Modeling and Control of the Paper Machine Drying Section

Slätteke, Ola

2006

Document Version:
Publisher's PDF, also known as Version of record

Link to publication

Citation for published version (APA):

Slätteke, O. (2006). Modeling and Control of the Paper Machine Drying Section. Department of Automatic
Control, Lund Institute of Technology, Lund University.

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors
and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the
legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study
or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove
access to the work immediately and investigate your claim.

LUNDUNI
VERSI
TY

PO Box117
22100L und
+4646-2220000
Modeling and Control of the
Paper Machine Drying Section

Ola Slätteke

Automatic Control
 Modeling and Control of the
Paper Machine Drying Section
 Modeling and Control of the
Paper Machine Drying Section

Ola Slätteke

Department of Automatic Control

Lund University
Lund, January 2006
 To Kristin

Department of Automatic Control

Lund University
Box 118
SE-221 00 LUND
Sweden

ISSN 0280í5316
ISRN LUTFD2/TFRT--1075--SE

” 2006 by Ola Slätteke. All rights reserved.

Printed in Sweden by Media-Tryck
Lund 2006
Abstract

The topic of this thesis is modeling and control of the last part of the paper
machine – the drying section. Paper is dried by letting it pass through a
series of steam heated cylinders and the evaporation is thus powered by
the latent heat of vaporization of the steam. The moisture in the paper is
controlled by adjusting the set point of the steam pressure controllers.
There exist several commercial incentives to focus on the performance
of the moisture control. The time to perform a grade change is often
limited by the moisture and shorter grade change time is directly
correlated to economic profit. Studies have shown that the drying section
uses Ҁ of the total energy requirement in paper making. Reduced
variations in moisture gives opportunity for target shifts (changed set
point) which reduces the amount of raw material and steam requirement.
It also creates opportunity for increased production rate.
The thesis is divided in two parts. The first part deals with the control
of the steam pressure inside the cylinders. Both a black-box model and a
physical model are given for the steam pressure process. A tuning rule for
both PI and PID control is derived and various other controller structures
are investigated. Many of the results are verified by experiments on paper
machines at different paper mills.
The second part of the thesis treats the moisture controller. The
physical model from the first part is expanded with a model for the paper.
This gives a complete simulation model for the drying section that is
implemented in the object-oriented modeling language Modelica. Two
new approaches to control the moisture by feedback are evaluated. The
first utilizes the air around the paper in combination with the drying
cylinders to improve the controller performance. The second uses only the
last part of the drying section to control the moisture, while the first part is
put at an appropriate level. Finally, feedforward of a surface temperature
signal is examined.
Acknowledgements
There are a number of people who have contributed to this thesis. First of
all I would like to thank my advisors Björn Wittenmark, Tore Hägglund,
and Krister Forsman. Our regular meetings have been very constructive
and fruitful, and this thesis would not have been possible without their
outstanding support. At the same time, I have been given a large amount
of independence in my research which is something I have appreciated.
I would also like to acknowledge some of the people at ABB; Per
Sandström, Jonas Warnqvist, Jonas Berggren, and Alf Isaksson. It has
been a great experience working with all of you.
There are many people I have come in contact with at different paper
mills during my research. I would particularly like to mention all of my
old colleagues at Stora Enso Nymölla. It has also been a pleasure getting
acquainted with Stefan Snygg at Stora Enso Hylte, Stefan Ericsson and
Lars Jonhed at AssiDomän Frövi.
I have had the opportunity to work with a few people at the
Department of Chemical Engineering in Lund; Magnus Karlsson, Stig
Stenström, Bernt Nilsson, and Erik Baggerud. Magnus really deserves an
extra salute for the work we have done together; I have learnt a lot from
him.
Much of the work on physical modeling in the last chapter was carried
out on account of a large amount of inspiration by Karl Johan Åström. It
all started as a minor discussion and ended up as a major piece of work.
Working at the Department of Automatic Control in Lund is an honor
and it is a great atmosphere to operate in. I would like to thank all my
colleagues for the years we have had together. I will miss you.
During the last year of my PhD-studies I had the privilege to work for
a month at the Pulp and Paper Centre, University of British Columbia,
Vancouver, under direction of Prof. Guy Dumont. This was an instructive
and very interesting time for me.
Finally I would like to thank ABB and the Swedish Foundation for
Strategic Research (SSF) within the project CPDC for the financial
support of the project.
Preface
I first encountered process control in the summer of 1990. I was working
as a summer intern at a pulp and paper mill at one of their winders (a
machine that slits and winds the paper from the paper machine into the
roll widths ordered by the customer). A winder does not have much
process control but one night shift I was assigned to manage a pulper (a
unit for slushing paper into pulp). I got a two minute crash course in
control theory by one of the operators. For the first time in my life I heard
words like set point and control signal. I remember that I did not
understand much of it at that time. There were two important control
loops to keep an eye on, the level control and the consistency control.
Both were controlled by single-loop controllers, manufactured by Fisher
& Porter, if I remember it correctly. A dangerous operating point was if
the consistency was too high to physically empty the pulper at the same
time as the level was too high to dilute the pulp mix. I promised the
operator to not reach that point and hoped that I was right. Luckily I
managed to do fine through the night and I was placed there the following
nights too.
The next summer I was working at the same site but this year at the
instrument department. One day we were replacing a malfunctioning flow
gauge at the pulp dryer and I was watching a level controller at the
instrument panel, trying to understand how it worked. I noticed that the
level was too low but the controller only opened the valve by 40% and it
was increasing slowly. I asked the maintenance guy who was dismounting
the flow meter, why the valve was not fully opened. I thought that was the
appropriate thing for the controller to do if the level was low. He then
explained to me the concepts of dynamics, overshoot and stability, and
from that day on I was hooked on the exciting field of process control.
During my studies I continued to work at the instrument department
each summer. I learned a lot, things that are still useful for me today,
every thing from repairing old pneumatic controllers with liquid solvent,
programming the DCS-system and understanding different control
structures. After my degree I worked there for a few years more before I
went back to the university to become a PhD student.
Contents

1. Introduction ........................................................................................ 13
1.1 Introduction and motivation .................................................... 13
1.2 Outline and contribution of the thesis ..................................... 19
2. Fundamentals of the Paper Drying Process ..................................... 22
2.1 Cylinder configurations in the drying section ......................... 23
2.2 The steam and condensate system........................................... 25
2.3 The moisture control loop ....................................................... 29
2.4 Disturbances in the drying section .......................................... 38
2.5 A note on the choice of units................................................... 40
PART 1.
Modeling and Control of the Steam and Condensate System
3. Black-box Models and Controller Structures .................................. 45
3.1 A black-box model structure í the IPZ transfer function ....... 46
3.2 PID control of the steam pressure ........................................... 53
3.3 Improved set point response by feedforward .......................... 58
3.4 A state feedback controller...................................................... 63
3.5 A two-pole model of the steam pressure ................................. 69
3.6 The differential pressure loop ................................................. 74
3.7 Summary ................................................................................. 77
4. A Physical Model of a Steam Heated Cylinder................................ 79
4.1 The model................................................................................ 80
4.2 Time and frequency domain analysis...................................... 89
4.3 Comparisons with plant data ................................................... 91
4.4 A modified model ................................................................... 94
4.5 Summary ................................................................................. 97
5. A Tuning Method for IPZ Models .................................................... 98
5.1 A design method based on optimization ................................. 99
5.2 The IPZ tuning rule for PI control......................................... 103
5.3 The IPZ tuning rule for PID control...................................... 109
5.4 Stability regions..................................................................... 113
5.5 Industrial verification of the tuning rule................................ 115
5.6 Comparison between PI and PID control .............................. 118
5.7 Comparison to other design methods .................................... 121
5.8 Summary ............................................................................... 138

PART 2.
Modeling and Control of Paper Moisture in the Drying Section
6. Enhanced Moisture Control Using the Air System ....................... 143
6.1 A literature review of drying section models ........................ 144
6.2 The model.............................................................................. 144
6.3 A prestudy ............................................................................. 149
6.4 Mid-ranging........................................................................... 150
6.5 Moisture control by mid-ranging the air system ................... 155
6.6 Summary ............................................................................... 166
7. Feedforward from a Paper Surface Temperature Measurement 168
7.1 The peak position í the position of a dry surface ................. 169
7.2 Design of a feedforward controller ....................................... 175
7.3 Simulations............................................................................ 178
7.4 Summary ............................................................................... 182
8. Object-Oriented Modeling and Predictive Control of the Moisture
Content .............................................................................................. 183
8.1 The model.............................................................................. 184
8.2 Steady-state model validation ............................................... 194
8.3 Open loop simulations........................................................... 196
8.4 Control of moisture by mid-range MPC................................ 199
8.5 Summary ............................................................................... 207
9. Conclusions ....................................................................................... 209
9.1 Summary ............................................................................... 209
9.2 Future work ........................................................................... 211

A. Glossary ............................................................................................ 213

B. Conservation Balances for Energy in Compartmental Models ... 218
C. Solution to the One Dimensional Heat Equation .......................... 223
References ............................................................................................. 230
List of Symbols...................................................................................... 245
 Chapter 1. Introduction

1
Introduction

1.1 Introduction and motivation

Paper is used for printing and writing, for wrapping and packaging, and
for a variety of other applications ranging from kitchen towels to the
manufacture of building materials. It simply comes in an enormous
variety of qualities. Some common types of paper qualities include the
following:

x Copy paper for printers, copying machines and writing

x Newsprint
x Cardboard
x Light-weight coated paper for magazines
x Wrapping and packaging paper
x Hygienic tissue paper
x Currency paper

In modern times, paper has become a basic material, commonly found in

almost all parts of the world. Just try to imagine a day without paper in
your life. No newspaper in the morning, no tissue to clean up the coffee
you spilled out on the breakfast table. No books to read in your hammock
on a sunny day. No notepad to write your shopping list on before you go

13
Chapter 1. Introduction
to the super market. An empty mailbox each day you come home from
work. No thesis to hold in your hand right now. The world simply became
a better place to live in with the advent of paper some 2000 years ago.
The pulp and paper industry is a highly competitive and capital-
intensive market that is under increasing price pressure. The price
pressure on the finished products implies that the margins are often small
and a producer can only be profitable by manufacturing high volumes
[Duncan, 2003]. In Europe, the total production of paper in 2003 was 95
million tonnes with a turnover of €72 billion [CEPI, 2004]. Customers are
demanding lower costs, better terms of delivery, and higher product
quality. In the last decade a large number of company acquisitions and
mergers has taken place in the forest industry all over the world as an
answer to the high competition [FFIF, 2004], see Table 1.1. Compared to
other industries such as food, chemical, and pharmaceutical, the paper
industry has delivered a relatively low return on capital employed
(ROCE). As a result the forest industry companies have grown by size and
the industry has become more consolidated. The main objectives behind
the mergers and acquisitions are lower production costs, less sensitivity to
economic fluctuations, reduced transportation costs, reduced labor costs,
and other positive synergy effects. Companies have realized that it might
be cheaper (and certainly quicker) to buy production capacity rather than
building it. At the same time there is a steady overcapacity in the world,
the industry is facing increasing environmental requirements and there is
an increased competition from other industries as alternatives to fiber
products appear [Dumont, 1988]. The plastic packaging demand is e.g.
expected to have a rapid growth in coming years. Therefore the
production of paper requires constant attention on process efficiency,
increasing productivity, and lower costs.

Table 1.1 Figures illustrating the consolidation trend in Europe with less number of
companies and paper machines, and yet a higher capacity [CEPI, 2004].
1991 2001 2002 2003
Number of companies 1 042 918 901 884
Number of paper machines 2 181 1 863 1 811 1 815
Employment 362 100 288 700 285 000 279 400
Turnover (million euros) 39 263 77 028 74 235 71 866
Capacity (1000 tonnes) 72 343 100 713 103 489 104 978
Consumption (1000 tonnes) 62 140 83 306 85 674 86 186

14
 Chapter 1. Introduction
The function of a paper machine is to form the paper sheet and remove the
water from the sheet. A paper machine is divided into three main parts,
the wire section, the press section, and the drying section, see Figure 1.1.
When the stock enters the head box in the wire section, it contains roughly
1 % of fibers or less. This low viscous mix is dispensed through a long
slice onto the wire. As it travels on the wire, much of the water drains
away by gravitational forces or is pulled away by suction from
underneath. As the water disappears, the cellulose fibres start to adhere to
one another by hydrogen bonds and form a paper web. When the paper
web leaves the wire section and enters the press section, the dry solids
content is around 20 %. In the press section, the newly formed sheet is
pressed between rotating steel rolls and water is displaced into a press felt.
After a few press nips the web enters the drying section with a solid
content of approximately about 50 %. It now encounters the dryer
cylinders. These are large hollow metal cylinders, heated internally with
steam, which dry the paper as it passes them. Finally, the paper is wound
up on a big roll and removed from the paper machine. The moisture
content is now roughly 5í10 %.
Although the drying section is only responsible for removing less than
1 % of the water volume in the original stock to the head box, this is the
part of the paper machine that, by far, consumes most energy. Studies
have shown that the drying section uses around Ҁ of the total energy
requirement in paper making [Fellers and Norman, 1998]. This implies
that the drying section is the most expensive part of the paper machine in
terms of energy use per kg removed water. Moreover, the drying section
affects a lot of the important physical properties of the final product, such
as paper sheet elasticity, twist, and curl.
~ 140 m
Wire section Press section Drying
Drying section
section

99 % 80 % 50 % 5%

Figure 1.1 The principle of paper production is simple. The water is separated from the
original stock which is smoothened out to a thin and endless paper sheet. By adding
different types of fillers the paper surface obtains different properties. Typical values of
moisture content are indicated. By courtesy of Skogsindustrierna.

15
Chapter 1. Introduction
For a paper mill, and even for a group of companies, erecting a new paper
machine is a large investment. A high production rate and capacity is
therefore essential to achieve a high return on the investment. One of the
most important quality variables in paper manufacturing is moisture
content. Below are a few reasons of why a well tuned moisture control
system provides economic yield.

x Large variations in moisture can adversely affect post

processing units like calendering, the converting or packaging
line, or even the customer’s printing press (worsen
printability). During production, moisture content is therefore
measured and monitored online, and the paper product is
rejected if it deviates outside the specified limits. A stable and
uniform moisture content during normal operation guarantees
low reject and consequently high production rates.
x With reduced variance the moisture set point can be increased
without changing the probability for an off-spec product, see
Figure 1.2. In plain language, the paper mill is selling more
water at an excessive price (paper is sold according to weight).
A modern paper machine makes around 1000 tons of paper per
day. A reduction of moisture by 0.1 % corresponds to 365 tons
of raw material per year. With a production cost for pulp
roughly around €500 per ton [Dagens Industri, 2004], this in
turn means a large economical saving for the mill. An increase
in moisture also gives a reduction in energy use (steam
consumption). If the specific paper machine is dryer limited
this also gives an opportunity to increase the machine speed,
see below.
x An obvious way to increase production is to increase the
machine speed. Then the drying section often becomes a bottle
neck by lacking the required capacity. Maximum production is
achieved by operating at maximum speed while remaining
within the control constraints. Reduced variations in moisture
then implies that the speed can be increased without reaching
the maximum available steam pressure.
x A well tuned moisture control system will reduce the time to
carry out a grade change (state transition). In practice, the
moisture feedback loop is often turned off during a grade
change and the process is run in open loop (feedforward). Due

16
 Chapter 1. Introduction
to model errors in the feedforward loop the moisture will
deviate from the set point when the feedback is turned on
again. Hence, the moisture control is important in the last part
of a grade change and a shorter grade change time is directly
correlated to economic profit.
x The moisture control loop is indirectly involved during a web
break by the steam pressure in the steam cylinders. A very
common problem is that the cylinders become overheated since
there is no longer any cooling paper around them. When the
paper web is put back, picking and new web breaks easily
occurs. By having an optimized steam control system during a
web break, the time it takes to get the paper sheet back on the
reel can be reduced.
x Many paper properties depend on moisture content, e.g. curl,
stretch, tear, strength and stiffness [Gavelin, 1972].
These are some of the reasons why the drying section plays a vital role in
paper manufacturing. As the title reflects, this thesis is focused on both
modeling and control of the drying section of a paper machine.

0.8

0.6 Tolerance
limit
Probability density

0.4

0.2

0.0
5 6 7 8 9 10 11 12 13 14 15
Moisture (%)
Figure 1.2 The reduction of moisture variation makes it possible to increase the set point
(target shift). The solid curve represents a condition where the standard deviation has been
reduced by 50 % compared to the dashed curve. Hence, it is possible to increase the set
point from 10 to 11 %. The difference between the tolerance limit and mean value is
sometimes called the give-away.
17
Chapter 1. Introduction

Figure 1.3 A drawing of the first paper machine from 1808, also known as the Fourdrinier
paper machine [Clapperton, 1967]. It was invented in 1798 by Nicholas-Louis Robert,
while working for the French paper mill owned by the Didot family. His machine used a
belt of wire screen to produce a continuous web of paper. He was backed in England by
the Fourdrinier brothers, who built and sold the first paper machines. By 1810, the
Fourdrinier brothers found themselves in bankruptcy and Bryan Donkin, their engineer,
continued to improve the basic design. Soon he was successfully manufacturing a machine
that mechanized the process of making paper. A water and pulp mixture flowed across a
moving, vibrating web of woven wire cloth, forming a wet mat of interlocking fibers.
From the wire, the newly formed paper transferred to a moving web of woolen cloth (the
felt), before being dried.

18
 Chapter 1. Introduction
1.2 Outline and contribution of the thesis

Although the fundamental principle of producing paper has not changed

since the invention of the paper machine, see Figure 1.3, very much has
happened since then in terms of quality and efficiency. In those days there
did not exist much automatic control in a paper mill. The pulp and paper
industry first initiated computer applications to process control in the early
1960’s. This was in a time when major changes occurred in the area of
control theory, new concepts like state-space theory, Kalman filtering, and
optimal control were introduced. Today, a large majority of all paper
machines in the world are computer controlled. Some of the major
breakthroughs in advanced control theory have been tested first in the
pulp and paper industry, e.g. the minimum-variance controller [Åström,
1967] and the self-tuning controller [Borisson and Wittenmark, 1974], see
also [Dumont, 1986] and [Bialkowski, 2000]. Since then, a significant
amount of papers have been written on the subject of quality control in the
paper machine but there are still opportunities for further improvements
and the contributions of this thesis are

x Analysis of different controller structures for the steam

pressure loop based on a previously proposed black-box model.
x Presentation of a physical steam cylinder model with the same
structure as the black-box model. The purpose is to gain deeper
understanding in the physics behind the process.
x A new tuning method for both PI and PID controllers based on
optimization of disturbance rejection, subject to a robustness
constraint. The method has one tuning parameter that adjusts
the trade-off between performance and robustness. It is
compared to a few other design methods and tested on a real
paper machine.
x A new approach to control the moisture content in the paper
sheet by using both steam pressure in the cylinders and the
supply air to the hood as actuator signals. This control
challenge is solved by using mid-ranging of two IMC-
controllers.
x A new approach to control the moisture content in the sheet by
manipulating the last steam group independently of the others.
It is solved by a model predictive controller (MPC).

19
Chapter 1. Introduction
x Introduction of a new signal which can be used in feedforward
to improve control performance. The signal is based on the
temperature profile in the machine direction.
x Design of a model library in Modelica® including components
for a drying section, with possibility to easily build a dynamic
simulation model of a whole drying section.

Publications
The thesis is based on the following publications:

x Stenström, S., M. Karlsson, O. Slätteke, B. Wittenmark, and K.

Forsman (2002): “Productivity increase from a better
understanding of dynamic processes and control of the paper
dryer,” Preprints 7th New Available Technologies, pp. 70–73,
Stockholm, Sweden.
x Slätteke, O., K. Forsman, T. Hägglund, and B. Wittenmark (2002):
“On identification and control tuning of cylinder dryers,”
Proceedings Control Systems 2002, pp. 298í302, Stockholm,
Sweden.
x Karlsson, M., O. Slätteke, B. Wittenmark, and S. Stenström
(2003): “Evaluation of models for the steam supply system,”
Tappi Spring Technical Conference & Trade Fare, Chicago, IL.
x Slätteke, O. (2003): Steam and condensate system control in paper
making, licentiate thesis, ISRN LUTFD2/TFRT--3231--SE,
Department of Automatic Control.
x Karlsson, M., O. Slätteke, B. Wittenmark, and S. Stenström
(2005): “Reducing moisture transients in the paper machine drying
section with the mid-ranging control technique,” Nordic Pulp and
Paper Research Journal, 20(2), pp. 150í156.
x Slätteke, O., and K. J. Åström (2005): “Modeling of a steam
heated rotating cylinder í A grey-box approach,” Proceedings
American Control Conference 2005, Portland, OR.
x Karlsson, M., O. Slätteke, T. Hägglund, and S. Stenström:
“Feedforward control in the paper machine drying section,”
submitted to American Control Conference 2006.

20
 Chapter 1. Introduction
x Slätteke, O.: “Object oriented modeling and predictive control of
the moisture content in paper production,” submitted to American
Control Conference 2006.

Outline
This thesis is divided into two main parts. The first treats modeling and
control of the steam and condensate system, and the second focuses on
modeling and control of paper moisture. The different chapters are
organized as follows.

