin and Badgwell (2003),
the largest controller has 693 inputs and 283 outputs. The number of potential
relationships (thus models) is 693 × 283 = 196,119!
In reality, however, most of the relationships do not exist or are negligible. Deter-
mining the significant input/output process variables and identifying the non-
negligible cause and effect relationships is a vital step in MPC applications. In
complex PID control solutions (Sect. 4.1.2), this step is tedious and is thus more
meticulous because each relationship will eventually be implemented explicitly
with a PID control loop. On the other hand, with MPC, the multivariable interac-
tions are addressed implicitly with a model structure selection. This step becomes
considerably easier but is more prone to mistakes and oversights if the necessary
cautions are not taken.
5.4 Model Predictive Control—The Concept 191

3. Estimating the model parameters: The determination of the model structure is

still an art to a certain extent. After the model structure is decided, the parameter
estimation is relatively more straightforward. There are many mature techniques
for parameter estimation, so it is more science than art. Most commercial software
tools are capable of producing the model parameters once the model structure is
given.
In practical applications, the multivariable interactions are usually represented in
an intuitive and convenient table format called the cause and effect table (CET), as
illustrated in Table 5.2.
The most challenging step in MPC is to formulate operating requirements into a
control strategy. The knowledge and skills required for controller variable selection
and model structure determination are the same for complex PID control discussed
in Sect. 4.1 and MPC. An in-depth understanding of the operating objectives and
the process dynamics is essential, and a systematic process analysis is vital to select
the appropriate model structure to and ensure that the critical process characteristics
are captured in the control design. Chapter 6 is dedicated to a structured approach in
process analysis applicable to both complex PID control and MPC.
Inferential property is often an integral part of MPC. Once the inferential prop-
erty is built, it can be used the same way as with other directly measurable process
variables. Many controlled variables are inferential properties instead of direct mea-
surements. It is vital to have the inferential properties developed and commissioned
before MPC step testing starts.
As an example, Fig. 5.13 shows all the models produced by the process identifi-
cation tool as a full matrix, with both the step response model and the parametric
model superimposed on each other. Unfortunately, many models are not of high qual-

Table 5.2 Multivariable interactions

192 5 Advanced Process Control

Fig. 5.13 Raw model matrix for MPC

ity (low confidence). The decision for the final model matrix is somewhat subjective,
depending on many factors such as the understanding of the process dynamics and
the trust level on the plant test data. The final model matrix may be like Fig. 5.14 or
different depending on the preference and competency of the engineer designing the
MPC controller.

Fig. 5.14 Final model matrix for MPC

5.4 Model Predictive Control—The Concept 193

One of the weaknesses of most MPC software is that the calculation does not
differentiate between good and bad models. All models in the controller are given
equal trust. Discriminative treatment of the models based on the model quality is
an open challenge for improving MPC. One approach is to separate the process
models into prediction models and control models. The control model is a subset of
prediction models that represent the key cause and effect relationships with higher
quality. The implicit “matrix inversion” inside the MPC only involves the control
models.
All commercial software has the built-in feature of model validation or model
quality assessment (the goodness-of-fit), which is an essential checkpoint on the
model quality. No corners should be cut related to model quality.

5.4.2 MPC as Predictive Control

In addition to decoupling multivariable interactions, the dynamic process models can

predict future process behavior (Rossitier 2003).
Unlike purely reactive feedback control, MPC can “predict” the future process
output with the process model. The dynamic nature of the model means that the
current output is a function of the past inputs and outputs. The future outputs can be
predicted based on the present and past process inputs with the same model. With this
predictive capability, APC can predict where the controlled variables are moving to,
then back-calculates which, when, and how much to move the manipulated variables
to correct the undesired deviations from control targets. As a result, the MPC control
actions are calculated based not only on where the process is currently running but
also on where it is predicted to be moving to in the next few minutes or even hours.
This predictive capability, a vast improvement over PID controllers, makes the control
performance much smoother than purely feedback-based control. A reactive control
compared to a predictive control is like a blind person walking with a stick (one-step-
ahead prediction) compared to a normal person walking in daylight (multi-step-ahead
long-range prediction).
With Eq. 5.2, the future prediction of the output y are given by
⎧
⎪
⎪ b u(k − d − 1) − a y(k − 1) = y(k)
⎪
⎪
⎪
⎪ b u(k − d) − a y(k) = y(k + 1)
⎪
⎪
⎪
⎨b u(k − d + 1) − a y(k + 1) = y(k + 2)
(5.17)
⎪
⎪ b u(k − d + 2) − a y(k + 2) = y(k + 3)
⎪
⎪ ..
⎪
⎪
⎪
⎪ .
⎪
⎩
b u(k − d + m − 1) − a y(k + m − 1) = y(k + m).

The corresponding finite-impulse-response (FIR) model with n parameters is given

by
194 5 Advanced Process Control
⎧
⎪
⎪ bn u(k − d − n) + · · · + b2 u(k − d − 2) + b1 u(k − d − 1) = y(k)
⎪
⎪
⎪
⎪ n u(k − d − n + 1) + · · · + b2 u(k − d − 1) + b1 u(k − d)
⎪
⎨
b = y(k + 1)
bn u(k − d − n + 2) + · · · + b2 u(k − d − 2) + b1 u(k − d + 1) = y(k + 2)
⎪
⎪ ..
⎪
⎪
⎪
⎪ .
⎪
⎩
bn u(k − d − n + m) + · · · + b2 u(k − d + m − 2) + b1 u(k − d + m − 1) = y(k + m)
(5.18)
where the three parameters a, b, d in the first-order-plus-delay-time (FOPDT) are
expanded to n + 1 parameters bn , . . . , b2 , b1 and d in the finite-impulse-response
model.
A crucial design parameter (sometimes exposed as a tuning parameter) is the so-
called prediction horizon. The prediction defines how far MPC predicts into the future
on the process response. The farther it predicts, the smoother (or more sluggish) the
control performance is. An analogy is driving at night versus driving in daylight.
The former has a short prediction horizon, while the latter has a longer one. Night
driving must be slower since the car can easily overrun the range of the headlights
at a higher speed.

5.4.3 MPC as Constrained Control

One of the distinctive MPC advantages is the ability to handle multivariable inter-
actions and constraints in a unified manner, thanks to the explicit process model.
The internal mathematical model opens the opportunity to utilize optimization tech-
nologies such as quadratic programming (QP) to produce control actions that meet
operating limits and constraints. There are several types of constraints as follows:
1. Constraints on manipulated variables: These are the hard limits on the manipulated
variables, such as valve saturation limits.
2. Constraints on the move size of the manipulated variables: These are the limits
on the rate of change in the manipulated variables for stability.
3. Constraints on controlled variables: These are the limits imposed on controlled
variables to avoid overshoots or minimize deviations from setpoints.
An MPC algorithm without considering constraints is a straightforward linear algebra
problem of solving matrix equations. However, MPC needs an optimization engine
to solve the equations with constraints. The quadratic programming (QP) algorithm
is a popular choice.
Many a time, it is not feasible to meet all the constraints specified. For this reason,
the constraints are further classified as hard constraints and soft constraints. Hard
constraints, such as valve limits, cannot be violated, while soft constraints are allowed
to be broken with a penalty. The overall objectives are thus two folds:
1. Fast response for setpoint tracking and disturbance rejection.
2. Minimum moves (typically valves) to conserve energy and reduce wear and tear.
5.4 Model Predictive Control—The Concept 195

The two objectives are often contradictory, and thus the optimal behavior is merely
a compromise between the two. If represented as an objective function, it would be
like the following:

[Objective Function] = [Total Control Error] + [Total Control Energy]

 2
J= r 2 y(k) − ysp (k) + s 2 (u(k) − u(k − 1))2 .
k

The tuning aims to strike a good balance between fast response and minimum effort.
Mathematically, this balance is determined by the relative weighting factors r and s.
As an example, consider the four reactors in Fig. 5.15. The reactors are exothermic.
The operating requirement is to maintain the temperatures in all four reactors at
152 ◦ C. However, at the start of the catalyst life cycle, the reaction heat generated by
reactor R-103 is more than enough to raise the T-104 temperature above its setpoint,
even after the heating medium flow FC104 is reduced to zero.
Without changing process design and the type of catalyst, one option for correcting
the temperature profile is distributing the control error among the four reactors,
especially between T-103 and T-104. That is, within the allowable range of the
temperature variation, lowering the temperature setpoints of T-101 to T-103 to below
the desired 152 ◦ C so that less heat is generated and T-104 temperature can come
closer to 152 ◦ C. This error distribution is a coordinated effort among all four reactors.
The temperature profiles with and without MPC are illustrated in Fig. 5.16.

Fig. 5.15 Example of constrained control

Fig. 5.16 Constrained control with MPC

196 5 Advanced Process Control

An alternative PID-based approach is an override control on R-103 temperature

control by R-104 temperature. Another option is to use valve position control (see
Sect. 4.1.4) to prevent the R-104 control valve from saturation. However, MPC is a
better approach to achieve a balanced temperature profile reliably. The optimization
engine in MPC can optimally distribute the control error among the four reactors
based on the temperature setpoint range of each reactor.

5.4.4 MPC as Optimizing Control

MPC provides a unified solution for addressing several process challenges (see
Sect. 5.1.1) elegantly in a single controller and can replace the complex PID control
schemes in most applications. However, the more deep-rooted reasons for MPC’s
massive success in the petrochemical industry are attributed to its capability to push
for more profitable operations (See Fig. 1.9).
MPC stabilizes the process operation and thus creates the opportunity to push
the operating point closer to the most profitable operation region to maximize eco-
nomic benefits. Usually, another optimization layer is above MPC to calculate the
more profitable operation points and passes them down to MPC as setpoints. The
optimization is based on some economic targets. The role of this profit optimization
layer is the same as the real-time optimization layer in Fig. 2.25. The optimization
problem can be formulated as follows:

minimize: P(X) (5.19)

subject to: f (X) = 0
and: Xlow < X < Xhigh

where X is the vector of model variables, and P(X) is the cost-based objective func-
tion. The equality constraints, f (X), are the open equation models. The inequality
constraints on the model variables, X, are the high and low constraints placed on the
system to represent plant operating constraints. Mathematical optimization software
such as quadratic programming (QP) minimizes the economic cost objective func-
tion while driving the model residuals to zero and honoring the process operating
constraints.
In MPC, instead of a full-fledged real-time optimization (RTO) application, the
profit maximizer is a light optimization engine, typically linear programming (LP)
algorithm, to drive a few key economic variables to the more profitable region. It is
usually not a global rigorous optimization function to replace RTO.
On the downside, the closer the operating point is pushed to the constraint limit,
the higher is the risk of violating the constraints. Therefore, a vital requirement for
profit pushing is that there must be a reliable constraint variable that can stop the
pushing. For every pusher, there must be a reliable stopper!
5.5 Model Predictive Control—Practical Considerations 197

5.5 Model Predictive Control—Practical Considerations

MPC application is more about the work process than technology. Compared with
traditional PID control, MPC is a change in the “control mindset.” This mindset
change is made through the exercise of rigorously formalizing and analyzing the
underlying problems, selecting input and output variables, representing complex
physical phenomena by descriptive yet simplified mathematical models, and using
mathematical tools for design and analysis.
Generally speaking, MPC has been an enormous success over the last few decades.
However, not every control problem is suitable for MPC (see Chap. 7). There has
been a tendency of overuse and even abuse of MPC technology. “To a man with
a hammer, everything looks like a nail.” There are many examples of disastrous
failures. The quality of implementation is highly dependent on the competence of
the implementors, much more so for MPC than for PID controls.
One of the crucial reasons for the failures is that most commercial MPC products
have packaged the MPC theory and algorithm into the software tools and usually are
not sufficiently transparent to the user. Many practitioners overly depend on the tools
and have treated the MPC application like a software application. The importance of
good process knowledge and the best practice work process is grossly undervalued or
conveniently bypassed. Under the pressure of project delivery, they tend to cut corners
on some critical steps in the MPC work process. Instead of spending sufficient effort
to obtain a high-quality process model and deciding on a sound controls structure,
they often prematurely rush to controller design with questionable model structure
and parameters. Late in the project, they end up spending much time tinkering and
twisting the controller tuning parameters betting on good luck to arrive at the “best”
performance and thus fall into the same trap as in the “trial-and-error” approach in
PID loop tuning (Sect. 9.2.2).
The process aspects of process control are the key to success; the systems aspects
(the bits and bytes) are largely irrelevant. If we can do an application with one
commercial system, then we can do it with any of them (Smith 2009).

5.5.1 Process Noises and Disturbances

Process noises and disturbances come from many sources and exhibit different fre-
quency, magnitude, repeatability, and stochastic properties. Most process noises and
disturbances do not meet the textbook descriptions, such as Gaussian distribution or
stationary trends. Thus a large part of the control effort is to handle the varying and
uncertain process dynamics caused by noises and disturbances. Table 5.3 lists some
of the familiar sources of disturbances in a typical operating plant.
The long-range model prediction in MPC has successfully addressed the dynamic
uncertainty caused by noises and disturbances. The control action is calculated based
on the overall long-range prediction trajectory and is not un-proportionally biased
by individual noise or disturbance along this trajectory.
198 5 Advanced Process Control

Table 5.3 Examples of Source Examples of noises and disturbances

process disturbances
Process Steam pressure changes, recycle loops
Measurement Measurement noises, failures
Valves Stiction, slip, deadband, saturation
Controllers Tuning, interactions, windup
Operator Manual interventions, operator “Sweet Spot”
Environment Daily and seasonal changes in temperature/pressure
Market Supply and demand balancing, feed/product prices

5.5.2 Baselayer Control Versus Advanced Control

For multivariable process control, there is a significant overlap between baselayer

control and MPC in terms of applicability. The decision between baselayer control
and MPC solution is often a difficult one, and sometimes even a “political” one. One
crucial lesson in process control is, the best solution is not the most sophisticated
one, but the one that is simple and fit-for-purpose. As a result, it is not always good to
go with MPC when a PID control scheme can serve the purpose. There is a threshold
to pass to start using MPC, typically with three prerequisites as follows:
1. A commercial software license to apply for MPC: Most real-world MPC applica-
tions are based on commercial software packages. It is not as trivial as it sounds
to choose the right vendor.
2. Hardware infrastructure to run the application: Some MPC software is built in
DCS modules, while others require dedicated machines.
3. Engineering expertise to build or maintain the MPC application and the necessary
training to run (and the mentality to accept) MPC applications by operations.
In an operating site, the first MPC application should be a carefully selected one with
a sufficiently high return on investment (ROI) to justify the application and showcase
the benefits. A failure in the first application may set back the MPC acceptance by
many years (Smith 2009).
There are two schools of thought regarding the division between baselayer control
and advanced control. One advocates keeping as much as possible baselayer PID
controllers since PID control is more reliable in performance, and with multiple
PID controllers, the risk is distributed. The role of MPC is primarily for handling
constraints and pushing for profits. The other school of thought is to put everything
under MPC to the extent that MPC directly manipulates all the control valves and
other final control elements. All the PID controllers will be bypassed. The argument
is that this approach provides total control and eliminates the inefficiency caused
by the PID controllers in between. For example, a PID controller introduces extra
dynamics between setpoints and controlled variables and thus slows the response or
introduces limitations.
5.5 Model Predictive Control—Practical Considerations 199

There are pros and cons to both approaches. In practice, it often becomes a personal
preference on where to draw the line. The final control strategy is often a healthy
mix of baselayer PID and advanced MPC for most applications: the baselayer PID
controller keeps the individual process variables on target, while MPC supplies the
setpoints for these PID controllers.
Considering the number of valves and control variables in the fractionator example
(Fig. 5.11), it would be very challenging, albeit not impossible, to bypass baselayer
PID loops, including the level controls, and put everything under MPC. A realistic
approach is to use PID controllers to close those loops that are not very interactive
with other loops. For example, the levels LI-101 and LI-102 are better off being
put under PID control, so is the accumulator pressure PI-101. The top product
draw, FI-102, can become a flow control loop. The baselayer control scheme will
then be as shown in Fig. 5.17.
If the operating conditions require that MPC be taken offline, such as during a
significant process upset, it would safely and seamlessly “shed” to the next layer
down, i.e., baselayer control.
After the MPC technology becomes widely available, many existing PID control
schemes are upgraded to MPC for better performance. Although the two control
technologies are significantly different, they are based on the same process analysis
and operating requirements. The difference is that the control objectives may be
fulfilled more fully or more elegantly with the upgrade.

Fig. 5.17 A fractionator column, baselayer control scheme

200 5 Advanced Process Control

5.5.3 MPC Maintenance and Sustainability

As with any sophisticated application, sustained performance and benefit are always
desired but not easy to achieve. Experience shows that the controller performance
will deteriorate after a year or two, and the deterioration would accelerate if proper
maintenance measure is not taken to turn the trend around. The reasons that regular
MPC maintenance is required include the following:
1. Change in process dynamics requires review and potential update of process
models or even control structure.
2. Change in operating limits or constraints may require an update of the control
structure.
3. Wear and tear in control valves or other final control elements may demand model
updates.
An example of the MPC benefit prediction based on well-maintained and poorly-
maintained MPC applications is given in Fig. 5.18. The data is based on a 3-year
APC implementation project in an ethylene plant comprising over a dozen APC
applications. Without proper maintenance, the APC benefits are expected to diminish
after a year or so. The loss of benefits will accelerate until all MPC controllers are
taken out of service for good.
The ongoing monitoring and maintenance of the MPC controllers is as crucial as
the initial implementation but are often ignored or undervalued. The lack of main-
tenance has been an industry-wide problem and is often the key reason for reduced
benefits and even the decommissioning of MPC applications. The plant managers
and plant operations must be clear about MPC’s benefits, limitations, and potential
problems, and make necessary operational adjustments to adapt to the change.

Fig. 5.18 MPC benefits forecast based on maintenance level

5.5 Model Predictive Control—Practical Considerations 201

The first step toward good maintenance is proper monitoring of the controller
performance. It is recommended that a controller performance monitoring system
(CPM) be implemented to provide 24/7 monitoring on all the baselayer (PID) con-
trollers and advanced process control (MPC) applications. See Sect. 8.5.2 for more
discussions on CPM.
The CPM system should be proactive, systematic, and automated to provide the
basis for both predictive and reactive maintenance. The CPM system typically pro-
duces daily, weekly, monthly reports on the controller performance indicator (KPI)
targets such as the on-stream factor (e.g., 90%) and CV compliance (e.g., 95%). The
KPIs should be regularly reviewed as part of the maintenance routine:
1. Daily monitoring: The goal of daily MPC performance monitoring is to timely
troubleshoot any problems that may have arisen during the previous day and
ensure that MPC applications are not unnecessarily turned off.
2. Weekly performance reporting: A weekly performance review should be per-
formed and include the following actions:
a. Review the weekly trend of the on-stream factors for all the applications.
b. If any of the on-stream factors is lower than expected or displays an abnormal
pattern, check the MPC maintenance logbook and talk to the operators to
determine why.
c. If deemed necessary, retrieve history data to take a closer look at the trends.
d. If necessary, call a meeting with operations and engineers to decide on actions
to take.
3. Annual performance auditing: MPC performance deteriorates over time. As part
of the MPC maintenance, it is recommended that proactive performance evalua-
tion be conducted on every controller every six months to a year by performing
the following tasks:
a. Meet with engineers and operations and discuss if the controller performance
is still acceptable and if there is a need to improve the controller.
b. If any controller performance is in question, collect two weeks of normal
operating data, import them into the modeling tools, and check for significant
model-plant-mismatch.
c. The maintenance could be as simple as an update of model gain or as complex
as completely revamping the control strategy. Depending on the above out-
come, perform an update of the MPC controller following the standard work
process.
d. Determine the maintenance budget for the coming year.

5.5.4 Knowledge Management and Training

Although the MPC penetration rate in refining, petrochemical, and chemical plants
has been pretty high, there is still resistance in accepting MPC implementation in
many plants, especially in other industries. The key concerns include:
202 5 Advanced Process Control

1. Operations feel that MPC is too complex and does not trust the moves that it
makes.
2. Operations love MPC and rely on MPC so much that they lose familiarity with
the process operation over time.
MPC focuses on control objectives such as throughput, production specifications,
and economics, and less on the smooth and stable operation. For this reason, MPC
can be unpopular among operation personnel since MPC moves the process point
away from their “comfort zone.” The operators often switch MPC off (or part of it)
due to concerns about stability over profitability.
The ultimate solution is to deliver high-quality MPC applications that are intuitive,
reliable, and transparent, to facilitate understanding and operation. However, MPC
applications are inherently more complicated than baselayer control applications
both in the understanding and operation. Therefore, adequate training and education
is crucial.
1. Knowledge Management. The MPC knowledge and skills must be appropriately
retained and managed within the support team, including proper management and
maintenance of
a. software manuals;
b. training manuals;
c. controller documentation (MPC narratives and Operating Manuals);
d. work processes;
e. best practices and lessons learned.
2. Training. Training is the primary means to maintain the MPC expertise. A good
interest level and learning potential will help to motivate and retain MPC engineers
and operators.
Another common concern on MPC is that the operators may gradually lose their
intimate familiarity with the plant operation with MPC controlling the plants.
This gradual loss of skills is a general problem across the industry, and there have
been no standard or effective solutions developed yet. One mitigating option is to
make full use of an operator training simulator (OTS) system to train (or regularly
refresh) the operator on handling abnormal conditions.

5.5.5 Future Developments

Adaptive control is the hope of the future for addressing the time-varying process
dynamics and some nonlinearity (Challenge #6). The foundation of the adaptive
control is the continuous online update of the process model, including the model
parameters and potentially the model structure, to adapt to the loop dynamics changes.
Although academic research has been active, a genuinely reliable, practical, and
general-purpose algorithm similar to MPC is still unavailable. Nevertheless, some
emerging exciting progress in MPC development includes the following:
5.5 Model Predictive Control—Practical Considerations 203

1. Periodic model updates: A locally linearized model may become inadequate over
time due to operating condition changes, a common maintenance challenge for
both PID and MPC applications.
The self-tuning PID has been available for decades, where the tuning parameters
are automatically updated via a one-button click. However, self-tuning has not
been widely accepted because there is still much to expect in the performance.
PID tuning is still an art requiring engineering insight and practical experience
(see Chap. 9).
For MPC applications, the recommendation is to review and update the MPC
models every six months to sustain the MPC performance. However, most MPC
applications do not receive the model update as recommended due to budget
limitations. As seen in the MPC work process in Sect. 5.6, MPC modeling is
usually the most challenging and time-consuming part of MPC implementation
and is thus the most costly. It is usually easier to get approval for starting a
new MPC implementation in many operating plants, often with much fanfare, but
challenging to receive the continuing budget support for regular maintenance. The
best practice is to include the annual maintenance in the initial budget planning.
One recent development is the automated step testing, which allows online model
updates behind the scene while MPC is in control. Automated step testing provides
some limited adaptive control capability, although it is still manually initiated and
requiring a high level of expertise to carry out.
2. Nonlinear Models: The MPC model is inherently linear and time-invariant, mainly
because of its simplicity and low computation load. Some MPC implementation
includes nonlinear gain (e.g., via neural network models) while keeping the rest
of the model in linear form. There are incremental advantages over a purely linear
model.
3. First-principle models: Most industrial processes are inherently nonlinear and
time-varying. A very sophisticated and complex model would be required to cap-
ture this complicated dynamic behavior fully. The current state-of-the-art model
for process description is dynamic first-principle models based mainly on partial
differential equations (PDE). It has been used for dynamic simulation studies and
operator training simulators (OTS).
The complexity of first-principle models makes it impossible for real-time con-
trol, at least for now. However, as the online computational power increases, the
next generation of model-based process control may be based on high-fidelity
first-principle models.

Another direction of advancement is simplifying the process dynamics and improv-

ing the measurement reliability (Challenge #5) via inferential properties, a.k.a. pro-
cess analytics. Inferential property estimation is capable and cost-effective for many
practical problems and deserves a lot more attention than it is now receiving. Static
inferential properties calculation has been in extensive use. Dynamic inferential prop-
erty estimation has been a crucial supplement to MPC for simplifying or enhancing
the MPC models. In addition to the soft sensor, many nonlinear problems are trans-
204 5 Advanced Process Control

formed into linear ones through transformations or characterization, backed by sound

engineering insight.

5.6 Model Predictive Control—The Work Process

Model predictive control has been the de facto standard for process control solutions.
On the one hand, MPC is taken as the panacea for almost all control problems. On
the other hand, many practitioners are sent to the field without adequate background
and training on process control fundamentals. As a result, the results of MPC are a
mixed bag. There are many failure examples due to inappropriate implementation.
As discussed in Sect. 5.4, MPC is just another power tool in the process control tool-
box. It can solve more control problems than the conventional PID-based feedback
control; it can also cause more damage than a PID-based solution can if not correctly
implemented.
The applications are typically based on commercial MPC packages (Qin and
Badgwell 2003) that contain all the necessary tools for model identification, con-
troller design, performance tuning, and real-time monitoring. The theory and tech-
nology are built into the software tools and are typically not transparent to the user.
Compared with baselayer control, MPC is inherently multivariable and concerns a
much larger area of the process; it is thus even more vital to understand the underlying
process. In addition, the control design requires more careful considerations due to
constraint handling and economic target pushing. It is thus essential to follow a well-
defined methodology and procedure to ensure the proper execution of MPC projects.
The methods of process analysis in Chap. 6 and control design in Chap. 7 are all
applicable. As a result, the level of success of MPC applications highly depends on
the competence of the implementors and their compliance with the standard method-
ology and best practice procedures. A proven work process is one way of guiding
and ensuring that the necessary steps are followed, and no dangerous shortcuts are
taken.
Some key points that need to pay close attention to are highlighted as follows:
1. MPC is built on top of the baselayer controllers. Improving the baselayer control
loops is the prerequisite for MPC applications. More than half of the MPC benefits
come from improved baselayer control.
2. The quality of the process model is the most critical factor for MPC’s success. A
good model depends on the engineering insight and the best practice procedure
(Marlin 2015). The cause and effect analysis and dynamic response analysis in
Chap. 6 is essential in arriving at the correct model structure and parameters.
High-quality step-test data is instrumental in confirming the model structure and
estimating the model parameters.
The rest of the section provides an MPC best practice work process for MPC appli-
cations of various complexities. For some tasks, such as model update during regular
maintenance, only part of the work process needs to be followed.
5.6 Model Predictive Control—The Work Process 205

5.6.1 Project Preparation

Before starting an MPC application implementation, it is vital to understand the

project definition, the process to be controlled, and the objectives to achieve. The
following is a checklist of the things to know before calling for a kickoff meeting:
1. Understand the project scope, schedule, budget, deliverables, and benefit expec-
tations. There are typically some documents available such as feasibility studies
or master-plan studies that define the MPC project from a high-level perspective.
Collect and study these documents to get a basic understanding of the commercial
side of the project.
2. Check if there is any software and hardware to be procured before the project starts.
Since the procurement of hardware and software often requires considerable lead
time, plan well in advance to allow sufficient time for delivery.
3. Obtain a copy of the following documents where available, and spend as much
time as necessary to better understand the process flow and plant operating phi-
losophy.
a. Process control narratives (PCN) document.
b. Process description and operator training manuals.
c. Latest process PFD and P&ID documents.
d. Vessel sizing information and level transmitter data sheets to be used for tuning
level controllers.
e. For product quality variables to be controlled, collect at least six months of
lab data if online analyzers are not available.
f. For revamping-project, obtain all documentation from previous MPC imple-
mentation.
4. Define, or validate if already defined, both the control and economic objectives.
5. Identify the stakeholders of the project that need to be included in meetings and
communication. These typically include process engineers, economists, instru-
ment technicians (analyzer team), and operations.

5.6.2 Project Kickoff

MPC application is a multi-team effort and always requires close collaboration

between several parties, including process engineers, economists, control system
engineers, instrumentation technicians, and panel operators.
A project kickoff meeting is essential for team mobilization and alignment. A
recommended agenda for an MPC kickoff meeting is listed as follows:
1. Team introduction, including roles and responsibilities.
2. Review the MPC scope, schedule, and deliverables.
3. Review the control and economic objectives.
206 5 Advanced Process Control

4. Explain project execution methodology (schedule, staffing, key project steps,

milestones, reporting, interim review points).
5. Conduct a review and discussion on the following:
a. Major operation and control challenges.
b. Operating limits, constraints, and bottlenecks.
c. Known problems in instrumentation and equipment.
d. Significant disturbances to the unit (e.g., feed change, ambient temperature).
6. Check and note the unit blackout dates, such as scheduled shutdowns and turn-
down operating mode. Also, check and record the vacation plans of key players.
7. Agree on MPC benefit expectations. Discuss and agree on a baseline period for
benefit-review purposes, if not yet done.
8. Issue meeting minutes with action items (with item owners and due dates).

5.6.3 System Installation

The MPC infrastructure should be set up as soon as feasible since certain activities
may need a long lead time to complete (e.g., procurement). A checklist of the activities
related to infrastructure is listed as follows:
1. Set up the MPC server machines if MPC runs on a separate and dedicated machine.
2. Install and configure MPC software.
3. Configure and test the communication between DCS and the MPC machines.
4. Install and configure the MPC communication watchdog on DCS, if applicable.
5. Configure MPC offline design tools to gain access to the site historian database.
If a direct connection to the database is not possible, make sure that alternative
means of data acquisition are set up and functional. For example, the historical
data can be exported to Excel or CSV files and imported into the MPC modeling
tool.
6. Set up the project file structure on a shared location and make it accessible to all
concerned parties.
7. Determine how the process data will be collected. If a historian database is used,
check the data compression to see if the compression limit needs to be reduced
or removed.

5.6.4 Baselayer Review and Improvement

MPC relies on the baselayer controllers to function. A properly designed and well-
tuned baselayer is a critical requirement for the success of MPC. As part of the MPC
activities, a thorough review and improvement of the baselayer control are necessary.
A recommended list of items to check is as follows:
5.6 Model Predictive Control—The Work Process 207

1. Check the required analyzers’ availability, reliability, and accuracy and determine
whether inferential property estimator (IPE) applications are needed.
2. Create any new MPC-required calculation tags on DCS as needed and historize
them for both DCS and historian database.
3. Obtain a complete list of the relevant DCS tags for data collection. Identify all
the candidates for manipulated variables (MV) and disturbance variables (DV).
4. Obtain a complete list of relevant PID controller tags, including PV, SP, OP, and
Mode. Also, include the essential properties such as tuning parameters, equation
type, filter parameters, engineering units, and measurement ranges.
5. Check history data of the process variables and determine if excessive measure-
ment noises are present. If so, consider adding measurement filtering to smooth
out the noises.
6. Find out why any controller that has run in the manual (MAN) mode for an extended
period.
7. Identify the controllers whose performance is not satisfactory and re-tune them
by following the standard troubleshooting procedures.
8. Define and document the baseline condition of operation conditions based on the
agreed baseline period.
9. Implement and commission the inferential property estimator (IPE) applications
if required. Historize the inferred property values.

5.6.5 Pre-testing Activities

Pre-testing is essential for understanding the underlying process dynamics and iden-
tifying potential problems in baselayer control, instrumentation, equipment. Besides,
coupled with historical data, the pre-test can help optimize the step testing plan, sub-
stantially reduce the time spent on step testing, or help avoid some unnecessary step
tests. A checklist of things to watch out for during pre-testing is provided as follows:
1. Make a list of all parameters for which historical information will benefit MPC
implementation, and check to ensure that these parameters are configured and
historized in the historian database.
2. Check and make sure that the data compression setting in the historian database is
appropriate. The historian database’s default compression band (typically 0.1%)
is typically too high for the desired data resolution; it is often necessary to
change to a smaller deadband (e.g., 0.01%) for process modeling. Compile a
list of variables whose compression settings need to be changed, and ask the
database administrator to make the changes.
3. Obtain and document the product specifications and operating targets relevant
to the MPC application.
4. Determine the approximate time to steady-state (TTSS) and the move size of
the step changes for each MV and DV variable based on historical data. The
general principle is to use move sizes as large as possible to achieve the highest
208 5 Advanced Process Control

possible signal-to-noise ratio. However, many factors, primarily operating safety

and local linearity, limit the size of step change that can be made. Always consult
the operator before making any moves.
5. Operating condition: Before step test, check the operating condition against the
condition that MPC intends to run. If needed, ask the operator to make necessary
changes to bring the unit to the desired condition (e.g., unloading the column).
6. Controller modes: Make sure the baselayer controllers are in the correct modes
intended for MPC. For example, if one of the PID controllers in cascade mode
will be run with the cascade loop open when MPC is implemented, the cascade
loop should be kept open while conducting step tests.
7. PID algorithm type: For specific DCS systems, check and ensure that control
algorithms of the PID controllers to be manipulated by MPC are set to the right
algorithm type. For example, PID with algorithm “I-PD” shall be changed to
“PID” or “PI-D” to ensure proper setpoint tracking (see Sect. 3.2.1).
8. Ambient temperature: Ensure that a reliable ambient temperature tag is included
in data collection, as the ambient temperature is commonly a vital disturbance
variable (DV).
9. Step changes: Make at least one step change in each direction on every MV/DV
variable that is expected to move during formal step testing. Validate/confirm
the move sizes and TTSS with operators before making any moves.
10. Interaction with operators: Engage operators as much as possible in the test. Ask,
record, and cross-validate their opinions as follows:
a. Are there any valve, transmitter, or tuning problems?
b. What handles do they usually use to control each CV, and what the control
ranges are?
c. What do they typically move each MV for?
d. How much grace time does the operation allow for off-spec products?
11. Inferential properties: Build inferential property estimator (IPE) models and
applications if required.
12. Modeling tool: Setup the offline modeling tool and import pre-test data and
baseline data as necessary.
13. Preliminary cause and effect table: With process knowledge and test data, identify
the cause and effect relationship between the MVs, DVs, and CVs. Show them
in a cause and effect matrix, and mark the confidence on the relationship with
certain symbols. For example, if a strong relationship with a positive gain is
expected, mark the cell with +++. If a reasonably strong relationship is expected,
mark it as ++. If a positive relationship may exist, mark it as +. Do the same for
relationships with negative gains. If uncertain, mark the cell as ?.
After the pre-test is completed and the data is analyzed, a pre-test report or summary
should be issued to the operations and engineers to prepare the unit for formal step
testing. Depending on the nature of the problems discovered, this may take some time
to complete all the fixes. The pre-test report should include the following content at
a minimum:
5.6 Model Predictive Control—The Work Process 209

1. Equipment and instrumentation problems, especially valve problems.

2. Any changes in the baselayer control strategy to be implemented. For example, the
column’s sensitive tray for temperature may be shifted after column unloading,
and the temperature controller may need to be relocated and re-configured.
3. A list of the controllers that have been re-tuned and the controllers that are planned
to be re-tuned.
4. Any new instrumentation that is required to be added.
5. Evaluation of the analyzers and the decision on whether inferential property esti-
mators (IPE) need to be implemented.
6. Key control objectives and control handles that are identified for MPC.
7. A preliminary model structure showing the expected cause and effect relationship
and dynamic response characteristic (approximate gain, delay, and lag) between
each cause and effect variable pair.
8. Detailed step-test plan including the variables to be tested, such as the step size,
test duration, variables to watch during step testing.

5.6.6 Plant Test

A formal step-test practice aims to systematically and meticulously disturb the pro-
cess to observe the dynamic responses of the interested variables. Step test is a critical
step in MPC implementation and is also the most time-consuming part of an MPC
project. A carefully designed and conducted step test yields high-quality process data
and consequently leads to high-fidelity models. As emphasized already, the model
quality (both the structure and the parameters) is the most important determining
factor for MPC’s success (or failure).
Some recommendations on step testing are provided as follows:
1. Check and make sure that the inferential property estimators (IPE), if any, are
functioning with acceptable reliability and accuracy.
2. Check if analyzers need to be calibrated. Request for calibration if necessary.
3. Ensure the data collection mechanism is appropriately set up and reliably working,
and data compression is removed or reduced.
4. Use a standard template to create a test log file with a list of the MVs and DVs
and their move plans.
5. Ensure that the unit is in the same operating condition as that MPC intends to run
at, including process values and PID controller modes. If not, ask the operator to
make changes to bring the unit to the desired operating condition.
6. Check and request extra lab samples during plant tests as needed.
7. Brief the panel operators on what will be done and request their cooperation to
a. inform the MPC team of any significant upcoming disturbances, e.g., feed
change, fieldwork on equipment that could impact the test results and
b. inform the MPC team of any key control moves they make that could impact
the test results.
210 5 Advanced Process Control

8. In the test log file, record all key disturbances and operator-made moves that may
affect test results.
9. Frequently review the test data in offline modeling tools (if feasible) and make
adjustments on the move plan such as move size, move frequency, and test dura-
tion.

5.6.7 Process Modeling

The power of MPC comes from the process models, which capture the cause and
effect relationships and dynamic response behavior between the manipulated vari-
ables and controlled variables. Although the quality of the models is primarily deter-
mined by the data available, experience and skills are essential in extracting the max-
imum information from the data at hand. Some general recommendations regarding
process modeling are provided as follows:
1. Collect test data. Note that data can be from various sources, including historical
data, pre-test data, and step-test data. Step test data is obtained under controlled
conditions and is typically most trusted. However, pre-test and historical data can
often be used as supplementary information or for cross-validation of the models.
2. Build models. Use process understanding and engineer knowledge in all steps,
especially in data pre-processing, e.g., slicing and dicing.
3. Assign a quality indicator to each model identified, e.g., A for good models, B
for fair models, and C for low-quality models.
4. Use the simplest models (e.g., first-order plus time delay) whenever possible, even
if a more complex model (e.g., second-order model) can offer marginally better
performance.
5. If a critical model (based on process understanding) is expected, but the model
quality is questionable, re-do the step test to improve the model. Otherwise, leave
it out of the controller. A flawed model may be worse than no model.
6. Decide whether a nonlinear transform such as a logarithm function is needed on
certain variables.
7. Perform a relative gain array (RGA) analysis to determine that the resulting control
matrix is not singular or ill-conditioned (Skogestad and Phostlethwaite 2005).

5.6.8 Controller Design

The controller design part is relatively straightforward with a sound control strategy
(following the analytical steps in Chap. 6) and high-fidelity models (see Chap. 7).
The following are some of the things to watch out for during controller design:
1. Obtain economic values (market values) on the following streams for economic
function (EF) calculation:
5.6 Model Predictive Control—The Work Process 211

a. Feed Stocks.
b. Products.
c. Utilities such as steam, fuel oil, and electricity.
d. Waste (disposal cost).
2. Calculate the economic values of the variables if available. Otherwise, assign
economic values based on their relative economic significance.
3. Decide and define sub-controllers. Note that change in sub-controller definition
at a later time may require more re-work, such as the update of the watchdog.
4. Simulate, test, and fine-tune the controller offline with unconstrained move size
limits.
5. Simulate the controller offline in different cripple modes, e.g., removing some
MVs and CVs from control.
6. Update the control matrix to show the following information at a minimum:
a. All the MVs, DVs, and CVs.
b. The cause and effect relationships between the MV/DV and CVs, including
the gains.
c. Model Quality rating.
d. Setpoint values or setpoint ranges.

5.6.9 Model and Controller Review

After the models are identified and the controller built, it is necessary to conduct
a thorough review of the models and detailed control strategy with operations and
process engineers before commissioning the controller. Some general guidelines for
the review meeting are as follows:
1. Invite process engineers, operation representatives, and senior operators to the
meeting.
2. Review the detailed control and economic objectives.
3. Review the CVs, MVs, DVs and their cause and effect relationship. Use the control
matrix for convenient presentation.
4. When a CV is controlled by multiple MVs, decide which MV(s) is the prin-
cipal handle for the CV. Prioritize the MVs if possible and tune the controller
accordingly. Decide whether some of the MVs should be duplicated as DVs.
5. Review/confirm the product specifications, operating targets, and critical con-
straints.
6. Review the commissioning plan. Discuss and request operation support for com-
missioning.
7. Ask Operations to notify operators of the upcoming commissioning and training.
8. Issue meeting minutes with the following information at a minimum:
a. The agreed changes to the controller strategy and models.
b. The product specifications, operating targets, and critical constraint limits.
c. Commissioning Plan.
212 5 Advanced Process Control

5.6.10 Controller Commissioning

After the control strategy and controller models are reviewed and agreed upon, the
controller can be tested online.
The controller commissioning typically consists of two steps: pre-commissioning
and commissioning. Pre-commissioning aims to set up the proper communication
between MPC software and DCS and identify any errors in controller configuration.
Following the pre-commissioning is the formal commissioning. The main objective
of the commissioning phase is to fine-tune the controller to achieve the desired control
performance.
Below is a list of the general steps for controller commissioning:
1. Create a complete list of MPC MVs with the following modes: normal mode,
MPC mode, and Shed Mode. Validate the list with operators. Note that the shed
mode of a controller may not necessarily be the same as its normal mode.
2. Build and configure the MPC/DCS communication watchdog.
3. Run the MPC control in online open-loop mode, i.e., MPC calculates the next
moves as if it is online, but the moves are not actually implemented in DCS. After
every execution, do the following:
a. Check the calculated MV moves. These are the control actions to be sent
to DCS for all the MPC MVs. Based on the current CV errors, validate the
sign and magnitude of the moves and make sure they agree to the process
understanding and the values are in the expected range.
b. Check the predicted steady-state values of all MVs if available. These are
the total planned MV changes at the new steady state based on current CV
errors. Make sure the direction and size of the MV changes are reasonable in
comparison with current CV errors.
c. Check the predicted steady-state values of all CVs if available. These are the
predicted steady-state values for all CVs, if the planned moves above are all
implemented. Check whether the predicted values are within expectation.
d. Check the innovations on the PVs and make sure they are not too large. Each
PV has an innovation value, which is the difference between the controller’s
prediction of the PV and the actual measurement of the plant. This innovation
is an indicator of the “goodness” of the prediction model for the PV.
If any of the above is not as expected, stop the commissioning and double-
check on the controller tunings, DCS mappings, or even the controller models
to find the cause of the discrepancies.
4. Before turning on the controller in the closed-loop mode for the first time,
review/confirm again with the operations and engineers on the MPC MV lim-
its and CV set ranges, MV/CV predictions, and next MV move sizes. Adjust the
limits as necessary.
5. For critical MV, consider temporarily reducing their move limits to avoid unex-
pected large moves due to potentially too aggressive tunings. Remember to put
5.6 Model Predictive Control—The Work Process 213

these limits back to normal ranges once the controller has been “proven” to be
working correctly online.
6. Compare the CPU and network loads before and after the controller is turned on.
Verify that the load is within the expected range.
7. During pre-commissioning (day-time-only), make an MPC screen capture show-
ing all the current MV and CV values before turning on the MPC. Leave a copy
to the panel operator in case the operator needs to bring the unit back to where it
was.
8. If available, the online monitoring tool shall be configured and used parallel to
monitoring the MPC status and performance.

5.6.11 Operator Training and Documentation

Before the MPC application is commissioned, it is necessary to introduce the appli-

cation to all shifts of operators. Operator training typically involves the following
activities:
1. Prepare the operating manual using the standard document template.
2. Prepare operator training presentation slides based on the operating manual.
3. Check operator shift schedule and work out a training plan (group training or
individual training). Inform the operators in advance about the training schedule.
Recommended time for operator training is when the controller is commissioned
and relatively stable (for hand-on training). All operator training needs to be
completed before running the controller round-the-clock.
The operator training should cover the following contents at a minimum, with an
appropriate quiz to assess the training efficacy:
1. The purpose of the MPC application, along with historical data showing pre-MPC
performance.
2. Overview of all MPC variables (MVs, CVs, DVs).
3. Overview of modifications to existing graphics, as well as new MPC graphics.
4. Criticality of variables in the controller, for example, which variables are essential
for control.
5. Any special instructions for the MPC.
6. Operation of the controller, especially how to shed the MPC controller to the
baselayer in abnormal conditions.

5.6.12 Controller Handover to Operators

After completing the operator training and a few weeks of stable operation, the
controller can be officially handed over to operators. For doing that, make sure the
following items are covered:
214 5 Advanced Process Control

1. Make sure the operators have received a copy of the MPC operating manual.
2. Ensure the operators understand that they are assuming the responsibility of MPC
control once it is on 24/7, even though engineers may be remotely monitoring.
3. Make sure operators have the emergency contact numbers of the MPC engineers.
Even after controller handover to operators, close monitoring and maintenance sup-
port is still necessary to correct any issues discovered, improve the controller per-
formance, and in the long run, ensure the sustained performance of the controller.
Typical maintenance duties include the following:
1. Conduct regular (e.g., daily) performance reviews of the controller.
2. Discuss with operations and engineers for feedback and make fine tunings as
necessary.

5.6.13 Post-implementation Review (PIR)

After two to three months of stable operation, a post-implementation review (PIR)

can be conducted, with the following recommended agenda:
1. Review of MPC benefits.
2. Unit performance comparison before and after MPC.
3. Feedback on control performance and added values.
4. Future value opportunities identified during the project.
5. Issue meeting minutes.
The benefits analysis for the plant will involve calculating the post-MPC economic
performance against the baseline performance over a specific period. Since the base-
line data were gathered over two weeks of operation, it makes sense that this same
period is used for the PIR. An appropriate two-week period of process operation with
MPC will need to be identified for each plant as the “MPC period.”

5.6.14 Project Close-Out

To close out the project, ensure that all the required documents have been delivered
and relevant data files have been appropriately archived for each MPC application.
The checklist is to include the following items:
1. Organize and clean up the project file repository and make sure all the required
project deliverables and application-related files are properly archived and in the
correct format, including data files.
2. Finalize and distribute the MPC operating manual.
3. Finalize and distribute the MPC engineering manual (MPC narrative).
5.7 Summary 215

5.7 Summary

Advanced process control (APC) refers to any control scheme that is more complex
than a simple PID control loop. Now that model predictive control (MPC) has become
prevalent, APC is often used interchangeably with MPC. However, this book uses
APC separately from MPC and reserves the word APC for all advanced technolo-
gies, including, but limited to, model predictive control, nonlinear control, adaptive
control, and robust control.
Most advanced control applications are based on models of a specific type, from
the heuristic model, input/output black-box model, state-space model, to the first-
principle model (Sect. 5.2). The more advanced the technology is, the higher require-
ment it has on the underlying model. Inferential property estimator (Sect. 5.3) as an
open-loop calculation has flexibility in terms of model requirement. It is a reliable
and cost-effective approach to improve operation and control and is thus an area
deserving more attention in both theory and practice.
Model predictive control (Sect. 5.4) is a step-change from traditional PID feedback
control. Introducing explicitly and separately identifiable process models into the
control algorithm has been a breakthrough that successfully breaks the limitation
inherent to PID-dominated feedback control. MPC applications yield better and more
reliable control performance and have gained widespread use in the process industries
such as chemical plants and oil refineries since the 1980s.
One crucial change that MPC has brought about is that the MPC work process
forces the practitioners to pay more attention to the process analysis and overall
control strategy as part of the implementation requirement. MPC is inherently mul-
tivariable and covers a larger area of the process than the conventional PID control
approach. The control design must see the “bigger picture,” i.e., the overall control
strategy. Chapters 6 and 7 will elaborate on the “bigger picture” while discussing the
details of process analysis and control design. Due to the complexity of the MPC
application, a field-proven work process (Sect. 5.6) is critical to ensure that the key
application steps are not bypassed or compromised.

References

Åström KJ, Hägglund T (1995) PID controllers: theory, design, and tunings, 2nd edn. Instrument
Society of America
Blevins TL, McMillan GK, Wojsznis WK, Brown MW (2003) Advanced control unleashed—plant
performance management for optimum benefit. ISA
Box GEP (1976) Science and statistics. J Am Stat Assoc 71(356):791–799
Clarke DW, Mohtadi C, Tuffs PS (1987) Generalized predictive control—part I. The basic algorithm.
Automatica 23(2):137–148
Corriou JP (2018) Process control—theory and applications, 2nd edn. Springer
Cutler CR, Ramaker BL (1980) Dynamic matrix control—a computer control algorithm. In: Pro-
ceedings of American control conference
216 5 Advanced Process Control

Garcia CE, MPrett D, Morari M, (1989) Model predictive control: theory and practice—a survey.
Automatica 25(3):335–348
Gordon L (2005) Advanced regulatory control: decoupling. Control Eng
ISO-5167 (2003) Measurement of fluid flow by means of pressure differential devices inserted in
circular cross-section conduits running full, 2nd edn. ISO
Lee JH (2011) Model predictive control: review of the three decades of development. Int J Control
Autom Syst 9(3):415–424
Ljung L (1999) System identification: theory for the user, 2nd edn. Prentice-Hall, Englewood Cliffs,
NJ
Marlin TE (2015) Process control—designing processes and control systems for dynamic perfor-
mance, 2nd edn. McGraw-Hill
Morari M, Lee JH (1999) Model predictive control: past, present and future. Comput Chem Eng
23(4–5):667–682
Niu S, Fisher DG, Xiao D (1992) An augmented UD identification algorithm. Int J Control
56(1):193–211
Niu S, Fisher DG, Ljung L, Shah SL (1994) A tutorial on multiple model least-squares and aug-
mented UD identification. Technical report LiTH-ISY-R-1710, Department of Electrical Engi-
neering, Linkoping University, Linkoping, S58183, Sweden
Qin SJ, Badgwell TA (2003) A survey of industrial model predictive control technology. Control
Eng Pract 11(7):733–764
Rawlings JB, Mayne DQ, Diehl MM (2017) Model predictive control: theory, computation, and
design, 2nd edn. Nob Hill Publishing, LLC
Rossitier JA (2003) Model predictive control—a practical approach. CRC Press
Samad T, Bauer M, Bortoff S, Cairano SD, Fagiano L, Odgaard PF, Rhinehart RR, Sanchez-Pena R,
Serbezov A, Ankersen F, Goupil P, Grosman B, Heertjes M, Mareels I, Sosseh R (2020) Industry
engagement with control research: perspective and messages. Ann Rev Control 49:1–14. open
access
Shinskey FG (1994) Feedback controllers in the process industries. McGraw-Hill Publishing, New
York
Skogestad S, Phostlethwaite I (2005) Multivariable feedback control—analysis and design, 2nd
edn. Wiley
Smith CL (2009) Practical process control—tuning and troubleshooting. Wiley
Smith OJM (1957) Closed control of loops with dead time. Chem Eng Prog 53(5)
 Part III
Analytic Skills and Problem-Solving
Methodologies

Process control is about the control of processes. The practical challenge to process
control is that there are countless processes in numerous industries, and there is also
a rich collection of control technologies. As a result, process control design is by no
means a trivial job. For instance, a process control engineer in a service company
implementing model predictive control (MPC) applications may have to deal with a
completely different process in each new project. It is unrealistic to expect him or her
to have the required process knowledge to start with, nor is it advisable for him or her
to blindly copy and paste from existing projects. Even with the same process design,
every process typically has very different operating requirements and constraints
and must be treated differently. It is more practical to develop the essential and
generic process analysis skills applicable to all new processes encountered. After
all, the process knowledge required by process control engineers is different and less
comprehensive than that of process engineers.
A good understanding of the process characteristics and in-depth knowledge of
control technologies is essential for designing and executing the control solution. As
a cross-disciplinary specialty, process control requires knowledge in many related
but distinct areas such as process engineering, rotating equipment, instrumentation,
control systems, operations, maintenance, and project execution, in addition to the
core technologies of automatic control. For this reason, a process control practi-
tioner must be very selective in the acquisition of knowledge and skills, and more
importantly, pursue a more efficient approach to gain the necessary expertise.
For this purpose, we have recommended a learning process in two phases: build
the core knowledge framework as the foundation (Part II of this book) and then
develop the necessary skills to perform the job efficiently (Part III). This learning
process includes the following steps:
1. Have the basic knowledge of the necessary tools and enabling platforms, such as
control system infrastructure (DCS/PLC), sensors, transmitters, control valves,
network and communication, and information technology.
2. Master the core process control technologies, including the baselayer control and
advanced process control.
218 Part III: Analytic Skills and Problem-Solving Methodologies

3. Develop the analytic skills for problem-solving such as process analysis, control
design, troubleshooting, and performance monitoring and improvement.
4. Follow the best practice methodologies to tackle process control design or trou-
bleshooting tasks.
5. Eventually, reach the mastery level of process control instinct or mindset.
Structured process analysis (Chap. 6) helps in gaining the necessary process
insight. The process analysis skills such as process flow analysis, supply and
demand analysis, cause and effect analysis, degree-of-freedom analysis, and dynamic
response analysis provide a systematic approach to extract the necessary informa-
tion about the process as required by process control. Tools such as process overview
diagram (POD) and cause and effect table (CET) can help improve the consistency
and efficiency of the process analysis.
The process control design is the next step following process analysis. A structured
approach to analyzing the control needs and guiding the control technology selection
and implementation is critical. The techniques such as overall control strategy for
plant-wide control and protective/regulatory control for layered protection can help
design fit-for-purpose control solutions (Chap. 7). Control overview diagram (COD)
is an effective visual tool to assist the design.
The principles and skills of process analysis and control design can also apply
to control problem troubleshooting (Chap. 8) and improvements (Chap. 9), which
demand the most knowledge, skills, and experience. It requires a sound knowledge of
how the control scheme works, and a good understanding of how the control scheme
may fail. The skills are built through many years of hands-on experience that has no
substitute for. The guidelines and procedure in Chaps. 6 and 7 can make the loop
troubleshooting and improvement more focused and less painful.
Chapter 6
Methodology of Process Analysis

Process control is about process and control. The previous chapters have discussed
the different control technologies, from simple PID control to advanced process con-
trol. However, they are just various tools in a process control engineer’s toolbox.
Effectively using these tools to perform meaningful work depends on a good under-
standing of the process design and operating requirements (Sect. 6.1). As there are
countless processes in various industries, it is unrealistic for a process control prac-
titioner to become intimately knowledgeable on all the processes encountered. The
more practical approach is developing generally applicable skills and methodologies
to help with process analysis. This chapter presents a structured work process for
efficiently and consistently extracting the required process information into the form
suitable for process control. These skills include process flow analysis (Sect. 6.2),
supply and demand analysis (Sect. 6.3), cause and effect analysis (Sect. 6.4), degree
of freedom analysis (Sect. 6.5), and dynamic response analysis (Sect. 6.6).

6.1 Operating Requirements

Process control serves the operation and production in two main areas: process safety
and production efficiency. Familiarity with the operating objectives and operating
requirements is essential for understanding the process operation and designing the
control solution (Downs and Skogestad 2011; Marlin 2015; Seames 2018).

6.1.1 Operating Objectives

Generally speaking, the operating objectives are about producing the right products
with the right quality and quantity at the right time (Sect. 1.2). For example, in the
oil & gas industry,

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 219
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_6
220 6 Methodology of Process Analysis

1. Upstream exploration & production (E&P) is primarily concerned about the move-
ment of the various fluids (gas, oil, and water) for simple separation, light process-
ing, and transport. So the operating requirements are to ensure the various streams
flow to the right location with the right quantity. The general control requirement
is thus capacity (quantity) control.
2. Downstream operations, such as petrochemical and chemical production, aim to
produce specific products that meet the given specifications. So quality control is
the top priority. The market determines the product rate and can be unpredictable.
Large inventory buffers such as feed tanks and product storage tanks usually are
utilized to facilitate capacity control.
3. Mid-stream operations such as refineries have a good mix of quality control and
capacity control.
The following reference documents can provide a good start in understanding the
operating objectives at either the plant or unit level:
1. Existing process control narratives (PCN) document, if available. A good PCN
document is the best source of information to understand the process from a
process control perspective.
2. Process flow diagram (PFD): The PFDs provide the details of the process streams
and equipment. They typically also include each stream’s process conditions and
chemical compositions, such as the steady-state material balance data (Turton
et al. 2018).
3. Piping & instrumentation diagram (P&ID): P&ID provides detailed information
for the plant’s construction, including instrumentation and control.
4. Operating philosophy document and operating manual.
5. DCS Schematics.
This chapter will use the following process example to explain the methodology and
procedures of process analysis.
Example 6.1 Gas compression. Figure 6.1 shows a gas compression process in an
upstream production facility. The produced gas from the production separators is sent
to a produced gas compressor (PGC, K-101) for compression to the desired pressure.
The compressed gas is used as fuel gas by four heat recovery steam generators
(HRSG) as the top priority. The remaining gas is used as lift gas by multiple gas
lift wells. The diagram in Fig. 6.1 is a process overview diagram (POD) simplified
from the actual PFDs and P&IDs, with the non-control-relevant details removed for
clarity.
The operating objective is to provide a continuous gas supply at the desired pres-
sure (quality) and flow rate (quantity) for the steam generators and gas lift wells.
With Example 6.1, the process analysis methodology is explained step-by-step in the
following sections.
6.1 Operating Requirements 221

Fig. 6.1 A gas compression process

6.1.2 Product Specifications

All process operations aim at producing something useful, either as final products
or as semi-products. Most of these products have specifications on their quality or
quantity. Even the produced wastes would have to meet specific requirements before
they can be disposed of. For example, the produced water in Example 6.1 must be
sent to the injection wells or the deep disposal wells, as required by environmental
regulations.
The product specifications are vital to process engineering and process control
design; so it is crucial to get familiarized with the relevant product specifications and
their impact on control design.

6.1.3 Limits and Constraints

Another critical aspect of the operation is the design limits and operating constraints
that the process control scheme must honor. These limits and constraints may be
related to product quality or operating safety. They can significantly influence the
control design decisions. Each plant has its own operating limits and constraints,
even if it is based on the same design. The difference in the limits and constraints
is critical, and a control scheme designed for one plant cannot be copy-and-pasted
to another plant without going through detailed process analysis. The differences
must be considered in the overall control strategy rather than treated in a piecemeal
approach. See Sect. 1.3.5 for an explanation of how the operating setpoints and limits
stack up.
The control design for the compressor in Example 6.1 must observe the following
constraints to protect the equipment or piping:
222 6 Methodology of Process Analysis

1. Trip limit of the vapor recovery unit (VRU): The operating pressure of the vapor
recovery unit floats with the compressor suction pressure. The VRU has a trip
limit at 2 bar pressure, while the compressor suction pressure is around 1.4 bar
for normal operation. The flaring pressure for preventing high suction pressure
from tripping the VRU is set to 1.8 bar. There is very little room between the
normal operating pressure and the tripping pressures.
2. Trip limit at compressor discharge: Another constraint is the compressor discharge
pressure. High pressure is a safety risk to the compressor and piping and must be
avoided. For this purpose, a protective pressure controller is provided to prevent
the discharge pressure from going over the operating limits.
3. Compressor surge: Compressor surge is an abnormal condition caused by insuffi-
cient flow through the compressor. It must be prevented because it can potentially
lead to severe mechanical damages to the compressor or the associated piping.

6.1.4 Mode of Operations

It is essential to know all the modes of operation under which the plant/unit is expected
to operate. Typical modes of operation include startup mode, normal operating mode,
turndown mode, crippled mode, and shutdown mode. There are often other operating
modes specific to the operational requirements. For example, different feed requires
different operating conditions. Reduced capacity due to low demand or partial shut-
down may force the operation into a turndown mode. Failure in specific equipment
may cause the overall process into crippled-mode operation.
The challenge is the safe and smooth transition between the different modes of
operations. One example is polymer plant operation, where the production is a batch
operation. Each product is based on a different “recipe.” The transition from one
product to another, called “recipe” change, happens frequently. The transition is
expected to be quick and smooth to minimize off-spec products.
For the process control solution in Example 6.1, at least the following modes of
operation need to be considered:
1. Normal operation: During normal operation, the wells and compressor have suf-
ficient capacity to supply high-pressure gas needed by both the steam generators
and gas lift wells.
2. Compressor down: If the wells are shut-in or the compressor goes down, the entire
operation will enter crippled mode. The steam generators will draw gas from the
main gas line.
3. Steam generators down: If one or more of the HRSGs are offline, the operation
will transition to turndown mode with reduced capacity. The injection wells will
take all the produced gas, and the production will be subject to the injection well
capacity.
6.2 Process Flow Analysis 223

6.2 Process Flow Analysis

A modern operating plant design can often produce dozens or even hundreds of pages
of process flow diagrams (PFD) and piping & instrumentation diagrams (P&ID).
There are also a comparable number of DCS schematics. An inexperienced or unpre-
pared engineer can easily be overwhelmed by the daunting number of schematics
and fails to the “bigger picture.”
Process operation is mainly about material movement and energy transfer, gov-
erned by physical laws such as conservation of mass, energy, and momentum. The
mission of process control is to enforce these physical laws, not to defy them. Thus,
it is essential to start with a structured process flow analysis to identify the sys-
tem boundary, locate all the incoming and outgoing material and energy streams
of interest, identify any recycle and loop-back flows, list all possible process flow
line-ups, and enumerate all potential modes of operations (Arnold and Steward 1999;
Campbell 2004; Marlin 2015).

6.2.1 Process Flow and System Boundary

The process flow analysis is about the process topology. The study starts with a
process overview diagram (POD) derived from the PFDs and P&IDs. The POD
focuses on the “bigger picture” and provides a “full view,” in one single page, of
all major process flows, essential devices, equipment, final control elements, and
key process variables. The control-relevant content is extracted from the PFDs and
P&IDs and included in the POD. All the details not relevant to process control are
left out. The following information is generally expected in the POD:
1. The system boundary and the boundary conditions.
2. All the streams that may continuously flow into the system boundary (the supplies)
during normal operation, along with the stream specifications.
3. All the streams that may continuously flow out of the system boundary (the
demands) during normal operations, along with the stream specifications.
4. All the control elements such as switches and on/off valves potentially affecting
the process flow line-up during normal operation.
In Example 6.1, the system boundary is around the compressor, with three streams
coming in and four streams going out. It can serve as the initial POD and progressively
update with more information as the analysis continues:
1. Gas supply from the separator is the main process flow feeding the compressor.
The inlet pressure floats with the compressor suction.
2. Gas from the vapor recovery unit is a side stream with a much lower flow rate.
The pressure also floats with the compressor suction pressure.
3. The primary consumers of the compressed gas are the steam generators, with a
specification on the supply pressure.
224 6 Methodology of Process Analysis

4. The next consumers in line are the gas lift wells; each has a specification on the
gas flow rate.
5. Any excess gas is sent to the injection wells.
6. Sweet gas from the main gas pipeline is provided as make-up gas in case the
produced gas is insufficient.
7. Flaring is provided as an emergency measure to relieve excess gas during abnormal
conditions, e.g., compressor shutdown.

6.2.2 Key Disturbances

Noises and disturbances are a matter of life in the process industry (see Sect. 5.5.1).
Disturbances are the inputs or external conditions from the surrounding environ-
ment that the operator or the control system cannot control. Some typical process
disturbances include ambient air temperature, feed temperature, feed flow rate, feed
composition, steam pressure fluctuations, and cooling water temperature swings. For
example, the feed (flow, temperature, composition) is usually a critical disturbance
in distillation operation. The ambient temperature often has a significant impact on
the operation as well.
Disturbances can be measurable or unmeasurable. Both can drastically affect the
unit operation. For proper control design, all the primary sources of disturbances
must be identified and considered in the control strategy.
In Example 6.1, the process flow may be disrupted or altered by the following
events or devices:
1. Well shutdown: The compressor will run in turndown mode due to insufficient
gas when some production wells are shut down. Make-up gas from the main gas
line will be brought in to fill the gap in supply.
2. Compressor shutdown: When the compressor goes down, the gas will be drawn
from the main gas line to meet the demands of the HRSG and gas lift wells.
3. HRSG shutdown: When the steam generators are offline, the excess gas will be
diverted to the injection wells.

6.3 Supply and Demand Analysis

Process operation is about the continuous (or batch) movement of material, with
energy added or removed along the way. Any stream coming to the system boundary
is a supply, while any stream going out is a demand. Generally speaking, any stream
coming to a point or node is a supply, while any stream going away from the point is
a demand. At any point in the process flow, the material and energy must be balanced
statically and dynamically; otherwise, the imbalance will be internally accumulated
as inventory change or externally released abnormally, e.g., to flares.
6.3 Supply and Demand Analysis 225

A stream can be supply-driven or demand-driven, depending on how the flow

rate is determined. Consequently, the control scheme can be drastically different for
supply-driven or demand-driven operations. Some early discussions on supply and
demand control can be found in Bequette (2002); Luyben et al. (1998).

6.3.1 Material and Energy Balance

Material and energy balance follows the fundamental physical laws. It is one of the
primary considerations for process design, demanding the proper sizing and routing
of the piping, vessels, and equipment:

[Inventory Increase] = [Material In] − [Material Out]

= [Supply] − [Demand]. (6.1)

Since gas is compressible, the imbalance between supply and demand in a gas
processing unit impacts the pressure first, caused by the so-called line-packing. For
incompressible liquid flow, the imbalance between supply and demand is reflected
in liquid inventory, indicated by the fluid levels in tanks and vessels.
Sustained steady-state imbalance in material or energy flow is a process design
mistake. A common consequence is continuous flaring in the gas operation, over-
flow/emptying of surge tank in liquid operation, or choked operation with reduced
capacity. During process control design, it is essential to review the stream data,
such as the design values of the flow, temperature, and pressure, and validate the
steady-state material/energy balance against the different modes of operations, such
as startup, normal operation, turndown operation, and depleted operation. The stream
data can typically be found in the PFDs.
A crucial concept is the swing stream. Along any process flow path, the supplies
and demands must not be all fixed. At least one supply or demand stream must
have sufficient capacity to “absorb” the fluctuations or imbalances between supply
and demand. That is, the flow rate must have room to swing up and down. For
Example 6.1, the gases from the wells and the VRU are the supplies, with their
flow rates determined by upstream operation. The gases to the wells and HRSG are
demands, with their flow rates set by the respective wells and HRSGs. Based on the
reservoir simulation results, the gas supply may be less than the demand for some
operating scenarios; thus, the sweet gas stream is available as make-up gas when
needed, which serves as a swing stream.

