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: [Link]@[Link]
or the
In-house Editor
Mr. Oliver Jackson
Springer London, 4 Crinan Street, London, N1 9XW, United Kingdom
e-mail: [Link]@[Link]
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: [Link]
publishing-ethics/14214

More information about this series at [Link]

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)
[Link]

MATLAB is a registered trademark of The MathWorks, Inc. See [Link]

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, [Link]., [Link].

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 [Link]., [Link]., 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 [Link] 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,
[Link]
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

[Link] 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.

[Link] 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
[Link]
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,
[Link]
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. [Link]
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,
[Link]
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
[Link], [Link], [Link] and [Link].
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.,
[Link] = 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 ([Link]) would typically
track the flow controller PV or setpoint ([Link]) 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,
[Link]
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 ([Link] and [Link]) 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:

[Link] = [Link]
= [Link] · [Link]. (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
([Link] 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
[Link] = (4.6)
2
TCV
[Link] = · 100%. (4.7)
FCV + TCV
124 4 Complex PID Control Loops

FCV and TCV can be easily solved to yield

FCV = 2 · [Link] · ([Link]/100) (4.8)

TCV = 2 · [Link] · ([Link]/100). (4.9)

The calculations can be interpreted as follows:

1. The normal control position is [Link] = 50% and [Link] = 50%, leading to FCV
= 50% and TCV = 50%.
2. Suppose that the temperature control needs to increase its output from [Link] =
50 to 75%. To keep the flow rate unchanged at [Link] = 50%, the two valves will
need to move in opposite directions, to FCV = 25% and TCV = 75%.
3. Similarly, assume that [Link] = 50% and [Link] = 50% and a temperature
setpoint change require that [Link] 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 [Link] 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 [Link] = 50% and [Link] = 50% and the level controller
requires a change of [Link] = 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
[Link] = (4.10)
C1 + C2
C2 TCV
[Link] = · 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
 
[Link]
FCV = [Link] · (1 + K ) · 1 − (4.13)
100
 
1 [Link]
TCV = [Link] · 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
[Link] (%) 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 ([Link]>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 + [Link] + [Link] (4.21)

where [Link] is a fixed bias term and [Link] 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
[Link] is initialized to this gap value and gradually ramps to zero:

[Link] = [Link] − 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 ([Link]) takes 22 seconds to come down and become lower than the level
controller output (i.e., to have [Link] ≤ [Link]). Only after that point (time
55) does the pressure controller output ([Link]) 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