Chapter 2 gives the fundamentals of the paper drying process. Much of

the nomenclature used in other chapters is introduced here.

Chapter 3 presents a black-box model for the steam pressure process and
a few different controller structures are investigated.

Chapter 4 derives a physical model for the steam pressure process that is
compared with the black box model. It is also validated against plant data.

Chapter 5 presents a tuning rule for the steam pressure controller. It

assumes PI or PID control and is compared to other tuning rules found in
the literature.

Chapter 6 shows how the air system in the dryer hood can be combined
with the conventional steam pressure control to enhance the moisture
control performance.

Chapter 7 introduces a new feedforward signal that is based on

temperature measurements of the paper surface.

Chapter 8 presents a physical model of the drying section that is

implemented in the object-oriented modeling language Modelica. The
model is used to evaluate a new approach to manipulate the steam
pressure in the drying section to improve performance of the moisture
controller.

Chapter 9 gives a conclusion and suggests possible future work.

21
Chapter 2. Fundamentals of the Paper Drying Process

2.
2
Fundamentals of the Paper
Drying Process

The most common way to evaporate water from the paper web is to use
the latent heat of vaporization in steam. A steam-filled dryer is a cost
effective method to transfer heat into the sheet. The energy in steam has
proven to cost less than a quarter of any other available method [Pauksta,
1998]. Other advantages are low toxicity, easy of transportability, and
high heat capacity. Since most of the heat content of steam is stored as
latent heat, large quantities of heat can be transferred efficiently at a
constant temperature, which is a useful attribute in paper drying and many
other heating applications. Also, the energy can be extracted as
mechanical work through a turbine which makes many mills more or less
self-supporting in terms of electricity. For chemical pulp mills, steam is
obtained simply as a by-product in the chemical recovery process line.
The moist paper can be led around a single large steam heated
cylinder, called Yankee cylinder (mainly used for the drying of tissue) or
a large number of steam heated cast iron cylinders in series (commonly
called cans), called multi-cylinder drying. In this thesis, attention is only
given to the multi-cylinder dryer but most of the theory can also be
applied to the Yankee cylinder, see Figure 3.3. Two thorough textbooks
about paper drying are [Karlsson, 2000], and [Gavelin, 1972]. A glossary
can be found in Appendix A.

22
 Chapter 2. Fundamentals of the Paper Drying Process
2.1 Cylinder configurations in the drying section

When the steam enters the cylinder it releases its thermal energy to the
cast iron shell and condenses into water. This condensate is drawn off by
suction with a siphon and fed back to the boiler house. The steam is
typically fed to the cylinders on the backside of the machine (called the
drive side), and the condensate is evacuated on the front side (called the
operator side) or the backside. At some machines, especially on wide
machines, the condensate is removed on both sides. Effective condensate
removal is important for the heat transfer of the dryer cylinder. Therefore,
it is desirable to let some steam pass through the siphon together with the
condensate. This so-called blow-through steam ensures removal of
condensate, air, and other noncondensable gases from the cylinder.
Noncondensable gases reduce the partial pressure of the steam in the
cylinders and lower the condensation temperature at a given total
pressure. In addition, the air molecules tend to accumulate at the cylinder
surface as they can hardly diffuse fast enough against the direction of flow
of the steam and as a result the heat transfer between the steam and the
cylinder shell is reduced. The effect of air in a cylinder is therefore much
greater than would be expected from the average percentage of air in the
cylinder [Gavelin, 1972].
On slow machines (< 300í400 m/min) the condensate forms a pool at
the bottom of the cylinder. It is mainly old board machine running at these
speeds. As the speed increases, the condensate starts cascading

Figure 2.1 A two-tier configuration.

23
Chapter 2. Fundamentals of the Paper Drying Process

Figure 2.2 A single-tier configuration.

and suddenly a rim of condensate forms around the circumference. One

immediate effect of this is that less energy is required to rotate the
cylinder and the dryer load drops. If the speed is reduced, the rim will
break down again but at a much lower speed than where it was first
formed (hysteresis).
Almost every dryer in a modern paper machine has dryer bars on the
inside of the cylinder shell. They are also called turbulent bars or spoiler
bars. These provide higher and more uniform heat transfer from the steam
to the cylinder by increasing the turbulent behavior of the condensate.
[Peng, et al, 1997] show that the condensate film thickness can be
greatly reduced, and nearly eliminated, by exchanging the siphon by a
rubber scraper. The condensate is then mechanically removed from the
cylinder. To the author’s knowledge, this technique has not yet been
installed or verified on a real paper machine.
To support and transport the paper web through the drying section,
dryer fabrics are utilized. The dryer fabric is also used to press the web
onto the cylinders to provide good thermal contact between the two
surfaces. The dryer fabrics are woven with synthetic yarns and do not
absorb any water, as one might think. The water present in the web is
moving directly as vapor through the fabric into the air.
There are mainly two types of dryer arrangements today, the single-tier
design (single-felted) and the two-tier (double-felted). The two-tier
configuration, which is the older one of the two, is shown in Figure 2.1.
Here two separate fabrics are used, one is used on the top cylinders and
the other on the bottom cylinders (marked as a dashed line in the figure).
Wet paper is transferred unsupported from one dryer to the next, and this
can cause problems like wrinkles and sheet breaks. To prevent these

24
 Chapter 2. Fundamentals of the Paper Drying Process
runnability problems at higher machine speeds, the single-tier
configuration was invented (in 1975 at Stora Enso Hylte mill [Carlberg,
1989]), see Figure 2.2. Using this technique, a single fabric is supporting
the web on both the top and the bottom cylinders, as well as in the passage
between them. Since the fabric is between the web and the cylinders in the
bottom row, no significant drying occurs there. In modern machines, the
bottom row of cylinders is therefore replaced by smaller vacuum rolls to
increase the runnability even more [Asensio and Seyed-Yagoobi, 1992].

2.2 The steam and condensate system

The purpose of the steam and condensate system is to provide a sufficient

amount of steam to the dryers and to handle the condensed steam. The
cylinders in a drying section are divided in separate dryer groups,
normally between five and ten groups, see Figure 2.3. The steam pressure
in the different dryer groups can then be controlled individually to obtain
the desired pressure profile through the drying section, from the first
group to the last one. Since the steam inside the cylinder can be regarded
as saturated because of the continuous condensation at the cylinder wall,
there is a direct correlation between the steam pressure and steam
temperature and you could also talk about a temperature profile. For most
paper grades, dryer steam pressure is increased gradually for drying
Main steam header

PC Steam header
of the group
CYLINDER 1

CYLINDER 2

CYLINDER 3

CYLINDER 4

CYLINDER 5

CYLINDER 6

Condensate
pipe
Figure 2.3 Sketch of a dryer group consisting of 6 cylinders and one common pressure
gauge and controller.

25
Chapter 2. Fundamentals of the Paper Drying Process
Steam header

PC2 PC1

Dryer group B Dryer group A

PDC
To the condenser

Tank B Tank A

LC1
LC2

Figure 2.4 Part of a drying section with a cascade system. Sometimes the PDC-valve is
installed in the pipe between group A and tank A.

capacity and runnability reasons. [Perrault, 1991], [Hill, 1993], and

[Krumenacker, et al, 1997] give a good review of the steam and
condensate system, from simple troubleshooting to advanced control
schemes.
The simplest, but least energy efficient, way to supply the steam to the
different steam groups would be to let them all take steam from the header
and dump the blow-through to the condenser (a heat exchanger unit that
heats process water by the left over steam). However, the fact that the
dryer groups operate at different pressures can be utilized and this is done
in the cascade system, which is an efficient arrangement from an energy
usage perspective. The blow through steam from one dryer group at
higher pressure is reused in a group operating at lower pressure. In
Figure 2.4 we see a simple example of a system with two dryer groups.
The blow through steam from Group A and flash steam (when some of the
condensate meets the lower pressure in the condensate tank, it vaporizes
and forms new steam) from Tank A is piped to Group B, which operates
at a lower pressure. Controller PC2 then adjusts its valve, and adds some
extra make-up steam from the header, to maintain the desired pressure in
Group B. This means that there must be some minimum pressure
difference between Group A and Group B in order to get a steam flow
through the PDC-valve. This minimum pressure difference depends on
both the operating point and machine specific properties. The differential

26
 Chapter 2. Fundamentals of the Paper Drying Process
Steam header

2
1 PC
Thermo
compressor

Dryer group
1
PDC

To the condenser
Tank B Tank

LC

Figure 2.5 A steam group with a thermo compressor unit. Both the PC and PDC controller
work in split-range and the numbers indicate in which order the actuators are manipulated.
Some care must be taken since noncondensibles might be recirculated and accumulated.

pressure over Group A is preserved by controller PDC to guarantee a

satisfactory condensate evacuation. In this example, Group B has the
lowest pressure in the machine and therefore Tank B must dump its steam
to the condenser. Often this dryer group has an operating point just above
atmospheric pressure (sometimes below) and the condenser is a necessity
to obtain the required low pressure in the condensate tank.
From a control perspective, the cascade system is an inconvenience,
since it introduces additional interconnections between the different
control loops, and provides extra pathways for disturbance distribution
through the system. In general, material recycling can also severely affect
the overall dynamics and in most cases leads to positive feedback [Morud
and Skogestad, 1996]. Naturally, the energy perspective has higher
priority and the cascade system is the dominating configuration.
The disadvantage of the interconnections from the cascade
configuration can be resolved by a thermo compressor unit. A thermo
compressor is a device that uses high-pressure steam from the steam
header to compress blow through steam to a desired pressure. In this way,
the blow through steam can be recirculated to the same steam group,
making the different steam groups independent, see Figure 2.5. The PC-

27
28
system.
PC FI TI PI 1
PC
TC 1 1 2
PC PC PC
2
Condenser TC PI 2

PC PC
2 PC
1
To steam box

LC SG1 SG2 SG3 SG4 SG5 SG6

5 cyl. 7 cyl. 3 cyl. 17 cyl. 14 cyl. 12 cyl.

PDC PDC
2 PDC
PDC PDC 1
1 PDC To steam box

To condenser
2

LC T1
3.4 m3
FI
Chapter 2. Fundamentals of the Paper Drying Process

From boiler

LC T2 LC T3 LC T4 LC T5
LC 2 m3 6 m3 6 m3 3 m3
T0
3.1 m3

FI

Figure 2.6 A typical piping and instrumentation diagram of the steam and condensate
 Chapter 2. Fundamentals of the Paper Drying Process
controller primarily uses the recirculated steam from the thermo
compressor but have the possibility to also use make-up steam from the
header, if necessary. In the same manner, PDC leads the blow-through
steam to the compressor and dumps the steam to the condenser only in
exceptional cases e.g. web breaks or grades with low drying demand.
Sometimes the thermo compressor is used in cascading configurations too.
Naturally, there are many other ways to structure a steam- and
condensate system, and Figure 2.4 and Figure 2.5 are purely simplified
cases to illustrate some basic ideas. The steam and condensate system
almost always uses a case-by-case design. Figure 2.6 shows a typical, but
still somewhat simplified, P&ID of a drying section, indicating both the
complexity and the control engineering challenge.

2.3 The moisture control loop

The measuring principle

To control something, you must be able to measure or estimate it. Quality
parameters, such as basis weight, moisture, caliper, ash content, fibre
orientation, color, and brightness are measured on-line in a paper
machine. The quality control system (QCS) is divided in two separate
dimensions, the machine direction control (MD) and the cross direction
control (CD). The conventional technique is to measure the MD and CD
signals by scanning the sheet with a single sensor. The sensor is mounted
in a scanner platform, where it moves back and forth in the cross
direction, see Figure 2.7. Due to the MD movement of the paper, the
measurements form a zigzag pattern on the paper sheet, as shown in
Figure 2.8. This implies that the MD and CD variations are mixed
together by the measuring principle and the two signals must be separated

Figure 2.7 The scanner platform moves the measuring sensor back and forth across the
sheet. By courtesy of ABB Ltd.

29
Chapter 2. Fundamentals of the Paper Drying Process

~1000 m

~10 m

Figure 2.8 The path of the scanning sensor. The large arrow points out the direction of
machine speed. Notice the different length scale in the machine direction and cross
direction.
[Kastanakis and Lizr, 1991]. In [Natarajan, et al, 1988] an algorithm is
developed, which uses least squares to estimate the CD component and
Kalman-filtering for the MD component. It is then further developed in
[Dumont, et al, 1991], and [Chen, 1992]. A similar decomposing
algorithm based on Karhunen-Loeve expansion is [Rigopoulos, et al,
1997], and [Chen and Subbarayan, 1999]. [Chang, et al, 2000] proposes
an elliptic sensor trajectory by variable scanning speed or the use of two
scanners traveling in opposite direction to improve the MD/CD-
estimations.
As stated above, the ultimate objective of these measurements is
control. The primary mechanism today for the control of the moisture MD
variations is the dryer steam pressure. Other methods have been proposed,
like infrared drying [Kuang, et al, 1995], and [Seyed-Yagoobi, et al,
2001], impulse drying [Orloff and Crouse, 1999], and [Martinez, et al,
2001], and Condebelt drying [Lehtinen, 1995], and [Retulainen, 2001].
Most of these methods have been tested in lab-scale for many years but
have not yet found acceptance in industry for various reasons [Crotogino,
2001]. The exception is Condebelt who has one installation in Finland,
which has been running since 1996.
The CD profile, on the other hand, is controlled either by
remoisturizing showers, steam boxes (a device that improves the
vaporization in the paper by adding superheated steam directly onto the
sheet), or by infrared heating boxes located at intervals across the
machine’s width [Dumont, et al, 1993]. An even moisture content in the
CD is easiest to achieve if it is low, therefore it occurs that the paper is
over-dried and then remoisturized. Of course, then the gain in higher
quality has to be weighted against the cost of higher energy use. This

30
 Chapter 2. Fundamentals of the Paper Drying Process
4.40
Moisture (%)

4.35

4.30

4.25
0 500 1000 1500 2000 2500 3000 3500

80.2
Basis weight (g/m )
2

80.0

79.8

79.6

79.4
0 500 1000 1500 2000 2500 3000 3500
t (s)
Figure 2.9 Moisture content and basis weight measurements taken from a fine paper
machine. The set point for moisture in this case was 4.3% and the basis weight set point
was 80 g/m2. The 2-sigma values were 0.056% and 0.3 g/m2 respectively.
thesis focuses solely on the MD-control. More details about CD
estimation and control can be found in [Stewart, et al, 2003], [Heaven, et
al, 1994], and [Kjaer, et al, 1995].
The performance of the control system has, in the pulp and paper
industry, historically been described in “2-sigma” or two standard
deviations of the controlled variables. All produced reels of paper leaves
the paper machine together with a “reel-report” that include statistics like
“2-sigma MD”, “2-sigma CD”, and “2-sigma total” for both the moisture
Scanner

Dryer Dryer Dryer Dryer Dryer Dryer

Group 1 Group 2 Group 3 Group 4 Group 5 Group 6

Machine direction
Figure 2.10 Structure for the moisture control loop with one scanner device and six steam
groups.

31
Chapter 2. Fundamentals of the Paper Drying Process
Steam pressure

ǻp

Group 1 Group 2 Group 3 Group 4 Group 5 Group 6

Figure 2.11 Example of feasible steam pressure distribution of the drying section in
Figure 2.6 and Figure 2.10. The minimum pressure difference, ǻp, between cascade
groups depends on machine speed, siphon types, and steam and condensate pipe size. A
typical value of ǻp is 50 kPa.

and basis weight [Sell, 1995]. These are the average values for the whole
reel, but it is the short term 2-sigma values that are used to make the
decision if the product meets the quality requirements. To focus on the
variability in this way makes sense since a consistent and uniform product
is an important objective, as pointed out in Chapter 1, and the set points of
the quality variables are constant during long periods and only altered at
grade changes. For grade change control, see [Murphy and Chen, 1999],
[Kuusisto, et al, 2002], and [Viitamäki, 2004]. An example of scanner
measurements, in machine direction, during a normal run are shown in
Figure 2.9, taken from a machine producing 80 g/m2 of high quality copy
paper. At 1500 s, there is a short period of time when the measurements
are not updated, most distinct in the basis weight. This is due to the
automatic calibration of the scanner, performed at constant intervals,
when the measuring head is positioned at one of the ends in the CD. Also,
see Section 2.5 for a comment on moisture units.

The paper moisture loop

As explained previously, the moisture in the sheet is controlled by the
steam pressure in the cylinder groups. Since the drying section is divided
in separately controlled groups, this is a multi-input-single-output (MISO)
system. This means that the drying process has many degrees of freedom

32
 Chapter 2. Fundamentals of the Paper Drying Process
in terms of control. Traditionally, this has been solved by letting all steam
pressure controllers follow the same signal. The moisture controller then
manipulates the steam pressure set point of one dryer group and the others
follow that one, yielding a SISO system for the moisture controller.
Figure 2.10 shows how this can be arranged with one scanner device, also
called measuring frame, and the six dryer groups in Figure 2.6. Dryer
group 5 (called lead group) operates at the highest steam pressure and
receives the control signal from the moisture controller. The set points of
the other groups are then calculated from that value, either as a ratio or a
difference, see Figure 2.11 and Figure 2.12. The purpose of this is
twofold. Firstly, the constant relation between the pressure in the groups
gives good conditions for the function of the cascade system, and
secondly it is also important for both runnability and the quality of the
paper.
The functions f in Figure 2.12 are given by

­0  k n d 1, m n 0
°
fn k n r  mn where ® or , (2.1)
° k 1, m d 0
¯ n n

where index n refers to group number. These expressions can be used to

achieve pressure differentials between the groups as in Figure 2.11. A
combination of the two function alternatives (ratio/difference) in (2.1) is
of course possible but not common. Some machines use two scanners, one
in the middle of the drying section and one at the end, to improve the
control. The middle scanner then controls the first part of the machine and
r

f1 ( r ) f 2 (r ) f 3 (r ) f 4 (r) f 6 (r)

PC1 PC2 PC3 PC4 PC5 PC6

Figure 2.12 The set point r from the moisture controller is distributed to the steam
pressure controllers by passing it through a ratio/bias-function except for the lead group,
in this case group 5.

33
Chapter 2. Fundamentals of the Paper Drying Process

Setpoint Setpoint Steam

Moisture steam pressure pressure
Moisture
PI- Steam
IMC 6 Dryer
controller system

–1

Figure 2.13 A block diagram of the moisture control loop.

the scanner at the end of the machine controls the second part. The middle
scanner can of course also be used for feedforward control.
As indicated above, the moisture control loop is a cascade loop. Drawn
as a block diagram, it looks like in Figure 2.13. The inner loop controls
the steam pressure in the dryer groups. This is in general accomplished by
a PI- or PID controller. In the outer loop there is in general a model based
dead-time compensating controller, typically of the internal model control
(IMC) concept [Morari and Zafiriou, 1989] or based on the Dahlin type
[Dahlin, 1968] (which is a subset of IMC). The performance of these
controllers are evaluated in [Bialkowski, 1996] and [Makkonen, et al,
1995]. The IMC controls the moisture in the paper sheet, by giving set
point values to the PI-controllers in the inner loop. In Chapter 6, a mid-
ranging control structure by combining two IMC-controllers is investi-
Steam pressure
Condition of fabrics
Process air
Web tension
Leakage air
Condensate distribution
Moisture Tuning of controllers
Dry weight
Production speed Dryer section
Layer distribution
Formation
Bulk
Web temperature
Retention
Filler
Refining
Freeness
Blow through steam
Condensate flow
Exhaust air
Figure 2.14 A list of variables that affect the final moisture in the paper during the drying
process. A short description of some of the terms can be found in Appendix A.

34
 Chapter 2. Fundamentals of the Paper Drying Process
gated. In this way, two single-loop controllers form a quasi-multivariable
controller. The same structure is also implemented by a true multi-
variable controller but with different manipulated variables, see Chapter 8.
Other non-conventional moisture control schemes can also be found in
[Åström, 1967], [Brown and Millard, 1993], [Xia, et al, 1993], [Rudd and
Schweiger, 1994], [Murphy, et al, 1996], [Wang, 1996] and [Wells,
1999].
Apart from the steam pressure in the cylinders, there are a large
number of variables that determine the moisture in the paper sheet. To
indicate the complexity of the problem some of them are listed below and
given in Figure 2.14.

x Production Speed: Affects the amount of steam needed, since high

production also involves higher vaporization.
x Dry Weight: A thick sheet is more difficult to dry than a thin sheet
at the same production speed, see Figure 2.15 and Figure 2.16.
x Inlet Moisture: The moisture content of the sheet after the press
section is a disturbance variable that normally is unknown.
x Degree of Refining: This parameter naturally affects both the
freeness (measure of the drainability) and the ability to dry the
sheet.
x Broke Quotient: This is defined as the amount of broke being
blended into the pulp. The broke pulp (if dried before) can be more
easily dried than the new pulp.
x Air Dew Point: A high dew point inhibits effective evaporation.
x Dryer Fabric Condition: An old fabric can be clogged and give a
higher evaporation resistance.
x Bulk: High bulk means that the water inside the web has a longer
transport distance to the surface and ambient air.
x Retention aids: It is easier to dry the web when the retention is
high since it then contains more filler.
x Web tension: High web tension increases the heat transfer
coefficient and the drying rate.
x Leakage air: The air from the machine room is cooler that the
preheated supply air and therefore impair the drying conditions.