6.3.2 Supply-Driven and Demand-Driven Operation

The supply and demand model describes how the material and energy are moved
with the process flow. Each stream can be a supply or a demand, and each node on
the flow path can have a mix of supply and demand coming in and going out.
226 6 Methodology of Process Analysis

1. Supply-driven stream: A stream is supply-driven if the upstream operation dic-

tates the flow rate, and the current node has no control over it. A supply-driven
stream is also called a supply-pushing stream since it is pushed into the node. For
example, the produced gas in Example 6.1 is a supply-driven stream since the
production of the wells determines how much gas is produced and comes into the
system boundary. The operating objective is to maximize the (oil) production, and
the surface facility is expected to adapt to the production variations. Similarly, the
gas from the vapor recovery unit (VRU) is supply-driven since the compression
unit has no control over how much gas will come in from the VRU.
2. Demand-driven stream: If the consumer determines how much it takes from the
system, it is a demand-driven or demand-pulling stream. The primary consumer
of the compressed gas is the heat recovery steam generators (HRSG). The HRSGs
operate at a pre-defined load. So the gas flow stream to HRSG is determined by
the HRSG load and is thus demand-driven. Similarly, the gas lift determines the
gas lift wells’ gas flow rate; therefore, this stream is also demand-driven.
3. Reversible supply and demand model: A supply-driven (“pushing”) stream can
become a demand-driven (“pulling”) stream, and vice versa, depending on the
material balance. In Example 6.1, the HRSG pulls gas from the compression unit
based on how much it needs. However, if the valve to the HRSG is already fully
open, but HRSG still cannot get all it needs, then the HRSG will have to settle
with whatever it can receive. It then becomes supply-driven. This reversal may
be only temporary if the steady-state material balance is adequate. It may also
become permanent if the supply and demand model has changed permanently.
In summary, for Example 6.1, the process inside the system boundary is a mixed
supply-driven and demand-driven operation with two supply-driven streams (pro-
duced gas and VRU gas), two demand-driven streams (fuel gas for HRSG and lift
gas for wells), and two swing streams.

6.3.3 Supply-Driven and Demand-Driven Control

Process control enforces the supply and demand model. The control strategy can be
supply-driven control or demand-driven control corresponding to the supply-driven
or demand-driven operation. It is essential to know how to recognize or design a
supply-driven control or a demand-driven control. Typical control configurations are
shown in Fig. 6.2 and may include the following common scenarios:
1. No control or controller out of service ⇒ supply-driven.
2. Controller in manual mode or control output saturated ⇒ supply-driven.
3. Standalone flow controller upstream of system boundary ⇒ supply-driven (Case
1a in Fig. 6.2).
4. Standalone flow controller downstream of system boundary ⇒ demand-driven
(Case 2a).
5. Pressure controller upstream of a control valve ⇒ supply-driven (Case 1b).
6.3 Supply and Demand Analysis 227

Fig. 6.2 Control options for supply and demand operation

6. Pressure controller downstream of a control valve ⇒ demand-driven (Case 2b).

7. Level controller upstream of a control valve ⇒ supply-driven (Case 1c).
8. Level controller downstream of a control valve ⇒ demand-driven (Case 2c).
9. Standalone flow controller inside system boundary ⇒ attention! There cannot be
two active flow controllers (over-specified!) on any continuous flow path unless
there is also a swing stream.
The above rules can be used as a guide for placing process measurements and final
control elements. These rules can also be followed when analyzing an existing process
design.
The imbalance between supply and demand is sensed by pressure or level. Capac-
ity control for gas processing aims to maintain a desired pressure–flow profile along
the process flow path, and capacity control for liquid processing maintains a healthy
level–flow profile.
The supply and demand analysis starts from the incoming streams at the system
boundary and propagates downstream for supply-driven operation. Conversely, the
analysis starts from the demand point at the system boundary and propagates back-
ward to upstream for demand-driven operation. For example, for the supply-driven
operation Fig. 6.3, the flow and level control are so placed that the fluctuation in
228 6 Methodology of Process Analysis

Fig. 6.3 Supply-driven liquid processing operation

supply is reflected in level changes. The flow controller FC-101 sets the supply
flow rate. Level controllers (LC-102, LC-103) adjust the outlet flows and effec-
tively propagate the fluctuations to downstream operation. The downstream (after
FCV-103) must have the swing capacity to “absorb” the fluctuations.
On the other hand, if the capacity is set by downstream demand, the control design
would be entirely different (See Fig. 6.4). In this demand-driven control, the flow
controller FC-104 sets the demand. Any changes in the demand will affect the tank
levels, and the level controllers (LC-102, LC-103) would have to adjust the inlet
flows to keep the level constant. This level–flow interaction propagates backward all
the way to the inlet flow of the first tank, which must have sufficient swing capability
to absorb the demand fluctuations.
For gas processing, the focus is the pressure–flow profile instead of the level–
flow profile. Figure 6.5 is an example of a supply-driven gas processing operation.
The fluctuation in gas supply to the compressor is reflected in the pressures. The
pressure controllers force more flows through the compressor and piping. This flow–

Fig. 6.4 Demand-driven liquid processing operation

6.3 Supply and Demand Analysis 229

Fig. 6.5 Supply-driven gas processing operation

pressure–flow–pressure–flow propagation goes on until reaching the final consumer.

The consumer of the gas needs to have the capacity to absorb the swing in supply.
For the same process, if driven by demand, the control strategy would have to be
changed to that shown in Fig. 6.6. The critical requirement is that the fluctuation in
demand must be propagated to, and accommodated by, the supply.
A noteworthy scenario is that a supply-driven operation can become demand-
driven temporarily or permanently, and vice versa. In this case, the control strategy
will need to be changed to supply-driven control with demand override or demand-
driven control with supply override. This concept is illustrated in Fig. 6.7. Overriding
control is prevalent in processing facilities in E&P operation (See Sect. 2.2.7).
Adding override protection against supply and demand reversal, Fig. 6.3 would
become Fig. 6.8. In case the demand side cannot take all the flows (e.g., the valve is
already fully open), the operation automatically reverts to demand-driven, and the

Fig. 6.6 Demand-driven gas processing operation

Fig. 6.7 Supply and demand operation with override protection

230 6 Methodology of Process Analysis

Fig. 6.8 Supply-driven control with demand override

overriding controllers (LC-103B and LC-102B) take over, propagating the demand
back to the supply side. Again, the upstream operation must be able to reduce supply
to adapt to the demand change.

6.3.4 Swing Capacity and Swing Control

The swing capacity is critical in balancing supply and demand dynamically. In pro-
cess design, control design, and control troubleshooting, the same question should be
asked repeatedly: where is the swing? The swing capacity can come from multiple
sources. The process inventory can accommodate minor variations in supply and
demand. For example, for gas processing, packing up the pipeline and equipment
can provide some swing capacity. In liquid processing, because the liquid is not
compressible, there is not much swing capacity in the piping. The swing capacity is
mainly provided by the surge vessels and storage tanks. In practice, the line-packing
and liquid inventory may not be sufficient for swing control. For reliable operation,
proper swing streams need to be provided in the design:
1. For a supply-driven process, such as in Fig. 6.3, the final demand must have
sufficient swing capacity to take the fluctuation in supply. Level control LC-103
will propagate and push the fluctuations to the downstream operation and expect
that the downstream operation has the appropriate swing control to accommodate
the swing.
2. For the demand-driven process in Fig. 6.4, the capacity is determined by the final
demand and propagates back to the supply. The level controller LC-102 will
back-propagate the demand variations upstream of the FCV-101 and expect that
upstream has swing control to adjust the supply.
6.3 Supply and Demand Analysis 231

3. For a process where the supply and demand model may reverse, both the supply
and demand will have to be able to swing. The case with Fig. 6.8 is an example.
Although seemingly complex, this reversal in supply and demand is common in
practical applications.
4. For a mixed model with both supply-driven and demand-driven streams, as in
Fig. 6.1, dedicated swing streams such as the flows to injection wells, the make-
up gas, and the gas to flare are provided.

6.4 Cause and Effect Analysis

Process control, either open-loop or closed-loop, is based on the assumption that

adjusting some variables (control handles) can cause changes to some other variables.
The cause and effect analysis aims to identify all key process variables (the effect) and
ensures the necessary control handles (the cause) are available to affect the desired
changes.

6.4.1 Controlled Variables

Although the number of process variables in scope can be overwhelming, a small

subset of the variables can often provide a sufficiently reliable indication of the
overall operating health. Keeping these process variables at their desired values
ensures process stability (Juliani and Garcia 2017). We will call these variables the
key process variables (KPV). Most of them will become the controlled variables for
the process control.
The key process variables typically include the following:
1. Quality control variables: All measurements with commercial or operational spec-
ifications on their quality or quantity, such as composition, impurity, dew point,
vapor pressure, qualify as quality control variables. Pressure, temperature, and
flow rate can also be product specifications. In Example 6.1, the gas supply pres-
sure (PT-105) for the HRSGs and gas lift wells is a quality control variable.
2. Capacity control variable: Capacity control is for maintaining the dynamic supply
and demand balance. Some critical variables are the relay points for the control
action propagation, typically pressures (for gas) or levels (for liquid). In Exam-
ple 6.1, the suction pressure PT-101 is a capacity control point since the pressure
is an indicator of the supply and demand imbalance and can be used to adjust the
compressor speed to achieve the required capacity. The discharge header pressure
PT-105 is another capacity control point that needs monitoring and control.
3. Constraint variables: These are the limits and constraints that define the operating
and safety envelope. They can lead to the shutdown of the operation if not con-
trolled or limited. In Example 6.1, the suction pressure PT-101 is a constraint
variable since an excessively high pressure may trip the vapor recovery unit.
232 6 Methodology of Process Analysis

Fig. 6.9 A gas compression process: key process variables

The compression process is shown again in Fig. 6.9, with the key process variables
added.

6.4.2 Manipulated Variables and Disturbance Variables

The causes for process changes could be active or passive. A final control element
such as a valve or electric switch can actively cause a change to the process, while
ambient air temperature may passively affect the process variables. The causes that
we can manipulate are called the manipulated variables (MV), and those that we
cannot manipulate are called the disturbance variables (DV).
Let us examine all the control handles in Example 6.1. Temporary tag names are
used for now. They will be assigned permanent tag names later once the control
strategy is finalized. The manipulated variables include the following:
1. MV-01, Flaring valve: A control valve to send excess gas to flares.
2. MV-02, Compressor speed: Usually, an electric motor drives the compressor via
a variable speed gearbox (VSGB) to adjust the compressor speed.
3. MV-03, Compressor recycle valve: This valve is primarily for anti-surge control
(see Sect. 11.3).
4. MV-04, Make-up gas valve: A control valve to bring in sweet gas as make-up
gas.
5. MV-05, Supply valve for injection wells.
6. MV-06, Supply valve for gas lift wells.
7. MV-07, Supply valve to HRSG.
Figure 6.10 shows the process flow with the control elements (valves and motor
drives). The placement of the control elements is based on the supply and demand
analysis and is in line with supply-driven or demand-driven operating requirements.
6.4 Cause and Effect Analysis 233

Fig. 6.10 A gas compression process: with control valves

6.4.3 Cause and Effect Relationships

We use the manipulated variables (MV) to keep the controlled variables (CV) at the
targeted values to maintain the material and energy balance. The vital requirement
is that the cause and effect relationship between the MVs and CVs must exist and
must be known. When designing or analyzing a process control scheme, it is crucial
to identify all the key manipulated and controlled variables and examine the cause
and effect relationships.
The relationship may be single-input single-output (SISO) or multi-input and
multi-output (MIMO). Most of the cause and effect relationship is simple and obvi-
ous. For example, it is evident that the flow control valve is the cause for a flow
controller, and the flow rate is the effect. However, in many cases, the flow rate may
be the cause for other effects, e.g., in a level-to-flow cascade loop, the flow is a cause
and an effect simultaneously. In complex control schemes with multiple MVs and
multiple CVs, the cause and effect relationship can become quite complicated and
need to follow a structured approach for the analysis:
1. Focus on the “bigger picture.” The challenge for a multivariable process is in the
interactions between process variables. A change in one MV can cause changes in
multiple CVs. Similarly, the desired CV change can be effected by adjusting mul-
tiple MVs. In the compressor example (Fig. 6.10), the suction pressure PC-102
can be controlled by changing the compressor speed or adjusting the recycle
valves. Using two MVs to control one CV is a design decision, as discussed in
Sects. 2.2 and 4.1.
2. Simplify. The purpose is to examine the interaction between the MVs and CVs,
and certain localized variables can be simplified for better clarity. For example,
in a temperature–flow–valve cascade control, the cause and effect relationship
between the control valve and the flow rate is well-known and localized to the
flow. The effect of the valve on other variables goes through the flow; therefore,
the flow is considered a manipulated variable instead of the control valve.
234 6 Methodology of Process Analysis

Table 6.1 PGC example: cause and effect table, preliminary

CV-01 CV-02 CV-03 CV-04
Suction pressure Discharge pressure Lift-gas flow Fuel gas pressure
MV-01 Valve, flare ×
MV-02 Speed × × ×
MV-03 Valve, recycle × × ×
MV-04 Valve, makeup gas × ×
MV-05 Valve, injection ×
MV-06 Valve, gas lift ×
MV-07 Valve, HRSG × × ×

Following the supply and demand analysis in Sect. 6.3, let us examine the cause and
effect relationship among the manipulated variables and the key process variables in
Fig. 6.9. A cause and effect table (CET) is produced in Table 6.1, where “×” indicates
that a cause and effect relationship exists between the corresponding variables.
From the table, it is clear that a final control element (e.g., compressor speed
MV-02) can cause multiple process variables (CV-01, CV-02, and CV-03) to
change, while a process variable (e.g., CV-02) may be affected by several control
elements.
“Correlation does not imply causation” (Aldrich 1995). Many times the cause and
effect relationship is just a correlation in process input/output data. The cause and
the effect must be further examined and validated from the physical principles of the
process dynamics.

6.5 Degree of Freedom Analysis

Degree of freedom (DOF) is the number of variables that must be specified to define
a process completely; for process control design, every manipulated variable (MV)
or control handles, such as a valve, motor speed, or a slave controller, is counted as
one DOF.

6.5.1 Degree of Freedom Requirements

Depending on the number of controlled variables and manipulated variables, the

control problem can be as follows:
1. Over-specified: The number of controlled variables is more than the number of
manipulated variables. All the controlled variables cannot be maintained, and
some must be neglected or compromised.
6.5 Degree of Freedom Analysis 235

2. Under-specified: The number of manipulated is more than the number of con-

trolled variables. There may not be a unique solution in terms of control action,
and some of the manipulated variables will eventually be driven to saturation to
exhaust the DOFs.
3. Fully specified: The number of manipulated variables is the same as the number
of controlled variables.
A DOF must be a variable that can be independently manipulated and is usually
provided by a final control element such as a valve or a motor. Typically the DOF
has a cost in terms of CAPEX and OPEX, and an unconsumed DOF is a waste. It is
expected that process control design will fully utilize all DOFs. A control valve that
remains saturated (fully open or fully closed) renders this DOF unavailable. Leaving
a valve in the manual mode also removes the valve as a DOF. On the other hand,
sometimes multiple DOFs are required to control one process variable, such as in a
split-range control or fan-out control scheme where multiple valves are counted as
one DOF.
For proper control design, all DOFs must be consumed by pairing up all the vari-
ables and controlled variables. A controlled variable, once paired with a manipulated
variable, consumes one DOF. Two controlled variables must be sufficiently indepen-
dent of each other to consume two DOFs. If they are strongly correlated, they may
be considered one controlled variable and consume only one DOF:
1. For simple control (Chap. 3), DOF analysis is usually straightforward. For exam-
ple, a standalone flow controller has the flow rate as the controlled variable and
the control valve as the manipulated variable and is thus fully and properly paired
up.
2. In complex PID control schemes (Chap. 4), such as those containing split-range
control, or selective control, the DOF condition is less obvious and often over-
looked. The consequence is poor control performance.
3. In model predictive control (MPC, Chap. 5), the DOF condition is handled implic-
itly in the model structure and needs to be very careful. Relative gain analysis
(RGA) is typical to verify the DOF condition and variable pairing are adequate.

6.5.2 Pairing of Process Variables

The manipulated variables (MV) and controlled variables (CV) must be fully paired
up to leave no DOF unconsumed. The MV and CV pairing needs to consider many
factors, with the supply and demand model and the cause and effect relationship as the
critical considerations. In the multivariable control scheme, the following questions
are required to be answered:
1. Which MV to control which CV? Generally speaking, a CV should be controlled
by the MV with the strongest and most direct effect.
2. What if there are not enough MVs to control all the CVs? When there are more
CVs than MVs, the typical approach is to reduce the number of CVs via cascade
control, override control, selective control.
236 6 Methodology of Process Analysis

3. What if there are more MVs than CVs? When there are more MVs than CVs, the
typical approach is complex PID schemes with split-range or fan-out control.
4. What if multiple MVs affect multiple CVs? Strongly interacting MVs and CVs
constitute a multivariable control problem, and explicit or implicit decoupling is
typically needed.
Let us continue the compressor example. Table 6.1 has provided all the critical cause
and effect relationships between the final control elements and the key process vari-
able. Some of the variables will pair up as manipulated variables and controlled
variables through a DOF analysis. Others may need further actions to ensure the
DOF conditions are met. We will go through all the variables one by one as follows:
1. CV-01, the suction pressure: The suction pressure is a key process variable. It is
the primary indicator for supply and demand balance and thus should be upgraded
to a controlled variable. During normal operation, the compressor is supply-driven
by design. Therefore, this pressure controller serves as the capacity controller. The
compressor speed MV-02 is the obvious choice to pair with this CV based on
engineering insight. In terms of supply and demand control, this is Case 2a in
Fig. 6.2.
2. CV-02, the discharge pressure: The discharge header supplies the compressed
gas to multiple consumers and should be another controlled variable. By design,
the compressed gas is supply-driven, and the injection gas is the swing stream.
As a result, the discharge header pressure CV-02 is paired with the injection
gas flow valve MV-05. The supply and demand model is Case 3b in Fig. 6.2 for
supply and demand balancing.
3. CV-03, the lift gas flow: Per operating requirements, the lift gas flow is demand-
driven. Therefore, the gas flow CV-03 pairs up with the flow control valve MV-06.
The supply and demand model is Case 2a in Fig. 6.2.
4. CV-04, fuel gas to HRSG: The make-up gas is another demand-driven stream,
and the flow rate is dictated by gas pressure. So the CV-MV pair-up is CV-04
with MV-07. The supply and demand model is Case 2b in Fig. 6.2.
With the above CVs and MVs paired up, the cause and effect diagram is now shown
in Table 6.2, where the symbol ⊗ indicates the paired variables.
Now let us look into the remaining variables that are not yet paired:
1. MV-01, flare valve: This valve sends excess gas to flare in case of imbalance
caused by more supply than demand. The controlled variable to pair with should
be the suction header pressure, which has already been paired with compressor
speed MV-02. The flare valve is an extra DOF to be consumed.
2. MV-03, recycle valve: The recycle valve is dedicated by design for anti-surge
control (see Sect. 11.3). It also has a strong effect on the suction header pressure
CV-01. From engineering insight, the compressor speed also strongly affects the
surge condition of the compressor. This control problem thus has two MVs and
two CVs to pair up, and a multivariable control scheme is needed to handle the
complex cause and effect relationship.
6.5 Degree of Freedom Analysis 237

Table 6.2 PGC example: cause and effect table, updated

CV-01 CV-02 CV-03 CV-04
Suction pressure Discharge pressure Lift-gas flow Fuel gas pressure
MV-01 Valve, flare ×
MV-02 Speed ⊗ × ×
MV-03 Valve, recycle × × ×
MV-04 Valve, makeup gas × ×
MV-05 Valve, injection ⊗
MV-06 Valve, gas lift ⊗
MV-07 Valve, HRSG × × ⊗

3. MV-04, make-up gas valve: This make-up gas flow valve is open only when the
overall supply is less than demand, and in normal operation, it is fully closed by
design. The effect of this valve is on the discharge pressure header CV-02, which
has already been paired up with injection valve MV-05. This make-up gas valve
is an extra DOF to be consumed.
The different control options will need to be evaluated to pair up all the process
variables completely. We will defer the remaining DOF analysis to Chap. 7.

6.6 Dynamics Response Analysis

The cause and effect analysis in Sect. 6.4 reveals the cause and effect relationships
between the process variables. For process control, the bare existence of a cause and
effect relationship is not enough. A quantitative description of the relationship is
essential to process control design.
Most process control technologies assume that the process dynamics are rela-
tively linear and do not (significantly) change with time. However, most industrial
processes are inherently nonlinear and constantly change with time. In addition, this
nonlinearity and time-variance are often unpredictable. Therefore, it is critical to
know the process dynamic characteristics and have a good idea of the applicability
range of the control solution.

6.6.1 Process Disturbances

Disturbances and fluctuations are inevitable in process operation! They are the main
reasons for the need to control. There are many types of process disturbances from
different sources and at different levels.
238 6 Methodology of Process Analysis

Disturbances can be measurable or unmeasurable. Process control can reduce

the impact of measurable disturbance by resorting to feedforward control or model
predictive control. For unmeasureable disturbance, the only means to combat it is
probably feedback control, after it has impacted our controlled variables.
It is important to remember that process control cannot reject disturbances. It
simply transfers the noise and disturbances from one place to another, from a critical
variable to a less critical one. For example,
1. A flow is stabilized by shifting the fluctuations to valve movements. As per the
design, if the valve is driven by instrument air, the flow fluctuations are shifted to
the instrument air consumption.
2. A temperature in a heater is maintained constant by shifting the temperature
variations to the fuel oil flow.
3. With a surge tank, inlet disturbance is absorbed by level (less critical) in exchange
for a smoother outlet flow (critical variable); conversely, the tank level can be
maintained stable by shifting the level fluctuations to the outlet flow and then to
downstream (e.g., a storage tank).
4. Poor control itself can generate disturbances as a result of, for example, valve
malfunction and poor controller tunings.
When a disturbance is included in the control design, it is crucial to consider where
it is expected to be transferred to.

6.6.2 Dynamics Response

Process dynamics refer to the time trajectory of a variable in response to a change

in another variable. All control actions are based on known and predictable process
dynamics. The type of dynamic response determines the applicability of the different
process control technologies and affects the control strategy.
We all have the inherent appreciation that the effect of a cause takes time to
manifest itself. It takes seconds to see the flow change after a valve position change,
but it may take many minutes or even hours before its impact on product quality can
be observed. In general, the dynamic response of a stable (self-regulating) process
has three key characteristics, that is, once there is a change in input variable u(t),
when, how much, and how fast the output variable y(t) will respond:
1. Process gain (K p ): By moving the input variable by one unit, how many units of
change (direction and magnitude) will be eventually caused on the output variable.
A stable process will eventually reach a new steady-state, while an integral (ramp)
process does not have a new steady-state.
2. Time delay or dead time (θ ): After a change is made to the input (the manipulated
variable, or MV), when will the output variable (the controlled variable, or CV)
start to move? Time delay is typically caused by transport delays. Large time
delay is one of the most significant challenges for control.
6.6 Dynamics Response Analysis 239

3. Transient behavior: After the time delay elapses and the output starts to move,
how fast will the output reach a new steady-state, and what trajectory will it
follow?
Although a first-order-plus-delay-time (FOPDT) model can approximate most
dynamic response behaviors, many processes exhibit higher-order or nonlinear char-
acteristics. The different process dynamics differ mainly in the transient behav-
ior, including the response speed and trajectory. Figure 6.11 shows the commonly
encountered simple process dynamics:
1. First-order dynamics (Fig. 6.11a): Most industrial processes can be sufficiently
approximated by a first-order-plus-dead-time (FOPTD) characteristic, where the
transient behavior is defined by process gain K p , delay θ , and time constant τ ,
and d is the time delay in the number of samples:

K p −θ s
y(t) = K p e− Ts
t−θ
G(s) = e → (6.2)
τs+1
b1 z −θ−1
G(z −1 ) = → y(k) = −a1 y(k − 1) + b1 u(k − d − 1). (6.3)
1 + a1 z −1

The time constant is the amount of time that takes from the moment that the
output variable starts to respond until the output value reaches 63.2% of the new
steady-state value.
2. Higher-order dynamics (Fig. 6.11b): Although most process dynamics can be
approximated with a first-order-plus-dead-time, some processes exhibit a truly
high-order characteristic (characterized by the “S” shaped rising curve). Second-
order process dynamics may be sufficient to approximate most high-order
dynamics:

Fig. 6.11 Step response of common process dynamics

240 6 Methodology of Process Analysis

Kp b1 z −1 + b2 z −2 −d
G(s) = e−θ s → G(z −1 ) = z
(τ1 s + 1)(τ2 s + 1) 1 + a1 z −1 + a2 z −2
(6.4)
y(k) = −a1 y(k − 1) − a2 y(k − 2) + b1 u(k − d − 1) + b2 u(k − d − 2).
(6.5)

3. Integrating process (Fig. 6.11c): Some processes, such as the liquid level in a tank
or the pressure in a fixed volume, are not self-stabilizing. They behave like a ramp,
and there is no steady-state in step response:

K p −θ s Kp
G(s) = e or G(s) = e−θ s . (6.6)
s s (τ s + 1)

4. Inverse response (non-minimum phase, Fig. 6.11d): Some processes may show
strange behaviors and pose a significant challenge to control. For example, when
cold feed-water is added to a steam boiler, the water level may go in the “wrong”
direction before returning to the “right” direction. The control solution thus needs
to address this behavior.
5. Unstable: Some processes, such as an exothermic reaction process, are inherently
unstable. If not adequately controlled, it can diverge quickly and lead to loss of
containment and possible explosion.
6. Nonlinear process: A nonlinear process exhibits different dynamics at different
operating conditions. A notorious example is pH control, where a flow of acid or
base (called the reactant) is added to the product stream to control the pH value to
a given value. The process is very nonlinear, and an example of the process gain
is illustrated in Fig. 6.12. A simple PID control is almost impossible for proper
pH control. A feedforward control, ratio control, or preferably model predictive
control is often needed to battle the nonlinearity.
Another example is the high purity distillation. The impurity level is approx-
imately a linear function of the temperature at a particular tray. However, the
relationship changes to a logarithm function when the purity level is very high
(low impurity).
7. Time-varying: Most processes are time-varying by nature due to, for example,
fouling of piping and equipment, de-activation of catalysts, wear of tear of valves,
and other moving parts. The dynamic response will be different, although the
change is typically gradual.
In practice, the determination of the process dynamics starts with a sound engineering
understanding of the cause and effect relationship, confirmed by experimental or plant
tests, and quantitatively described by process models. Process models are facilitated
with process identification software. Decent identification software supports both
model structure identification and parameter estimation that help supply and demand
analysis and dynamics analysis. See Chap. 5 for discussions on process identification.
6.6 Dynamics Response Analysis 241

Fig. 6.12 Example of nonlinear process: pH process gain

Depending on the type of dynamics, different models may be required. Subse-

quently, other control technologies may have to be considered. More discussions will
follow in Chap. 7.

6.7 Process Analysis Documentation

We conclude this chapter by summarizing the best practice for process analysis and
formalizing the two simple visual design tools for the visualization and documenta-
tion of the analysis results.
When an unfamiliar process is encountered, it is helpful to follow a structured
process analysis methodology to familiarize oneself with the process quickly. The
information required by process control is quite different from that required by pro-
cess engineering; therefore, it is vital to know what to look for:
1. Understand the operating objectives and requirements.
2. Identify all the flow streams that may affect material and energy balance.
3. Decide on the key process variables to be controlled.
4. Identify all the significant disturbances and constraints.
5. Determine the supply and demand model for each stream.
6. Decide on the placement of final control elements based on the supply and demand
model.
7. Perform a cause and effect analysis to pair up the controlled variables and manip-
ulated variables. Add or relocate the variables as necessary.
8. Gain a basic idea of the dynamic response of each cause and effect relationship,
and decide on the appropriate type of control needed.
9. Repeat the above steps if the process design is modified or new information is
obtained.
242 6 Methodology of Process Analysis

Fig. 6.13 Visualization tools for process analyses

The process engineers are the best resources for in-depth discussions during process
analysis.
Two visual design tools are helpful to document the analysis results and for more
effective communication with other teams. One is the process overview diagram
(POD) for providing a holistic view of the process flow, and the other is the cause
and effect table (CET) for visualizing the cause and effect relationship and optionally
the dynamic response characteristics. Both have been extensively used in this chapter
(Fig. 6.13).
Figures 6.1, 6.9 and 6.10 are examples of POD for unit operation. Figures 1.3
and 12.27 are POD examples for large-scale plant operations.
The CET can be simple or complex. Tables 6.1 and 6.2 are simple CET examples
showing the model structure and the cause and effect relationships without the process
dynamics. Table 5.2 and Fig. 5.14 are examples of the CET with dynamic responses.
The crucial information in the CET include the following:
1. The controlled variables (CV), the manipulated variables (MV), and the distur-
bance variables (DV).
2. The existence of a cause and effect relationship between an MV/DV and CV.
3. The approximate characteristics of each identified cause and effect relationship.
4. The quality (credibility) of each identified cause and effect relationship.
Both the POD and CET are progressively developed as the process analysis proceeds.

6.8 Summary

There is an infinite number of processes in various process industries. Even the

same process design may have different operating requirements and constraints in a
different plant (Sect. 6.1). Although we strive for simplification and standardization,
we must validate each control solution to ensure it is fit-for-purpose. On the other
6.8 Summary 243

hand, it is unrealistic, nor necessary, to expect a process control person to achieve the
same level of process understanding as a process engineer does. As a result, one of
the main challenges facing process control engineers or researchers is, given a new
and unfamiliar process unit, where to start with the process analysis to quickly reach
the level of understanding required to perform process control duties.
Fortunately, the process knowledge required by a process control is very differ-
ent from that of a process engineer. We can quickly and systematically gather what
we need for the underlying process following well-defined procedures and guide-
lines with the basic knowledge in process engineering such as thermodynamics, heat
transfer, fluid mechanics, mass transfer, and chemical reaction.
Chapter 5 already provides a glimpse into the process flow analysis methodol-
ogy for model predictive control. This chapter formalizes this work process and
presents the essential skills and practical methodology for understanding the process
dynamics. These include process flow analysis (Sect. 6.2), cause and effect analysis
(Sect. 6.4), supply and demand analysis (Sect. 6.3), DOF analysis (Sect. 6.5), and
dynamic responses analysis (Sect. 6.6).
The scope and requirements of process analysis are summarized in Fig. 6.14.
Based on the complexity and specificity, all or part of the procedure can be fol-

Fig. 6.14 Scope and checklist for process analysis

244 6 Methodology of Process Analysis

lowed when an unfamiliar process is encountered. Although the process engineers

are always the best source for in-depth knowledge when needed, we need to know
enough to know what we do not know.

References

Aldrich J (1995) Correlations genuine and spurious in Pearson and Yule. Stat Sci 10(4):364–376
Arnold K, Steward M (1999) Surface production operations—design of oil-handling systems and
facilities, vol 1, 2nd edn. Imperial College Press
Bequette BW (2002) Process control-modeling, design, and simulation. Prentice Hall
Campbell J (2004) Gas conditioning and processing—the equipment modules, vol 2, 8th edn. John
M, Campbell and Company
Downs J, Skogestad S (2011) An industrial and academic perspective on plantwide control. Ann
Rev Control 99–110
Juliani RCG, Garcia C (2017) Plantwide control: a review of design techniques, benchmarks, and
challenges. Ind Eng Chem Res 56(28):7877–7887
Luyben WL, Tyréus BD, Luyben ML (1998) Plant-wide process control. McGraw Hill, New York
Marlin TE (2015) Process control—designing processes and control systems for dynamic perfor-
mance, 2nd edn. McGraw-Hill
Seames W (2018) Designing controls for the process industries. CRC Press
Turton R, Shaeiwitz JA, Bhattacharya D (2018) Analysis, synthesis, and design of chemical pro-
cesses. Prentice Hall
Chapter 7
Methodology of Process Control Design

Experience shows that a large percentage of operational difficulties originate from

poor process control design. Once in operation, the issues caused by flawed design
are typically difficult and expensive to correct. A modern operating plant is usually
very complex, so is the control solution (Sect. 7.1). The flawed control design is
often caused by the lack of a coherent overall control strategy (Sect. 7.2). It is not
unusual that all the individual control loops perform well in silos but fail to work in
concert when put together. So a structured design methodology should be followed
to achieve optimality at all plant levels (Sect. 7.3). The control solution is an integral
part of the plant operation; the documentation and visualization of the control scheme
(Sect. 7.4) are critically important but have been long neglected or undervalued. This
chapter provides some general discussions and guidelines to address these issues.

7.1 Process Hierarchy and Process Control Scope

A control solution must serve a clearly defined control objective. A modern produc-
tion process can be overwhelmingly complex, so is the control solution. It is essential
to follow a “divide and conquer” approach and break the process and control into
more manageable chunks while not losing sight of the “bigger picture.” (Åström and
Kumar 2014; Marlin 2015; Seames 2018).
This design methodology and the link between process analysis and control design
are illustrated in Fig. 7.1. The critical requirement for process control design is the
strong connection and interaction between process design and control design. Process
design must keep the process control strategy in mind so that the appropriate control
handles and measurements are available and with adequate capacity or rangeability
(Marlin 2015). The control design may have to go back to process analysis frequently
to validate and improve the design. In control troubleshooting and improvement, the
same methodology applies.
This tight integration between process and control is exemplified by model pre-
dictive control (Chap. 5). It is worth re-emphasizing that the process analysis is
the basis for control design. Whether selecting a base layer control strategy or an
advanced control strategy depends on the result of process analysis. As discussed

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 245
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_7
246 7 Methodology of Process Control Design

Fig. 7.1 From process analysis to control design

before, control design is always a trade-off of many factors, and the process analysis
result is the basis for the decision.

7.1.1 Process Hierarchy

A production plant can be viewed as a process hierarchy. The ISA-88 standards

define the physical model of operation as an enterprise, site, area, process cell, unit,
equipment module, and control module. However, we will only consider three levels
for control design, i.e., plant-level, unit-level, and loop-level operations.
An operating plant typically consists of one or more process units. Each unit
comprises one or more control loops, and each loop may be a simple PID controller,
a complex PID loop, or a model predictive control scheme. At each level, the operating
requirements are different, and the control objectives are thus also different:
1. Plant level: The control objectives at the plant level are typically related to main-
taining adequate material and energy balance. The material balance is reflected by
the pressure–flow profile (for gas) or the level–flow profile (for liquid). The mate-
rial balance analysis may run across multiple operating units and sometimes even
cross the battery limit. Maintaining the correct supply and demand profile, both
static and dynamic, is the primary consideration in the overall control strategy.
2. Unit level: At the unit operation level, product quality control is prevalent. Pres-
sure, temperature, and quality control are commonplace. Energy integration and
optimization have resulted in more compact and complex process design and thus
made process control more challenging. Most of the process challenges discussed
in Sect. 5.1.1 are the unit-level considerations, which have the most significant
impact on the control structure.
3. Loop level: A control loop can be a simple standalone PID loop, a complex PID
scheme, or a model predictive controller (MPC). The control loop performs a
specific control task that is part of the overall unit-level control strategy, which
in turn is part of the plant-level control strategy. Whether a PID or MPC, the
7.1 Process Hierarchy and Process Control Scope 247

controller is an integral part of the control loop and must always be treated that
way.
A PID controller is like a well-trained soldier. One soldier may be good enough
to guard a post. To hold a fortress, it needs a team of well-trained soldiers, while
to protect a city, it would need an army of well-organized and highly-coordinated
teams. Controlling a process unit is like guarding a fortress while controlling a plant is
like protecting a city. Individual well-trained soldiers are not enough. It needs plans,
strategy, tactics, and most importantly, coordination to achieve a higher goal. The
best process control solution is always a fit-for-purpose solution based on process
design and meets the operating requirements.

7.1.2 Plant-Wide Control

An operating plant typically consists of multiple interconnected units. The mate-

rial moves from upstream to downstream, often with recycles. Along with material
movements, there are energy changes and exchanges.
For example, in the E&P production plant illustrated in Fig. 1.3, the operating
objective is to send the separated gas, oil, and water to the different destinations,
subject to the capacity (quantity) constraints. Purity requirements (quality control)
are also becoming increasingly common. The plant operation comprises multiple
processing units, including gas processing, condensate processing, and water treat-
ment. Similarly, there are many separation and fractionation columns in a refinery
or olefin plant, and each is a separate unit operation.
Traditionally, control textbooks and academic teachings provide very little guid-
ance on achieving a sound design for the overall control at the plant level. The study
of plant-wide control started three or four decades ago (Foss 1973; Stephanopou-
los 1983; Luyben, Tyréus, and Luyben 1998) and has received more attention lately
(Downs and Skogestad 2011; Juliani and Garcia 2017; Rangalah and Kariwala 2012).
However, the literature on plant-wide control is still scarce compared to the impor-
tance of this subject, and most of them focus on plant-wide steady-state optimization
(corresponding to the real-time optimization and planning layers in the automation
pyramid in Fig. 1.10) rather than dynamic control.
HAZOP (HAZard + OPerability) review is compulsory at multiple phases of new
or revamping process design projects. It is an ideal opportunity to catch flaws in
plant-level control design, including safety hazards and plant operability. However,
in reality, the HAZOP sessions typically devote most effort to hazard analysis and
spare very little time and attention to operability reviews. The primary reason is the
lack of process control expertise or awareness. Many practical challenges in operation
can be traced back to poor operability in design. Therefore, operability analysis at
the plant level deserves more attention than it currently receives.
The plant-level control design is based on process flow analysis, supply and
demand analysis, and degree of freedom (DOF) analysis in Chap. 6.
248 7 Methodology of Process Control Design

Process Topology and Process Flows

The supply and demand model is one of the plant-level concerns that can cross
the plant or unit boundary. A change in process flow in one process unit may alter
the supply and demand relationship and require changes in other units. Similarly, a
change in the piping layout in another plant may force a change in the process design
or control design in the current plant.
Plant-wide control requires a holistic view of the plant operation, comprising all
the interconnected unit operations. The job of plant-wide control design is often cus-
tomarily assigned to process engineers alone. However, as mentioned in Sect. 1.1.4,
a process engineer’s control knowledge and mindset are significantly different, flaws
are common in the overall control strategy designed by process engineers. Many of
them are even against basic process control principles. It is usually too late to correct
the flaws after the control solution is finalized or the plant is built; a more elaborated
control solution must be designed to accommodate or compensate for the controlla-
bility issues in the process design. Many times plant operations have to live with the
flaws until the next major turnaround. Therefore, process control should be involved
as early as possible in the process design to ensure the controllability requirements
are adequately addressed (see Sect. 1.1.3).

Plant-Wide Supply and Demand Control

The primary control consideration at the plant level is the dynamic balance between
supply and demand. Take the typical E&P production process in Fig. 1.3 as an exam-
ple. The process illustrates an entire surface production plant, from the production
wellhead all the way to the final sales point. The operating objective is to maximize
the oil production; therefore, the well production rate determines the overall through-
put (capacity). The separator separates the multi-phase fluid into gas, oil, and water,
and sends them to the respective processing units under pressure or level controls.
Following the guidelines on supply and demand analysis in Sect. 6.3, a supply-
driven control design is illustrated in Fig. 7.2. The gas processing unit compresses
the gas ➀ to the desired pressure, and sends the gas to downstream users ➁, following
a supply-driven propagation. Similarly, the oil ➂ and water ➃ processing is supply-
driven as well. The supply and demand model is maintained through the pressure–
flow or level–flow profiles.
Now assume that the consumer of the produced gas has a flow or pressure speci-
fication on the gas. A flow controller is installed to adjust the intake of the supplied
gas. This change effectively turns the supply-driven operation into demand-driven
and would cause significant changes in the process design and control design.
By requirement, at least one swing stream must exist along the gas flow path from
the separator to the sale point, to absorb the supply and demand fluctuations. The gas
stream to sales used to be the swing stream in the supply-driven operation model. It
is now dictated by downstream consumption and can no longer freely “swing.” The
supply-driven model has effectively changed to demand-driven. With the demand-
7.1 Process Hierarchy and Process Control Scope 249

driven operation, the demand change must propagate back to the separator and then
to the wells; the production rate must swing to adapt to the demand changes. The
re-designed overall control strategy is shown in Fig. 7.3, with stream ➁ being demand-
driven and stream ➀ a swing.
Swings in the well flow affect the overall oil production. If the oil production is
a target to be maximized, choking back on the oil production at the wells would not
be an option. Another gas consumer must be added to serve as the swing stream to
ensure the supply-driven operation. Otherwise, the imbalance in gas will have to be
continuously flared, which is against environmental regulations.
Similarly, a change in environmental regulation has banned the wastewater dis-
posal to shallow water wells. All the processed water has to be disposed of in deep-
water wells. Before completing the deepwater disposal (DWD) facility, the plant
operation is constrained by water disposal and becomes demand-driven. Temporary
changes to the overall control strategy are required to adapt to the reversal of the
supply and demand relationship. Once the DWD facility is complete, the supply and

Fig. 7.2 Overall control strategy for E&P process, supply-driven

Fig. 7.3 Overall control strategy for E&P process, demand-driven

250 7 Methodology of Process Control Design

demand analysis must be carried out again to assess if the control design needs to be
modified again.
The above discussion shows that the controllability at the plant level can signif-
icantly impact both the control design and the process design. The actual process
flow might be more complex than what is shown above.
Some general procedures are provided below for determining or selecting supply-
driven or demand-driven control strategies:
1. For each incoming and outgoing stream, determine if they are supply-driven or
demand-driven, or both. Whether a stream should be supply-driven or demand-
driven is primarily dictated by operating requirements.
2. Along the process flow path, identify/define some propagation nodes. These nodes
are typically the key process variables and key control points. The supply and
demand imbalance is “sensed” by these process variables and propagates forward
via control loops. Thus we can also call these variables the relay points, around
which the supply and demand analysis will be performed. For the compression
process in Example 6.1, we can have three nodes, one at the compressor suction,
one at the compressor discharge (injection pressure header), and the third at the
HRSG inlet.
3. Identify and validate the swing capacity. The fluctuations in gas pressure or liq-
uid level reflect the dynamic imbalance between supply and demand at the point
of pressure or flow measurement. The swing capacity at these points is typically
provided by gas line-packing or liquid surge volume for minor fluctuations. A sig-
nificant supply and demand imbalance would require dedicated swing stream(s).
The supply and demand analysis discussed in Sect 6.3.2 should be performed first.
As part of the control design, an additional operability analysis should be conducted
on the supply and demand propagation in both directions to ensure the adequacy of
the supply and demand model. Some guidelines are provided below.
For each supply-driven stream, perform a structured hypothetical what-if analy-
sis. This analysis is similar to what is done in a HAZOP session, where the standard
scenarios such as “more flow,” “less flow,” “reverse flow,” and “no flow” are con-
ducted for each stream. For Example 6.1, compressor suction is the first relay node.
The gas entering this node is the produced gas from wells:
1. Supply increase (“more flow”): If the incoming gas flow increases, more produced
gas will flow to the compressor suction header, causing the pressure to rise. To
keep the pressure constant, the pressure controller at the compressor suction would
increase the compressor speed to push through more gas.
2. Supply decrease (“less flow”): Similarly, if the well production rate decreases,
less gas will come to the compressor, causing the suction pressure to decrease.
The suction pressure controller would then slow down the compressor to reduce
the compressor capacity. If the minimum speed or minimum flow limit is reached,
then recycle valve is opened to protect the compressor.
3. Limited demand (“reversed supply and demand model”): If the compressor
throughput is limited and cannot take all the gas coming in, then the supply
7.1 Process Hierarchy and Process Control Scope 251

and demand model at this node will reverse to demand-riven. The capacity of the
compressor now sets the demand. The pressure at the suction will increase. The
excessive gas will have to be sent to flare. Otherwise, well production has to be
reduced.
At the compressor discharge, which is the second node of our analysis, the inlet is
supply-driven, while the outlet is a mix of supply-driven (injection gas) and demand-
driven (lift gas and HRSG gas). More flow from the compressor will cause the
pressure to go higher, and the discharge header pressure controller would then force
more gas to the injection wells. Less flow from the compressor will cause the pressure
header to decrease; the pressure controller will reduce the gas flow to the injection
wells. If the gas to injection is reduced to zero, and the pressure still goes down
further due to insufficient gas from the compressor side, the make-up gas stream will
open and fill the gap.
For each demand-driven stream, perform similar hypothetical what-if analyses
on all streams and check the capacity changes from the demand point propagating
backward all the way to the supply point. For example, the fuel gas to HRSG (the
third node in Example 6.1) is demand-driven. More demand will force the pressure
controller to open the make-up gas valve and cause the header pressure to decrease.
The pressure controller detects the pressure change and compensates it by reducing
the gas flow to injection wells.

Transition Between Different Mode of Operations

Changes in operation mode are common in all operations. For example, during plant
startup, the operation transitions from shutdown mode to normal operating mode.
This transition has impacts on all plant levels. For a typical operating plant, the
startup and shutdown is a complicated process with significant risks. Coordination
at the plant level is essential. Sequential control with sequence and logic, coupling
with regulatory control, is a practical yet challenging need.

7.1.3 Unit-Level Control

Unit-level control is typically focused more on quality control than capacity control.
Since all commercial operations are for-profit, quality control is the highest priority,
which can in turn affect or dictate the capacity control strategy at the plant level.
For example, in the fractionator example in Fig. 5.17, the top product draw has a
quality requirement on the composition. The quality controller will determine the
product flow and forces the control to be demand-driven. The feed is supply-driven;
therefore, the bottom product flow must be supply-driven to meet the dynamic supply
and demand balance. The bottom product has no specification, and thus a simple level
control is sufficient to meet the need for this supply-driven stream.
252 7 Methodology of Process Control Design

Stringent HSE regulations and energy optimization have led to leaner and
more integrated process designs, reflected by higher purity specifications, higher
yields/selectivity, more recycle streams, tight energy integration, and reduced swing-
ing capacities. Control of material and energy flow at one location will more likely
cause changes and disturbances to other locations due to interconnection and recy-
cles. They all make process control more challenging.
Unit-level control design relies on the results of the cause and effect analysis
(Sect. 6.4), DOF analysis (Sect. 6.5), and dynamic response analysis (Sect. 6.6). The
challenge is often the large number of possible control options, which can grow com-
binationally as the number of controlled targets and DOFs increase. Therefore, the
critical control concern at the unit level is the control strategy or the control structure.
It is vital to focus the attention on the unit-level issues and not be distracted by local
details. “Even though control engineering is well developed in terms of providing
optimal control algorithms, it is clear that most of the existing theories provide little
help when it comes to making decisions about control structure” (Skogestad 2000).

Example 7.1 Control of an Acetylene Hydrogenation Unit. Figure 7.4 shows an

acetylene hydrogenation process discussed in Al-Dawery and Dakhil (2012), a pro-
cess unit in a polyethylene plant. The feed to the polymerization unit is a mixture of
hydrocarbons from the olefin plant containing the undesired impurity of acetylene
(C2 H2 ), which should be controlled to below 2–3 ppm. Acetylene is a harmful con-
taminant in high-purity polymer-grade ethylene and must be removed. The typical
process to remove the acetylene is via selective hydrogenation over the catalyst in
multi-bed adiabatic reactors.
From a quick process flow analysis, it is clear that the unit operation is strictly
supply-driven. The supply is from the C2/C3 separator going straight through the
heat exchangers and reactors. The material balance should have been addressed in

Fig. 7.4 An acetylene hydrogenation process

7.1 Process Hierarchy and Process Control Scope 253

the plant-level control design. There is no capacity control requirement in this unit
operation, and thus, there is no need for any flow control elements along the process
flow.
On the other hand, it is also clear that the primary goal for the unit operation
is quality control, i.e., on the impurity level of the acetylene. The control strategy
manipulates the heat exchange at the heaters and coolers to achieve the desired
temperature and composition. Therefore, this unit control is strictly an energy balance
problem, and energy optimization can be a consideration in the control design.

7.1.4 Loop-Level Control

At the loop level, a critical decision is in choosing the appropriate control algorithm.
A control algorithm can be a simple standalone PID control (Chap. 3), a complex
control loop consisting of multiple PID controllers and supporting functional blocks
(Chap. 4), and can also be an advanced process control scheme like MPC (Chap. 5).
A simple standalone PID loop is sufficient for most applications, and the complex
PID control loops provide an economical solution for multivariable problems. Model
predictive control (MPC) provides a new technology that is more versatile than PID-
based control.
For the example in Fig. 7.4, the options for the control strategy can include the
following:
• Simple PID temperature controls after each heater/cooler.
• Cascaded temperature controllers from reactor outlet to heater/cooler outlet tem-
peratures.
• Model predictive control where one single controller is used to control the reactor
outlet composition, and perform energy optimization.
The process analysis and control design will be elaborated in Sect. 12.5.
The best control is always the fit-for-purpose control. It is crucial to keep the
control solution as simple as possible. Simpler solutions tend to be more reliable. A
simple solution is also easier to understand and thus more acceptable by the operators
and plant managers, at least conceptually. The control solution that the operators do
not accept will likely be turned off and eventually abandoned (also see the discussions
in Sect. 5.5.2).

7.2 Control Hierarchy and Control Strategy

The control design is divided into multiple layers, including continuously regulatory
control for normal operation, protective control for handling abnormal conditions,
sequential control for transition and mode changes, and instrumented safeguarding as
necessary (see Sect. 1.3.4). The purpose and function of each layer typically include
the following:
254 7 Methodology of Process Control Design

1. Normal operation: Maintain the operating point at the desired location or within
an acceptable region.
2. Abnormal condition: Protect the operating point from going outside the operating
or safety envelope during abnormal situations.
3. Mode transition: Ensure a safe and smooth transition during the change of oper-
ating modes.
In line with the above operating requirements, the process control function can be
classified into three main categories: normal regulatory control, abnormal protective
control, and sequential control. Each type of control may have close interaction and
integration with instrumented safeguarding (SIS):
⎧ 
⎪
⎪ Quality Control
⎪
⎪Normal Regulatory Control
⎪
⎪
⎪
⎪ 
Capacity Control
⎪
⎨ Protective Control
Process Control Abnormal Protective Control (7.1)
⎪
⎪ Overriding Control
⎪
⎪ 
⎪
⎪
⎪
⎪ Control Sequence
⎪
⎩Sequential Control Logic and interlock.

This layered control solution can be achieved by either baselayer PID control or
model predictive control (MPC). The targets for regulatory control loops are usually
the controlled variables in MPC. The targets for the protective control loops are
often included in MPC as constrained variables. Although the implementation differs
between PID-based control and model predictive control, the control objectives and
control strategy are the same.

7.2.1 Overall Control Strategy

The control objectives can typically be finalized after the process analysis. However,
the control strategy can have many options for the same control objectives, depending
on many practical factors. When designing new control solutions or troubleshooting
existing ones, it is essential to always have the “full picture” in mind. The high-level
control objectives and overall control strategy determine how the next level of details
should be laid out. For this reason, a top-down approach for process control design
is recommended following the different phases of project execution (Rangalah and
Kariwala 2012).
The control strategy is dictated by the control objectives and is based on a thor-
ough process analysis discussed in Chap. 6. There is no “one size fits all” solution for
process control. PID control is very effective, economical, and robust but has limita-
tions for more complex process dynamics. Model predictive control (MPC) has been
very successful for complex process control problems but has a higher requirement
7.2 Control Hierarchy and Control Strategy 255

on expertise and resources. There are always compromises to be made in process

control design.

7.2.2 Normal Regulatory Control

Normal Regulatory Control is responsible for the continuous adjustment of the

manipulated variables to keep the controlled variables at the desired setpoint val-
ues. The normal regulatory control is also the control we commonly refer to.
Normal regulatory control consists of quality control and capacity control (quan-
tity control). Quality control is prevalent in downstream operations, while capacity
control is more commonplace in upstream E&P operations. As the HSE require-
ments become more stringent, quality control in E&P is also becoming a common
requirement.

Quality Control

All production processes mean to produce something useful, either as final finished
products for sale or as semi-product to send to other units or plants for further
processing. They all have certain specifications or quality standards to meet, typically
represented by product quality such as composition, impurity level, dew points, and
pH value. Any stream flowing out of the system boundary is considered a product,
and the requirement on pressure, temperature, or flow rate are also considered product
specifications. The acetylene concentration in Fig. 7.4 is a typical example of quality
control.
Although already a common practice in downstream operations, moving from
process variable control (temperature, pressure, level, and flow) to product quality
control is still a value proposition in many upstream operations. Significant chal-
lenges for quality control include the following:
1. Availability: Many product qualities are not easily measurable online, while others
depend solely on lab samples. Most of them are not available at the frequency
required by online control.
2. Credibility: The accuracy of most online measurements is not on par with the
conventional process measurements such as flow and pressure. Common problems
such as analyzer freezing (“flat-lining”), spikes, drifting render it not viable for
online control.
3. Reliability: Online quality measurement devices are prone to malfunctions and
require regular calibrations and maintenance. They typically have a much lower
uptime than conventional devices.
The slow and unreliable online quality measurements can be enhanced with infer-
ential property estimation (“soft sensors”). The product quality can be “inferred”
from the fast and reliable conventional measurements based on the cause and effect
256 7 Methodology of Process Control Design

relationship and dynamic response characteristics. The inferred values are regularly
corrected by feedback from online analyzer or lab results to improve accuracy. See
Sect. 5.3 for inferential property estimation (IPE).

Capacity Control

Capacity control is a vital aspect of the process control design in the general plant-
wide control, especially in E&P operations. By definition, a process is the contin-
uous or semi-continuous (batch) processing of certain materials. Maintaining the
inlet/outlet material and energy balancing is an essential requirement. Capacity con-
trol is the continuous regulatory control of the plant/unit throughput, ensuring the
dynamic material and energy balance. Capacity control design follows the supply
and demand analysis in Sect. 6.3, and maintains the desired supply-driven operation,
demand-driven operation, or mixed-operation.
For gas processing, the imbalance in supply and demand is reflected in the gas
inventory in the process and can be sensed by the gas pressure. The capacity control
is thus focused on maintaining the desired pressure–flow profile. For liquid handling,
capacity control is concerned with maintaining the proper level–flow profile along
the process path, as the liquid levels in the vessels reflect the liquid inventory. For
example, the suction pressure controllers in Fig. 7.2 are responsible for capacity
control for the supply-driven operation. If the operation is demand-driven, as in
Fig. 7.3, the discharge pressure controllers will take care of capacity control.
For energy balance, maintaining the proper temperature–flow profile is the objec-
tive of capacity control. However, since the energy balance is typically managed by
heat exchange, and the heating or cooling media are typically part of the utility flow
that is not directly coupled with the process flow, the plant-wide energy balance is
not a significant concern. The typical challenge with energy balance is the saturation
in heating or cooling supply, especially if associated with exothermic processes.
For the compression process in Example 6.1, the cause and effect table (CET) is
updated and shown in Table 7.1, where ⊕ indicates a positive process gain and  a
negative gain.

7.2.3 Protective Control

The optimal operating points are often located at or near the boundaries of the oper-
ating envelope, which results in more profitable production but also increases the
risk of going out of control and trip the unit to shut down. A protective control layer
is often included in the control design to avoid unnecessarily tripping the operation.
This protective layer stays dormant during normal operation but becomes active if
the operation approaches too close to the boundary of the operating envelope. The
abnormal protective control layer is typically tuned much more aggressive than the
normal regulatory controllers for fast action.
7.2 Control Hierarchy and Control Strategy 257

Table 7.1 The PGC example: cause and effect table, with process gain
CV-01 CV-02 CV-03 CV-04
Suction press Discharge press Liftgas flow Fuel gas press
MV-01 Valve, flare ×
MV-02 Speed  × ×
MV-03 Valve, recycle × × ×
MV-04 Valve, make-up × ×
MV-05 Valve, injection 
MV-06 Valve, gas lift ⊕
MV-07 Valve, HRSG × × ⊕

As discussed in Sect. 1.3.5 on the layered approach for control and safeguarding,
a protective control layer should be considered, if feasible, to avoid unnecessary
trips whenever there is a safeguarding trip limit for a particular control point. The
protective control can be a standalone PID controller or an overriding PID controller
(see Sect. 4.1.3 for detailed discussions on protective control).
Now let us continue from where we left in Sect. 6.5.2 and finish the MV-CV
pairing for the compressor control example (Example 6.1). From the cause and effect
relationships in Table 6.2, the MV-CV pairs marked with ⊗ are all normal regulatory
control loops for capacity control. The MV-01 (flare valve) is evidently for protective
control of the suction pressure CV-01 during normal operation, which is already
paired with MV-02. We now have a control problem with one controlled variable
and more than two manipulated variables. From Sect. 4.1.6, the extra DOF can be
consumed by split-range control or dual-pressure control design. A dual-pressure
control scheme is ideal because flaring pressure and normal operating pressure are
different by design. As a result, a second pressure controller from the same pressure
measurement CV-01 is created as a protective controller for the flaring. Let us name
it CV-01B for now.
The same analysis applies to the make-up gas valve (MV-04). The make-up gas
valve is for protective control. The make-up intake is only expected when the dis-
charge pressure is below the normal operating pressure and reaches the pre-specified
setpoint. Again, a dual-pressure control loop is selected over split-range control. A
protective pressure controller CV-03B is created to pair up with the make-up gas
valve MV-04.
The recycle valve is designed explicitly for anti-surge control (see Sect. 11.3 for
details on anti-surge control), so another controlled variable, CT-01C, is created to
pair up with MV-03, the compressor anti-surge control valve. The anti-surge control
is a standalone protective controller.
Now that all the DOF are exhausted, let us assign the permanent tag names to the
controlled variables and manipulated values following the naming conventions, as
shown in the updated CET (Table 7.2).
258 7 Methodology of Process Control Design

Table 7.2 PGC example: cause and effect table, final

PC-101 PV-102 UC-103 PV-104 PV-105 FC-106 PC-107
Flare gas Suction P. Surge Discharge Injection gas Lift gas HRSG gas
PCV-101 Flare valve ⊗
SCV-102 Speed ⊗ × ×
UCV-103 Recycle × ⊗ ×
PCV-104 Make-up valve × ⊗
PCV-105 Injection valve ⊗
FCV-106 Lift gas valve ⊗
PCV-107 HRSG × × ⊗

7.2.4 Sequential Control

Sequential control such as automated startup, shutdown, or transition is an integral

part of process control and is a vast topic by itself. We will provide some brief dis-
cussions on this topic, emphasizing its interaction and integration with regulatory
control. Sections 12.3 and 12.4 provide two unit-control examples where regulatory
control and sequential control are integrated to meet the unique operating require-
ments.
Another type of sequential control is the transition between the different operating
modes. For example, a feed change in an ethylene plant may require many adjust-
ments to the operating points and operating envelope. The safe and smooth transition
between the two operating conditions is crucial.
Crippled mode operation is another common scenario. A piece of equipment
or device may malfunction, forcing the operation to shut down partially and enter
the so-called crippled mode operation. Critical considerations include the smooth
transition to the crippled mode, the safe operating in the crippled mode, and the
quick transition back to the normal operating mode. For example, with a complex
integrated compressor control scheme (Sect. 11.3) consisting of multiple compressor
trains, a compressor automatically opens the recycle flow valve to stay in operation
when another compressor in parallel trips.
Conventionally, dedicated control sequence or logic handles the sequential control
scenarios. However, it is preferable if a control solution can gracefully handle the
transition.

7.2.5 Instrumented Safeguarding

Instrumented safeguarding interacts with process control in two prominent areas:

1. The proper configuration of setpoints and limits: There is typically a trip limit for
every key process variable (KPV) to prevent the process from going out of the
7.2 Control Hierarchy and Control Strategy 259

operating envelope or safety envelope in case of abnormal operating conditions.

The normal regulatory control loops maintain the operating point during normal
conditions. Protective control loops are implemented to keep the operating points
inside the operating envelope in case of disturbance. The normal control setpoint,
the protective control setpoint, the alarm limit, and the trip limit must be appro-
priately validated and rationalized for stable operation. For example, a setpoint
for protective control close to the trip limit does not offer much protection.
The rationalization of the setpoints and limits is a joint effort between multiple
teams, including process engineering, operations, rotating equipment, and process
control, based on the specific operating conditions and requirements.
2. Control logic embedded in the control loops that interact with safeguarding:
Although back-end logic is strongly discouraged for the sake of clarity in design
and ease of maintenance, behind-the-scene control logic does exist as the “neces-
sary evil” and can not be avoided entirely. They are a part of the control scheme
and need to be carefully considered during process control design and adequately
documented.
The level of sophistication in automation, control, and safeguarding is dictated by
the nature of operating units and the associated economic benefits. The safest plant
would be a plant in stopped condition; however, such a plant does not make money.
Implementing lots of safety features may cause the plant to trip too frequently with-
out a particular reason (nuisance trips), which leads to unnecessary production loss
or deferment. So reliable safeguarding (more trips) and high availability (fewer nui-
sance trips) are often conflicting goals. A careful compromise is needed to reach the
“optimal” balance.

7.3 Process Control Functional Design

The process hierarchy and the control hierarchy present the “big picture” of process
control design. This representation is illustrated in Table 7.3, where “x” indicates
the prevalence of the type of control. For example, capacity control is typically
considered at the plat level, while sequential control is often a requirement at the unit
level.

Table 7.3 Process control hierarchy

Process control requirements
Quality Capacity Override Protect Sequence Logic and interlock
Plant-level control x x x
Unit-level control x x x x
Loop-level control x
260 7 Methodology of Process Control Design

7.3.1 Variable Pairing and Degree of Freedom Validation

A vital step in control design is determining the control structure, which involves
variable pairing and DOF analysis. The DOF analysis is typically an explicit step
with PID-based complex control. For model predictive control, the DOF analysis is
often built into the software tools. Although the control structure may seem entirely
different between complex PID control and model predictive control, the analytic
skills and results are the same.
As discussed in Sect. 4.1.2, if there are more control targets than control handles,
the typical approach is to reduce the number of control targets through selective
control such as overriding control, ratio control, or cascade control. On the other
hand, if there are fewer targets than handles, extra DOFs are not consumed. The
solution is typically to reduce the number of control handles by making them non-
independent via split-range, fan-out, and switching techniques. The extra DOFs can
also be consumed by increasing the number of targets by making the same variable
as the control target of a new controller (but with a different setpoint).
Now let us come back to the compressor example in Example 6.1 and validate
the adequacy of the manipulated variables against the supply and demand model and
check on the DOF requirements:
1. Flaring: Since gas flaring is supply-driven, a pressure controller upstream of the
control valve is appropriate. Pressure measurement PT-101 will become pressure
controller PC-101.
2. Compressor capacity: Overall production is supply-driven; therefore, the com-
pressor is also supply-driven. The suction pressure is an indicator of supply
change, and thus a suction pressure controller PC-102 controls the compres-
sor capacity by manipulating the compressor speed.
High load on the compressor may cause the discharge pressure PT-103 to go
excessively high and over the safety limit. A pressure controller PC-103 based
on discharge pressure measurement PT-103 is added to override the capacity
controller PC-102 in case of high discharge pressure.
3. Recycle: When compressor flow (capacity) goes below a low limit, the compressor
may enter into an unsafe condition of compressor surge. A minimum flow through
the compressor is maintained by recirculating part of the gas from the discharge
side back to the suction to prevent the surge from happening. Therefore, the recycle
valve is reserved for anti-surge control (more on anti-surge control in Sect. 11.3).
4. Gas supply to HRSG: The gas supply to HRSG is demand-driven; thus, a pressure
controller PC-106 downstream the valve meets the requirement.
5. Gas supply to gas lift wells: The gas lift determines the gas demand; thus, it is
demand-driven. A flow controller upstream of the valve fulfills the demand.
6. For the gas header pressure PT-104, the gas flow to injection provides the
swing capacity, clearly a supply-driven operation. A pressure controller PC-104
upstream of the valve is appropriate.
7. Make-up gas flow is demand-driven. Based on the rules, a pressure controller
downstream of the valve is an obvious choice. Since the header pressure PT-104
7.3 Process Control Functional Design 261

has been used to manipulate the valve to injection wells, another pressure con-
troller can be spawned from the same measurement, but with a higher setpoint,
to adjust the makeup gas in-take.
The final challenge with the control design is the interaction between multiple
MVs and CVs. The cause and effect relationship is solid for all the controllers except
the suction pressure and surge, and the dynamics are very fast, so the interactions
are not a serious concern. However, the compressor speed and recycle valve both
affect the suction pressure. From the compressor dynamics analysis, the speed alone
also limits the controllability range of the capacity control, so a split-range control
is used for the suction pressure control.
The final control structure is shown in Table 7.4, considering the supply and
demand model, the cause and effect relationship, and the dynamic response charac-
teristic of each causal relationship.
The overall control strategy is presented in a control overview diagram shown in
Fig. 7.5.

Table 7.4 PGC example: cause and effect table, updated

PC-101 PV-102 UC-103 PV-104 PV-105 FC-106 PC-107
Flare gas Suction Surge Discharge Injection gas Lift gas HRSG
PCV-101 Flare valve ⊗
SCV-102 Speed ⊗ × ×
UCV-103 Recycle ⊗ ⊗ ×
PCV-104 Make-up × ⊗
PCV-105 Injection ⊗
FCV-106 Lift gas ⊗
PCV-107 HRSG × × ⊗

Fig. 7.5 A gas compression process: control strategy

262 7 Methodology of Process Control Design

7.3.2 Control Algorithms

With the widespread adoption of model predictive control, a critical decision in con-
trol strategy selection is the control algorithm, typically between PID-based baselayer
control and MPC-based advanced process control. PID control is based on existing
infrastructure and expertise and is thus cost-effective. The downside is that PID-
based solutions have limited capability in handling very complex control problems
(Sect. 5.1.1). Model predictive control is significantly more versatile and power-
ful but has a higher requirement on expertise and resources, especially for the first
application.
As already mentioned, even for the simple example in Fig. 7.4, there can have
several different control strategies, including simple PID control loops, complex
PID control schemes, and full-blown MPC solutions. In addition, there can be many
ways to pair up the variables. For example, Chin (1979) shows 21 ways to control
the pressure of a fractionator; each has advantages and disadvantages.
Riding with the vast success of MPC, there is a tendency of over-use or abuse in
MPC applications, mainly in two aspects:
1. Treating MPC as a panacea for all control needs. Many times a baselayer control
scheme can do equally well or even better than an MPC solution.
2. Blindly applying the MPC technology without sufficient process insight. For many
MPC applications driven by tight project schedules, the proper process analysis
discussed in Chap. 6 is often bypassed or short-cut.
Careful considerations of all the factors need to be taken when deciding between
baselayer control and advanced control. Once the decision is made, the discussions
and guidelines in Chap. 4 can be followed for complex PID control loops and Chap. 5
for advanced process control.