35
Chapter 2. Fundamentals of the Paper Drying Process
x Ply loading: In paperboard, different layers consist of different
pulps, hence different physical properties. This influences the
drying.
x Blow through steam: In case of improper amount of blow through
steam, the cylinder may be flooded. This has a large influence on
the heat transfer to the paper (and the load on the drives).
Some of these disturbances are controlled variables and can therefore be
regarded as known. This opens for possibilities of feedforward, which
often is the case for production speed and dry weight (see below). Other
variations like inlet moisture, leakage air, or amount of condensate in the
cylinders can be very difficult to measure, and can only be reduced by
feedback. Variations in dryer performance due to conditions of dryer
fabrics can be considered as constant since this is a slowly degrading
process, unless it is unevenly distributed on the fabric.

13
Moisture (%)

12

11

10

0 200 400 600 800 1000 0 200 400 600 800

142
Cond. weight (g/m )
2

141

140

139

138

137
0 200 400 600 800 1000 0 200 400 600 800
time (s)
Figure 2.15 A case study from a fluting machine. At a first inspection it was found that
there was a very large moisture variation with a period time of two minutes and the steam
control system was thoroughly examined to find the cause. Later it turned out that the
source of the disturbance was a large variation in dry weight and since the drying demand
is correlated with the amount of fibers, the moisture is also affected. The set points are
indicated with dotted lines.

36
 Chapter 2. Fundamentals of the Paper Drying Process

1.2

1.0
2-V moisture (%)

0.8

0.6

0.4
0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
2-V dry weight (g/m2)
Figure 2.16 Each ‘*’ corresponds to the mean value of a 30 minute sample from the
machine in Figure 2.15. It clearly shows the strong correlation between variations in weight
and moisture.

Steam
a)
pressure
Drying Moisture
process

Paper sheet
process Total weight
Stock
flow

Steam
b) pressure
Drying Moisture
process

Paper sheet
Stock process Dry weight
flow

Figure 2.17 There are cross-connections between both steam pressure í total weight and
stock flow í moisture (a), but one of these is easily eliminated by using dry weight as a
controlled variable (b).

37
Chapter 2. Fundamentals of the Paper Drying Process
Apart from moisture, basis weight is also measured at the reel and
controlled by using the stock flow as the manipulated variable (this is the
conventional configuration, there exist others where also the machine
speed is included). The total weight is naturally affected by the amount of
water in the web. However, this coupling is eliminated by instead
controlling the dry weight and leaving only the cross-coupling stock-
moisture behind, see Figure 2.17. Since the moisture is measured, the
amount of water is easy to deduct from the total weight measurement.
As a matter of curiosity, it can be mentioned that one of the first paper
companies to use digital computers to control one of their machines was
Billerud AB in Sweden [Åström, 1967] and [Åström, 2000a]. This was in
the middle of the 1960’s, and the system was an IBM 1710 with a CPU
running at 100 kHz and 80 kB of memory. The system had a special real
time operating system, written as a part of the installation project. All
control was done in a supervisory mode, the digital computer provided set
points to the analog system and it was based on stochastic control theory.
The history of process control in relation to the development of computers
is overviewed in [Balchen, 1999].

2.4 Disturbances in the drying section

It is important to have a good knowledge of the distribution of

disturbances in the drying section when evaluating different tuning
methods and control structures. By estimating an ARMAX model for
closed loop data, a noise model is obtained together with a process model.
Figure 2.18 illustrates an example of the power spectrum for the noise in
steam pressure and moisture, taken from a fine paper machine. In
Figure 2.19, a part of the corresponding time series is given, where the
effect of the controller is removed. It can be seen that there is an apparent
difference in frequency content between the two variables.
The steam pressure shows low frequency variations and a frequency
peak at 0.1 Hz. The cut off frequency for the moisture is a decade lower
(around 0.01 Hz) and there is small frequency peak at 0.003 Hz.
Figure 2.18 will be used in different chapters throughout the thesis when
analyzing different aspects of the control of both steam pressure and sheet
moisture.

38
 Chapter 2. Fundamentals of the Paper Drying Process
3
10

Power spectrum pressure

1
10

-1
10 -4 -3 -2 -1 0
10 10 10 10 10

0
10
Power spectrum moisture

-2
10

-4
10 -4 -3 -2 -1 0
10 10 10 10 10
Frequency (Hz)
Figure 2.18 Power spectrum for steam pressure (above) and sheet moisture (below), from a
fine paper machine. The dotted lines indicate the 95 % confidence interval.

215
Steam pressure (kPa)

210

205

200

195
0 100 200 300 400 500 600 700 800 900 1000

4.4
Sheet moisture (%)

4.3

4.2
0 100 200 300 400 500 600 700 800 900 1000
Time (s)
Figure 2.19 Time series for the steam pressure and moisture, used for spectrum estimations.

39
Chapter 2. Fundamentals of the Paper Drying Process
2.5 A note on the choice of units

Moisture
There exist two alternatives to express the amount of moisture in the
sheet. Which one is used depends on the context. Books and articles
treating control of the sheet moisture use moisture content, defined as

100mwc
w [%] (2.2)
mwc  mds

where mwc is the mass of the water content in the sheet and mds is the mass
of dry solids. This is the quantity used by most control system vendors
and also by staff at the mills.
Alternatively, the amount of moisture can also be expressed as
moisture ratio, defined as

m wc
u [kg/kg] (2.3)
m ds

Moisture ratio is often used in chemical engineering and literature on

physical modeling of paper drying. The relation between the two
quantities is

100u w
w , u (2.4)
1 u 100  w

The advantage of the moisture ratio is the linearity and it better reflects
variations of water content. A change in moisture content by 98 % í 99 %
corresponds to a change of 50 kg water, while a change in moisture
content by 8 % í 9 % corresponds to a change of 0.012 kg of water, see
also Figure 2.20.
Throughout this thesis, moisture content is used in all figures and also
in black-box models regarding sheet moisture. The reason for this choice
is that the, by the author, expected target group for thesis, is much more
familiar to that unit and different results will therefore be more prominent.
The exception for this choice is in Chapter 8 where instead moisture
ratio is used in a few physical relations, since this simplifies them slightly.

40
 Chapter 2. Fundamentals of the Paper Drying Process

80

70

60
Moisture content (%)

50

40

30

20

10

0
0 0.5 1.0 1.5 2.0 2.5 3
Moisture ratio (kg/kg)

Figure 2.20 Relation between moisture content and moisture ratio.

However, it is clearly notified in those equations and the corresponding

figures still show moisture content.

Pressure
All pressures are given in Pa. The value is always given in absolute
pressure apart from a few exceptions where it is given in gauge pressure.
It is then clearly notified. It is assumed that the absolute value is 101.325
kPa above the gauge pressure.

Temperature
Temperatures in formulas are given in K. In figures, temperatures are
shown in °C for simple interpretation.

41
Chapter 2. Fundamentals of the Paper Drying Process

42
Part Modeling and control
of the steam and
1 condensate system

Papermaking in the early 1900s, painted by Thomas M. Dietrich. It shows

the wet end of the paper machine at the Fox River Paper mill in Appleton,
Wisconsin. This painting illustrates the forerunner of the head box where
stock is sprayed onto the wire. This man is adjusting valves in order to
adjust the amount of stock sprayed. By courtesy of Fox Valley
Corporation.
 Chapter 3. Black-box Models and Controller Structures

3.
3
Black-box Models and
Controller Structures

The pulp and paper industry is a highly competitive and capital-intensive

market that is under increasing cost pressure. Customers are demanding
lower costs, better terms of delivery, and higher product quality. To meet
these requirements, much effort is spent on process modeling [Foss, et al,
1998]. The purpose of the models is varying. Some examples are (i)
improved process understanding from experiments with “what-if”
scenarios, (ii) identify the bottlenecks in a process and suggest
modifications, (iii) creating process simulators for operator training, or
(iv) improved control system design.
The word model is derived from the Latin modus, which means a
measure [Bequette, 1998]. There are different classes of models and
which class is best suited depends on the problem. Models can be divided
into first-principles versus black-box (also known as statistical or
empirical), or steady-state versus dynamical, or linear versus nonlinear,
or continuous versus discrete, or lumped versus distributed.
In this thesis, preferably continuous-time dynamical models are used.
They will be either black-box or first-principles, linear or nonlinear, and
lumped or distributed. It all depends on the context.
This chapter introduces a linear black-box model structure for the
steam pressure in a drying cylinder that is based on step response data. It
45
Chapter 3. Black-box Models and Controller Structures
will be used to analyze the closed loop system for a few different
controller structures. In the next chapter, a corresponding first-principles
model is presented.

3.1 A black-box model structure í the IPZ transfer

function

From experiments on a large number of different industrial paper

machines, producing a whole range of different paper qualities, it has
been found that a linear process model can describe the dynamics from
the steam valve to the steam pressure. This model has an integrator, one
pole, and one zero, therefore it is called the IPZ-model. This model
structure has also been suggested in [Sell, 1995], and [Nelson and
Gardner, 1996]. The IPZ-model is represented by the transfer function

1  sT1   sL
G IPZ ( s ) kv e T1 ! T2 . (3.1)
s 1  sT2 

210

200
Pressure (kPa)

190

180

170
0 20 40 60 80 100 120 140 160 180 200

85
Valve position (%)

80

75

70

65
0 20 40 60 80 100 120 140 160 180 200
Time (s)
Figure 3.1 Open loop step response, taken from a liner machine. The response can be
approximated by (3.1) with the process parameters kv=0.0027, T1=58, T2=3.1, and L=2.0.

46
 Chapter 3. Black-box Models and Controller Structures
Note that T1 is always larger than T2, typically by a factor of 5 to 50. The
transfer function can then be regarded as an integrator in series with a
lead-network. This gives a characteristic open loop step response. It is
different from most processes normally encountered in the process
industry, which are often approximated by first order systems or pure
integrators, possibly with a time delay. Figure 3.1 shows a step response
from a liner machine. It is performed on the second of totally nine
cylinder groups, which consisted of six dryer cylinders. The basis weight
is 246 g/m2 and the moisture content set point is 7.5 %.
Initially, the steam pressure makes a rapid decrease after the first step
in the valve position, and then there is a distinct breakpoint in the curve
with a significant change in pressure decrease rate. The same thing occurs
at the second step in the control signal. This is the characteristic
appearance of the IPZ-process.
[Stenström et al, 2002] indicates the physics behind this phenomenon.
When the steam valve position is increased it will increase the pressure
inside the cylinder, and consequently the saturation temperature of the
steam and the condensation rate. However, the increasing condensation

525

520
Pressure (kPa)

515

510

505

500
0 100 200 300 400 500 600

31
Valve position (%)

30

29

28

27

26
0 100 200 300 400 500 600
Time (s)
Figure 3.2 Open loop step response, taken from a board machine. The response can be
approximated by (3.1) with the process parameters kv=0.0020, T1=73, T2=21, and L=1.0.

47
Chapter 3. Black-box Models and Controller Structures
rate lags behind the increasing steam inlet flow as the condensate layer
heats up to the new steady state temperature. Therefore, there will be a
fast initial build-up in steam pressure, before the steam consumption has
reached its new value. This and many other physical properties in the
drying section will be further analyzed in Chapter 8. As will be shown
later, the integrator in (3.1) is an approximation of a real pole close to the
origin.
In Figure 3.2, the effect of the integrator in the model structure is
particularly apparent. The valve is first opened a small amount and then
equally closed. When the control signal is put back to the original
position, a new level of steady state pressure is reached. This is the same
behavior as for the level in a tank when the influent or effluent is
manipulated. In closed loop control, this implies that the controller output
always returns to the original level in steady state, in absence of
disturbances acting on the process.
To examine if the dimension of the cylinder has any impact on the
model structure, a step response has been done on a Yankee dryer. These
are mainly used for drying of tissue and have diameters up to 5.5 m

710

700
Pressure (kPa)

690

680

670

660
0 50 150 250 350 450 550

74
Valve position (%)

72
70
68
66
64
0 50 150 250 350 450 550
Time (s)

Figure 3.3 Open loop step response taken from a Yankee dryer. The model corresoinds
to (3.1) with the process parameters kv=0.0026, T1=269, T2=87, and L=2.0

48
 Chapter 3. Black-box Models and Controller Structures
[Karlsson, 2000]. This should be compared with multi-cylinder dryers,
which normally are 1.5 or 1.8 m in diameter. The result is shown in
Figure 3.3. It is not a big surprise that the response has an IPZ structure
since the fundamental physical principles must be the same as in the case
of a smaller cylinder. In Chapter 4, we will investigate how physical
properties of the steam and dimensions of the cylinder will affect the
parameters of the linear process model.

Graphical process identification from a step response

Even though there are a few commercial software tools for process
identification (most DCS vendors have their own, suited for their specific
control system), these are seldom systematized for identification of the
IPZ-process or work preferably in discrete time. Exceptions are [Ljung,
2004] and [Wallén, 2000]. Therefore a graphical identification procedure
for the IPZ model is helpful and it will now be illustrated how to obtain
the four parameters, kv, T1, T2, and L from a simple open loop step
response.
Figure 3.4 shows a step response of an IPZ-model (3.1), where L is
chosen to zero for simplicity. Start by drawing two straight lines, the
3.5

q 2(t)
2.5
(t0 ,y0)
Output (kPa)

q 1(t)
1.5

0.5

0
-20 0 20 40 60 80 100 120
Time (s)
Figure 3.4 Simulated unit step response of an IPZ-process with kv=0.01, T1=220, T2=20,
and L=0.

49
Chapter 3. Black-box Models and Controller Structures
tangent to the step response at time t = 0 (called q1(t)) and the asymptote
as time tends to infinity (called q2(t)), both marked in the figure. Suppose
that the size of the step in the control signal is u0 and that the slope of
q1(t) is k1 and the slope of q2(t) is k2. Also, suppose that the two lines
intersect at the coordinates (t0,y0). Then we have

k2 y0
kv , T1 , T2 t0 . (3.2)
u0 k2

The time delay L is obtained in a standard fashion as the time that elapses
between the time when the controller output is changed and the time at
which the response of the process output begins.
To derive the expressions in (3.2), start by denoting the step response
by y(t). To get the final slope of the step response (which also is the
steady state value of the impulse response), we use the final value
theorem

dy (t ) u0 k v 1  sT1 
k2 lim lim s 2 G ( s ) lim u kvu0 (3.3)
t of dt s o0 s s o0 1  sT2  0
By the initial value theorem, the initial derivative is

dy (t ) dy (t ) u0 kv 1  sT1  kvT1
k1 lim lim s 2G ( s ) lim u u0 .
dt t 0
t o0 dt s of s s of 1  sT2  0 T2
(3.4)

Thus the tangent of y(t) at t = 0 is

k v T1u 0
q1 ( t ) t, (3.5)
T2

since the response starts in the origin. By the inverse Laplace transform,
the step response can also be written as

y (t )  
k v u 0 T1  T2  1  e  t / T2  t ,   (3.6)

and to get the equation for the asymptote of the step response we observe
that

50
 Chapter 3. Black-box Models and Controller Structures
lim  y (t )  k v u 0 t  k v u 0 T1  T2 , (3.7)
t of

which is the value of q2(0). Consequently, we have

q 2 (t ) k v u 0 t  T1  T 2 , (3.8)

and the two lines q1(t) and q2(t) intersect at t=T2, since

q1 (T2 ) q 2 (T2 ) k vT1u 0 . (3.9)

A note on the identification procedure

When doing graphical identification of a step response of a steam
cylinder process, it has often been observed that the individual parameters
may vary between different experiments on the same cylinder process
while the products kvT1 and kvT1/T2 are more or less constant. This is
probably due to difficulties with finding the slope of the asymptote q2, see
Figure 3.4. If the effect from the low-pass filter part with time constant T2
has not vanished from the response, the final slope is hard to acquire and
the identified value will depend the length of the response.
By analyzing the sensitivity of the parameter estimation, given in
(3.2), the observation can be verified analytically. Let the two lines in
Figure 3.4 be given by

q1 (t ) k1t , q2 (t ) k 2t  y2 . (3.10)

The intersection of the two lines is the given by

y0 k1t0 , y0 k 2t0  y 2 , (3.11)

which can be rewritten as

k1 1
y0 y2 , t0 y2 . (3.12)
k1  k 2 k1  k 2

We also have

k2 k1 y2 y2
kv , T1 , T2 , (3.13)
u0 (k1  k 2 )k 2 k1  k 2

51
Chapter 3. Black-box Models and Controller Structures
and

k1 y2 kvT1 k1
E1 kvT1 , E2 , (3.14)
(k1  k 2 )u0 T2 u0

where E1 and E2 are the two expressions of interest. Assume an error in

k2, and differentiate T1 with respect to k2.

dT1 k1 y2 (k1  2k 2 )
 , (3.15)
dk 2 (k1  k 2 ) 2 k 22

and rewrite as a relative error

dT1 k1  2k 2 dk 2
 (3.16)
T1 k1  k 2 k 2

Similar calculations for the other parts of (3.13) gives

dk v dk 2 dE1 k 2 dk 2 dE2
, , 0. (3.17)
kv k2 E1 k1  k 2 k 2 dk 2

The first thing to notice is that E2 is independent of the slope k2, so it is

therefore obvious that kvT1/T2 tends to be independent of different
experiments, if there is an uncertainty in k2. Also, since, k1 is larger than
k2, the relative error of E1 is smaller than the relative errors of kv and T1.
Consequently the observation is confirmed. It is also clear from (3.5) that
kvT1/T2 is independent of line q2.
The product kvT1 can be interpreted as follows. Rewrite (3.1) as

§ kv ·  sL
G IPZ ( s ) T 0
¨  k vT1 ¸e (3.18)
2
© s ¹

The step response of this function is shown in Figure 3.5. A constant

product kvT1 is equal to a constant initial height in the response. If the
identification procedure tends to get a good estimate of the height but a
slightly altered kv, this will give a small variation in T1 too. It is not only
short durations of the step response that can cause this difficulty. As we
will see later, long durations have the same effect due to the

52
 Chapter 3. Black-box Models and Controller Structures

kv

kvT1

Figure 3.5 A unit step response of the IPZ process, if the low-pass filtering part is
removed or equivalently T2 Ł 0.
approximation done when the slow time constant is assumed to be an
integrator.

3.2 PID control of the steam pressure

It is probably fair to say that in today’s paper machine control systems,

the standard PI or PID controller is the most frequent controller element
used to regulate the steam pressure. In general, more than 95% of the
control loops in process industry are of PID type [Åström and Hägglund,
2005]. The advantage of PID control is that its function is easy to
understand, even for people without knowledge in control theory. It is
also possible to tune it, at least with an acceptable performance, without
having any process model (you need to know the direction of the process
response though). The PID controller can be found in all major DCS
systems, sometimes with more advanced features like auto-tuning, gain
scheduling, and loop performance analysis.
The PID controller, in ideal form, can be written as

kc t d
uc (t ) kc E r (t )  y (t )   ³ r (W )  y(W )dW  k T
c d J r (t )  y (t ) 
Ti 0 dt
(3.19)
where uc(t) is the control signal, y(t) is the process output, r(t) is the set
point, and kc, Ti, Td, E, and Ȗ are controller parameters. Often the set point
is not included in the derivative part (Ȗ = 0) and there is also a first order
filter on the derivative part. Written in the frequency domain using the
Laplace operator, (3.19) becomes

§ 1 · § 1 ·
C ff k c ¨¨ E   J Td s ¸¸, C c k c ¨¨1   Td s ¸¸, (3.20a)
© Ti s ¹ © Ti s ¹

53
Chapter 3. Black-box Models and Controller Structures

r
Cff (s)
d n
uc y
Cc (s) Ȉ Ȉ P(s) Ȉ

1
Figure 3.6 Block diagram of the PID-controller used in this thesis, together with the
process. r is the set point, uc the control signal, y the measurement, d a load disturbance,
and n measurement noise.

where Cff and Cc are transfer functions, defined by the block diagram in
Figure 3.6. A low-pass filter added to the derivative part gives

§ · § ·
¨ ¸ ¨ ¸
1 J T s 1 T s
C ff ( s ) kc ¨ E   d ¸, C ( s )
c k c ¨1   d ¸,
¨ Ti s Td ¸ ¨ Ti s Td ¸
¨ 1  s ¸ ¨ 1 s¸
© N ¹ © N ¹
(3.20b)

where N = 2í20 [Åström and Hägglund, 2005]. The purpose of the filter
is to attenuate the high frequency gain of the derivative part but it can
also be regarded as an inherent part of the control design, like the other
parameters in the controller [Isaksson and Graebe, 2002].

The closed loop behavior

The pressure loop is in cascade control with the moisture control, as
noted before, and therefore its set point varies more or less continuously
during normal operation. But it is not unusual that the steam pressure in
the first dryer group is run with a constant set point, given by the
operator, to avoid sheet picking, fibre rising, dusting, and other
runnability problems caused by too rapid heating of the sheet. Figure 3.7
shows a typical behavior of the pressure when there is a step in the
reference value. The step response has an overshoot, which turns out to
be caused by the integral action in the controller. To see this, let the

54
 Chapter 3. Black-box Models and Controller Structures
745

740
Pressure (kPa)

735

730

725
0 50 100 150 200 250 300

56
Valve position (%)

54

52

50

48

46
0 50 100 150 200 250 300
Time (s)
Figure 3.7 Typical closed loop step response of the pressure loop when using a PID
controller, taken from a fluting machine.

controller be given by (3.20a). Now, assume that the process has an

integrator. Write the process transfer function as

1 ˆ
P( s) P( s ), (3.21)
s

where

Pˆ (0) z 0 and lim sPˆ ( s ) 0, (3.22)

s o0

and since

E (s) R( s )  Y ( s ) 1  G ry ( s ) R(s) G re ( s ) R ( s ), (3.23)

where Gry(s) is the transfer function from set point to process output and
Gre(s) is the transfer function from set point to control error, we can write
the transfer function from set point to control error as

55
Chapter 3. Black-box Models and Controller Structures

§ 1 ·1
kc ¨¨ E   J Td s ¸¸ Pˆ ( s )
Ti s ¹s
Gre ( s ) 1  Gry ( s ) 1  ©
§ 1 ·1
1  k c ¨¨1   Td s ¸¸ Pˆ ( s )
© Ti s ¹s
§ 1 ·1 § 1 ·1
1  k c ¨¨1  Td s ¸¸ Pˆ ( s )  kc ¨¨ E   J Td s ¸¸ Pˆ ( s )
© Ti s ¹s © Ti s ¹s
§ 1 ·1
1  k c ¨¨1   Td s ¸¸ Pˆ ( s )
© Ti s ¹s
   
Ti s 2  k c Ti s  1  TiTd s 2 Pˆ ( s )  k c E Ti s  1  J TiTd s 2 Pˆ ( s )
Ti s 2  k c (Ti s  1  TiTd s 2 ) Pˆ ( s )

Ti s 2  1  E  (1  J )Td s  Pˆ ( s ) kcTi s
.
Ti s 2  k c (Ti s  1  TiTd s 2 ) Pˆ ( s )
(3.24)

Now let us study the integral of the controller error as time goes to
infinity and the set point is a step.

f
1
³ e(t )dt
0
lim s ˜
s o0 s2
Gre ( s )
(3.25)
T s  1  E  (1  J )Td s Pˆ ( s )kcTi
lim i 2
s o0 T s  k (T s  1  T T s 2 ) P ˆ ( s)
1  E Ti .
i c i i d

In the classical case, when E=1, the integral of the control error is zero.
This implies that there must be an overshoot in the set point step response
to compensate the initial positive area in the time plot. We can also
conclude that E>1 will give an overshoot in the step response since the
integral has a negative value. The range of E is normally defined as

0 d E d 1, (3.26)

56
 Chapter 3. Black-box Models and Controller Structures
1.5

1
G ry

0.5

0 -4 -3 -2 -1 0 1
10 10 10 10 10 10

1.2
1.0
0.8
G ny

0.6
0.4
0.2
0 -4 -3 -2 -1 0 1
10 10 10 10 10 10
Frequency (1/s)
Figure 3.8 Frequency response from set point to output (above) and frequency response
from noise to output (below) for the closed loop system in Figure 3.7, when governed by
a PID-controller.

and extensive simulation results have indicated that there will be an

overshoot in this case too. Note that the integral error is independent of
the controller gain, kc, in (3.25).
Figure 3.8 shows two closed loop frequency plots for the process in
Figure 3.7. Gry is the response from set point r to output y. Gny is the
response from noise n to output y (the sensitivity function) and has
implications for the robustness as will be discussed later. An interesting
observation is the resonance peak in Gry, showing how some frequencies
are amplified by the controller. This is obviously an undesired feature and
problems might occur if all steam pressure loops amplify r in the same
frequency region. In practice it has been observed that often the pressure
variations in the steam and condensate system are in that specific
frequency region, possibly because of this response amplification. We
will now show how this peak can be avoided by having controllers with
more degrees of freedom than the standard PID.

57
Chapter 3. Black-box Models and Controller Structures
3.3 Improved set point response by feedforward

Normally feedforward is introduced as a method to reduce the impact of

measurable disturbances. However, it can also be used to improve set
point responses, see [Åström and Wittenmark, 1997] and [Åström and
Hägglund, 2005]. In fact, this has already been introduced for the PID
controller in Section 3.2 with the set point weights ȕ and Ȗ. Here, we will
have a more systematic treatment for a general controller but assume that
the process is given by (3.1).
Assume that we have a controller structure as in Figure 3.9, also
known as a two-degree-of-freedom controller. The transfer function from
set point to process output is then

P(CM y  M u ) PM u  M y
G ry My  , (3.27)
1  PC 1  PC

where My is the desired response from r to y. The condition for ideal

feedforward is

Mu P 1 M y . (3.28)

If we require that Mu is realizable, My should have a time delay that is

equal to or larger than the time delay of the process and the pole excess of
My should be at least as large as the pole excess of P. In addition, My
should have the same zeros in the right half plane as P.

Mu
r
uff
e ufb y
My Ȉ C Ȉ u P
c

í1
Figure 3.9 A two-degree-of-freedom controller (2DOF), where P is the process, C the
controller, and Mu and My are feedforward filters. The advantage of this configuration is
that the servo and regulation problems are separated.

58
 Chapter 3. Black-box Models and Controller Structures
Let P be given by the IPZ transfer function

1  sT1  sL
P kv e , (3.29)
s (1  sT2 )

and choose the desired set point response to

1
My e  sL , (3.30)
1  sTcl

then

s (1  sT2 )
Mu
k v (1  sT1 )(1  sTcl )
(3.31)
T2 Tcl  T2 1 T1  T2 1
  ,
k v Tcl T1 k v Tcl (T1  Tcl ) (1  sTcl ) k v T1 (T1  Tcl ) (1  sT1 )

and uff can be generated as the output of the sum of a constant gain and
two low pass filters. This is necessary if the DCS-systems have no
component for high order filters or no possibility to implement the pure
derivative. Observe that the calculations made above are independent of
the feedback controller C. Under ideal conditions, the control signal ufb
remains constant during a set point change and the purpose of C is simply

CONTR

y MV
OUTP uc
My SP
r
Mu FF

Figure 3.10 Implementation of the feedforward structure in Figure 3.9 in a DCS system.
To let the operator manipulate the correct set point, the objects should be grouped
together and presented as one controller element in the operator station, as indicated by
the dashed line.

59
Chapter 3. Black-box Models and Controller Structures
to give good disturbance rejection and robustness. In reality, there will be
a control error during a set point change due to modeling errors which the
feedback loop takes care of by driving e to zero. Figure 3.10 shows how
the feedforward part could be implemented in an industrial control
system to make use of its anti-windup, bumpless transfer, and other
functions. Observe that normally there is a pure time delay in My, which
also should be implementable in the control system.
How should Tcl be chosen? Looking at the shape of the control signal
gives a hint. The transfer function from set point to control signal is

s (1  sT2 )
G ru Mu , (3.32)
k v (1  sT1 ) (1  sTcl )

and we have lead-network (low-pass filtered derivative) followed by a

zero-pole-network. If Tcl < T2, more lead action is added to the first lead-
network, and Tcl > T2 gives a low-pass filter. Observe that T1 is always
greater than T2.

1.2
1.0
Process output

0.8
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70 80 90 100

0.8
Control signal

0.6

0.4

0.2

0
0 10 20 30 40 50 60 70 80 90 100
Time (s)

Figure 3.11 Set point step response of the 2DOF-controller. Tcl = 5 (solid), Tcl = 10
(dashed), and Tcl = 20 (dash-dotted). The dotted line is the set point.

60
 Chapter 3. Black-box Models and Controller Structures
If r is a unit step, the initial value theorem gives

T2
u c ( 0) . (3.33)
kvTclT1

This can also be seen in the partial fraction expansion of Mu in (3.31).

Figure 3.11 shows a simulation where the process is given by kv=0.01,
T1=200, T2=10, L=2. It clearly shows how the relation between Tcl and T2
affects the shape of uc. The selection of Tcl will be a trade-off between
performance and control action, and letting Tcl • T2 is a good choice.
Since there is an integrator in the process, the control signal will
always go to zero in absence of disturbances. However, due to the
cancellation of the slow process zero in í1/T1 by (3.32), the control signal
is slowly brought back to zero long after we have reached the set point.
But the control signal must behave like this to maintain the output at the

1.2
1.0
Process output

0.8
0.6
0.4
0.2
0
0 50 100 150 200 250 300

0.8
Control signal

0.6

0.4

0.2

0
0 50 100 150 200 250 300
Time (s)
Figure 3.12 Dashed í 2DOF-controller Tcl = 10, solid í PI with aggressive tuning, and
dash-dotted í PI with robust tuning. The thin dashed curve is the set point. Note that the
2DOF-controller cancels process dynamics from r ĺ y but not from load disturbance d,
therefore there is a slight overshoot when the disturbance acts on the system.

61
Chapter 3. Black-box Models and Controller Structures

1.0
Process output

0.5

0
0 20 40 60 80 100 120 140 160 180

1.0
Control signal

0.5

0
0 20 40 60 80 100 120 140 160 180
Time (s)
Figure 3.13 Robustness analysis to modeling errors of the 2DOF-controller. Perfect
process model (solid), doubled time delay in process (dotted), and doubled process gain
(dashed). The dash-dotted line is the set point.
desired value. The physical explanation is that the energy flow from the
steam to the cylinder and paper is slowly increasing which implies that
the steam consumption is slowly increasing. To compensate for this the
controller must add extra steam to the cylinder long after the set point has
been reached. Therefore, even though the set point has been reached the
system has not reached steady-state, and the cylinder and paper
temperatures are still increasing.
Figure 3.12 shows how a well-tuned PI controller behaves without the
feedforward compared to the 2DOF-controller. What is meant by well
tuned is obviously relative. The tuning method used for the PI controller
is introduced in Chapter 5. The method has one user parameter that
determines the robustness of the loop and two different settings are used
in the figure, here denoted as aggressive and robust. The two settings are
chosen to give both a faster and slower response compared to the 2DOF-
controller. The 2DOF-controller gives a smoother performance in set
point response. Since the transfer function from r to y is a first order
system, the frequency response does not have a peak as in Figure 3.8,
which is a nice feature of the 2DOF-controller structure. It is also an
advantage that this structure is easy to implement into most commercial

62
 Chapter 3. Black-box Models and Controller Structures
DCS-systems, which is not the case for the controller structure presented
in the following section.
Finally, the sensitivity to modeling errors is investigated. Figure 3.13
shows closed loop simulations where the time delay and gain in the
process is increased by a factor two. The response in process output is no
longer equal to the desired response in presence of modeling errors since
the feedback loop is active, and the second term in (3.27) is not canceled
by Mu. However, the control system proves to handle model errors well.

3.4 A state feedback controller

As we have shown, when the steam pressure is governed by a standard

PID controller there will be an overshoot in the step response and a peak
in the frequency response of the closed loop system from r to y. In the
previous section, a feedforward filter was introduced which gives a
closed loop response without this overshoot and peak. This is also
possible to obtain by a state feedback controller as will be shown here.
The zeros of the closed loop systems can not be positioned with state
feedback but the poles can be located arbitrarily, which is sufficient to
avoid the overshoot. Since a time delay in continuous time can not be
represented by a finite number of poles and zeros, and the purpose here is
merely to show the idea, the delay is assumed to be zero. In practice, the
design should be done in the discrete time domain and the delay will then
augment the transfer function with extra poles in the origin and possibly
an extra zero.
Let the process be given by

k v T1 k
s v
sT1  1 T2 T2 b1 s  b2
G(s) k v . (3.34)
s ( sT2  1) § 1 · s ( s  a1 )
s¨¨ s  ¸¸
© T2 ¹

Assume that the model is written in observable canonical form

­ ª a1 1º ªb º
° x (t ) Ax(t )  Buc (t ) « » x(t )  « 1 » uc (t )
® ¬ 0 0¼ ¬b2 ¼ . (3.35)
°
¯ y (t ) Cx (t ) >1 0@ x(t )

63
Chapter 3. Black-box Models and Controller Structures
By definition this system is observable. It is also controllable unless T1 =
T2 which implies that we have a pure integrator instead of an IPZ process.
Let the state feedback be given by

u c (t )  Lx(t )  l r r (t ) >l1 l 2 @ x(t )  l r r (t ). (3.36)

The closed loop system then becomes

­ ª a1  b1l1 1  b1l 2 º ªb º
°° x (t ) «  b l » x(t )  l r « 1 » r (t )
® ¬ 2 1  b2 l 2 ¼ ¬b2 ¼ , (3.37)
°
°¯ y (t ) >1 0@ x(t )

which in input-output form becomes

(b1 s  b2 )l r
G ry ( s ) . (3.38)
s  a1  b1l1  b2 l 2 s  a1b2 l 2  b2 l1
2

By choosing

lr l1  a1l 2 , (3.39)

we get unit steady-state gain. If we choose to place the poles in Ȝ1 and Ȝ2,
we get the equation system

ª b1 b2 º ª l1 º ª O1  O 2  a1 º
«b ». (3.40)
¬ 2 a1b2 »¼ «¬l 2 »¼ «
¬ O1O 2 ¼

Note that we have assumed that the two states of the system are
measurable, which practically is unlikely. What is also needed then is a
Kalman filter to estimate the states, which is feasible since the system is
observable. However, since an observer will not affect the dynamics of
the closed loop system from set point r to output y, except for initial
transients in the estimation, it is adequate to consider simply the state
feedback in this analysis.

64
 Chapter 3. Black-box Models and Controller Structures
A Simulation Example
Assume that we have the system

50s  1 0.1s  0.002 b1 s  b2

G s  0.02 . (3.41)
s 10s  1 s 2  0.1s s 2  a1 s

The open loop poles are in p1 = 0 and p2 = í0.1, and the zero is in
z = í0.02. To examine a cancellation of the process zero, place one pole
at Ȝ1 = í0.2, and let the other one be at Ȝ2 = í0.04, í0.02, and í0.01 (in
three different simulations). In other words, we have one case where both
closed loop poles are faster than the process zero, one case where one
pole cancels the zero, and finally one case where one pole is slower than
the zero.
The controller parameters can then be calculated as

l1 1.25  12.5O 2 ,
l2 12.5  1125O 2 , (3.42)
lr 100O 2 .

Figure 3.14 shows the simulation results. We can see that when both
poles are faster than the process zero we get an over-shoot due to the lead
action of the zero (which is closer to the origin). When Ȝ2 = í0.02, we
cancel the zero and get a first order response. In this case, the
performance of the system is given only by the position of the other pole.
Finally, we let Ȝ2 = í0.01, and the closed loop system has one fast and
one slow pole. The slow pole is clearly visible in the step response.

Adding Integral Action

The feedback loop in Figure 3.14 has one drawback. In presence of
disturbances, there will be a steady state error in the output. By
introducing integral action, the asymptotic error becomes zero. This can
be accomplished by augmenting the system with an extra state, called x3

x 3 ry r  Cx. (3.43)

The new system can then be written as

65
Chapter 3. Black-box Models and Controller Structures

1.5

1.0
y, r

0.5

0
0 20 40 60 80 100 120 140 160 180 200

3
uc

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)
Figure 3.14 Simulation of the three different cases of state feedback. Set point (solid), Ȝ2
= í0.04 (dashed), Ȝ2 = í0.02 (dotted), and Ȝ2 = í0.01 (dash-dotted).

ª x1 º ª a1 1 0º ª x1 º ª b1 º ª0 º
« x » « 0 0 0» « x 2 »  «b2 » u c  ««0»» r.
» « » « » (3.44)
« 2» «
«¬ x 3 »¼ «¬  1 0 0»¼ «¬ x3 »¼ «¬ 0 »¼ «¬1»¼

This augmented system is also controllable from uc unless T1 = T2. By

using the standard state feedback law

uc  Lx  l 3 x3 , (3.45)

we get the following closed loop system

ª x1 º ª a1  b1l1 1  b1l 2  b1l 3 º ª x1 º ª0º

« x » « b l  b2 l 2  b2 l 3 »» «« x 2 »»  ««0»» r , (3.46)
« 2» « 2 1
«¬ x 3 »¼ «¬ 1 0 0 »¼ «¬ x3 »¼ «¬1»¼

66
 Chapter 3. Black-box Models and Controller Structures
r ki uc y
Ȉ Ȉ Process
s

íL

í1
Figure 3.15 Adding integral action to a state feedback system

which has the transfer function

 l3 (b2  b1s )
Gry ( s ) . (3.47)
s  (a1  b1l1  b2l2 ) s 2  (a1b2l2  b2l1  b1l3 ) s  b2l3
3

The new state, x3, can be added by having an additional feedback loop as
shown in Figure 3.15. The integrator gain, ki, is then given by

ki l 3 . (3.48)

A Simulation Example, Cont’d

We add an integrator to the state feedback controller and let one closed
loop pole be placed in Ȝ = í0.02 (cancellation) and a double pole in
Ȝ = í0.2 . In other words, we want the characteristic equation to be given
by

s 3  0.42 s 2  0.048s  0.0008 0, (3.49)

and get the following system of equations

­ a1  b2 l 2  b1l1 0.42
°
®a1b2 l 2  b2 l1  b1l 3 0.048 , (3.50)
°  b2 l 3 0.0008
¯

which has the solution l1 = 3, l2 = 10 and l3 = í0.4. Figure 3.16 shows the
simulation result. Since there is an extra pole in the closed loop system

67
Chapter 3. Black-box Models and Controller Structures

1.0
0.5
y, r

0
1.0
0.5
0
0 20 40 60 80 100 120 140 160 180 200

1.5
uc

0.5

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)
Figure 3.16 A simulation where a disturbance acts on the controller output for the case
when there is a cancellation of the process zero. Without integral action in the controller
(dotted í same as the dotted curve in Figure 3.14), and with integral action (dash-dotted).
Observe that the dynamics are different in the two cases since the number of closed loop
poles is different.

due to the integrator state, x3, the dynamics of the set point step response
is different from the system in (3.42).

A Remark
In Section 3.3 the overshoot in set point step response, as the standard
PID-controller introduces, is eliminated by canceling process dynamics
by feedforward. Here we have shown how the process zero can be
cancelled by one of the closed loop poles introduced by feedback. By
letting the remaining two poles (in the case of integral action in the
controller) be real, we avoid both overshoot in the step response and
resonance in the frequency response.
The practical disadvantage with state feedback is that most
commercial DCS do not have any support for this controller type. Instead,
it must be implemented by the user together with features like windup
protection, bumpless transfer, and bumpless parameter change.

68
 Chapter 3. Black-box Models and Controller Structures
3.5 A two-pole model of the steam pressure

Since the steam pressure in a cylinder group finally must reach some
steady state value, the integrator in (3.1) is more likely to be a long time
constant. This implies that the model does not capture the low frequency
behavior well and this has been one of the criticisms against it. However,
the purpose of the model is not simulation and the question is if its
complexity is sufficient for a control design usage. With an extra pole, we
get

1  sT1 
G PPZ ( s ) kp e  sL , T3 ! T1 ! T2 , (3.51)
1  sT2 1  sT3 
instead of (3.1). Here the integrator is replaced by a new pole close to the
origin, and instead we call the model structure PPZ (pole-pole-zero).
Since it in most practical cases is not desirable to let the steam pressure in
an open loop step response reach steady state, because of the influence on
the paper qualities, this slow pole can be difficult to identify graphically

280
Pressure (kPa)

260

240

220
1600 1800 2000 2200 2400 2600 2800 3000 3200

45
40
Valve position (%)

'u
f
35
30
25
20
15
1600 1800 2000 2200 2400 2600 2800 3000 3200
Time (s)
Figure 3.17 Closed loop step response. Taken from a liner machine.

69
Chapter 3. Black-box Models and Controller Structures
from a single step. It can be identified from well excited closed or open
loop data with system identification techniques for linear parametric
models, such as the prediction-error methods, though.
In Figure 3.17 a closed loop step response from a steam group in a
liner machine, consisting of 13 cylinders, is shown. We will now see that
the fact that the control signal does not return to the same steady state
value as before the step in the set point, exposes the PPZ-structure of the
process.
Start by defining the change in control signal as time goes to infinity,
and assume a step in the set point with size 'r

§ 'r ·
'uf { lim uc (t )  u0 lim¨ s ˜ Gru ( s )  u0 ¸  u0 Gru 0 'r , (3.52)
t of s o0 © s ¹

where Gru(s) is the transfer function from the set point r to the control
signal uc. Assume that the process dynamics are described by the IPZ-
model given in (3.1). Also, assume that the controller is given by (3.20a).
Then we have

§ 1 ·
kc ¨¨ E   sJTd ¸¸
C ff ( s ) © sTi ¹
Gru ( s )
1  Cc ( s )GIPZ ( s ) § 1 · sT1  1  sL
1  kv kc ¨¨1   Td s ¸¸ e (3.53)
© Ti s ¹ s sT2  1
 
kc s s 2JTiTd  sE Ti  1 sT2  1
.
2
 2

s Ti sT2  1  kv kc s TiTd  sTi  1 sT1  1 e  sL

The steady state value of the control signal is then

'uf Gru 0 'r 0, (3.54)

and we can see that the final value of the control signal must be equal to
zero. This is obviously because of the integrator in the process and it
means that the control signal in Figure 3.17 should have returned to its
original steady state value (to the level it had before the step in the set
point). If we instead let the process be of PPZ-type, see (3.51), then the
steady state value of the control signal will become

70
 Chapter 3. Black-box Models and Controller Structures
C ff 0 
'r
1  Cc 0GPPZ 0 
(3.55)
 
kc 1  sETi  s 2JTiTd 1  sT2 1  sT3 'r 'r

sTi 1  sT2 1  sT3   kc k p 1  sTi  s 2TiTd 1  sT1 e  sL  s 0
kp

and it depends of the process gain and the size of the step.
Normally, the difference between the PPZ- and the IPZ-model is only
exposed in long time series. Figure 3.18 shows an example from a paper
board machine (a group consisting of 12 cylinders) where almost two
430

420
Validation
410

400
Steam pressure (kPa)

390

380

370

360

350
Identification
340

330
60
valve position (%)

40

20

0
0 1000 2000 3000 4000 5000 6000
Time (s)
Figure 3.18 Model validation showing the difference between the IPZ (dashed) and the
PPZ (dotted) process. Measurements taken from a board machine.