7.4 Process Control Visualization and Documentation

Process control is a cross-disciplinary engineering specialty. The control solution

is a collaborative result of multiple teams, including process engineering, rotating
equipment, operations, control systems, instrumentation and measurements, project
engineering, process safety and integrity. The control solution is expected to control
the plant 24/7 throughout the plant’s life cycle, operated and maintained by multiple
teams. Therefore, it is crucial to document the control design simply and concisely
to be easily understood by all the concerned parties of diverse backgrounds.
Based on the author’s experience with countless projects, from minor trou-
bleshooting tasks to multi-billion dollar mega-projects, one of the most challenging
tasks, which is also a task that most people have no interest in doing, is the process
control documentation. When performing process control maintenance and trou-
bleshooting, the most needed document is the process control document. However,
7.4 Process Control Visualization and Documentation 263

most of the time, it is challenging to find a decent control document. The control
document most likely lies in the brain of the original designer, who has left the com-
pany a long time ago.1 Even if there is a control document, it typically does not have
the expected quality to be very useful.
The process control narrative (PCN) document is a critical delivery from process
control. The PCN document is authoritative on the process control design, imple-
mentation, operation, and maintenance. It is also an essential communication tool
among all the teams regarding the control solution. Therefore, the quality of the PCN
document has a profound impact on the plant operation.
However, the development of a high-quality PCN is a highly challenging task.
It requires a holistic view and intimate knowledge of the process, operation, and
control to cover all the critical information. It requires an even higher skill to present
the information in a simple and concise fashion.
This section provides some guidelines on standardizing and simplifying this chal-
lenging task of process control documentation. A control overview diagram (COD)
is recommended to facilitate the understanding of the control solution. A standard
document structure is provided to improve the consistency and completeness of the
information presented.

7.4.1 Process Control Narrative Document

The PCN is the authoritative document on the design and implementation of the con-
trol solution. It typically consists of the process description, overall control strategy,
narratives of complex control loops, implementation considerations, and operation
recommendations. The following contents are recommended to be included in a PCN
document.
1. Process description: It is vital to include a concise process description to provide
the background for the process control design. The guidelines on developing the
process overview diagram (POD) in Sect. 6.7 can be followed as the starting point
for COD:
a. The design intent and operating requirements: A brief description of the oper-
ating objectives and requirements is essential in providing the background of
the process operation.
b. The process streams and specifications: The major process flow streams com-
ing into and going out of the system boundary, with the critical specifications
for each stream.
c. The design limits and operating constraints: The limits and constraints can
have a significant impact on the control design.

1In the USA, in the industries that are subject to FDA (food and drug administration) review, it is
commonly said that “if you didn’t document it, you didn’t do it”.
264 7 Methodology of Process Control Design

2. Plant-level control design (if applicable):

a. The plant control objectives.
b. The narratives on the overall control strategy at the plant level.
c. The control overview diagram at plant level.
3. Unit-level control design (if applicable):
a. Unit control objectives.
b. Narratives of the overall control strategy at the unit level.
c. Control overview diagram (for complex units).
d. Control narratives are required for any complex loop whose intent or design
cannot be unambiguously understood from the control diagrams such as COD
or P&ID.
A common mistake in writing control narratives is elaborating on the simple
control loops but neglecting the more complex control schemes. For example, it is
very often to see detailed descriptions of how each simple flow feedback controller
works but lacks any description of the override control scheme comprised of these
very flow controllers.
4. Implementation and Configuration:
a. Implementation aspects: Many implementation considerations require a good
understanding of the process design, operation requirements, control strategy,
and the capability of the control systems. Process control is arguably the only
specialty in the operating plant that has a holistic view of all these aspects.
b. Controller configuration: The controller implementation is the task of con-
trol systems engineers/technicians. However, many configuration details such
as anti-reset windup and bumpless transfer are best specified by the process
control engineers (Sect. 3.5). MPC configuration is the sole responsibility of
process control engineers. Including all the configuration details in PCN is
unnecessary, but some critical or unique configurations must be adequately
documented.
5. Special Considerations: Many practical scenarios must be taken into account.
a. What if something goes wrong? The control design challenge is not how to
make the control solution work under normal conditions but how the control
system should respond to the different operating scenarios including abnormal
situations. Special considerations must be taken to prevent or mitigate the
negative impact of these abnormal scenarios.
b. Special features.
The PCN document is progressively developed through the different phases of the
project. The expected content from different phases of a project is shown in Table 7.5.
The PCN can be elaborated or succinct, depending on the actual requirement and
the preparer’s preference. However, the following three components are strongly
recommended to be part of the PCN document:
7.4 Process Control Visualization and Documentation 265

Table 7.5 The phased process control deliverables

Project phase Process control narratives Check point
Conceptual design Process flow analysis Design review
+ Overall control strategy
+ Control overview diagrams
Front-end engineering + Key process control loops Design review
(FEED) + Process control schematics HAZOP review
Detailed engineering + Detailed function specifications Design review
+ Loop configuration details HAZOP review
Implementation + DCS control modules/sequences Factory acceptance
+ Logic flow diagrams
Commissioning + “As-Built” updates Site acceptance
Maintenance/improvement + Management of change (MOC)

1. Control overview diagram (COD). A one-page overview to provide a holistic view

of the overall control strategy.
2. Narratives of overall control strategy and complex control loops.
3. Key configuration parameters of the controllers.

7.4.2 Control Overview Diagram

A modern operating plant/facility design can often produce dozens or even hundreds
of pages of process flow diagrams (PFD) and piping and instrumentation diagrams
(P&ID), along with a comparable number of DCS schematics. An inexperienced
engineer can easily be overwhelmed by the daunting number of schematics and fails
to see the forest for the trees, thus loses the “bigger picture” (the overall control
strategy). A flawed process control design is often caused by the lack of a sound
overall control strategy. It is not unusual that all the individual control loops perform
well in silos but fail to work when put together.
A control overview diagram (COD) is one way to force the design engineers
to think about the bigger picture and thus make sound decisions. The COD is a
one-page “visual summary” of the process flow and control strategy, evolved from
the process overview diagram (POD) discussed in Chap. 6. The COD captures the
process analysis results in Chap. 6 and control design in this chapter and aims to
provide a simplified yet concise high-level summary of the overall process design
intent and essential operating requirements, focusing on the dynamic material/energy
balance. The COD focuses on the overall control strategy, emphasizing the integrity
of the control scheme and interaction among different parts of the operation.
266 7 Methodology of Process Control Design

The COD should include the following essential contents:

1. The critical process flow streams with stream specifications, including all the
process streams, equipment, and devices that may affect the dynamic material
and energy balance.
2. Key process variables (KPVs): The key process variables are a subset of process
variables whose values and statuses provide a sufficiently complete and reliable
indication of the overall control status and operating health when viewed together.
The KPVs can include, for example, the quality control variables, capacity control
variables, and critical protective control variables.
3. Key process control loops: The key control loops are the control loops that rep-
resent the overall control strategy at the plant level and unit level. They reflect
the supply and demand model and cause and effect relationships among the key
process variables.
The COD should include only the minimum and essential details for an uncluttered
view of the process flow and control strategy (“the big picture”). Any details that do
not contribute to improving the presentation clarity of the control scheme should be
excluded. The details of process engineering and control systems should refer to the
P&IDs or individual control schematics in the PCN document. Figures 7.2 and 7.5 in
this chapter are examples of COD. All the application examples in Chap. 12 include
a COD.
Due to the cross-disciplinary nature, process control engineers need to collaborate
closely with several other teams during control design and maintenance. The COD
serves as a very effective visual communication tool when working with process
engineers, control systems engineers, and instrumentation engineers, operations per-
sonnel, and operation manager, especially for complex control strategies. From the
authors’ experience with hundreds of applications, from small brownfield projects to
multi-billion dollar greenfield projects, the COD has been an indispensable tool for
visualization, communication, and documentation of the process control strategy.

7.4.3 Control Loop Narratives

Simply put, the PCN provides the answers for the essential questions on all the
control schemes or control loops whose intent or function is not immediately clear
from the control schematics such as PFD or P&ID. The questions and purposes are
listed in Table 7.6.
The same questions apply to a controller, a control loop, or a control scheme, and
should be asked at different plant levels. For example,
1. At the plant level, the questions are typically about the overall control strategy.
2. At the unit level, the questions are often on the control design, either a complex
baselayer scheme or a model predictive control.
7.4 Process Control Visualization and Documentation 267

Table 7.6 Essentials elements of control narratives

Intent Sample narratives
WHERE? Where is the control scheme or loop located/placed?
The placement of the measurements and valves are based on process design, and
is a result of supply and demand analysis, cause and effect analysis, and dynamic
response analysis
WHY? Why is the control scheme/loop needed?
This is the process reason for justifying the needs for this particular scheme/loop,
such as maintaining product quality, enforcing material balance, addressing process
non-linearity
WHAT? What does the control scheme/loop provide?
This is the control strategy and functional specification for the control scheme, based
on the WHY and WHERE analysis. The control strategy addresses the design intent
and operating requirements, while the functional specification is the implementation
details that can be implementable by control system engineers
HOW? How is the control scheme/loop implemented?
WHEN? When is the control expected to be in operation, and when should it be taken offline?
WHAT IF? What if things go wrong?
In case of an abnormal operating condition such as measurement failures, emergency
unit shutdowns

3. At the control loop level, the questions are more about implementation and con-
figuration, whether a PID loop, an on/off control loop, model predictive control,
or a logic sequence.
All control loops, sequences, or logic whose purpose or function are not immedi-
ately evident from the COD or P&ID require a narrative in the PCN document. For
example, for the compressor anti-surge control loop in Fig. 4.2, a minimum control
narrative can be as brief as shown in Table 7.7.
The control narratives are expected to be short and to the point. Engineers in the
field do not have the time and patience to flip through a 500 page PCN document to
find the information they urgently need. It is much more effective to document the
control design in drawing such as a COD or bullet-point list of the critical information
listed above. For example, the complex capacity control scheme for the entire gas
compression process unit in Fig. 4.2 can still be as brief as shown in Table 7.8, as
long as the main points are captured.

7.4.4 Loop Configuration Details

A PID controller has many configuration parameters, all having system default val-
ues. However, the default values are rarely adequate for the intended application and
must be reviewed and assigned with appropriate values per control design.
Usually, the PID configuration is the scope of work of control system engineers or
technicians. However, a subset of the configuration parameters are closely related to
the operating requirements and overall control strategy, typically beyond the knowl-
268 7 Methodology of Process Control Design

Table 7.7 Sample control narratives for compressor anti-surge control

Intent Sample narratives
WHERE? An anti-surge controller (UC-111) shall be implemented at each stage of the
compressor …
WHY? …to prevent the compressor from going into surge
WHAT? It calculates a surge parameter with UY-111 based on suction pressure
PT-111, discharge pressure PT-112, and discharge flow FT-111, and use
the anti-surge control valves ASCV-111 to regulate the recycle flow and keep
the compressor flow above the minimum flow requirement
HOW? The control loop is a PID controller with I-PD setting and in DIRECT action,
and runs at 100 ms …
WHO? … and shall be operated in AUTO mode during all operation operation modes
except for crippled mode. The controller mode and setpoint are locked from the
operators
WHAT IF? In the case of BAD PV, the output shall be ramped up to the fail-safe position
of the valve; during ESD/PSD, the control valve shall open to the fully open
position in one shot

Table 7.8 Sample control narratives for compressor capacity control

Intent Sample narratives
WHERE? A pressure controller PC-002 is implemented at the compressor suction header
…
WHY? …as the capacity control loop to enforce the supply-driven operation
WHAT? It manipulates the compressor speed SC-101 and recycle valve ASCV in split-
range to maintain the pressure setpoint PC-002
HOW? The pressure controller manipulates the compressor speed via a cascade control
through the load control XC-101 and is the first leg of the split-range control.
If the compressor speed is insufficient to achieve the desired capacity turndown,
the recycle valve ASCV-111 and ASCV-121 are used as the second leg in the
split-range control setting
WHO? The capacity control should normally operate in AUTO. Mode change is part
of the compressor startup logic. Operators are allowed to change control mode
after normal startup
WHAT IF? The capacity control PC-002 can be overridden by the discharge header pres-
sure protective controller PC-003 via a low selector. The capacity control
output PC-002.OP to the speed control can be overridden by motor power
protective controller EC-101 via a low selector. The pressure control output to
the recycle valve can be overridden by the anti-surge controller UC-111 and the
discharge pressure controller PC-112 via a high selector; both are protective
controllers

edge of control system engineers, and must be decided by process control engineers.
For this reason, it is recommended to include a table of the loop configuration details
in the PCN document to guide the control systems engineer on the configuration of
those critical configuration parameters. This table contains the typical and critical
configuration details for the PID controllers, either specified by the process control
engineers or extracted from other master sources.
A sample table is provided in Table 7.9 and the guidelines can be found in Sect. 3.5.
Some additional remarks on the PID loop configurations are provided here:
7.4 Process Control Visualization and Documentation 269

Table 7.9 Control loop configuration (sample)

Properties Example
Loop name on P&ID 15FICA-111 …
Loop name in DCS 15FC111
Loop description K1510 stage 1 suction flow control
SP/PV engineering unit M3/HR
OP engineering unit %
Loop type (regulatory or protective) Regulatory
PV measurement range high 500
PV trip limit high (if applicable) –
PV alarm limit high (if applicable) 450
SP range limit high (if applicable) 300
Typical PV value (e.g., nominal value) 250
SP range limit low (if applicable) 100
PV alarm limit low (if applicable) –
PV trip limit low (if applicable) –
PV measurement range low 0
OP range limit high 100
OP range limit low 0
Algorithm (P, PI, PID, PID, I-PD) PI-D
Direction of action (direct/reverse) REVERSE
SP tracking PV when in MAN YES
Anti-reset windup (Y/N) YES
Explicit bumpless transfer (Y/N) NO
Initial PID tuning: gain 3.0
Initial PID tuning: integral (minute) 0.5
Initial PID tuning: derivative (minute) 0.0
Has custom calculations/logic? (Y/N) NO
Operator access (MODE/SP/OP) YES/YES/YES
Mode during normal operation AUTO
Mode during startup MAN
Mode during shutdown/OP = 0% MAN
Mode when entering crippled operation MAN/OP = Last good
Mode when resuming from crippled mode AUTO, by Operator
270 7 Methodology of Process Control Design

1. Tag in DCS often has a shorter name than that on P&ID (see Sect. 1.3.3).
2. For the process values (PV), the typical value, alarm value, trip value, and mea-
surement range, along with the setpoint range limits and output range limits,
are collected and shown in the same table to facilitate the rationalization of the
setpoints and limits (see Fig. 1.14).
3. Anti-reset windup: Anti-reset windup is required if the controller output may be
limited or disconnected (see Sect. 3.5). The standard built-in anti-reset windup
configuration should be used whenever possible. If the built-in anti-reset function
is not sufficient, an external anti-reset windup is required, detailed narratives must
be provided in the main context of the document (see Sect. 3.5).
4. Explicit bumpless transfer support: In most modern control systems, the bumpless
transfer function has been a standard feature for most control blocks. However, if
the standard built-in functions do not support the bumpless transfer requirement,
an explicit bumpless transfer block must be added and configured (see Sect. 3.5).
5. Initial PID settings: The system default for the PID settings is typically not ade-
quate for most of the PID loops. Initial PID tuning should be provided in the table
to avoid defaulting to the system values.
6. If custom calculations, sequences, or logic are present, indicate YES in the table
and provide narratives in the relevant section of the PCN document. A common
source of frustration is the lack of documentation on back-end logic in the con-
troller.
7. Operator access: Many PID controllers may need to be locked from the operator
to avoid mal-operation. The lock-up can include MODE, SP, or OP. See Sect. 3.6
for more information.
8. Mode of operation: The controller MODE, SP (in AUTO), or OP (in MAN) shall
be specified for the following operation modes if the default behavior is not
acceptable or the standard practice is not adequate:
• Normal operation.
• Startup.
• Shutdown.
• Entering crippled mode.
• Recovering from crippled mode.
• Any other major operation modes.

7.5 Building Control Overview Diagram—The Work

Process

Visualization and documentation of the process control solution are essential require-
ments but are often neglected. One reason is that good visualization and documen-
tation require that the developer have both a thorough understanding of the overall
control strategy and also intimate knowledge of the details.
The control overview diagram (COD) is one recommendation for simplifying the
job. A COD should typically be a one-page diagram with a top-down layout. The
7.5 Building Control Overview Diagram—The Work Process 271

COD must provide the top-level holistic view of the entire facility within the process
control scope. If there is room for more details, the responsible engineer can decide
what additional details to include at his/her discretion, provided that the added details
will not inadvertently affect the clarity of the overview.
A sample procedure for COD development is recommended as below. Typically,
this procedure needs to be repeated multiple times before reaching a satisfactory
outcome.

7.5.1 Collect Relevant Information

The following information or documents are essential for the process analysis and
control design. Lack of reliable documentation is a major headache to process control
engineers, so it is beneficial to collect as much as possible for completeness and
cross-validation of the information:
1. Existing PCN document, if available.
2. Process flow diagrams (PFD) (with the material balance sheets).
3. Piping & instrumentation diagrams (P&ID).
4. Operating philosophy document and operating manual.
5. DCS schematics.

7.5.2 Build Process Overview Diagram

Follow the process analysis process and build the process overview diagram (POD)
as the backdrop of the COD:
1. Define the project or process boundary.
2. Identify and add all the major streams flowing into the system boundary.
3. Identify and add all the major streams flowing out of the system boundary.
4. Identify and add the piping and equipment essential for showing the process flow
and field instrumentation.
5. Identify and add all the final control elements such as valves and motors that can
affect the product quality or production capacity.
6. Identify and add all the control elements such as switches and on/off valves that
can affect the process flow line-up during normal operation.

7.5.3 Add Regulatory Control Loops

1. Identify and add the essential quality control and capacity control loops to the
diagram.
272 7 Methodology of Process Control Design

2. Perform a cause and effect analysis between all final control elements and key
process variables, and verify that the control PV and OP selection are appropriate.
3. Perform a dynamic response analysis of the control loop. Assuming a step-change
in the controller output, through the final control elements and the process itself,
assess the dynamic response of the controller PV of the same controller in terms of
gain, lag, and delay. Note that transient response characteristics from the controller
OP to the controller PV give the loop dynamics.
4. Perform a DOF analysis to verify that there is no surplus or deficit in the DOF for
the current selection of manipulated variables and controlled variables, e.g., mul-
tiple controllers (implicitly) controlling the same variable or one valve attempting
to control multiple variables.

7.5.4 Add Overriding and Protective Control Loops

1. Identify and add overriding protection control to complement the capacity con-
trol loops in case the supply and demand relationship may change, such as the
following:
a. Supply-driven operation may temporarily become demand-driven due to pro-
cess upset or operator intervention.
b. Demand-driven operation may permanently become supply-driven over time
due to depletion of supply or process modifications.
c. Operation may be unexpectedly limited by critical equipment (e.g., compressor
becomes the bottleneck), and the capacity becomes equipment-driven.
2. Identify and add any critical protective control loops as needed. Generally, for
any key process variables with safeguarding trip configuration, it is essential to
evaluate the necessity and feasibility of adding a protective control layer between
normal regulator control and safeguarding trip.
3. For each control loop, determine if it requires special configurations such as anti-
reset windup and bumpless transfer. Start from the controller output and trace the
control signal to the final control elements, reset windup protection is needed if the
signal could be disconnected or limited anywhere along the path. If the change due
to disconnection or re-connection of the signal can cause an undesirable bump
in the control components, then an explicit bumpless transfer block should be
considered.
4. Perform a rationalization check on the operating ranges for each key control
variable, including normal regulatory control setpoint, protective control setpoint,
alarm limit, trip limit, and measurement range.
7.5 Building Control Overview Diagram—The Work Process 273

7.5.5 Add Special Configuration Requirements

1. Add a “#” mark to the controller if the controller has any of the following special
requirements:
a. Supporting calculations (other than standard built-in functions such as filtering,
gap control).
b. Back-end logic that affects the operation of this controller or other controllers.
c. External logic that may affect the operation of this controller.
d. Anti-reset windup (ARWU) via external feedback.
e. Explicit bumpless transfer blocks.
Any controllers with the “#” mark should have more detailed control narratives
in the PCN document.
2. Add notes to indicate any other additional information required to facilitate the
design or understanding. For example, for level control of a large vessel, show
the vessel dimension (height and diameter) and the flow range to help vessel-size-
based controller tuning (see Sect. 9.5).

The progression from Figs. 6.1, 6.9, 6.10, and finally Fig. 7.5 illustrates the work
process. The plant or unit-control examples in Chap. 12 all have a COD as the primary
means of presentation. After all, “a picture is worth a thousand words.”

7.6 Summary

The process analysis in Chap. 6 helps build a solid understanding of the process flow,
design intent, operating requirements, and operating challenges. It is the prerequisite
for a sound process control design, either baselayer control or advanced control.
This chapter continues from the process analysis and shows how to perform pro-
cess control design based on the process analysis results. Because of the diversity of
process flows, providing ready-to-copy control solutions for various process control
problems it is not practical nor efficient. Instead, this chapter provides some essential
analytical skills and generic processes that can be followed by any process control
design. The same skill and methodology also apply to troubleshooting and improve-
ment of existing control solutions. The key considerations in process control design
are summarized in Fig. 7.6.
Section 7.1 discusses the process hierarchy and the corresponding control scope
at the different operating levels. While in Sect. 7.2, the overall control strategy and
the layered control structure are explained. Section 7.3 provides some guidelines on
control functional design, with emphasis on the control algorithm selection. Finally,
Sect. 7.4 discusses the typical process control deliverables in a new project and
emphasizes the critical importance of the visualization and documentation of the
process control design.
274 7 Methodology of Process Control Design

Fig. 7.6 Scope and checklist for process control design

References

Al-Dawery SK, Dakhil HM (2012) Plantwide control of acetylene hydrogenation process. J Eng
Sci Technol 7(1):11–24
Åström KJ, Kumar PR (2014) Control: a perspective. Automatica 50(1):3–43
Chin TG (1979) Guide to distillation pressure control methods. Hydrocarbon Processing
Downs J, Skogestad S (2011) An industrial and academic perspective on plantwide control. Ann
Rev Control 99–110
Foss AS (1973) Critique of chemical process control theory. AICheE J 19:209–214
Juliani RCG, Garcia C (2017) Plantwide control: a review of design techniques, benchmarks, and
challenges. Ind Eng Chem Res 56(28):7877–7887
Luyben WL, Tyréus BD, Luyben ML (1998) Plant-wide process control. McGraw Hill, New York
Marlin TE (2015) Process control - designing processes and control systems for dynamic perfor-
mance, 2nd edn. McGraw-Hill
Rangalah GP, Kariwala V (eds) (2012) Plantwide control: recent developments and applications.
Wiley
Seames W (2018) Designing controls for the process industries. CRC Press
Skogestad S (2000) Plantwide control: the search for the self-optimizing control structure. J Process
Control 10:487–507
Stephanopoulos G (1983) Synthesis of control systems for chemical plants - a challenge for cre-
ativity. Comput Chem Eng 7(4):331–365
Chapter 8
Methodology of Control Problem
Troubleshooting

Control problems can range from sub-optimal performance, crippled operation, to

complete breakdown (Sect. 8.1). The causes may be simple instrumentation failures,
dynamics changes, and the control design flaws (Sect. 8.2). For this reason, trou-
bleshooting is typically the most challenging work for a process control engineer
demanding the highest level of expertise and experience. The troubleshooter not
only must understand how things work but also needs to know how things may fail
(Sect. 8.3). Consequently, the solution can be simple instrument repair, controller tun-
ing improvement, controller configuration, to a control scheme re-design (Sect. 8.4).
Overall, control problem troubleshooting is a mix of art and technology with no fixed
recipes. Nevertheless, some general guidelines and best-practice procedures make
the troubleshooting effort more technical and less artistic.

8.1 Common Control Problems

Process control is a “hidden” technology (Åström 1999). When the control scheme
works well, nobody notices. However, when it fails, the results can be catastrophic!
Some major accidents include Three Mile Island nuclear reactor (PA, USA, March
1979), Chernobyl nuclear reactor (April 1986), and the Texas City Refinery fire
(March 2005).
A typical operating plant can have hundreds to even thousands of control loops
monitored by only a small and shrinking number of operators. The staffing level might
be sufficient for normal operation but can be easily overwhelmed during major plant
upsets. It is critical to minimize the abnormal operating conditions and detect the
process control problems early before it deteriorates from a performance issue into
a safety problem.
Some commonly reported problems (or symptoms) on process operation include
the following:
• Frequent process upsets, trips, or breakdowns.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 275
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_8
276 8 Methodology of Control Problem Troubleshooting

• Excessive number of alarms leading to frequent operator interventions.

• Controller not in the desired normal mode most of the time.
• Persistent cycling in one or more process variables.
• Difficulties in achieving product specifications.
• Excessive flaring and recycling.
• An excessive number of starts/stops of critical equipment such as pumps or com-
pressors.
• Difficult startups.
• Process values diverge and become unstable, and thus controller has to be taken
out of control.
• A controller takes too long to bring PV to its setpoint or exhibits excessive PV
deviation from the setpoint at steady-state.
• Controller outputs in saturation for a prolonged period.
• Excessive valve movements leading to excessive wear & tear.
For operation and troubleshooting, it is crucial to know what the normal per-
formance is to be expected. This normal performance serves as the baseline when
identifying abnormal performance issues.
It is also critical to make the initial judgment if the operating problem is indeed
a control problem. Many of the problems are clearly control problems, while many
others may not be that obvious. In an operating plant with a well-established pro-
cess control team, many operating problems are customarily reported to the process
control team for troubleshooting because of trust. Thus, the process control engineer
needs to be aware that many alleged control problems may turn out to be non-control
related at all.
Most control textbooks provide detailed theoretical analysis on the control stabil-
ity using advanced tools such as Nyquist plots, root locus analysis, Lyapunov criteria.
However, these analyses are of minimal practical use because they all assume a known
linear time-invariant process dynamics. In reality, however, the process dynamics are
rarely linear or time-invariant. Besides, the control design is based on loop dynam-
ics rather than process dynamics alone. Other changes in the control loop, such as
the valve characteristics, measurement noise, disturbances, and especially the con-
trol structure, play a more detrimental role in the loop stability. Table 8.1 lists the
common issues with a complex control scheme, from a broken loop, bad data, con-
strained data, to drifting dynamics. These issues may be caused by process dynamics,
measurements, valves, or control design.

Table 8.1 Common control issues and possible causes

Issues Process Measurement Valve Objective Controller
Broken data flow X X X X
Bad data value X
Limited data value X X X X
Drifting dynamics X X X X
8.1 Common Control Problems 277

A control loop has five essential components (Sect. 2.1.2). Issues with the process
dynamics are only part of the potential causes of control problems.

8.2 General Troubleshooting Procedure

Control scheme troubleshooting is a mixed bag of arts and technologies, with no

cookbook recipes to follow. The best weapons are engineering common sense, pro-
cess insight, control knowledge, practical experience, and enough patience. Despite
the uncertainties, some general guidelines and procedures can help make the trou-
bleshooting effort less painful.
The control problems can be divided into several levels, corresponding to the
analysis in Chaps. 6 and 7. The general troubleshooting procedure follows this order:

design → implementation → operation. (8.1)

1. Control design: Flaws in the control strategy or control structure have a significant
impact on the overall operations. It requires a higher skill to locate and is typically
expensive to correct.
2. Control implementation: For complex PID control loops or model predictive con-
trol, implementation is another area that can often go wrong. A thorough under-
standing of the control scheme and some help from control systems engineers are
most useful.
3. Control operation: During operation, the loop dynamics may change for many
reasons. System troubleshooting procedures can help to detect and diagnose the
issues.
A general troubleshooting procedure is illustrated as a flow diagram in Fig. 8.1
for easy visualization. The following sections discuss the critical steps of loop trou-
bleshooting and provide guidelines on what to check for each component of the PID
loop.

8.2.1 Obtain a Problem Statement

Troubleshooting a process control problem is very much like a doctor seeing a patient.
The first step is to obtain a problem statement, i.e., the symptoms or complaints. Con-
trol problems are typically reported by operating personnel but may also come from
other sources, such as a proactive control performance monitoring (CPM) application
(see Sect. 2.3.6).
To have a good understanding of the complaints, try to find out from operations
(e.g., the panel operators) and instrumentation engineer for answers to the following
questions:
278 8 Methodology of Control Problem Troubleshooting

Fig. 8.1 PID loop troubleshooting procedure

1. What is wrong with the loop? That is, what are the complaints and symptoms of
the reported control problems?
2. When did the problem start? Is this the first time happening, or is it a recurring
issue?
• If this is a new problem, what has been changed recently to process, equipment,
operating condition, control valve, or instrumentation?
• If this is a pre-existing problem, is it just a simple repetition of the last occur-
rence, or is it getting worse with new symptoms?
3. Have there been any significant changes to the plant or operation requirements?
For example, are there any recent process modifications, feed/load changes, turn-
down operations?
4. Have there been any major repairs or maintenance work on the instrumentation
or control valves?
5. What is the severity of the problem? That is, how critical is this control loop to
safety, production quality, or process operability?
6. How urgent is a solution for the problem? That is, what is the risk if no action is
to be taken immediately?
8.2 General Troubleshooting Procedure 279

The problem statement provides a basic understanding of the severity and urgency
of the reported problem and serves as a starting point for planning the troubleshooting
effort. However, the description of the control problems from the operational person-
nel is often ambiguous, sometimes exaggerated, or understated. On the other hand,
alerts from control performance monitoring (CPM) applications are typically more
objective and accurate. Depending on the location and nature of the problem, the
statements may need to be double-checked and cross-validated. The troubleshooter
needs to research further into the reported problem to obtain a complete picture and
verify it before starting troubleshooting:
1. Collect the process control narrative (PCN) document, process flow diagrams
(PFD), and piping & instrumentation diagrams (P&ID) to understand the process
design, operating philosophy, the control objective, and the interactions with other
control loops.
2. Check history data and the maintenance record to verify the problem history.
3. Evaluate the criticality of the loop relating to process safety, product quality, and
process operability.
4. Verify if there have been any significant process changes that can affect the loop(s)
in question.
5. Verify if any operating condition changes (e.g., a significant shift in control set-
points) have moved the operating points outside the normal operating envelope.
6. Verify if multiple loops are suffering from the same or similar problems (e.g.,
oscillations).

8.2.2 Identify Human Errors

Human error is a common cause of process control problems. The consequence of

human errors can be from poor control performance, crippled operation, to sometimes
catastrophic incidents. For example, the Chernobyl nuclear reactor (April 1986)
incident was a combination of mechanical and human error. The Texas City Refinery
incident is caused by instrumentation error and human error (CSHI 2005).
Human error is more related to violations of standard operating procedure (SOP)
and is thus beyond the scope of this book. However, it is essential to rule out human
error preliminarily before immersing oneself in technical troubleshooting too quickly.
Besides, if human errors have caused the issues, many can be prevented or mitigated
by improved process control design:
1. Operator mal-operation: Normal plant operation is commonly repetitive and bor-
ing, and over time the operators can become less vigilant, and human errors start
to occur. As an example from personal experience, while doing step tests for a
model predictive control, a request was made to the operator to step change a
temperature from 110◦ C to 100◦ C. When entering the new setpoint, the operator
carelessly entered 10 instead of 100. Although he realized the error immediately
and re-entered the correct setpoint right away, the vast step changes from 110 to
280 8 Methodology of Control Problem Troubleshooting

10 and back to 100 already caused a significant upset to the column operation,
which took a few hours to re-stabilize. The small incident definitely ruined the
step-test plan for that day. Fortunately, it did not result in a costly shutdown.
If this part of historical data is investigated as part of a troubleshooting effort for
something else, it might be challenging to find out that the unit-wide fluctuation
in the data was caused by a human error rather than any control problems.
Naturally, this problem could have been avoided if the operator stayed more vig-
ilant when operating the plant. However, on the process control side, a high and
low setpoint limit could have been configured for the PID controller (see Sect. 3.5)
to prevent such an abnormally large setpoint change from the operator.
2. Controller not in the desired mode: Another common cause of poor performance
is that operating personnel switches controllers away from their intended mode
of operation (the normal mode). For example, many PID controllers are put in
MAN mode instead of the desired AUTO or CAS mode for no good reasons. Each
experienced operator usually has his or her own “comfort zone,” which may
not always be the best operating choice. It is common to see that after a shift
change, the first thing the operators do is to make many setpoints changes to
bring the process operation to their own “comfort zone.” Model predictive control
(MPC) is often switched off due to unfamiliarity with the technology and lack
of understanding of the moves produced by MPC. For complex control loops
involving multiple controllers, a poorly performing controller could impact the
performance of the entire control scheme or even an entire unit and results in
many PID controller running in manual mode.
The best solution is to improve the control performance so that the operators will
become enthusiastic about keeping the controllers running in automatic mode
since automated control typically makes their work easier after all. At the same
time, the operating procedure should include a tracking mechanism to record
why the operator chooses to take the controller offline. For example, control
performance monitoring (CPM, Sect. 2.3.6) can produce automated reports on
the statistical information on the number of changes made to PID controller mode,
setpoint, or output.
Problems resulted from human error can be very difficult to troubleshoot because,
by human nature, it is common to delay, forget, or even cover up the true cause of
an incident or abnormal condition. Troubleshooting needs to be based on a good
understanding of the process design and a thorough investigation of the historical
data to make informed decisions.

8.2.3 Isolate the Problem

The next step is to narrow down the problem to a specific area or a particular loop. For
example, it is crucial to decide if the problem is a plant-wide problem, a unit-level
problem, or is local to a control loop.
8.2 General Troubleshooting Procedure 281

A plant-level problem is typically caused by a dynamic imbalance in supply and

demand. The inventory fluctuation is reflected by gas pressure or liquid level. Exces-
sive flaring or overflow/underflow of vessels are the consequences. Understanding
the process flow (Sect. 6.2) and the cause and effect relationships (Sect. 6.4) is vital
in isolating the problems.
A common symptom of unit-level or loop-level control problems is the oscillation
in the controlled variables. The oscillation often occurs to multiple loops and even
to a whole unit. Consequently, the problem may often show up as a “chicken or
egg” dilemma, challenging to determine which comes first and which comes after.
The cause and effect relationships (Sect. 6.4) and dynamic response characteristics
(Sect. 6.6) between the key control variables are crucial for troubleshooting.
A common approach to isolate the problem to a particular control loop is to put
the most suspicious control loop in manual (MAN) mode for a few minutes to see if
the oscillations in key process variables start to go away. Continue monitoring the
loop to find out the following:
1. Is the variability in the controlled variables (PV) less than when the loop is in
AUTO? If yes, it indicates that the controller is doing more harm than good.
2. Does the variability diminish or decrease in other loops when the loop is in
manual? If yes, that means the interaction exists, and the loop under investigation
may be the root cause.

8.2.4 Identify the Root Cause

The root cause of a control problem can be pretty challenging to determine, depending
on the complexity of the control scheme. For example, oscillating loops can be caused
by many reasons, including the following:
1. Interaction between the loops, where the output of one controller affects another
controller and vice versa.
2. Valve stick–slip from excess friction in the packing or sealing surfaces.
3. Incorrect PID tuning.
Although the causes of the control problems can be many, process control prob-
lems can be classified into three categories or levels: design, implementation, and
operation. It is often a combination of two or more problems lined up to cause a
major incident. See the Swiss cheese model (Euro-Control 2006):
1. Design problems: Many chronicle control problems are due to flaws in control
design, especially in the overall control strategy (see Sect. 1.1.3). Once in opera-
tion, these types of problems are difficult or expensive to correct. For example, a
supply-driven control scheme for a demand-driven operation may result in con-
tinuous swings in process variables or permanent offset in steady-state values.
A flawed control strategy is unlikely to achieve satisfactory performance even
with all the individual controllers perfectly functioning. The troubleshooting thus
282 8 Methodology of Control Problem Troubleshooting

should start with a good understanding and careful validation of the overall control
strategy.
2. Implementation problems: The configuration and tuning of a complex PID con-
trol loop or a model predictive control scheme are not trivial and often require
many careful considerations. Incorrect configuration is a common source of per-
formance issues. Reset windup or bumpy transition are examples of loop integrity
problems in complex PID control schemes.
3. Operation problems: Operation problems are typically related to the changes in
the loop dynamics. As discussed in Sect. 4.3.1, loop dynamics are determined by
the data flow, subject to four types of common problems:
a. Broken connection of the data flow.
b. Bad data in the data flow.
c. Limited or constrained data values.
d. Drifting in loop dynamics.
The control loop comprises five essential components, and each can contribute to
the problem, with the final control elements, the measurements, and the process
dynamics being the most common culprits. See Table 8.1. A structured review of
the five essential components should be carried out if the problem is pinpointed
to a specific PID control loop. Process measurement can be erroneous but can
typically be fixed easily. Final control elements, especially control valves, are the
most common causes of control problems and are often hard to resolve without
plant shutdown.
Controller tuning is less likely the cause for existing control loops. A control
loop is designed and configured for specific loop dynamics. Unless there is a
significant change in the loop dynamics, the controller is expected to deliver
the same performance. As explained in Chap. 4, many factors affect the loop
dynamics; it is crucial to determine which part of the loop dynamics has changed
before jumping to control loop tuning.

8.2.5 Propose a Solution

For complex control schemes, proposing a solution to fix the existing problem can
follow the same process control design methodology discussed in Chaps. 6 and 7.
The solution can be a permanent fix or a temporary workaround. In response to
an urgent request from operations, a temporary workaround is often proposed and
implemented first to allow the operation to continue. In reality, however, it is often
found that once the temporary workaround is in place, operations are content with it
and stop pursuing a proper permanent solution. As a result, a temporary workaround
becomes a permanent solution. As a lesson learned, when the temporary workaround
is implemented, make sure that the permanent solution will follow through.
For standalone control loops, depending on the specific cause, process control may
propose solutions ranging from modification of the process, cleaning of equipment,
8.2 General Troubleshooting Procedure 283

calibration of instrumentation, maintenance of control valves, or fine-tuning PID

controls.
Except for PID tuning, all other remedies need to be carried out by other depart-
ments, a proof that process control is a cross-disciplinary engineering specialty
(Sect. 1.1.1), and close collaboration with other teams is a must. On the other hand,
only process control offers this cross-disciplinary view of the operating problems
(Sect. 1.1.4).

8.3 Troubleshooting of Complex Control Scheme

The crucial step for a complex control scheme consisting of multiple variables and
multiple PID controllers is to ensure that the overall control strategy is correct and
fit-for-purpose.

8.3.1 Overall Control Strategy

The overall control strategy must meet the control objectives, which in turn serve the
operating objectives. It is vital to have a good understanding of the operating objec-
tives and operating requirements. The different components in a properly designed
control scheme must perform their duties; they must work together seamlessly with
other components to achieve the high-level goals. Even the best configured or tuned
loops will not deliver the desired performance with a flawed control strategy.
Chapters 6 and 7 have provided a systematic discussion on reaching a sound
overall control strategy. The same process analysis and control design methodol-
ogy discussed in Chaps. 6 and 7 also apply to control scheme troubleshooting and
improvement. For any control loop, whether a simple standalone PID loop or com-
plex control scheme, and for either design or troubleshooting purposes, the following
questions can be asked to gain the necessary knowledge about the loop.

WHERE Is the control scheme/loop located?

WHY Is the control scheme/loop needed (process reason)?
WHAT Functionality does (should) the control scheme/loop provide?
WHEN Is the control scheme/loop expected to respond?
HOW Was (will be) the control scheme/loop implemented in DCS?
WHAT IF The operating condition is not met?

The troubleshooting should include the following reviews and analysis to under-
stand and validate the control strategy:
1. Process flow analysis.
2. Supply and demand analysis.
284 8 Methodology of Control Problem Troubleshooting

3. Cause and effect analysis.

4. Degree of freedom analysis.
5. Dynamic response analysis.
6. Layered control design.
Plant-level troubleshooting depends on a good analysis of the process flow and
the supply and demand model; unit-level control relies on a good understanding of
the cause and effect relationship and dynamic characteristics; the loop-level trou-
bleshooting depends on the type of control scheme, whether PID-based control or
advanced process control.

8.3.2 Dynamic Supply and Demand Balance

In upstream E&P operation, the most common operating requirement is the physi-
cal separation and movement of the produced fluid. Maintaining a healthy balance
between the supplies and demands, both static and dynamic, is mandatory at any
point on the process flow path.
For gas processing, a prolonged imbalance between gas supply and demand will
result in continuous flaring. Environmental regulation has been calling for zero-
flaring during normal operation, and thus flare reduction is a pressing target. Constant
flaring indicates poor process design, while temporary flaring indicates process upset
that has exceeded the controllability range of the normal regulatory control loops.1
For liquid processing, imbalance in the supply and demand is typically buffered in
surge tanks but up to a limit. Over the limit, the unit operation has to turn down the
throughput; otherwise, the unit will be tripped to a shutdown.

Example 8.1 Continuous flaring. Let us re-visit the produced gas compression
example from Sect. 6.3. The original process design is as shown in Fig. 8.2.
There is one gas supply from the gas/liquid separator, and there are two consumers
for the compressed gas, namely the gas lift wells and four heat recovery steam
generators (HRSG). The demand is forecast to be greater than the supply, so a make-
up gas stream is provided from a sweet gas source if needed.
Following the analysis in Sect. 6.3, the supply side is production-driven, thus
supply-driven. The two consumers are either on flow control or pressure control, so
they are both demand-driven. The make-up gas provides the swing control to regulate
the dynamic supply and demand balance.
After the plant was built and brought online, many operating problems were
encountered, including frequent compressor tripping, severe and continuous flaring,
and failure to maintain compressor suction pressure at the setpoint. The heavy flaring
went on for two years. At the peak time, the flaring rate reached one million standard

1 It is common to see large flares in a refining or ethylene plant during startup or shutdown, when
the supply and demand balance cannot be met.
8.3 Troubleshooting of Complex Control Scheme 285

Fig. 8.2 Bad process design: gas compression

cubic meters per day! The operation had to rely on repeated renewal of government
waiver on flaring to continue operation.
The process control group was approached to troubleshoot the flaring problem.
After a careful review of the process flow, material balance, and operating condition,
it was found that the root cause is the imbalance in supply and demand, due to the
following causes:
1. The original design was based on the assumption that the produced gas is insuffi-
cient to meet the needs of the gas lift wells and the steam generators. A continuous
make-up gas stream was expected to fill in for the supply shortage and serve as
the swing stream. However, after startup, difficulties with the steam generators
delayed the commissioning of two of the four steam generators. The gas demand
was lower than the forecast and resulted in gas oversupply.
2. Subsurface study has underestimated the GLR (gas–liquid ratio) of the reservoir.
For every barrel of oil produced, there was higher associated gas production than
expected. This higher GLR made the gas oversupply problem worse.
The control overview diagram (COD) for the original control design is illustrated
in Fig. 8.3.
The control scheme clearly shows that the excess gas has nowhere to go but to
flares. The over-supply in the produced gas causes the compressor discharge pressure
to increase. The elevated pressure causes the pressure controller PC-103 to take
charge and overrides the capacity controller PC-102, which leads to higher pressure
at suction. The suction pressure reaches the setpoint of the protective controller
PC-101, which kicks in and opens the valve to flare.
In other words, due to the supply and demand imbalance, the supply-driven oper-
ation was reversed to a demand-driven operation at the suction side of the compres-
sor. The protective controller PC-101 overrode the normal regulatory controller
PC-102 and became the normal regulatory controller. What makes things more
challenging is that the operating pressure of the vapor recovery unit (VRU) floats
with the compressor suction. The suction pressure operated much closer to the trip
286 8 Methodology of Control Problem Troubleshooting

Fig. 8.3 Bad process design: supply/demand imbalance

limit of the VRU. Process upsets in the compressor train had caused several trip
incidents in the VRU.
Plant operation was under pressure to maximize oil production, and thus, scaled
down the overall capacity was not an option. This steady-state supply and demand
imbalance could only be corrected by process modification. Based on the recommen-
dation from process control, a plant shutdown was scheduled, and an extra stream
was added to send excess gas to gas injection wells. The new process design and
control scheme has been discussed in Sect. 6.3.

Flaws in plant-level control design are often related to inadequate supply and demand
handling, and the fix is typically costly once the plant is built.

8.3.3 Dynamic Cause and Effect Relationship

As shown in Fig. 7.1, unit-level problems are often caused by incorrect or unreliable
cause and effect relationships. Feedback control is based on a known and credible
causal relationship between the controlled and manipulated variables. If this causal
relationship is weak or changing, the control performance would be unpredictable.
Let us use a unit operation example to illustrate the criticality of the cause and
effect relationship.
Example 8.2 Figure 8.4 is a simplified process flow for the true vapor pressure
(TVP) control loop in a crude oil stabilizer. TVP is an important quality criterion for
crude oil, and too high a TVP value is a safety hazard, while too low a TVP value
means loss of oil to gas.
The corresponding control scheme is illustrated in Fig. 8.4. The control scheme
shows that a PID control loop is implemented to control the TVP by manipulating
the stabilizer bottom temperature.
However, the complaints about the stabilizer operation are multiple:
8.3 Troubleshooting of Complex Control Scheme 287

Fig. 8.4 Bad process design: unreliable cause and effect

1. The TVP controller can never be put into automatic AUTO operation for an
extended time due to inferior control performance.
2. The TVP is manually controlled by manipulating the bottom temperature, and
the TVP of the crude oil in the storage tank fluctuates so much and at one time
even popped up the floating roof and caused a gas leak and oil spill.
Again, the problem was brought to the attention of process control. A systematic
troubleshooting effort arrived at the following findings, among others:
1. The TVP online analyzer is highly unreliable due to technology limitations and
poor maintenance. It is far from acceptable for closed-loop control.
2. The TVP analyzer measurement lags the actual changes in the stabilizer by about
40 min due to poor choice of location. As mentioned in Sect. 2.3.1, a long time
delay is the number one challenge for PID control, and a 40-minute delay is
beyond impossible for PID control.
3. The process variables that affect the TVP are not just temperature. The pressure
in the stabilizer also significantly impacts the TVP value and is thus an essential
component in the cause and effect relationship.
Based on the discussions in Sect. 8.3.3, to properly control the TVP value in the
stabilizer, the control scheme must include all the critical cause and effect relation-
ships related to the controlled variable, and the dynamic response corresponding to
each cause and effect relationship must be fast, predictable, and reliable. The above
control scheme has violated all the above requirements.
The improvement is implementing an inferential property estimator (soft sensor),
where faster and more reliable temperature and pressure measurements are used to
predict the TVP value. A PID controller controls TVP based on the predicted TVP
value instead of the value from the online analyzer.
288 8 Methodology of Control Problem Troubleshooting

8.3.4 Degree of Freedom Review

All the controlled and manipulated variables must pair up to leave no degree of
freedom (DOF) unconsumed in a multivariable control design. If not, the unconsumed
DOF will find its way to get consumed, typically by going to saturation. A DOF
analysis (Sect. 6.5) can reveal many design problems. Some are obvious, and many
are not.
We will make use of two examples to show how DOF analysis is applied in
troubleshooting.
Example 8.3 Over-specified pressure control. Figure 8.5 shows a gas pressure let-
down circuit that aims to maintain a 10 bar header pressure of gas supply to down-
stream operation. Two pressure controllers are installed in parallel, with setpoints set
to 10 bar.
The two individual pressure controllers both look innocent and are also properly
configured and tuned. The two valves offer two DOFs, and there are two pressures
as controlled variables.
However, this control scheme will not perform well. The problem is that the two
pressure variables are the same header pressure. They are not sufficiently independent
and can only consume one DOF. The overall control strategy of using two controllers
to control the same header pressure is flawed. There are two valves for one pressure,
which is one extra DOF unconsumed. As expected, at the steady-state, there is always
one valve at saturation.
The solution to this control scheme is to re-do the control design. The improved
design should have the header pressure as the only controlled variable and use either
a split-range control or a fan-out control to manipulate the two control valves.
Sometimes this DOF over-specification is not apparent. For example, if the two
controllers are far apart along the path, they may not appear on the same P&ID page
or DCS screen (a key reason that a COD is necessary).
Example 8.4 Over-specified flow control. Figure 8.6 shows a water treatment pro-
cess where the entrained sand is removed by educers upstream and a downstream
hydro-cyclone.

Fig. 8.5 Bad control design:

over-specified pressure
control
8.3 Troubleshooting of Complex Control Scheme 289

Fig. 8.6 Bad control design: over-specified flow control

In an early design, flow controllers are implemented at the educers in upstream.

While downstream at the hydro-cyclone, the differential pressure (DP) across the
hydro-cyclone is controlled to ensure the proper operation of the hydro-cyclone.
The educer and hydro-cyclone are supplied by two different vendors. Both vendors
insist that the control scheme for their equipment is adequate with a proven track
record.
Since the two systems are far apart, they seem to be okay when viewed in isolation.
However, flow rate is a function of differential pressure across the hydro-cyclone.
The DP control is the same as flow control. This design has two flow controllers on
the same stream. One extra DOF is not consumed, and one of the valves will end up
in saturation (fully open).

8.4 Troubleshooting of PID Control Loop

If the control scheme is a simple standalone PID control, or a complex control scheme
where the control problem has been pinpointed to a specific PID loop, then some
systematic PID troubleshooting procedure can be followed.
As explained in Chap. 3, a PID control loop comprises five essential components,
process dynamics, process measurements, final control elements, controller setpoint
(objective), and the controller. The loop dynamics, determined by all the components
on the data flow, dictate the control performance (King 2016; Luyben and Lyuben
1997; Marlin 2015). When troubleshooting, always keep the complete loop in mind
and not take any component out of its loop context.
The most vulnerable component in the data path is the final control elements,
specifically, the control valves. Process measurement also creates many troubles for
control.

8.4.1 Process Dynamics

Control performance is determined by the loop dynamics instead of the process

dynamics alone. However, as mentioned in Sect. 3.3.2, process dynamics are the
290 8 Methodology of Control Problem Troubleshooting

Fig. 8.7 Process dynamics verses loop dynamics

most critical part of the loop dynamics. Change in process dynamics can render
the process control configuration inadequate, especially the PID tuning parameters.
Figure 8.7 shows how the loop dynamics and process dynamics relate to each other.
Depending on the type of change to process dynamics, from slow aging/fouling
of equipment to significant modification of process flow, its impact on control per-
formance can vary. Some of the changes in dynamics are obvious, while others may
be subtle.

Example 8.5 Control tuning for fan-out control. Suppose a water tank level is con-
trolled by two pumps running in parallel at normal conditions, and the controller
is adequately tuned. If the processing capacity reduces and only one pump needs
to be in operation, the original PID tuning would be no longer adequate because
the process gain has changed. Depending on how much performance degradation is
acceptable, adjustment of the control gain may be required.
This example shows that although the individual pump dynamics remain the same,
the process flow has caused the process dynamics (two pumps vs. one pump) to
change and thus requires a change in PID tuning.

A more subtle example is with a distillation column where the process dynamics
for control can vary with the operating condition and significantly affect the control
design.

Example 8.6 Fractionator bottom temperature control. For the bottom impurity con-
trol, a tray temperature is typically selected as the manipulated variable. That is, the
process dynamics are from the tray temperature to the bottom impurity level.
It is well known that the relationship between the tray temperature and the impurity
is very nonlinear at a high purity level. It is also known that the most sensitive
temperature (at the so-called sensitive tray) shifts to a lower tray if a higher purity
(lower impurity target) is desired. When implementing advanced process control,
which is supposed to improve the separation and lower the impurity level, a different
tray temperature often needs to be selected to control the bottom impurity.

Heat exchanger fouling, filters/strainers blockage, distillation column flooding,

and catalyst deactivation can all cause the process dynamics to change over time.
8.4 Troubleshooting of PID Control Loop 291

Some of the changes are within the range of robustness of the PID controller, while
others may degrade the control performance so much that adjustment of the PID
tuning or change of control strategy is needed.

8.4.2 Process Measurement

Process measurement typically includes a primary device (e.g., a sensor) and a sec-
ondary device (e.g., a transmitter). The sensor senses the changes in the process
variables while the transmitter converts and conditions the sensor output to an elec-
tric signal appropriate for transmission to other devices such as DCS or PLC. Signals
for transmission are typically in 4–20 mA current signals.
The analog input (AI) block in DCS receives the transmitter signal (e.g., 4–20 mA).
It converts the signal to a value with the appropriate engineer unit and scale, based
on the design datasheet and via calibration and re-calibration. This value becomes
the process value, or PV, for the control loop.
All measurements have errors. For example, the measurement resolution is limited
by the analog/digital (A/D) converter, which is at least 0.05% of the span for a 12 -bit
A/D conversion (1/212 ). Instrument error can occur due to various factors such as
calibration, environment, electrical supply, the addition of components to the output
loop, and process changes.
The drift in the zeros, incorrect span, and linearization errors are typical calibration
errors and can be corrected by re-calibration. Most instruments are provided with a
means of adjusting the zero and span of the instrument. Many instruments offer a
linearization adjustment to correct linearization errors.
Some common troubleshooting steps on measurement problems include the fol-
lowing (Brannan 2018):
1. Check whether the transmitter “freezes.” A frozen transmitter could drive the
controller to saturation and thus out of control.
2. Check for any significant non-uniform fluctuations in the measurements. For
example, a spike or step change in flow or temperature that is physically impossible
indicates a measurement problem.
3. Calibration: Check whether the instruments, especially analyzers, are correctly
calibrated and serviced. Transmitter zero shift is a common issue.
4. Check the ranges and scan times to make sure they are adequate.
5. Check the location of the transmitter. For example, for a flow controller, the flow
measurement should be located in front of the valve to avoid the turbulent flow
caused by the valve. For liquid temperature measurement, the sensor should be
immersed in the liquid, not in the vapor or the splashing liquid.
6. If PV is very noisy, consider adding a small PV filter to smooth out the high-
frequency measurement noises.
Some noises and disturbances with slow frequency can be corrected by PID control,
while others, such as those with higher frequencies, must be corrected at their origin.
292 8 Methodology of Control Problem Troubleshooting

Although high-frequency noises can be smoothed out with a small filter, the filter time
will become part of the process dynamics and must be considered when designing
process control loops.

8.4.3 Final Control Elements

The feedback controller calculates the control action to eliminate or minimize the
control error by bringing the process value closer to its setpoint at each execution.
The control output is either in the percentage of scale (0–100%) or in the actual
engineering unit. The output signal is then converted to 4–20 mA by the analog
output (AO) block and sent to the final control element. The final control element,
which typically is a valve consisting of a positioner, actuator, and the valve, receives
the control signal in the range of 4–20 mA, converts it to 3–15 psi pressure signal to
drive the air-operated valve.
The control valve is the component in a control loop with the most troubles in a
closed-loop control system, primarily because of wear and tear. The valve problems
degrade the integrity of the control loop and render the valve dynamics (which is
the part of the loop dynamics) uncertain or time-variant. Common valve problems
include the following:
1. Valve stiction: Stiction refers to stickiness and friction. With a sticky flow valve,
there is no change in flow until the accumulated changes exceed a certain amount
(“the amount of stiction”). Valve stiction adds delay to a loop for any changes in
the controller output.
2. Valve overshoot (slip): The change in flow is more discrete than continuous and
is much larger than demand. Valve slip causes the PV to pass by the setpoint.
3. Dead band: No PV response is observed until the manipulated variable has
changed more than a certain amount. Dead band can be caused by many fac-
tors, such as backlash, excessive friction, or shaft deflection.
4. Step resolution: The minimum step change in the input signal to which the control
valve will respond while moving in the same direction. This phenomenon is
typically caused by the tendency for the control valve to stick after coming to
a rest and takes a more remarkable change in controller output to initiate the
correction at the valve. Degraded step resolution means more lag and error that
lead to the stick–slip limiting cycle problem, characterized by continuous cycling
at a steady-state under closed-loop control, even with correct tuning.
5. Hysteresis: Hysteresis indicates that the valve position can take any valve position
within a range of values for a given input signal. There is no change in flow when
the controller reverses direction until the change exceeds a certain amount.
6. Nonlinearity: All valves have nonlinearity. The valve characteristics are different
from valve to valve. The nonlinearity becomes more severe towards both ends of
the characteristic’s curve. It is advised to limit the normal operating range to the
middle section of the valve curve, typically 10% to 80%.
8.4 Troubleshooting of PID Control Loop 293

7. Valve Rangeability: Valve is too small or too large for the loop to be controlled.
For example, a column temperature control with an undersized reboiler is very
likely to find that the valve stays saturated most of the time and is effectively out
of control. On the other hand, an oversized or undersized valve may force the
operating point to operate outside the valve’s normal operating range, becoming
more nonlinear.
The normal process operating range should typically fall between 10% and 80% of
the valve operating range, and the valve should be sized accordingly. In practice,
however, it is not uncommon to see the contrary. For example, in a reciprocating
compressor, the recycle valve is typically used for capacity control. Since recy-
cling is expensive, many control specifications place the normal operating point
too close to the upper range and thus lose the controllability in that direction. It
is crucial to allow the valve sufficient room to move around the normal operating
point in both directions.
Valve hysteresis plus dead band is the total friction in the valve. All valves have
hysteresis and resolution limits. For control, valve slip is typically worse than valve
stiction, and valve stiction is worse than dead band. To check for valve hysteresis or
stiction, make several controller output changes: at least two steps in one direction
and two in the other, followed by several tiny steps to check stiction. Using the data,
run a hysteresis check on the loop. If the hysteresis is more than 1% for valves with
positioner and 3% for valves without positions, repair or replacement of the valve
should be considered. Hysteresis of 1–4% degrades loop performance, while with
tight tuning, hysteresis greater than 3% may start to cause oscillations.
Some valves (e.g., smart valves in Fieldbus control system) can provide the actual
valve position as read-back. The controller can use the valve read-back value to
effectively eliminate many of the control problems related to the valve. For example,
the anti-reset windup configuration shall use the valve read-back where available.
The following checklist can help troubleshoot some common valve problems:
1. Check valve history data (SP, PV, OP) to look for unusual behaviors or patterns.
2. Check if the controller output is high or low limited when the loop problems
occur.
3. Check valve read-back, if available, to see if the positioner follows the controller
output well.
4. Put the valve in manual mode, and see if the variability in the controlled variable
is less than when the loop is in automatic mode AUTO. If yes, it indicates that the
control loop is doing more harm than good.
5. Is the valve oversized or undersized for the loop? An oversized valve reduces
the sensitivity of the controller, while an undersized valve limits the controllable
range.
6. Is the valve sufficiently linear within the normal operating range. Put the loop
in manual mode (MAN) and make a few steps in controlled output in both direc-
tions. The linearity check includes a symmetry check to investigate if the valve
responds uniformly in both directions. If the valve nonlinearity cannot be cor-
294 8 Methodology of Control Problem Troubleshooting

rected, then more conservative tuning or valve characterization can be considered

to compensate for the nonlinearity.
7. Is the loop problem limited to only a specific range of valve output, or can it occur
at the full range of output?
8. Check the PV response to control output changes and determine the severity of
the valve slip, stiction, and dead band.
9. Check the bypass valve, if any, to see if it is in the appropriate position. The bypass
valve opening can shift the controllable range of the control valve.
A scatter plot of the controller PV vs. OP can reveal many valve performance
issues. See Sect. 8.5 for examples of using scatter plots for valve troubleshooting.
Note that although the ultimate solution is to fix the root cause, many times, modi-
fications to the controller tuning (or the control strategy if necessary) can temporarily
reduce the severity of the problem and allows the operation to carry on until the next
scheduled shutdown or maintenance. However, ensure that the temporary solution
will not be taken as a permanent solution!

8.4.4 Control Objectives

The operating objective dictates the control objective. For complex control loops that
involve multiple controllers, it is imperative to view the control scheme and control
loops in their entirety and understand the overall control strategy. With the whole
picture in mind, it is easier to appreciate how the loops work together to achieve the
overall control objectives and understand how they interact and interfere. It often
requires validating the following to troubleshoot the control objectives at the PID
control level:
1. Know what to expect of the control scheme or control loop. For instance, a level
control on a surge drum has a very different objective than that on a knockout
drum.
2. Check the control strategy to see if it serves the control objective. It is not uncom-
mon in practice to find that the control loop design does not meet the operating
need at all times.
3. The control target is realistic and achievable. For example, controlling the steam
pressure at its saturation temperature is hard to achieve. Similarly, nobody would
expect the room temperature to reach freezing in a hot summer by setting the A/C
thermostat to 0◦ . As another example, tight control of a delay-dominant process
with PID may not be realistic.
4. Maintain an adequate balance between loop performance, stability, and tuning
effort. The tightest control setting is often the one that can lead the process to
unstable conditions during a process upset. Besides, the extra effort spent on
reaching the “optimal” tuning may not be justified because the “optimal” tuning
often changes with time and operating conditions.
8.4 Troubleshooting of PID Control Loop 295

8.4.5 The Controller

Some of the control design and configuration problems with the control algorithm
or control design are listed as follows:
1. Cause and effect relationship: Is the controlled variable sensitive enough to the
manipulated variable for decent control? If not, consider controlling a different
process variable. For example, for a distillation column, if the current tray temper-
ature is not at control sensitive tray in a column, the control performance would
not be good, and a different tray may need to be used for control.
2. Process dynamics: Is the dead time of the process short enough for control? Loops
with long time delays are notoriously hard for PID loops to handle.
3. Control structure: Is the controller designed to handle all the operating modes,
such as overriding, split range, fan-out? see Sect. 4.3. Check whether a complex
PID control scheme such as feedforward control and cascade control can help
improve the performance.
4. Measurement: Consider other controlled variables such as pressure compensated
temperature, internal reflux, separation index, heat duty, pressure/temperature
compensated flow, reflux/feed ratio, and reflux/draw ratio.
5. Nonlinearity: Adding nonlinear transformation can often improve the control per-
formance if the relationship between the manipulated and the controlled variable
is very nonlinear. More on signal characterization in Sect. 10.1.
If a controller worked well in the past and stopped working all of a sudden,
the most likely cause is typically not the controller itself (including the tuning) but
something else in the control loop.
Many preventive and protective features are included in controller design and
implementation to deal with abnormal operating conditions. The most common and
essential features include anti-reset windup and bumpless transfer. While these fea-
tures provide the flexibility to handle abnormal conditions, they are also familiar
sources of loop problems.
For each PID controller, troubleshooting should include the validation of the con-
troller configuration listed in Table 3.9. Some of the typical configuration errors that
can result in persistent poor performance of the control loops include the following:
1. Incorrect algorithm type: An I-PD algorithm may not deliver the desired per-
formance for setpoint tracking, while a full-form PID algorithm for a noisy flow
loop can produce many fluctuations.
2. Wrong tuning parameters: During project commissioning, many PID controllers
are left with inadequate system default tunings.
3. Reset windup in complex control loops such as override control involving selec-
tors or switches.
4. Bumpy transfer in complex control loops that involve splitters and split-range
controls.
In the context of a complex control scheme that consists of multiple complex
control loops, it is crucial to account for all the operating scenarios that the controllers
are expected to run.
296 8 Methodology of Control Problem Troubleshooting

Fig. 8.8 Amount of flaring before and after PID tuning

Example 8.7 Improved control performance via tuning. Figure 8.8 shows an exam-
ple of the difference in the amount of flaring in a production station before and after
PID tuning improvement. The flaring rate is controlled by a protective pressure con-
troller at a production separator. The separator level control caused severe oscillations
due to highly aggressive tunings. Those tunings were probably left unchanged from
the DCS system default (see Sect. 3.5.6). The improved tuning reduced the flaring
from about 1,000 m3 /h to under 100 m3 /h, with additional room to improve.

Oscillations caused by poor loop tuning typically show sinusoidal behavior. Saw-
tooth-shaped oscillations are often caused by valve hysteresis, stiction, and nonlinear-
ity. Level loop on a vessel that feeds another unit (downstream) is often a significant
cause of variability. Similarly, a poorly performing pressure control loop is often a
common source of oscillations in a large area.
For oscillations caused by valve stick–slip, the ultimate solution is to fix the valve.
However, if the valve maintenance requires an unscheduled unit shutdown, it may
have to be delayed unless the valve is in critical service. Meanwhile, de-tuning the
loop (smaller controller gain or less integral action) can temporarily improve the
control performance to an acceptable level to wait for the next scheduled shutdown.

8.5 Troubleshooting Tools

Software tools are becoming widely available to help monitor and diagnose loop
performance problems. Most tools can continuously monitor all the control loops in
a plant or unit and proactively report those under-performing loops based on pre-
defined statistical or analytical criteria. Some software tools can even provide an
excellent initial diagnosis.
8.5 Troubleshooting Tools 297

Most of the tools, however, only detect and report the problems. To pinpoint
and fix, the problem is the most challenging part of the effort. Knowledge, skills,
experience, plus sufficient patience are still the key.

8.5.1 Data Trending and Scatter Plots

Process control engineer’s most effective and frequently used tools are still the basic
tools available on everyone’s desktops:
1. Trending: Data visualization and trending is an essential feature in any decent
data analysis software. Trending the setpoint (SP), process value (PV), and con-
troller output (OP), sometimes also with controller mode (MODE), can reveal many
problems in measurements and control valves.
2. X-Y Plot (scatter Plot): An X-Y plot of the process variable (PV) and the controller
output (OP) with a spreadsheet software or most monitoring tools can often expose
many control valve problems and controller nonlinearity.
3. Auto-correlation and cross-correlation analysis: Correlation analysis is a valuable
tool for performance evaluation but is not as readily available as trending and x-y
plots.
Below are a few examples of using scatter plots to identify valve problems.
Example 8.8 Normal control valve characteristics. Figure 8.9 shows a PV vs. OP
scatter plot for a properly tuned PID control loop of flow. A good correlation exists

Fig. 8.9 Scatter plot of PV verses OP for a normal PID loop

298 8 Methodology of Control Problem Troubleshooting

between the PV and OP, and the line is almost straight over the normal operating
range of the valve openings.