71
Chapter 3. Black-box Models and Controller Structures
hours of logged data is used. Both an IPZ (dotted line) and a PPZ (dashed
line) model have been identified and their simulations are plotted together
with the measured value (solid line). The first part of the time series is
used for identification (marked in the figure) and the second part is a pure
simulation. The estimated PPZ model is

G1 ( s ) 15.5
1  291s  e  2.83 s (3.56)
1  797 s 1  74.2 s 
and the estimated IPZ model is

G2 ( s ) 0.00421
1  1896 s  e  2.83 s (3.57)
s 1  138 s 

The difference between the two models is apparent. The IPZ-model does
not capture the low frequency component of the process as well as the
PPZ-model does. By manipulating the parameters in the IPZ-model, the
graphical fit can be improved in the validation part of the figure, to the
other part’s disadvantage.
The question that we posed in the beginning of this section was if the
IPZ model is sufficient for control design purposes or do we need to use a
PPZ model instead? To answer that question we start by quoting [Ljung,
1999]: “Feedback control is both forgiving and demanding in the sense
that we can have good control even with a mediocre model, as long as it
is reliable in certain frequency ranges. Loosely speaking, the model has to
be reliable around the cross-over frequency ( | the bandwidth of the
closed loop system), and it may be bad where the closed loop sensitivity
function is small.”
Figure 3.19 shows the open-loop Bode plot for the models in equation
(3.56) and (3.57). Both in the magnitude and the phase plot, there is a
large discrepancy for low frequencies but the disparity between the
models is negligible in the high frequency region. Also around the cross-
over frequency, the two different process models have similar
appearance. This means that the IPZ-model is adequate and well suited
for control design usage. We also note that compared to the PPZ-model,
the IPZ-model in general has larger phase lag. Thus, we can, by using the
IPZ-model for design, be sure to have a stable controller even if the
process is of PPZ-type.

72
 Chapter 3. Black-box Models and Controller Structures
3
10
Magnitude (abs)

0
10

-3
10 -5 -4 -3 -2 -1 0 1
10 10 10 10 10 10 10

-20
Phase (deg)

-40

-60

-80

-100 -5 -4 -3 -2 -1 0 1
10 10 10 10 10 10 10
Frequency (rad/s)
Figure 3.19 Bode plot for the IPZ (dashed) and PPZ (dotted) models.

We have previously referred to the identification issue and pointed out

the advantage with the IPZ model. From an open loop step response it is
simple to graphically obtain the parameters of the transfer function.
Practically it is more difficult to get a good PPZ model since it has more
parameters to identify.
Let us finally regard another quotation that agrees with the
conclusions we have already made. In [Ljung, 2004] it is indicated that a
model validation as in Figure 3.18 in terms of a pure simulation, can
portray a good model in an unpromising manner. “If the model is
unstable, or has integration or very slow time constants, the levels of the
simulated and the measured output may drift apart, even for a model that
is quite good (at least for control purposes). It is then a good idea to
evaluate the model’s predicted output rather than the simulated one”.

73
Chapter 3. Black-box Models and Controller Structures
3.6 The differential pressure loop

The main objective in this thesis is to give a systematic treatment of the

moisture control loop, shown in Figure 2.13. That involves a steam
pressure process with a steam pressure controller, and a paper process
with a moisture controller. But there is a third controller loop described in
Section 2.2, the differential pressure, which here will be shortly analyzed.
It is easy to realize that there ought to be cross-connections between
the pressure and differential pressure, since they are both acting on the
same physical unit. We then have a two-input-two-output (TITO) system.
To examine this, an experiment has been performed on a fine paper
machine, see Figure 3.20. The processes are excited by step responses in
the control valves, one loop at a time, see Figure 3.21. The model is
found to be

ª 29 s  1 1.7 s º
ª yP º « 0.186 s (1.9s  1) e 0 » ª uP º
«y » « » (3.58)
¬ dP ¼ «0.0395 122 s  1 e  2 s 343s  1  2.6 s » «¬udP »¼
0.0029 e
«¬ s (2.3s  1) s (3.1s  1) »¼

Steam header

3
2 PC

Flash steam 1

Dryer group

PDC
Flash steam

To boiler house Tank

LC

Figure 3.20 A P&ID of the dryer group where the TITO experiment was performed. The
pressure controller PC primarily uses flash steam (from another group operating at higher
pressure), secondly steam from the compressor, and thirdly live steam from the header.

74
 Chapter 3. Black-box Models and Controller Structures
300

280
yP (kPa)

260

240

220
2200 2400 2600 2800 3000 3200 3400

16

14

12
u P (%)

10

6
2200 2400 2600 2800 3000 3200 3400

80

70
ydP (kPa)

60

50

40
2200 2400 2600 2800 3000 3200 3400

26
24
22
u dP (%)

20
18
16
14
2200 2400 2600 2800 3000 3200 3400
Time (s)

Figure 3.21 The step response experiment of the pressure and differential pressure loops.
The identified model is given in (3.58). Taken from a fine paper machine.

75
Chapter 3. Black-box Models and Controller Structures

340
yP (kPa)

320

300

280
2000 2200 2400 2600 2800 3000 3200 3400 3600

14

12

10
u P (%)

4
2000 2200 2400 2600 2800 3000 3200 3400 3600

90

80
ydP (kPa)

70

60

50
2000 2200 2400 2600 2800 3000 3200 3400 3600

8
u dP (%)

2
2000 2200 2400 2600 2800 3000 3200 3400 3600
Time (s)

Figure 3.22 A second step response model of a similar process as in Figure 3.20. Taken
from a fine paper machine.

76
 Chapter 3. Black-box Models and Controller Structures
where yP and uP are output and input of the steam pressure process,
respectively, and ydP and udP are output and input of the differential
pressure process, respectively. Most conspicuous of this model is that the
differential pressure valve, udP, has no influence on the pressure, yP. This
is also very clear from the figure. This is a nice property since the
pressure controller can then be tuned without regarding the differential
pressure, ydP. The differential pressure, however, is significantly affected
by the pressure valve, most easily seen in Figure 3.21. This
interconnection has to be regarded when tuning the PDC controller. The
purpose of this loop is to maintain good condensate evacuation and even
short term deviations from the set point will influence the drying effect.
However, since the PC loop has a larger effect on the drying, good
regulation of this is more important and the performance of the PDC loop
is of minor importance. Figure 3.22 shows a similar experiment as in
Figure 3.21, but on a different dryer group. This experiment gives
equivalent results.
It is not a big surprise that also the differential pressure can be
modeled as an IPZ process, just like the pressure. Both are acting on the
same system and must, to some extent, be governed by the same physical
equations. However, it should be noted that in some cases it has shown
sufficient to model it as a first order system. Deep analysis of the
differential pressure loop is not covered in this thesis, and from here on
only the moisture cascade loop will be treated

3.7 Summary

In this chapter a black-box model structure has been presented and some
controller structures have been analyzed. First, the closed loop properties
when using a PID controller are examined. It is then followed by two
different controller structures. The main purpose of this is to avoid
resonances in the frequency response, r ĺ y, as feedback with a simple
PID gives, see Figure 3.8. In Section 3.3 this is accomplished by
canceling process dynamics by feedforward in terms of a two-degree-of-
freedom controller. In Section 3.4 dynamics are instead cancelled by state
feedback. The 2DOF controller is best suited for implementation since
the existing DCS today seldom support state feedback controllers.
A different type of model for the steam pressure has also been
investigated where the integrator is replaced by a pole. The conclusion is,
however, that the integrator model is sufficient for controller tuning
purposes and therefore the natural choice.

77
Chapter 3. Black-box Models and Controller Structures
Finally, a multi-variable model for the pressure and differential pressure
loop has been given. The main conclusion of this is that the pressure loop
can be tuned without regarding the differential pressure.

78
 Chapter 4. A Physical Model of a Steam Heated Cylinder

4
4.

A Physical Model of a Steam

Heated Cylinder

In Chapter 3, an empirical black-box model structure for the steam

pressure in a dryer cylinder is presented. This class of models is adequate
for controller tuning purposes but does not tell anything about the physics
that generate the dynamic behavior. Here we will present a first principles
model, initially proposed by [Åström, 2003]. The foundation of the model
is simple mass and energy balances. The primary model is a nonlinear
differential-algebraic equation set, but by algebraic manipulations and a
linearization, it will have the same structure as the IPZ model. The steam
pressure and the cylinder shell temperature are chosen as state variables,
since both these variables are possible to measure. One of the main
purposes of this grey-box model is to gain insight into which physical
characteristics and mechanisms have key effect on the parameters in the
IPZ model. A similar approach for a drum boiler has been presented in
[Åström and Bell, 2000b]. For more on grey-box modeling, see [Allison et
al, 1997], [Bohlin and Graebe, 1995] and [Bohlin, 1994]. Two
comprehensive books on physical modeling and model analysis is
[Hangos and Cameron, 2001] and [Thomas, 1999]. A nomenclature can
be found at the end of the thesis.
The nonlinear model will be further examined in Chapter 8, where it is
expanded with dynamics for the paper web to give a complete simulation
model for a whole drying section.
79
Chapter 4. A Physical Model of a Steam Heated Cylinder
4.1 The model

Let qs (kg/s) be the mass flow rate of steam into the cylinder, qc (kg/s) be
the condensation rate, and qw (kg/s) be the siphon flow rate. Also, let Vs
(m3) and Vw (m3) be the volume of steam and water in the cylinder, and let
Us (kg/m3) and Uw (kg/m3) be the densities of steam and water. The mass
balances for water and steam are then

d
U sVs  qs  qc , (4.1a)
dt
d
U wVw  qc  qw , (4.1b)
dt

where no blow-through steam is assumed. Sometimes the blow-through

steam is modeled as a fraction of qs [Karlsson et al, 2002]. This does not
affect the dynamics of the system, only the steady state gain. The energy
balances for steam, water and metal are

d
U susVs  qs hs  qc hs , (4.2a)
dt
d
U wuwVw  qc hs  qwhw  Qm , (4.2b)
dt
d
dt

mC p , m Tm  Qm  Q p , (4.2c)

where Qm (W) is the power supplied from the water to the metal, Qp (W)
is the power supplied from the metal to the paper, hs (J/kg) is the steam
enthalpy, hw (J/kg) is the water enthalpy, m (kg) the mass of the cylinder
shell, Cp,m (J/(kg˜K)) the specific heat capacity of the shell, Tm (K) the
mean temperature of the metal, us (J/kg) and uw (J/kg) are the specific
internal energies of steam and water. For the period of one revolution, the
cylinder is in contact with the paper or dryer fabric one part of the time,
and in contact with the surrounding air the other part. Due to higher
resistance for heat transfer, the energy loss to the air is only a fraction of
the energy flow to the paper or fabric. However, [Janson and Nordgren,
1958] found that the temperature variation during one revolution is
negligible (less that 0.3°C) and the metal temperature is therefore assumed
to be independent of the cylinder rotation. The energy flow to the metal is
given by

80
 Chapter 4. A Physical Model of a Steam Heated Cylinder

Paper web

Ts

Tm

Qm Tp
Qp

Steam Condensate
Dryer shell
Figure 4.1 A piece of the cross-section of a drying cylinder, visualizing the assumption on
the temperature profile and the energy flows.

Qm D sc Acyl Ts  Tm , (4.3)

where Dsc (W/(m2˜K)) is the heat transfer coefficient from the steam-
condensate interface to the centre of the cylinder shell, Acyl (m2) is the
inner cylinder area, and Ts (K) the steam temperature. The outer surface
area of the cylinder is assumed to be equal to its inside area. The error is
negligible (less than 5%), because the thickness of the cylinder shell is
much smaller than the outer cylinder diameter. The energy flow to the
paper is given by

Qp D cp AcylK Tm  T p , (4.4)

where Dcp (W/(m2˜K)) is the heat transfer coefficient from the center of the
cylinder shell to the centre of the paper sheet, Ș (unitless) is the fraction of
dryer surface covered by the paper web, and Tp (K) is the paper
temperature. Fraction Ș is between 0.5 and 0.7.
For simplicity, all steam within the cylinder cavity is assumed to be
homogeneous with the same pressure and temperature. From (4.3) and
(4.4), we make the assumption of a temperature gradient in the condensate
layer, cylinder shell, and paper web, as illustrated in Figure 4.1.

81
Chapter 4. A Physical Model of a Steam Heated Cylinder
Equations (4.1), (4.2), and (4.3) are a crude nonlinear model for the steam
and condensate system in the cylinder cavity. To obtain a linear second-
order model, we make a few simplifications.

i. First assume that the steam in the cylinder is saturated. This

because there is a continuous condensation occurring at the
cylinder wall. This means that the state of the steam can be
characterized by one variable only and that it is sufficient to use
either the mass balance or energy balance. Therefore, we leave out
the energy balance (4.2a).
ii. When the inflow of steam is varied, the temperature of the paper,
Tp, is likely to vary slowly compared to the steam- and cylinder
dynamics, due to the low pass effect of the cylinder shell.
Therefore, we assume that Tp is constant (otherwise we would
also need an energy balance for the paper web).
iii. In addition, the thermal dynamics of the water is very fast
compared to the cylinder, so we replace it by a static model.
iv. Observing that the volumes are constrained by Vs + Vw = V, where
V is the total cylinder volume, the second mass balance in (4.1)
can be eliminated. Since the water volume is small we also have
V § Vs .

Summarizing, we find that the system can be described by the equations

d
U sV  qs  qc ,
dt
d
dt
 
mC p , mTm Qm  Q p ,
(4.5)
0 qc hs  qw hw  Qm ,
Qm D sc Acyl Ts  Tm ,
Qp D cp AcylK Tm  Tp ,

which are a mass balance for the steam, an energy balance for the metal, a
static energy balance for the water, and an algebraic equation for the
energy flow. Eliminating the variables qc and Qm, the model becomes

82
 Chapter 4. A Physical Model of a Steam Heated Cylinder
d
hs U sV  qs hs  qwhw  D sc Acyl Ts  Tm ,
dt (4.6)
d
dt

mC p , mTm  D sc Acyl Ts  Tm   D cp AcylK Tm  Tp .

Assuming that the steam in the cylinder is saturated, enthalpies hs, hw,
density Us and the temperature Ts, are all functions of the pressure p. The
model can thus be written as

dU s dp
hs ( p )V qs hs ( p )  qw ( p )hw ( p )  D sc Acyl Ts ( p )  Tm ,
dp dt (4.7)
dT
mC p ,m m D sc Acyl Ts ( p )  Tm   D cp AcylK Tm  T p ,
dt

where the states are pressure p and mean metal temperature Tm. The steam
inlet flow, qs, is the input. The equilibrium gives the relations

0 
qs0 hs ( p 0 )  qw ( p 0 )hw ( p 0 )  D sc Acyl Ts ( p 0 )  Tm0 ,
(4.8)
 
0 D sc Acyl Ts ( p 0 )  Tm0  D cp AcylK Tm0  T p0 .  
Hence

q s0 hs ( p 0 )  q w0 hw ( p 0 )
T p0 Tm0  . (4.9)
D cp AcylK

The numerator in (4.9) is the amount of energy delivered to the cylinder

shell (difference between inflow and outflow of energy). Dividing this by
the conductivity for the surface covered by paper gives the reduction in
temperature from cylinder to paper. Linearizing around the equilibrium
gives

83
Chapter 4. A Physical Model of a Steam Heated Cylinder

dU s d'p
hs ( p 0 )V
dp p p0
dt
§ 0 dhs dh dq dT ·
¨¨ qs  qw ( p 0 ) w  hw ( p 0 ) w  D sc Acyl s ¸¸ 'p
© dp dp dp dp ¹ p p0 (4.10)
 D sc Acyl 'Tm  hs ( p 0 )'qs ,
d'Tm dTs
mC p ,m D sc Acyl 'p  D sc Acyl 'Tm  D cp AcylK'Tm ,
dt dp p p0

where the states are expressed in terms of deviations and the equilibrium
point in (4.8) have been used to simplify the expression. Assuming that

dhs dh dq dT
qs0  qw ( p 0 ) w  hw ( p 0 ) w  D sc Acyl s , (4.11a)
dp dp dp dp

and

D cpK  D sc , (4.11b)

the model becomes

dU s d'p dTs
hs ( p 0 )V  D sc Acyl 'p  D sc Acyl 'Tm  hs ( p 0 )'qs ,
dp p p0
dt dp p p0
d'Tm dTs
mC p , m D sc Acyl 'p  D sc Acyl 'Tm .
dt dp p p 0

(4.12)

Assumption (4.11b) is another way of saying that Qp is varying much

more slowly than Qm when the inlet steam flow is changed. The interface
between the cylinder and paper then acts as a large heat transfer barrier.
From experimental values in [Karlsson, 2000], it is found that the right
hand side is 2 í 20 times larger than the left hand side in (4.11b). For the
lower region of that range, it is probably unsatisfying to use model (4.12).
However, experiments show that large values of Dcp only occur for high
moisture contents ( > 40 %) and high fabric tensions. It is therefore the
belief of the author that this is an uncommon situation and the right hand

84
 Chapter 4. A Physical Model of a Steam Heated Cylinder
side of (4.11b) is generally much larger that the left hand side. This is
further discussed in Section 4.4 and Chapter 8. The inequality in (4.11a)
will be commented and examined later in the simulations, and also in
Chapter 8.
Writing the system in standard state-space form, we find that

x Ax  B'qs ,
(4.13)
y Cx  D'qs ,

where x >'p 'Tm @ T and

ª dTs º
« Acyl dp Acyl » ª 1 º
« » « dU s »
« h V dU s dU
hsV s » «V »
A D sc «
s
dp dp », B « dp »,
« dTs » « »
« Acyl » « »
« dp Acyl » « »
 ¬ 0 ¼
« mC mC p ,m »¼
¬ p ,m

(4.14)
C >1 0@, D 0.

The steam properties are here assumed to in given in their equilibrium

values. The matrix A has an eigenvalue at the origin and one eigenvalue
on the negative real axis. The transfer function from steam flow to
pressure is

s  a 22 sz b1 § z O  z ·
G ( s ) b1 b1 ¨  ¸, (4.15)
ss  a11  a 22  ss  O  O ©s sO¹

where

85
Chapter 4. A Physical Model of a Steam Heated Cylinder

1 D sc Acyl
b1 , z  a22 ,
dU s mC p
V
dp
§ dTs · (4.16)
¨ Acyl ¸
dp Acyl ¸
O a11  a22 D sc ¨  .
¨ dU s mC p ,m ¸
¨ hsV dp ¸
© ¹

Note that the model (4.15) has an IPZ structure. The essential parameters
of the model are

x Cylinder volume V
x Cylinder mass m
x Specific heat capacity of metal Cp,m
x Area of the cylinder surface Acyl
x Steam properties hs , dU s dp, dTs dp
x Heat transfer coefficient Dsc

All parameters, except the heat transfer coefficient, Dsc, are known
beforehand, either by machine specifications or from a physics handbook
which includes a steam table and heat capacities. The heat transfer
coefficient depends on both amount of condensate and its degree of
turbulence, and is very difficult to predict. Therefore it is used to fit the
model to the measured data. Note that it is only the last two items in the
parameter list that depend on the operating point.
The following assumptions have been made in the development of the
model

x No blow-through steam
x The steam in the cylinder is saturated
x Paper temperature is constant
x The thermal dynamics of the condensate is fast compared to the
cylinder shell
x The condition (4.11)

The pole O and the zero z are both proportional to the heat transfer
coefficient Dsc. For large s the transfer function (4.15) is approximated by

86
 Chapter 4. A Physical Model of a Steam Heated Cylinder
b1
G ( s) | , (4.17)
s

where b1 does not depend on Dsc. For small s the transfer function can be
approximated by

b1 z
G(s) | , (4.18)
Os

where

b1 z hs
, (4.19)
O dTs dU
mC p , m  Vhs s
dp dp

does not depend on Dsc. Therefore neither the initial part of a step response
nor the slope of the asymptote depend on the heat transfer coefficient.
Finally, the relations between the black-box and grey-box parameters are

hs mC p , m
kv , T1 ,
dTs dU s D A
mC p , m  Vhs sc cyl
dp dp
dU (4.20)
mC p , m hsV s
dp
T2 .
§ dTs dU s ·
D sc Acyl ¨¨ mC p , m  Vhs ¸
© dp dp ¸¹

Note that the velocity gain, kv, above is not normalized by the measuring
or any actuator range.
As can be noticed, the grey-box model does not explain the time delay
often seen in the black-box model. It is important to remember that we
often are dealing with sampled measurements and it has been observed, in
practice, that the time delay of the system often is close to the sampling
time. Moreover, we have seen from the derivation of the grey-box model
that there are neglected dynamics in the model, which may possibly give
rise to an estimation of the dead-time that is larger than the true value.