It is also observed that the operating range of the valves is between 30% and 60%. If
the data represents the normal operating condition, the valve may have been a little
oversized.

Example 8.9 Sticky control valve. Figure 8.10 depicts the real-time trending of the
setpoint, process value, and output of a flow controller with a sticky valve. The
setpoint (flat line) and process value are at the upper part of the chart and the output
at the bottom.
The trending plot shows that the controller output is ramping up and down, but the
process value does not follow the valve until the output change reaches a threshold.
When the valve suddenly starts to move, the output change has already been too
excessive, and causes the process value to jump to the other side of the setpoint,
which immediately drives the output to move in the opposite direction.
Figure 8.11 is another example of a PID loop with a sticky valve. The trending
plot shows that the valve stickiness is 8–9%, too much for a PID controller.
Plotting controller PV against its output in a scatter plot (xy-plot) immediately
reveals the poor performance of the control loop. See Fig. 8.12 and compare it to the
normal loop performance shown in Fig. 8.9.

Example 8.10 Valve Slip. A slippery valve can significantly degrade the control
performance. For example, in a crude propylene oxide (PO) column in a styrene

Fig. 8.10 PID data trending: sticky valve

8.5 Troubleshooting Tools 299

Fig. 8.11 PID data trending: severe stiction

Fig. 8.12 PID data trending: poor performance

monomer and propylene oxide (SMPO) plant, the reboiler steam flow loop has infe-
rior performance, leading to instability of the unit operation. The unstable operation
of the crude PO column was believed to be the cause of the problems in polymeriza-
tion in a downstream column, which once led to a costly plant shutdown. Figure 8.13
shows the trend of the history data.
The plot shows that the flow has been spiky by as much as 7%, causing unstable
column operation (period ➀). The flow loop was temporarily switched to manual
(MAN) mode for troubleshooting purposes, and the flow immediately becomes more
300 8 Methodology of Control Problem Troubleshooting

Fig. 8.13 PID data trending: slippery valve

steady. A few step changes on the valve output were performed, and the PV/OP data
revealed that the valve has difficulties positioning itself. Valve is also found to be
significantly oversized (period ➁). When the loop was put back to the automatic
model (AUTO), the spiking in flow re-appeared immediately. Consultation with the
instrumentation engineer confirmed that the valve had difficulties positioning around
45% (period ➂) due to wear and tear.
The ultimate solution for the valve slip problem was to take the valve offline and
repair it. However, the valve could not be serviced without a unit shutdown, and the
next turnaround was still one year away. As a temporary workaround, the controller
was de-tuned with integral time from 16 to 45 s. The bypass valve was also adjusted
so that the control valve would stay away from 45–50% opening (see Fig. 2.9).
The resulting performance is substantially improved (period ➃). Although not
optimal, the control performance is sufficiently well, awaiting the next plant
turnaround.

8.5.2 Control Performance Monitoring Tools

A properly designed, implemented, and maintained process control solution is critical

for safe and efficient plant operation. Once a plant is built and running, the challenge
is tuning the control performance to the optimal level and sustaining it. For this
8.5 Troubleshooting Tools 301

reason, a control performance monitoring (CPM) system is becoming increasingly

popular.
The CPM system provides a convenient, continuous, and reliable way of moni-
toring the performance of the control loops in operation. The objective of loop per-
formance monitoring is to improve the control performance through the following
steps:
Monitor ⇒ Diagnose ⇒ Improve. (8.2)

Proactive monitoring and troubleshooting are arguably more effective and efficient
than reactive maintenance to achieve or retain higher control performance. A good
CPM software application is valuable to help this transition from reactive to proactive
maintenance.

Monitoring

Loop performance monitoring provides proactive detection and automated reporting

of the loop performance issues. However, control loop performance has been an
active area with many theories and practices but still lacks standard definitions or
implementations. The practical requirements include at least the following:
1. Provide a statistical summary of the control loop status at any operational level
and time period. For example, what percentage of control loops were running in
their intended mode for a particular unit during the last month or the last eight
hours? What is the daily up-time trending for all the controllers in the entire plant
for this month? These are crucial information for management.
2. Detect the poorly performing control loops, or “bad actors”, generate a “bad actor
report.” The bad actor report is a list of the worst-performing loops automatically
generated by the system. It is typically sent to the mailbox of the responsible
engineers for troubleshooting and improvement. As the worst-performing loops
are fixed, the less severe ones bubble up and make into the bad actor report. Over
time, the overall control performance is improved. Consequently, chasing the bad
actors is a systematic and sustainable approach to improve loop performance.
The loop performance is usually represented as intuitive and reliable KPIs. For
example, the statistical information may include the following KPIs:
1. Availability: A controller may not be in operation due to operational reasons but
can be turned on if needed. For example, all the controllers on standby equipment
are not actively controlling but are in ready-to-run condition, and thus, they are
considered available.
2. Up-time: A controller running in its intended mode is called in-control. A con-
troller in manual mode is not in control, while a controller in automatic mode or
cascade mode is. The percentage of time a controller is in control is called the
controller up-time (see Fig. 3.10 for PID controller modes).
302 8 Methodology of Control Problem Troubleshooting

3. Compliance: A controller that keeps the control error within the desired range
is defined as complying with the control performance requirement. There are
many ways of defining the compliance standard; a simple error band is the most
common.
The definitions of loop performance differ between simple standalone PID con-
trollers, complex PID control schemes, and model predictive controls. The control
performance index (CPI) has been a popular KPI for simple PID controllers, com-
paring the closed-loop response with the ideal minimum-variance control algorithm.
A good CPM tool is expected to include most of the following functions:
1. Proactively and continuously calculate the key performance indicators (KPI) for
all control loops.
2. Detect any performance issues based on pre-defined performance criteria.
3. Support quick “roll-up” (aggregation) of the KPIs to any operational levels, from
individual loop to a particular unit to the entire plant, and for any selected period.
4. Provide a convenient web-based interface to allow on-demand access to the var-
ious KPIs by any authorized user from anywhere and at any time.
5. Generate customized performance reports and automatically deliver them to the
email or mobile device of the stakeholders or responsible persons. The report can
be a high-level statistical view of the control loops for operation management or
an exception-based report containing bad actor lists for process control engineers.
6. Provide relevant details (“drill-down”) on the poorly performing control loops to
facilitate root cause analysis.

Diagnosis

The diagnosis part of a control performance monitoring (CPM) tool provides accurate
information and analytical tools to facilitate the root causes analysis for the problems
detected and reported. For example, some software can pinpoint the problem to valve
stiction or measurement error, while most software offers cross-correlation analysis
to help isolate process variable oscillations. CPM software typically provides the
following diagnostic information at the individual control loop level:
1. PID loop dashboard to show the running status and performance information,
highlighting the detected problems.
2. Single-point access to the critical configuration information, performance history,
and other dependent information to facilitate the root cause analysis.
3. Access to all the required loop configuration information.
4. Access to the change history of key PID configuration parameters, including the
controller mode, setpoint, tuning parameters, and filter value.
5. Online access to the relevant P&IDs, PCN, and operating manuals.
6. Convenient analytical tools such as trending charts, cross-correlation analysis,
and essential dynamic modeling capability.
8.5 Troubleshooting Tools 303

Improvement

The ultimate purpose of the monitoring and diagnosis is to improve the control loop in
question. Improvement is a highly complex process and can follow the same method-
ology and procedure for process analysis (Chap. 6) and control design (Chap. 7).

8.6 Summary

Depending on the complexity of the control scheme, the area of attention can also
be quite different. For simple standalone PID controllers, the focus is on the loop
dynamics, i.e., the five essential components of a PID loop. However, for complex
control schemes, the troubleshooting should start with a good understanding and
careful validation of the overall control design, the interaction between the different
controllers, and the integrity of the control scheme configurations.
Troubleshooting is typically the most challenging work demanding the highest
expertise. It requires that the troubleshooter not only understands how the control
scheme work but also knows how it may fail.
Some general guidelines and best-practice procedures are discussed in this chapter.
The basis is the knowledge presented in Chap. 3 for simple PID loops and Chaps. 4
and 5 for complex and advanced control schemes. The process analysis skills and
control design methodologies in Chaps. 6 and 7 provide the methodology and guide-
lines.
When troubleshooting a control loop, the best support is typically from the opera-
tors. As a process control engineer, it is critically important to maintain a good work
relationship with the panel operators. The operators can make the troubleshooting
effort very enjoyable, and they can also make it miserable.

References

Åström KJ (1999) Automatic control - the hidden technology. In: Frank PM (ed) Advances in
control - highlights of the ECC’99. Springer
Brannan C (2018) Rules of thumb for chemical engineers - a manual of quick, accurate solutions
to everyday process engineering problems, 6th edn. Elsevier
CSHI (2005) Report no. 2005-04-I-TX: Investigation report – refinery explosion and fire. Techni-
cal report, U.S. Chemical Safety and Hazard Investigation Board. https://www.hsdl.org/?view&
did=234995
Euro-Control (2006) Revisiting the Swiss Cheese model of accidents. Technical report, EuroControl
Experimental Center
King M (2016) Process control: a practical approach, 2nd edn. Wiley
Luyben WL, Lyuben ML (1997) Essentials of process control. McGraw-Hill
Marlin TE (2015) Process control - designing processes and control systems for dynamic perfor-
mance, 2nd edn. McGraw-Hill
Chapter 9
Control Loop Tuning and Improvement

There are many ways to tune a PID controller. PID loop tuning starts with a realistic
expectation of the closed-loop control performance to achieve (Sect. 9.1). Trial-and-
error and empirical methods (Sect. 9.2) were popular during the early days, but they
rely heavily on the practitioner’s experience and patience and the results are often
hit-and-miss. Model-based PID tuning methods (Sect. 9.3), although a little more
involved, significantly reduce the uncertainties in the PID tuning process and are
gaining increased popularity to become the preferred methods. The work process
of the model-based tuning method can be simplified for some control loops such as
flow loops (Sect. 9.4) and vessel level control loops (Sect. 9.5), where the process
gain can be calculated via some “shortcut” methods to save time and effort.

9.1 PID Control Performance

Different applications have different expectations of the ideal PID performance. The
optimal PID response in robotic control may be unacceptable in industrial process
control and vice versa. The desired control performance is also severely limited by the
inherent weakness of the PID algorithm. Therefore, control performance is always
a trade-off between desire and practicality.

9.1.1 Performance Criteria

The control performance refers to the closed-loop response (Fig. 9.1), typically
gauged by setpoint tracking and disturbance rejection.
The quantitative measurements of the dynamic response characteristics of the
closed-loop response typically include the following:
1. Dead time: After a control action is taken, the time it takes for the process output
to respond.
2. Overshoot: How much the peak level is higher than the steady-state, normalized
against the steady-state value.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 305
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_9
306 9 Control Loop Tuning and Improvement

Fig. 9.1 Closed-loop transfer function

3. Steady-state error: The difference between the steady-state value and the desired
value.
4. Settling time: The time it takes for the system to converge to its steady-state
(within 2 or 5% of the final steady-state value).
5. Rise time: The time it takes for the process output (controlled variable) to reach
the new steady-state value (typically 95%).
6. Peak time: The peak time is the time required for the response to reach the first
peak of the overshoot.
In setting the control objectives, a good balance should be maintained between
performance and simplicity. Simple often means reliable, and reliability is far more
important than performance in the process industry. Ideally, we would like the closed-
loop response to meet the following expectations:
1. Closed-loop stability: The closed-loop response must be stable.
2. Disturbance rejection: The effects of disturbances need to be minimized.
3. Setpoint tracking: Fast and smooth response to setpoint changes.
4. Steady-state error: There should be no steady-state offset.
5. Control actions: There are no excessive control actions.
6. Robustness: The control design is insensitive to a wide range of changes in process
dynamics and operating conditions.
PID control performance tuning is achieved by the proper setting of the three
parameters: proportional gain (K c ), integral time (Ti ), and derivative time (Td ). How-
ever, due to the simplicity of the three-parameter formula, it is often difficult to
achieve the desired best performance. The tunings that provide the fastest setpoint
tracking are the settings most likely to cause instability if the loop dynamics change.
PID tuning thus always seeks a compromise between performance and robustness
instead of performance only. Some processes must not allow an overshoot of the
9.1 PID Control Performance 307

process variable beyond the setpoint if, for example, this would be unsafe. Other
processes may require that the manipulated variable be limited to save energy.
Mathematically there are many criteria for deriving the tuning parameters and
measuring the controller performance. We will not go into any details on these
criteria. For industrial process control, most of the time, the desired PID control
response has the following characteristics, similarly to line ➁ in Fig. 9.2:
1. Stable and smooth response in the controlled variable, with no overshoot or slight
overshoot.
2. Relatively fast response to reject disturbances.
3. Relatively smooth response in the manipulated variable.
Quarter wave decay (Line ➂ in Fig. 9.2) has long been accepted as the most popular
performance criteria for PID response performance because of the good compromise
between oscillation and settling time. However, this response performance usually
is unacceptable in the process industry due to concerns of stability and overshoot.
Sluggish response (Line ➀) is not acceptable either. The desired response is quick
and smooth, with a small (maximum 5%) overshoot (Line ➁).

9.1.2 Practical Considerations

Control performance can be affected by many factors such as noises and disturbances,
nonlinearity and time variation in process dynamics, uncertainties in the final control
elements, control path changes due to branching and merging of control signals. Poor
PID tuning is not the primary cause of bad performance.
Because of the industrial process’s inherent nonlinear and time-varying nature,
there is no way to tell whether a given set of tuning parameters is the best. The
best tuning today may become sub-optimal tomorrow or next week. Therefore, it is
impossible, nor is it necessary, to reach the so-called optimal tunings. The tuning only

Fig. 9.2 PID closed-loop response

308 9 Control Loop Tuning and Improvement

needs to provide decent and robust performance within a sufficiently large operating
range. Any effort to achieve an improvement from 90% satisfaction to 100% is a
waste of time.

9.2 Classical PID Controller Tuning Methods

Proper controller tuning is the quickest, easiest, and least expensive process control
improvement to reduce process variability. As a result, most inexperienced engineers
are often eager to jump to PID tunings when faced with a control performance issue.
The loop dynamics dictate the closed-loop transfer function instead of the process
dynamics alone. All the data flow components contribute to the loop dynamics (see
Sect. 3.3.2). The process and control valves are the dominant factors affecting the
loop dynamics, while other components such as the analog/digital converters and
transmitters may only have insignificant and relatively deterministic dynamics. Other
factors such as branching and merging the data flow through selectors and splitters
may significantly impact the loop dynamics.
The closed-loop transfer function is mainly determined by the process dynamics
G p (s) and the controller dynamics G c (s) for a simple standalone PID loop. The
desired closed-loop transfer function G cl (s) is given by

Y (s) G c (s) G p (s)

G cl (s) = = (9.1)
Ysp (s) 1 + G c (s) G p (s)

and the desired response characteristic for this closed-loop transfer function (line ➁
in Fig. 9.2) is achieved by adjusting the two (for PI) or three parameters (for PID) in
G c (s), which is the so-called PID tuning process.
Due to the complexity of the process dynamics and the added dynamics of all the
components in the control loop, the loop dynamics are usually complex and non-
deterministic. Thus, achieving the desired response performance with merely two or
three tuning parameters in the PID algorithm is challenging. Consequently, all the
tuning principles are just approximations (Smith 2009).

9.2.1 Typical Tunings

A good starting point for PID tuning is to have a general idea about the tunings’
ballpark values. In the process industry, by following the best practice design, the
different process variables often have a consistent range of values for the process
gains and response time. These are the natural results of the governing physical
laws. For example, the typical residence time for separators/accumulators are from 3
to 20 min (see Brannan 2018; Luyben and Lyuben 1997, and also API12J standards).
The residence time is achieved through the vessel and flowline sizing, which dictates
9.2 Classical PID Controller Tuning Methods 309

Table 9.1 Typical values for process gains and controller tunings
Process type Gain (%/%) Response time (min) Controller gain (%/%) Integral (min) Derivative (min)
Flow 1.0−5.0 0.25−1.0 0.3 0.5 0.0
Temperature 0.5−2.0 0.50−5.0 1.5 5.0 1.0
Pressure/liquid 1.0−5.0 0.25−1.0 0.5 0.5 0.0
Pressure/vapor 0.1−2.0 0.50−2.0 2.0 5.0 0.0
Level 1.0−5.0 5.0−30 1.0 10.0 0.0
Quality 1.0−5.0 0.5−30 0.3 40.0 0.0

the process gain and time constant (see Sect. 9.5.2). Table 9.1 lists the typical range of
values for the process gains and response time for typical process variables (Blevins
et al. 2003; Ford 2013; Svrcek et al. 2006).
The typical PID tuning parameters can be obtained in line with the typical process
gain and time constants and are shown on the right side of Table 9.1. These values
are on the conservative side and will generally result in a sluggish response. They
can serve as the initial tunings to start a green-field plant if nothing is known about
the loop and no test can be performed. They are by no means the final acceptable
tunings (Brannan 2018; Ford 2013). The tunings must be improved with a proper
tuning method later on.
These initial tuning parameters can also serve as an order-of-magnitude sanity
check on existing PID tunings. For example, an average level control loop would
expect an integral time in the order of minutes rather than seconds. If a level loop
has an integral time of 10 s, it definitely raises a flag for further investigation.
Control systems all have their system default values for PID tunings. For example,
in the DeltaV system, the system default for PID controller gain is 0.25 (%/%), and
the integral is 10 s. However, these values are not appropriate for most applications
and must be modified before putting the control loop online.

9.2.2 Tuning by Trial and Error

Tuning by trial-and-error is a “tune by feel” method and has been widely used in
practice. Although quick, convenient, and sometimes effective, it has been proven
time and again to be inefficient, inaccurate, and often dangerous. This method should
be avoided whenever possible.
Below we give two examples showing where the trial-and-error method can be
successful for one problem and ineffective and dangerous for another.

Example 9.1 Oscillating PID performance. A typical performance problem is oscil-

lations in both process value and controller output while the setpoint is constant.
This oscillation typically has a sinusoidal pattern with a fixed frequency or period.
Figure 9.3 shows the historical trends of the process value and control output of a
flow controller. The clear sinusoidal pattern suggests that the swing is due to the PID
310 9 Control Loop Tuning and Improvement

Fig. 9.3 Oscillating loop due to too strong integrating action

controller’s excessive control action. Is it because the controller gain is too large or
the integral action too strong?
One further indication is the phase shift. The integral term in the PID controller
(Eq. 3.8) introduces a 90◦ phase shift between the process value and the controller
output, while the proportional action does not cause a phase shift. A PI controller has
a combination of both, and consequently, the phase shift is typically between 0 and
90 ◦ C. Based on the phase shift (if pronounced), we can often determine whether the
oscillation is caused mainly by too strong proportional action or too strong integral
action.
In the example of Fig. 9.3, the process value and the controller output have been
swinging with a 30 min cycle. Process value leads (due to closed-loop control) the
control output by about 6 min, i.e., 1/5 of a cycle. Since a complete cycle has a phase
shift of 360 ◦ C, the 6 min phase shift between the process variable and controller out-
6
put is equivalent to 360 × = 72◦ , which suggests that the oscillation is primarily
30
caused by too strong integral action.
Increasing the integral time from 180 to 1200 s (larger value means less strong
integrating action) improved the control performance. The amplitude of the oscilla-
tion decreases almost immediately, as shown in Fig. 9.4.
The loop performance can be further improved by increasing the integral time
or reducing the gain. However, many more trial-and-error rounds may be needed
before reaching the final satisfactory tunings. It is also possible that the satisfactory
performance may never be reached because the criterion for “satisfaction” is personal
and subjective. Since the new tuning parameters are based purely on data patterns via
trial-and-error, the confidence in the new tunings is low, and a temporary deviation
from the “good” performance may trigger another round of trial-and-error tuning.
9.2 Classical PID Controller Tuning Methods 311

Fig. 9.4 Oscillating loop after tuning change

Now let us look at an extreme case of misuse or abuse of this trial-and-error method.

Example 9.2 Failure example of the trial-and-error tuning method. Figure 9.5 shows
the tuning history of two control loops from a petrochemical plant, one pressure loop,
and one temperature, with strong interactions between each other. Both are critical
control loops on a high-pressure steam system in the utility plant.
The list shows the changes made to the two controllers’ tuning parameters over
one year:
1. Many tunings were tried, possibly by different engineers, with no success. Most
of the tuning changes seem to be following the trial-and-error approach since
there was no sound basis for any of the tuning parameters experimented. Little
trust was given to any of them, and thus they were abandoned and replaced easily.
“Easy come easy go!”
2. Some of the changes were very drastic without a clear rationale to support the
decision. For example, on Feb 10, 2009, the integral time for loop PC-811 was
changed from 3800 s to 60 and the proportional band from 16.1 to 50.1.
3. On the other hand, some of the changes are so small that they should not have any
noticeable impact on the control performance to worth the change. For example,
the proportional band was changed from 101 to 96, then to 91, then to 96, 91, 81,
76, 91, and 81. This pattern of change indicates the lack of confidence in the new
tuning parameters.
Unfortunately, the final tuning change on loop TC-811 coincided with a unit trip
that caused an expensive unscheduled shutdown.
312 9 Control Loop Tuning and Improvement

Fig. 9.5 Failure example of trial-and-error tuning

Table 9.2 Loop dynamics Loop dynamics Control gain Integral Derivative
and controller tunings
Kc Ti Td
Loop gain K p  
Delay/lag ratio θ/τ    
Delay θ    

Table 9.3 Controller tuning versus controller response

PID tuning Rise time Overshoot Settling time Offset Stability
Gain K c    –  
Integral Ti     ↓ 
Derivative Ti  –   – 

The general qualitative indication between the loop dynamics and the controller
tunings are given in Tables 9.2 and 9.3. More discussions along this line can be found
in Smith (2009).
9.2 Classical PID Controller Tuning Methods 313

Table 9.4 Ziegler–Nichols tuning method

Parameter Symbol P-only controller PI controller PID controller
Gain Kc K u /2.0 K u /2.2 K u /1.7
Integral Ti Pu /1.2 Pu /2.0
Derivative Td Pu /8.0

9.2.3 Ziegler–Nichols Method

There are many empirical tuning methods in use, with variable levels of success.
Most tuning methods developed in the early days had focused on fast closed-loop
response, that is, on how to track the setpoint changes as fast as possible.
The Ziegler–Nichols method, or simply the Z–N method (Ziegler and Nichols
1942), is the best known empirical tuning method. It has had a significant influence on
the practice of PID control. The Z–N method can be applied as follows: under P-only
control, gradually increase the controller gain K c until the control loop is in sustained
oscillation. The corresponding controller gain K u is called the ultimate gain, and the
process oscillation period Pu is called the ultimate period. The tuning parameters are
then calculated from the ultimate gain and period, as shown in Table 9.4.
Although popular and with a proven record in other industries, the Z–N method
is rarely used in the process industry for many reasons:

• The controlled variables in the process industry are often interacting with each
other. Putting the controlled variable to sustained oscillation could lead to unsafe
operation to many loops and even an entire unit.
• The method is based on a quarter-wave-decay tuning objective and thus produces
aggressive control tunings that often lead to process upsets.
• The significant overshoot in the controlled variable is typically not allowed because
many controlled variables are already operated close to the operating limit, and
overshoot could drive the controlled variable over the limit.

The Z–N method is limited to processes with relatively small process delays (com-
paring to process lag time) and integrating processes such as level control (Ziegler
and Nichols 1942).

9.3 Model-Based Tuning

The trial-and-error method is simple and convenient because it does not require much
knowledge about the underlining process dynamics. On the other hand, this lack of
insight into the process dynamics is the exact reason that little confidence is placed
in the tuning result.
314 9 Control Loop Tuning and Improvement

Model-based tuning is based on a sufficient understanding of the cause and effect

dynamic relationship between the manipulated variable (e.g., control valve) and the
controlled variable. A mathematical model is used to represent this relationship, and
this tuning method is thus called model-based controller tuning.

9.3.1 Model-Based Tuning Concept

The generic approach of model-based tuning can be simplified as

Model Identification Tuning Rules

Data −−−−−−−−−−→ Model −−−−−−→ Tunings.

Compared with the trial-and-error or empirical tuning methods, the model-based

approach has a sound theoretical basis. As a result, much higher trust and confidence
can be placed in the resulting tuning parameters. It might require more effort and
self-discipline to reach the tuning parameters, but overall it is more efficient and less
time-consuming because it needs less re-work later.
The general model-based PID tuning involves three simple steps:
1. Identify the process model. There are many ways to find the process model,
depending on the process type, model type, and personal preferences.
2. Calculate the tuning parameters. Based on the model parameters, apply a favorite
tuning rule to find the controller parameters.
3. Apply the new tunings. Apply the new controller parameters online and make
tuning adjustments as necessary.
A detailed tuning procedure in block diagrams is shown in Fig. 9.6.

9.3.2 Step Testing

Model-based tuning starts with a good process model. The model is typically derived
from credible input/output response data obtained through step tests . One or more
carefully planned step changes are applied to the manipulated variable and cause the
controlled variable to respond. The dynamic response reveals the process dynamics
for calculating the process model. A general procedure is recommended as follows:
1. Based on process knowledge and history data, obtain a guesstimate of the gain
and time constant of the loop dynamics. Then decide on the step size and switch
time of the step changes.
2. Explain to the operator the intent of the step test. Confirm with the operator that
the proposed move size, move direction, and switch time are acceptable.
3. Put the control loop in MAN and let the process value stabilize. If loop output has
been swinging, wait for the valve to come to its average value before opening the
loop.
9.3 Model-Based Tuning 315

Fig. 9.6 Model-based PID

tuning procedure
316 9 Control Loop Tuning and Improvement

4. Make at least two step changes in opposite directions, with switch time about the
same as the settling time. For noisy loops, make at least four-step changes (two
up’s and two down’s), two moves with a switch time of one settling time, and two
moves with about half the settling time.
5. Watch and record any valve problems such as stiction, deadband, and slippage.
Remember that the control calculation is based on the loop dynamics rather than
the process dynamics, so the step changes should be applied from the controller
output rather than the final control element. See Fig. 3.11 for the location of the
control output and the final control element.
A single step change on the manipulated variable, and its response, is shown in
Fig. 9.7.
The data quality directly determines the model quality and thus the control per-
formance. Some disadvantages associated with this approach include the following:
1. Safety concerns: Running the process in an open-loop condition always risks
losing control to a significant process upset during the test. Close monitoring is
thus required for open-loop testing.
2. For a nonlinear process, the test results may be sensitive to the step change’s
direction and magnitude. The starting point, the step size, and the direction are
all important considerations.
3. For an open-loop unstable system such as a level loop, it is possible to perform
open-loop testing but must be doubly careful in the preparation and execution.
For a process with significant noise or disturbances, multiple step changes can
minimize the influence of noisy measurement and drifting steady-state, as illustrated
in Fig. 9.8. The modeling can be accomplished with dedicated identification tools.

Fig. 9.7 Step response of a stable process

9.3 Model-Based Tuning 317

Fig. 9.8 Step testing with multiple step changes

The general guideline is that the step size should be large enough to be outside the
noise band (typically at least five times larger than the noise band, or 1∼2% of the
measurement range). On the other hand, the step size should not be too large to drive
the process to an unstable or undesirable operating region, e.g., nonlinear region, or
trigger an alarm. The first step should be made in the direction of safer operation.
For example, to test a reflux flow, step up the flow valve first.

9.3.3 Model Identification

A process model is a mathematical description of the cause and effect relationship

between the process input and output. In other words, with the same process input
u(t), we would like to see that the model produces an output ẑ(t) that is identical,
or as close as possible, to the process output z(t). See Fig. 9.9. The technology to
extract the model from process data is process identification or process modeling.

Fig. 9.9 Process and process model

318 9 Control Loop Tuning and Improvement

Many types of models can be used to describe a given process. For loop tuning,
the most widely used is the simple first-order-plus-time-delay (FOPTD) model:

K p −θ s
G p (s) = e (9.2)
τs + 1

where K p is the process gain, τ the time constant, and θ the time-delay (or dead-time).
The process model is typically obtained in one of the two methods. Visual inspec-
tion and manual calculation are sufficient for a process with a clean response (well-
behaved and low noise). For more complex dynamics with noisy data, dedicated
process modeling software tools should be used.
Figure 9.10 shows how to derive a process model from the dynamics response to
a single step change. Although the signal is not noise-free, its trend is clean enough
to calculate the three parameters manually.
The process gain (normalized) is calculated as

CV/CVspan CV2 − CV1 MVspan

Kp = = · . (9.3)
MV/MVspan MV2 − MV1 CVspan

The time delay and time constant are obtained as shown in the time response. As
shown in Sect. 3.3.2, the PID tuning is based on the loop gain, not the process gain.
In other words, the MV in the gain calculation in Eq. 9.3 should be the controller
output (OP), and the CV should be the controller PV. In other words, the step changes
should be applied to the controller output (in manual mode), not at the valve. The
rationale of this requirement is evident when performing step testing on a cascade
control loop.
Suppose other components on the data flow path do not have significant dynamics.
In that case, the difference between the process dynamics and the loop dynamics may
be negligible, and either can be used for the gain calculation.

Fig. 9.10 Step response of a stable process

9.3 Model-Based Tuning 319

Fig. 9.11 Step response of a ramp process

Figure 9.11 illustrates the step response of a ramp process (e.g., level). The process
gain (normalized) for a ramp process is calculated as follows:
   
CV CV
−
CVSlop/CVspan t 2 t 1 MVspan
Kp = = · . (9.4)
MVSlop/MVspan MV2 − MV1 CVspan

When the above-discussed simple modeling methods are not feasible (e.g., for
step test results from multiple step-moves), dedicated modeling tools should be used.
The model identification step often needs to go back and forth several times before
arriving at a satisfactory model. Process knowledge is critical in deciding if the model
is satisfactory.

9.3.4 Tuning Parameter Calculation

The tuning parameters are calculated based on the process model. The calculation
from model to tuning parameters is called the tuning rules and can be empirical or be
based on sophisticated theoretical derivation. For convenience, the tuning rules are
often simplified into a look-up table. Once the process model is known, the tuning
formula is conveniently looked up from the tuning table.
There are many established tuning rules. Some rules may be better than others
for different loops, but often, which rule to use is merely a personal preference.
Among model-based tuning, the internal model control (IMC) (Garcia and Morari
1982; Garcia et al. 1989; Rivera et al. 1986)-based methods are becoming prevalent
because of their sound theoretical basis and field proved track record. The IMC
structure is a transparent framework for control design. It makes it easy to understand
how the process characteristics (such as time delays and unstable transfer function
320 9 Control Loop Tuning and Improvement

components) affect the inherent controllability of the process. The IMC-based tuning
procedure results in an intuitive single tuning parameter (Garcia et al. 1989).
The tuning parameters calculated with any tuning rules only give a starting point.
Adjustment and fine-tunings based on process responses and control objectives are
almost always a necessary next step:
1. Modified Cohen–Coon method: The modified Cohen–Coon method is based on a
first-order-plus-time-delay (FOPTD) model representation of the process dynam-
ics. Once the model parameters K p , τ, and θ are known, the tuning parameters
can be calculated with the formulas (tuning rules) in Table 9.5.
Although the modified Cohen–Coon method is model-based, it uses the quarter-
wave-decay response as the optimal tuning criteria, and the resulting performance
is typically less stable than expected with excessive overshoots, thus not widely
used in the process industry.
2. IMC-based tuning and λ-tuning rules: Internal model control (IMC) for PID tuning
has been one of the most popular model-based tuning rules (Rivera et al. 1986).
The IMC tuning rules have proven to produce robust performance when used in
the control of common processes.
It is beyond this book’s scope to go into the detailed theory and derivations of
IMC. We list the resulting tuning rules and show them in Tables 9.6 and 9.7 for
easy reference (Bequette 2002; Smith 2009). These tuning rules are based on a
first-order-plus-dead-time (FOPDT) model, sufficient for most processes. More
sophisticated tuning rules should be followed for truly high-order or nonlinear
dynamics, and many references are available (Bequette 2002; Chien and Freuhauf
1990; King 2011; Morari and Zafiriou 1989; Rivera et al. 1986).
The λ value can be considered the closed-loop time constant and is a user-selected
value. For this reason, this tuning method is also called λ-tuning. The value is
typically set to be one to three times the process dead time θ . λ = 3τ can be used
to achieve a very stable closed-loop response. Set the λ as low as τ for a faster
response.
Compared with Ziegler–Nichols and Cohen–Coon tuning rules, the IMC-based
tuning rules have some clear advantages as follows:
• The closed-loop response exhibits a first-order dynamic response, and the con-
trolled variable has no or minimum overshoot. The process industry welcomes
this type of response.
• The tuning is robust and is less sensitive to changes in process dynamics than
other tuning methods.
• It is more intuitive since the user can specify the desired closed-loop time
constant, which serves as the only tuning parameter required.
The drawback is that the resulting controller tuning may produce a large integral
value for slow process dynamics, resulting in a slow recovery from disturbances.
The reason is that the controller’s integral time is assumed to be the same as the
process’s time constant.
9.3 Model-Based Tuning 321

Table 9.5 Modified Cohen–Coon method for PID tuning

Parameter Unit PI controller PID controller
Gain K c (%/%) 0.45  τ  0.67  τ 
· + 0.092 · + 0.185
Kp θ Kp θ
   
τ + 0.092 θ τ + 0.185 θ
(min) 3.33 θ · 2.5 θ ·
Integral Ti τ + 2.22 θ τ + 0.611 θ
–  
Derivative Td (min) τ
0.37 θ ·
τ + 0.185 θ

Table 9.6 The IMC tuning rules for stable process

Parameter Unit PI controller PID controller
with λ ≥ max(0.2, τ, 1.7 θ) λ ≥ max(0.2 τ, 1.7θ)
1 τ 1 τ + θ/2
Gain K c (%/%)
Kp λ+θ Kp λ+θ
θ
Integral Ti (min) τ τ+
2
τθ
Derivative Td (min)
2τ +θ

Table 9.7 The IMC tuning rules for ramp process

Parameter Unit P-Only controller PI controller PI controller PID controller
√ √
with λ > 0, θ = 0 λ ≥ 10 θ λ ≥ 10 θ
1 1 1 2 1 2λ + θ 1 2λ + θ
Gain K c (%/%)
Kp λ+θ Kp λ K p (λ + θ )2 K p (λ + θ/2)2
Integral Ti (min) 2λ 2λ+θ 2λ+θ
λ θ + θ 2 /4
Derivative Td (min)
2λ+θ

9.3.5 Tuning Implementation

Once the tuning parameters are available, apply them to the PID controller with care.
The recommended procedure is recommended as follows:
1. After a new tuning is put online, make one or two setpoint changes to verify that
the control performance is adequate.
2. If the tuning changes are more than five times different from the existing tunings,
especially in the more aggressive direction, stop and double-check the new tunings
before implementation.
3. When making significant tuning changes, apply the changes in several steps,
especially in the more aggressive direction. Leave enough time in between to
observe the performance improvement.
322 9 Control Loop Tuning and Improvement

4. When making changes to both gain and integral time, always change the one that
will make the loop more stable (slower). For example, apply the gain change first
if the controller gain needs to be changed from 1 to 0.3 and the integral time from
60 to 40 s.
5. For a nonlinear process, identify the highest gain and the largest dead-time and
tune the control loop based on that worst-case scenario.
6. Unless in an emergency, never apply dramatic changes to a control loop without
being able to test and monitor its performance for some time. For example, never
apply a drastic tuning change to a loop at the last minute of the week and leave
for the weekend.

9.3.6 PID Self-Tuning

PID auto-tuning (or self-tuning) has been extensively researched and is available
in almost all major control systems. The so-called PID auto-tuner automatically
identifies the process model online and calculates the tuning parameters to keep
up with the changed loop dynamics. Typically the tuning procedure is on-demand,
where the operator initiates the tuning procedure via a user interface.
The popular self-tuning method proposed by Åström and Hägglund (1984) uses
a relay switch to create on/off changes in the manipulated variable to perturb the
process. The input/output data from the artificial perturbation is then used to infer
the process dynamics and calculate the PID tuning parameters.
Although the PID self-tuning function is available in most major control systems,
its acceptance is not as well as anticipated. One of the primary reasons is that self-
tuning is not fool-proof. It still requires a good knowledge of all the five control loop
components, a sound judgment on the condition suitable to start the self-tuning, and
proper validation of the resulting tuning parameters. Another reason is that the extra
perturbations added are disturbances to the normal operation that the operator may
not feel comfortable with. Blindly applying the auto-tuning function to all control
loops inevitably leads to many failures and hurts the credibility of auto-tuners. If a
controller cannot be tuned manually, it will not be tuned by the auto-tuner either.
Designing a fully automated and continuously running PID auto-tuning procedure
involves many critical issues, especially with data quality and model quality, and thus
remains an open challenge for research. A fully automated PID auto-tuner falls under
the category of adaptive control (Visioli 2006).

9.3.7 Best Practice of PID Tuning

In a nutshell, model-based PID tuning is a simple three-step procedure:

1. Identify the process model.
9.3 Model-Based Tuning 323

2. Calculate tuning parameters.

3. Implement the new tuning.
The critical requirement for applying model-based tuning is the practitioner’s self-
discipline to adhere to the best practice work process. A checklist for model-based
PID tuning is provided below for convenience:
1. Identify the process model:
a. Record the original tuning parameters for later reference and book-keeping
purposes.
b. Decide if alternative tuning methods are applicable for the loop to be tuned.
For example, flow loops may be tuned via the x-y plot of the history data
(Sect. 9.4). Level loops may be tuned based on vessel sizing information
(Sect. 9.5).
c. Review history data and check if the normal operating data has sufficient
information for modeling.
d. If step testing is deemed necessary, conduct open-loop testing of the loop
if possible. If it is too risky to operate the process in an open-loop, some-
times a closed-loop step testing can be performed instead. Closed-loop step
testing requires that the identifiability conditions be satisfied for the data to
be valid. Modeling identification in a closed-loop is more involved in both
theory and practice. The interested readers should consult books on process
identification, such as Ljung (1999), for more details.
e. From the step test data, determine the process gain, time constant, and delay.
2. Calculate tuning parameters:
a. Normalize the process gain to make it dimensionless, i.e., adjust the gain by
input/output signal spans.
b. Apply the tuning rules (e.g., IMC tuning) to obtain the controller gain (or
proportional band) K c , integral time Ti , and derivative time Td .
3. Implement the new tuning:
a. If the tuning is significantly different from the existing tuning, stop and
double-check the data and calculation. Apply the changes in multiple steps
to minimize risk and disturbance to operation.
b. Test the loop to see how it responds to small setpoint changes in both direc-
tions. Also observe the controller’s response to noise and disturbances. Make
fine-tunings as necessary until satisfactory.
c. If the new tuning is consistently worse than the original tuning, restore the
original tunings, and re-validate the tuning procedure and calculations.
There are many empirical rules and best practices accumulated from practical
experience. One of them is the rule of five’s, which is provided as follows:
1. For step testing, the changes in PV step response should be at least five times
larger than the noise band.
324 9 Control Loop Tuning and Improvement

2. The PV filter constant for any control variable should be no larger than 1/5 of the
process’s time constant (or the controller integral time Ti ).
3. If the tuning change is more than five times different from the existing tunings,
especially in the more aggressive direction, stop and double-check the new tunings
before implementation.
4. Mass balance (inventory) control loops (e.g., level controls) should be tuned at
least five times as slow as energy balance control loops (e.g., temperature control).
5. For a cascaded control loop or a nested control loop, the closed-loop time constant
for the outer loop should be larger than five times that of the inner loop.
6. For PID controller tunings, the derivative action, if activated, should be no more
than 1/5 of the integral action. That is, Ti /Td > 5.
7. The positioner in a smart device (field bus) is a control loop by itself. Any control
loop cascaded to the positioner should be at least five times slower.

9.4 Tuning Flow Control Loops with History Data

When the setpoint of a flow control loop has frequent setpoint changes, the gain for
the flow loop can often be conveniently estimated using normal closed-loop operating
data without needing step tests. One good example is the flow control loop in a level-
to-flow cascade control. The outer level loop produces the setpoints for the inner flow
loop with continuous changes. The PID tuning for flow control loops can thus be
simplified. The general model-based tuning procedure discussed in Sect. 9.3 should
be followed for difficult flow loops with delays and longer time constants.

9.4.1 Flow Loop Dynamics

Under particular circumstances, the flow control loop can be tuned with historical
data without step testing. However, the effectiveness of this approach depends on the
following assumptions and should be used with discretion:
1. Most flow control loops are so fast in response that the dead time is typically
negligible. That is, we can assume the delay equals zero.
2. The time constant of a typical flow loop is about 0.25–0.5 min. If the time constant
is much larger than this, then this method is not recommended (see Table 9.1 for
typical process dynamics).
3. The process data meets the identifiability condition. For open-loop data, it is
required that there be sufficient output changes to reveal the process dynamics.
If using closed-loop data, there must be a sufficient number of setpoint changes,
such as the inner loop of a cascade loop, to meet the model identifiability criteria.
9.4 Tuning Flow Control Loops with History Data 325

9.4.2 Flow Loop Tuning Procedure

The recommended general procedure for flow loop tuning based on closed-loop
PV-OP data is provided as follows:

1. Identify the process model parameters K p , τ , and θ :

a. Select the period of data that the controller has sufficient changes in setpoint
SP (frequency and magnitude) and is free of saturation in the controller output
OP.
b. Draw a scattered plot of the controller PV versus OP.
c. Visually check the plot for valve problems such as stiction, dead zone, friction.
d. Estimate the normalized process gain:

PV/[PV Span]
Kp = . (9.5)
OP/[OP Span]

e. For normal flow controller, assume delay θ = 0, and time constant τ = 0.25
or 0.50 min. Note that if a PV filter is present, the filter value needs to be
added to the selection of τ .
2. Calculate the controller tuning parameters K c , Ti , and Td :
a. Decide on the desired closed-loop constant. Use λ = 0.5 min for the
inner/nested control loop and λ = 1.0 min for the standalone loop or the
primary loop in cascade control.
b. Using tuning rules in Table 9.6 to calculate the tuning parameters.
3. Apply the new tuning parameters:
a. Apply the new tuning parameters. If change is drastic, apply the change in
multiple steps.
b. Monitor and fine-tune the loop performance.

This “shortcut” approach can be demonstrated with a practical example.

Example 9.3 Flow loop tuning via PV-OP Plot. Consider the flow loop FC-303 in
Fig. 9.12, the inner loop of the level-flow cascade control.
Figure 9.13 shows the setpoint SP, process value PV, and controller output OP for
the flow control loop. The cycling is caused by the primary controller and provides
constant setpoint changes to meet the requirement for using this shortcut approach.
Although the data shown in the trending plot (Fig. 9.13) is quite oscillatory and
noisy, the PV-OP plot in Fig. 9.14 reveals that the valve-to-flow dynamics are pretty
decent at the normal operating range (30–80%).
Because of the loop’s fast dynamics, the time delay can be assumed to be zero, and
the time constant is assumed to be known and small, say 0.25 min. What is remaining
is just the process gain, which can be simply performed as the slope of the curve
in the PV-OP plot. A relatively linear range, e.g., from 40 to 80% valve opening, is
326 9 Control Loop Tuning and Improvement

Fig. 9.12 A three-phase separator with level-flow cascade loop

selected to calculate the process gain. This selected range should fall into the normal
operating window and be as wide as possible for better accuracy. The corresponding
flow is between 900 and 3,350 m3 /d. The process gain in engineering units is then
calculated as follows:
PV2 − PV1 3350 − 900
Kp = = = 81.5 m3 /d/%. (9.6)
OP2 − OP1 80 − 40

Tuning calculation is based on normalized (dimensionless) process gain. Knowing

that the PV range is 0–3,500 m3 /d, and the MV range is 0–100%, the normalized
process gain is given as

(PV2 − PV1 )/[PVSPAN]

Kp =
(OP2 − OP1 )/[OPSPAN]
(3350 − 900)/(3500 − 0)
=
(80 − 40)/(100 − 0)
= 2.33 (%/%). (9.7)

Now we have a flow process model with K p = 2.33 %/%, τ = 0.25 min, and
θ = 0. Also assume the desired closed-loop response time is λ = 0.5 min (cascade
loop), then from Table 9.6, we have
9.4 Tuning Flow Control Loops with History Data 327

Fig. 9.13 Trending the flow controller data

⎧
⎪
⎪ 1 2τ + θ 1 2 ∗ 0.25 + 0
⎨Kc = K
⎪
2 λ
=
2.33 2 ∗ 0.5
= 0.21 %/%
p
(9.8)
⎪
⎪ Ti = τ + θ/2 = 0.25 + 0/2 = 0.25 (min)
⎪
⎩T = 0 (min).
d

Example 9.4 Flow loop tuning. For loop FC-301, the time series plot is given in
Fig. 9.15. Again, it is seen that the flow loop oscillates due to poor level control
LC-0304. In addition to the setpoint changes driven by the level loop, the flow PV
and OP signals are also noisier than FC-303 in themselves.
A PV-OP plot for the flow controller, shown in Fig. 9.16, indicates that the flow
control valve does not perform well. There is excessive stick–slip with the valve
dynamics, and service is needed. Knowing that the flow range is 2,200 m3 /d, as seen
from Fig. 9.16, the valve might be oversized as well for its intended operation.
Nevertheless, to proceed with the tuning, let us pick two points within the normal
operating range, say, at 20 and 60% valve openings. The corresponding flow rates can
be approximately assumed to be 500 and 2,100 m3 /d, respectively. The normalized
process gain can then be calculated as
328 9 Control Loop Tuning and Improvement

Fig. 9.14 PV-versus-OP plot for a flow loop

Fig. 9.15 Trending the flow controller data (incomplete)

9.4 Tuning Flow Control Loops with History Data 329

Fig. 9.16 X-Y plot for a flow loop

(2100 − 500)/2200
Kp = = 1.82 %/%. (9.9)
(60 − 20)/100

Since the valve dynamic is not ideal, let us assume the time constant to be 0.5 min
and the delay to be 0. With a desired closed-loop time constant λ = 1.0 min, the
controller tuning parameters are then given, from Table 9.6, as
⎧
⎪ 1 2τ + θ 1 2 ∗ 0.5 + 0
⎪
⎪Kc = = = 0.27(%/%)
⎪
⎨ Kp 2λ 1.82 2 ∗ 1 + 0
θ 0 (9.10)
⎪
⎪ Ti = τ + = 0.5 + = 0.5 (min)
⎪
⎪ 2 2
⎩T = 0 (min).
d

The level control loops LC-301 and LC-304 cannot be tuned easily with history
data. However, for many level control, if the vessel sizing information is available,
there is also “shortcut” approaches to calculate the tuning parameters, as shown
below.

9.5 Tuning Level Control Loops with Vessel Sizing Data

Level control is critical in maintaining the inventory in the process operation. Unfortu-
nately, a vessel level loop is an inherently unstable process, and PID control parameter
tuning is thus also tricky.
Luckily, the process gain can often be calculated based on the vessel sizing infor-
mation and the flow rate. Consequently, the PID tuning parameters can be obtained
330 9 Control Loop Tuning and Improvement

based on the calculated gain. This level tuning approach is another shortcut method
for model-based PID tunings that can produce an accurate result without going
through the time-consuming procedure of step testing.

9.5.1 Level Loop Dynamics

A level loop is an integrating process rather than a self-regulating stable process.

Take Fig. 9.17 as an example.
Assuming we start with a perfect mass balance between inlet and outlet flow,
the vessel’s level stays constant. Now we increase the outlet flow; the level starts
to decrease. If the flow rate does not change, the level will fall until the vessel is
emptied. The vessel’s volume (vessel size) and the outlet flow rate determine the
level’s decrease rate. If we change the flow rate back to the original value, the level
will stop decreasing and stay constant. However, it does not come back to where it
was as the flow does since the vessel’s inventory has decreased.
A pure level loop has a transfer function of

Kp
G(s) = (9.11)
s
that is, only one parameter, the process gain, is in the model. For a pure integrating
process, the normalized process gain is calculated as
   
CV CV
−
CVSlop/CVspan T 2 T 1 MVspan
Kp = = · . (9.12)
MVSlop/MVspan MV2 − MV1 CVspan

For the level process in Fig. 9.17, we can calculate the gain based on the two data
segments ➀ and ➁. The flow change from segment ➀ to ➁ is 0–100% of the flow

Fig. 9.17 Level in a vessel

9.5 Tuning Level Control Loops with Vessel Sizing Data 331

range, while level change is from 100 to 0% over a period of T . Assuming the level
is H , and the flow is F, the normalized process gain is then given by

0− H 0 − 100%
−0 F −0 − 0 100% − 0 1
Kp = T · = T · =− . (9.13)
F −0 H −0 100% − 0 100% − 0 T

This T is the duration it takes to empty the vessel from 100 to 0% with the
maximum outlet flow (100%). This duration is also called the residence time, denoted
as Tr . The residence time is given as the control volume (0–100% level) divided by the
maximum volumetric flow rate (at 100% flow). The control volume can be calculated
with the vessel sizing information such as shape, height, and diameter. The maximum
flow rate can typically be found from the flow meter datasheet or valve datasheet.

9.5.2 Level Control Loop Tuning Procedure

Once the process gain K p is known, the level controller tuning can follow the generic
tuning rules in Table 9.7. A general procedure for level loop tuning based on vessel
sizing information is summarized as follows:
1. Calculate the vessel residence time based on vessel sizing information:
a. Calculate the control volume, V , of the vessel.
b. Find or calculate the maximum flow rate (flow range), F.
c. Calculate the residence time:

Control Volume V
Residence Time Tr = = . (9.14)
Maximum Flow Rate F
2. Compute the normalized process gain, which is the reciprocal of the residence
time (see Eq. 9.12):

1 1
Process Gain K p = = . (9.15)
Residence Time Tr

3. Calculate the controller tuning parameters based on the tuning rule in Table 9.7:
a. Calculate the recommended closed-loop time constant λ = Tr = 1/K p , or
choose a different λ value of choice.
1 1
b. Calculate controller tunings K c = · . Note that K c ≡ 2 if λ = 2 Tr is
Kp λ
used.
c. Calculate the controller integral time Ti = 2 λ.
4. Apply the new tunings online and fine-tune the integral time until satisfaction.
332 9 Control Loop Tuning and Improvement

Example 9.5 Level loop tuning with sizing information. Consider the level control
loop in Fig. 9.17. Assume that the vessel has an inner diameter of 3.0 m. The height
of the vessel is 11.5 m, with a tap-to-tap distance of 10.0 m.
To calculate the tuning parameters for the level controller LC-101, let us first
calculate the residence time. The vessel is a standard vertical cylindrical vessel. The
control volume, i.e., the volume between the two level-measuring taps, is given by
 2  2
D 3
V = A·H =π · ·H =π· · 10 = 70.7 m3 . (9.16)
2 2

The maximum flow rate is given as a 2500 kg/min mass flow, with a density
of 960 kg/m3 . Therefore, the maximum volumetric flow rate is 2500.0/960 = 2.6
m3 /min. The residence time is then given by

V 70.7 m3
Tr = = = 27.2 min . (9.17)
F 2.60 m3 / min

The process gain is thus

1 1
Kp = = = 0.0367. (9.18)
Tr 27.2

If we would like the closed-loop response to having a time constant the same as
the residence time, then set λ = Tr = 27.2. From Table 9.7, the controller tunings
are given as
⎧
⎪
⎪
1 1
⎪Kc =
⎨ · = 2 (%/%)
Kp λ
(9.19)
⎪
⎪ Ti = 2 λ = 2 × 27.2 = 54.4 (min)
⎪
⎩T = 0 (min).
d

For less tight control, the λ value can be set to λ = 2, or Tr = 54.4 min. The
resulting controller tunings would be K c = 1.0 and Ti = 108.8 min.
In this example, the vessel volume is a linear function of the level (height). In
case the relationship is not linear, this method provides only an average value for the
process gain. Adjustment of the result based on actual conditions should be applied.
Below is another example illustrating the case of nonlinear gain.
Example 9.6 Vessel with nonlinear gain. Consider the level loop tuning for the
three-phase separator in Fig. 9.12. There are two level/flow cascade control loops on
the separator, one loop for the total oil level (LC304/FC301) and the other is the
interface level control (LC302/FC303). The vessel sizing information is supplied
in the vessel manufacturer’s datasheet. The relevant information for the loop level
tuning is extracted from the datasheet and shown in Fig. 9.18. The flow ranges for the
two flow controllers are 0–2,200 m3 /d for FC-301 and 0–3,500 m3 /d for FC-303.
9.5 Tuning Level Control Loops with Vessel Sizing Data 333

Fig. 9.18 Vessel sizing information for level tuning

Following the loop tuning procedure, let us first calculate the vessel’s residence
time, which requires calculating each vessel’s control volume. By definition, the
control volume is the volume in the vessel between the two level taps. For LC-302,
the control volume is between height h 1 and h 2 ; and for LC-304, between height
h 3 and h 4 . Fig. 9.19 shows the cross-section areas corresponding to the four level
transmitter taps. The blue/dashed horizontal line designates the liquid level. The
level transmitter range for LC-304 is between the level lines in Fig. 9.19a and b,
while the level transmitter range for LC-302 is between the level lines in Figs. 9.19c
and d.
The cross-section areas are calculated as follows:
1. Control volume for level loop LC-304: From Fig. 9.19a, the liquid area is half
the circle, and thus the vertical cross-section area is given by
π 2 π 2
A1 = D /2 = 4 /2 = 6.28 m2 . (9.20)
4 4
From Fig. 9.19b, the vertical cross-section area up to height h 2 = 3.22 m can be
calculated as
   
2 h2 2 × 3.22
α2 = arcsin − 1 = arcsin −1
D 4
= arcsin(0.61) = 0.66 (9.21)
1 1
A2 = (π/ + 2 α/2 ) D 2 + D (2h 2 − D) cos(α)
8 4
1 1
= (π/ + 2 × 0.66) × 42 + × 4 × (2 × 3.22 − 4) × cos(0.66)
8 4
= 10.83 m2 . (9.22)
334 9 Control Loop Tuning and Improvement

Fig. 9.19 Vessel cross-section area calculation

The area corresponding to the level transmitter range is thus

A = A2 − A1 = 10.83 − 6.28 = 4.55 m2 (9.23)

and the control volume (the middle cylindrical part without the elliptical ends) is
thus given by
V = A · L = 4.55 × 10 = 45.5 m3 . (9.24)

Assume that the two elliptical bulges at the two ends contribute 4.5 m3 to the
control volume without detailed calculation. That makes the total control volume
to be about 50.0 m3 .
2. Control volume for level loop LC-302: The control volume for the LC-302
controller can be calculated similarly. From Fig. 9.19b, the vertical cross-section
area up to level transmitter tap h 3 = 1.53 m can be calculated as
   
2 h3 2 × 1.53
α3 = arcsin 1 − = arcsin 1 − = 0.24
D 4
1 1
A3 = (π − 2 α3 )D 2 − D (D − 2 h 3 ) cos(α3 )
8 4
9.5 Tuning Level Control Loops with Vessel Sizing Data 335

1 1
= (π − 2 × 0.237) × 42 − × 4 × (4 − 2 × 1.53) × cos(0.24)
8 4
= 4.42 m2 .

The calculation of the area up to height h 4 in Fig. 9.19d is similar to that in

Fig. 9.19b, with h 4 = 2.75 m
   
2 h4 2 ∗ 2.75
α4 = arcsin − 1 = arcsin − 1 = 0.384
D 4
1 1
A4 = (π + 2 α4 ) D 2 + D (2 h 4 − D) cos(α4 )
8 4
1 1
= (π + 2 × 0.384) × 42 + × 4 × (2 × 2.75 − 4) × cos(0.384)
8 4
= 9.21 m2 .

The area between the level transmitter taps is then given by

A = A4 − A3 = 9.21 − 4.42 = 4.79 m2 . (9.25)

The control volume of the vessel without counting the bulges on the two ends is
given as
V = A · L = 4.79 × 10 = 47.9 m3 . (9.26)

The volume of the two bulges can be calculated separately but is more involved.
However, without loss of much accuracy, assume it accounts for 5.1 m 3 to make
the total control volume 53.0 m3 .
In summary, the level controller tuning, whether by the generic step testing method
or the vessel sizing method, has significant challenges due to the nonlinearity.

Now follow the level tuning rules (Table 9.7) and calculate the two level con-
trollers’ tuning parameters. The results are shown in Table 9.8. For comparison, the
tunings for Level LC-301 are also included. To demonstrate the ease and flexibil-
ity on achieving different closed-loop performance, we choose λ = 1.5Tr instead of
λ = Tr for calculating the tuning parameters.

9.5.3 Characterization of Level Loop Dynamics

Depending on the vessel’s shape, the level is often a very nonlinear function of the
flow rate. This nonlinearity poses a significant challenge to PID control. We now
show another example to demonstrate how nonlinear the level process dynamic can
become. Fig. 9.20 shows a conical vessel that requires level control. The volume is
336 9 Control Loop Tuning and Improvement

Table 9.8 Level tuning parameters

Symbol Unit LC-101 LC-302 LC-304
Control volume V m3 70.7 53 50
Max flow range F m3 /d 3,744 3,500 2,200
Residence time Tr = V /F min 27.2 21.8 32.7
Process gain K p = 1/Tr %/% 0.037 0.046 0.031
Closed-loop λ = 1.5 Tr min 40.8 32.7 49.0
1 2
Controller gain K c = %/% 1.33 1.33 1.33
Kp λ
Integral time Ti = 2 λ min 81.6 65.4 98.0
Derivative time Td = 0 min – – –

Fig. 9.20 Nonlinearity of

conic shaped vessel

given by
2
π D
volume up to level h : V = · · h3. (9.27)
12 H

The relationship between the volume and the level is a cubic function and is highly
nonlinear. For example, assuming D = 10 m and H = 5 m, a level change of one
meter (10%) at the lowest liquid level leads to a volume change of
 2
π 10 1
V = × 13 − 0 = π (m3 ) (9.28)
12 5 3

while a level change of one meter (10%) at the highest level (from 9 to 10 m) will
cause a volume change of
9.5 Tuning Level Control Loops with Vessel Sizing Data 337
 2  2
π 10 π 10 271
V = × 10 −
3
× 93 = π (m3 ). (9.29)
12 5 12 5 3

Also, for comparison, for the same vessel, the volume at 50% level (5 m) is
 
π 10 2 125
V50% = × 53 = π (m3 ) (9.30)
12 5 3

while at level 100%, the volume is

 
π 10 2 1000
V100% = × 103 = π (m3 ) (9.31)
12 5 3

which is eight times more liquid than at 50% level.

Between Eqs. 9.28 and 9.29, there is a difference of 271 times! It is almost impos-
sible for a PID controller with constant tuning parameters to cope with such a wide
range of dynamic changes.
In practice, level characterization (transformation) can be applied to help PID
control. A new DCS tag can be created to calculate the vessel volume from the
vessel height h with Eq. 9.27. A volume controller replaces the level controller to
control the volume instead of the level. The calculated volume serves as the process
value (PV) for the PID controller. Since the volume and flow relationship is linear,
this volume PID control should be much more robust across the whole vessel level
range. After all, we are interested more in the volume than the height for inventory
control.

9.6 Summary

A control loop consists of five essential components (Fig. 2.2). The controller per-
formance is determined by the loop dynamics rather than process dynamics alone.
As long as the loop dynamics stay the same, the PID tuning should have no reason to
change. The possibility that the PID tuning causes the loop problem is surprisingly
lower than commonly believed. Most of the time, the poor performance is caused by
other reasons such as issues with the control valve or the data flow in the control loop.
For this reason, a systematic troubleshooting procedure (see Chap. 8) is the recom-
mended first step to rule out other problems before starting PID tuning. Jumping too
quickly to PID tuning often proves to be counter-productive. Nevertheless, tuning a
PID control loop is an essential skill that a competent process control engineer must
possess.
PID tuning is still a mixed bag of art and technology and requires knowledge,
experience, and patience. Controller tuning starts with a clear expectation of the tar-
geted performance to be achieved (Sect. 9.1). The best PID performance in automatic
control, such as the “quarter-wave decay,” may not necessarily be acceptable in the
process industry.
338 9 Control Loop Tuning and Improvement

Among the many different loop tuning methods, trial-and-error and empirical
methods (Sect. 9.2) are quick and simple to start. However, it typically takes longer
to complete since there is low confidence in the tuning parameters and thus no
“endpoint” to call a stop. Model-based PID tuning methods (Sect. 9.3) are based on
an explicit process model, representing a direct insight into the process dynamics,
and thus offer much higher confidence on the tuning parameters produced. As a
result, even though the model-based method may sound more time-consuming than
other approaches, the overall effort is typically less since there is less repeated effort.
There are some “shortcut” methods for tuning flow (Sect. 9.4) and level (Sect. 9.5)
control loops. They can produce sufficiently good performance while saving time.
However, a good understanding of their principle and limitations should be in order.

References

Åström K, Hägglund T (1984) Automatic tuning of simple regulators with specifications on phase
and amplitude margins. Automatica 20(5):645–651
Bequette BW (2002) Process control—modeling, design, and simulation. Prentice Hall
Blevins TL, McMillan GK, Wojsznis WK, Brown MW (2003) Advanced control unleashed—plant
performance management for optimum benefit. ISA
Brannan C (2018) Rules of thumb for chemical engineers—a manual of quick, accurate solutions
to everyday process engineering problems, 6th edn. Elsevier
Chien IL, Freuhauf PS (1990) Consider IMC tuning to improve controller performance. Chem Eng
Prog 86:33–41
Ford J (2013) Initial settings for PID controllers. Control Eng. https://www.controleng.com/articles/
initial-settings-for-pid-controllers
Garcia CE, Morari M (1982) Internal model control—1: a unifying review and some new results.
Ind Eng Chem Process Des Dev 21:308–323
Garcia CE, MPrett D, Morari M, (1989) Model predictive control: theory and practice—a survey.
Automatica 25(3):335–348
King M (2011) Process control—a practical approach. Wiley
Ljung L (1999) System identification: theory for the user, 2nd edn. Prentice-Hall, Englewood Cliffs,
NJ
Luyben WL, Lyuben ML (1997) Essentials of process control. McGraw-Hill
Morari M, Zafiriou E (1989) Robust process control. Prentice Hall
Rivera DE, Skogestad S, Morari M (1986) Internal model control 4: PID controller design. Ind Eng
Chem Process Des Dev 25:252–265
Smith CL (2009) Practical process control—tuning and troubleshooting. Wiley
Svrcek WY, Mahoney DP, Young BR (2006) A real-time approach to process control, 2nd edn.
Wiley
Visioli A (2006) Practical PID control. Advances in industrial control. Springer, London
Ziegler JG, Nichols NB (1942) Optimum settings for automatic controllers. J Dyn Syst Meas
Control-Trans ASME 64:759–768
 Part IV
Process Control Typicals

With the essential knowledge in Part II and core skills and methodologies in Part
III, this book’s final part devotes to some typical control applications. The goal is to
demonstrate how the knowledge and skills are applied to practical applications. The
specific applications selected include the following:

1. Common control-related calculations (Chap. 10): Many online calculations are

needed to support control solutions, from signal processing, measurement com-
pensation, to inferential property estimation. Online calculations, especially infer-
ential property estimation, are very cost-effective ways of enhancing and improv-
ing control performance. Online calculations are also essential for real-time mon-
itoring applications such as data validation, performance monitoring, abnormal
situation management, and operator advisory systems.
2. Control of typical equipment (Chap. 11): Pumps and compressors are critical and
prevalent equipment in the process industry, and their optimal control and reliable
protection are essential for safe operation and efficient production. Traditionally,
equipment control is regarded as the specialty of equipment manufacturers or
third-party specialists. However, equipment control is not more mysterious than
a distillation column or chemical reactor as part of the process flow. The standard
process analysis and control design skills and methodology apply.
3. Plantwide and unit-level control (Chap. 12): Most process control books cov-
ered distillation columns or chemical reactors to showcase the process analysis
and control design. However, this book purposely selects several less discussed
yet pervasive processes to demonstrate the thinking and analytical processes for
designing decent control solutions. The purpose is not to provide control solu-
tions as go-by templates, which is helpful but inefficient since even the typical
processes are countless. This chapter attempts to demonstrate a more practical
approach, which is to learn the general thinking process and structured analytical
skills that can be applied to any process control problem.

All the selected applications are typical in practical settings, and their proper
implementation or control has a significant impact on HSE or production profitability.
340 Part IV: Process Control Typicals

Each application demonstrates one significant aspect of process analysis, control

design, or control improvement. The selected applications are modestly complex so
that each can be covered in its entirety in one section to show the complete thinking
and analysis process, at the same time are not too trivial. As the discussions are
based on the knowledge and skills in Part I–Part III, cross-references to the relevant
sections are provided wherever appropriate. It is recommended to check back on
these sections as needed.
Designing, troubleshooting, and improving a process control solution is a mix of
art and technology. We strive for simplification and standardization to improve reli-
ability and sustainability. Therefore, the solutions should be as simple as practically
possible. On the other hand, we must also pay attention to the distinct characteristics
of each process to avoid blind copy and paste. The standardization should be on the
philosophy and methodology, not on the solution.
There is no standard solution in process control, and the best solution is always
the one that is simple and fits the purpose. As a result, the final solutions still depend
to a large extent on the practitioner’s knowledge, skill, experience, and personal
preference for complicated control problems such as plant-level or unit-level control.
Chapter 10
Control Typicals: Common Calculations

Online calculations are essential to real-time control. This chapter presents some
commonly encountered control-related calculations, such as signal transforma-
tion (Sect. 10.1), data quality assessment (Sect. 10.2), measurement compensation
(Sects. 10.3–10.5), and internal reflux (Sect. 10.6). They are critical components of
process control but are often misunderstood or incorrectly implemented. For con-
sistency and ease of understanding, each section follows a standard format, starting
with application background, theory and design, implementation and operation, and
finally, some application examples.