87
Chapter 4. A Physical Model of a Steam Heated Cylinder
Remark 1
A unit analysis of the expressions in (4.20) shows that T1 and T2 are given
in seconds, while kv is given in Pa/kg. Since kv is a velocity gain, one
would expect its unit to include seconds. However, the inflow of steam is
given in kg/s and the time unit is therefore canceled.

Remark 2
The energy flow to the metal (4.3) can be derived as the solution of
Fourier’s law of heat conduction

Q  kA’T , (4.21)

where k is the thermal conductivity, A is the area, T the temperature, and

Q the energy flow. The negative sign indicates that the temperature
gradient is in the opposite direction of the energy flow. Assuming there is
no build-up of heat at any point along the path of the heat flow, Q is
constant along the path. Consider a one-dimensional and homogeneous
system, and integrate (4.21) from point “0” to point “1”

x1 T1

³ Qdx
x0
³
kA dT
T0
(4.22)

or equivalently

T0  T1 k
Q kA A(T0  T1 ). (4.23)
x1  x0 'x

Writing the quotient k/ǻx as a heat transfer coefficient Į, we have the

relation in (4.3). It might appear as a large restriction to assume constant
energy flow along the path but it can still vary with time, there is simply
no build-up. By combining the energy balance for the metal in (4.2c) with

Qm D sc Acyl (Ts  Tm ), (4.3)

we get a first-order system from steam temperature to mean metal

temperature, which also corresponds to the solution given by the
distributed heat equation (with constant energy flow to the paper as
boundary condition), see Appendix C (Figure C.5).

88
 Chapter 4. A Physical Model of a Steam Heated Cylinder
Remark 3
The steady-state solution to (4.1) and (4.2) is

qs qc qw
Qm qc (hs  hw ) (4.24)
Qm Qp

As expected, both the mass flows and energy flows are equal. The energy
flow to the metal, Qm, is given by the condensation rate times the
difference in enthalpy due to condensation. The drying is, in other words,
powered by the latent heat of vaporization of the steam.

4.2 Time and frequency domain analysis

To investigate the dynamic behavior of the linearized model, given by

(4.13), we will look at the Bode plot and a step response. The machine
dependent parameters are taken from a steam group of a fluting machine,
running at an operating point with a steam pressure of 90 kPa (gauge
pressure), a nominal speed of 450 í 600 m/min and a basis weight
between 110 í 200 g/m2. The machine dependent values used for the
simulation are

x Cylinder volume: V = 12.6 m3

x Cylinder mass: m = 7610 kg
x Cylinder area: Acyl = 37.2 m2
x Heat capacity for cast iron: Cp,m = 500 J/(kg˜K)
x Steam properties for the given operating point

The nominal steam mass flow rate to each cylinder is approximately

0.25 kg/s. This value is obtained by simply dividing a measurement of the
total machine steam consumption by the number of cylinders. Figure 4.2
shows the Bode plot of the process, where the gain is normalized by the
measuring ranges of the input and output to make it unitless.
The amplitude gain is independent of Dsc, both at high and low
frequencies, as shown in (4.17) and (4.18). The heat transfer coefficient
has a considerable influence on both the gain and phase in the mid-
frequency range. For the purpose of designing a PID-controller, this
difference in gain and phase influences the controller parameters.
Otherwise it would be possible to tune the controller solely from cylinder
89
Chapter 4. A Physical Model of a Steam Heated Cylinder
1
10
Magnitude (abs)

0
10

-1
10

-3 -2 -1 0
10 10 10 10

-45
Phase (deg)

-65

-90 -3 -2 -1 0
10 10 10 10
Frequency (rad/s)
Figure 4.2 Frequency properties for different values of the heat transfer coefficient Dsc =
500 (dotted), 1000 (dashed), and 2000 (solid).

dimensions and a steam table. It can also be seen that, a higher heat
transfer coefficient yields a lower steam pressure gain, since there is a
larger heat transfer to the cylinder and a higher condensation rate.
The linearized model has also been simulated with a step in the input
signal, shown in Figure 4.3. As the figures illustrate, a higher heat transfer
coefficient gives a lower steam pressure, at a given time instance, due to a
larger heat transfer to the cylinder shell. This has also been pointed out in
[Nelson and Gardner, 1996]. A greater energy flow through the cylinder
shell gives a higher cylinder temperature even though the steam has a
lower temperature, at any time index. The effect of a larger Dsc, in basic
terms, dominates over the effect of a lower steam temperature. It is
essential to remember that there are two effects taking place in this
pressureítemperature process. The heat transfer coefficient varies
between the simulations and the steam temperature is changing with time,
and the cylinder temperature depends on both.
In (4.11a) an inequality that depends on the operating point and
cylinder dimensions was utilized to make a significant simplification.
Using values from this example, we can examine its justification. Apart

90
 Chapter 4. A Physical Model of a Steam Heated Cylinder
6000

4000
' p (Pa)

2000

0
0 10 20 30 40 50 60 70 80 90 100

0.6

0.4
' T (K)

0.2

0
0 10 20 30 40 50 60 70 80 90 100
Time (s)
Figure 4.3 Step responses for different values of the heat transfer coefficient Dsc = 500
(dotted), 1000 (dashed), and 2000 (solid). The upper graph shows the steam pressure and
the lower shows the cylinder temperature.

from the steam properties, we also need an expression for the derivative of
siphon flow rate, qw, with respect to the cylinder pressure. Some
experimental values are given in [Stenström and Svanquist, 1991] and
using those, we find that the right hand side of (4.11a) is 10 to 20 times
larger than the left hand side. The next section shows that the model has a
good fit to experimental data.

4.3 Comparisons with plant data

To evaluate the accuracy of the grey-box model it has been calibrated and
validated against measurements from a steam- and condensate system.
The experiments have been carried out on a paperboard machine and
signals have been measured with a sampling time of 1 s. The cylinder data
is

x Cylinder volume: V = 18.4 m3

x Cylinder mass: m = 8300 kg
x Cylinder area: Acyl = 45.5 m2

91
Chapter 4. A Physical Model of a Steam Heated Cylinder
Since the input signal (steam input flow) is not manipulated directly,
neither measured, a model for a steam valve has to be added to (4.13). A
simple approach is to assume a static linear relationship between the
controller signal, uc, and the steam input flow, qs, namely

qs d v uc , (4.25)

where dv is a valve constant which will be the second calibration

parameter together with the heat transfer coefficient, Dsc. Using this valve
description, we keep the linearity and IPZ-structure in the model, given in
(4.13).
To calibrate the model, the functions idgrey.m and pem.m in
System Identification Toolbox for Matlab, were used to find the optimal
calibration parameters. The optimization method is based on minimizing
the prediction error, see [Ljung, 1999].
Figure 4.4 shows an open loop response together with the calibrated
model, where the control signal and model output are bias corrected (by
the linear model, a steady-state pressure level can only be reached for a
closed valve). The calibration parameters obtained are

Dsc = 1820 W/(m2˜K), dv = 0.00308 kg/(s˜%). (4.26)

To compare the result with nominal values cited in literature, we need a

heat transfer coefficient through only the condensate film, D scc . From
[Karlsson, 2000] the relationship

1
D sc , (4.27)
1 G cyl

D scc Ocyl

is given, where Gcyl is the distance into the cylinder where the temperature
is equal to the mean cylinder temperature, and Ocyl is the thermal
conductivity of the cylinder shell. In this example, the cylinder thickness
is 25 mm (the mean temperature, Tm, occurs in the middle of the cylinder
shell), and the thermal conductivity is 50 W/(m˜K). The heat transfer
coefficient through the condensate is then

D scc = 3340 W/(m2˜K). (4.28)

92
 Chapter 4. A Physical Model of a Steam Heated Cylinder
425

p (kPa)

420

415
0 50 100 150 200 250 300 350 400 450 500 550

60

55
u c (%)

50

45
0 50 100 150 200 250 300 350 400 450 500 550
Time (s)
Figure 4.4 Calibrated model (dotted) and measured data (solid).

375

370
p (kPa)

365

360

0 200 400 600 800 1000 1200 1400 1600 1800 2000

55

50
u c (%)

45

40

35
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (s)
Figure 4.5 Validation of model (dotted) against measured data (solid).

93
Chapter 4. A Physical Model of a Steam Heated Cylinder
In [Karlsson, 2000], typical values of the heat transfer coefficient are
given. They depend strongly on the condensate thickness and turbulence,
and can vary between 1000 and 4000 W/(m2˜K), where 2000 is a nominal
value. Nevertheless, the fact that our estimated value from the model is
within that range, gives support for the legitimacy of the model.
The total mass flow rate of steam, during the experiment, to the drying
section is 85 ton/h. The machine has 93 cylinders, so the average steam
flow per cylinder is 0.254 kg/s. The parameter dv, in (4.26), and the
average valve opening gives the steam flow to the particular cylinder in
the model, namely 0.154 kg/s. The steam flow is likely to vary a great
deal between different drying groups but by comparing the two values, we
know that also the second calibration parameter is realistic.
In (4.29) the grey-box model is compared with the corresponding
black-box model, adjusted on the same data set. The velocity gain in the
models is normalized with the measuring range of the pressure gauge. The
black-box model is not shown in the figures but it gives a slightly better fit
to the data, since it has more degrees of freedom (three parameters to
adjust instead of two). There is also a difference in the parameters of the
transfer functions. The time delay is equal to the sampling time and comes
from the identification procedure (pem.m in System Identification
Toolbox for Matlab).

G grey ( s ) 0.00243
50.1s  1 e  s
s 20.4s  1
(4.29)
Gblack ( s ) 0.00176
77.8s  1 e  s
s 21.3s  1

The grey-box model has also been validated graphically by using the
control signal values to simulate an output. The model output is then
compared with the measured steam pressure. Figure 4.5 shows such an
evaluation. The excitation in the control signal is generated by a series of
steps in the set point (closed loop), which is not shown in the figure to
keep it clear.

4.4 A modified model

In Chapter 3, it was discussed that in some cases the IPZ-structure is not

sufficient to describe the pressure dynamics in a steam cylinder. This can
be resolved by changing the integrator to a real pole. In the grey-box

94
 Chapter 4. A Physical Model of a Steam Heated Cylinder
model this can be accomplished by disregarding the assumption from
(4.11b), saying that

D cpK  D sc . (4.11b)

It was previously discussed that for high moisture contents and high fabric
tensions, this assumption might not be valid. The system matrix in (4.14)
now becomes

ª dTs º
« D sc Acyl dp D sc Acyl »
« »
« h V dU s hsV
dU s »
s
Ac « dp dp », (4.30)
« dT »
« D sc Acyl s »
« dp Acyl (D sc  D cpK ) »

« mC mC p ,m »
¬ p ,m ¼

and the other matrices are unchanged. By examining time series where the
IPZ-structure is sufficient with cases where it is not, an explanation to the
modeling problem is found. If Dsc >>Dcp then model (4.13) is adequate
and Dsc § Dcp means that (4.30) is a better structure. This has also been
verified by simulation of the primary DAE system (4.1)í(4.3) in
Modelica, see Chapter 8. Closer examination of (4.30) shows that
Dsc >>Dcp gives a system with one fast pole and one close to the origin.
When Dcp is increased, the slow pole moves along the real negative axis
towards the other pole, and it can then no longer be regarded as an
integrator.
It can be shown, knowing that all factors in the elements of (4.30) are
positive, that the eigenvalues of A´ are real and negative. This is nice since
it would be unsatisfying to have a model that is unstable for some
combinations of physical parameters, when it is clear that the process is
stable in reality. The characteristic equation of (4.30) is

s 2  (a11  a22 ) s  a11a22  a12 a21 , (4.31)

where a11, a12, a21, and a22 are the elements of matrix A´. Identification of
the parameters gives

95
Chapter 4. A Physical Model of a Steam Heated Cylinder
dTs
D sc Acyl
dp Acyl (D sc  D cpK )
a11  a22    0,
dU s mC p , m
hsV
dp
(4.32)
2 dTs
KD scD cp Acyl
dp
a11a22  a12 a21 ! 0,
dU s
hsVmC p , m
dp

and since the coefficients of the polynomial (4.31) are all positive, the
system is stable. The roots of (4.31) are

a11  a22 (a11  a22 ) 2

s1, 2 r  a12 a21 , (4.33)
2 4

and since both a12 and a21 are positive, the solution has no complex parts.
The relation between the position of the poles and the physical parameters
is a bit more complicated than in (4.16).
It can also be shown that the initial dynamics of the modified model in
(4.30) are equal to (4.17). For large s we have

1
G(s) | . (4.34)
dU s
V s
dp

The low frequency properties will be different for the two models
however, since (4.13) contains an integrator and has no steady state gain.
For small s the modified model becomes

hs (D sc  D cpK )
lim G ( s ) . (4.35)
s o0 dT
D sc AcylKD cp s
dp

96
 Chapter 4. A Physical Model of a Steam Heated Cylinder
4.5 Summary

Chapter 3 introduced a black-box model structure for the steam pressure

dynamics. In this chapter an additional model has been presented, a grey-
box model. Both have the same structure but different purposes. The
intention of the black-box model is purely controller tuning and the main
purpose of the grey-box model is to gain insight into the physical laws
behind the black-box model. Issues like how the heat transfer coefficient,
cylinder diameter, condensate removal, selection of materials or steam
pressure affect the dynamical performance of the system can then be
answered. This might have effect on the mechanical design of the drying
section, such as the siphon shape and form, dryer bars, cylinder
dimensions etc.
There is also a potential to make a recursive identification of the heat
transfer coefficient for fault detection with respect to condensate
evacuation. It would then be beneficial to have a separate pressure meter
and mass flow meter installed at the drying cylinder of interest, to acquire
an accurate estimate. An important thing to remember here is that the
model is an approximation of the real process. Model errors will therefore
be included in the calibration parameter and its absolute value might be
uncertain. The relative value is a useful parameter though.
The grey-box model has been validated by measurement from a paper
machine with good results. Further examination of the model can be found
in Chapter 8, where it is combined with a paper web model.

97
Chapter 5. A Tuning Method for IPZ Models

5.
5
A Tuning Method for IPZ
Models

From practical experience, it has been seen that there can be significant
disturbances in the drying section, which makes disturbance attenuation
an important issue. Since the prevalent machine design standard is to
connect all the dryer groups to the same steam header, a disturbance in
one group easily affects the other groups. In this section, a simple tuning
rule for both PI and PID control is presented. Many tuning methods for PI
and PID control have been proposed previously but most of them
exclusively suppose a first-order system with dead time or an integrating
process with dead time. Since this does not fit very well to the IPZ-
process, a new method is necessary. The design goal is to obtain good
load disturbance response. The tuning rule is based on the four process
parameters of the IPZ transfer function and is therefore denoted as IPZ-
tuning. The idea is to provide operators and control engineers at the paper
mill with a simple tuning tool, which can be used without any complex
optimization calculations. To give the user the option to balance between
robustness and performance, the tuning rule has a design parameter. This
parameter is derived from the Nyquist stability theory and is
dimensionless, which is a nice property since it is then independent of unit
selections. The IPZ-tuning rule is tested and evaluated on an industrial
paper machine. It is also compared with a few other design methods.
98
 Chapter 5. A Tuning Method for IPZ Models
5.1 A design method based on optimization

The word optimize comes from the Latin word optimus which means the
best (maximize comes from maximus which means the highest).
Optimization methods are an important tool in nearly all engineering
domains, and the control field is indeed no exception. Many control
methods are based on optimization of some vital criteria, subject to one or
a few constraints. However, it is important to consider how the
optimization problem is set up, since this obviously very much affects the
result and it is not necessarily a good controller that comes out of an
optimal solution. If the optimization problem is not correctly formulated,
the optimal controller might not even be stable. The formulation of the
problem also affects how easily it is solved. There exist very powerful
numerical tools that solve linear or quadratic programming problems,
while non-convex optimization is much more demanding due to less
effective numerics and the existence of several local optimal solutions.
The derivation of the IPZ-tuning rule is based on a design proposal in
[Åström, et al, 1998] and [Åström and Hägglund, 2004]. The idea is to
maximize the integral gain, defined as ki = kc / Ti, subject to a robustness
constraint. Due to the form of the constraint, the optimization problem is
non-convex. Therefore, much effort is put on finding a simple relation
between the process parameters and the optimal solution of controller
parameters, so that the user is relieved from the optimization issue.
Instead, he or she will get an approximate solution from just a few button
pushes on a pocket calculator.
By maximizing the integral gain, the absolute value of the integrated
error (IE) of a step load disturbance is minimized. This can easily be seen
by using the nomenclature in Figure 3.6 and writing the error as

E (s) R( s)  Y ( s)
§ P( s) P( s )C ff ( s ) ·
R( s )  ¨¨ D( s)  R( s ) ¸¸ (5.1)
© 1  P ( s )C c ( s ) 1  P ( s )C c ( s ) ¹
§ P ( s )C ff ( s ) · P( s)
¨1  ¸
¨ 1  P ( s )C ( s ) ¸ R ( s )  1  P ( s )C ( s ) D( s ).
© c ¹ c

Assume a step unit load disturbance and that the set point is zero. Evaluate
the integral of the error by the final value theorem

99
Chapter 5. A Tuning Method for IPZ Models
f
P( s) 1
IE ³ e(t )dt
0
lim 
s o0 1  P ( s )Cc ( s ) s
P( s) 1
lim  2
(5.2)
s o0 1  Ti s  TiTd s s
1  P( s) kc
Ti s
Ti P ( s ) Ti
lim   .
s o0 Ti s  kc (1  Ti s  TiTd s 2 ) P ( s ) kc

By maximizing the integral gain, this integrated error is minimized if the

sign is disregarded. However, to only maximize the integral gain is not
sufficient since there is no guarantee that the control loop is stable, see
Figure 5.1. Therefore, an additional constraint is needed. The robustness
constraint used here is characterized by a circle with its centre at the point
í1 in the Nyquist diagram, see Figure 5.2. By avoiding the point of
instability with a certain distance, R0, stability is guaranteed. The radius of
the circle will then be the design parameter, and the smaller R0 is the more
aggressive the controller will be. This can also be expressed in terms of
the sensitivity function, defined as

Positive area
Control error

Negative area

Time

Figure 5.1 A characteristic response to a load disturbance on the IPZ process, controlled
by a PID controller. It is crucial add a constraint to the optimization problem to avoid an
oscillatory solution, since equally large positive and negative areas cancel each other by
the IE criteria.

100
 Chapter 5. A Tuning Method for IPZ Models

Im

R0
Re
–1

Figure 5.2 The robustness constraint of the design method. By varying the radius of the
circle, the degree of robustness is changed.

1
S (s) , (5.3)
1  L( s )

where L(s) is the loop transfer function. The maximum sensitivity, Ms, is
then given by

1
Ms max S (iZ ) max . (5.4)
Z Z 1  L(iZ )

Since |1+L(iZ)| is the distance from a point on the Nyquist curve to the
critical point í1, the shortest distance from the Nyquist curve to the point
í1 is thus 1/Ms. Therefore, we get

1
Ms , (5.5)
R0

and we can then use Ms as our design parameter, when deriving the tuning
rule. The nice thing about the maximum sensitivity function is that it
connects the open loop Nyquist curve with a closed loop property. It is
also dimensionless which is a nice property. The disadvantage is that it
can be difficult to relate to for the unfamiliar user. A common question,
e.g., can be what Ms = 1.2 means in practice. Simulations and
familiarization is one answer to that issue.

101
Chapter 5. A Tuning Method for IPZ Models

Figure 5.3 The complete Nyquist curve for the IPZ-process and a PID controller if the
dead time is assumed to be zero. Otherwise, there will be the typical circular appearance in
the origin.
The shape of the Nyquist curve is changed by varying the controller
parameters, in order to avoid the Ms-circle while achieving a high integral
gain. The derivative time Td, affects the curve close to the origin (high
frequencies), the integral time Ti affects low frequencies, while the gain kc
affects all frequencies equally. However, a few properties of the Nyquist
curve are independent of the controller parameters. Since both the IPZ-
process and a PID controller contain one integrator each, the Nyquist
curve starts at the phase lag íS. At high frequencies the phase is íS/2, if
we assume no dead time. The complete Nyquist plot then appears as in
Figure 5.3.
Note that this design method does not give any suggestion about the
feedforward parameters E or Ȗ of the controller, see (3.19), since these
parameters have no influence on the sensitivity function or the disturbance
rejection.
The design method puts emphasis on disturbance rejection. In
[Slätteke, et al, 2002] it was shown that for a certain value of the
maximum sensitivity (Ms=1.2), maximizing the integral gain or
maximizing the bandwidth of the closed loop system (from set point to
measurement) gave essentially the same controller parameters. That
means that emphasis is put on both the regulator and servo problem. This
does not always hold though, even if the resulting controllers from the two
criteria prove to be close.