10.1 Signal Transmission and Transformation

Signal processing is an essential and standard component in the control data flow
and may include input processing, calculation processing, output processing, and
alarm processing (Yokogawa 2004). In a closed control loop consisting of multiple
function blocks, the signal transmission and processing can be a source of confusion
and configuration errors, negatively impacting the control design and performance.
A basic understanding of the calculations involved is required.

10.1.1 Application Background

In a typical PID loop, the process variable (PV) originates from the sensors in the
field as a measurement signal, transmitted from the sensor, through the transmitter,
to the control system, and back to the actuator and valve in the field.
Along this transmission path, signal scaling and conversion are performed at mul-
tiple places, some being simple range scaling to deal with the change of engineering
unit, while others may involve more complex signal transformation such as square-
rooting and signal characterization.
The control design is based on the loop dynamics, including everything on the loop
path, from the controller output, through the process, and back to the controller PV

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 341
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_10
342 10 Control Typicals: Common Calculations

(see Sect. 3.3.2). The signal transmission and transformation may have a significant
influence on control design and performance. For instance, a PV filter on the process
value adds additional dynamics to the loop. The filter’s impact may be insignificant
for most loops but must be considered for fast control loops such as flow control.

10.1.2 Theory and Design

1. Signal transmission: As shown in Fig. 10.1, the signals from the sensors in the
field are typically converted to a 4–20 mA electric current signal for transmitting
to DCS (e.g., as the PV for a PID controller). Similarly, the PID controller output
is converted to a 4–20 mA electric current signal before transmitting from DCS
to the field device. This 4–20 mA electric current signal is further converted to
a 3–15 psi pressure signal to operate a pneumatic valve. With 4–20 current mA
signal or 3–15 psi pressure signal, all values must be converted from the real-zero
to the live-zero (see Sect. 3.3.1) and back to the real-zero.
One example is a differential pressure (DP)-based (e.g., orifice) flow meter. A
differential pressure signal (e.g., 0–10 kPa) is produced by the sensor. This signal
is converted to 4–20 mA at the transmitter for sending to DCS. On the receiving
side, which typically is the analog input (AI) module in DCS or PLC, the 4–20
mA signal is converted back to a digital flow signal (e.g., 0–40 kg/s). Square-root
transformation is done along the way, either in the transmitter or the AI function
block.
The latest Fieldbus technology allows the digital control signal to extend directly
to the field equipment before converting to an analog signal.
2. Signal transformation: A signal goes through many transformations when trans-
mitted from a sensor in the field to the controller in the DCS. The transformation
can be linear or nonlinear. Many times, both linear and nonlinear transformations
are performed simultaneously.

Fig. 10.1 Transformation of control signals in a PID control loop

10.1 Signal Transmission and Transformation 343

A linear transformation changes the signal range but keeps the same percentage-
in-range (PIR) value; However, a nonlinear transformation changes the PIR but
may retain the signal range. Some common transformations are discussed as
follows:
a. Signal scaling: Signal scaling is a linear transformation. It adjusts the signal
range but retains the percentage-in-range (PIR). For example, converting a
level signal of 0–5 m to a percentage value of 0–100% is a linear transforma-
tion.
Assume the incoming signal is x with range xmin to xmax . The signal is to be
converted to y with a range from ymin to ymax . The conversion is performed
as
y − ymin x − xmin
PIR = = (10.1)
ymax − ymin xmax − xmin

which gives

x − xmin
y= · (ymax − ymin ) + ymin . (10.2)
xmax − xmin

If both xmin and ymin are zeros, then Eq. 10.2 is simplified into
x
y= · ymax . (10.3)
xmax

b. Signal characterization: Signal characterization is a nonlinear transformation

that changes the signal characteristics. It typically retains the range of the
signal but changes the PIR.
Most modern DCS has built-in support for signal transformation in many of
its basic control function blocks such as the transmitters, analog in (AI) and
analog out (AO) blocks (see Sect. 4.2.1 for DCS function blocks). Trans-
formation can be performed with a mathematical formula or via a lookup
table (characterization table). Although a simple mathematical equation is
preferred, an analytic formula cannot always represent the characteristics;
a lookup table can be used. Most transmitters and DCS I/O blocks support
signal characterization.
Assume the signal transformation (both linear and nonlinear) is defined by
five mapping points as in the following table:
x (input) y (output)
xmax ymax
x4 y4
x3 y3
x2 y2
xmin ymin
344 10 Control Typicals: Common Calculations

When a new measurement value x is received, the table is searched for an

exact match with x in the first column. If found, the corresponding y value
in the second column is the result. Most of the time, there would be no exact
match to the valve x in the table, but rather in between two values. Let us
say, for an input value x2 < x < x3 , a linear interpolation using the two
neighboring points is performed to produce the corresponding y value as

x − x2 x − x3
y= · (y3 − y2 ) + y2 , or y = · (y2 − y3 ) + y3
x3 − x2 x2 − x3

which is the same operation as in Eq. 10.2.

The advantage of the lookup table is that it is extremely flexible to represent
any transformation as long as the relationship between the two values is
monotonic. For example, a valve with linear characteristics can be converted
to fast open via a nonlinear transformation.
c. Square-rooting: Square-Rooting is a nonlinear signal conversion that is com-
monly applied to a differential pressure type of flow measurement. Square-
rooting is based on

y − ymin x − xmin
=
ymax − ymin xmax − xmin

which leads to

x − xmin
y= · (ymax − ymin ) + ymin . (10.4)
xmax − xmin

If both xmin and ymin are zeros, then Eq. 10.4 can be simplified as

x
y= · ymax . (10.5)
xmax

A square-rooting transformation changes the PIR of the signal. If the signal

is converted to a different engineering unit simultaneously, the range of the
signal will be changed as well. That is, signal characterization and signal
scaling are performed at the same time.
d. Signal filtering: A signal filter, most commonly a first-order exponential filter,
is equivalent to a nonlinear signal characterization. While keeping the range
intact, it modifies the PIR of the signal as the signal goes through the filter.
A first-order filter is given as

y f (k) = (1 − α) y f (k − 1) + α x(k) (10.6)

where x(k) is the raw signal and y f (k) is the filtered signal; y f (k − 1) is the
filtered signal saved from the previous execution (see Eq. 5.2).
10.1 Signal Transmission and Transformation 345

The filter constant α is related to the filter time constant τ by α = e−τ/Ts ,

where Ts is the sampling frequency. For example, assuming Ts = 1 min, a
2-second filter is equivalent to α = e−2/60 = 0.967, while a 10 second filter
is equivalent to α = 0.846.
On a modern control system, filtering is provided as a standard configuration
option. Adding a filter to a PV is just a matter of activating this function
by specifying a non-zero filter constant. However, beware that a PV filter
introduces extra lag to the loop dynamics and potentially affects the controller
tuning. Generally speaking, the filter constant should be at least five times
smaller than the time constant of the loop dynamics.
e. Signal compensation: The signal generated by the sensor often does not reflect
the actual physical value, e.g., due to different operating conditions. It often
requires compensation with other information to arrive at the correct mea-
surement value. Signal compensation, such as flow compensation, adjusts the
signal’s PIR but retains the measurement range. See Sects. 10.3 and 10.4 for
examples on flow and temperature compensation.

10.1.3 Implementation and Operation

The signal scaling and transformation are essential for signal processing. Support
for this operation is provided in multiple devices or control modules such as the
transmitters in the field, the analog in (AI), or the analog out (AO) function blocks
in DCS. The decision is often on where to perform the signal processing.
Normally, signal scaling should be performed closest to the signal source and
preferably with the built-in functions. However, more complex signal transformation
shall be performed in a location best for the application if multiple options are
available. For example, the square-rooting transformation from DP to flow shall
typically be performed in the flow transmitter. One exception is compressor anti-
surge control, where the transformation shall be done in DCS so that both the DP
and flow signals are accessible from DCS for purposes of display and troubleshooting.
Signal transformation via table look up shall be used with discretion. For a com-
plex relationship that requires a large table to define, it is recommended that an
analytic equation be obtained via curve fitting and be used in place of the large
lookup table.
All complex signal transformations shall be appropriately documented with
detailed narratives. Due to the lack of documentation, maintenance work on back-end
calculation often relies on reverse engineering and guesswork in practical applica-
tions, which is highly inefficient and error-prone.
346 10 Control Typicals: Common Calculations

10.1.4 Application Examples

Online calculations are ubiquitous. Here are a few examples illustrating the basics.

Signal Scaling

Assume a level sensor is calibrated to measure a level from 0 to 2.0 m. The level
signal in % is then given, following Eq. 10.2, by

LT − 0
LI = · (100% − 0) + 0 = 50% · LT.
2.0 − 0

For example, an LT = 1.2 m of the level at the sensor will produce LI = 50% ×
1.2 = 60% of the level.
In reality, however, the level signal may be transmitted as an analog electric current
signal of IT = 4–20 mA from the field to the DCS, and then converted to a digital
level indicator value of LI = 0–100%. The conversion is performed as follows: from
LT = 0–2 m to IT = 4–20 mA in the transmitter:
LT − 0
IT = · (20 − 4) + 4 = 8 · LT + 4
2−0

and then from IT = 4–20 mA to LI = 0–100% in DCS:

IT − 4 IT − 4
LI = × (100 − 0) + 0 = × 100%.
20 − 4 16

A 1.2 m of level will produce an electric current signal of 8 × 1.2 + 4 = 13.6 mA,
and then to a level indication of (13.6 − 4)/16 × 100 = 60%.

Square-Rooting for Flow Measurement

A differential pressure (DP)-based flow meter, such as an orifice or a Venturi tube,

directly measures the pressure difference across the device. The pressure signal
requires square-rooting to produce a flow signal. The flow FT and P are related by

P − Pmin
FT = · (FTmax − FTmin ) + FTmin
Pmax − Pmin

P
= · FTmax . (10.7)
Pmax
10.1 Signal Transmission and Transformation 347

Equation 10.7 follows Eq. 10.5 with both Pmin and FTmin being zero. For example,
assume the differential pressure (DP) measurement range is 0 ∼ 50 kPa under a given
condition. The flow meter is calibrated to produce a flow signal of 0–20,000 kg/h cor-
responding to the 0 ∼ 50 kPa range at this given condition. The signal transformation
can be performed following Eq. 10.7 as

P
FT = × 20000.
50
√
That is, a P signal of 12.5 kPa indicates a flow of 12.5/50 × 20, 000 = 10, 000
kg/h.
Internally, the P signal is first converted to a 4–20 mA electric current signal in
the flow transmitter for transmission in the field. The analog signal is converted to a
digital signal in DCS before transforming into a flow signal. The transformation can
be performed in two ways:
1. Square-rooting in the transmitter: Linear scaling with square-rooting from P =
0–50 kPa to IT = 4–20 mA in the transmitter, followed by linear scaling from IT
= 4–20 mA to FT = 0–20,000 kg/hr in the AI block:
 
P − 0 P
IT = · (20 − 4) + 4 = · 16 + 4
50 − 0 50
IT − 4 IT − 4
FT = · (20000 − 0) + 0 = × 20000.
20 − 4 16
√
The 12.5 kPa of P signal will produce an electric current signal of 12.5/50.0 ×
16 + 4 = 12 mA for transmission, and then a value of (12 − 4)/16 × 20,000 =
10,000 kg/hr as flow indication.
2. Square-rooting in DCS: Linear scaling from P = 0–50 kPa to IT = 4–20 mA in
the transmitter, followed by linear scaling with square-rooting in the AI block:

P − 0 P
IT = × (20 − 4) + 4 = × 16 + 4
50 − 0 50
 
IT − 4 IT − 4
FT = × (20000 − 0) + 0 = × 20000.
20 − 4 16

That is, the 12.5 kPa P signal produces√an electric current signal of 12.5/50 ×
16 + 4 = 8 mA, and then a flow signal of (8 − 4)/16 × 20,000 = 10,000 kg/hr.
One advantage of the first approach is that the square-rooting of the sensor signal
in the transmitter results in a larger electric current (12 mA vs. 8 mA in this
example) for transmission. This analog electric current tends to have a higher
signal-to-noise ratio and thus offers higher accuracy during transmission.
Although the square-rooting can be performed in multiple locations along the
signal transmission path, it shall only be performed once!
348 10 Control Typicals: Common Calculations

Signal Characterization

In this example, a crude oil pipeline travels over mountainous terrain to reach the
storage tanks in the tank farm at the coastal port. The maximum flow rate is close to
a million barrels a day.
A feedback + feedforward control scheme is implemented to control the arriving
pressure at the storage tank. Flow rate is the feedforward variable, and the change
in flow rate will cause pressure friction loss along the pipeline. The pressure friction
loss in the crude oil pipeline is given in Table 10.1 as a function of the flow rate.
This pressure–flow relationship was implemented as a lookup table in Foxboro
DCS using a CHARC (for signal characterizer) function block. This approach works
as expected.
Alternatively, a simple mathematical formula can be used to approximate the
relationship defined in this table. Using a formula instead of a large lookup table
allows a cleaner implementation and faster execution in DCS. A trivial curve-fitting

Table 10.1 Pressure friction Flow rate Frictional losses

loss versus flow rate
F P
(bpd × 106 ) (kPa)
0 0.00
0.0525 30.74
0.1050 105.02
0.1575 215.59
0.2100 359.24
0.2625 533.92
0.3150 738.13
0.3675 970.71
0.4200 1230.72
0.4725 1517.36
0.5250 1829.97
0.5775 2167.94
0.6300 2530.74
0.6825 2917.93
0.7350 3329.07
0.7875 3763.80
0.8400 4221.79
0.8925 4712.64
0.9450 5228.14
0.9975 5769.36
1.0500 6335.11
10.1 Signal Transmission and Transformation 349

exercise reveals that a simple second-order polynomial function between the pressure
and flow can reasonably approximate the lookup table:1

[pressure loss] ≈ 858.9 ∗ [flow rate] + 4959.4 ∗ [flow rate]2 . (10.8)

10.1.5 Special Considerations

In practical applications, there are more scenarios involving single transformations

and processing than discussed here. Many are typically hidden from process control
engineers, but it is beneficial to know their existence and the potential impacts.

Raw Counts

The word raw count is often encountered when dealing with more primitive PID
loops such as those implemented in a remote terminal unit (RTU) (see Sect. 3.1.9).
This raw count involves another step of signal scaling during signal transmission
from the field to DCS. This step converts the signal from analog to digital by the
so-called A/D converter and is part of the AI block’s functionality. It is an essential
step but is usually not interested by process control engineers.
With a 12-bit A/D converter, the range of the digital signal (raw counts) is between
0 and 4095 (= 212 − 1). Therefore, the electric current signal is transformed into the
digital signal of the range 0–4095 as the A/D converter output. This digital signal is
typically shifted by 2 to 4 bits by different vendors for improved resolution. So the
range may become 0–16383 (14 bits), 0–32767 (15 bits), or even 0–65535 (16 bits).
This raw count signal is then scaled to the correct engineering range and becomes
the output of the Analog Input (AI) block. A detailed illustration of the signal scaling
and transformation is provided in Table 10.2.
However, for historical reasons, the raw count range varies between PLC/DCS
vendors and rarely matches the ranges in Table 10.2. For example, Allen Bradley
PLC uses 6241–31206 for 4–20 mA mapping. Following Eq. 10.2, the conversion is
given by

I −4
R= × (31206 − 6241) + 6241 (10.9)
20 − 4
R − 6241
I = × (20 − 4) + 4. (10.10)
31206 − 6241

The maximum raw count value of 32,767 corresponds to a current signal of 21 mA,
which is defined as the high range (not compliant with NAMUR NE43), and the
live-zero of 4 mA in current corresponds to 6241 in the raw count.

1 With MATLAB/Octave: x=[ones(size(F)),F,F.^2]\P; where F is the flow rate and P

is the pressure loss.
350 10 Control Typicals: Common Calculations

Table 10.2 Example of flow signal scaling and transformation

Transmitter A/D raw count DPI FI
kPa PIR% mA 12-bit 15-bit kPa PIR% m3 /h PIR%
Span 105% 22.0 4,095 32,767
Burnout (BAD) 21.0 3,909 31,278
Over range 102.5% 20.5 3,816 30,533 10.31 40.62
Range high 10.0 100% 20.0 3,723 29,788 10.00 100% 40.0 100%
75% range 7.5 75% 16.0 2,978 23,831 7.50 75% 34.6 86.6%
50% range 5.0 50% 12.0 2,234 17,873 5.00 50% 28.3 70.7%
25% range 2.5 25% 8.0 1,489 11,915 2.50 25% 17.9 50.0%
Range low 0.0 0 4.0 745 5,958 0.00 0 0.0 0
Under range −1.25% 3.8 707 5,660 −0.13
Burnout (BAD) 3.6 670 5,362 −0.25
Zero 0.0 0 0

Pneumatic Signals

Most control valves are driven by instrument air and controlled with pneumatic
signals (i.e., pressure). The controller output, typically 0–100%, is transmitted to the
field as a 4–20 mA analog signal. This electric current signal of 4–20 mA is converted
to a 3–15 psi pneumatic pressure signal through an I/P transducer to drive the control
valve. This I/P conversion is another signal scaling step involving two signals that
both have live-zeros.
The 4–20 mA signal transmission has long been adopted as the standard for signal
transmission. With the live-zero set to 4 mA, the over-range and under-range reading
is 3.8 mA and 20.5 mA (i.e., −1.25% and 103.125%), while the burn-out out-of-
range values are 3.6 mA and 21.0 mA (−2.5% and 106.25%), respectively, per the
widely accepted standards of NAMUR NE 43 and API 551.

10.2 Data Quality and Bad Data Handling

Data quality is a measure of the condition of data based on factors such as accuracy,
completeness, consistency, reliability, and timeliness. Measuring data quality can
help identify data errors that need to be resolved and assess whether the data is
adequate for the intended purpose.
Low data quality is one of the critical factors that prevent many advanced appli-
cations from online 24/7 operations. Many complexities in applications are catered
to deal with reliability issues in real-time data.
10.2 Data Quality and Bad Data Handling 351

10.2.1 Application Background

Nothing lasts forever. Measurement can fail and will fail; it is just a matter of proba-
bility. Bad process values (BAD PV) are one of the most common issues in practice.
How to gracefully respond to bad process values is a crucial consideration in process
control design. Below are some examples:
1. In a simple level-flow-valve cascade control loop, both the process variables (PV)
for the level controller and the flow controller can fail, so do the control valve
positioner and network communication. How should the control loop detect the
failure and respond? In a much more complex control solution, such as compressor
control in Fig. 4.2, what is the mitigating action if one of the process values
becomes unavailable or erroneous?
2. Many model predictive control (MPC) applications are implemented in a stan-
dalone machine. The MPC application communicates with DCS to receive the
process values, calculates the setpoint for the PID controllers, and sends them
back to DCS. How should MPC monitor the network communication’s health?
How should MPC and DCS respond if the communication is lost or becomes
suspicious?
3. Online analyzers are becoming prevalent, especially in petrochemical and chemi-
cal plants. However, their availability and reliability are still not sufficiently good
to support online real-time control. How to assess the quality of data received from
the analyzer to decide whether the closed-loop control should stay in control or
temporarily go off-line?
These questions must be considered, and mitigation actions must be included in
the control design.

10.2.2 Theory and Design

Measurement failure refers to either loss of data or data with erroneous value. The
control system may react unpredictably to measurement failures and is thus a poten-
tially serious hazard. There have been no standard solutions for bad data handling,
but some general guidelines or approaches exist:

1. Reactive approach: Once a bad measurement is detected, the control scheme

should safely and gracefully downgrade itself to a less sophisticated operating
mode, and at the same time, alarm the operations personnel to intervene. This
degraded operating condition is called crippled mode operation.
2. Proactive approach: For mission-critical process variables, redundant measure-
ment is commonly implemented to reduce the probability of failure. Real-time
monitoring and fault detection are also an active research area.
352 10 Control Typicals: Common Calculations

3. Intelligent handling: In a modern plant, the instrumentation level has improved

to such an extent that redundant information, both spatial and causal, is available
to cross-validate each other and provide quality indications and early warnings.
One key area to improve is reliably detecting and indicating the quality of mea-
surement. One change in recent years was the generation and transmission of data
quality as part of the process data. For example, with the open platform communi-
cation (OPC2 ) protocol, all the data from the field has a data quality attribute with
one of the following three statuses:
1. Good: The tag value is trustworthy.
2. Uncertain: There is a non-critical abnormality associated with the tag value.
3. Bad: There is a critical abnormality associated with the tag value.
See the OPC data access specification from the OPC Foundation for a detailed defi-
nition of data quality.
A general recommendation is that only good data is acceptable for real-time and
automated decision-making such as closed-loop control. Uncertain data cannot be
trusted for automated decision-making but may still have reference values and thus
can be used for information purposes such as monitoring and display.
The OPC quality indication, however, is related chiefly to device-level measure-
ment issues. More advanced data quality indicators suitable for process control are
still not widely available. For example, there are no standard and reliable approaches
in detecting common data anomalies and generating credible data quality indicators.
Below are some common data anomalies:
1. Out-of-range value: The measurement value is out of the measurement range.
2. Frozen value (flat line): A dead sensor often produces a constant (frozen) value,
especially from online analyzers.
3. Spikes and outliers: A sudden and significant change in value.
4. Runaway value: A measurement value that is within range but gradually running
away from its correct value.

10.2.3 Implementation and Operation

A careful evaluation of the potential causes and consequences of bad measurements

is needed. This exercise is part of the hazard and operability (HAZOP) review. Bad
measurement and poor measurement are the two prevalent abnormal scenarios that
have to be addressed from the beginning.
The field implementation infrastructure (DCS/PLC) has provided various means
for handling abnormal situations. However, many of the decisions are dependent on
the operating objectives and must be dealt with case by case. Here are a few practical
options available for handling bad data issues:

2The abbreviation OPC has changed from the original OLE for Process Control to Open Platform
Communication.
10.2 Data Quality and Bad Data Handling 353

1. Middle-of-three voting: The middle-of-three configuration provides higher avail-

ability and is thus favored in practical applications. The voting mechanism ensures
that the correct signal is always selected even if one of three input signals fails.
Challenge remains when two or all the three input signals are not available or
erroneous.
2. Measurement comparison and validation: For specific measurements such as
online analyzers, validation algorithms must detect analyzer malfunction such
as flat-lining and spikes.
When redundant measurements are available, for example, one for control and one
for safeguarding, the two measurements can be constantly compared, and an alarm
is generated if the difference exceeds a predetermined value. The comparison
should consider the inherent noise in the signals and apply a realistic threshold to
avoid nuisance alarms.
3. Dual transmitters: All meters have an effective range. For example, a flow meter
has a typical turndown ratio of 3:1 for DP-based measurement or 10:1 for other
types. The measurement reading below this turndown ratio will either be unreli-
able (e.g., orifice meter) or unavailable at all (e.g., vortex meter).
For measurement with a large span, dual flow meters can be used to improve the
accuracy. See Fig. 10.2 for a typical arrangement. The high-range transmitter loses
accuracy when it measures below its turndown limit (typically 10% of the full
range). Below the turndown limit, the source of the measurement value switches
to the low-range transmitter.
To achieve a smooth transition between the two signals, which typically takes
place between 80% and 100% of the range of the low- range transmitter, and to
maximize the measurement accuracy over the entire operating range, the following
calculation can be used, where k defines the transition factor:

Fig. 10.2 Dual-range measurement

354 10 Control Typicals: Common Calculations
   
FTX2 − Fmin
k = max min ,1 ,0
Fmax − Fmin
F = (1 − k) · FTX1 + k · FTX2

where
k: Transition range factor.
FTX1 : Measurement from low range transmitter.
FTX2 : Measurement from high range transmitter.
Fmin : Minimum of the transition range (typically 80% of FTX1 range).
Fmax : Maximum of the transition range (typically 100% of FTX1 range).
When the measurement FTX2 is below the minimum transition range (80% of
low-range transmitter), k = 0, the final measure is determined only by the low-
range transmitter. When the actual flow is higher than the maximum transition
range (100% of the low-range transmitter), k = 1, then the high-range transmit-
ter provides the measurement. When the actual measurement value is within the
transition range, the measurement value is a weighted average of the two mea-
surements, determined by the above equation (0 ≤ k ≤ 1).
If flow compensation needs to be performed, it should be applied to the individual
flows rather than the combined flow unless the calibration conditions are identical
for both measurements.
4. Soft sensors: Certain process variables such as product quality (compositions,
cut-points, vapor pressures, and dew points) are inherently difficult to measure,
and online analyzers are notorious for low availability and reliability. Soft sen-
sor technology (see Sect. 5.3) has become popular where more reliable process
measurements are used to infer the process values for supplementing the online
measurement. The soft sensor technology can also be used to cross-validate some
online measurements.

The PID controller and other function blocks in DCS have provided many configu-
ration options to allow the control scheme to respond to various abnormal conditions.
Some examples include the following:
1. PID Controller: A PID controller typically sheds itself to the manual (MAN) mode
of operation when a BAD PV is detected. In the MAN mode, the controller skips
the PID calculation, and the operator sets the controller output (OP) directly. See
Sect. 3.6.2 for crippled mode handling.
2. Selector: A bad input to a selector will typically exclude itself from the selection.
It can also be configured to propagate the BAD PV status directly to the selector
output.
3. Totalizer: A totalizer does the accumulation or summation of the input to produce
a total value over time. Multiple options exist for the totalizer to respond to bad
input values, including using the last good value, skipping the input accumulation,
or simply stopping the totalizer.
10.3 Flow Compensation and Conversion 355

10.3 Flow Compensation and Conversion

Flow is one of the most fundamental measurements. Due to the diversity of mea-
suring principles and devices, the raw measurement signal may need various signal
processing to provide an accurate estimate of the flow rate at conditions different
from the conditions at which the flow meter was calibrated.

10.3.1 Application Background

Flow is not directly measurable. Instead, it is always inferred from other directly mea-
surable values such as differential pressure (DP), time, and frequency. Based on the
measurement principle, there are many different flow meter types. Correspondingly,
there are many ways of inferring the flow from the raw measurements.
Flow meters typically consist of a primary element (e.g., orifice plate, and vortex
meter), one or more secondary devices such as pressure and temperature transmitters,
and sometimes tertiary devices such as a flow computer. See Fig. 10.3 for a depiction.
All measurements have errors. Some may be inherent to the measuring principle,
while others may be due to the environment, such as heat expansion. For example,
a differential pressure (DP)-based flow meter such as an orifice plate produces a
DP signal. The DP signal is then converted to a flow signal with the assumed fluid
property and flow-line condition. If the flow-line condition changes, the meter will
produce the correct DP value, but the flow value derived from the DP may no longer be
accurate since the correlation between the two has changed under the new condition.
Flow compensation is thus required with a modified correlation equation.
If the flow-line temperature changes significantly, the orifice’s heat expansion
may cause the orifice to produce an inaccurate DP value. A wrong DP leads to
an incorrect flow value. For precise measurement, temperature compensation for
the flow is also required. However, this error is relatively insignificant and is not
commonly a concern.

Fig. 10.3 Flow meter

components
356 10 Control Typicals: Common Calculations

10.3.2 Theory and Design

A perfect flow meter would be reliable, accurate, inexpensive, with a high turndown
ratio, requiring little calibration/maintenance, and can apply to any fluid at any con-
dition. In real life, of course, this perfect flow meter does not exist. The best flow
meter is always the one that is fit-for-purpose.

Type of Flow Meters

Based on the measurement principles, flow meters can be broadly divided into linear
and square-root types:
1. Linear type meters: The so-called linear type flow meters are typically based on
velocity or density, with which the flow is a linear function:
a. Velocity-based: Both vortex meter and ultrasonic meter are velocity-based.
Volumetric flow rate is the product of the flow velocity v and cross-section
area of the pipe A, i.e., F = v A. By measuring the flow velocity, the flow
rate can be readily calculated.
b. Density-based: The Coriolis flow meter directly measures the density of the
fluid. The mass flow can then be calculated.
2. Square-root-type meters: Square-root-type flow meters such as those based on
orifice plate, Venturi tube, or flow nozzle measure the pressure drop (P) across
the device. A fixed relationship between the pressure drop and flow rate can
be established at a controlled condition (calibration condition). Consequently,
the flow rate is calculated from the actual measured P value based on this
relationship. The DP-based flow meter is called the square-root type since the
flow rate is proportional to the squared root of P.

Type of Flow Values

There are two types of flow measurements: mass flow and volumetric flow. For incom-
pressible fluid such as liquid, both mass flow and volume flow are not influenced by
the operating condition. For compressible flow such as gas and steam, however, the
volumetric flow rate does change with many factors, primarily:
1. Gas property, such as the gas molecular weight (M), compressibility Z , and to a
lesser extent the ratio of specific heat κ = C p /Cv .
2. Operating condition, including the operating pressure P and temperature T .
For clarity and consistency of presentation, let us define the following types of
flow conditions:
1. Design condition: The design condition is the expected normal operating condi-
tion specified during process design.
10.3 Flow Compensation and Conversion 357

2. Calibration condition: The calibration condition is the reference condition under

which the flow meter is calibrated. Typically, the calibration is chosen to be the
same as the design condition of the process.
3. Reference condition: The reference condition is a pre-specified pressure and tem-
perature condition. The reference condition is often selected to be the same as the
design condition and the calibration condition.
4. Flowing condition, line condition, or actual condition: This condition refers to
the actual pressure and temperature that the fluid is flowing under.
5. Standard condition: The definition of the standard condition varies by industry
and country. The standard condition for the oil and gas industry is typically P = 1
atm and T = 15◦ C.
6. Normal condition: The definition of the normal condition varies by industry and
country. A typical normal condition is P = 1 atm and T = 0◦ C.
For clarity, let us also define the different types of flow values:
1. Indicated flow (FI): The indicated flow refers to the flow that is directly read from
the flow transmitter. Sometimes, this is also called the raw flow. For a DP-based
flow meter, the indicated flow is not the correct flow unless the flowing condition
is the same as the calibration condition.
2. Actual flow: The actual flow refers to the true process flow, compensated to the
flowing condition if necessary. The actual flow is also called the compensated flow
or true flow. This flow is only meaningful when the associated flowing condition
is mentioned.
3. Equivalent flow: The equivalent flow refers to the same actual flow converted to
a different baseline condition such as standard condition or normal condition.
The baseline condition may be chosen as the design condition or the standard
condition.

Flow Compensation

A DP-based flow meter can produce correct results only if the fluid density is the same
as that under the calibration condition. Otherwise, the indicated flow is inaccurate
and needs to be compensated or corrected with the density to arrive at the correct
flow measurement. On the other hand, velocity or density-based linear flow meters
directly produce the correct volumetric or mass flow, and thus no compensation is
needed.
Assuming that the fluid density is ρa under actual condition (line condition), ρr
under reference/calibration condition, and ρs under standard condition, the compen-
sation formulas are given in Table 10.3.
Since density is a function of pressure P, temperature T , and molecular weight
M, in case the fluid density is unavailable or remains constant, it is often acceptable
to perform with the available measurements (P, T , and M). The accuracy of the
compensation is reduced accordingly. That is, in the order of preference,
358 10 Control Typicals: Common Calculations

Table 10.3 Flow compensation and conversion with density

Meter type Raw Flow type Indicated Compensated Standardized
 
√ ρa ρa 1
Orifice P Mass FI = C P ρr Fm = FI · Fs = FI · ·
 ρ ρ ρs
 r
√ r
P ρr ρr ρa
Orifice P Volume FI = C Fv = FI · Fs = FI ·
ρr ρa ρs
ρa
Vortex v Volume FI = v · A Fv = FI Fs = FI ·
ρs
ρa
Ultrasonic v Volume FI = v · A Fv = FI Fs = FI ·
ρs
1
Coriolis ω Mass FI = f (ω) Fm = FI Fs = FI ·
ρs

Table 10.4 Flow compensation/conversion with pressure/temperature

Meter Flow Indicated Compensated Standardized
 
√ Pa Tr Pa Tr Z s R Ts
Orifice Mass FI = C P ρr Fm = FI · Fs = FI · · ·
Pr Ta P Ta Ps Ms
   r
P Pr Ta Pr Pa Ts Ts
Orifice Volume FI = C Fv = FI · Fs = FI ·
ρr Pa Tr Ps Ps Tr Ta
Pa Ts
Vortex Volume FI = Fv Fv = FI Fs = FI ·
Ps Ta
Pa Ts
Ultrasonic Volume FI = Fv Fv = FI Fs = FI ·
Ps Ta
1
Coriolis Mass FI = Fm Fm = FI Fs = FI ·
ρs

ρ1 P1 T2 P1 T2
→ · → →
ρ2 P2 T1 P2 T1

The pressure and temperature are commonly available. The compensation formulas
with pressure and temperature are given in Table 10.4, with the density ρ replaced
by P and T .

Flow Conversion

Flow conversion is a technique to convert the flow from one condition to another.
Flow conversion assumes that the flow measurement is accurate and correct, and the
same flow value is to be represented in a different engineering unit. For example, the
flow rate can be converted from actual flow rate (m3 /h) to standard flow rate sm3 /h
or normal flow rate nm3 /h.
Certain flow meters such as vortex and ultrasonic meters directly produce vol-
umetric flow, which is the true actual flow, then there is no need to compensate.
However, the flow can be converted to other desired conditions, such as standard
10.3 Flow Compensation and Conversion 359

flow or normal flow. Coriolis meters directly produce mass flow, and typically no
need to convert unless volumetric flow is expected.

Compensation versus Conversion

A flow transmitter provides a raw flow (or indicated flow) under actual conditions.
This raw actual flow is compensated to become the compensated flow under actual
conditions and then can be converted to compensated flow under standardized con-
ditions. Alternatively, the raw flow under actual condition can be converted to stan-
dardized flow first to get the raw flow under standard condition (uncompensated) and
then compensated to the compensated flow under standard condition. Either way, it
arrives at the same final compensated standardized flow.
This conversion procedure is illustrated in Table 10.5. It is essential to know what
type of flow representation is needed.
For DP-based volumetric flow measurement, assuming the indicated flow is rep-
resented by FI, the formulas for performing the compensation and conversion are
summarized in Table 10.8.
For example, if the orifice is calibrated to produce a flow value of FI at the
calibration (reference) condition, to compensate
√ the flow to the actual condition, we
simply multiply the indicated flow FI by ρr /ρa . Sometimes, the flow meter is
calibrated to provide standardized actual flow directly. To obtain the compensated

Table 10.5 Flow compensation verses flow conversion

Raw Actual Flow ⇒ compensation ⇒ Compensated Actual Flow

⇓ ⇓
conversion conversion
⇓ ⇓
Raw Standard Flow ⇒ compensation ⇒ Compensated Standard Flow

Table 10.6 Linear volumetric flow meter

Raw Flow ⇒ compensation ⇒ Compensated Flow

Actual Condition 
FI = Fv ×1 = FI
(kg/s)
⇓
ρa ρa
conversion × ×
ρs ρs
⇓
Standard Condition ρa ρa
(sm3/h) = FI · ×1 = FI ·
ρs ρs

Raw Flow ⇒ compensation ⇒ Compensated Flow

360 10 Control Typicals: Common Calculations

Table 10.7 Linear mass flow meter

Raw Flow ⇒ compensation ⇒ Compensated Flow

Actual Condition 
FI = Fm ×1 = FI
(m3/h)
⇓
1 1
conversion × ×
ρs ρs
⇓
Standard Condition = FI ·
1
×1 = FI ·
1
(sm3/h) ρs ρs

Raw Flow ⇒ compensation ⇒ Compensated Flow

Table 10.8 DP-based volumetric flow meter

Raw Flow ⇒ compensation ⇒ Compensated Flow

  
Actual Condition  P ρr ρr
(m3/h) FI = C × = FI ·
ρr ρa ρa
⇓
ρa ρa
conversion × ×
ρs ρs
⇓
 √
Standard Condition ρa ρr ρr ρa
(sm3/h) = FI · × = FI ·
ρs ρa ρs

Raw Flow ⇒ compensation ⇒ Compensated Flow

√
actual volume flow, we need to multiply the flow by ρr /ρa (compensation), and
then divide it by ρa /ρs . Similarly, the formulas for performing the compensation and
conversion for DP-based mass flow measurement are summarized in Table 10.9.

10.3.3 Implementation and Operation

There are many different ways of performing flow compensation, depending on what
is available online.

Dynamic and Static Compensation/Conversion

Flow compensation can be either static, dynamic, or quasi-dynamic depending on the

number of relevant parameters. Since the density of the flow is given by ρ = ZPRMT ,
10.3 Flow Compensation and Conversion 361

the compensation is done either by density ρ, or by one or more of the individual

parameters in the formula, including pressure (P), temperature (T ), molecular weight
(M), and compressibility (Z ), if the density is not available:
1. Static compensation: Suppose it is known that the density at the actual operating
condition is different from that at the calibration condition but remains relatively
constant. In that case, a fixed compensation term can be used in place of ρa in the
above formulas in Tables 10.7, 10.8 and 10.9. Similarly, if the molecular weight
is known and is different at the flowing line condition, then static compensation
on molecular weight can be performed.
2. Dynamic compensation: Since the pressure and temperature at flow conditions
are typically different from that at calibration conditions, online compensation
can be performed with the actual pressure and temperature measurements. See
Tables 10.3 and 10.4.
3. Quasi-dynamic compensation: Some of the properties are known and relatively
constant while others vary with time, then mixed compensation can be applied.
For example, the gas’s molecular weight and compressibility are different from
that of the calibration gas but do not vary much with time, but the pressure and
temperature vary significantly. The flow compensation can then be performed
with static compensation on molecular weight and compressibility, and dynamic
compensation on pressure and temperature.

10.3.4 Application Examples

To correctly apply compensation and conversion, start with the type of flow meter
used, the type of flow (mass vs. volume); it is calibrated to measure under what con-
dition it operates, and for what process condition to display the value. Table 10.10
provides a lookup table for finding the right formula for compensation and conver-
sion. Several examples are provided below to demonstrate the procedure.

Table 10.9 DP-based mass flow meter

Raw Flow ⇒ compensation ⇒ Compensated Flow

 
Actual Condition  √ ρa ρa
(kg/s) FI = C Pρ × = FI
ρr ρr
⇓
1 1
conversion × ×
ρs ρs
⇓
 
Standard Condition 1 ρa ρa 1
(sm3/h) = FI · × = FI ·
ρs ρr ρr ρs

Raw Flow ⇒ compensation ⇒ Compensated Flow

362 10 Control Typicals: Common Calculations

Table 10.10 Compensation For volumetric flow For mass flow

method lookup table
Linear type meter See Table 10.6 See Table 10.7
Square-root-type meter See Table 10.8 See Table 10.9

Pressure and Temperature Compensation on Flow

The most common application of flow compensation is with pressure and temperature
on the gas flow. Assuming a DP-based flow meter is calibrated to the design condition
of 2,200 kPa(g) for pressure and 60 DegC for temperature. With the actual flow-line
condition at 2,400 kPa(g) and 65 ◦C, the indicated flow rate is 89,000 m3 /h. Since the
actual pressure and temperature condition is different from the calibration condition,
we know that the indicated flow of 89,000 m3 /h is not the correct value.
To convert from indicated flow to compensated flow, we can start from Table 10.10.
Since the compensation is on volumetric flow with a square-root type of meter, the
compensation formula is in Table 10.8. However, since density is unavailable, the
online compensation is commonly simplified to pressure and temperature compen-
sation only:
PM ρ P Ts
ρ= , ≈ . (10.11)
Z RT ρs Ps T

To obtain the compensated actual flow, the following formula from Table 10.8 can
be used:

ρr
FIcomp = FI ·
ρa

Pr Ta
≈ FI · ·
Pa Tr

2200 + 101.35 65 + 273.15
= 89, 000 × ×
2400 + 101.35 60 + 273.15
= 86, 006 m3 /hr (10.12)

where 101.35 and 273.15 are added to the actual pressure and temperature, respec-
tively, to arrive at the absolute pressure and temperature.
Converting the indicated flow to compensated standard flow is a two-step calcu-
lation, as shown in Table 10.8:

ρr ρa
FIcomp = FI ·
ρs ρs
 
Pr Pa Ts Ts
≈ FI ·
Ps Ps Tr Ta
10.3 Flow Compensation and Conversion 363
 
2200 + 101.35 2400 + 101.35 15 + 273.15 15 + 273.15
= 89, 000 ×
101.35 101.35 60 + 273.15 65 + 273.15
= 1, 808, 793sm3 /hr (10.13)

where the standard condition is defined as Ps = 101.3 kPa and Ts = 15◦ C.

Vortex Meter Flow

A vortex meter produces the correct actual volumetric flow at the flow-line condi-
tion with no need for compensation. In practice, the flow may typically need to be
converted to some common base condition for comparison and understanding. For
example,
1. Mass flow: There is no distinction between actual mass flow and standard mass
flow, so there is no conversion necessary between the two mass flows. However,
converting the indicated volumetric flow to a mass flow requires the gas density
ρ. In addition to the pressure and temperature, the gas molecular weight M and
compressibility Z are also required:

Fm = FI /ρ.

2. Standard volume flow: Table 10.10 suggests that the formula in Table 10.6 be used
to convert the actual volume flow to standard flow with a linear flow meter:
ρa Pa Ts
Fv,s = Fv,a ≈ Fv,a . (10.14)
ρs Ps Ta

For standard condition Ps = 101.3 kPa and Ts = 15◦ C, sometimes the following
simplified formula is given:

Pa Ts 273.15 + 15
Fv,s = 2.844 Fv,a · since = = 2.844.
Ta Ps 101.3

In addition, if the molecular weight and compressibility are known, static compen-
sation with molecular weight and compressibility can be applied.

Compensated Flow to Raw Flow

In some applications, such as compressor anti-surge control (details can be found in

Sect. 11.3), the required flow measurement is the raw volumetric flow at the actual
condition.
Assume the orifice-based flow meter is calibrated to provide mass flow measure-
ment. The design condition of the pressure is 2200 kpa(g) and temperature of 60 ◦ C.
364 10 Control Typicals: Common Calculations

The fluid is natural gas with a molecular weight of 21.5 kg/kmol, compressibility
0.96, and the ratio of specific heat of 1.26.
For square-root-type flow meter calibrated for mass flow measurement, Table 10.10
suggests that the formula in Table 10.9 be used for compensation and conversion.
In actual operation, the flow meter gives a flow reading of 24 kg/s, with a pressure
measurement of 2,100 kpa(g) and a temperature of 58 ◦ C:
1. This 24 kg/s is the indicated flow (raw flow) at the actual condition.
2. To get the compensated mass flow (true flow) at the actual condition, the formula
given in Table 10.9 is as follows:

ρa
FIm,a = FI (10.15)
ρr

where ρr is the design density, and ρa is the actual density. Since the molecular
weight is not known under the actual condition, the compensation is reduced,
based on ρ = ZPRMT , to the following:

ρa
FIm,a = FI
ρr

Pa Tr
≈ FI
Pr Ta

2100 + 101.325 60 + 273.15
= 24 ×
2200 + 101.325 58 + 273.15
= 23.5 kg/s.

The indicated flow from the meter is 24 kg/s, while true compensated flow at
actual condition is 23.5 kg/s.
3. If the compensated volumetric flow at standard condition (pressure = 101.325 kPa
and temperature = 15 ◦ C) is required, then again from Table 10.9, the following
compensation formula can be used:

ρa 1
FIm,a = FI
ρr ρs
 
Pa Tr Ps Ms
≈ FI
Pr Ta Z s R Ts
 
2100 + 101.325 60 + 273.15 101.325 × 21.5
= 24 ×
2200 + 101.325 58 + 273.15 0.96 × 8.314 × (15 + 273.15)
= 24.855 sm3 /s.

The compensated volumetric flow rate at standard condition is thus 24.855 m3 /s.
10.3 Flow Compensation and Conversion 365

4. Now assume that the flow meter is calibrated to produce a standard volumetric flow
of 24.855 sm3 /s; that is, the indicated value at the flow meter is the true volumetric
flow at standard condition. Anti-surge control requires that the raw value at actual
condition be provided. Therefore, the indicated flow must be converted to raw
flow at actual condition.
The same Table 10.9 is used but in reverse order. The raw value at the actual
condition is given by

ρa 1
FI = Fv,s .
ρr ρs

It is easy to verify that a standard flow at 24.855 sm3 /s is 24 kg/s at actual condition.

10.4 Temperature Compensation with Pressure

Gas and steam temperature are heavily correlated with their pressure. For better con-
trol, the temperature measurement often needs to be compensated with the pressure
variation to improve the accuracy.

10.4.1 Application Background

Gas and steam temperature are a function of their pressure. Temperature measure-
ment is calibrated to be accurate for a certain pressure. In practice, the pressure often
fluctuates and could lead to significant temperature measurement errors. Temperature
compensation with pressure is often required. For instance, in a distillation column,
the temperature inside the column is a critical controlled variable because the tem-
perature is used to infer purity. Due to the correlation with pressure, the column
temperature will also increase if the column pressure increases due to inadequate
pressure control. This temperature increase due to pressure change will incorrectly
indicate that the composition has changed, although it is the same composition. Thus,
for more accurate temperature control, which leads to better composition control, the
temperature is often compensated with pressure to eliminate the pressure influence.

10.4.2 Theory and Design

The objective is to derive a pressure-compensated temperature that correlates better

with composition.
366 10 Control Typicals: Common Calculations

Pressure compensation of the vapor temperature is based on an equation to remove

the correlation between pressure and temperature. That is, if the temperature increase
is caused by pressure, this will be automatically excluded. If a composition change
causes an increase in both pressure and temperature, then the part of the temperature
change caused by pressure change will be nullified by the calculation.

Temperature–Pressure Relationship

The basis for temperature compensation is the relationship between temperature and
pressure described by vapor pressure equations. Most of them are semi-empirical
equations derived from the integration of the Clausius–Clapeyron equation. These
equations are for pure components, and the coefficients are obtained by experimental
procedures (Towler and Sinnott 2012).
One of the best known equations is Antoine’s equation given by

B B
log P = A − , or T = −C (10.16)
T +C A − log P

where P is the vapor pressure (absolute) in mmHg, T is the temperature in ◦ C, and

log is the base-10 logarithm function.
Antoine’s equation uniquely relates the pressure and temperature for a pure com-
ponent in vapor–liquid equilibrium. The base-10 logarithm of the pressure is a linear
function of the inverse of the temperature. If there is a pressure swing, the tem-
perature will change immediately according to the ideal gas law. The vapor–liquid
equilibrium will shift, and the pressure and temperature will eventually re-balance
as the purity changes.
The Antoine constants A, B, and C are readily available from chemical engi-
neering data books or the Internet for pure compounds. For example, for the follow-
ing compounds, the constants are given in Table 10.11 (see physical properties in
Appendix C of Towler and Sinnott 2012).

Table 10.11 Antoine Coefficients for some compounds (data from http://ddbonline.ddbst.com and
Towler and Sinnott (2012))
P in mmHg, T in ◦ C P in Pascal, T in Kelvin Valid T Range in◦ C
A B C A B C Tmin Tmax
Methane 6.612 259.63 265.99 8.737 259.63 −7.16 −180 −153
Ethane 6.803 656.40 255.99 8.928 656.40 −17.16 −143 −74
Propane 6.830 813.20 247.99 8.955 247.99 −25.16 −109 −24
n-Butane 6.809 935.86 238.73 9.058 935.86 −34.42 −78 17
iso-Butane 6.748 882.80 240.00 8.873 882.80 −33.15 −86 7
Pentane 6.876 1,075.78 233.21 9.001 1,075.78 −39.94 −50 58
Steam 8.140 1,810.94 244.49 10.265 1,810.94 −28.67 99 374
Water 8.071 1,730.63 233.43 10.196 1,730.63 −39.72 1 100
10.4 Temperature Compensation with Pressure 367

Even though SI units are preferred, the coefficients are customarily given in mmHg
for pressure and ◦ C for temperature. To convert from P in mmHg and T in ◦ C to P in
Pascal and T in Kelvin, use the following equations (the converted results are shown
in Table 10.11):

101325
A + log = A + 2.124903 → A
760
B→B
C − 273.15 → C.

A linear approximation approach may be used for compounds whose exact com-
position is unknown or changes with time. Knowledge of linear regression (or data
fitting) is required. Since Antoine’s equation (Eq. 10.16) is nonlinear between pres-
sure P and temperature T , it needs to be linearized first to use data-based regression.
Rearranging Eq. 10.16 gives the equation that is linear in parameters:

AC − B log P
log P = A + −C
T T
1 log P
= K1 + K2 + K3 . (10.17)
T T
A simplified form of the Antoine equation is the August equation:

B B
log P = A − , or T = . (10.18)
T A − log P

Temperature Compensation with Pressure

There are two common methods for performing pressure compensation on temper-
ature measurement. The rigorous approach is based on accurate knowledge of the
composition of the fluid. For complex compounds, this requirement may not be
realistic, especially if the composition changes online. An approximate data-based
empirical compensation may be applied.
The pressure-compensated temperature (PCT) is the measured temperature, minus
the part caused by pressure change. That is,

Tc = T − T (10.19)

where Tc is the compensated temperature and T is the indicated temperature. T is

the temperature change caused by pressure deviation from baseline pressure Pb and
should be removed from the indicated temperature.
368 10 Control Typicals: Common Calculations

The correction term T can be calculated in several ways, depending on the

equation adopted for the temperature–pressure relationship:
1. Direct compensation based on Antoine’s equation:
   
B B
Tc = T − −C − −C
A − log P A − log Pb
 
B B
=T− − . (10.20)
A − log P A − log Pb

2. Compensation based on Antoine’s equation via local linearization:

dT
Tc = T − (P − Pb )
dP
B
=T− (P − Pb ) (10.21)
ln(10) P (A − log P)2

where log is the base-10 logarithm and ln is the base-e natural logarithm. The tem-
perature gradient dT /d P is calculated from Antoine’s equation (Eq. 10.16):
 
B
d −C
dT A − log P B
= = . (10.22)
dP dP ln(10) P (A − log P)2

This method works only within the base operating point’s close vicinity due to
the assumption of a linear relationship between pressure and temperature, which is
technically not true. The range of validity of the compensation should be specified
in actual implementation, and the calculation result should be clamped accordingly.
If the pressure moves outside this validity range, the clamped value should be used,
and the data quality should be marked as uncertain.

10.4.3 Application Example

Consider a debutanizer column as an example. The bubble points of the vapor inside
the column are produced via dynamic simulation, and the temperatures/pressures are
shown in the first two columns in Table 10.12 (Stephan 2013).

Calculation of Antoine’s Coefficients

Let us calculate Antoine’s coefficients based on the bubbling point data. First, convert
the pressure from kg/cm2 (gauge) to mmHg (absolute) by multiplying the pressure
10.4 Temperature Compensation with Pressure 369

Table 10.12 Bubble points: pressure and temperature in a debutanizer

Temperature Pressure → Pressure Estimated Compensated
T P P T̂ (1) T̂ (2) T̂ (3) T̂ (4) T̂ (5)
◦C kg/cm2 (g) mmHg(a) ◦C

90.00 4.83 4312.8 89.99 90.73 99.963 100.663 98.770

91.25 4.98 4423.1 91.25 91.78 99.965 100.492 98.984
92.50 5.13 4533.4 92.47 92.84 99.987 100.372 99.198
93.75 5.29 4651.1 93.76 93.94 99.950 100.210 99.343
95.00 5.45 4768.8 95.03 95.05 99.935 100.098 99.488
96.25 5.61 4886.5 96.27 96.17 99.940 100.030 99.634
97.50 5.77 5004.2 97.50 97.28 99.966 100.050 99.779
98.75 5.94 5129.2 98.77 98.46 99.936 99.945 99.855
100.00 6.10 5246.9 99.96 99.57 100.000 100.000 100.000
101.25 6.28 5379.3 101.27 100.82 99.937 99.948 100.007
102.50 6.45 5504.4 102.49 101.99 99.967 100.007 100.083
103.75 6.63 5636.8 103.76 103.24 99.945 100.034 100.090
105.00 6.81 5769.2 105.02 104.49 99.943 100.098 100.097
106.25 6.99 5901.6 106.25 105.74 99.959 100.197 100.104
107.50 7.17 6034.0 107.47 106.99 99.994 100.330 100.111
108.75 7.36 6173.7 108.73 108.32 99.981 100.433 100.049
110.00 7.55 6313.5 109.97 109.65 99.986 100.569 99.987
111.25 7.75 6460.6 111.26 111.05 99.945 100.680 99.856
112.50 7.94 6600.3 112.47 112.38 99.986 100.878 99.794
113.75 8.14 6747.5 113.73 113.79 99.982 101.049 99.663
115.00 8.35 6901.9 115.03 115.27 99.934 101.198 99.463
116.25 8.55 7049.0 116.24 116.68 99.966 101.427 99.332
117.50 8.76 7203.5 117.51 118.17 99.954 101.632 99.132
118.75 8.97 7358.0 118.75 119.67 99.960 101.864 98.931
120.00 9.19 7519.8 120.04 121.25 99.925 102.075 98.662

values with 735.559 and adding the atmospheric pressure of 760 mmHg. The result
is in the third column of Table 10.12.
Using the linearized Antoine equation (Eq. 10.17), with the pressure and tem-
perature data in Table 10.12, the coefficients K 1 , K 2 , K 3 can be obtained by
solving 25 simultaneous equations.3 The result is given as K 1 = 6.349, K 2 =
546.630, and K 3 = −217.585. Antoine’s coefficients are then given by Eq. 10.17
as A = K 1 = 6.349, C = −K 3 = 217.585, B = AC − K 2 = 6.349 × 217.585 −
546.630 = 834.730. The values are close to that of n-Butane in Table 10.11. The
estimated temperatures using Eq. 10.16 are shown as T̂ (1) in Table 10.12, very close
to the true temperatures in Column 1. In other words, Antoine’s equation with the

3 With MATLAB/Octave:
K=[ones(size(T)),1./T,log10(P)./T]\log10(P); A=K(1); C=-K(3); B=A*C-K(2); .
370 10 Control Typicals: Common Calculations

above coefficients is an excellent approximation. This close match is not surprising

since the dynamic simulation is likely based on the same equation.
Now let us try the simplified August equation in Eq. 10.18. Solving for parame-
ters A and B with the data in Table 10.12, we have A = 4.594, B = 87.036.4 The
estimated temperature with August’s equation is given as T̂ (2) in Table 10.12. The
simplified August equation produces a close approximation as well.

Temperature Compensation

Up to now, the effort has been on obtaining and validating Antoine’s coefficients
from pressure and temperature data. The data in Table 10.11 is for the vapor in the
debutanizer with a fixed composition. The different pressure/temperature combina-
tions in the table indicate the same composition. The temperature variation is caused
entirely by the pressure change. Therefore, the T term in Eq. 10.19 accounts for the
temperature change caused by pressure change. By subtracting T from the actual
temperature, we obtain the pressure-compensated temperature, which correlates to
the composition and is expected to be a constant value since the composition is
constant in this example.
To use Antoine’s equation for temperature compensation, we have several
approaches as explained in Sect. 10.4.2. Assuming that the base pressure is Pb = 6.10
kg/cm2 (g) (5246.9 mmHg(a)) indicated by the line in bold text in the table, with
Eq. 10.20, the compensated temperature Tc is calculated, and the results are listed as
T̂ (3) in Table 10.12. As expected, the compensated temperature is nearly constant at
100◦ C, since the composition is fixed.
The simplified equation in Eq. 10.21 can also be used, but the results are accurate
only at the vicinity of the base pressure due to the local linearization. See T̂ (4) in
Table 10.12 for the estimated results and Fig. 10.4 for an illustration. This method is
based on piece-wise linearization. The slope dT /d P around T = 100 as calculated
by Eq. 10.22 is 0.0099988.
The methods mentioned above are all based on Antoine’s equation. They assume
that the Antoine coefficients A, B, and C are explicitly available either by table lookup
(for pure compounds) or via data analysis with Eq. 10.20 or 10.21. In practice, there
is a more direct data-based linearization method based on the following:

Tc = T − K (P − Pb ) (10.23)

where K is the slope of the T-P curve in Fig. 10.4. The slope can be calculated by
solving the following equation:

T = K P + K0. (10.24)

4With MATLAB/Octave: x=[ones(size(T)),-1./T] \log10(P); A=x(1);

B=x(2);.
10.4 Temperature Compensation with Pressure 371

Fig. 10.4 Nonlinearity in pressure compensation

For the data in Table 10.12, the K value is given by K = 0.009351.5 K is the overall
T-P slope in Fig. 10.4 based on all the data, which is different from the local slope of
0.0099988 calculated above. The compensated temperatures with Eq. 10.23, assum-
ing a base pressure of 5246.9 mmHg(a), are given as T̂ (5) in Table 10.12, which is
also close to the expected value of 100 ◦ C.

10.4.4 Special Considerations

The example in the previous section assumes that the composition is constant, and
the temperature variation is solely caused by pressure variation. In practice, if actual
operation data is used, either historical data or step test data, the most likely scenario
is that the temperature change results from both composition change and pressure
variation. The challenge is how to separate the two. The recommended approach is
to select a period where the composition is relatively constant, but sufficient pressure
variations are present.

5 With MATLAB/Octave: x=[P,ones(size(T))] \T; K=x(1); K0=x(2);.

372 10 Control Typicals: Common Calculations

10.5 Level Compensation

Liquid level measurement is generally accurate and straightforward and does not
need compensation. However, the multi-phase interface level is notoriously difficult
to measure, and compensation is often necessary to correct the reading. Boiler drum
steam/water interface level is one example.

10.5.1 Application Background

A boiler is prevalent in many operating plants for generating high-pressure steam,

where the boiler drum water level is the most crucial parameter to measure and
control. The detailed process analysis and control strategy are discussed in Sect. 12.2,
and here we only focus on how to improve the level measurement.
One of the challenges for control is the measurement accuracy of the steam/water
interface level in the boiler drum. Any external level measurement device connected
to the steam drum will operate below the actual drum steam/water saturation tem-
perature. The water in the measurement device will thus have a higher density than
that in the drum. Due to the difference in density, the indicated level may differ
significantly from the drum’s actual level.

10.5.2 Theory and Design

The boiler level is indicated either by direct reading from a gauge glass or through
a remote level indicator of various types (ASME 2019). Here, we will explain the
working principles of density compensation for the two most widely used methods:
1. Gauge glass: Gauge glass is a direct reading method. A gauge glass is more
commonly called a gage glass, and is a mandatory requirement by ASME code
(ASME 2019) for any boiler.
2. DP-based level measurement: This indirect measurement is deduced from a dif-
ferential pressure measurement.
Level measurement based on other principles such as displacers, conductivity, and
radar is also common but will not be covered here.

Gauge Glass for Steam/Water Interface Level

As shown in Fig. 10.5, a gauge glass is for direct visual reading of the level in the
drum.
Because the density of the water inside and outside the drum is different, the
indicated drum water level is always lower than its true level. The indicated level
10.5 Level Compensation 373

Fig. 10.5 Boiler drum level

indicator: gauge glass

is h w while the true level is h. H is assumed to be the height between the level
measurement taps, and ρs the steam density. The difference h − h w is the error
caused by the difference in density and is commonly called the density error. For
accurate level reading and control, the density error must be corrected.
The correction is performed by inferring the true level h from the indicated level
h w . The static pressure is related to static height via

P = ρgh (10.25)

where ρ is the density, h is the height, and g is the gravitational acceleration constant.
Based on pressure balance at the steam/water interface level in the gauge, we have

ρ g h = ρw g h w + ρs g (h − h w ) (10.26)

solving for h gives

ρw − ρs
h= hw . (10.27)
ρ − ρs

Here, ρρ−ρ
w −ρs
s
is the compensation factor. It is always greater than 1.0 because ρ ≤ ρw .
Note that if the water inside the drum is the same cold water as in the gauge glass,
i.e., ρw = ρ, then the indicated level is the same as the true level, h = h w .
At the design pressure, the density of water and steam can be looked up from
steam tables. For example, at atmospheric pressure, the densities of the water in the
gauge glass (at 35 ◦ C) and inside the drum (at 100 ◦ C) are ρw = 994 kg/m 3 and ρ =
958 kg/m 3 , respectively, and the density of the saturated steam is ρs = 0.598 kg/m 3
(at 1 atm). The compensated level is then given by

994 − 0.6
h= h w = 1.038 h w . (10.28)
958 − 0.6
374 10 Control Typicals: Common Calculations

An indicated level of 50% is actually 51.9% inside the drum. The density error is
given by

h − hw ρw − ρ 994 − 958
= × 100% = × 100% = 3.6%. (10.29)
h ρw − ρs 994 − 0.6

At 40 bar pressure, the density ρ is 798.4 kg/m3 for water, and 20.10 kg/m3 for
steam. We can easily verify with Eq. 10.29 that the density error can be as high as
20%. A 100% level in the drum may only show as 80% inside the glass gauge!

DP-Based Level Measurement

DP-based transmitters are most commonly used for drum level control. It is based
on the difference in level between the water inside the drum and the external static
column (Fig. 10.6). The level difference generates a pressure difference. This column
is also called the reference leg (or wet leg), which is filled with sub-cooled water from
the drum up to the condenser pot.
The drum level can be inferred from an online DP measurement. A higher DP
indicates a lower water drum level, while a higher DP indicates a lower drum level.
Assume that the water in the reference leg has height H , and the steam density
is ρs . The pressure on the drum side, LP, and the pressure on the reference leg side,
HP, are given by

Fig. 10.6 Boiler drum level measurement: DP-based transmitter

10.5 Level Compensation 375

LP = ρ g h + ρs g (H − h) + ρ0 h 0 (10.30)
HP = ρw H + ρ0 h 0 (10.31)

the pressure difference DP is then given by

P = HP − LP
= (ρw H + ρ0 h 0 ) − (ρ g h + ρs g (H − h) + ρ0 h 0 )
= (ρw − ρs ) g H − (ρ − ρs ) g h. (10.32)

When the drum is filled with the same liquid (e.g., cold water) as in the reference leg
(i.e., h = H , ρ = ρw ), the differential pressure P should be zero.
From Eq. 10.32, the water level h is easily derived from P as

(ρw − ρs ) g H − P
h= (10.33)
(ρ − ρs ) g
ρw − ρs P
= H− . (10.34)
ρ − ρs (ρ − ρs ) g

10.5.3 Implementation and Operation

The water and steam density is a function of the pressure inside the Drum. Therefore,
density compensation is mainly pressure compensation. The densities of the steam
ρs and water ρ are a function of the drum pressure. They are assumed to be known
at design time and remain unchanged after calibration.
However, the drum pressure changes significantly during operation, especially
during startup, shutdown, and major upsets. Real-time online compensation based
on pressure changes becomes mandatory. Typically, the drum pressure measurement
is provided, and an online calculation of the compensation factor is needed. See
Fig. 10.6, where the pressure reading PT-101 is used to look up the water and steam’s
actual density. The calculation is performed in the calculation block LY-101.
The water and steam density can be looked up from the steam tables widely
available in reference books or the internet. Table 10.13 shows some sample values
for the density of water and steam under several different pressures.

10.6 Internal Reflux Versus External Reflux

Reflux flow is a crucial manipulated variable in distillation column control and

directly affects the product quality. However, it is also challenging to use reflux to
achieve the desired control performance because of the strong correlation between
energy and mass.
376 10 Control Typicals: Common Calculations

Table 10.13 Density of water Drum pressure Temperature Water density ρw Steam density ρs
and steam at different pres- ◦C
sures bar(a) kg/m3 kg/m3
1.000 120.2 958.6 0.59
5.000 151.8 915.3 2.67
10.000 179.9 887.1 5.15
15.000 198.3 866.6 7.60
20.000 212.4 849.8 10.05
25.000 224.0 835.1 12.52
30.000 233.9 821.9 15.01
40.000 250.4 798.4 20.10

Internal reflux is a combination of mass transfer and heat transfer and often proves
to be a better handle than external reflux for column control.

10.6.1 Application Background

A distillation column is a standard unit operation involving multi-phase, multi-stage,

and counter-current mass and heat transfer. For many columns, the reflux is used
to control the top and upper tray temperature. The temperature control loop serves
as the inner loop for a cascade control with top product composition control. See
Fig. 5.17 for a typical fractionator column, where the reflux is responsible for the
column overhead control.
Control with reflux can be challenging due to the strong interaction between vapor
temperature and pressure in the column, determined by the heat duty change brought
up by the reflux. Operators often break the cascade loop and directly manipulate the
reflux flow, leading to inconsistent performance.

10.6.2 Theory and Design

Reflux provides down-flowing liquid throughout the rectification section and contacts
the up-flowing vapor to achieve stage-by-stage equilibrium heat and mass transfer.
The heat exchange and mass transfer are complicated and interact with each other.
For this reason, control with external reflux is not accurate.
Heat duty is better than temperature as a measure of heat content. For sub-cooled
reflux flow, the heat duty is defined as

Q = Fm C p T (10.35)
10.6 Internal Reflux Versus External Reflux 377

where Q (in Watt) is the heat duty or the total heat transferred, Fm (in kg/s) is the mass
flow rate for the reflux flow, C p is the heat capacity of the reflux fluid going through
temperature change (at constant pressure), and T is the temperature change in the
reflux after entering the column.
As reflux enters the column, the mass/heat transfer and phase changes are compli-
cated, and it is impossible to be described by flow or temperature alone. The reflux
is typically sub-cooled at the condenser before entering the column. The degree of
sub-cooling varies with the cooling medium, ambient air, or rainstorms and is a sig-
nificant disturbance. Internal reflux is a derived variable that combines mass transfer
and heat transfer and often provides a better indicator of the top product’s purity than
external reflux (Shinskey 1984).
The internal reflux is calculated as

Cp
Ri = R e 1 + (To − Tr ) (10.36)
H

where
Ri internal reflux,
Re external reflux,
Cp heat capacity of liquid reflux,
H heat of condensation of overhead vapor,
To overhead temperature,
Tr reflux temperature, and
the heat of condensation H ≈ heat of vaporization of reflux. Overhead temperature
To is used to approximate the temperature inside the column. The reflux temperature
Tr is assumed to be available.
The internal reflux can then be used as a derived or inferred variable to control the
overhead temperature or top product draw quality, depending on the control strategy.
The internal reflux is a nonlinear function of flow and temperature and can be deemed
a static inferential property (Sect. 5.3.1).

10.6.3 Implementation and Operation

The temperature change T = To − Tr in Eq. 10.36 is assumed to be the differ-

ence in the reflux temperature before and after it enters the column. In practice, the
temperature at the best location may not be measured, but the overhead tempera-
ture is almost always available. So using the overhead temperature as T0 is a close
approximation.
The heat capacity C p and heat of vaporization H can be easily looked up based
on the reflux fluid property. The values depend on the degree of sub-cooling of the
reflux flow. As long as the reflux stays sub-cooled in one phase, it is acceptable to
378 10 Control Typicals: Common Calculations

use a constant value for both. The exact value is not critical since feedback control
can eliminate the error introduced by the C p value inaccuracy.

10.7 Summary

The chapter presents several typical calculations that are commonly encountered in
process control. Most of the calculations in this chapter can be regarded as special
cases of static inferential properties discussed in Sect. 5.3:
1. Signal transformation and conditioning (Sect. 10.1): Signal transformation is
ubiquitous in process control but often hidden in applications. Following the
signal transmission path in the control loops, many types of signal transforma-
tions or processing occur. It is essential to know how these transformations can
impact control performance.
2. Data quality and bad data handling (Sect. 10.2): Reliable measurement is a nec-
essary condition for good control. It is crucial but challenging to monitor the data
quality in real time and assess its usability for closed-loop control. For example,
if a measurement signal is deemed bad or cannot be trusted, the control loop must
take appropriate mitigation measures.
3. Flow compensation and conversion (Sect. 10.3): Flow is not directly measurable
and is always deduced from other measurable quantities such as pressure or veloc-
ity based on a known correlation. The correlation is established under a specific
operating condition. Many times, compensation must be performed to correct or
improve the measurements due to the change in operating conditions. Flow com-
pensation is a requirement and technique that is widely used but often incorrectly
implemented.
4. Temperature compensation with pressure (Sect. 10.4): Vapor or gas temperature
changes with pressure. In many applications, the measured temperature must be
compensated with pressure to arrive at the correct temperature value.
5. Level compensation with density (Sect. 10.5): Steam/water interface level is heav-
ily influenced by the relative density of the steam and water. Inaccurate level mea-
surement is one of the key reasons for poor control in boiler drum level. Online
compensation of the level measurement is essential, especially for high-pressure
steam-generating boilers.
6. Simple inferential properties such as internal reflux (Sect. 10.6): Similar to
pressure-compensated temperature, in an actual application, many process vari-
ables correlate and interact with each other, and a derived variable that “links” or
“lumps” the variables together can provide cleaner and improved process dynamic
responses.
References 379

References

ASME (2019) BPV complete code – 2019, ASME Boiler and pressure vessel code. American
Society of Mechanical Engineers
Shinskey FG (1984) Distillation control: for productivity and energy conservation. McGraw-Hill
Stephan D (2013) How to compensate pressure changes in temperature control. Process-
Worldwide.com. https://www.process-worldwide.com/how-to-compensate-pressure-changes-
in-temperature-control-a-394278/
Towler G, Sinnott R (2012) Chemical engineering design - principles, practice and economics of
plant and process design, 2nd edn. Butterworth-Heinemann
Yokogawa (2004) CS1000/CS3000 reference – function block details. Yokogawa, 11th edn
Chapter 11
Control Typicals: Equipment Control

There are all kinds of equipment in an operating plant, with pumps and compressors
being the most common and essential. This chapter chooses ejectors (Sect. 11.1),
centrifugal pumps (Sect. 11.2), and centrifugal compressors (Sect. 11.3) as typical
equipment examples for detailed analysis and control design. Equipment is part of
process flow, but equipment control has traditionally been the specialty of equipment
manufacturers. With sufficient engineering insight into the working principle, the
same structured process analysis (Chap. 6) and control design (Chap. 7) skills and
methodology can be followed to produce a sound control solution, as demonstrated
with the selected equipment in this chapter.

11.1 Ejector and Educer

Ejectors and educers have been widely used for the pumping and compressing of gas
and liquids, respectively. They have a simple geometry, have no moving parts, and
do not require mechanical or electric drivers; therefore, they are very compact and
cost-effective in construction. They are also very reliable and efficient in operation
(Berkeley 1958; Keenan et al. 1950; Liao 2008).
Ejector and educer are selected as typical equipment in this chapter because they
are probably the simplest and most reliable liquid pumping and gas compressing
equipment. However, their design and operation are backed by very sound theories.
They serve as an excellent introduction to the pump and compressor discussions in
the sections that follow. Ejectors and educers work on the same principle, so we will
use a gas ejector as an example for our discussions in this section.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 381
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_11
382 11 Control Typicals: Equipment Control

11.1.1 Process Description and Analysis

In E&P operations, a gas ejector is commonly found in a glycol dehydration process

(Campbell 2004) to compress the waste gas to a higher pressure to serve a useful
purpose. High-pressure motive gas that might otherwise be throttled for use in an
evaporator or heat exchanger can be used to actuate the ejector. For example, high-
pressure natural gas at 6 bar or higher as motive gas can bring the pressure of the
process gas at atmospheric pressure to one bar or higher.

Operating Requirements

A gas ejector can serve one or more of the following purposes:

1. Transport and compress a weight of process gas from the suction pressure to the
exit pressure.
2. Throttle the high-pressure gas to the desired pressure.
3. Blend two gases to the desired proportion.
The operating requirements are typically specified on the ejector capacity, suction
pressure, and discharge pressure, depending on the operating objectives.

Process Flow Analysis

A gas ejector consists of an actuating nozzle, a suction chamber, and a diffuser. The
high-pressure motive gas (activating gas) is expanded from its initial pressure to a
pressure equal to that of the secondary fluid. See Fig. 11.1 for typical construction.
The process gas enters the suction side and becomes entrained in the motive
gas. The high-pressure motive gas elevates the pressure of the process gas and exits
the ejector mixed as the discharge gas. Inside the ejector, the motive gas induces
a region of low pressure and high-velocity gas flow and causes the process gas to

Fig. 11.1 A gas ejector

11.1 Ejector and Educer 383

become entrained and mixed with the motive gas. During the mixing process, the
actuating fluid is retarded, and the secondary fluid is accelerated. As the mixture
enters the diffuser, it is compressed to the exit pressure by rapid deceleration.
Ejectors can be used singly or in series to create a wide range of vacuum conditions.

Supply and Demand Analysis

As usual, the supply and demand analysis needs to include the entire process sur-
rounding the ejector operation, from the source of supply to the final destination of
the demand. In a simple ejector operation, there are two inlet flow streams and one
exit flow stream:
1. The process gas flow comes from an operating unit such as a glycol dehydration
unit, a vacuum deaerator, and a flash cooler, so the operation is typically supply-
driven by the process gas production.
2. The motive gas compresses the process gas and is thus demand-driven by the
process gas.
3. The exit gas goes to another ejector or a downstream operation unit and is supply-
driven. The downstream operation must have full capacity to consume the exit
gas and absorb its fluctuations.
A recycle flow path is sometimes provided to facilitate capacity control.
The potential measurement instruments and valves are appropriately placed in
line with this supply and demand model. See Fig. 11.2 for the potential valves and
measurements and their locations (see guidelines in Fig. 6.2). A recycle line from
discharge to process gas inlet is sometimes used to help with the turndown operation.

Fig. 11.2 Instrumentation for an ejector

384 11 Control Typicals: Equipment Control

The relationship between the flow streams is given by

[exit flow] = [motive gas flow] + [process flow] + [recycle flow]. (11.1)

Not all valves are needed for all applications. Which one to be in the final control
design is a joint decision by process engineers and process control engineers in early
process design.

Cause and Effect Analysis

The ejector capacity is usually the key control target and is thus a controlled variable.
The motive gas pressure may need to be controlled depending on the supply pressure
and whether the ejector operates at a fixed or variable capacity. The suction pressure
may be a protective control target, depending on the process configuration. The
discharge pressure affects the ejector stability and may need to be a control target.
The control handles for ejector control include one or more of the following:
1. The suction valve PCV-102: The suction valve is primarily for capacity control.
2. The recycle valve PCV-103: The recycle valve adjusts the recycle flow from
discharge to suction side to help capacity control.
3. The flare valve PCV-104 prevents excessively high suction pressure.
4. For variable capacity ejectors, the pressure of the motive gas is often used as one
additional handle for process control, and a throttling valve PCV-101 is thus
provided.
There are three pressures of interest: suction pressure of the process gas PT-102,
inlet pressure of the motive gas PT-101, and the discharge pressure of the mixed
exit gas PT-103. The full cause and effect table with all the potential control handles
and control targets is given by Table 11.1.
Only a subset of the control targets and control handles are necessary, depending
on the final process design. The selected targets and handles will be the controlled
variables and manipulated variables.
For downstream operation, the ejector’s exit gas is supply-driven; a control handle
may need to be provided by the downstream operation to control the ejector discharge
pressure. That is a relay point for the propagation of the supply/demand model further
downstream.

Table 11.1 Cause and effect table for ejector control

Suction pressure Process flow Motive gas pressure
Suction valve ↑   –
Recycle valve ↑   –
Motive gas valve ↑   
11.1 Ejector and Educer 385

Dynamic Response Analysis

A key concept of ejector operation is critical flow and critical condition (Brannan
2018). When a flowing fluid at a given pressure and temperature passes through
an ejector (or any constriction such as an orifice or nozzle in a pipe) to a lower
pressure environment, the velocity of the fluid increases. The flow rate increases
as the upstream pressure increases or the downstream pressure decreases. However,
when the pressure ratio between the upstream pressure and downstream exceeds a
specific limit, the velocity of the fluid would reach sonic speed, and the flow becomes
the so-called critical flow or choked flow, which indicates that the velocity is choked
or limited.
The volumetric flow is the product of gas velocity and cross-sectional area A;
thus, the volume flow becomes constant once the flow is choked at sonic speed:

Fv = A · v (11.2)

where Fv is the volumetric flow, A is the cross-section area of exit, and v is the flow
velocity.
The flow at the critical condition is called the critical flow. Assuming single-phase
ideal gas behavior, the pressure ratio at which the flow becomes choked is given by
  κ−1
κ
P2 2
= (11.3)
P1 κ +1

where P1 is the upstream (motive gas) pressure and P2 is the downstream (exit gas)
pressure; κ = C p /Cv is the ratio of heat capacity, with C p and Cv as the heat capacity
at constant pressure and constant volume, respectively. For most gases, the value of
κ ranges from 1.09 to 1.41. For natural gas with κ ≈ 1.3, a pressure ratio of about
0.5 is required for the gas to reach critical flow (Campbell 2004).
Although the volumetric flow is constant under critical conditions, the choked
mass flow is not. It is a function of the gas density that varies with pressure and
temperature. The choked mass flow rate is given by Perry and Green (1984):

  κ+1
κ−1
2
Fm = Cd A κ ρ1 P1 (11.4)
κ +1

where Cd is the discharge coefficient, ρ1 is the gas density at absolute pressure P1 and
absolute temperature T1 upstream. For an ideal gas, ρ = ZPRMT . Note that the mass
flow rate is primarily dependent on the cross-section area A of the nozzle throat and
the upstream pressure P1 , and is weakly dependent on the upstream temperature T1 .
The flow rate does not depend on the downstream pressure at all.
For completeness, the volumetric flow rate can be derived from Eq. 11.4 as
 
  κ+1
κ−1
  κ+1
κ−1
Fm P1 2 Z RT 2
Fv = = Cd A κ = Cd A κ (11.5)
ρ1 ρ1 κ + 1 M κ +1
386 11 Control Typicals: Equipment Control

where it is seen that the choked volumetric flow is independent of the upstream or
downstream pressures. However, the choked velocity is a function of the gas property
and the temperature.
An ejector can operate in two different modes because of choked flow which are
fixed capacity and variable capacity:
1. A fixed capacity ejector typically runs at the sonic speed with the critical flow.
The pressure of the motive gas does not affect the capacity as long as the pressure
is above the minimum pressure required for stable operation.
2. Variable capacity ejectors typically operate at a subsonic speed with a relatively
lower compression ratio than fixed capacity ejectors. For variable capacity ejec-
tors, the pressure of the motive gas does affect the capacity. That is, the ejector’s
capacity is a function of both the pressure of the motive gas and pressure at the
suction inlet.
Note that the capacity here refers to the exit gas flow, which combines the process
gas flow and motive gas flow (Eq. 11.1). The exact rate of process gas flow is not
easy to calculate. Also, note that the fixed capacity refers to the volumetric flow. The
mass flow changes with the motive gas pressure even after reaching the critical flow
condition (see Eq. 11.4).
Depending on the pressure ratio, the ejector can be running in different operating
conditions:
1. Stable operation: For most ejectors, the pressure of the motive gas must be higher
than a threshold to sustain the stable operation. The value of this minimum pres-
sure is a direct function of the discharge pressure and also depends on whether
the pressure is increasing from unstable operation region to stable region or is
decreasing from stable region to unstable region.1
2. Nonlinear behavior: Before the pressure ratio reaches the critical flow threshold,
a higher motive gas pressure will increase the discharge flow. After reaching the
critical pressure ratio, a further increase in the motive pressure may decrease
the process flow due to the choking effect of the excess gas in the diffuser throat,
because the exit gas flow is the total of motive gas flow and process flow (Berkeley
1958). The motive gas pressure setpoint should be carefully selected.
3. Discharge pressure: Ejectors are sensitive to discharge pressure. The ejector may
become unstable if the discharge pressure exceeds a high-pressure limit, and the
relationship between process flow and suction pressure may change.

1 The former is called “pickup” pressure, while the latter is called “break” pressure. The “pickup”
pressure is always higher than the “break” pressure. Operation between the “pickup” pressure and
the “break” pressure can be either stable or unstable depending on which direction the pressure is
changing (Berkeley 1958).
11.1 Ejector and Educer 387

11.1.2 Control Strategy

The control objective serves the operating objective, and the control strategy is subject
to the process design and operating constraints. The available motive gas pressure,
and downstream pressure, and capacity are all factors to consider.

Control Structure

The control requirements include both normal regulatory control and abnormal pro-
tective control; thus, a layered control design is recommended:
1. Normal regulatory control: Capacity control is the primary control objective in
normal operation. Capacity control is achieved through suction pressure control
as the capacity change is reflected in pressure change.
The requirements for suction pressure control are determined by upstream oper-
ation. Some process operations may require that a constant suction pressure be
maintained, while most others do not.
2. Abnormal protective control: Depending on the operating mode (fixed capacity
versus variable capacity) and process configuration, protective control against
high and low pressures may be needed (see control schematics in Fig. 11.2):
a. Low-pressure protection: If the process unit that produces the off-gas is fully
rated for vacuum operation, low-pressure protection is not needed, and the
recycle valve PCV-103 is not required. The suction side can float with the
pressure of the process unit, even if it draws a vacuum.
b. High-pressure protection: In case of process upset or ejector shutdown, the
suction pressure may go excessively high. High pressure protection should be
provided. PCV-104 is provided to send the excess gas to flare.
c. Motive gas pressure: If the ejector operates in fixed capacity mode, low-
pressure protection on the motive gas inlet may be required to ensure that
the pressure ratio of the ejector is above the critical flow threshold. In the
case of a variable capacity ejector, the pressure of the motive gas needs to be
controlled for stable operation of the ejector.

11.1.3 Functional Design

Figure 11.3 is a typical control scheme for fixed capacity ejectors where the pressure
of the motive gas is controlled by PC-101 at the desired pressure to maintain stable
ejector operation, and the suction valve is used to control the suction pressure via
PC-102. If a high turndown ratio is required, the discharge recycle line can be added.
The suction pressure control PC-103 can manipulate the recycle valve to maintain
388 11 Control Typicals: Equipment Control

Fig. 11.3 Control design for an ejector: fixed capacity

the desired turndown ratio. The high-pressure protective controller PC-104 provides
high-pressure protection.
Figure 11.4 is a control scheme for variable capacity ejectors. In comparison with
Fig. 11.3, the recycle line is removed. The pressure PC-101 of the motive gas is
used under cascade control to achieve the desired turndown ratio since recycling is
inefficient. Note that a minimum pressure (via low setpoint limit or minimum valve
opening) should be imposed on the motive gas to ensure that the ejector operates in
the stable region.
If the suction side is fully rated for vacuum operation, the control scheme can
be simplified by eliminating the suction pressure controller PC-103. The high-
pressure protective controller PC-104 is still needed to flare the excess gas if the
suction pressure does go too high. See Fig. 11.5.

11.1.4 Implementation and Operation

All controllers are either standalone PID controllers or standard complex PID loops.
The configuration of the final control scheme can follow the checklist in Table 3.9
and is omitted here.
Multiple ejectors can be used in series to boost the pressure. The process and
control design can be analyzed in the same way.
11.1 Ejector and Educer 389

Fig. 11.4 Control design for an ejector: variable capacity

Fig. 11.5 Control design for an ejector: fully rated for vacuum

11.2 Centrifugal Pump

Pumps are extremely widely used. A 2001 report by the United States Department
of Energy claims that “... pumping systems account for nearly 20% of the world’s
electrical energy demand and range from 25 to 50% of the energy usage in cer-
tain industrial plant operations” (DOE 2001). Therefore, the reliable and optimal
operation of pumps is of great economic value.
390 11 Control Typicals: Equipment Control

11.2.1 Process Description and Analysis

Pumps are prevalent in process facilities to move liquid from one location to another
or raise the liquid pressure to the desired value. There are many types of pumps for
different application needs. Here we will focus on centrifugal pumps only.

Operating Requirements

The operating objective of a pumping process is to deliver the desired head and
flow during normal operation and protect the pump from potentially unsafe and
inefficient conditions such as surge and high pressure during abnormal operating
operations. The head and flow of the pump define the operating point. The minimum
speed, maximum speed, minimum flow, maximum flow, and maximum motor power
typically define the operating envelope.
1. Minimum continuous stable flow (MCSF):2 Operation of centrifugal pumps below
their minimum flow requirements is the primary cause of premature pump failure.
Hydraulic instability occurs at low flows and can cause surge, cavitation, and
excessive vibration in the pump. The type and extent of damage to the pump
depend on how long the low flow persists and the magnitude of the generated
forces and vibrations. The surge phenomenon is a common threat to all centrifugal
and axial rotating equipment. The typical solution is to provide automatic bypass
circulation through a minimum flow controller and recycle control valve.
2. Operation of centrifugal pumps above their maximum flow tolerance (end-of-
curve) results in very inefficient operation and can cause vibrations and typically
should be avoided as well.

Process Flow Analysis

Figure 11.6 shows a typical pumping process. The center of the process is a centrifugal
pump, where the mechanical energy from the driver (e.g., a motor) is injected into
the liquid to increase its potential and kinetic energy. The potential energy is the
increase in the pump head, while the kinetic energy is reflected in the change of the
pump flow.
Pumps are generally motor-driven and operate at either fixed speed or variable
speed. Based on operating needs, multiple pumps can operate in parallel or serial.
Therefore, depending on the type of pump and the operating requirements, the piping
and instrumentation diagrams (P&ID) can be very different and complex.

2 There are three types of minimum flows for a centrifugal pump: minimum continuous thermal flow
(MCTF), intermittent minimum flow, and minimum continuous stable flow (MCSF). By minimum
flow, we typically refer to the minimum continuous stable flow (MCSF). MCSF is a value that can
range from roughly 10% to 80% of the best efficiency point flow depending on pump size and type,
operating speed, impeller suction geometry, liquid density, and other factors.
11.2 Centrifugal Pump 391

Fig. 11.6 A typical pumping process, with difference in elevation

Supply and Demand Analysis

A pumping process can be as simple as a single pump with one stream coming in
and going out. It can also be as complicated as multiple pumps operating in parallel
or serial with multiple streams merging and branching out. The supply and demand
model can also vary from supply-driven, demand-driven, to mixed.
Figure 11.7 is a simple process flow configured as a supply-driven operation with
a loop-back. The same process flow can easily be configured as a demand-driven
operation by relocating the measurements and control valves.
The loop-back is a recycle flow from pump discharge to the tank to ensure that
the pump flow is above the minimum flow requirement. Obviously,

[pump flow] = [process flow] + [recycle flow]. (11.6)

The recycle flow remains closed when the pump flow is above the minimum flow
limit. In case the process flow falls below the minimum flow limit, e.g., during startup
or turndown operation, the recycle valve opens, and the recycle flow supplements the
process flow to bring the pump flow above the minimum flow limit. With a properly
sized recycle line and valve, the pump operation can achieve 100% turndown, i.e.,
full recycle and zero process flow.

Fig. 11.7 Flows in pumping process

392 11 Control Typicals: Equipment Control

Cause and Effect Analysis

When analyzing the cause and effect relationship of a pumping process, both the
pump and the surrounding process flow components must be considered altogether.
The key to understanding the pump characteristics can be summarized as three
variables, two curves, and one point. The three variables are the pump head, flow,
and speed; the two curves are the performance curve and resistance curve, and the
one point is the operating point:
1. Head: The pump head is the liquid force measured by the height of elevation,
typically by meters or inches of the liquid.
2. Flow: Pump flow is the liquid volume flow through the pump. The pump flow
rate is also called the pump capacity.
3. Speed: The pump speed is the pump shaft speed measured in revolutions per
minute (rpm).
The relationship between the head, flow, and speed defines the pump operation.
The relationship can be graphically represented by the so-called pump performance
curves or simply the pump curves. An example of the pump curve is given in Fig. 11.8.
The head–flow relationship at a constant speed is represented by one head–flow
pump curve. If the pump speed can vary, then there would be a range of parallel head–
flow pump curves. As pump speed changes, the operating point moves up and down,
following roughly a parabolic path. This trajectory is called the system resistance
curve.
The pump curves do not change once the pump is built. However, the resistance
curve can move or be moved by changing the process flow components such as a
throttling valve position or recycle flow. These adjustable components serve as the
control handles to influence the operating conditions:

Fig. 11.8 A typical pump curve

11.2 Centrifugal Pump 393

Table 11.2 Cause and effect table for a centrifugal pump

Pump head Pump flow Process flow Recycle flow
Pump speed ↑    –
Discharge valve ↑    –
Recycle valve ↑  –  

1. Discharge valve: For fixed-speed pumping, a discharge control valve is typically

used to adjust the operating point.
2. Recycle valve: The recycle control valve is typically reserved to maintain the
minimum pump flow.
The pump performance curve determines where the operating point can be, while
the system resistance determines where the operating point should be; the former is
determined by the mechanical characteristic of the pump while the latter is desired
by process operation. Thus, the operating point is always at the intersection of the
pump performance curve and the system resistance curve. By changing the pump
speed and system resistance, the operating point can be moved to anywhere on the
performance map (subject to constraints).
The controlled variable is typically the pump flow (pumping capacity). The manip-
ulated variable is the pump speed in a variable-speed pump or discharge throttling
valve for a fixed-speed pump. The cause and effect analysis of a pumping process
can be simplified to four control targets and three control handles. Depending on the
application, all the control handles are not necessary.

Dynamic Response Analysis

The power, head, flow, and speed are related by the so-called pump affinity laws,
which are more popularly known as the fan laws. The fan laws state the following:
1. Change in pump flow is directly proportional to change in velocity or speed.
2. Change in pump head is directly proportional to the square of the change in
velocity or speed.
3. Change in Power is directly proportional to the cube of the change in velocity or
speed.
In equation format,
 3  2
W2 N2 H2 N2 Fv,2 N2
= , = , = (11.7)
W1 N1 H1 N1 Fv,1 N1

where W is the brake horsepower (BHP), N is the speed, H is the pump head, and
Fv is the volume flow rate. Subscripts “1 ” and “2 ” denote two operating conditions.
The affinity laws imply that if the speed of a pump is increased by 10% (i.e.,
N1 /N2 = 1.10), then the changes in flow, head, and power are given as follows:
394 11 Control Typicals: Equipment Control

1. The volumetric flow increases by 10%: Fv,2 /Fv,1 = 1.10.

2. The head increases by 21%: H2 /H1 = 1.12 = 1.21.
3. The power consumption increases by 33%: W2 /W1 = 1.13 = 1.33.
The best efficiency points (BEP) are the flow rates at which the pump unit’s com-
bined hydraulic losses are the lowest. Moving the operating point along the system
resistance curve does not cause drastic changes in pump efficiency. On the other
hand, moving the operating point along the performance curve can cause significant
changes in pump efficiency. See the pump efficiency curves in Fig. 11.8.
Capacity control is usually achieved through pump speed or via a discharge valve;
see Table 11.3. In specific applications, the capacity control may be through discharge
pressure control or direct flow control. One decisive factor is the dynamic character-
istics. For example, a flat performance curve results in a low process gain for pressure
change and makes the capacity control a challenge. It is typically a joint decision
between process engineering, rotating equipment, and process control to determine
which control targets to use the controlled variables and which control handles to
use as manipulated variables.
For completeness, the performance curves from pump manufacturers typically
include the following information, as shown in Fig. 11.8. They are all important
considerations for abnormal protective control:
• Head versus mass or volume flow.
• Efficiency versus mass or volume flow.
• Brake horsepower (BHP) versus mass or volume flow. Brake power is the energy
needed to pump the liquid.
• Best efficiency point (BEP).
• Minimum flow limit.
• Choked flow limit.
• Operating limits: Minimum speed and maximum speed.
• Operating limits: Minimum flow (surge line) and maximum flow (end of curve
line).

11.2.2 Control Strategy

Centrifugal pumps are widely used in many industries and process operations. The
operating objectives and requirements can significantly vary. It is impossible to cover
all the application scenarios. As part of process control, pump control must be consid-
ered within the overall control strategy. The critical connection with the plant-wide
control is the supply and demand balancing, which may affect the supply-driven or

Table 11.3 Cause and effect Discharge pressure Pump flow

table for pump capacity
control Pump speed ↑  
Discharge valve position ↑  
11.2 Centrifugal Pump 395

demand-driven control. The best approach is to follow the systematic methodology

discussed in Chap. 7 to arrive at the best fit-for-purpose design. Here, we will only
present some typical control designs.

Control Objectives

The main objective of pump control is to provide the desired throughput (capacity)
during normal operation, and keep the pump operation within the operating envelope
during abnormal operating conditions.

Control Structure

The control design follows the layered approach of normal regulatory control and
abnormal protective control. Equation 11.8 provides a list of the full scope of pump
control and protection. Not all features are required for all applications:

⎧ ⎧
⎪
⎪ ⎪
⎪ Capacity Fixed Speed Pump
⎪
⎪ ⎪
⎪
⎪
⎪ ⎨ Control
⎪
⎪ Regulatory Variable Speed Pump
⎪
⎪
⎪
⎪ Control ⎪
⎪ Pumps in Parallel
⎪
⎪ ⎪
⎪
Load
⎪
⎪ ⎩ Balancing
⎪
⎪ Pumps in Series
⎪
⎪
⎪
⎪ ⎧ ⎧
⎪
⎪ ⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪
Minimum Flow (Surge)
⎪
⎪ ⎪
⎪ ⎨Maximum Flow (End of Curve)
⎪
⎪ ⎪
⎪ Envelope
⎪
⎪ ⎪
⎪
⎪
⎪ ⎨ Protection ⎪
⎪High Speed
Pump ⎨ Protective ⎪
⎩
Control ⎪ Low Speed
Control ⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ Discharge Pressure
⎪
⎪ ⎪
⎪
Constraint
⎪
⎪ ⎩ Handling
⎪
⎪ Motor Power
⎪
⎪ ⎧
⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ Startup Cold Startup
⎪
⎪ ⎪
⎨ Sequence
⎪
⎪ Sequential Hot Swapping
⎪
⎪
⎪
⎪ Control ⎪
⎪
⎪ ⎪ Shutdown
⎪ Emergency Shutdown
⎪
⎪ ⎪
⎩ Sequence
⎪
⎪ Planned Shutdown
⎪
⎪
⎪
⎩ Instrumented
Safeguarding
(11.8)

1. Normal regulatory control: Normal regulatory control is typically about quality

and capacity. There is no quality specification for pumping operation, so the only
normal regulatory control requirement is pump capacity control.
For fixed-speed pumps, capacity control is typically achieved by discharge throt-
tling with a control valve. Throttling the discharge value increases the discharge
pressure and causes the operating point to move to the left along the pump curve.
396 11 Control Typicals: Equipment Control

That is, higher pressure leads to lower flow and vice versa.
However, valve throttling results in permanent pressure losses. The more efficient
way is to use a variable-speed pump with which the pump capacity is adjusted by
varying the pump speed. Reducing the pump speed will cause both the head and
flow to decrease.
The recycle flow can also be used to adjust the pump capacity, although it is very
inefficient.
2. Abnormal protective control: The objectives of protective controls are to protect
both the pump and the process. Protective controls for the process depend on
the overall process flow, and thus are application-specific. Typical pump protec-
tion includes minimum flow (surge) control and maximum flow (end-of-curve)
control:
a. Minimum flow: Centrifugal pump suffers from the damaging phenomenon of
surge when the flow throughput is below the minimum flow requirement. The
minimum flow is typically maintained by recirculating a portion of the process
flow back to the suction side to prevent surge via a simple PID controller. For a
fixed-speed pump, the minimum flow limit is a fixed value, and the controller
has a fixed setpoint. For a variable-speed pump, however, the minimum flow
value is different at different speeds, and thus a more elaborated control scheme
is needed for efficient control.
b. Maximum flow: The maximum flow protection is also called the end-of-curve
protection. The pump is designed to operate in a specific range of back pres-
sures. In real applications, the discharge pressure may be different during
startup and at different operating conditions. If the back pressure falls too low,
i.e., the system resistance is too little, it could drive the pumps to run at the end
of the performance curve or beyond and results in very inefficient operation
and even potential damage to the pump. See Fig. 11.8. End-of-curve control
is typically achieved by reducing the pump speed or throttling a control valve
on the discharge.
3. Sequential Control: Pump startup is a routine operation and sometimes involves
risks in itself. An automated startup sequence is desired to ensure a consistently
safe and smooth startup of the pump.

11.2.3 Functional Design

The functional design of process control varies with process design. This section
provides several typical designs to illustrate the general design principle.
11.2 Centrifugal Pump 397

Capacity Control

Pump capacity control is to maintain the operating point at the desired flow and head.
The pump discharge flow control usually serves as the capacity control. From the
cause and effect table (Table 11.2), there are two common approaches:
1. For variable-speed pumping, manipulating the pump speed to control the pump
capacity (flow).
2. For fixed-speed pumping, throttling the discharge valve to change the pump flow.
Figure 11.9 is a typical control design for a supply-driven fixed-speed pumping
process. FC-101 is the process flow, and FC-102 is the pump flow. The control
target for the flow is usually supplied by the tank level control for supply-driven
operation. If the operation is demand-driven, which is less common, the flow set-
point may be supplied manually by the operator or automatically by a downstream
controller.
The performance curve in Fig. 11.8 correlates the pump head, flow, and speed.
If two are specified, the third is determined. A higher tank level demands a higher
discharge flow, and a high flow will result in a lower head (and thus the discharge
pressure).
For a variable-speed pump, the pump speed is the manipulated variable for capac-
ity control. Figure 11.10 shows a typical control design. With a low tank level, the
level controller demands a lower flow rate. The flow controller will decrease the
pump speed to achieve a lower pump flow (also a lower head). If the required flow
rate is below the pump’s minimum flow limit, the recycle valve will open to increase
the recycle flow, helping reduce the process flow further while maintaining the pump
flow above the limit. The process flow, recycle flow, and pump flow are related by
Eq. 11.6 and illustrated in Fig. 11.11a.
From the pump performance curves in Fig. 11.8, the minimum flow limit decreases
with speed. The amount of recycle flow also decreases. This change is illustrated in
Fig. 11.11b. Since recycling is a very inefficient operation, the recycling flow should
be reduced to the minimum possible.

Fig. 11.9 Capacity control for fixed-speed pumping

398 11 Control Typicals: Equipment Control

Minimum Flow Control

One of the leading causes of pump damage is running the pump below the minimum
flow limit. Maintaining the pump flow (not the process flow! see Fig. 11.11) above
the minimum flow limit is mandatory for most large-capacity centrifugal pumps.
The standard approach is to recirculate a portion (or all if necessary) of the discharge
flow back to the suction side. See the flow controller FC-102 in both Figs. 11.9 and
11.10. The minimum flow controller FC-102 is a standalone protective controller
discussed in Sect. 4.1.3.
The challenge for minimum flow control, however, is in determining the flow
control setpoint. There is one single performance curve for a fixed-speed pump, and
thus one fixed minimum flow limit value, as shown by the point marked as MIN

Fig. 11.10 Capacity control for variable-speed pumping

Fig. 11.11 Process flow, recycle flow, and pump flow

11.2 Centrifugal Pump 399

Fig. 11.12 Head–flow pump curve

FLOW in Fig. 11.12. This minimum flow value serves as the setpoint for the flow
controller.
However, for a variable-speed pump, there is a performance curve for every speed,
and the minimum flow limit is a function of the pump speed. The parabolic line  3
in Fig. 11.12 indicates the minimum flow values at different speeds. Suppose a fixed
flow setpoint is used, indicated by the vertical dashed line .
1 In that case, the area
between line  1 and  3 will not be reachable by the pump operation, because if the
pump needs to operate at the flow rate lower than indicated by line , 1 then recycle
starts from line 1 instead of . 3 See Fig. 11.11b for the two scenarios where the
recycle flow F1>F2. This unnecessarily high setpoint is a waste of energy since
recycling is very inefficient for capacity control.
The approach to address this inefficiency is to use a variable flow control setpoint
as indicated by line .
3 There are multiple ways of achieving this:
1. Variable setpoints for the flow controller can be used to maintain the minimum
recycle flow. It is easy to derive a function that relates the flow rate with the pump
head or speed from the performance curve. The pump affinity laws state that the
pump head change is a quadratic function of the change in the flow rate. That is,
 2
H Fv H0
= → H= 2
· Fv2 = K · Fv2 . (11.9)
H0 Fv,0 Fv,0

In other words, the parabolic resistance curves in the head (H ) versus flow (Fv )
coordinate in Fig. 11.13a become a straight -line in the head (H ) versus squared
flow (Fv2 ) coordinate in Fig. 11.13b.
The line that links all the minimum flow points at the different pump speeds is
called the surge line. A straight line approximation to the surge line is given by
fitting Eq. (11.9) to the surge points. Picking one of the surge points (for example,
1220 m, 114 m3 /hr) as (H0 , Fv,0 ), then from Eq. (11.9), the minimum flow setpoint
is derived as
400 11 Control Typicals: Equipment Control

H0 Fv,0 √
H= 2
· Fv2 → Fv = √ H
Fv,0 H0
114 √ √
≈√ H = 3.26 H (m3 /hr). (11.10)
1220
Pump manufacturers typically provide the pump curves as head-to-flow relation-
ships. However, the pump head H is not directly measurable in practical appli-
cations, and Eq. 11.10 cannot be used online. The pump suction and discharge
pressures are usually used in place of the pump head. The pump head is related
to the differential pressure by

Pd − Ps = ρ g H (Pa = kg/m3 · m/s2 · m) (11.11)

where Ps and Pd (in Pascal) are the suction and discharge pressure, respectively;
H (in meters) is the pump head; ρ (in kg/m3 ) is the density of the liquid being
pumped. Assuming that the density of the liquid is 960 kg/m3 , then the setpoint
for the minimum flow controller given by Eq. (11.10) becomes

Fv,0 √ Fv,0 Pd − Ps Fv,0
Fv = √ H=√ =√ Pd − Ps (11.12)
H0 H0 ρg H0 ρ g
114
≈√ Pd − Ps = 0.0336 Pd − Ps (m3 /hr). (11.13)
1220 × 960 × 9.8
The pump speed can also be used to relate to the minimum flow limit. However,
the online speed measurement is not as fast and reliable as the head measurement
and is not recommended for real-time control.
2. Minimum flow control via a surge indicator. Another approach is to define an
intuitive surge indicator. On the surge reference line, this surge indicator has a
constant value regardless of pump speed. For any other points on the pump curves,
there is a corresponding surge indicator that indicates its location relative to the
surge reference line.
One definition of this indicator is the ratio of the slopes between the operating
point and the surge reference line (see Fig. 11.13b). The slope in the Head–Flow
coordinate is given by
S = H/Fv2 . (11.14)
Assume that the slope of the surge reference line is S0 = H0 /Fv,0
2
, and the slope
of the current operating point is S = H/Fv , then the ratio of slope between the
2

surge reference line and the operating point is given by

11.2 Centrifugal Pump 401

Fig. 11.13 Head–flow and head–flow-squared curve

 √
S0 H0 /Fv,0
2
H0 Fv
KS = = 2
= √ (11.15)
S H/Fv Fv,0 H
√ √
H0 Fv ρ g H0 Fv
= = √ . (11.16)
Fv,0 Pd − Ps Fv,0 Pd − Ps
ρg
Obviously, Eq. 11.16 is a different form of Eq. 11.12. The absolute value of K S
indicates the current operating point:
⎧
⎪
⎨= 1.0 on the surge reference line
K s > 1.0 on the right of the surge reference line (11.17)
⎪
⎩
< 1.0 on the left of the surge reference line.

Fig. 11.14 Protective flow control for centrifugal pump

402 11 Control Typicals: Equipment Control

The values of H0 and Fv,0 are obtained from the pump curves, while the values of
Pd , Ps , and Fv are from the online measurement. If the pump curve is provided as
head-versus-flow, then Eq. 11.15 should be used. However, in the less common
circumstances, the pump curve may be provided as a pressure-versus-flow rela-
tionship, then Eq. 11.16 should be used.
Figure 11.14 shows the control schematic, where PI-101 is the suction pressure
Ps , PI-102 is the discharge pressure Pd , and FI-102 provides the pump flow
Fv . The calculation block XY-102 performs the calculation in Eq. 11.16, whose
result is fed to the minimum flow protective controller XC-102 as PV.
The surge reference line is assumed to have a value of 1.0. For safety reasons, the
controller setpoint is set to be 10% higher than the surge reference line value, i.e.,
1.10. The controller output is sent down to recycle valve FCV-102.
If the flow is measured with a DP-based flow meter such as an orifice
√ plate, the
differential pressure P across the orifice is available as Fv = Cv P/ρ where
Cv is the flow constant. The above formula (11.16) can be further simplified into

P
√ √ Cv
H0 Fv H0 ρ
KS = √ =
Fv,0 H Fv,0 Pd − Ps
ρg
√  
g H0 P P
= Cv =C . (11.18)
Fv,0 Pd − Ps Pd − Ps

We will come back to this formula when we discuss the compressor anti-surge
control in the next section.

End-of-Curve Control

The end of curve for a pump is also called stonewall in compressor operation. At
the end of the pump curve, the pump flow approaches sonic speed and can no longer
increase beyond this point. The end-of-curve flow varies with pump speed, as shown
in Fig. 11.12. If a fixed maximum flow limit is used, line 2 would be the maximum
flow that meets the maximum flow limits for all speeds. However, this line is too
conservative. If the minimum and maximum flow limits are both chosen as fixed
values, the feasible operation region would be limited to the shaded area between the
two vertical lines  1 and ,
2 which does not even enclose the best efficiency point
(BEP). Therefore, variable setpoints similar to that in minimum flow control must
be used. The flow setpoints approximately follow line . 4
The end-of-curve controller XC-101 is an overriding protective controller on the
pump discharge flow; see Fig. 11.14. If the capacity control requires a flow rate that
is over the end of the pump curve, the override control will override the capacity
control and limit the flow to below the maximum flow limit.
11.2 Centrifugal Pump 403

The surge indicator calculated by Eqs. 11.16 and 11.18 produces a value√of 1.0 for
√ produce a value of approximately 10 at the
the surge line. The same equation would
end of the curve, consistent with the 10:1 turndown ratio of pressure measurement.
We will use the end-of-curve control to show an alternative procedure to calculate
the surge indicator. Different engineering units for pressure and flow measurements
are used on purpose in this example to demonstrate the unit conversion required in
the calculation.
Example 11.1 Surge indicator calculation. Figure 11.14 is the pumping process
schematic, and the pump curve is given in Fig. 11.12. The pressure measurements
PI-101 and PI-102 are provided in kPa(g). The flow measurements FI-101 and
FI-102 are provided as mass flow in kg/s. Now we use a simplified form Eq. 11.16
as √
Fm Pd − Ps
KS = C √ → C = (with K S = 1) (11.19)
Pd − Ps Fm
where C is a constant to be determined. We have assumed that the surge points have
a surge indicator value K S of 1.0; we can readily determine the constant value C
from the pump curve.
Let us choose a few surge points from the pump curve to calculate the C value. The
pump curve is given as head (meter) versus volume flow (m3 /hr). The head and flow
data must be converted to kPa(g) and kg/s to be the same as the online measurements.
Conversion from head to differential pressure can follow Eq. 11.11, while mass flow
and volume flow are related via

Fm = ρ Fv (kg/s = kg/m3 · m3 /s). (11.20)

Assuming the liquid density is 960 kg/m3 , the raw data for the surge points are read
from the pump curves. The converted data with the same engineering units as online
measurements are listed in Table 11.4. The calculated values for the constant C with
Eq. 11.19 are listed in the last column.
For safer operation, let us take the most conservative value of C from the last
column, 3.42. The surge indicator is then given by
Fv
K S = 3.42 √ . (11.21)
Pd − Ps

Table 11.4 Surge points from pump curves

Pump head Volume flow Pd − Ps Fm C from Eq. (11.19)
√
(m) (m3 /hr) (kPa) (kg/s) ( kPa kg/s)
Surge point 1 1220 114 11477.8 31.0 3.45
Surge point 2 1000 104 9408.0 28.3 3.43
Surge point 3 780 90 7338.2 24.5 3.50
Surge point 4 600 79 5644.8 21.5 3.49
Surge point 5 440 68 4139.5 18.5 3.48
Surge point 6 300 57 2822.4 15.5 3.42
404 11 Control Typicals: Equipment Control

Table 11.5 Pump operating range by surge indicator

Pump head Volume flow Pd − Ps Fm K S from Eq. (11.21)
(m) (m3 /h) (kPa) (kg/s) (–)
Surge point 1 1220 114 11477.8 31.0 0.99
Surge point 2 1000 104 9408.0 28.3 1.00
Surge point 3 780 90 7338.2 24.5 0.98
Surge point 4 600 79 5644.8 21.5 0.98
Surge point 5 440 68 4139.5 18.5 0.98
Surge point 6 300 57 2822.4 15.5 1.00
Choke point 1 950 315 8937.6 85.8 3.10
Choke point 2 770 285 7244.2 77.6 3.12
Choke point 3 605 252 5691.8 68.6 3.11
Choke point 4 470 220 4421.8 59.9 3.08
Choke point 5 340 190 3198.7 51.7 3.13
Choke point 6 240 158 2257.9 43.0 3.10

Equation 11.21 is the formula for the online calculation block XY-102 in Fig. 11.14.
We can verify the surge points and choke (end-of-curve) points, as shown in
Table 11.5.
The end-of-curve control is a standard protective controller with the PV calculated
from Eq. 11.21 and a setpoint set to 3.10. See XC-101 in Fig. 11.14.
In summary, by limiting the surge indicator value K S to between 1.0 and 3.1,
the pump effectively operates between the minimum flow limit defined by the surge
reference line and the maximum flow limit defined by the end-of-curve points. For
safety reasons, a margin of approximately 10% should be given to both setpoints,
resulting in SP = 1.1 for the minimum flow controller and SP = 2.8 for the end-of-
curve controller.

11.2.4 Implementation and Operation

The actual implementation depends on the functional design and can vary from appli-
cation to application. Here, a few key considerations are presented and discussed.

PID Configuration

The minimum recycle flow control FC-102 is a simple protective flow control
(see Sect. 4.1.3 for protective control), with the setpoint provided by Eq. 11.12. The
setpoint calculation is typically implemented in a standard CALC function block in
DCS. To the minimum flow controller, the setpoint is provided by another program,
and thus the normal controller mode should be RCAS or equivalent (see Sect. 3.2.1 for
RCAS mode). Controllers FC-101 and XC-101 may override each other, and anti-
11.2 Centrifugal Pump 405

Table 11.6 Loop configuration details for pump control

Configuration LC-101 FC-101 XC-101 XC-102
Description Tank level Pump discharge Minimum flow End-of-curve flow
Equation type I-PD PI-D I-PD I-PD
Normal mode AUTO CAS AUTO AUTO
Action direction DIRECT REVERSE REVERSE REVERSE
Anti-reset windup n/a YES n/a YES
PV-tracking Disabled Enabled Disabled Disabled
Output scaling 0% = fully closed, 100% = fully open

reset windup needs to be correctly configured (see Sect. 4.3.3 for anti-reset windup)
(Table 11.6).

Multi-pump Operation

Multiple pumps can run in parallel or series to deliver a higher pressure or flow.
Typical pump configurations and control include the following:
1. Pumps in series: Booster pump + main pump. Pumps in series share the same flow,
but the pump head (thus pressure) is extended. The startup of serially operating
pumps is challenging.
2. Pumps in parallel: Pumps running in parallel extend the overall capacity. One of
the considerations is load balancing, i.e., how to distribute the capacity among
the pumps. Common approaches include split-range control and fan-out control.
Switching the pumps on and off creates a significant disturbance to operation and
is a challenge for control design.
Due to the length limit, we will not detail the control complexity of multi-pump
operations.

Sequential Control

Pump startup is a routine operation. However, frequent startup and shutdown of the
pump are harmful to the pump due to the higher electric current required to start the
pump, which leads to motor heating. The electric current required during startup can
be ten times higher than keeping the pump at a normal speed. Since motor heating
is proportional to the current squared (I 2 R), the heat generated during startup may
be 100 times higher than normal. Typical startup scenarios include the following:
1. Single pump: An operator manually starts and stops the pump based on operational
requirements.
406 11 Control Typicals: Equipment Control

2. Duty/standby: Standby pump automatically starts upon failure of duty pump or

electrical/vibration trip of the duty pump. Pump switching may cause a temporary
disturbance to the flow control.
3. Normal auto-start: With two pumps, the standby pump starts automatically based
on a process variable that provides the switching point corresponding to a specific
operating condition. For example, the process variable could be a low flow or low-
pressure switch point.
4. Special auto-start: With three or more pumps operating in parallel, the standby
pump can be selected via a selector switch. A duty/standby switch may be required
to designate the standby pump.
The startup can be cold or hot. One particular challenge for a cold startup is the
lack of initial system resistance. From the pump curves, it is clear that the pump’s
power and pressure are related to pump resistance. If the discharge pressure (thus
pump head) is low, the operating point may be over the end of curve; then the pump
flow would be very high, and so is the power required. One extreme example is deep
water disposal (DWD), where the pump discharge connects via a long pipeline to the
deep disposal wells. The pipeline, which can be several kilometers long, is initially
empty and has almost zero resistance. During a cold start, the discharge pressure
remains very low until the pipeline is filled with liquid.
With a fixed-speed pump, once the pump starts, the operating point rapidly moves
to the far right on the performance curve due to the low system resistance (low head).
The large flow rate quickly overwhelms the motor and trips the motor by power or
vibration.
Pump startup is a complex sequence. The trajectory of the operating point during
startup is a crucial consideration for process and process control design. Adequate
resistance should be provided by design. The pump response during startup is briefly
discussed as follows:
1. Pump starts with the downstream block valve closed and the recycle valve fully
open.
2. Once the pump is started, the fluid quickly fills the recycle line and enters full
recycle mode. Depending on the size of the recycle piping and valve, the operating
point will typically be on the far right side of the performance curve.
3. As more process flow enters the suction side, the discharge pressure (thus the
head) starts to build up.
4. For a warm startup or a pump joining an established header, the block valve would
be opened as soon as the discharge pressure is higher than the downstream line
pressure, and thus the transition is bumpless.
5. The downstream pipeline may be a long empty pipeline with virtually zero pres-
sure. During a cold startup, if the block valve (an on/off valve!) is suddenly opened,
and the pump discharge is faced with virtually zero resistance, the operating point
would abruptly move beyond the end of curve and trips the pump.
It is not uncommon to see careless process engineering and process control designs
that have overlooked the system resistance requirement and find themselves in a
situation that the pump simply cannot be started.
11.2 Centrifugal Pump 407

Fig. 11.15 Pump startup

with a throttling valve

Fig. 11.16 Pump startup

with double block valves

The process solution, and with proper accompanying control design and startup
sequence, is to create the required system resistance artificially during cold startup.
Two popular designs include the following:
1. Discharge throttling valve: The discharge control valve for capacity control can be
used to create the initial system resistance during startup. Once the pump reaches
full recycle mode, the valve can be manually and slowly opened to build up the
back pressure in the downstream pipeline. See the piping configuration at the top
in Fig. 11.15.
2. For a variable-speed pumping process, there may not be a discharge valve
installed. Even if available, the control valve may not be rated for the type of
high pressure drop across the valve in high-pressure pumping applications. For
this scenario, a working example is to utilize two block valves in parallel, one large
and one small. At the initial phase of the startup, the large valve is closed, and the
small valve is open. The small valve provides enough resistance to keep the pump
running and prime the pipeline. Once the back pressure in the pipeline reaches a
specific value, the large valve is opened. See the configuration in Fig. 11.16. The
startup sequence is responsible for the opening and closing of the valves.

11.2.5 Practical Considerations

Many practical considerations are dictated by process design and operational needs.
However, these process decisions often have an enormous impact on the process
control strategy.
408 11 Control Typicals: Equipment Control

Fixed-Speed Versus Variable-Speed Pumping

One major decision in process design is between using a variable-speed or a fixed-

speed pump. A fixed-speed pump with a throttling discharge valve is lower in capital
investment and also requires lower maintenance. However, there are also significant
drawbacks as follows:
1. Throttling the valve reduces the discharge flow and increases the head. The
increased head (pressure) is the energy wasted/lost across the throttling valve.
2. Throttling the valve moves the operating point along the performance curve and
away from the best efficiency point (BEP).
3. The startup of a fixed-speed pump is more challenging. Typically, a much larger
motor is needed to address the initial power spike.
Variable-speed operating is typically provided by either a motor with variable
frequency drive (VFD) or a variable-speed gearbox (VSGB) between the pump and
a fixed-speed motor. Although the initial capital investment (CAPEX) is higher,
variable-speed drives are more energy-efficient, and thus the operating cost (OPEX)
is lower.
The startup of a variable-speed pump is also much more manageable. The pump
speed can be quickly ramped up to the minimum stable speed and then slowly be
brought to the normal operating point. This startup procedure allows the operating
point not to go too far to the right on the performance curve, so less initial motor
power is required to allow the use of a smaller motor.

Suction Throttling Versus Discharge Throttling

When pumping incompressible (liquid) flow, discharge throttling is always preferred.

The main reason is that suction throttling could potentially cause low pressure and
increase the risk of cavitation.
The formation of bubbles causes cavitation as the liquid flashes under low pressure
(below the vapor pressure) and can damage a pump. Therefore, it is imperative to
operate the pump only in its operating region where the net positive suction head
required (NPSHR) is satisfied. In addition to proper design, low suction pressure
protective control and safeguarding can be implemented to avoid low suction pressure
if needed.

11.3 Centrifugal Compressor

A gas compressor is a vital equipment in many operating plants. The gas compression
system has the following unique properties (Bloch 2006; Hanlon 2001):
1. Large capital investment: A gas compressor can easily cost millions of dollars to
purchase and install.
11.3 Centrifugal Compressor 409

2. Critical to operation: The gas compression system is typically an indispensable

part of the process operation. Tripping or shutting down the compressor often
results in a significant interruption to production, which may directly lead to
production loss or deferment.
3. High operating cost: Gas compression also constitutes a considerable part of the
operating cost. It is common to see compressors rated at over ten megawatts of
power consumption in the oil and gas industry.3
4. Maintenance intensive: A compressor is a piece of maintenance-intensive equip-
ment and requires routine and costly maintenance work.
For these reasons, a mandatory requirement is that a reliable compressor control
system be implemented to protect the capital investment and maintain efficient oper-
ation.
A centrifugal compressor is similar to a centrifugal pump in many aspects. Many
of the discussions on pumps in Sect. 11.2 also apply to compressors. However, a
significant difference between a pump and a compressor is that the former works on
incompressible liquid and the latter on compressible gas. Most of the complexities
of compressor control originate from the compressibility of the gas.
Chapter 6 has used a compression process for process analysis. This section
focuses more on the compressor itself with the overall process flow in mind, as
shown in Fig. 6.1.

11.3.1 Process Description and Analysis

A compressor is a piece of rotating equipment that compresses the gas from a lower
pressure to a higher pressure and moves it from one location to another for further
processing, storage, or transportation.
A compressor is part of the overall gas compression process. Process analysis
should be carried out on the entire process flow rather than the compressor alone.
Since process configurations vary from operation to operation, it is impossible to
analyze all possible variations. Instead, the generic analysis methodology provided
in Chap. 6 can be followed to perform the analysis. Here in this section, we only
focus on the process flow scheme around the compressor.

Operating Requirements

The operating objectives of a gas compressor include the following:

1. During normal operation, maintain the desired pressure and flow (operating point)
at their target values. The compressor is part of the process flow and serves the

3 A 10 MW compressor consumes more than 80 million kWh of power (10.0 MW ×

1000 kw/MW × 24 h/day × 365 day/yr = 87,600,000 kWh), or over 8 million dollars per year
(assuming $0.1/kwh for electricity).
410 11 Control Typicals: Equipment Control

overall process operation. As a result, the operating requirements from the process
side determine the operating point.
2. During abnormal operating conditions such as process upsets or severe turndown
operation, keep the operating point within the safe operating region (operating
envelope).
3. During the transition between different operating conditions, including startup
and shutdown, ensure that the compressor quickly and smoothly reaches the new
steady-state.
A compressor must operate within many constraints and limits imposed by both
the compressor and the process to ensure safe and efficient operation. The typical
constraints that are inherent to the compressor include the following:
1. Surge limit: When the compressor flow falls below a specific low limit, an unstable
operating condition may occur where the flow rapidly reverses direction. This flow
reversal may cause violent vibrations that can potentially damage the compressor
and associated piping and fittings. This phenomenon is called compressor surge
and must be prevented.
2. Stonewall: On the other hand, when the compressor flow exceeds a specific high
limit, the compressor may reach a condition called stonewall or choke condition,
and the flow can no longer increase. The impact of a choke on the compressor
operation is not as severe as that of a surge. However, the compressor efficiency
drops significantly at the stonewall, and thus compressor operation should stay
away from the stonewall if possible.
3. Maximum and minimum speed: Another two inherent constraints with the com-
pressor are the maximum and minimum rotating speed. Beyond the speed limits,
the compressor operation may become less stable and thus should be prevented.
The process side can impose additional operating constraints. They vary from
operation to operation subject to the process design. Common operating constraints
from the process side include the following:
1. High limits on discharge pressure and temperature: Compression increases the
gas pressure, and at the same time, also increases the gas temperature. As the
process flow components such as compressor bearing, sealing, process piping,
and vessels all have pressure and temperature ratings, keeping the temperature and
pressure within the safe operating limits is imperative. For example, for centrifugal
compressors in process service, the maximum actual discharge temperature shall
generally not exceed 180 ◦ C.
2. Motor power: For motor-driven compressors, the motor is another expensive piece
of equipment to protect. A motor has a rated power limit that the compressor
operation should not exceed. There is usually a spike in the motor load during
startup, which can be several times higher than regular demand.
3. Gearbox torque: A variable-speed gearbox (VSGB) has a limit on the torque
rating. Compressor operation should monitor and limit the torque it generates on
the gearbox.
11.3 Centrifugal Compressor 411

Fig. 11.17 A single-stage gas compression unit

The limits and constraints mentioned above define the operating envelope. They
reduce the feasible operating region, and some may even change over time.

Process Flow Analysis

A typical gas compression system consists of the gas source, a suction scrubber,
the compressor, a discharge cooler, a discharge knockout drum, the recycle line and
valve, and the gas destination. Figure 11.17 shows the essential components in a
simple single-stage variable-speed gas compression circuit.
Three types of flows are critical to our analysis for the compressor: the process
flow, compressor flow, and the recycle flow (see Fig. 11.17). The process flow, also
called the forward flow, is the process gas to be moved forward to downstream
operation. The recycle flow is typically not flowing and is needed only for surge
prevention (see later). The three flows are related to each other by

[compressor flow] = [process flow] + [recycle flow]. (11.22)

Compressor flow is usually measured at the suction side using an orifice plate or
a Venturi tube but can also be measured at the discharge side.
A compressor can operate at a fixed speed or variable speed. A compressor
driven by a fixed-speed motor is a fixed-speed compressor. A compressor driven
by a variable-speed motor, or a fixed-speed motor with a variable-speed gearbox,
becomes a variable-speed compressor.

Supply and Demand Analysis

The material balance in the compression process follows the general principles below:
412 11 Control Typicals: Equipment Control

1. Gas is compressible. The mass flow through the compressor is assumed to be the
same4 from suction to discharge. However, the volumetric flow rate depends on
operation conditions and can be very different due to compression. Assuming no
gas loss during compression, we have

Fm,d = Fm,s (11.23)

ρd Pd Md Ps Ms
Fv,d = Fv,s , ρd = , ρs = (11.24)
ρs Z d R Td Z s R Ts
where Fm is the mass flow and Fv is the volume flow. ρ is the gas density. The
subscripts “s ” and “d ” indicate the suction side or the discharge side, respectively.
P is the operating pressure, and T is the operating temperature. M is molecular
weight, and R is the gas constant.
2. The imbalance in supply and demand causes the gas accumulation to change,
reflected in the pressure. The pressure–flow profile5 is thus the focus for supply
and demand analysis.
The compression process’s energy balance is about converting the motor power
to compressor impellers’ mechanical energy and then imparting it to the process gas
as potential energy (head or pressure) and kinetic energy (gas flow). Some energy
is lost as heat, causing temperature increases in the gas and the casing/bearing. The
temperature ratio and pressure ratio are linked via
  n−1
Td Pd n
= (11.25)
Ts Ps

where n is the polytropic exponent that will be discussed later.

The compressor is part of the supply and demand propagation. The capacity of
the compressor can be set by either the supply or the demand. The compressor has to
take whatever gas flow comes in, compress it, and send it further downstream with
the supply-driven operation. The pressure–flow profile (Sect. 6.3) would require that
suction pressure be the controlled variable (with discharge pressure as an override).
The supply/demand propagation goes from upstream to downstream. On the other
hand, if the downstream user determines how much gas the compressor should supply,
this is a demand-driven operation. The controlled variable would be the discharge
pressure, and the supply and demand relationship would propagate backward from
downstream to upstream.
The supply and demand model may become very complicated for a complex com-
pressor system involving multiple compressor trains, gas suppliers, and consumers.
The supply and demand analysis can follow the guidelines in Sect. 6.3. A real-world
example is provided in Chap. 6.

4 There can be liquid knockout or side streams that cause the mass flows to differ at suction and
discharge.
5 In liquid operation, the material balance is reflected by the level-flow profile, while for gas pro-

cessing, the material balance is reflected by the pressure–flow profile.

11.3 Centrifugal Compressor 413

Fig. 11.18 The three variables that define compressor dynamics

Fig. 11.19 Compressor performance curves (Left: fixed speed; Right: variable speed)

Cause and Effect Analysis

Similar to pump operation, the cause and effect relationship of compressor operation
can also be simplified and summarized as three variables, two lines, and one point:
1. Three variables: The gas flow through the compressor, the head (or pressure
increase) that the compressor generates, and the compressor speed are the three
variables that fully define the compressor operation. See Fig. 11.18 for an illus-
tration.
2. Two curves: The relationship between the three variables is defined by the so-
called performance curve(s). A fixed-speed compressor has one performance
curve, while a variable- speed compressor can have a series of curves within
the operating range. See Fig. 11.19.
At a constant speed, the compressor head is inversely proportional to the com-
pressor flow. The more the gas is compressed, the less the head it will generate,
and vice versa. The exact relationship is complicated and is determined by the
414 11 Control Typicals: Equipment Control

Fig. 11.20 Compressor resistance curve(s)

mechanical construction of the compressor. Once the compressor is built, the per-
formance curves stay unchanged (barring minor changes from wear and tear).
A higher speed will generate more flow and head for a variable-speed compres-
sor, while a lower speed will produce less flow and a lower head. This head/flow
versus speed relationship is called the system resistance curve(s). The resistance
curve approximately follows the affinity law (fan law) and with the head being
roughly a quadratic function of the gas flow (see Fig. 11.20).
The diagram showing both the performance curves and the resistance curves is
also called the compressor map.
3. One point: The operating point of the compressor must fall on both the perfor-
mance curve and the system resistance curve. The operating point is always at the
intersection of the performance curve and the resistance curve. The compressor
map in Fig. 11.21 shows the possible locations of the operating points.
Unlike the performance curve, which is fixed after the compressor is constructed,
the system resistance can be changed dynamically by many factors such as valve
position, recycle, supply pressure, and discharge back pressure, and thus the sys-
tem resistance curve can move right or left on the compressor map as the system
resistance changes.
With a fixed-speed compressor (Fig. 11.21 left), the operating point always moves
along the performance curve through system resistance changes. For a variable-speed
compressor, the operating point can be moved continuously to any location within a
region by changing the system resistance and compressor speed.
Example 11.2 A two-speed hairdryer. Figure 11.22 shows a two-speed hair dryer
with a concentrator to increase the air pressure.
A hairdryer is essentially a low-pressure air compressor. With the choice of two
operating speeds, and with or without the concentrator, there are four operating
points, as shown in Table 11.7.
11.3 Centrifugal Compressor 415

Fig. 11.21 Compressor operating points

Fig. 11.22 A two-speed hair

dryer with an add-on
concentrator

The performance curves of the hair dryer can be illustrated as in Fig. 11.23.
Based on the cause and effect relationship, at the low speed and without the
concentrator, the operating point is at A. Switching to the higher speed will move the
operating point up along the system resistance to point B. Adding the concentrator
increases the system resistance but does not change the motor speed, so the operating
point moves along the performance from A to C or B to D. The concentrator functions
as a discharge valve.
There are design limits and operating constraints that restrict the operating point
to a specific region. To the left on the compressor map is the surge line. When the
flow goes too low, the compressor can enter the surge cycle and lose stable operation.

Table 11.7 Operating points Concentrator

of a two-speed hair dryer
OFF ON
Low speed Point A Point C
High speed Point B Point D
416 11 Control Typicals: Equipment Control

If the flow rate is too high, the compressor may hit the stonewall and lose operating
efficiency. The limits and constraints define the operating envelope.
The surge limit and stonewall are illustrated by the surge line and the stonewall
line on the compressor map in Fig. 11.24.
Typical control handles for compressor operation include one or more of the fol-
lowing: the compressor speed, inlet guide vane, control valves at suction or discharge,
and recycle valves. Other control handles may exist to enforce the supply-driven or
demand-driven material flow upstream and downstream of the compressor.
The compressor head and flow are the primary control targets. Additional control
targets may include protective targets such as surge and stonewall, power, and gearbox
torque, as discussed above.
Surge prevention is a mandatory requirement for centrifugal compressor opera-
tion. Surge happens when the compressor flow (not the process flow!) runs below a

Fig. 11.23 Performance curves of a two-speed hair dryer

Fig. 11.24 Compressor operating envelope

11.3 Centrifugal Compressor 417

low limit. The challenge, however, is that this minimum flow limit is not constant. It
is different at different compressor speeds and varies with gas property (molecular
weight and compressibility) and operating conditions (pressure and temperature).
In practice, an inferential property called surge indicator is usually defined as an
online variable to indicate how far the operating point is from the surge limit (see
Sect. 5.3 for inferential property estimation). Surge indicator is inferred from fast and
reliable measurements such as flow, pressure, and optionally temperature. A general
formula is as follows:
K s = f (Ps , Pd , Fv , Ts , Td ) (11.26)
where Ps is the suction pressure, and Pd the discharge pressure. Fv is the raw vol-
umetric flow (Sect. 10.3) at either the suction or discharge side. Ts and Td are the
temperatures at the suction and discharge sides, respectively. There are many ways
to define the surge indicator, and there are also different names for surge indicator.
A more detailed discussion will be provided later in the control design section.
The surge indicator is the most critical protective control variable for a centrifugal
compressor. Other protective targets such as motor power or gearbox torque can be
controlled variables as well.
The potential manipulated and controlled variables, and the cause and effect rela-
tionship between them, are summarized in Table 11.8.
Not all the control handles are available for a gas compression system. The speed,
suction valve, and discharge valve are mutually exclusive. For a fixed-speed com-
pressor, the typical configuration is with the suction valve and the recycle valve. For
a variable-speed compressor, the typical manipulated variables are the compressor
speed and the recycle valve.
The compressor speed or suction valve as a manipulated variable is primarily
for maintaining the compressor capacity. The recycle valve, also called the anti-
surge control valve (ASCV), is mainly for surge prevention (and choke prevention
if needed).

Degree-of-Freedom Analysis

Table 11.8 shows that there are more targets than handles for compressor control.
Per the degree-of-freedom requirement, the controlled variables and manipulated
variables must be fully paired up and exhaust all the degrees of freedom (DOF)
available (Sect. 6.5).

Table 11.8 Cause and effect relationship for compressor operation

Head Compr flow Proc flow Surge margin Power Dschg press
Speed ↑    –  
Suction valve ↑      
Discharge valve ↑      
Recycle valve ↑   –  – 
418 11 Control Typicals: Equipment Control

The available handle is typically the suction valve, discharge valve, or compressor
speed for normal regulatory control. Usually, only one of these control handles is
installed. It is the manipulated variables responsible for maintaining the compressor
head and flow (the capacity).
The compressor performance curves (Fig. 11.19) show that the pressure and flow
are completely correlated at a given speed, so the compressor head and flow are
treated as one controlled variable and consume one DOF. Typically, the polytropic
head is selected as a controlled variable to indirectly control the compressor flow
(capacity).
The practical challenge is that the polytropic head is not directly measurable.
However, since the polytropic head is a function of the pressure ratio between dis-
charge and suction pressure, the pressure is typically chosen as the controlled vari-
able instead. Either the suction pressure or the discharge pressure will serve as the
capacity control variable depending on whether the operation is supply-driven or
demand-driven.
The recycle flow is provided primarily for surge protection. Therefore, choosing
the recycle valve as the manipulated variable and the surge indicator as the controlled
variable is an easy decision.
Now we still have two remaining challenges with the process variable selection
and pairing:
1. There are still more control targets to be paired up with control handles. The
targets include the compressor power limit, discharge pressure, gearbox torque,
and maybe more.
2. There are strong interactions between the manipulated variables and controlled
variables and remaining control targets.
Based on the guidelines in Sect. 2.2 and Chap. 7, complex control loops such
as split-range control and overriding control can be considered to meet the DOF
requirements. The pairing of the process variables will be further explored in the
next section’s control strategy discussion.