102
 Chapter 5. A Tuning Method for IPZ Models
Remark
Apart from IE, a few other minimization criteria have been proposed in
different contexts, e.g. the integrated absolute error (IAE), integrated
square error (ISE), and integrated absolute time weighted error (IATE). A
disadvantage with ISE is that it penalizes large errors and gives a small
but long-term error. IAE have the advantage to avoid oscillatory
responses. However, if the system is well damped, IE and IAE give
similar results. Keep in mind that IE = IAE, if the error is positive. This is
also noted in [Åström, et al, 1998]. IE is therefore chosen as the
optimization criterion, since it also is computationally effortless.

5.2 The IPZ tuning rule for PI control

A large number of different IPZ models, see (3.1), have been tuned
according to the method described in the previous section, with respect to
different Ms-values. The Ms-values have been chosen to give practically
reasonable and sound controller settings. The models have been chosen to
represent a large variety of cylinder pressure processes encountered at
different paper machines, producing different grades and qualities. The
models cover many different cases, from L << T2 to L > T2, T1 from 50 to
800, and T2 from 2 to 400. However, process parameters T1, T2, and L are
presumed to be positive. This implies that, e.g. non-minimum phase
systems are not included. Furthermore, T1 is always larger than T2 but
cases where T2 is almost as large as T1 are also included in the batch of
models. The purpose of this is to cover the whole range of models found
in the industry. This is also of interest in other process areas since a PPZ
structure appears when two first order models are added (parallel paths in
integrated plants) and if the two time constants differ in magnitude, it can
be modeled as an IPZ process.
The result for Ms = 1.1 is plotted in Figure 5.4, where each circle
represents one set of process parameters and the corresponding optimal
controller parameters. In the plot, the controller gain and integral time
have been drawn in combination with the process parameters to give a
linear dimensionless relationship. The advantage with dimensionless
parameters is that they are independent of the selection of units. Whether
the unit of the velocity gain kv, for example, is given in kPa/(%˜s) or
simply sí1, does not affect the plot in Figure 5.4.
Even though there is some experimental work in finding combinations
that give a clear relationship between the parameters, there is also a
structure in the selection. Due to their units, kv and kc belongs together, in

103
Chapter 5. A Tuning Method for IPZ Models
0.07

0.06

0.05

0.04
kv kc L

0.03

0.02

0.01

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(T 2 +L /3)/T 1

1.0

0.9

0.8
T i / (T i+L )

0.7

0.6

0.5

0.4

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

T 2 / (T 2 + L )
Figure 5.4 The linear relationship between the PI controller parameters and process
parameters for Ms = 1.1. Each circle represents one set of process parameters from the
investigated batch and corresponding control parameters from the optimization routine.

combination with a parameter that is given in seconds (T1, T2, Ti, or L). As
will be obvious shortly, it must be L because of the influence it has on the
controller gain. Moreover, larger T1 compared to T2 give more lead action
in the process, which must results in a smaller controller gain, given a

104
 Chapter 5. A Tuning Method for IPZ Models
certain robustness. Therefore the ratio between T1 and T2 (together with a
correction term L/3, that is found ad hoc) is related to kc. Also, it seems
natural to expect Ti depend on the ‘time constant’ T2, compare e.g. lambda
tuning [Morari and Zafiriou, 1989], and here it is also affected by L in a
way that the expressions on both axes, in Figure 5.4, ranges between zero
and one.
The figure shows an apparent linearity. The least squares method has
been used to get an equation for this relationship, given in the controller
parameters. Obviously, linearity is not a necessity. Curves of higher
degrees are possible to use, as long as the desired parameters can be
analytically solved for.
The chosen Ms-values are 1.1, 1.2, 1.3, and 1.4. The resulting
controller parameters are given by

L
T2 
3, T 6T2  L
Ms 1.1 : k c 0.09 i 4L ,
T1 k v L T2  21L
L
T2 
3, T 23T2  4 L
Ms 1.2 : k c 0.16 i 5L ,
T1 k v L 9T2  105 L
(5.6)
L
T2 
3, T 100T2  17 L
Ms 1.3 : k c 0.23 i L ,
T1 k v L 11T2  94 L
L
T2 
3, T 88T2  15 L
Ms 1.4 : k c 0.28 i L .
T1 k v L 12T2  85 L

By looking at the controller gain for the different Ms-values, it is easy to

interpret the implication of the tuning parameter. The larger the maximum
sensitivity is chosen to be (remember that a large value means a less
robust controller); the larger is the controller gain, as expected. The
interpretation of the integral time is not so obvious by simply looking at
the equations. In Figure 5.5 the integral time is plotted against the time
constant T2 of the IPZ model, for a specific time delay L. Here we can see
that the larger the value of the design parameter, Ms, is the smaller will the
integral time be, for a given value of T2. As in the analysis of kc above,
this is also expected.

105
Chapter 5. A Tuning Method for IPZ Models
30
Solid: M = 1.1
s
Dashed: M = 1.2
25 s
Dotted: M = 1.3
s
Dash-dotted: M = 1.4
s
20
i
T

15

10

0
0 10 20 30 40 50
T
2
Figure 5.5 The integral time, Ti, as function of the time constant T2, for the PI tuning rule.
For small T2, the two variables are almost equal, otherwise Ti is smaller that T2. The thin
solid line shows Ti = T2, as a comparison. This is also suggested in a tuning method by
[Nelson and Gardner, 1996].

If the time delay in the IPZ-model is very short, the gain will become
increasingly large and the integral time short, in the formulas above. This
is theoretically correct, since the constraint is still fulfilled for the control
loop. By considering the root locus of the feedback loop and assuming
L = 0, shown in Figure 5.6, we can see that an increasing gain does not
move any of the closed loop poles into the right half plane. Neither by
varying the integral time, the loop will become unstable at any time as
long as it is positive. There is no limitation because of system dynamics. It
has infinite gain margin and any combination of positive kc and Ti can be
achieved. But, high controller gain amplifies measurement noise and gives
large control signals, which will saturate actuators. Limitations are instead
given by [Åström, 2000b]

x Sensor noise
x Unmodeled process dynamics
x Actuator saturation
x Admissible control signal variations
x Sensor and actuator resolution

106
 Chapter 5. A Tuning Method for IPZ Models
We can conclude that the dead time of the process is directly related to the
stability of the feedback system. However, it is unusual to obtain a dead
time equal to zero, when identifying the steam pressure process, partly
because of the sample time in today’s discrete DCS-systems. Thus, this
should not impose large restrictions to the usefulness of the method.
In Figure 5.7 and Figure 5.8 simulations of an IPZ-tuned PI controller
for different values of Ms and with ȕ = 1 or ȕ = 0. The process is given by

(200 s  1)  s
P ( s ) 0.01 e . (5.7)
s (20 s  1)

With ȕ = 0 the control signal smoother and the overshoot smaller for the
step in set point compared to the case with ȕ = 1. This can also be
accomplished with a more sophisticated set point filter. Remember that
the IPZ tuning is based on optimization of disturbance rejection and can
with advantage be combined with e.g. the 2DOF-controller described in
Section 3.3.

Remark 1
By looking at the root locus in Figure 5.6, an interesting observation can
be made. If the dead time is not dominating the process dynamics, higher
controller gain makes the closed loop system less oscillatory. This
counter-intuitive phenomenon can also be seen when controlling a pure
integrating process with a PI or PID controller.

1 1 1
  
T2 Ti T1

Figure 5.6 The root locus for the IPZ-process in connection with a PI controller for the
case T1 > Ti > T2. Good tuning normally requires T1 >> Ti § T2.

107
Chapter 5. A Tuning Method for IPZ Models
1.5

1.0
r, y

0.5

0
0 20 40 60 80 100 120 140 160 180 200

3
uc

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)
Figure 5.7 Closed loop response of (5.7) and an IPZ-tuned PI controller for Ms = 1.1
(solid), Ms = 1.2 (dashed), Ms = 1.3 (dotted), and Ms = 1.4 (dash-dotted). ȕ = 1.

1.0
r, y

0.5

0
0 20 40 60 80 100 120 140 160 180 200

1.5
uc

0.5

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)
Figure 5.8 Simulation with the same process and controller as in Figure 5.7 but with ȕ = 0.

108
 Chapter 5. A Tuning Method for IPZ Models
Remark 2
The root locus also shows why the open loop zero in the process can not
be canceled by a PI controller, as discussed in Chapter 3. It is only by
having infinite controller gain that one of the feedback poles will reach the
process zero in í1 / T1.
Remark 3
The feature of the IPZ-tuning, to give a non-realizable controller when the
process dead time is zero, is not unique for this method. Many other
tuning methods for minimum phase systems have the same property, see
e.g. [Ziegler and Nichols, 1942], [Chien, et al, 1952], [Cohen and Coon,
1953], [Ho, et al, 1995], [Poulin and Pomerleau, 1999], and [Visioli,
2001].
Remark 4
Small values of T2 will generally give a short integral time, Ti by the IPZ-
tuning. Then it is also important to consider the sample time so that the
process is sampled sufficiently fast [Åström and Hägglund, 2005].

5.3 The IPZ tuning rule for PID control

The same technique used in the previous section was also used to obtain
tuning parameters for the PID controller. Using (3.20a) the high frequency
gain for the open loop system is

§ 1 · 1  iZT1 kc kvT1Td
lim kc kv ¨¨1   Td iZ ¸¸ e  iZL . (5.8)
Z of
© Ti i Z ¹ iZ (1  iZ T2 ) T2

Physically this means that the open loop system has infinite bandwidth
which is unrealistic. To obtain a finite bandwidth, (3.20b) is used instead,
where N is chosen to 10. The result for Ms = 1.2 is shown in Figure 5.9.
The integral time proves to have a more complex appearance that in the
case of PI control, compare to Figure 5.4. This is also the case for the
derivative time and therefore second-degree curves have been fitted to the
optimal parameters, except for the controller gain. The resulting controller
parameters are given in (5.9). We see that the equations are a bit more
complex than compared to PI control (5.6), but have the same property of
giving infinite controller gain and zero integral time for zero dead time.
In Figure 5.10 and Figure 5.11 simulations of an IPZ-tuned PID
controller for different values of Ms and with ȕ = 1 or ȕ = 0.
109
Chapter 5. A Tuning Method for IPZ Models
0.18

0.16

0.14

0.12

kv kc L 0.1

0.08

0.06

0.04

0.02

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(T 2+L/3)/T 1

1.0

0.9

0.8
T i / (T i+L)

0.7

0.6

0.5

0.4

0.3
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
T 2 / (T 2+L)

1.0

0.8

0.6
Td / T2

0.4

0.2

0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
T 2 / (T 2+L)

Figure 5.9 The relationship between the PID controller parameters and process parameters
for Ms = 1.2.
110
 Chapter 5. A Tuning Method for IPZ Models
3T2  L
kc
22k v T1 L

L(472T22  471T2 L  146 L2 )

Ms = 1.1: Ti
(28T22  529T2 L  354 L2 )
T2 (7T22  264T2 L  1359 L2 )
Td , N 10
1000(T22  2T2 L  L2 )

3T2  L
kc
12k v T1 L

L(883T22  875T2 L  286 L2 )

Ms = 1.2: Ti
3(39T22  375T2 L  238L2 )
T2 (T22  361T2 L  1400 L2 )
Td , N 10
1000(T22  2T2 L  L2 )
(5.9)
3(3T2  L)
kc
26k v T1 L

L(835T22  842T2 L  277 L2 )

Ms = 1.3: Ti
3(55T22  386T2 L  241L2 )
2T2 (3T22  176T2 L  736 L2 )
Td , N 10
1000(T22  2T2 L  L2 )

3(3T2  L)
kc
20k v T1 L

L(786T22  851T2 L  278L2 )

Ms = 1.4: Ti
214T22  1149T2 L  722 L2 )
T2 (8T22  367T2 L  1443L2 )
Td , N 10
1000(T22  2T2 L  L2 )

111
Chapter 5. A Tuning Method for IPZ Models
1.5

1.0
r, y

0.5

0
0 20 40 60 80 100 120 140 160 180 200

4
uc

0 20 40 60 80 100 120 140 160 180 200

Time (s)
Figure 5.10 Closed loop response of (5.7) and an IPZ-tuned PID controller for Ms = 1.1
(solid), Ms = 1.2 (dashed), Ms = 1.3 (dotted), and Ms = 1.4 (dash-dotted). ȕ = 1 and Ȗ = 0.

1.0
r, y

0.5

0
0 20 40 60 80 100 120 140 160 180 200

2
uc

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)
Figure 5.11 Simulation with the same process and controller as Figure 5.10 but with ȕ = 0.

112
 Chapter 5. A Tuning Method for IPZ Models
5.4 Stability regions

A nice tool to graphically examine stability is stability regions [Åström

and Hägglund, 2001]. It will be used here to relate the parameters
suggested by the IPZ-tuning and the stability boundary. It will later be
applied in Section 5.7 also, when comparing other tuning methods.
Obviously, staying inside this boundary does not automatically imply
good tuning, only that the closed loop is stable. Start by writing the
process as

G (iZ ) r (Z )eiM (Z ) r (Z )cos M (Z )  i sin M (Z ) , (5.10)

and let the controller be given by

ki
C ( s ) kc   k d s, (5.11)
s

where kd = kcTd. The condition for oscillation is then given by Nyquist’s

theorem

§ k ·
G (iZ )C (iZ ) r (Z )cos M (Z )  i sin M (Z )  ¨ k c  i i  ik d Z ¸ 1. (5.12)
© Z ¹

Identifying the real and imaginary parts, we find that the boundary of the
stability region can be represented by

cos M (Z ) Z sin M (Z )
kc  , ki Z 2kd  . (5.13)
r (Z ) r (Z )

For the IPZ process, we have

(1  Z 2T12 ) 0.5
r (Z ) kv ,
Z (1  Z 2T22 ) 0.5
(5.14)
S
M (Z ) arctan(ZT1 )  arctan(ZT2 )   ZL.
2

Figure 5.12 and Figure 5.13 show examples of stability regions and IPZ-
tuning. The IPZ-tuning gives parameters rather distant from the boundary

113
Chapter 5. A Tuning Method for IPZ Models
4 4

3 3

2 2
ki

ki
1 1

0 0
0 5 10 15 0 5 10 15
kc kc
Figure 5.12 Stability regions for the IPZ-process controlled by a PI controller. Left:
kv = 0.005, T1 = 50, T2 = 5, L = 3; Right: kv = 0.001, T1 = 200, T2 = 1, L = 2. The points
indicate the position of the IPZ-tuning for the different Ms-values in (5.6). The indications
are moving from the origin to larger kc and ki with larger Ms.

10 10
M s = 1.1 M s = 1.2
8 8

6 6
ki

ki

4 4

2 2

0 0
0 5 10 15 0 5 10 15
kc kc
10 10
M s = 1.3 M s = 1.4
8 8

6 6
ki

ki

4 4

2 2

0 0
0 5 10 15 0 5 10 15
kc kc
Figure 5.13 Stability region for the IPZ-process controlled by a PID controller. The
process is given by kv = 0.005, T1 = 50, T2 = 5, L = 3. The stability region forms a three
dimensional surface in the kcíkiíkd plane. Here, it is visualized in two dimensional plots
by looking at the kcíki plane for the optimal kd (constant in each plot) given by the IPZ-
tuning rule.

114
 Chapter 5. A Tuning Method for IPZ Models
of stability. With higher robustness (lower Ms-values), the tuning moves
closer to the origin. With lower robustness, the proposed tuning moves in
direction to the maximum ki-value in the stability region. This implies that
if we assume a stable controller, there is lower boundary for the
achievable IE, for a given process. Also, note that, if the dead time of the
IPZ-process is zero, the stability region fills the whole first quadrant, and
any and ki is feasible.

5.5 Industrial verification of the tuning rule

To verify the tuning rule it has been tested on a real paper machine, for a
PI controller and different Ms-values. It is the first group in a fluting
machine consisting of five cylinders. The process transfer function has
been obtained from a step response and is given by

P ( s ) 0.0196
51.6s  1 e 1.30 s . (5.15)
s 7.79 s  1

The controller settings are calculated for all the values of the maximum
sensitivity function in (5.6) and the obtained values are given in Table 5.1.
The results are presented in Figure 5.14 and Figure 5.15 by a series of
closed loop step responses.

Table 5.1 The controller settings for (5.15)

Kc Ti
Ms=1.1 0.56 7.1
Ms=1.2 1.0 5.8
Ms=1.3 1.4 5.0
Ms=1.4 1.8 4.5

The step responses in the figures and the table show very well the
difference between various values of the tuning parameter Ms. A higher
value provides a more quick response and tighter control, but the valve
must also work harder. This is essential to remember since an aggressive
control signal can introduce disturbances in adjacent dryer groups through
the steam header and must be considered when tuning a whole drying
section of a paper machine.

115
Chapter 5. A Tuning Method for IPZ Models

Pressure (kPa) 200

195

190

185
0 100 200 300 400 500 600 700 800 900 1000

55
Valve position (%)

50

45

40

35
0 100 200 300 400 500 600 700 800 900 1000
Time (s)

200
Pressure (kPa)

195

190

185
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800

55
Valve position (%)

50

45

40

35
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800
Time (s)
Figure 5.14 Experiment on a real paper machine showing a closed loop step response,
using IPZ-tuning for a PI controller. Ms = 1.1 (above) and Ms = 1.2 (below). Controller
parameters ȕ = 1.

116
 Chapter 5. A Tuning Method for IPZ Models
200

Pressure (kPa)
195

190

185
0 100 200 300 400 500 600 700 800 900 1000

55
Valve position (%)

50

45

40

35
0 100 200 300 400 500 600 700 800 900 1000
Time (s)

200
Pressure (kPa)

195

190

185
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800

55
Valve position (%)

50

45

40

35
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800
Time (s)
Figure 5.15 Experiment on a real paper machine showing a closed loop step response,
using IPZ-tuning for a PI controller. Ms = 1.3 (above) and Ms = 1.4 (below). Controller
parameter ȕ = 1.

117
Chapter 5. A Tuning Method for IPZ Models
Another important issue is the steam load to the boiler house. Large
variations in the paper machines steam demand can put out the recovery
boiler and this has to be considered when tuning the drying section too.
For Ms = 1.4, a high frequency component becomes evident, most
easily seen at t = 1000 í 1400. This is probably due to actuator
nonlinearities or disturbances (as a result of the tuning, some disturbance
frequencies are amplified by 40 %).

5.6 Comparison between PI and PID control

An interesting question is if PID control gives any enhanced performance

compared to PI control, in the case of an IPZ-process. It is a well-known
fact that for a first-order process, PI control is adequate (two closed loop
poles and two controller parameters) and for high order process dynamics,
derivative action can speed up the response [Åström and Hägglund, 2005].
Also, if the dynamics are delay dominated, derivative action gives modest
performance improvements compared to PI control but derivative action
gives significant improvements for processes that are lag dominated.
To evaluate this for the IPZ-process the integral error (IE) is calculated
for a few different processes and Ms-values, when a unit step disturbance
acts on the system, see Table 5.2. The table clearly shows a large
difference in IE when comparing PI and PID control for the same IPZ-
process. The integrated area is more or less reduced by a factor two by
derivative action. What happens is that the derivative-term increases the
phase margin by adding phase lead to the open loop system, making it
possible to increase the integral gain without changing the stability
margin. So it seems that a PID control is superior to PI control in this
sense. But there are a few more other central aspects to regard.
Table 5.2 Controller settings for both PI and PID control, giving minimal IE w.r.t the
maximum sensitivity constraint. Row three is marked in italics to highlight the settings
used for the example in Figure 5.12.
Ms kv T1 T2 L kc Ti kc Ti Td IE PI IE PID
1.2 0.01 400 40 4 0.42 22.9 0.64 15.89 1.88 54.52 24.83
1.2 0.01 100 8 1 1.35 5.0 2.05 3.57 0.48 3.70 1.74
1.2 0.01 50 15 3 1.74 13.7 2.63 10.15 1.37 7.87 3.86
1.2 0.01 25 2 1 1.47 1.9 2.33 1.68 0.45 1.29 0.72
1.4 0.01 400 40 4 0.72 17.0 1.10 10.45 1.97 23.61 9.50
1.4 0.01 100 8 1 2.29 3.8 3.52 2.42 0.50 1.66 0.69
1.4 0.01 50 15 3 2.91 10.5 4.51 6.98 1.45 3.61 1.55
1.4 0.01 25 2 1 2.51 1.7 3.92 1.33 0.49 0.68 0.34

118
 Chapter 5. A Tuning Method for IPZ Models

1.0
y

0.5

0
0 50 100 150 200 250 300 350 400

2
uc

-2
0 50 100 150 200 250 300 350 400
Time (s)
Figure 5.16 Comparison between a PI controller (solid) and a PID controller (dotted)
regulating (5.16). Both controllers are tuned for Ms = 1.2, see Table 5.2. The other
controller parameters are chosen as ȕ = 1, Ȗ = 0, N = 10, see (3.20b). There is a set point
change at t = 10, and a load disturbance starting at t = 250 s.