Dynamic Response Analysis

Typically, step tests are recommended to learn more about the dynamic characteris-
tics of each causal relationship in the cause and effect table (Table 11.8). Centrifugal
compressor has been a piece of well-studied equipment, and some general under-
standings of the compressor dynamic characteristics include the following:
1. Compressor dynamics are extremely fast. Among the cause and effect relation-
ships in Table 11.8, the dynamic response between the recycle valve and the surge
indicator is typically in sub-seconds. The dynamics between the compressor speed
and capacity are orders of magnitude slower.
2. The dynamic response characteristics between all the control handles and control
targets are assumed to be linear, time-invariant, and self-stabilizing. For example,
11.3 Centrifugal Compressor 419

by increasing the compressor speed, the operating point will move to and settle at
a new location by itself provided that the operating points are within the operating
envelope.
3. For a multi-stage compressor, opening the recycle valve of one stage can inad-
vertently interfere with the operation of other stages.
A compressor is a self-regulating process. That means if the system resistance
changes, the operating point automatically moves to a new location and settles there
by itself. This behavior is reflected by the negative slope between the head and the
flow in the performance curve and the positive slope of the system resistance curves
(see Fig. 11.19). As long as the operating point stays inside the operating envelope,
the head and the flow will automatically find a new balance point, and thus the
compressor operation is relatively stable.
The surge phenomenon occurs when the operating point moves too far to the left
on the performance curve. To the left of the surge point (i.e., below the minimum
flow limit), the head–flow slope turns positive and thus becomes self-accelerating
and unstable. This phenomenon is like pushing a cart up a hill (see Fig. 11.25). The
energy from the motor is like the force pushing the cart, while the system resistance
is like the gravitational force pulling the cart down. Once the cart is over the top, the
two forces work in the same direction, and the cart quickly falls over the top.
Therefore, the surge point’s location (the top of the hill) at each compressor speed
is the most critical information for compressor control.
The compressor is typically designed and built based on off-line simulation and
calculation data, per the client’s specifications. The predicted (estimated) perfor-
mance data is provided in advance for front-end engineering of the overall process
design (including process control). After the compressor is constructed, surge tests
are usually conducted to confirm that the actual dynamic responses (e.g., perfor-
mance curves) match the predicted performance. The test is typically performed

Fig. 11.25 Surge and reverse flow

420 11 Control Typicals: Equipment Control

with common gases such as nitrogen (N2 ) or fuel gas because the actual process gas
is not available yet.
After the compressor is installed at the site and before startup, a formal surge
test is usually required to confirm the compressor dynamics response with the actual
process gas.

11.3.2 Control Strategy

The process analysis paves the way for control design. Chapter 7 provides some
guidelines for process control design at both plant level and individual loop level.

Control Objectives

Control objectives serve operating objectives. The generic objectives of compressor

control are maintaining the operating point at the desired value during normal oper-
ation and keeping the operating point inside the operating envelope during abnormal
process conditions. Protecting the compressor is vital; therefore, integrated control
and safeguarding is also a common requirement.
The compressor dynamic response, particularly the surge phenomenon, is
extremely fast. Compressor surge causes the reversal of compressor flow, which
happens in tens of milliseconds to hundreds of milliseconds. The control solution
must execute at a sub-second frequency, typically as fast as the DCS system can sup-
port. In olden days, dedicated hardware and software were utilized to support fast
execution. As technology advances, many DCS systems can now execute at 100ms
scan time or faster. Consequently, many compressor control applications are now
implemented in DCS using the standard function blocks. Compressor control is no
more mysterious than a complex PID control loop in Chap. 4.
The fast execution requirement has ruled out model predictive control (MPC) and
other sophisticated control schemes requiring intensive online calculation. Complex
PID control loops implemented in DCS are the only feasible solution based on
standard hardware and software.

Control Scope

The overall control strategy follows the layered control structure described in Chap. 7,
with continuous regulatory control for normal operation and overriding control as
protection during operating upsets. There are strong interactions between control and
safeguarding, and therefore integrated control also needs to incorporate safeguarding
in the design.
The startup and shutdown of a gas compression system is an operational chal-
lenge, and automated startup and shutdown procedures greatly help operations. The
11.3 Centrifugal Compressor 421

Fig. 11.26 Operating requirements of gas compression process

automated startup sequence is part of sequential control and is typically included in

the integrated compressor control design.
Figure 11.26 provides a list of the components in a fully integrated compressor
solution.
It is beyond this book’s scope to provide an in-depth discussion of all the compo-
nents in the control strategy. A discussion on the essential components is provided
as follows:

1. Normal regulatory control: Normal regulatory control is typically about quality

and capacity. There is typically no requirement on product quality for gas com-
pression; thus, the normal regulatory control is mainly about capacity control.
For capacity control, a clear understanding of the supply/demand model is
required for placing the appropriate pressure/flow controllers at suitable locations
to ensure proper supply/demand propagation. For example, the control configu-
ration for a supply-driven operation of a variable-speed compressor is shown in
Fig. 11.27. The blue dashed line indicates the direction of supply/demand propa-
gation. The two relay points for the propagation are the suction pressure and the
discharge pressure.
The downstream consumer of the gas is assumed to have sufficient capacity to
absorb the fluctuation in supply. Otherwise, another swing stream needs to be
provided as swing control.
If the operation is demand-driven, the control configuration would be as shown
in Fig. 11.28. In this case, the supply side needs to adjust the supplied gas rate
in response to the demand change. Otherwise, a separate swing stream must be
provided.

Suppose the supply-driven operation may become demand-driven, either tem-

porarily or permanently. In this case, override control must be provided in the
422 11 Control Typicals: Equipment Control

Fig. 11.27 Supply-driven control of a variable-speed compressor

Fig. 11.28 Demand-driven control of a variable-speed compressor

design. Figure 11.29 shows an example of supply-driven control with demand

override. The dashed line indicates the propagation during regular supply-driven
operation, while the dotted line illustrates propagation when the supply-driven
operation reverses to demand-driven. In terms of control configuration, Fig. 11.29
is a combination of Figs. 11.27 and 11.28. The automatic reversal of the sup-
ply/demand model is via the control selector function block and the proper con-
figuration of the normal regulatory control setpoints and the protective control
setpoints.
2. Overriding protective control: The most crucial protective control for a compres-
sor is the anti-surge control. The focus of this section is anti-surge control only,
as other protective controls listed in Fig. 11.26 are relatively straightforward.
Compressor surge is a dangerous condition to compressor operation and must
be prevented. Low process flow (see Fig. 11.17) causes low compressor flow and
subsequently leads to compressor surge. The most effective approach to prevent
the compressor surge is to open the recycle valve and recirculate part of the com-
pressed gas back to the suction side. As the compressor flow is the sum of the
11.3 Centrifugal Compressor 423

Fig. 11.29 Supply-driven control of a variable-speed compressor, with demand override

process flow and recycle flow, the compressor does not know, neither does it care,
whether the gas is from the process or recycle.
Reducing system resistance can increase the surge margin as well. For example,
opening the suction or discharge valve (if available) moves the operating point to
the right side of the performance curve.
3. Instrumented safeguarding: If both capacity and protective controls fail, the safety
instrument system (SIS) shuts down the compressor operation. The safeguarding
design should be included in the overall control design because the anti-surge
tripping must be consistent with the anti-surge control. The integrated control
design has the following requirements on the safeguarding function:
a. The same calculation for the surge indicator must be used.
b. The pressure and flow measurement must be from the same locations. They
are from independent sensors and transmitters than those used for anti-surge
control to increase reliability by avoiding common source failures.
Emergency shutdown of the compressor has its own risk, so the trip logic needs
to be carefully calibrated to safeguard the compressor effectively, and at the same
time, minimize nuisance shutdowns. For example, an isolated surge of short dura-
tion (e.g., less than 3 s) is not a serious concern, but a frequent or sustained surge
is very hazardous.
4. Sequential Control: Compressor startup is a complicated task. There are many
checks and balances in an automated startup sequence. Multiple steps in the startup
sequence require coordination with the process control scheme. For example, the
startup sequence needs to quickly bring the operating point to a stable location
inside the operating envelope. The anti-surge control will then be activated. After
the compressor reaches normal operating conditions, the capacity controller can
be put into automatic operation.
424 11 Control Typicals: Equipment Control

11.3.3 Functional Design

With the process analysis in Sect. 11.3.1 and the overall control strategy in Sect. 11.3.2,
we can move on with the functional design. We take a variable-speed compressor
as the typical setup in discussing control design. For fixed-speed compressor, the
control design is similar but more straightforward.

Capacity Control

The capacity control around the compressor is provided in Figs. 11.27, 11.28,
and 11.29. However, the actual gas compression process is much more compli-
cated than a simple single-train operation. Example 6.1 is a compression process of
medium complexity. A more complex process may have multiple compressor trains,
and each may have multiple stages, with multiple streams in and multiple streams
out. See Fig. 12.29 for an example.
The overall capacity control design can follow the generic supply and demand
analysis methodology in Sect. 6.3. The capacity control propagation may span across
multiple operating units or even beyond the project boundary. As mentioned in
Sect. 7.1.2, capacity control is a consideration at the plant operation level. A holistic
view of the supply and demand model must be kept in mind, while a process overview
diagram (POD) is a handy tool to keep this big picture at sight.

Surge Indicator Calculation

A healthy surge margin needs to be maintained in compressor operation to prevent

the compressor from going into surge. Fast and reliable calculation of the surge
indicator is thus a requirement. There have been many efforts and solutions for
the surge indicator calculation (White 1972; Staroselsky and Ladin 1979). A good
summary is provided in Mirsky et al. (2012).
In a nutshell, the surge point is the extreme operating point beyond which the
compressor would enter a surge cycle. When the compressor flow falls below a
minimum flow limit, the flow can reverse direction in less than 200 ms seconds and
goes back and forth. It is like running up a hill. Right over the top is a cliff. Once
at the top, there is no time to stop from going over the edge. So it is crucial to stay
away from the top, but be able to get near the top.
The line connecting all the surge points is called the surge line, which may have a
highly irregular shape due to noises and irregularities in data (Fig. 11.24). For online
calculation, a simple mathematical formula is desired to approximate the surge. This
simple formula represents a smooth line going through some or most of the surge
points. This line is called the surge reference line or surge limit line (see Fig. 11.30).
The selection and definition of the surge reference line is the primary difference
between the different compressor control vendors’ technologies.
11.3 Centrifugal Compressor 425

Fig. 11.30 Surge reference line

Compressor affinity law states that the head is a quadratic function of the flow:
 2  2
H p,1 Fv,1 Hp Fv
≈ , or ≈ (11.27)
H p,2 Fv,2 H p,0 Fv,0

where H p and H p,0 are the polytropic head of two arbitrary points along a system
resistance line; Fv and Fv,0 are the corresponding volumetric flows. In the head-
versus-flow coordinate, the fan law formula represents a parabolic line. That is, the
system resistance line is parabolic in theory.
If H p,0 and Fv,0 are assumed to be a known point on the surge reference line, then
any point on the surge reference line can be represented by the following equation:

H p = K · Fv2 . (11.28)

Again, the surge reference line does not necessarily go through all the surge points
due to measurement noises and some non-conformity to theory. With a conservative
design, the surge reference line should be drawn in such a way that all the surge
points are either on or to the left of the line.
In practical application, the polytropic head is not directly measurable, nor is the
compressor flow. Therefore, they are not available for online calculation of the surge
indicator. Other reliable and online measurements must be used instead. From the
definition of polytropic head and volumetric flow (Eqs. 11.29 and 11.31),
  n−1   σ 
n Z s R Ts Pd n 1 Ps Pd
Hp = −1 = −1 (11.29)
n−1 M Ps σ ρs Ps
426 11 Control Typicals: Equipment Control
   
Pd Pd Ts PM n−1
with n = ln ln , ρ= , σ = (11.30)
Ps Ps Td Z R, T n

 
Ps Pd
Fv = C or Fv = C (11.31)
ρs ρd

the polytropic head can be expressed in pressure ratio Pd /Ps , while the compressor
flow is typically inferred from the pressure drop P across an orifice plate.
In the above equations, n is the polytropic exponent, Z is the gas compressibility,
R is the universal gas constant, ρ is the density of the gas, and P is the pressure
drop across the orifice flow meter. Subscript “s ” indicates the suction side while
subscript “d ” the discharge side.
With Eqs. 11.29 and 11.31, the compressor affinity laws can be reformulated as
⎡ ⎤
 n − 1
n Ps ⎢ Pd n − 1⎥ ⎛ ⎞2
⎣ ⎦ Ps Ps
n − 1 ρs Ps C C2
⎜ ⎟
ρs ⎟ ρs
≈⎜
⎝ Fv,0 ⎠ = Fv,0 . (11.32)
H p,0

Lumping all the constant values H p,0 , Fs,0 , and C together into a single constant K ,
and reformulating the equation, we have
 σ 
1 Pd Ps
−1 ≈ K · . (11.33)
σ Ps Ps

Most surge indicator calculation formulas are based on this relationship, all with a
certain degree of simplification or approximation.
Head versus flow is known as an invariant coordinate, meaning that the head–flow
relationship is determined by the compressor’s mechanical construction and does
not vary with gas property or operating condition. It is proven that the compression
ratio Pd /Ps versus P/Ps is an alternative coordinate that is almost invariant and
suitable for online applications since it is based on available online measurements
only (Batson 1996; Nägeli et al. 1973; White 1972).
Anti-surge technologies based on Eq. 11.33 require online temperature measure-
ments of Ts and Td , because polytropic exponent n is needed (see Eq. 11.30) (Kurz
et al. 2016; Mirsky et al. 2012). Since temperature is a slower and less reliable mea-
surement than pressure and flow, it is not ideal to be included in online calculations.
Equation 11.33 can be simplified to eliminate the need for temperature:
 σ 
1 Pd Pd
− 1 ≈ C1 + C2 . (11.34)
σ Ps Ps
11.3 Centrifugal Compressor 427

Some early anti-surge control utilizes two differential pressures, one across the
orifice plate (P), the other between the compressor discharge and suction (Pd −
Ps ). It is evident that if Eq. 11.34 is simplified further by assuming σ = 1, then
 σ 
1 Pd Pd
− 1 ≈ C1 + C2 ≈ Pd − Ps . (11.35)
σ Ps Ps

The typical approach is to define a more intuitive variable to indicate the operating
point’s location relative to the surge reference line for online anti-surge control. When
the operating point is on the reference line, the value is set to a fixed value called
surge reference value. When the operating point is at another location on the map, its
value is compared with the surge reference value. The relative value indicates how
far the point is from the surge limit.
The parabolic surge reference line becomes a straight line if redrawn in the coor-
dinate of H p versus Fv2 . The surge parameter can be defined as the ratio between the
slope at the operating point and the slope of the surge reference line:
⎧
⎪= 1.0, at surge point
Slope of Surge Point ⎨
S= > 1.0, at right of surge point
Slope of Operating Point ⎪ ⎩
< 1.0, at left of surge point
 σ !
1
σ
Pd,0 /Ps,0 − 1

H p,0 H p P0 /Ps,0 P/Ps
= 2 = 1 σ
! = C  σ  (11.36)
Fv,0 Fv2 (P d /Ps ) − 1 1 Pd
σ −1
P/Ps σ Ps

where S is the surge indicator.6 The surge reference line has a surge value of S ≡ 1.
A value of S > 1 indicates that the operating point is on the right-hand side of the
surge reference line, while S < 1 is on the left-hand side (Kurz et al. 2016; Mirsky
et al. 2012).
The surge value can also be based on the so-called equivalent flow, with a definition
derived from Eqs. 11.33 and 11.34:
"
# P/Ps
# FTE FT
S = #  σ ≈ ≈√ (11.37)
$1 Pd Pd C1 Ps + C2 Pd
−1 C1 + C2
σ Ps Ps

P
FT = Fm,range · (11.38)
P range

P Ps,range
FTE = Fm,range · · (11.39)
P range Ps

6 When the polytropic exponent n approaches ∞, value of σ would approach 1. Equation 11.36 is
reduced to S ≈ C PdP
−Ps . This is the same as Eq. 11.18 for pump.
428 11 Control Typicals: Equipment Control

where FT is the raw actual flow (uncompensated actual flow as discussed in

Sect. 10.3.4), and
√ FTE is the equivalent flow. It can be proven that FTE is linearly
proportional to P/Ps ; therefore, the compression ratio Pd /Ps versus equivalent
flow is another choice of invariant coordinates. If the surge indicator S is set to 30%
for the surge reference line, then the stonewall line would have a value close to 100%.
The operating region lies between approximately 30% and 100%.7
A 10% safety margin is typically provided for control. So a surge control line
is typically chosen to be about 10% higher than the minimum flow rate. The surge
control line is the setpoint for the anti-surge controller. Above this setpoint, the
anti-surge controller drives the recycle valve to the fully closed position. Below
this setpoint, the anti-surge aggressively open the recycle valve to prevent surge.
Equation 11.36 is modified to become
⎧
⎪
⎨= 0.0, on control line
DEV = S − 1 − B1 > 0.0, above control setpoint (11.40)
⎪
⎩
< 0.0, below control setpoint

where B1 is the 10% safety margin for control. The objective of the anti-surge
controller is to maintain the DEV value above zero, corresponding to an operating
point to the right of the surge control line.

11.3.4 Implementation and Operation

The actual compressor control implementation can have many variations. This
section only provides a few typical configurations to give a flavor of the complexity.

Capacity Control

Capacity control is closely related to the supply and demand model and varies from
design to design. Figure 7.5 in Sect. 7.2.2 has already provided an excellent example
of capacity control design. Figures 11.27, 11.28, and 11.29 provide several typical
implementations. Therefore, no further discussions will be provided here.

7 A compressor behaves similar to an orifice in terms of restriction to flow. An orifice plate has a
turndown ratio of about 10:1 in DP. Below 10% of the DP measurement range, the flow may not
be stable enough to produce credible √
readings. Since the flow is a square-root function of DP, this
10:1 DP turndown corresponds to a 10 : 1 ≈ 3 : 1 turndown for flow. In other words, for flow
measurement with an orifice plate, the effective range of reading is approximately between 30%
and 100%.
11.3 Centrifugal Compressor 429

Fig. 11.31 Basic anti-surge control

Anti-surge Control

The anti-surge control is a protective PID control that monitors the surge indicator.
Whenever the surge indicator is below its setpoint, the anti-surge controller quickly
opens the recycle valve to bring the compressor above the minimum flow limit.
Figure 11.31 shows a basic anti-surge control configuration. The surge indicator is
calculated with Eq. 11.26 from the suction pressure, the discharge pressure, and the
compressor flow. The surge indicator is the process value (PV) for the anti-surge
controller. The controller setpoint is the desired surge indicator, while the controller
output is sent down to the anti-surge recycle valve. The compressor flow can be
measured at either the discharge side (FT-111) or the suction side (FT-121).

Other Protective Controls

Other protective controls such as high discharge pressure protection, high motor
power protection, and high gearbox torque protection are all overriding protective
control. They share the existing final control elements such as the compressor speed
and recycle valve, override the capacity control or other protective controls as needed.
For example, high discharge pressure protective controller can override the anti-surge
controller through a high selector to open the anti-surge recycle valve if it demands
a larger opening than required by anti-surge control (Fig. 11.32).

Integrated Control

Figure 11.33 shows a typical compressor control implementation.

430 11 Control Typicals: Equipment Control

The compressor is supply-driven. Thus, the capacity control is provided by suc-

tion pressure controller PC-102, cascaded to a load controller XC-101, and then to
speed controller SC-101, which manipulates the gearbox speed. The pressure con-
troller PC-004 manipulates a downstream control element and thus propagates the
supply change further downstream. High capacity may cause the discharge header
pressure PC-003 to go above its high limit; the pressure control PC-003 serves as
an override controller to protect the discharge header.
There is one anti-surge control for each stage (UC-111 and UC-121). The surge
indicator is calculated by the CALC function block (UY-111 and UY-121) based
on suction pressure, discharge pressure, and compressor flow.
Other protective controllers are implemented as needed. These include the suction
header pressure protective controller PC-001 and the motor power protective control
EC-101.
For optimal performance, a split-range arrangement can be implemented so that
the capacity control uses the compressor speed as the primary control handle, sup-
plemented by a recycle valve in case a severe turndown operation is needed. Due to
the length limit, we will skip the detailed discussion on this split-range arrangement.

11.3.5 Practical Considerations

Integrated compressor control is one of the most complicated baselayer control

schemes. In addition to the intricate control design, there are also numerous practical
considerations to be addressed. Below are some examples:

Fig. 11.32 Anti-surge control with high discharge pressure override

11.3 Centrifugal Compressor 431

Fig. 11.33 A typical integrated compressor control scheme

1. The fast compressor dynamics require high-speed calculation and control action.
This requirement poses some unique challenges on all the components in the
control loop, especially on the control valves.
2. PID configurations are quite involved because of the extensive use of override
controls, fan-out controls, and cascade controls. Anti-reset windup, bumpless
transfer, and back initialization are essential considerations.
3. Startup and shutdown. Starting the compressor is a complicated procedure. The
compressor goes through multiple stages of the transition process, from zero flow
and zero pressure ratio, through full recycling, quickly increasing the discharge
pressure, pushing for the forward flow, and finally reaching the steady-state. The
startup sequence is responsible for bringing the compressor to the normal operat-
ing condition safely and quickly. Process control is part of the startup sequence.

11.4 Summary

Equipment control such as pump and compressor control is traditionally regarded as

the specialty of equipment manufacturers or their contractors, assuming that equip-
ment is too complex and critical to leave their control to somebody else. However,
one of the issues with the control solutions from equipment manufacturers or third
parties is that their primary objective is to protect their equipment rather than the
overall process control. They do not have the required process knowledge and con-
trol expertise to produce an optimal integrated control strategy aiming at overall
optimality in control.
Process control depends on a good understanding of the process dynamics, equip-
ment characteristics, and operating requirements. For this reason, process control
engineers with a holistic view of both the process and the equipment can typically
432 11 Control Typicals: Equipment Control

produce a process-centric control solution that is more effective than an equipment-

centric solution in protecting both the process and equipment.
Equipment such as a compressor may seem overwhelmingly complicated. How-
ever, as mentioned in Chap. 1, the information required by process control is very
different from that for an equipment engineer. Following a systematic approach, the
control-specific information is not as hard to extract and generalize as expected.
When implemented in standard DCS with standard function blocks, the pump con-
trol (Sect. 11.2) and compressor control (Sect. 11.3) are merely another application
of the complex PID control schemes, as discussed in Sect. 2.2 and Fig. 4.1.

References

Batson BW (1996) Invariant coordinate systems for compressor control. In: Proceedings of the inter-
national gas turbine and Aeroengine congress & exhibition. The American Society of Mechanical
Engineers, Birmingham, UK
Berkeley D (1958) Ejectors have a wide range of uses. Petroleum Refiner December
Bloch H (2006) A practical guide to compressor technology, 2nd edn. Wiley Interscience, New
York
Brannan C (2018) Rules of thumb for chemical engineers - a manual of quick, accurate solutions
to everyday process engineering problems, 6th edn. Elsevier, Amsterdam
Campbell J (2004) Gas conditioning and processing - the equipment modules, vol 2, 8th edn. John
M, Campbell and Company
DOE (2001) Pump life cycle costs: a guide to LCC analysis for pumping systems. Technical Report
DOE-GO-102001-1190, Hydraulic Institute, Euro-Pump, and the US Department of Energy’s
Office of Industrial Technologies (OIT)
Hanlon PC (2001) Compressor handbook, 2nd edn. McGraw-Hill, New York
Keenan JH, Neumann EP, Lustwerk F (1950) An investigation of ejector design by analysis and
experiment. ASME J Appl Mech 72(17):299–309
Kurz R, White RC, Brun K, Winkelmann B (2016) Surge control and dynamic behavior for cen-
trifugal gas compressors. In: Proceedings of Asia Turbomachinery and Pump Symposium, Tur-
bomachinery Laboratory, Singapore
Liao C (2008) Gas ejector modeling for design and analysis. PhD thesis, Texas A&M University
Mirsky S, Jacobson W, Tiscornia D, McWhirter J, Zaghloul M (2012) Development and design of
anti-surge and performance control systems for centrifugal comrpessors. In: Proceedings of the
Forty-Second Turbomachinery Symposium, Houston, Texas
Nägeli JP, Spechtenhauser A, Aicher W (1973) Turbomachinary in base load natural gas liquidifi-
cation plants. In: Proceedings of the 12th World Gas Conference, Nice, France
Perry R, Green D (1984) Perry’s chemical engineers’ handbook, 6th edn. McGraw-Hill, New York
Staroselsky N, Ladin L (1979) Improved surge control for centrifugal compressors. Chem Eng
175–184
White MH (1972) Surge control for centrifugal compressors. Chem Eng 54–62
Chapter 12
Control Typicals: Plant-Wide Control
and Unit Control

The number of processes in operation or under design is innumerable. Therefore, it is

not practical to provide standard control designs for every typical process. The more
practical approach is to learn the general thinking process and structured analytical
skills that can be applied to any process control problem. This chapter presents
several commonly encountered but not much-discussed processes to demonstrate the
process analysis and control design skills and methodology. The selected processes
are all plant-level or unit-level operations with manageable complexity, from tank
blanketing (Sect. 12.1), boiler drum level (Sect. 12.2), well test separator (Sect. 12.3),
to firewater control (Sect. 12.3). An acetylene hydrogenation process is presented to
show the work process for unit-level control design (Sect. 12.5), and an E&P surface
production plant is used to illustrate the overall control strategy of a plant-wide
control solution (Sect. 12.6). The discussions focus not on the specific solutions but
rather on the thinking processes and design procedure.

12.1 Liquid Tank Blanketing Control

Gas blanketing control of a liquid tank is a common and straightforward application.

However, this application is chosen as a typical control problem to show that even a
simple control problem can have different control designs, from good, bad, to ugly,
with long-lasting consequences.
The ugly designs may work to some extent and would not impact the operation as
severely as tripping the plant operation. They are just less efficient or less reliable. For
this very reason, plant operations may not take the poor designs as severe problems
and choose to live with the poor performance. The bad control designs are often
carried over to the next project through careless copy and paste. Due to the massive
number of tanks in operation and the many incidents associated with tank operation,
the proper control design is critical.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 433
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3_12
434 12 Control Typicals: Plant-Wide Control and Unit Control

12.1.1 Process Description and Analysis

A tank is typically used for separation or storage. For example, there are many liquid
tanks of different sizes and shapes in the oil and gas processing facilities, such as
refineries, petrochemical plants, and terminals. Some tanks are for the separation
or storage of oil rich in hydrocarbon, while others are for produced water that has
entrained hydrocarbons.

Operating Requirements

Tank operation is typically a low-pressure operation (near atmospheric pressure).

The critical operating requirement for tanks with hydrocarbon content is to prevent
oxygen or air from entering the tank. The hydrocarbon content in the liquid must be
adequately isolated from oxygen or air to avoid potential fire and explosion. Because
the tank operation is seemingly straightforward, the importance of proper control is
often underestimated. Due to negligence in design and operation, many incidents are
linked to the collapse or explosions of liquid tanks.
One common approach to protecting the tank is to blanket the liquid in the tank
with oxygen-free gas. Tank blanketing, also referred to as tank padding, applies gas
to the empty space in a storage container. The purpose is to keep oxygen levels in and
around the product at a low level to reduce oxidation. The oxygen-free blanketing
gas can be inert nitrogen gas or processed natural gas from the same facility. As the
blanketing gas has its cost, the tank is expected to be protected with minimal gas
consumption (Brannan 2018; Kinsley 2001).

Process Flow Analysis

Figure 12.1 shows a free water knockout tank (FWKO) for produced water from
production separators. There are entrained hydrocarbons in the produced water that
may flash out and fill the tank with flammable hydrocarbon vapor. No contact with
oxygen is allowed.

Supply and Demand Analysis

There are three gas streams for the tank operation: the inlet blanketing gas, the flashed
gas from the liquid, and the outlet venting gas to flare. They are related to each other
by the following material balance equation:

[venting gas] = [blanketing gas] + [flash gas]. (12.1)

12.1 Liquid Tank Blanketing Control 435

Fig. 12.1 A free water knockout drum with gas blanketing

The operation is tank-driven (see supply and demand analysis in Sect. 6.3). To the
tank, the blanketing gas stream is demand-driven, while the venting gas stream is
supply-driven. This tank-driven operation requires that both the blanketing gas and
venting be able to swing. Two independent control valves meet the blanketing oper-
ation requirements.
The inlet blanketing gas flow and outlet venting gas flow are two independent
control handles serving as two manipulated variables for control. The tank pressure
is the control target chosen as the controlled variable.

Cause and Effect Analysis

The available control handles are the inlet blanketing gas and outlet flaring gas.
Increasing the blanketing gas intake will increase the tank pressure. Increasing the
venting gas flow to flare will decrease the tank pressure. This relationship can be
summarized by the simple cause and effect table (CET) shown in Table 12.1.

Degree-of-Freedom Analysis

There are two control handles and one control target, implying that one extra degree
of freedom (DOF) must be consumed. As discussed in Sect. 6.5, there are two ways to

Table 12.1 Cause and effect Tank pressure

table for gas blanketing
control Blanketing gas flow ↑ 
Venting gas flow ↑ 
436 12 Control Typicals: Plant-Wide Control and Unit Control

exhaust the DOF: increasing the number of control targets or decreasing the number
of control handles. The options will be explored with the different control designs
later on in the control design section.

Dynamic Response Analysis

The dynamic behavior of the cause and effect relationship in Table 12.1 is all fast
and straightforward. Besides, the dynamic impact on the tank pressure from the
inlet blanketing gas and outgoing venting gas is comparable; therefore, no special
consideration is required on the process dynamics.

12.1.2 Control Strategy

The control objective for the liquid tank operation is to maintain the tank pressure
within the desired range of 0–1.0 kPa for three reasons:
1. Keep the tank above atmospheric pressure to prevent oxygen ingress.
2. Avoid vacuum in the tank to prevent buckling collapse of the tank.
3. Minimize the consumption of the blanketing gas.
This application has one control target and two control handles; thus, as suggested
in Table 2.7, a complex PID control scheme is sufficient. There is no need to explore
more advanced control schemes.

12.1.3 Functional Design

The process is simple, and the control objective is clear; however, the detailed func-
tional design can vary from good, bad, to ugly.

The “Ugly” Design

A naïve control design uses two independent pressure controllers to maintain the
tank pressure, as shown in Fig. 12.2. One controller manipulates the inlet blanketing
gas, while the other controls the flaring gas rate. Both controllers are assigned with
the same setpoint of 0.5 kPa(g).
This design will not work as desired! The pressure measurements for the two
pressure controllers are highly correlated, so they are essentially the same variable.
As a result, this design has one controlled (tank pressure PC-101A ≈ PC-101B)
and two manipulated variables (PCV-101A and PCV-101B).
12.1 Liquid Tank Blanketing Control 437

As explained in Sect. 8.3.4 on troubleshooting, there is no unique solution for

this under-determined control problem. The two controllers will continuously fight
each other until one is driven to saturation because when the valve is saturated, it is
no longer a DOF. The worst-case scenario is that the blanketing gas control valves
saturate to the fully open position, and excessive blanketing gas is sent to flare.
This scenario can happen because a PID controller does not know or care about the
absolute opening of the valve. It only cares about the incremental change to be added
to the valve opening to reduce the control error (see Sect. 3.1.6). The two controllers
can maintain the 0.5 bar(g) setpoint with infinite combinations of valve openings,
including one fully open. This design is similar to the troubleshooting example of
Fig. 8.5 discussed in Sect. 8.3.4.
This design may have achieved the goal of maintaining the tank pressure but is
far from optimal due to the potential excessive blanketing gas consumption.

The Bad Design

Split-range control is a common approach to reducing the number of independent

control handles by linking the two valves in series (see Table 2.7). For this tank
pressure control example, as shown in Fig. 12.3, a split-range controller can improve
the control performance over the design shown in Fig. 12.2.
A single pressure controller PC-101 manipulates two control valves, PCV-101A
and PCV-101B. The control with PCV-101A is reverse action, while the one
with PCV-101B is direct action.
Since the dynamics from the two valves to the tank pressure are comparable, the
split point is set to the default 50%. When the tank pressure is below 0.5 kPa(g), the
pressure controller first reduces the venting gas flow until the valve PCV-101B is
fully closed. If the pressure is still below the setpoint, the controller will introduce

Fig. 12.2 Tank gas blanketing control: a naive design

438 12 Control Typicals: Plant-Wide Control and Unit Control

Fig. 12.3 Tank gas blanketing control: a improved design

Fig. 12.4 Tank gas blanketing control: split point

blanketing gas by opening valve PCV-101A. On the other hand, if the pressure is
higher than its setpoint, the controller will close the blanketing gas first before opening
the venting valve. See Fig. 12.4a for an illustration of the split-range pattern.
A careless design may result in a split-range control shown in Fig. 12.4b, where the
controller adjusts both valves in opposite directions simultaneously, unnecessarily
consuming the blanketing gas and increasing wear and tear on the valves.

The Good Design

Figure 12.3 is an improved design over Fig. 12.2. However, there is still room for
improvement. According to the operating requirement, there is no need to maintain
the tank pressure strictly at 0.5 kPa(g). It is perfectly acceptable to let the pressure
fluctuate without taking any control action, as long as the pressure fluctuation is
within an acceptable range.
12.1 Liquid Tank Blanketing Control 439

Along this line, a further improved design uses two pressure controllers with dif-
ferent setpoints, i.e., the dual pressure controller approach (Sect. 2.2.2). As shown in
Table 2.7, dual-controller control increases the number of control targets to consume
the extra DOF, contrary to the split-range control scheme just discussed above.
The process values for the two pressure controllers come from the same pressure
transmitter.1 See Fig. 12.5 for an illustration of the controller configuration and note
the difference with the naïve design in Fig. 12.2.
With the dual-controller control design, there are three operating scenarios:
1. If the tank pressure PT-101 is higher than the setpoint of controller PC-101B
(=0.8 kPa), the flaring valve will be opened to let out the excess gas and the inlet
valve for the blanket gas remains closed.
2. If the pressure is lower than the setpoint of PC-101A (=0.2 kPa), then the inlet
blanketing gas flow valve is opened to let in more gas, while the flaring valve is
fully closed.
3. When the pressure is between 0.2 and 0.8 kPa, which is expected to be the case
most of the time, both valves will be fully closed.
Compared with the split-range control scheme where the pressure controller acts as
the normal regulatory controller, the dual controllers perform as two independent pro-
tective controllers, one against low pressure, the other high pressure. At steady-state,
at least one valve is fully closed. When the pressure is between 0.2 and 0.8 kPa(g),
both valves are fully closed. Therefore, the gas consumption and venting are both
minimized.
See Sect. 4.1.6 for the pros and cons between split-range control and dual-pressure
control.

Fig. 12.5 Tank gas blanketing control: a good design

1 If two independent pressure meters are used, the measurement may have discrepancies that poten-
tially affect the two pressure controllers’ setpoint spacing. Besides, the use of two meters also
increases the cost.
440 12 Control Typicals: Plant-Wide Control and Unit Control

Table 12.2 Loop configuration details

Parameter Unit PC-101A PC-101B
Algorithm type PI PI
Equation type I-PD I-PD
Typical PV value kPa(g) 0.5 0.5
Setpoint kPa(g) 0.2 0.8
Setpoint limit kPa 0.2–0.3 0.6–0.8
Initial tuning: gain %/% 2.0 2.0
Initial tuning: integral Min 5.0 5.0
Initial tuning: derivative Min 0.0 0.0
Direction of action REVERSE DIRECT
Fastest scan rate/sequence Default Default
Output scaling 0% = Fully closed, 100% = Fully open
Fail-safe position Fail-close
PV filtering Min TBD TBD
PV tracking No No
Anti-reset windup N/A N/A
Bumpless transfer N/A N/A

12.1.4 Implementation and Operation

The implementation and operation are straightforward because the two pressure
controllers (PC-101A and PC-101B) are independent simple PID controllers. The
PID tunings can be treated the same way as for standalone PID controllers.
The key configurations for the two pressure controllers in Fig. 12.5 are listed in
Table 12.2. As suggested in Sect. 7.4.4, these types of configuration details should be
provided by a process control engineer. The DCS configuration engineer or technician
often does not know or care about the correct values. They most likely will leave
many of them to the system default. Setting the configurations right the first time is
much more efficient than troubleshooting to correct them later.

12.2 Boiler Drum Level Control

Boilers are prevalent equipment in many industrial plants for generating process
steam. Boilers are also central to thermal power generation. Boiler drum level control
is a vital requirement for operational safety and efficiency.
The boiler drum level control is chosen as a typical process to demonstrate the non-
minimum phase process behavior and the challenges it causes for process control.
12.2 Boiler Drum Level Control 441

For this application, we will again follow the work process in Figs. 6.14 and 7.6, but
with the focus on dynamic response analysis and functional design.

12.2.1 Process Description and Analysis

The boiler drum is where water and steam are separated. A typical steam boiler
application is shown in Fig. 12.6.

Operating Requirements

The steam must be superheated to avoid water carryover to the downstream consumer,
so the operating objective of the boiler drum is the proper separation of steam and
water. Maintaining the proper steam/water interface level is critical—if the level
becomes too low, the boiler can run dry, resulting in mechanical damage to the drum
and boiler water tubes. If the level becomes too high, water can be carried over into
the superheater and steam pipework, possibly damaging downstream equipment.
Therefore, for the steam-generating process, the critical requirement is to maintain
the required steam quality and ensure that the drum’s water level is within a safe
range.

Fig. 12.6 Boiler drum level process

442 12 Control Typicals: Plant-Wide Control and Unit Control

Process Flow Analysis

The process flow of the steam-generating process with a boiler is relatively straight-
forward. Boiler feedwater is pumped to the boiler, where it is heated into superheated
steam and feeds the downstream steam consumers. The streams of concern include
the inlet feedwater and the outlet steam flow. There are requirements on pressure and
the temperature inside the drum, but that is a separate topic and is not included in
the discussions here.

Supply and Demand Analysis

The mass balance in the boiler is between the boiler feedwater and the generated
steam. Feedwater supply is typically assumed to be unlimited.
Since the level in the drum is an unstable integrating process, the mass balance
between the feedwater and generated steam is reflected in the drum level. Any imbal-
ance will cause the drum level to ramp up or down continuously. The drum level is
thus a critical variable to control.
As a steam generator, the operation is demand-driven, the steam need from down-
stream dictates the production rate. The boiler feedwater is assumed to be unlimited
and can serve as the swing stream with unlimited capacity.

Cause and Effect Analysis

The primary concern of the boiler operation is the drum level. The variables that
affect the boiler level include the feedwater flow and steam consumption.
Directionally, an increase in the feedwater will cause the water level in the drum
to go up, while an increase in steam consumption will cause the level to go lower.
However, an extraordinary phenomenon with boiler water level is the so-called shrink
and swell effect, which leads to inverse responses in the process dynamics. The
inverse response is the main challenge for control.
The primary disturbance is the sudden change in steam demand, dictated by the
downstream steam consumers. The drum pressure can be another disturbance. The
bubble size is affected by pressure, so if a boiler experiences a sudden extra demand
for steam, its pressure drops.
The cause and effect relationship is shown in the cause and effect table (CET) in
Table 12.3.

Table 12.3 Cause and effect Boiler drum level Remark

table for boiler drum level
control Feedwater flow rate ↑  Inverse response
Steam flow rate ↑  Inverse response
Drum pressure ↑ 
12.2 Boiler Drum Level Control 443

Degree-of-Freedom Analysis

Naturally, the drum level is the controlled variable in the level control scheme. The
boiler feedwater is the main handle for the level control and is the manipulated
variable.
The steam flow rate is a key process variable (KPV) but is determined by down-
stream operation. The control handle for the steam flow is not available to the
drum level control scheme. However, the steam flow is measured online. Based
on Table 2.7, the option for including the steam flow in the control design is to treat
it as a disturbance variable, and the choices for control include feedforward control
and ratio control.

Dynamic Response Analysis

The dynamic behavior of the level response to the changes in feedwater flow and
steam flow is exceptional. The boiler drum level is an integrating process. Any mis-
match between the feedwater supply and steam consumption will cause a continuous
ramping in the drum level. To further complicate matters, the boiler drum level is
notorious for its inverse response. The inverse response poses a unique challenge for
control, and thus the complexity in control design:
1. From feedwater flow to the drum level: When the drum level is low, more feed-
water should be added to increase the level. However, because the feedwater is
cooler than the water/steam in the drum, the cooler water will cause some steam
to condense, which in turn causes the volume of water/steam to decrease. See
Fig. 12.7 for an illustration. As a result, adding more feedwater tends to decrease
the drum level instead of increasing it initially.
Once the temperature is restored, the drum level will start to increase. This inverse
response may confuse the controller with feedback control and cause it to take
the wrong actions.

Fig. 12.7 Inverse response of boiler drum level

444 12 Control Typicals: Plant-Wide Control and Unit Control

2. From steam flow to the drum level: A sudden steam load increase will cause a drop
in the steam drum pressure because the firing rate cannot increase fast enough to
match the steam production rate at the new demand level. The decrease in pressure
causes a fraction of the saturated water in the boiler to vaporize immediately and
increase the steam drum level. Once boiler pressure is restored, the steam bubbles
contract, and the measured water level drops suddenly. This behavior is an inverse
response to the load change.
As the net steam draw rate has gone up, the net water flow to the boiler is expected
to increase because the total mass of water inside the boiler is falling. The false
response in the level may cause a feedback control to respond in the wrong
direction, at least initially.
Inverse response (or non-minimum phase response) is illustrated in Fig. 6.11d and is
one of the challenges for feedback control. The difficult dynamics lead to different
control designs.

12.2.2 Control Design

The challenge in the drum level control design is not in the complexity of the pro-
cess flow scheme but rather related to the nonlinear dynamic behavior between the
feedwater and the drum level and the unreliable measurement of the drum level.

Control Scope

The boiler operation is a relatively isolated process. As explained above, the inter-
action with the upstream process flow is via the boiler feedwater, and the interaction
with the downstream process flow is the steam flow rate. The boiler feedwater is
assumed to be unlimited, and the steam flow is treated as a critical disturbance vari-
able.
Within the boiler operating unit, the drum level may interact with the drum pres-
sure. Good pressure control is a requirement but is not the focus of the discussion
here.

Control Strategy

The control objective for the level control loop is to maintain a steady water/steam
interface level in the drum for efficient steam generation and safe operation.
The normal regulatory control is the drum level control, with the feedwater flow
as the manipulated variable. The downstream steam flow has a substantial impact on
the drum level, and feedforward action is considered.
There are no process variables that need to be protected with protective controllers.
12.2 Boiler Drum Level Control 445

Control Algorithm

The challenge for boiler drum level is dealing with the nonlinear dynamic response
and the difficulties in level measurements. The control technology selection is limited
to DCS-based PID control schemes, typically with three options, from simple to more
complex: single-element, two-element, and three-element control:

1. Single-element control: The single-element control strategy is a simple standalone

PID-level control, as shown in Fig. 12.8.
The drum level is the controller PV. The level controller directly manipulates the
feedwater valve, which is at the pump discharge.
One or more boiler feedwater pumps push water through feedwater control valves
into the boiler drum. In the case of multiple pumps in parallel, split-range control,
fan-out control, or selective control (Sect. 2.2) can be considered to manipulate
the pumps. Variable-speed boiler feed pumps are sometimes used, in which case
the pump speed will be the manipulated variable instead of the flow control valve
(see Sect. 11.2 on pump control).
The simple feedback control design described above is called single-element
control because it uses only one feedback element—the drum level measurement.
2. Two-element control: The two-element control is a standard-level-flow-valve cas-
cade control and offers some incremental improvements over the single-element
control. The inner loop is a flow controller which stabilizes the pump discharge
flow, and the outer loop is the level controller manipulating the flow control set-
point. See Fig. 12.9 for the control schematic.

Fig. 12.8 One-element control for boiler drum level

446 12 Control Typicals: Plant-Wide Control and Unit Control

Fig. 12.9 Two-element control for boiler drum level

For large capacity boilers, there may be two or three feed pumps. The pumps
will be switched on or off to follow the variation in feedwater needs. As a result,
the overall dynamics of the pump discharge valve to the discharge flow can be
complex and slow. Introducing a flow control as the inner control loop for level
control can result in a noticeable improvement in the level control performance.
See Sect. 2.2.4 for the advantages of cascade control over simple control.
3. Three-element control: Like the feed flow, the steam flow changes can also cause
significant deviations in the drum level and may even trip the boiler. Changes
in steam flow will directly affect the drum level before the controller starts to
adjust the feedwater flow. Due to the slow and challenging level dynamics, the
feedback control performance of single-element and two-element controls is often
not satisfactory in responding to steam flow disturbance. In addition, the drum
level’s inverse response to feedwater flow and steam flow makes control even
more challenging.
The difficulty lies in that the PID controller needs a high controller gain and
strong derivative action for fast response. On the other hand, the PID control
loop requires a low gain, long integral time, and no derivative to achieve stable
performance. It is thus challenging to strike a balance between performance and
stability.
To more effectively compensate for the influence of steam flow variations, the
three-element control is commonly used. The three-element control is a cascade
control with feedforward (see Sect. 4.1.2 for feedforward control). Steam flow rate
is a measurable disturbance as a feedforward variable to the single-element or two-
element control scheme. Changes in the steam flow rate will almost immediately
be counteracted by similar changes in the feedwater flow rate (Fig. 12.10).
12.2 Boiler Drum Level Control 447

Fig. 12.10 Three-element control for boiler drum level

Although three-element drum level control is typically superior in performance

to single- or two-element control, the feedforward element is typically turned off
at low boiler loads. The reason is that steam flow measurement can be inaccurate
at low rates of steam flow. Once the boiler load is high enough for steam flow to
be measured accurately, the feedforward must be activated bumplessly (via the
incremental form).

Functional Design

The overall functional design of the level control scheme for the boiler drum is
straightforward. However, there are several things to watch out for, including the
level compensation and feedforward calculation:
1. Level compensation: The drum level is challenging to measure accurately due to
the turbulent condition with the water/steam interface and the density difference
between the water and steam phases. Both are inherent to the process.
Typical level measurement is by gauge glass for direct visual indication. Remote
level sensors based on differential pressure, displacers, conductivity, and radar
are typically used for closed-loop control. The DP-based level meter is the most
commonly used for control.
448 12 Control Typicals: Plant-Wide Control and Unit Control

However, due to the density difference between steam and water, the level mea-
surement is always lower than the actual level in the drum when using either gage
glass or DP-based sensors. Level compensation with density is usually required
to improve the level measurement. See Sect. 10.5 for details in boiler drum level
compensation.
2. Feedforward calculation: Feedforward is generally set up to maintain a 1:1 mass
balance between the steam flow and feedwater. If both flow meters are set up for
the same span in the engineering unit, the feedforward gain is usually set to 1.0.
For better performance, a dynamic feedforward approach may be more beneficial
than a straight gain. However, this requires a good understanding of the dynamic
response characteristic (gain, delay, and lag) from the feedwater flow and the
steam flow to the drum level and is thus not commonly implemented.

Implementation and Operation

Drum level is an integrating process, and integrating control loops are difficult to
tune. The control loop can quickly become unstable if the controller’s integral time
is too short (i.e., high integral gain). The process-imposed requirement for a long
integral time makes the loop slow to recover from disturbances to the drum level.
Following the tuning recommendation for cascade control loops, the inner flow
loop should be tuned first, then the level loop. The flow loop tuning is straightfor-
ward, but the level tuning can be tricky. Model-based tuning based on a proper step
test is recommended. The Lambda tuning method (Table 9.7) for controllers usually
provides stable control loops:

1 2λ+θ
Kc = . (12.2)
K p (λ + θ )2

12.2.3 Practical Considerations

A practical challenge is the credibility of the level measurement. Depending on

the technology used, the interface level measurement between the steam and water
is usually inaccurate and inconsistent due to the density difference between steam
and water. Level compensation with density is typically required, as discussed in
Sect. 10.5. After all, we can’t control what we can’t measure.
For boiler drum level control, some improvements in process design or operat-
ing procedure can simplify the response dynamics to facilitate process control. For
example, preheating the boiler feedwater and adding drum baffling can reduce the
inverse response behavior.
12.3 Well Test Separator Control 449

12.3 Well Test Separator Control

Well testing is an essential and routine operation in upstream production to acquire

production data on individual wells. The collected data is used to monitor the pro-
duction rate for better production management and improve the understanding of
the hydrocarbon properties of the underground reservoir where the hydrocarbons are
trapped. Well testing is chosen as a typical process in this chapter for the following
reasons:
1. There are some unique control challenges related to the inherent limitations in
measurement technology. Process control can help to alleviate the limitation.
2. Well test sequence is a typical sequential control. Sequential control is a gray area
that falls under the scope of process engineering, operation, instrumentation, and
process control. This application shows the connection and interaction between
sequential control and standard process control.
3. This application also demonstrates that the most effective improvement to control
solutions is often from improvement in the process design or advancement in the
instrumentation technology. A controller cannot do the impossible. The process
has to be designed to accommodate its control function. Without the process doing
its part, the controller cannot adequately do its job either.
The classical approach for well tests is through a well test separator where the
oil, gas, and water are separated and individually measured.

12.3.1 Process Description and Analysis

In upstream production, a three-phase test separator is standard equipment that mea-

sures the individual phase production of oil, gas, and water. However, the control
and operation of test separators have many practical issues, and there is much room
for improvement.

Operating Requirements

An upstream exploration and production (E&P) company may have thousands of

production wells in operation and dozens or hundreds of test separators for well
testing. Each test separator is responsible for dozens of wells with various production
capacities. The test separator can test only one well at a time, and each well test may
last up to 24 h.
Two operating requirements are of significant importance to us. One is the schedul-
ing and executing of well test in rotation. Well testing has traditionally been a labor-
intensive manual operation involving multiple dedicated teams. Automated well test-
ing has been a strong demand and is making progress as part of improved sequential
450 12 Control Typicals: Plant-Wide Control and Unit Control

control. The other requirement is to ensure accurate and reliable measurements of

the individual phase flows, which is not as trivial as commonly believed. The two
requirements dictate that the test separator operation needs to
1. emulate the normal production condition. The well production rate is affected by
the wellhead pressure. When the well is diverted from the production header to
the test header, the wellhead pressure should be maintained as close as possible
to the wellhead pressure during normal production so that the same production
rate is maintained while in test and normal production;
2. separate the incoming multi-phase flow into individual phases (gas, oil, and water).
The phase separation is affected by the pressure, oil level, and interface level.
Adequate control of the pressure and levels in the separator is required;
3. measure the individual phase flow rates. The separated oil, gas, and water are
individually measured and recorded. The production rate can vary from well to
well and over the well’s life cycle.

Process Flow Analysis

A typical production system from wellhead to test separator is shown in Figs. 12.11
or 12.12.

Fig. 12.11 From wells to test separator

Fig. 12.12 From wells to test separator, with multi-port selection valve (MSV)
12.3 Well Test Separator Control 451

The produced multi-phase fluid from a well is sent to a production manifold.

The fluids from multiple wells commingle at the manifold and flow under pressure
to one or more production separators (bulk separators) in the downstream facility
(processing station). The multi-phase fluid is separated into gas, oil, and water at the
production separator for further processing.
The multi-phase fluid enters the separator as a turbulent flow. The separator pro-
vides sufficient residence time for the oil and emulsion to form an oil layer (“oil
pad”) at the top. The free water settles to the bottom. The oil weir maintains the oil
level, and the controller maintains the water/oil interface level.
Although the total production rates of oil, gas, and water from multiple wells are
measured, each well’s production rate is typically unavailable online. For production
monitoring and planning, it is crucial to know the production rate at each well.
A test separator is used for measuring the individual well production, one well at
a time. The produced well fluid from a selected well is regularly diverted to a test
separator, typically monthly, where the fluid is separated into gas, oil, and water.
The flow rate of each phase is individually measured and totalized to calculate the
production rate. The diversion of the fluid from the production separator to the test
separator is typically automated, either via a range of isolation valves (Fig. 12.11),
or more commonly, a multi-port selector valve (MSV) (see Fig. 12.12).
Figure 12.13 is an illustration of a conventional single-weir test separator. The
oil is skimmed over the weir and flows to the oil chamber downstream of the weir.
The oil level is controlled by a level controller that operates the oil outlet flow. The
produced water flows from a nozzle located upstream of the oil weir. An interface
level controller senses the oil/water interface and manipulates the outlet water flow to
maintain the desired interface level. The gas flows horizontally and then out through

Fig. 12.13 A three-phase test separator

452 12 Control Typicals: Plant-Wide Control and Unit Control

a mist extractor. A pressure controller is usually provided to keep the pressure inside
the vessel. In some designs, the vessel pressure is let to float with the downstream
operation.

Operating Limits and Constraints

Even though the test separator may be running round-the-clock, individual wells are
tested only in turns, and thus the test data is very infrequent, typically once every
few weeks. Besides, the measurements with the test separator are very unreliable as
the results are affected by many factors. The key constraints that affect the operation
include the following:
1. Inconsistent well test procedure: A well test has many steps that are prone to
human operating errors. For example, wellhead pressure condition, well and sep-
arator line-up, separator purging, and reporting.
2. Inaccurate flow measurements: The test separator is designed to test multiple
wells, of which the production rate may vary drastically. For the same well, as the
well production depletes over time, the flow rates may also change substantially.2
The production flow is measured with online flow meters. Flow meters have
effective measurement ranges, typically between 10 and 100% of the full range
(typical turndown ratio of 10:1). Below the turndown limit, the measurement
would become less accurate (e.g., orifice-based) or even unavailable (e.g., vortex
meters). Since the flow meter is sized to cover the highest production rate, the
same flow meter may be significantly too large for low-producing wells, and thus
no accurate measurement may be obtained.
With the current well testing practice, the well test data is notoriously unreliable. The
test results typically require heavy post-processing based on human interpretation
and cleansing.

Supply and Demand Analysis

For the test separator, the well fluid flows into the separator under wellhead pressure.
Each phase flows out the separator under supply-driven operation once separated
into gas, oil, and water. Therefore, from inlet to outlet, separator operation is strictly
supply-driven. The placement of valves and instruments complies with the supply-
driven model. See Fig. 12.13.

2For example, a low gas–oil ratio (GOR) oil well may gradually deplete to become a high GOR
well after some years of production.
12.3 Well Test Separator Control 453

Cause and Effect Analysis

The controlled variables for the test separator include the pressure inside the vessel,
the oil level, and the oil/water interface level. The manipulated variables include the
flow control valve for the outlet oil, the flow control valve for the outlet water, and the
pressure control valve for the gas. The cause and effect relationship is straightforward
for the test separator (Table 12.4).

Dynamic Response Analysis

Both the level and pressure inside the vessel behave as an integrating process. Most
of the time, test separator control is treated as a standard level control application.
However, a practical challenge is in the control valve. Due to wear and tear, the flow
control valve may leak even when fully closed (also called passing). When testing
low production wells, even if the level control output is 0%, there may still be a
non-negligible amount of flow through the control valve. This flow is below the low
cutoff threshold of the flow meter and thus, cannot be accounted for. These abnormal
valve characteristics as part of the loop dynamics affect both flow measurement and
level control.
Another factor that affects the process dynamics is the wellhead pressure. When a
well is diverted from a production separator to a test separator, the wellhead pressure
may become different from its normal production pressure due to the different flow
line-up, which affects the flow rate, water cut, gas–oil ratio, and other parameters.
The test result will thus not reflect the true well behavior. The wellhead pressure
should be maintained as close as possible between the test and normal operating
modes to achieve more accurate production flow data.

12.3.2 Control Strategy

The main control objectives for the well testing are to achieve accurate and consistent
test results. The goals are achieved by
1. automating the well testing procedure and ensuring accurate production flow
measurement;
2. maintaining the level within the desired range by manipulating the outlet flow
while at the same time ensuring that the outlet flow does not fall below the flow
meter turndown limit when testing low producing wells.

Table 12.4 Cause and effect Vessel pressure Oil level Interface level
table for test separator
Outlet oil flow ↑ 
Outlet water flow ↑ 
Outlet gas flow ↑ 
454 12 Control Typicals: Plant-Wide Control and Unit Control

Well Test Sequence

The well tests need to follow a well-defined procedure to produce consistent and
accurate test results. Well test used to be a manual process but is being automated in
many operations. However, the generic test sequence is the same whether the test is
done manually or automatically.
The design of the well test sequence is a multi-team activity involving different
specialties. Well test sequence falls under sequential control and is closely integrated
with the standard process control requirements. A typical well test will go through
the following five stages:
1. Purge: Purging is intended to clear the test line of the previously tested fluid and
fill it with the fluid from the well to be tested.
2. Forced dump: Dump refers to the action of opening the outlet valve and letting go
of the fluid accumulated in the separator. Before the well test starts, the separator
level needs to be brought to a pre-defined value via a forced dump action. This
level reflects the initial inventory in the tank for flow totalization.
3. Well test: The flow data for all the phases from the metering devices are collected
and totalized during the well test.
4. Forced dump: After the well test, the separator level shall be brought back to the
same level as before the well test starts, via a forced dump action. The same level
before and after the test will ensure that the test results will not be affected by
inventory changes in the test separator.
5. Reporting: The test results shall be calculated, reported, and archived as needed.
We will not go into more details than the above general procedures. However, the
separator control interacts with the test procedure and requires a good understanding
of the context of the control requirements.

Test Separator Control

The separator level needs to be controlled in a relatively tight range for proper
separation of the multi-phase fluid. The fluctuation in the liquid phase in the inlet
flow, from well to well due to different capacities, is reflected in the outlet liquid
flow. The typical test separator control includes the following functionalities, and a
basic control scheme is illustrated in Fig. 12.14:
1. Maintain the test separator pressure so that the well production in test mode closely
mimics that in normal production mode.
2. Provide proper oil level control while ensuring that the flow rate is within the
effective range of the flow meter for accurate oil production measurement.
3. Provide adequate water level (i.e., oil/water interface level) control while ensuring
that the outlet water flow is within the effective range of the flow meter for proper
water production measurement.
12.3 Well Test Separator Control 455

Fig. 12.14 Basic control scheme for a test separator

The following measurement and control are part of the well test separator opera-
tion:
1. Pressure control to maintain a minimum differential pressure across the valve
to ensure the gas flows forward. The pressure is controlled by manipulating the
pressure control valve at the gas outlet.
2. Gas production measurement with a gas flow meter. The raw flow is typically
compensated with pressure and temperature for improved measurement accuracy.
The compensated gas flow is generally converted to standard volumetric flow (see
Sect. 10.3 for flow compensation and conversion).
3. Oil level control and oil flow measurement. Standard level-to-flow cascade control
is used to maintain the oil/condensate level. A flow meter is installed to provide
flow measurement.
4. Water level control and water flow measurement. Standard level-to-flow cascade
control is used to maintain the interface level. A flow meter is installed to provide
water measurement.
In practical applications, the extensive range of variations in the production flows
from well to well may demand a very high turndown ratio of the flow meters beyond
what the current technology can economically offer. Similarly, a control valve also
has an effective range or turndown limit, beyond which the control will become less
accurate.
Under these circumstances, one or more of the following enhancements can be
considered to improve measurement accuracy and ensure control stability:
456 12 Control Typicals: Plant-Wide Control and Unit Control

1. Dual gas flow meters: For improved pressure and differential pressure control,
two parallel gas flow meters (even lines) can be used in tandem to increase the
range of gas measurement. See Sect. 10.2 for a dual meter approach.
2. Semi-dump control for liquid level: The outlet flow is temporarily stopped if
the production rate goes below the flow meter’s turndown limit. Once sufficient
inventory is built in the separator, the outlet flow resumes. This scheme is similar
to on/off control (also called dump control), but is different in that the dump
control function is only activated when the flow rate is below a threshold, hence
the name semi-dump control.

12.3.3 Functional Design

A process control schematic showing an implementation of the semi-dump logic is

illustrated in Fig. 12.15.
The semi-dump logic provides the following functionalities to make the flow
measurement cover the flow range of all the wells:
1. For high-producing wells, the level is controlled like a standard level-flow-valve
cascade control loop.
2. For low-producing wells, if the outlet flow drops below the minimum flow limit,
the outlet flow line is temporarily closed to allow the liquid inventory in the
separator to build up. Once the level is high enough, the level-flow-valve cascade
loop is re-activated, and the outlet flow should be above the minimum flow limit
again.
3. Transition: The transition between the high and low producers is automatic and
bumpless.

Fig. 12.15 Semi-dump control test separator level

12.3 Well Test Separator Control 457

When testing low production wells, the level control starts with the standard
cascade control. Under level control, the outlet flow will decrease and be clapped at
the minimum flow rate Fmin. The level would then start to fall. Once the level is
below the level setpoint by a pre-defined margin, the semi-dump logic would kick
in and switch off the on/off valve USV-101. With the outlet flow stopped, the level
will then slowly rebuild.
Once the level is above a specific limit (e.g., 10% higher than the setpoint), the
on/off valve is opened again, and cascade control resumes with outlet flow above
the turndown limit. Figure 12.16 shows the possible level/flow interaction during
semi-dump control.
Semi-dump control is a unique control scheme designed for test separators. It is
not a universally recognized term. Nevertheless, the practice of semi-dump control
has been very successful and is thus provided here as a creative example of PID
control.

Integration Between Test Sequence and Control

The well test sequence and the semi-dump logic interact with each other and thus
need to be considered together:
1. Purge: Purging is to clear the existing fluid in the pipeline and replace it with the
fluid of the well to be tested. Depending on the production rate, the purge time
varies from well to well. The test sequence needs to set a fixed duration or define
a triggering event to end the purging.
During the purge stage, the level control is kept in cascade control. Since the
purpose of the purge step is to prepare the pipeline only, there are three scenarios:
a. There is sufficient purging fluid to replace all the liquid in the separator vessel.
The final level is at the setpoint under level control.
b. The separator starts at a low level, and there is not enough purging flow to fill
the separator to the setpoint. So the well test starts with the separator level that
is lower than the setpoint.

Fig. 12.16 Level and flow during semi-dump control

458 12 Control Typicals: Plant-Wide Control and Unit Control

c. The purging flow is so low that the separator level control is in semi-dump
mode with the on/off valve closed. At the end of the purge phase, the separator
level is above the setpoint (see Fig. 12.16) but below the dumping limit (e.g.,
10% above the setpoint).
The well test logic needs to provide appropriate considerations for all the above
scenarios.
2. Forced dump: A forced-dump step is necessary to ensure that the separator level is
at the level control setpoint. Forced dump stops automatically when the level is at
the setpoint. The criterion for stopping the forced dump is based on the separator
level and outlet flow.
3. Well test: During the well test step, the level control automatically transitions
between standard cascade control mode and semi-dump control mode.
4. Forced dump: Again, the semi-dump logic is designed to automatically close the
valve once the separator level reaches its setpoint. The forced dump ensures that
the well test starts and stops with the same liquid level in the separator. The logic
to reliably start and end the forced dump is an important consideration.
5. Reporting: The totalization of the different phases of flows stops now, and the
production rate is calculated. Presenting the results to the operator and recording
the results to an electronic report is part of the test sequence.

12.3.4 Implementation and Operation

Semi-dump control is an ad hoc approach to compensate for inaccurate flow mea-

surement caused by low flow rate. The side-effect is that the interface level is artifi-
cially disturbed and may inadvertently affect the oil/water separation as well as the
gas/liquid separation:
1. The flow controller FC101 should have a low setpoint limit configured to be
the same as the low flow limit Fmin. The level controller output will saturate at
this low limit. If the outlet flow remains low-limited, the separator level will start
to fall (out of control). When the level falls below the deadband (LC101.PV<
LC101.SP-x%), the second condition in the above logic is then met; the semi-
dump logic will then close the on/off valve and disable the flow controller, ensuring
zero outlet flow.
2. The interaction between control and safeguarding. When the process is in emer-
gency shutdown (PSD or ESD), the shutdown logic should switch off the on/off
valve USV101, put the FC101 controller to MAN mode, and set the FC101 con-
troller output to 0% (fully closed).
3. The setpoint of the flow controller FC101 is configured to track the process value
(PV-Tracking ON) in manual mode so that the level controller will be properly
initialized when the flow controller is put back to cascade.
12.3 Well Test Separator Control 459

Compared with a straightforward on/off dump control, the semi-dump control

scheme provides a smoother and more stable control on both the level and flow
variables.

12.3.5 Practical Considerations

To conclude this discussion on separator control, let us back off one step and think
about the separation process for improvement.
First and foremost, an inexpensive and reliable multi-phase flow meter (MPFM)
at every wellhead is the ultimate solution. MPFM provides continuous measurement
of the well flows, and can consequently open the door for many real-time control and
optimization opportunities. However, in the foreseeable future, an MPFM at every
wellhead is still not realistic due to affordability reasons. Test separator will continue
to be the primary choice for well testing.
However, some new separator designs, such as a double-weir separator, can
mechanically ensure the oil and interface level; therefore, no tight level control is
needed. For example, the so-called bucket and weir separator (Campbell 2004) is
illustrated in Fig. 12.17. The height of the oil weir controls the liquid level in the
vessel. The difference in the height of the oil and water weirs controls the oil pad’s
thickness based on specific gravity differences.

Fig. 12.17 A three-phase test separator, with double weirs

460 12 Control Typicals: Plant-Wide Control and Unit Control

It is easy to verify that

h 2 ρ2 − h 1 ρ1
(h 1 − h) ρ1 = (h 2 − h) ρ2 → h= (12.3)
ρ2 − ρ1

where h is the interface level, h 1 is the height of the oil weir, and h 2 is the water weir
height. ρ1 and ρ2 are the oil and water density, respectively.
With this design, the interface level is mechanically guaranteed, and thus the
water/oil interface level controller LC-102 in Fig. 12.14 is no longer needed. Both
the oil and water flow over the weirs to the buckets of their own, where simple
level control is needed for inventory control purposes only, and the level control
performance does not impact the separation.
This application demonstrates that process design has a considerably higher
impact on controllability than control design, whether classical or advanced (Luyben
et al. 1998).

12.4 Firewater Control

Adequate firewater supply is essential to the operation of any operating plant. How-
ever, the proper design of the firewater control system was not receiving the attention
it deserves.

12.4.1 Process Description and Analysis

A firewater network is the most common and economical fire suppression system,
and water is the best medium for controlling the spread of fire. A firewater system
consists of an independent fire grid main or ring main fed by permanently installed
fire pumps taking suction from a water source with suitable capacities, such as storage
tank, cooling tower basin, river, and sea.

Operating Requirements

The operating objective is to provide sufficient water with the desired pressure. Since
firefighting is an unpredictable emergent event, the firewater should be ready at all
times. There are two modes of operation for the firewater process which are normal
mode and fire mode:
1. Normal mode: When there is no demand for firewater, the ring pressure should
be maintained within a given range, e.g., 800–850 kPa. This pressure serves two
purposes; one is to keep the system ready for emergency fire demand. The other
12.4 Firewater Control 461

is for automatically detecting a fire demand and triggering the transition to fire
mode.
2. Fire mode: In case of an increase in firewater demand, for example, from the
opening of a fire hydrant, the ring pressure should be quickly brought up to
firefighting pressure, e.g., 1150 kPa(g), and this pressure should be maintained.
The actual pressure requirement is dependent on the height the firewater needs to
be delivered and thus differs case by case. The pressure of the ring main should be
protected against going overly high, e.g., over 1200 kPa(g), to potentially damage
the piping and pump.

Process Flow Analysis

A firewater system usually consists of three essential parts, the firewater storage tank,
fire pumps, and the firewater distribution system (Fig. 12.18):
1. The firewater is typically drawn from a large-capacity water source such as storage
tanks, cooling tower basins, rivers, and the sea. The capacity should be adequate
to meet peak demand for at least 4–6 hours (Vervalin 1981).
2. Centrifugal pump has been the standard choice for fire pumps because of its
compactness, reliability, easy maintenance, hydraulic characteristics, and variety
of available drivers. An outstanding feature of a centrifugal pump is the inverse
relation between discharge pressure and flow rate at a constant speed, whereby
flow rate is reduced when the discharge pressure (or head) increases. See Sect. 11.2
for detailed discussions on pump control.
The fire pumps are a group of pumps consisting of at least a jockey pump and
the main pump. The jockey pump runs on and off frequently to maintain the
distribution system’s correct pressure during normal operation. In case of a fire,
the main pump automatically starts and provides the firewater for fighting the fire.
Typically, there are at least two main pumps, one is electric motor-driven, and the
other is driven by a fully independent power source such as a diesel engine. The
latter serves as a backup. Each jockey pump typically also has a backup.
3. The distribution system is generally designed with a ring-main system. The ring
main is a primary loop permanently connected to the pumps so that there are two
routes for water to flow in case one side is blocked. Fire hydrants are distributed
along the ring main to deliver the high-pressure firewater.

Supply and Demand Analysis

The firewater process is a typical demand-driven operation. During normal mode, no

firewater is needed; the jockey pump automatically starts and stops to compensate for
the slow and gradual loss of water in the ring main due to leakage and evaporation.
462 12 Control Typicals: Plant-Wide Control and Unit Control

Fig. 12.18 Process flow scheme for a firewater process

In fire mode, the firefighting need determines the water demand. The main pump
automatically starts and supplies the water needed. The dump line sends excessive
water back to the storage tanks.

Cause and Effect Analysis

The control targets include the ring main pressure and the pump discharge pressure.
The available control handles include the ring main control valves and the dump
valve. The most critical control handle, however, is the on/off switch of the pumps.
Firewater demand causes the ring main pressure to decrease, which in turn causes the
pump discharge header pressure to decrease. In the normal mode (idle mode), this
sudden drop in pressure triggers the start of the main pump to increase the pressure
to the firefighting pressure quickly. Once in the fire mode, the ring main pressure
is regulated by the pressure controllers. A decrease in the ring pressure causes the
pressure valve to open and the dump valve to close. The cause and effect table is thus
given as in Table 12.5.

Dynamics Response Analysis

The centrifugal fire pump has an important characteristic for the capacity control of
the firewater. The discharge pressure (or head) and the flow have an inverse relation-

Table 12.5 Cause and effect table for firewater process

Firewater flow Ring pressure Discharge pressure
Ring main pressure valve ↑   
Dump valve ↑   
Pump starts ↑   
12.4 Firewater Control 463

Fig. 12.19 Pump curve in a firewater process

ship, as shown in Fig. 12.19. Throttling the ring main valves will reduce the flow but
increase the pump discharge pressure. Opening the valves will increase the flow and
cause the pressure to go low. The pump pressure and flow are thus self-regulating,
within a specific operating range.
These characteristics also enable the pump to serve as a dynamic swing control
to respond to the fluctuation in firewater demand during fire mode. The centrifugal
pump for firewater pumping typically has a relatively flat performance curve, offering
a decent range for flow variations. On the other hand, this flat curve also implies that
the main pump has a narrow range for operating pressure variations. As a result,
the pump capacity needs to be carefully considered to meet the operating pressure
requirements.

12.4.2 Control Strategy

The control requirement for the firewater process is relatively simple. The challenge
is in the integration with the operating logic and sequence.
The control objective is to ensure that an adequate supply of firewater is always
available, which is achieved with the following control design:
1. Normal regulatory control: The normal regulatory control is via the jockey pumps’
on/off switching to maintain a normal operating pressure. In fire mode, the regu-
latory control is the pressure controller for the ring main.
2. Protective control: A high-pressure protective controller protects the ring’s main
pressure from going excessively high. In case the ring pressure goes above the
high limit, the protective control opens the dump valve and sends the excess water
back to the storage tanks.
464 12 Control Typicals: Plant-Wide Control and Unit Control

3. Mode transition: The transition from normal mode to fire mode is initiated by
pressure change. When a fire event occurs, some hydrants will be opened, and the
pressure of the ring main will quickly go down. Once the pressure falls below a
specific value, the operating logic automatically starts the main pump to quickly
bring the ring main pressure to the required firefighting pressure. The operation
enters the fire mode. When firewater is no longer needed, manual intervention is
needed to switch off the pumps to return to normal mode.
A PID-based complex control solution and an automated control sequence are
sufficient to meet the control requirements.

Sequential Control

The critical design consideration of the firewater process is the mode transition.
Sequence and logic are required to monitor the ring pressure, initiate the mode
transition, and change the operation modes. The sequence is summarized as follows:
1. Normal mode: During normal mode, the main pump is switched off. The pressure
controllers remain in MAN modes. The jockey pump automatically starts and stops
to maintain the ring pressure between 800 and 850 kPa(g). The ring main pressure
control valve should stay at the low mechanical lock position (typically 10%) to
keep the water flowing.
2. Transition to fire mode: If the firefighters open the hydrants, the ring pressure will
drop. When the pressure drops below 750 kPa(g), the main pump starts automat-
ically to increase the pressure. If the pressure continues to fall after a short time
(e.g., 20 s) due to a failed start of the main pump, the backup pump should start.
3. Fire mode: The two pressure controllers, PC-103A and PC-103D, should now
be switched to automatic (AUTO) mode. PC-103A is responsible for maintaining
the firewater pressure at 1,150 kPa(a). If the supply from the pump is higher than
the demand and causes the pressure to go higher than 1,200 kPa(g), the dump
controller PC-103D opens the dump valve and sends some water back to the
storage tanks.
4. Transition to normal mode: When there are no more firewater needs, the main
pumps must be switched off manually, and both pressure controllers to man-
ual (MAN) mode. The ring pressure will slowly fall back to the range of 800–
850 kPa(g). The firewater system is in normal standby mode again.

12.4.3 Functional Design

Figure 12.20 is the control overview diagram (COD) for a typical control design. An
actual design may be more complex depending on specific requirements.
The pressures of the two firewater headers are individually measured with
PI-103A and PI-103B. The lower of the two pressure values is fed to the pressure
12.4 Firewater Control 465

Fig. 12.20 Firewater control scheme

controller PC-103 as the process value (PV) and ensures the firewater pressures of
both headers meet the operating requirement. The output of the pressure controller
PC-103 is fanned out to the two ring pressure valves, PCV-103A and PCV-103B.
Fan-out control is used because the two pressure measurements are highly correlated
and can only be counted as one controlled variable.
The higher value of the two pressure measurements is fed to the pressure protective
controller PC-103D. If the pressure goes excessively high, the protective controller
PC-103D simply increases the dump valve (PCV-103D) to let more water go back
to the storage tank. The excessive water can be caused by reduced firewater demand
that is below the pump turndown limit.
The operating sequence is illustrated in Fig. 12.21. The main pump starts when the
lower of the two pressure measurements at the pump discharge fall below 750 kPa(g).
During normal mode, the firewater is maintained between 800 and 850 kPa with
on/off control through the start and stop of the jockey pumps. The drop in pressure
transition triggers the transition to the fire mode through the startup of the main pump.
Once the main pump is started, the pressure controller PC-103 is switched to AUTO
and maintains the firewater pressure at 1,150 kPa. The operator manually initiates the
transition back to normal mode after confirming that firewater is no longer needed.

12.4.4 Implementations and Operation

Some practical considerations for the firewater control design are briefly explained
here:
1. Dump valve: Fail-open or fail-close? In case of loss of instrument air, a fail-open
dump valve goes to its fully open position and releases the pressure in the ring
main. This action protects the piping from potential damage due to high pressure;
however, the ring main loses the required water pressure for fire fighting. On the
other hand, a fail-close dump valve will maintain the ring main pressure, but the
466 12 Control Typicals: Plant-Wide Control and Unit Control

Fig. 12.21 Firewater control sequence

high pressure may be unacceptable to the piping and the pumps. Since firewater
readiness is vital, a fail-close valve is preferred. The ring main piping and the
pumps should be fully rated for the highest pressure to protect from high-pressure
damage. Otherwise, secure instrument air (SIA) should be considered.
2. Pressure surge: When entering fire mode, the main pump automatically starts.
Upon startup, the pump creates an initial pressure surge in the ring main. Depend-
ing on the pressure rating, this pressure surge may not be acceptable. Additional
logic is thus required to prepare the control valve for the pressure surge.

12.5 Unit Control: Acetylene Hydrogenation Process

We continue with the acetylene hydrogenation example in Example 7.1 to demon-

strate the control design of a simple unit operation. Although the process is relatively
straightforward, the control design can have many variations. Again, the thinking
process will follow the procedures in Chaps. 6 and 7.
12.5 Unit Control: Acetylene Hydrogenation Process 467

12.5.1 Process Description and Operating Requirements

The acetylene hydrogenation process saturates the acetylene (C2 H2 ) with hydrogen
into ethylene and ethane. The process is commonly found in olefin plant operations:

C2 H2 + H2 → C2 H4 reaction 1
(12.4)
C2 H2 + H2 → C2 H6 reaction 2.

The operating objective is to reduce the acetylene concentration to below 2 ppm

through two hydrogenation converters. The reaction is facilitated by catalysts and
is exothermic. See Fig. 12.22 for a process overview diagram (POD) with the key
process variables.
The critical process disturbances are the variations in the feed composition and
the fluctuations in the heat duty in the heaters and coolers.

12.5.2 Supply and Demand Analysis

The material balance is straightforward. There is only one stream and one flow path
through the unit, so the operation is strictly supply-driven.
The energy balance is primarily the heating and cooling of the process flow.
There is heat added from the heater to reach the reaction temperature, and there are
coolers after the reactors to remove the reaction heat. The primary constraint on the
supply and demand balance is the saturation of the heat exchangers; therefore, the
heat duties of the heater and coolers should be adequate to prevent saturation. An

Fig. 12.22 An acetylene hydrogenation process

468 12 Control Typicals: Plant-Wide Control and Unit Control

advanced operating requirement is the energy distribution between the heater and
coolers to maximize energy efficiency.

12.5.3 Cause and Effect Analysis

The measurements are typically abundant in a chemical reaction process. The key
process variables are the temperatures at various locations and the acetylene con-
centration at the outlet of the two reactors. The control handles are the temperature
control valves at the heater and coolers. The cause and effect relationships among
the key process variables (KPV) and the control handles are given in Table 12.6.
There are more KPVs than control handles; thus, the critical decision is selecting the
controlled variables.
The cause and effect table can be expanded to become Table 12.7, which is one
way to show the cascade cause and effect relationship between the variables. The
intermediate variable (IV) is similar in concept to the secondary variables in a PID
cascade loop. For example, the heat exchange valve TCV-01 directly affects the reac-
tor 1 inlet temperature TI-01, which in turn affects the reactor outlet temperature
TI-02, and thus impacts the outlet C2 H2 concentration AI-01.

Table 12.6 Acetylene TI-01 TI-02 AI-01 TI-03 TI-04 AI-02 TI-05 TI-06
hydrogenation: preliminary
cause and effect table TCV-01 ⊕ ⊕
TCV-02 ⊕
TCV-03

Table 12.7 Acetylene hydrogenation: cause and effect table with cascade
MV IV TI-01 TI-02 AI-01 TI-03 TI-04 AI-02 TI-05 TI-06

TCV-01

TI-01 1

TI-02 1

TCV-02

TI-03 1

TI-04 1

TCV-03

TI-05 1
12.5 Unit Control: Acetylene Hydrogenation Process 469

The ultimate goal of the unit operation is to achieve a 2 ppm acetylene concen-
tration at the outlet of the second reactor, which is AI-02. The reactor outlet tem-
perature TI-04 is the preferred variable to control, considering the low availability
and reliability of the online analyzers.

12.5.4 Dynamic Response Analysis

Simulation shows that the reactor outlet temperature and acetylene concentration are
highly correlated (Al-Dawery and Dakhil 2012). The transit response of the outlet
temperature shows a clean first-order-plus-delay-time (FOPDT) behavior.

12.5.5 Control Design

As discussed in Sect. 7.1.3, at the unit level, regulatory control is primarily concerned
with quality control.
In this example, the acetylene concentration sent down to the cold train is the
quality to be controlled. Based on the supply and demand analysis, capacity control
is not needed for this simple supply-driven unit operation consisting of a straight
flow process stream.
With the cause and effect table (Table 12.7), the control strategy has several
options, from simple standalone PID controllers to complex PID schemes, to model
predictive control:
1. Standalone PID controllers: The inlet temperature primarily determines the reac-
tion, so a simple control solution is to control the reactor inlet temperature.
The pairing of the variables would be TCV-01/TI-01, TCV-02/TI-03, and
TCV-03/TI-05. The control scheme is shown in Fig. 12.23.
However, as the concentration is also influenced by the feed flow rate and the feed
composition (e.g., CO), the improved control structure would be to pair the control
valves with the reactor outlet temperatures, TCV-01/TI-02, TCV-02/TI-04,
and TCV-03/TI-06. The control scheme is shown in Fig. 12.24.
2. Cascade PID control scheme: One of the disturbances to the unit operation is the
fluctuation in the heating (steam) duty. A cascade control scheme can improve
the control performance, as shown in Fig. 12.25.
Other disturbances to the unit operation include the variations in the reaction
heat caused by the changes in feed flow rate, feed composition, and catalyst
condition. Feedforward control can be considered if the variations are known and
measurable, for example, the feed rate or the CO concentration.
The acetylene concentration is the ultimate control target. If online analyzers
are available to measure the acetylene concentration, the temperature control
470 12 Control Typicals: Plant-Wide Control and Unit Control

Fig. 12.23 Acetylene hydrogenation process control: design #1

Fig. 12.24 Acetylene hydrogenation process control: design #2

can be further cascaded to the acetylene concentration for improved control. See
Fig. 12.26.
3. Model predictive control (MPC): Although valve position control (see Sect. 4.1.4)
can be considered to prevent the control valves from saturation, the design and
configuration can be tricky. For example, Reactor 1 outlet temperature setpoint
can be manipulated (within a specific range) to maintain the desired valve opening.
MPC can be considered for this unit operation to achieve the best performance.
MPC can optimize the heat duty balance between the heat exchangers (heaters
and coolers) by including the temperature controller outputs as constraints. If one
12.5 Unit Control: Acetylene Hydrogenation Process 471

Fig. 12.25 Acetylene hydrogenation process control: design #3

Fig. 12.26 Acetylene hydrogenation process control: design #4

of the heat exchangers saturates, MPC automatically shifts the load to other heat
exchangers based on the cause and effect relationship in Table 12.7.
The two critical requirements for model predictive control are the model structure
and model parameters. The model structure describes the cause and effect relation-
ship between the manipulated variables and the controlled variables. The model
parameter describes the dynamic behavior of each cause and effect relationship.
Table 12.7 can serve as the model structure for the control design.
472 12 Control Typicals: Plant-Wide Control and Unit Control

12.5.6 Practical Consideration

The process analysis clarifies the operating requirements and control objectives.
However, there can be many alternative control solutions based on the same process
analysis. To what extent the operating objectives can be met and requirements fulfilled
depends on many factors, such as the controllability by the process design and the
available handles and measurements for control. Unfortunately, the control design
is a mixed bag of arts and technologies and thus requires a high level of knowledge
and experience.

12.6 Plant-Wide Control: E&P Surface Production Process

We conclude this chapter with a plant-wide control example. The plant is an integrated
oil and gas production facility built to handle oil and gas production via miscible
gas injection (MGI)—an enhanced oil recovery mechanism. This multi-billion dollar
plant is a real-life version of the generic process diagram in Fig. 1.3 with a capacity
of 6 MMSCMD of sweet gas, 60,000 BPD of oil production, and 16 MMSCMD
of high-pressure sour injection gas. The operation is characterized by extremely
high-pressure and high-sour gases. For this reason, safe operation is of paramount
importance in the process and control design.
The process analysis methodology (Fig. 6.14) in Chap. 6 and control design
(Fig. 7.6) in Chap. 7 were followed throughout this project from design, implemen-
tation to commissioning and operation. As discussed in Sect. 7.1.2, capacity control
is the primary consideration at the plant level. This example has a very challenging
supply and demand model, and thus the primary consideration of our discussion is
in capacity control, i.e., dynamic material and energy balancing.

12.6.1 Process Description and Operating Requirements

A process overview diagram (POD) is developed from hundreds of pages of PFD

and P&ID and is shown in Fig. 12.27 for the plant-level overview.

Operating Objectives and Product Specifications

The operating objective is to produce oil and gas for sales and sour gas for injection.
The gas processing area, which is the focus of this discussion, comprises gas produc-
tion, separation, cleaning and conditioning, and gas compression. The produced gas
is extraordinarily high in H2 S and CO2 . Many packaged units by various vendors are
implemented to clean and condition the gas, condensate, and produced water. Each
12.6 Plant-Wide Control: E&P Surface Production Process 473

Fig. 12.27 Process overview diagram of an E&P surface facility

unit by itself is a complex unit operation, and all have their specifications to meet,
such as purity, dew point, vapor pressure, and delivery pressure.
The three consumers of the processed gas include, in the order of priority: export
to sales, injection wells in area B, and injection wells in area A.

Modes of Operations and Operating Constraints

In addition to the normal operating mode and startup/shutdown mode, the primary
challenge is the turndown mode, where insufficient production (supply) or consump-
tion (demand) causes the operation to run at a considerably lower capacity. Different
consumers have different priorities; thus, preferential treatments need to be provided
when faced with insufficient gas supply.
A commercial agreement requires the recovered off-gas from the sour gas treat-
ment units (GSU) split 50/50 between areas A and B. Two control valves are provided
to facilitate this flow split.
474 12 Control Typicals: Plant-Wide Control and Unit Control

12.6.2 Process Flow Analysis

This process comprises two areas: plant A is a surface processing facility, complete
from wellheads to export. Plant B wells are outside the system boundary, but some
produced gas is sent to plant A for processing. The produced gases are primarily from
area A wells, with about 1/3 from the existing gas plant B. The gases from plant B
are exclusively for area B injection wells, while the gas from plant A is prioritized to
be exported to the main gas line for sales. Excess gas is sent to area A for injection.
Any remaining gas is sent to area B injection wells. See Fig. 12.27 for the process
flows.
The significant disturbances to the plant operation are the fluctuations in the
gas production in area A (shutdown of wells) and the disruption in export sales.
The startup of the compressors is another significant disturbance and challenge to
operation.

12.6.3 Supply and Demand Analysis

Based on the process flow analysis, there are two supplies (plants A and B) and three
gas demands. Steady-state material balance is adequate since the gas from plant B
has sufficient capacity to swing in case of imbalance between supply and demand.
The gas production from area A wells is usually production-driven but can be
choked back in case of low demand. The gas from plant B is demand-driven, regulated
by the header B pressure. Export gas is demand-driven, dictated by sales. The flows
to the injection wells in both areas are demand-driven, with the flow rate decided by
injection needs.
The main swing stream is the gas from plant B, which is a demand-driven opera-
tion. If necessary, the well production from area A can be reduced, but the dynamics
are slow. If the supply is so low that the demands need to be reduced, the area A
injection is the first stream to cut back, followed by area B injection. Export gas is
the last one to be affected.
The recovered sour gas streams from the GSU units and the condensate processing
unit are supply-driven, and the suction pressures regulate the capacity.

12.6.4 Cause and Effect Analysis

At the plant level, the primary concern is capacity control for balancing supply and
demand. The supply and demand imbalance is reflected in the pressure–flow profile.
It is thus vital to select the appropriate pressure variables to propagate the pressure–
flow changes.
12.6 Plant-Wide Control: E&P Surface Production Process 475

The two supply header pressures, Header A pressure and Header B pressure,
are critical in the supply and demand propagation, and thus they are two controlled
variables. The export gas is demand-driven, and thus the gas flow is a controlled
variable. The gas to injection wells is also demand-driven, and thus the gas flow rates
are controlled variables. The produced gas from plant B operates as a demand-driven
swing stream; its flow rate is dictated by Header B pressure.
The key process variables (KPV) are shown in Fig. 12.28. The KPVs and control
handles are assigned temporary tag names before they are finalized as controlled
variables and manipulated variables.
The preliminary cause and effect table (CET) is given in Table 12.8. The cause
and effect relationships are marked with an “⊕” or “ ” for dominant dynamics with
a positive or negative gain, respectively, and a simple “+” or “-” for the non-dominant
ones.

12.6.5 Degree-of-Freedom Analysis

The pairing of the control targets and control handles are straightforward except
for the two critical header pressures, PC-01 and PC-02. There are more handles

Fig. 12.28 Plant-wide control: key process variables

 476

Table 12.8 Cause and effect table for the generic E&P process in Fig. 12.28
PC01 PC02 PV03 PV04 FC05 PC06 PC07 FC08 PC09 PC10 FC11 PC12 PC13 PC14 FC15 PC16
MV-01 ⊕
MV-02 ⊕
MV-03 ⊕
MV-04 ⊕ +
MV-05 ⊕
MV-06 ⊕
MV-07 ⊕ +
MV-08 ⊕
MV-09 ⊕
MV-10 ⊕
MV-11 ⊕ +
MV-12 ⊕
MV-13 ⊕
MV-14 ⊕
MV-15 ⊕ ⊕
MV-16 ⊕ ⊕
MV-17 − +
MV-18 −
MV-19 +
12 Control Typicals: Plant-Wide Control and Unit Control
12.6 Plant-Wide Control: E&P Surface Production Process 477

than targets, and therefore complex PID control schemes should be considered. The
analysis follows the methodology in Sect. 6.5.

12.6.6 Control Design

Our discussion here is limited to plant-level control. Each unit should have an overall
control strategy of its own.
The normal regulatory control at the plant-level control is primarily for the
dynamic supply and demand balance. With the cause and effect analysis above,
the control overview diagram (COD) with the normal regulatory control loops is
shown in Fig. 12.29.
There are 14 compressors with a total of 18 stages. Therefore, compressor control
and protection are at the center of the control design. The compressor control strategy
and techniques discussions in Sect. 11.3 are an integral part of the control design.
What is noteworthy in this control design includes the following:
1. The complex supply and demand model: For example, the export gas compressor
is demand-driven; the discharge pressure controller regulates the capacity. On the

Fig. 12.29 Plant-wide control: normal regulatory control

478 12 Control Typicals: Plant-Wide Control and Unit Control

other hand, the GSU off-gas compressor and the condensate flash gas compressors
are supply-driven, and therefore, the capacity control is performed by the suction
pressure controllers.
2. The two header pressures, PC-01 and PC-02, are the critical nodes for overall
capacity control. The fluctuations in these two pressures reflect the dynamic sup-
ply and demand swings. For Header A pressure, the normal operating pressure is
maintained by adjusting the wellhead pressure. However, the dynamics from the
wellhead to the pressure header are relatively slow; thus, protective controllers are
provided to protect this pressure header against both high and low pressures. The
low protective controller PC-01A is provided to cut back the gas to the injection
wells if the pressure is temporarily lower than the normal header pressure by 1.0
bar. During normal operation, since the PC-01A setpoint of 74 bar is lower than
the normal header pressure of 75 bar, the pressure controller PC-01A will drive
the valve MV-17 to open fully. However, if the header pressure falls below 74 bar,
the protective controller PC-01A would start closing the flow to area A injection
first, then the flow to area B, through the split-range calculation PY-01A.
If a severe plant upset drives the pressure to above 78 bar during an abnormal
situation, the excess gas will be sent to flare via a protective pressure controller
PC-01B. Header B pressure is maintained by manipulating the gas flow from area
B. This stream from area B is demand-driven. Sufficient swing capacity must be
provided inside the area B control solution to meet the demand change.
The supply and demand analysis in this example shows that the control design of
a new plant may require a change in an existing plant, which may even be outside
the project scope.
3. The equal split of the sour gas flow from two GSUs is a 2×2 multivariable control
problem. Based on discussions in Sect. 4.1, a straightforward ratio controller can
be implemented to ensure equal flow rates. However, the pressure loss across the
control valves is not acceptable. The flow balancing control scheme in Sect. 4.1.4
is implemented to minimize the pressure loss.
There are many protective controls in the design. For example, the supply-driven
streams and the compressors need to have demand override. Each compressor needs
multiple protective control loops to protect the operational integrity. We will not go
into the details here.

12.6.7 Practical Consideration

The actual process is much more complicated than discussed here since this is a
multi-billion dollar project involving many unit operations. However, the analysis
and design presented here provide an overview of the plant-wide control solution for
a highly complex plant with 14 centrifugal compressors and three preferential flow
streams.
12.6 Plant-Wide Control: E&P Surface Production Process 479

With this control solution, the plant startup was smooth, and the operation has
been free of significant problems. The uptime of the 800+ PID control loops in this
plant has reached and sustained a remarkable 80+% several months after the plant
startup.

12.7 Summary

This chapter starts with the example of tank blanketing control in Sect. 12.1. The
process flow is simple, and the control requirement is clear. However, the design
can still go from good, bad to ugly. Bad designs are accepted and lived with for a
long time because they may not be wrong; they are just sub-optimal. Many control
solutions in operation fall under this category. The incremental values in improving
this type of control solution are enormous considering the large number in operation.
Section 12.2 provides a structured discussion on boiler drum level control. This
example emphasizes that a good understanding of the process behavior is crucial,
even for a seemingly simple process.
Section 12.3 presents the example of a well testing process and shows that process
control must produce fit-for-purpose solutions to work around the practical limita-
tions in process design. This example also demonstrates that process improvement
can be the most effective way of improving control performance.
Section 12.4 presents the example of firewater process control. Firewater supply is
a critical requirement for any sizable operating facility, and the control requirement
is the high availability and reliability requirement. An integrated control solution
requires sequential control, normal regulatory control, and abnormal protective con-
trol.
Section 12.5 shows the analysis and design of a unit-level control solution. The
emphasis is on the importance of a good cause and effect relationship which is the
basis for control design, either PID-based base-layer control or advanced process
control.
This chapter is concluded with a complete multi-billion dollar E&P process and
demonstrates the thinking process and control design at the plant level, mainly on
enforcing the plant-level material balance and preferential flows.

References

Al-Dawery S, Dakhil HM (2012) Modeling and control of acetylene hydrogeneration process. Emir
J Eng Res 17(1):9–16
Brannan C (2018) Rules of thumb for chemical engineers - a manual of quick, accurate solutions
to everyday process engineering problems, 6th edn. Elsevier, Amsterdam
Campbell J (2004) Gas conditioning and processing - the equipment modules, vol 2, 8th edn. John
M, Campbell and Company
Kinsley G (2001) Properly purge and inert storage vessels. Chem Eng Prog
480 12 Control Typicals: Plant-Wide Control and Unit Control

Luyben WL, Tyréus BD, Luyben ML (1998) Plant-wide process control. McGraw Hill, New York
Vervalin CH (1981) Fire protection manual for hydrocarbon processing plants, 3rd edn. Gulf Pub-
lishing Co, Houston
Appendix
Special Calculations

A.1 Calculation of Vessel Volumes

For vessel level control, it is often necessary to know the vessel volume in order to
calculate the PID tuning parameters (Sect. 9.5) or perform level linearization action
(Sect. 10.1). This section provides the formulas for calculating the volume of some
commonly encountered vessels. For a more complete list of vessel types, see Brannan
(2018).
Some more complex vessels are often a combination of simpler vessels presented
in this section and can be calculated by breaking them down into components. Some-
times, the vessel is so complex that it is difficult (and also unnecessary) to get the
exact volume; reasonable simplification can be made to reduce the computation
complexity.
The following terminologies will be used for this section:
1. Vessel height H : The vessel height from the bottom to the top in the engineering
unit.
2. Liquid height h: The liquid height from the bottom to the surface of the liquid in
the engineering units.
3. Liquid level LT: The liquid height from the lower tap of the level transmitter to
the upper tap, in percentage (0–100%) (Fig. A.1).

A.1.1 Rectangular Tank

A rectangular tank is probably the simplest for volume calculation, although it is not
common to see them in practice. The volume can be calculated as follows:

© The Editor(s) (if applicable) and The Author(s), under exclusive license 481
to Springer Nature Switzerland AG 2022
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3
482 Appendix: Special Calculations

Fig. A.1 Volume of a

cuboid vessel

⎫
horizontal cross area at height h: A= L·W ⎪
⎬
volume up to height h: V = L ·W ·h (A.1)
⎪
⎭
full volume: V = L·W ·H

A.1.2 Spherical Vessel

The volume of a spherical vessel can be calculated as follows. The horizontal cross
area of the sphere at level h is calculated first, then the volume in the vessel up to
level h is calculated (Fig. A.2):
⎫
horizontal cross area at level h: A = π h (D − h) ⎪
⎪
⎪
π 2 ⎬
volume up to level h: V = h (3D − 2h) (A.2)
6 ⎪
π 3 ⎪
⎪
full volume: V = D ⎭
6

Fig. A.2 Volume of a

spherical vessel
Appendix: Special Calculations 483

Note that the volume is a cubic function of the height. It is thus highly nonlinear. A
regular PID controller with a single set of tunings would be complicated to control
the level of this type of vessel.
The process gain K p , assuming the volumetric flow rate is Fv , is given as

dh Fv
Kp = = (A.3)
dt π h (D − h)

which is a function of the liquid height.

A.1.3 Vertical Cylindrical Vessel

For level control, a vertical cylindrical vessel is probably the simplest vessel found in
a plant. Because of the uniform diameter along the vessel’s height, the relationship
between the level and volume is linear; thus, the relationship between the level and
the outlet flow (Fig. A.3).
The vessel volume can be calculated as follows:
⎫
π 2 ⎪ ⎪
horizontal cross area at height h: A = D ⎪
⎪
4 ⎪
⎪
⎪
⎬
π 2
volume up to height h: V = D h ⎪ (A.4)
4 ⎪
⎪
⎪
⎪
π ⎪
full volume: V = D 2 H ⎪ ⎭
4

Fig. A.3 Volume of a

vertical cylindrical vessel
484 Appendix: Special Calculations

where the area is the horizontal cross-section at level h and the volume is the bottom
up to level h.
Note that the liquid volume is proportional to the liquid level V ∝ h, so is the
liquid height to flow h ∝ F. The process gain K p , assuming the volumetric flow rate
is Fv , is given as
dh Fv 4Fv
Kp = = = (A.5)
dt A h D2
which stays constant at different heights.

A.1.4 Conical Vessel

The volume of a cone-shaped vessel can be calculated as follows. The horizontal

cross area of the sphere at level h is calculated first, then the volume in the vessel up
to level h is calculated (Fig. A.4):
 2 ⎫
π D ⎪
⎪
2 ⎪
Cross-section area at height h: A = h ⎪ ⎪
⎪
4 H ⎪
⎪
⎪
⎬
 2
π D 3 ⎪
(A.6)
volume up to height h: V = h ⎪⎪
12 H ⎪
⎪
⎪
⎪
π 2 ⎪
⎪
full volume: V = D H ⎭
12

Fig. A.4 Volume of a conical vessel

Appendix: Special Calculations 485

Note that the volume is a cubic function of the height, and is highly nonlinear. The
process gain is given by

dh Fv 4Fv H 2
Kp = = = . (A.7)
dt A π D2 h2

A.1.5 Horizontal Cylinder

Horizontal cylindrical vessels are common in operation plants, typically with ellip-
tical bulges. Here for simplicity, only a cylinder shown in Fig. A.5 is discussed.
The calculation of volume can be divided into two scenarios, h < D/2 and h ≥
D/2. The vertical cross-section area A up to height h is calculated first. The volume
is then simply the vertical cross-section area multiplied by the length of the vessel:
1. When h < D/2
 ⎫
2h ⎪
Angle: θ = 2 arccos 1 − ⎪
D ⎪ ⎪
⎪
⎪
2 ⎬
D
Area up to h: A = (θ − sin θ ) ⎪
(A.8)
8 ⎪
⎪
⎪
⎪
D 2 ⎪
volume up to h: V = L · (θ − sin θ ) ⎭
8
2. When h ≥ D/2

Fig. A.5 Volume of a

horizontal cylindrical vessel
486 Appendix: Special Calculations
 ⎫
2h ⎪
Angle: θ = 2 arccos −1 ⎪
⎪
D ⎪
⎪
⎪
⎬
D2
Area up to h: A = (2π − θ + sin(θ )) ⎪ (A.9)
8 ⎪
⎪
⎪
⎪
L · D2 ⎪
Volume up to h: V = (2π − θ + sin(θ ))⎭
8

π 2
3. The full volume (at h = D) is given by V = D L.
4
Note that the volume is nonlinear to the height.

A.2 Calculation of Lambda Tuning Parameters

This section provides a simple illustration on how the λ tuning parameters are derived
for first-order-plus-delay-time processes.

A.2.1 First-Order-Plus-Delay-Time Process

The closed-loop control is related to the process and controller as illustrated in

Fig. 9.2. The close-loop transfer function in Eq. 9.1 is given as follows:

G c (s) G p (s)
G cl (s) = . (A.10)
1 + G c (s) G p (s)

From Eq. A.10, we can derive the controller transfer function as

1 G cl (s)
G c (s) = · . (A.11)
G p (s) 1 − G cl (s)

Assume the process dynamics can be represented with a first-order-plus-delay-

time model:

K p −θs
G p (s) = e (A.12)
τs + 1

where K p (s) is the normalized process gain, τ is the time constant, and θ is the delay
time. Also assume that the desired closed-loop transfer function is a first-order-plus-
time-delay (FOPDT) response with a desired time constant λ:
Appendix: Special Calculations 487

1
G cl (s) = e−θs . (A.13)
λs + 1

The time delay in the closed-loop response is assumed to be the same as the process
delay time since feedback control is unable to eliminate or reduce the delay time θ .
From Eqs. A.11, A.12, and A.13, we have

1
e−θs
τs + 1 λs + 1
G c (s) = ·
K p e−θs 1 − 1 e−θs
λs + 1
τs + 1 1
= · . (A.14)
Kp λs + 1 − e−θs

If a first-order Taylor expansion is used to approximate the term e−θs ≈ 1 − θ s,

we have
τs + 1 1
G c (s) = ·
Kp λs + 1 − (1 − θ s)

1 τ 1
= · 1+ . (A.15)
Kp (λ + θ ) τs

Compared with the PI controller formula in Eq. 3.8, we have

⎧
⎪ τ 1
⎪
⎨ K c = K (λ + θ )
⎪
p
(A.16)
⎪
⎪ Ti = τ
⎪
⎩T = 0.
d

This agrees with the tuning rule for PI controller in Table 9.6. The tuning parameter
derivation for PID controller would be considerably more complicated.

A.2.2 Pure Integrating Process

Similarly, assume a pure integrating process with delay θ :

K p −θs
G p (s) = e . (A.17)
s
From Eqs. A.11, A.13, and A.17, we have
488 Appendix: Special Calculations

1
e−θs
G c (s) =
s
· λs + 1
K p e−θs 1 − 1 e−θs
λs + 1
s 1
= · . (A.18)
K p λs + 1 − e−θs

With Taylor approximation e−θs ≈ 1 − θ s, we have

s 1
G c (s) = ·
K p λs + 1 − (1 − θ s)
1 1
= · . (A.19)
Kp λ+θ

This leads to a P-only controller with gain (see Table 9.7):

1 1
Kc = . (A.20)
Kp λ+θ

Reference

Brannan C (2018) Rules of thumb for chemical engineers - a manual of quick,

accurate solutions to everyday process engineering problems, 6th edn. Elsevier,
Amsterdam
Glossary

Actuator A component of an automatic valve used to open and close the valve.
May be pneumatic or motor-driven.
Affinity Laws The affinity laws (also known as the “Fan Laws” or “Pump Laws”)
for pumps/fans are used to express the relationship between performance vari-
ables such as head, volumetric flow rate, shaft speed, and power. They apply to
centrifugal and axial flows in pumps, fans, and hydraulic turbines.
Asset The buildings, plant, machinery, and other permanent items required to
produce and supply the product.
Availability The availability of an item to be in a state to perform a required
function under given conditions at a given instant of time or over a given time
interval, assuming that the required external resources are provided.
Analog Controller A type of controller as distinguished from a digital controller.
It operates on continuous signals such as voltages, pressures, or currents.
Automation The use of automatic control devices in a process so that human
supervision is minimized.
Basic Sediment and Water (BS&W) is a technical specification of certain impu-
rities in crude oil. When extracted from an oil reservoir, the crude oil will contain
some saltwater and particulate matter from the reservoir formation. The residual
content of these unwanted impurities is measured as BS&W.
Best Practice A method or technique generally accepted as superior to any alter-
natives because it produces results that are superior to those achieved by other
means or because it has become a standard way of doing things, e.g., a standard
way of complying with legal or ethical requirements. (from Wikipedia, https://
wiki2.org/en/best_practice)
Constraint Variable Process output that must be maintained within an operating
range.
Control Element A component of the control system that relates the error between
the desired value and the manipulated variable.
Cycling Periodic changes in a process variable. Also known as oscillation.

© The Editor(s) (if applicable) and The Author(s), under exclusive license 489
to Springer Nature Switzerland AG 2022
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3
490 Glossary

Deadband The largest range of values of a process variable (e.g., input variable)
within which the value does not respond to changes or cause changes to other
variables.
Disturbance An input signal that affects the controlled variable but is not directly
manipulated. Sometimes also called load change.
Fail Safe Fail-safe allows a system to return to a safe state after a breakdown of
the control. The fail-safe design allows the process to avoid hazardous conditions
that may otherwise occur.
Failure The termination of the ability of an item to perform a required function.
Fault The state of an item characterized by the inability to perform a required
function, excluding the inability during preventive maintenance or other planned
actions, or due to lack of external resources.
Final Control Element The control element that directly changes the manipulated
variable, such as a valve or motor.
HAZOP/FMEA A Hazard and Operability (HAZOP) study is a structured and
systematic analysis technique for examining a system to identify potential hazards
in the system and potential operability problems with the system and, in particular,
the causes of operational disturbances and production deviations likely to lead to
nonconforming products.
FMEA is a systematic procedure for analyzing the potential failure modes, their
causes, and effects on system performance.
Highway Addressable Remote Transducer (HART) is a standard for sending
and receiving digital information across analog wires between smart devices and
a control or monitoring system.
Instrument Range The region in which the instrument is designed to operate.
Jockey Pump Low capacity pressure maintenance pump working on the principle
of on/off control.
Life Cycle The time interval that commences with the initiation of the concept
and terminates with the disposal of the asset.
Life Cycle Cost The total cost of ownership of an item considering all the costs
of acquisition, personnel training, operation, maintenance, modification, and dis-
posal.
Linear A relationship showing output proportional to input and follows the prin-
ciple of superposition.
Linearize To substitute a nonlinear function with an approximate linear function.
Linear Motion Valve A valve with a sliding stem that pushes the obturator closed
or pulls it open. They are capable of being adapted to bellows sealed. See also
rising stem.
Load or Load Variable see disturbance.
Manipulated Variable Process input that is adjusted to maintain the controlled
parameter at the setpoint.
Manned Operations Operation of a Production Plant that requires constant
“human” supervision on location to manage the process and take immediate inter-
vention action during process upsets.
Glossary 491

Maximum Operating Pressure (MOP) The maximum pressure allowed in a

device or system. The design pressure (DP) is 110% of the MOP.
Net Positive Suction Head (NPSH) To ensure that the suction conditions for a
pump are such that no vaporization occurs in the pump inlet, there has to be
sufficient suction pressure above the vapor pressure of the liquid. This surplus
pressure is called NPSH available.
Nonlinear A relationship that does not conform to linearity;
Offset The steady-state deviation between the controlled variable and its setpoint.
Overshoot The difference between the final steady-state value and the value of
the first maximum (or minimum if the response is downward) in a step response
of an under-damped system, typically expressed as a fraction;
Pump Priming A pump should be primed to meet the NPSH requirements before
the pump can be started. This is typically done by keeping the pump elevation
below the lowest level of the liquid supply so that the pump remains primed at all
times and the NPSH requirements are met.
Quarter Turn Valve A valve (such as ball, plug, butterfly) that requires a 90-
degree operation of the stem to move from the fully open to the fully closed
position. They will readily accept any type of actuation (pneumatic, hydraulic,
electric).
Rangeability The ratio of maximum value to the minimum controllable value in
a final control element.
Redundancy The existence of more than one means of performing the same
intended function.
Reid Vapor Pressure Reid Vapor Pressure is the pressure exerted by a liquid to
calculate the volatile nature of gasoline. RVP is defined as the vapor pressure of
a substance at 100 ◦ F (37.8 ◦ C).
Reliability The ability of an item to perform a required function under given
condition for a given time interval.
Resistance The potential required to produce change such as against a back-
pressure.
Restriction Orifice (RO) A restriction orifice limits the flow through the pipe to
a set flow by choking the flow at its sonic velocity.
On the other hand, a flow orifice is used to determine the flow through the pipe
from the pressure drop over the plate. There a normally two pressure taps (one
on each side of the plate) and a pressure transmitter that determines the pressure
drop over the plate. A formula is then used to convert this pressure drop to a flow
rate for the specific fluid passing through the orifice.
Risk A combination of probability, or frequency, of occurrence of a defined hazard
and the magnitude of the consequences of the occurrence.
Safety Instrument System A function comprising one or more initiators, a logic
solver, and one or more final control elements whose purpose is to prevent or
mitigate hazardous situations
Self-Regulation The inherent characteristic of a system that produces a steady
state after a disturbance.
492 Glossary

Setpoint Value at which the controlled parameter is to be maintained by the control

system.
Settling Time The time required for the absolute value of the difference between
the output of the controlled variable and its final steady-state value to become and
remain less than a specified amount.
Specific Heat Specific heat is the amount of energy required to increase the tem-
perature of one unit of mass of a material by one degree.
Stable System A system whose response to a bounded input is also bounded.
Steady-State The final condition after the transient response having died out fol-
lowing all properties becoming constant with time.
Steady-State Error A control error at steady state.
Step Response The response of a process variable in response to a step change in
another process variable.
Throttling The intentional restriction of flow by partially closing or opening a
valve.
Time Constant The time required for the output of a first-order system to change
63.2% of the steady-state value.
Time to Steady State Same as Settling Time.
Transducer Any device that transmits, amplifies, or changes a signal.
Transfer Function A mathematical relationship that describes the ratio of an out-
put of a system to the input.
Two-Position Control See on/off control.
True Vapor Pressure True vapor pressure is the equilibrium partial pressure
exerted by a volatile organic liquid as a function of temperature as determined by
the test method ASTM D 2879. The true vapor pressure (TVP) at 100 ◦ F differs
slightly from the Reid vapor pressure (RVP) (per definition also at 100 ◦ F), as it
excludes dissolved fixed gases such as air.
Unattended Operations Operation of a plant in which personnel are absent from
the plant for short periods but operations are continuously monitored 24 h a day,
e.g., during the day shift, personnel are available for the control room and the
plant. During the night shift, a limited crew is available for plant intervention
operations that have been subjected to a risk-based assessment.
Unmanned Operations Operation of a purpose-designed plant that allows for
periods of operations in which there are no “human” supervision on location
and no immediate intervention to process upset while maintaining the required
design level of production availability. Periods of unmanned operations will be
followed by manned periods in which the planned maintenance routines and cor-
rective/breakdown work will be executed.
Valve Positioner A device that precisely controls the control valve stem position
by adjusting the instrument air pressure to the control valve.
Valve Seat The part of the valve that the valve plug rests upon when the valve is
fully closed.
Valve Stem A connecting rod between the diaphragm in a valve actuator and the
valve plug that allows air pressure on the diaphragm to control the valve plug
position in the valve and hence flow through the valve.
Index

A Batch control, 222

Abnormal conditions, 6, 13, 50, 86 Batch operation, 222
Abnormal situations, 254, 264, 352 Best efficiency point
Adaptive control, 51, 56 pump, 394
Advanced process control, 15, 16, 43, 52, Best practice, 322
172, 262 Big data, 180
Affinity law Black box, 143
compressor, 413 Block diagram, 26
pump, 393 Bumpless transfer, 96, 108, 141, 148, 161,
Alarm rationalization, 13, 259 162, 270
alarm management, 58 Bump test, see plant test
alarm suppression, 59
Alarms and alerts, 13
Ambient temperature, 208 C
Analog, 92 Capacity control, 220, 231, 255, 256, 383,
Analytics 387, 395, 421, 428, 474
descriptive, 175, 180 Capital expenditure, 5
predictive, 175, 180 Cascade control, 46, 116, 119, 161, 469
Anti-reset windup, 108, 156, 157, 270, 404 inner loop, 47
Anti-surge control, 236, 257, 422, 428 master loop, 47
Anti-surge trip, 423 outer loop, 47
Artificial intelligence, 57, 180 primary loop, 47
Automatic control, 27 secondary loop, 47
Automation pyramid, 16 slave loop, 47
Cause and effect analysis, 231
Cause and effect relationship, 11, 47, 53, 94,
B 132, 173, 233, 261, 266, 295, 317,
Back-calculation, 96, 108, 146 468
BKCALI, 144 Cause and effect table, 208, 242, 418, 435,
BKCALO, 144 442, 468
Back-end logic, 87, 151 Central control room, 17
Back-initialization, 96, 101, 163 Challenges
Bad actors, 301 abnormal situations, 264
Bad data, 92, 151 adaptive control, 56
Baselayer control, 16, 43, 198, 206, 262 analyzer issues, 255

© The Editor(s) (if applicable) and The Author(s), under exclusive license 493
to Springer Nature Switzerland AG 2022
S. S. Niu and D. Xiao, Process Control, Advances in Industrial Control,
https://doi.org/10.1007/978-3-030-97067-3
494 Index

automated startup, 251 Compressor surge, 222, 260

data flow tracking, 100 Compressor variables
data health, 96 flow, 413
data quality, 56, 350, 352 head, 413
disturbances, 30 speed, 413
documentation, 92 Configuration, 102
model predictive control, 56 Constrained control, 125, 194
mode transition, 222 Constraint, 50, 52, 194
MPC weighting on model quality, 193 hard limit, 194
nonlinearity, 171 soft limit, 194
online analyzer, 55, 351 Control algorithm, 28, 40, 262
operator training simulator, 58 Control design, 246, 277
performance monitoring, 57 control overview diagram, 265
process, 170 Control element, 90
quality control, 255 final, 90
real-time optimization, 57 primary, 90
Characterization, 337 secondary, 90
Choke Control handle, 25, 118
compressor, 410 Controller mode, 88
Choked flow, see critical flow automatic mode, 86
ejector, 385 cascade mode, 86
Closed-loop control, 26 criticality, 86
Comfort zone, 202, 280 higher mode, 86
Common source failures, 423 lower mode, 86
Complex PID control, 16, 43, 119, 143 manual mode, 85
cascade control, 46, 116, 119, 161 priority, 86
decoupling control, 122, 171 remote cascade mode, 86
dual-controller control, 45, 119, 142 remote-manual mode, 86
fan-out control, 45, 117, 161 Controller performance index, 302
feedforward control, 48, 119, 120 Controller up-time, 301
flow balancing control, 128 Control loop, 25
inferential control, 117 data flow, 91, 117
nested loop, 137 process flow, 91
override control, 50, 116, 118 simple, 25
protective control, 116 standalone, 25
ratio control, 49, 118, 122 Control loop integrity, 90
selective control, 51, 119 Control loop stability, 90
split-range control, 43, 116, 119, 138, Control objectives, 4, 28, 37
161 Control overview diagram, 261, 263, 265,
standard complex loops, 115 464
valve position control, 129 Control performance monitoring, 57, 88,
Compressor 201, 277, 278, 280, 301
anti-surge control, 93 Control philosophy, 59
fixed speed, 411 Control strategy, 11, 191
variable speed, 411 Control structure, 260
Compressor characteristics Control systems
performance curve, 413 distributed control system, 143
Compressor control, 116, 220, 363 Foundation Fieldbus control system, 143
Compressor flow, 411 supervisory control and data acquisition,
Compressor map, 414 143
Compressor performance Control systems engineering, 3, 9
head flow relationship, 413 Control target, 25, 118
system resistance curve, 413 Control valve
Index 495

butterfly valve, 34 two element, 445

globe valve, 34 Dual-controller control, 45, 119, 142, 439
leak, 453 Dump control, see on/off control
leaking, 36 Dump valve
passing, 36, 453 in test separator, 463
pneumatic actuator, 34 Dynamic response, 11
Crippled-mode operation, 258, 351 Dynamic simulation, 58, 107
Critical condition, 385
Critical flow, 385
Cross correlation, 297 E
Cruise control, 27 Educer, see ejector
adaptive, 50 Ejector, 381
fixed capacity, 386
motive gas, 382
variable capacity, 386
D
Emergency response, 14
Data flow, 91, 117, 151
Emergency shutdown, 14
bad data, 92, 94
Energy balance, 225
complex PID control, 151
Enterprise resource planning, 17
constrained, 93
Execution frequency, 106
disconnected, 92, 94
Exploration and production, 220
drifting dynamics, 94
health monitoring, 96
limited, 94, 106 F
tracking, 95 Fan laws, see affinity laws
Data quality, 188, 316, 350, 351 Fan-out control, 45, 117, 161
bad data, 351, 352 Fault detection, 53
bad measurement, 352 Feedback control, 26, 28
poor measurement, 352 components, 28
DCS schematics, 220 Feedforward control, 48, 119, 120
Deadband additive, 122
control valve, 37, 292 multiplicative, 122
on/off control, 40 Fieldbus, 30, 92, 342
Dead time, 30, 121 Filtering, 344
Decoupling control, 122, 171 Final control element, 11, 28, 34, 47, 84, 90
Degree of freedom, 118, 234, 417, 435 Fire and gas system, 14, 17
Demand-driven, 225 Fire water
boiler, 442 grid main, 460
control, 227 ring main, 460
ejector, 383 First-Order-Plus-Delay-Time (FOPDT),
firewater, 461 177, 239, 320, 469
Differential equation, 67 First principle, 61
Digital, 92 Flow balancing control, 128
Distillation column, 376 Flow coefficient, 35
Distributed control system, 16, 17 Flow compensation, 354, 355
Disturbance Flow measurement
load change, 29 cutoff threshold, 33
Disturbance rejection, 238 Flow meter, 32
Disturbance variable, 232 cutoff, 34
Documentation, 241 linear type, 33
DOF requirement, 260, 418 pressure loss, 33
Drum level control square-root type, 33
single element, 445 Forward flow, see process flow
three element, 446 Functional requirements, 4
496 Index

Function blocks, 90, 143, 144 Internal model control, 319, 320
arithmetic, 148 Internal reflux, 377
control functions, 144 Invariant coordinate, 426
control selector, 146 Inverse response, 443
control splitter, 147, 150 I/O block
gain, 148 analog input, 92
lead-lag, 148 analog output, 92
manual loader, 147 I/P transducer, 350
model predictive control, 148
non-control functions, 144
PID controller, 146 J
switch, 147 Jockey pump, 461
time delay, 149
Fuzzy logic model, 173
K
Key control loops, 266
G Key control points, 250
Gain Key Performance Indicator (KPI), 57, 87,
controller gain, 76 173
normalized gain, 76 Key process variables, 5, 231, 234, 250, 266,
process gain, 76 281, 467
Gain scheduling, 46, 72, 81, 133, 146
gap control, 134
quadratic gain, 134 L
Goal zero, 39 Lab sampling, 54
Goodness of fit, 193 Last good value, 111, 354
Level compensation, 372
Level–flow profile, 227, 246, 256
H Level measurement
HAZOP review, 247, 250, 352 density error, 373
Health Safety and Environment (HSE), 5, 39 Level of automation, 60
Historian database, 206 Limit pushing, 133
Human error, 279 Limits and constraints, 263
Human-machine interface, 17, 85 Linear programming, 196
Human safety, 13 Linear transformation, 343
Live zero, 92, 342, 350
Load disturbance, 73
I Loop dynamics, 90, 93
Inferential property, 32, 117, 119, 179, 180, Loss of primary containment, 14
256, 377
dynamic, 184
neural network, 188 M
static, 181 Management of change, 62
Inferential property estimate, 180, 255 Manipulated variable, 232
Inferential property estimator, 54, 173, 184, Mass balance
207, 209, 287 boiler, 442
Innovation value, 212 Material balance, 6, 225, 246
Instrument air, 14, 350 Measurement, 11, 28, 30
Instrumented safeguarding, 21, 22, 258 flow, 32
Instrument symbol, 19 Mechanical protection, 14
Integral windup, 154, see also reset windup Middle of three voting, 353
Interface level Mode change, 258
oil and water, 451 Model-based applications, 173
water and steam, 447 Model-based tuning, 314
Index 497

Model predictive control, 15, 16, 43, 55, 143, Non-minimum phase, 240
188, 262, 280, 351, 470 Normal operation, 21
architecture, 55 Normal regulatory control, 395, 421, 444,
commissioning, 212 463
constrained control, 194 Nuisance alarms, 13, 58
control design, 210 Nuisance trip, 259
documentation, 213
dynamic matrix control, 188
generalized predictive control, 188 O
modeling, 210 Online analyzer, 31, 54, 255, 354
multivariable control, 189 gas chromatograph (CG), 32, 71
optimizing control, 196 TVP, 286
plant test, 209 On/off control, 40
post implementation review, 214 Open-loop control, 26
predictive control, 193 Open Platform Commication (OPC), 352
pre-testing, 207 Operability analysis, 250
work process, 204 Operating envelope, 13, 38, 256, 258, 390
Model quality, 316 compressor, 416
Model type, 174 Operating expenditure, 5
Mode of operation, 222 Operating limits
crippled mode, 111, 222 alarm, 21
crippled operation, 222 setpoint, 21
normal operation, 222 trip, 21
shutdown mode, 112, 222 Operating objectives, 37, 219, 220
startup mode, 113, 222 Operating point, 13, 37, 258, 390
turndown mode, 222 best efficiency point, 394
turndown operation, 222 most profitable, 13
MPC benefits, 206 Operator control, 11, 27
Operator intervention, 13
baseline period, 206, 207
Operator training simulator, 58, 173, 202
expectations, 206
Optimizing control, 127, 196
MPC commissioning
Order of execution, 106
online closed-loop, 212
Overall control strategy, 246, 265, 281, 283,
online open-loop, 212
294
MPC watchdog, 206
Override control, 50, 76, 116, 118
Multi-input multi-output, 143, 233
protective, 50
Multi-phase flow meter, 30, 31
selective, 51
Multivariable control, 118
model predictive control, 189
P
Percentage in range, 343
N Performance curve
Naming convention, 18 compressor, 417
control variables, 27 pump, 392
tag names, 18 Performance monitoring, 201
Near misses, 39 Personal protection equipment, 13
Negative feedback, 104 PID algorithm, 72
Nested loop, 137 dimensionless, 75
Neural network, 188, 203 integral, 88
Noises and disturbances, 26, 29, 197, 224, proportional band, 87
291 reset, 88
Nonlinear control, 51 PID control, 16, 41, 169
Nonlinear gain, 203 closed-loop performance, 306
Nonlinear transformation, 180, 295, 337 constraint handling, 125
498 Index

disturbance rejection, 73 positional form, 80

history, 67 serial form, 72
limitations, 52, 262 standard form, 72
limit pushing, 132 variations, 72, 74
mode change, 109 velocity form, 80
multivariable control, 118 PID tunings, 72, 270
optimizing control, 127 auto tuning, 322
self-tuning, 56 cascade control, 47
stability analysis, 153 Cohen–Coon method, 320
uptime, 87 default tunings, 107, 309
PID control action dual-control control, 45
derivative action, 42, 69, 75 fan-out control, 46
incremental change, 78 flow loops, 324
integral action, 41, 69, 75 IMC tuning, 320
proportional action, 41, 69 initial tunings, 107, 308
PID controller, 69, 146 lambda-tuning rules, 320
configuration, 102 level loops, 329
control error, 72 model-based tuning, 314
controller modes, 88 quarter wave decay, 307
crippled mode, 111 self-tuning, 322
integral windup, 154 split-range control, 44
mode, 84, 153 trial and error, 309
output, 84 tuning rules, 319
process value, 72, 84, 341 typical tunings, 308
process valve, 428 Ziegler–Nichols method, 313
reset windup, 109, 154, 156 Piping and Instrumentation Diagram
setpoint, 84 (P&ID), 91, 220, 265, 271
transfer bump, 101, 109, 161 Planning and scheduling, 17
PID implementation, 79, 277 Plant automation system, 17
anti-reset windup, 108, 157, 270 Plant startup, 58
back-calculation, 96, 146, 155 Plant test, 53
back-initialization, 96, 101, 163 Plant-wide control, 247, 472
bumpless transfer, 96, 108, 148, 161, Prediction horizon, 194
162, 270 Predictive analytic, 54
digital, 79 Predictive control, 193
direct of action, 104 Preferential control, 127
discretization, 79 Pressure-compensated temperature, 181,
face-plate, 85 370
function block, 82 Pressure-flow profile, 227, 246, 256, 412
initialization, 101 Pressure loss
initial tunings, 270 flow meter, 33
PV filtering, 106 valve, 37
PV tracking, 107 Primary element, see primary control ele-
setpoint tracking, 73 ment
tag name, 84 Process, 3, 28
tracking and initialization, 144 MIMO, 42
variations, 87 operable, 10
PID structure, 70, 72 operative, 10
equation type, 73, 153 single-input single-output, 67
ideal form, 72 Process analysis
interacting form, 72 cause and effect, 233
non-interacting form, 72 degree of freedom, 234
parallel form, 72 supply and demand model, 225
Index 499

Process analytics, 175 white-box model, 176

Process control, 3, 13 Process overview diagram, 5, 91, 220, 223,
automatic control, 6 242, 263, 265, 424, 472
capacity control, 255 Process safety, 12
documentation, 241 Process value, 91
manual control, 6 Product specification, 221
normal regulatory control, 255 Programmable logic control, 17
objectives, 219 Proportional band, 87
operator control, 6 Protective control, 11, 51, 116, 125, 396, 422
plant-wide control, 247 Pump capacity, 392
quality control, 255 Pump control
sequential control, 258 end-of-curve, 396
unit control, 251 minimum continuous stable flow, 390
un-manned operation, 6 minimum flow control, 125, 396
Process control deliverable, 5 surge prevention, 396
Process control narrative, 5, 62, 205, 220, Pump curve, 392
263 Pump head, 392
Process design, 6, 13 Pump priming, 407
Process dynamics, 29, 93, 238, 295 PV filtering, 106, 163
first-order dynamics, 239 PV tracking, 107, 109
higher-order dynamics, 239
integral mode, 240
inverse response, 240 Q
nonlinear, 170, 240 Quadratic programming, 194
Quality control, 220, 231, 246, 255
non-minimum phase, 240
Quarter wave decay, 307
process gain, 238
time constant, 239
time delay, 238
R
time-varying, 240
Ratio control, 49, 118, 122
transient behavior, 239 wild stream, 49
unstable, 240 Real-time optimization, 15, 16, 57, 174
Process engineering, 3, 4, 8 Recycle flow
Process flow, 91 compressor, 411
compressor, 411 Regulatory control, 11, 21, 255
Process Flow Diagram (PFD), 91, 220, 223, Relative gain-analysis, 235
225, 265 Reset windup, 51, 79, 109, 154
Process hierarchy, 246 Residence time, 331
Process identification, 54, 176, 317 Root cause analysis, 302
fitting criterion, 176 Rule of five’s, 323
parameter estimation, 177 cascade loop tuning, 47, 324
process data, 176 energy balance vs. material balance, 324
process model, 176 PID derivative action, 324
structure determination, 177 positioner, 324
Process model, 53, 55, 173 PV filter, 107, 324, 345
black-box model, 176 step size, 317, 323
dynamic model, 175 tuning change, 321
finite impulse response, 177
finite step response, 177
first-order-plus-time-delay, 177 S
model parameters, 175, 191 Safeguarding system, 13
model structure, 175, 190 Safety envelope, 259
state-space, 177 Safety instrument system, 14, 17, 20, 423
static model, 174 Safety onion, 12
500 Index

Sampling time, 79 State space model, 68

Scan time, see execution frequency Steam table, 373, 375
Scatter plot, 294, 297 Step change, 29
Secondary data flow, see back-calculation Step response, 29
Secondary element, 90 Step test, 53, 314
Secure instrument air, 466 Stonewall
Selective control, 51, 119 compressor, 410
Self-tuning PID, 56, 88, 203 Supply and demand
Sensitive tray, 290 demand driven, 474
Sensors and transmitters, 11, 16 imbalance, 474
Sequential control, 26, 258, 449 production driven, 474
well testing, 450 supply driven, 474
Setpoint, 84 swing stream, 474
Shed, 199 Supply and demand analysis, 248
Shrink and swell, 442 Supply and demand balance, 230
Shutdown Supply and demand model, 11, 225, 248,
emergency, 112 260, 261, 266, 391
Signal demand-pulling, 226
analog, 349 demand-driven, 226
conversion, 341 demand driven with supply override, 229
digital, 349 mixed model, 226
electric current, 342 reversible, 226
pneumatic, 350 supply-driven, 226
scaling, 343 supply driven with demand override, 229
transformation, 341 supply-pushing, 226
Signal characterization, 343, 348 Supply chain management, 17
Signal transformation, 181, 342 Supply-driven, 225, 452, 467, 469
square-rooting, 347 ejector, 383
Signal transmission, 30, 342 Supply-driven control, 227, 248
Simulation, 53 Surge
Single element control, 445 compressor, 410
Single Input Single Output (SISO), 42, 233 pump, 390
Smart device, 83, 96, 100 Surge indicator, 119
Smith predictor, 171 Surge line, 399
Soft sensor, 32, 54, 173, 354 Surge margin, 181
Split-range control, 43, 116, 119, 127, 138, Surge test, 419
161, 437 Swing capacity, 230
split point, 44, 140, 437 Swing stream, 225
Standard complex loops, 115 System identification, 176
Standard operating procedure, 13 System resistance curve
Standards pump, 392
API-551, 350
ASME BPV, 372
EEMUA-191, 58 T
IEC-61804, 144 Tank blanketing, 434
ISA-88, 58, 246 Tank blanketing control, 433
ISA-S5.1, 18 Tank padding, see tank blanketing
ISA-S5.3, 18 Temperature compensation, 365
ISO-3511, 18 Three element control, 446
ISO-5208, 36 Time constant, 121
NAMUR-NE-43, 350 Time delay, 30
Open Platform Communication (OPC), Time to steady state, 207
352 Tracking and initialization, 144
Index 501

Transfer bump, 101, 161 check valve, 34

Transfer function, 68, 82 non-return valve, 34
Transient behavior, 239 on/off valve, 28
Trial and error, 309 relief valve, 14
True vapor pressure, 54, 286 solenoid valve, 35
Tuning rules, 319 throttling valve, 28
Two element control, 445 Valve sizing, 36, 293
Vena contracta, 182
Vessel volume
U conical vessel, 484
Unit operation, 3, 246 horizontal cylinder, 485
Unmanned operation, 111 vertical cylinder, 483
Uptime, 87 Visual communication, 266
Visualization, 266

V
Valve characteristics, 35 W
inherent, 36 What-if analysis, 250
installed, 36 White box, 143
manufacturer’s, 36 Work process, 204
nonlinearity, 36 control loop troubleshooting, 277
Valve issues control overview diagram, 271
passing, 107 model predictive control, 204
Valve position PID tuning, 323
air-to-close, 103 troubleshooting, 277
air-to-open, 103
fail-close, 36, 103
fail-locked, 36 X
fail-open, 36, 103 X-y plot, 297
fail-safe position, 36, 103, 105
read-back, 83, 100, 160, 293
Valve position control, 129, 470 Z
Valve positioner, 34, 47 Ziegler–Nichols tuning, 313
Valves Z–N tuning, 313
ball valve, 34 Z-transform, 68