Apart from the disturbance rejection, set point following is also an

important property for the closed loop system. Figure 5.16 shows both a
set point change and a load disturbance, acting on the process

1  50s  3s
P ( s ) 0.01 e . (5.16)
s (1  15s )

The figure clearly shows that the PID controller puts back the process
value to the set point more quickly than the PI controller during the load
disturbance. The process also reacts faster to a set point change with the
PID controller. However, in both those cases the control signal is slightly
more aggressively used by the PID, which might be a disadvantage due to
cross couplings to adjacent steam groups and increased variations in the
steam demand to the boiler house.
Figure 5.17 shows the gain of the transfer function from r to y and also
the sensitivity function. Figure 5.18 shows the corresponding Nyquist
curves. The PID controller gives a somewhat larger peak in the amplitude
curve for Gry. However, this can be adjusted by the feedforward
parameters ȕ and Ȗ of the PID controller, or by a more general set point

119
Chapter 5. A Tuning Method for IPZ Models
1.5

G ry 1.0

0.5

0 -4 -3 -2 -1 0
10 10 10 10 10

1.5

1.0
G ny

0.5

0 -4 -3 -2 -1 0
10 10 10 10 10
Frequency (Hz)
Figure 5.17 Amplitude diagram for the transfer functions from set point r to process
output y, Gry, and from noise n to process output y, Gny. Gny is equal to the sensitivity
function and Gry is equal to the complementary sensitivity function, if ȕ = 1 and Ȗ = 1. PI
í solid, and PID í dotted. Note that the maximum sensitivity is equal for the to
controllers as the tuning prescribes.

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4

-0.6 -0.6

-0.8 -0.8

-1.0 -1.0
-2.0 -1.5 -1.0 -0.5 0 0.5 -2.0 -1.5 -1.0 -0.5 0 0.5

Figure 5.18 Nyquist curves for the marked controllers in Table 5.2 and the process
in (5.16), PI í left, PID í right. An often occurring difference between the two optimized
Nyquist curves is that for the PID controller the curve touches the Ms-circle in two points
but only once for the PI controller. This can also be seen in the sensitivity function in
Figure 5.17.

filter, so this is actually not an issue of the feedback loop. The sensitivity
function shows that some disturbances are amplified more by the PID
controller. By looking at the spectrum of pressure disturbances in
Figure 2.18, it can be seen that this extra peak in the sensitivity function

120
 Chapter 5. A Tuning Method for IPZ Models
coincides with the variations found in that specific case study. Therefore,
the gain of higher performance by PID control is put in the shade by the
amplification of specific disturbances (the problem should evidently be
resolved by finding the root cause of the disturbances and resolve it, if
possible).
The IPZ-tuning, presented in this chapter, is based on stability of the
closed loop system and disturbance rejection, and does not give any
recommendations on the feedforward settings. However, in Chapter 3 a
two-degree-of-freedom controller structure with set point feedforward is
given that can be combined with the PI or PID controller. Then the
regulation problem and servo problem are separated in a nice way and a
design procedure is given for both.
The conclusion is that PID control of the IPZ-process is more effective
than PI control but it will give a controller with slightly more aggressive
use of the control signal, and also a bit more over-shoot in the set point
response for the standard PID structure. It is therefore difficult to
dogmatically say that the one or the other is better. It simply depends on
how sensitive the steam system is to disturbances and how severe the
cross-couplings between the different steam groups are.

5.7 Comparison to other design methods

This chapter has presented a tuning method for a PI and PID controller
regulating an IPZ-process. Many different tuning methods have been
presented previously, but they mostly suppose a first or second-order
process, see e.g. [Chien, et al, 1952], [Cohen and Coon, 1953], [Zhuang
and Atherton, 1993], [Ho, et al, 1996], and [Skogestad, 2003], or an
integrating process, see e.g. [Poulin and Pomerleau, 1999], [Wang and
Cai, 2001], [Visioli, 2001], and [Chidambaram and Sree, 2003]. Since the
IPZ-process is not a self-regulating process, it is most natural to use the
latter. However, it is not obvious how to match the IPZ-process to a pure
integrator and these methods do not always give a satisfactory result.
Nevertheless, there a few examples, which are modifications of the
Ziegler-Nichols method that give reasonable tuning, e.g. [Tyreus and
Luyben, 1992], and [Poulin and Pomerleau, 1999], but these methods
have no design parameter which is a deficiency.
In this section, a few design methods found in the literature are
examined to see how well they are compared to IPZ-tuning. The purpose
is not to discredit these methods but to demonstrate the need for a new

121
Chapter 5. A Tuning Method for IPZ Models
tuning method for the steam pressure process. All examples are based on
the process model

1  100s  s
P ( s) 0.05 e . (5.17)
s (1  20s )

The simulations of IPZ-tuning use Ms = 1.2 and Ms = 1.3, while the

obtained Ms-value for the other design methods are also given for a
comparison. Remember that these other methods are sometimes based on
a different criterion than the IPZ-tuning. The resulting parameters for the
IPZ-tuning are

Ms = 1.2 PI: kc = 0.67, Ti = 8.1,

Ms = 1.3 PI: kc = 0.92, Ti = 6.4,

(5.18)
Ms = 1.2 PID: kc = 1.0, Ti = 5.3, Td = 0.41,

Ms = 1.3 PID: kc = 1.4, Ti = 3.9, Td = 0.49.

5.7.1 Ziegler-Nichols two tuning methods

The Ziegler-Nichols rules are the most famous tuning rule for the PID
controller [Ziegler and Nichols, 1942]. It was presented in 1942 and
consists of two different methods. One is based on frequency response
data (a closed loop test) and the other one is a step test method (an open
loop test). Both these methods are known to give only reasonable
performance. Nevertheless, they are used here for comparison due to their
widespread use.

The frequency method

The frequency method uses information of one point in the Nyquist curve,
namely the frequency where kcP(iȦ) pass through the point í1. This point
can be found without knowing the transfer function P(s) by a
straightforward closed loop experiment. Simply disable any integral or
derivative action and increase the controller gain until a stable oscillation
is achieved. This gain is called the ultimate gain, k0, and the oscillation
period is T0. The controller settings are then

122
 Chapter 5. A Tuning Method for IPZ Models

1
y

0.5

0
0 50 100 150

2.5
2
1.5
uc

1
0.5
0
0 50 100 150
Time (s)
Figure 5.19 Evaluation of Ziegler-Nichols frequency response method for PI control (solid
line) given the process in (5.17). Also shown is the IPZ-tuning for Ms = 1.2 (dotted) and
Ms = 1.3 (dashed).

1
y

0.5

0
0 50 100 150

2.5
2
1.5
uc

1
0.5
0
0 50 100 150
Time (s)
Figure 5.20 Evaluation of Ziegler-Nichols frequency response method for PID control
(solid line) given the process in (5.17). Also shown is the IPZ-tuning for Ms = 1.2 (dotted)
and Ms = 1.3 (dashed).

123
Chapter 5. A Tuning Method for IPZ Models

y 1

0.5

0
0 50 100 150

1.5

1
uc

0.5

0
0 50 100 150
Time (s)

Figure 5.21 Evaluation of Ziegler-Nichols step response method for PI control (solid line)
given the process in (5.17). Also shown is the IPZ-tuning for Ms = 1.2 (dotted) and
Ms = 1.3 (dashed).

1
y

0.5

0
0 50 100 150

2.5
2
1.5
uc

1
0.5
0
0 50 100 150
Time (s)
Figure 5.22 Evaluation of Ziegler-Nichols step response method for PID control (solid
line) given the process in (5.17). Also shown is the IPZ-tuning for Ms = 1.2 (dotted) and
Ms = 1.3 (dashed).

124
 Chapter 5. A Tuning Method for IPZ Models
T0
PI: kc 0.45k0 , Ti , (5.19)
1 .2

and

T0 T0
PID: kc 0.6k 0 , Ti , Td . (5.20)
2 8

The ultimate gain and oscillation period can also be calculated directly
from the transfer function. Assume the IPZ process and a P controller. The
condition for the point of instability is

1  sT1  sL
kc e 1. (5.21)
s (1  sT2 )

By letting s = iZ, and equating the complex and real parts, the following
controller settings and Ms-values are obtained

PI: kc = 2.87, Ti = 3.28, Ms = 2.7, (5.22)

and

PID: kc = 3.83, Ti = 1.97, Td = 0.49, Ms = 2.55. (5.23)

As mentioned before, the Ziegler-Nichols method is known to give

oscillatory results and it is no surprise that the maximum sensitivity
functions are very high. Figure 5.19 shows simulation results for PI
control and Figure 5.20 shows results for PID control. In both cases, it is
related to IPZ-tuning, even though they are not completely comparable.
This is since the simulated IPZ-tuning has different Ms-values. Figure 5.21
and Figure 5.22 show the corresponding simulations for the step response
method, which is presented next.

The step test method

The step test method is based on two parameters of an open loop step test,
a and L, see Figure 5.23. They are achieved by finding the point of
inflection, which is the point where the response has the maximum
derivative. The tangent through this point gives the two parameters of
interest. The method now suggests the controller settings as

125
Chapter 5. A Tuning Method for IPZ Models

Figure 5.23 The two parameters used in the Ziegler-Nichols step response method.

0.9'uc
PI: kc , Ti 3.33L, (5.24)
aL

and

1 . 2 'u c
PID: kc , Ti 2 L, Td 0 .5 L , (5.25)
aL

where ǻuc is the size of the step. Like in the case of the frequency method,
(5.17) is examined by a simulation. The obtained settings are

PI: kc = 3.6, Ti = 3.33, Ms = 3.8 (5.26)

and

PID: kc = 4.8, Ti = 2, Td = 0.5, Ms = 4.3. (5.27)

The controller settings are somewhat more aggressive than in the

frequency response method. This is also confirmed in the simulations.

126
 Chapter 5. A Tuning Method for IPZ Models
5.7.2 Tyreus-Luyben’s modified ZN tuning rule
The Ziegler-Nichols tuning rules are derived to give decay ratio of ¼. For
many process control system this is too aggressive, which led [Tyreus and
Luyben, 1992] to derive a more conservative PI tuning rule. The method
is obtained by maximizing the closed loop resonant frequency given a
maximum complementary sensitivity function of 2 dB ( § 1.26 ). The
result is given by

k0
PI: kc , Ti 2.2T0 , (5.28)
3.22

where k0 and T0 is defined as in Section 5.7.1. This gives the following

values for the process in (5.17)

PI: kc = 1.97, Ti = 8.66, Ms = 1.65 (5.29)

Figure 5.24 shows the simulation results and the method gives a
reasonable good tuning. However, the maximum complementary
sensitivity function is different (in this case 1.16) from the value
prescribed by the method, since it is derived for a different process. Both
the set point response and disturbance rejection is faster than the IPZ-
tuning, but it is also less robust (larger Ms) and the control signal is much
more aggressive. As a comparison, Figure 5.25 shows IPZ-tuning for
Ms = 1.65. The disturbance rejection is then better for IPZ-tuning than
Tyreus-Luyben’s tuning rule. The time to reach a new set point is similar
for the two methods even though the size of the over-shoot is different.
Remember that the over-shoot should be dealt with the feedforward part
of the controller, if necessary.
As for the case of Ziegler-Nichols, the Tyreus-Luyben method lacks a
tuning parameter.

5.7.3 The AMIGO tuning rule

The AMIGO-rule [Åström and Hägglund, 2005] is developed in a similar
way as the IPZ-rule, with Ms = 1.4. However, a test batch of nine different
process structures have been used (IPZ not included), and the method
presumes that the process can be approximated by either a first-order
system or an integrator. Given the system

k  sL
P( s ) e , (5.30)
s
127
Chapter 5. A Tuning Method for IPZ Models

1.0
y

0.5

0
0 20 40 60 80 100 120 140 160 180 200

2
uc

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)

Figure 5.24 Evaluation of Tyreus-Luyben’s method for PI control (solid line) given the
process in (5.17). Also shown is the IPZ-tuning for Ms = 1.2 (dotted) and Ms = 1.3
(dashed).
1.5

1.0
y

0.5

0
0 20 40 60 80 100 120 140 160 180 200

2
uc

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)

Figure 5.25 Evaluation of Tyreus-Luyben’s method for PI control (solid line) given the
process in (5.17). Also shown is the IPZ-tuning for Ms = 1.65 (dotted) (kc = 1.57,
Ti = 4.09).

128
 Chapter 5. A Tuning Method for IPZ Models

1
y

0.5

0
0 20 40 60 80 100 120 140 160 180 200

2
uc

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)
Figure 5.26 Evaluation of the AMIGO tuning rule for PI control (solid line) given the
process in (5.17). Also shown is the IPZ-tuning for Ms = 1.2 (dotted) and Ms = 1.3
(dashed).

1
y

0.5

0
0 50 100 150

2.5
2
1.5
uc

1
0.5
0
0 50 100 150
Time (s)
Figure 5.27 Evaluation of the AMIGO tuning rule for PID control (solid line) given the
process in (5.17). Also shown is the IPZ-tuning for Ms = 1.2 (dotted) and Ms = 1.3
(dashed).

129
Chapter 5. A Tuning Method for IPZ Models
the proposed tuning rule is

0.35
PI: kc , Ti 13.4 L, (5.31)
kL

0.45
PID: kc , Ti 8L, Td 0.5L. (5.32)
kL

The next step is to match the IPZ-process to a pure integrator with time
delay. By letting k be equal to the maximum slope of a unit step response
of the IPZ-process, (3.5) gives

k vT1
k . (5.33)
T2

The time delay, L, is equal to the time delay of the true process. Assuming
(5.17), the controller settings and obtained maximum sensitivity are

PI: kc = 1.4, Ti = 13.4, Ms = 1.4, (5.34)

and

PID: kc = 1.8, Ti = 8, Td = 0.5, Ms = 1.4. (5.35)

Figure 5.26 and Figure 5.27 show the simulations. We observe the
AMIGO rule gives a pleasant performance for this process. There is less
over-shoot in the set point response, compared to the IPZ-tuning and the
controllers also bring back the process output to the set point nicely at a
load disturbance.
However, if we instead let the process be given by

1  100s  3s
P ( s ) 0.01 e (5.36)
s (1  10 s )

we get the tuning

PI: kc = 1.17, Ti = 40.2, Ms = 1.3, (5.37)

130
 Chapter 5. A Tuning Method for IPZ Models

1.0
y

0.5

0
0 50 100 150 200 250 300

1.5

1.0
uc

0.5

0
0 50 100 150 200 250 300
Time (s)

Figure 5.28 Evaluation of the AMIGO tuning rule for PI control (solid line) given the
process in (5.36). Also shown is the IPZ-tuning for Ms = 1.3 (dashed).

and

PID: kc = 1.5, Ti = 24, Td = 1.5, Ms = 1.4. (5.38)

Figure 5.28 shows the simulation for (5.37) compared with one case of
IPZ-tuning. The AMIGO tuning does not give a satisfactory result at all
for this process. Note that the two controllers in the figure have the same
robustness measure but still very different performance.

5.7.4 Pole placement

The idea of pole placement is to find a controller that gives a closed-loop
system with a specified characteristic polynomial. In Chapter 3, this was
introduced by state-feedback and all closed loop poles could be placed
arbitrarily. This is not always possible when the process is controlled by a
PI or PID controller. In the general case, a PI controller can give the
characteristic polynomial arbitrary values for a first-order process, and a
PID controller handles the same thing for a second-order process. It is
clear that it is possible to find a good controller with such a general tuning
method as pole placement. The difficulty can be to know where to place
the poles to obtain a satisfactory feedback loop. The design method is
shown by a few examples.

131
Chapter 5. A Tuning Method for IPZ Models
Nelson-Gardner’s pole placement rule for PI control
In [Nelson and Gardner, 1996] a tuning rule for the IPZ process is given,
that is derived from simple pole placement. It can also be found in one of
the exercises in [Sell, 1995], written by Bill Bialkowski, but not as
detailed. In both references the idea is given in words without carrying out
the calculations and the time-delay is disregarded. Let the process be
given by

(1  sT1 )  sL (1  sT1 )(1  sL)

P(s) k v e | kv , (5.39)
s (1  sT2 ) s (1  sT2 )

where the time delay have been approximated by the first-order Taylor
series. The controller is given by

1  sTi
C ( s) k c . (5.40)
sTi

Now, let Ti = T2. This implies that the process pole is canceled by the zero
of the controller. This can be compared with the IPZ-tuning rule, where
the integral time follows the parameter T2 in the sense that larger values of
T2 gives larger values of Ti, and for small T2 the two are almost equal, see
Figure 5.5. The closed loop system is now given by

kv kc (1  sT1 )(1  sL)

Gry ( s ) . (5.41)
(T2  kv kcT1L) s 2  (kv kcT1  kv kc L) s  kv kc

The two zeros of the denominators are

kv kc (T1  L) r (kv kc ( L  T1 )) 2  4kv kcT2

s  . (5.42)
2(T2  kv kcT1L)

By the controller gain kc we have some liberty to choose the position of

the closed loop poles. By choosing a double pole, the contribution from
the square root is zero, and the equation

(k v k c ( L  T1 )) 2  4k v k c T2 0 (5.43)

132
 Chapter 5. A Tuning Method for IPZ Models

1
y

0.5

0
0 20 40 60 80 100 120 140 160 180 200

2
uc

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)

Figure 5.29 Evaluation of Nelson-Gardner’s pole placement rule (5.44) for PI control
(solid line) given the process in (5.17). Also shown is the IPZ-tuning for Ms = 1.2 (dotted)
and Ms = 1.3 (dashed).

is fulfilled. This gives us the tuning rule

4T2
kc , Ti T2 . (5.44)
k v ( L  T1 ) 2

and the double pole is then positioned in

2
 (5.45)
T1  L

Assuming the process given in (5.17), the controller settings are

PI: kc = 0.16, Ti = 20, Ms = 1.04, (5.46)

The simulation result is shown in Figure 5.29 and we can immediately see
that this is not a satisfying controller. We can also see that it is a rather
robust controller, since the Ms-value is fairly small. Further investigations
show that this method gives a quite robust controller with Ms-values

133
Chapter 5. A Tuning Method for IPZ Models
below 1.05 for short time delays (L § 1). Larger time delays or shorter T2
give Ms § 1.1í1.2, and relatively fine simulation results. The disadvantage
with this method is the lack of any design parameter but it gives a
reasonable tuning.

PID control
The poles of a second order process with one zero, can be placed
independently of one another by a PID controller. By assuming that the
time delay can be neglected, the IPZ-process fits into this model structure.
Then the process is given by

(1  sT1 )
P( s) kv , (5.47)
s (1  sT2 )

and the controller is

1  Ti s  Ti Td s 2
C ( s) k c . (5.48)
sTi

The closed loop system then becomes

Gry ( s )
kv kc (1  Ti s  TiTd s 2 )(1  T1s )
.
Ti (kv kcT1Td  T2 ) s 3  Ti (1  kv kcT1  kv kcTd ) s 2  kv kc (T1  Ti ) s  kv kc
(5.49)

A suitable characteristic polynomial for a third-order system is

( s  Z1 )( s 2  2]Z 0 s  Z 02 ) (5.50)

By equating coefficients of equal power in s, we get the following system

of equations

134
 Chapter 5. A Tuning Method for IPZ Models

­ kv kc
° Z1Z 02
° k v k c T1Ti Td  T2 Ti
°° k v k c (T1  Ti )
® 2Z 0 Z1]  Z 02 (5.51)
k k T
° v c 1 i dT T  T T
2 i
° Ti (1  k v k c T1  k v k c Td )
° 2Z 0 ]  Z1
°¯ k v k c T1Ti Td  T2 Ti

The solution is given by

Z 0 (2T1Z1]  T1T2Z 0Z1  T1Z 0  2T2Z1]  T2Z 0  T12Z 0Z1 )

kc ,
k v (T13Z1Z 02  T1Z1  2T1Z 0]  2Z 0Z1]T12  T12Z 02  1)
2Z1]  Z 0  T1Z 0Z1
Ti , (5.52)
Z 0Z1
T12 T2Z 02 Z1  2T1T2Z 0 Z1]  T1T2Z 02  T2Z1  2T2Z 0 ]  1
Td .
Z 0 (2T2Z1]  T2Z 0  T1T2Z 0Z1  2T1Z1]  T1Z 0  T12Z 0Z1 )

Even though any Z0, Z1, and ȗ can be realized there should be some
relation between the process and the desired closed loop poles, to give a
controller with satisfactory performance. E.g. requesting a bandwidth that
is too high for the closed loop system gives very high controller gain. This
injects much noise into the loop, can be damaging to actuators, and might
make the system unstable due to neglected process dynamics.
From (5.52) the controller settings can easily be calculated for a given
closed loop response. As an example, we select to place the poles in

3 3 3
Z0 , Z1 , ] . (5.53)
2T1 4T2 4

The solution is then

9T2 (T1  T2 ) 4 T12

kc , Ti T2 , Td . (5.54)
k v T12 (3T1  4T2 ) 3 9(T1  T2 )

135
Chapter 5. A Tuning Method for IPZ Models
21 21
20 20
Steam flow (kg/s)

Steam flow (kg/s)

19 19
18 18
17 17
16 16
15 15
0 1000 2000 3000 0 1000 2000 3000
Time (s) Time (s)
Figure 5.30 A case study í before (left) and after (right) an aggressive retuning of the
steam pressure controllers in a drying section. Both figures show the total steam usage by
the drying section. To the left is before retuning of the steam pressure controllers and to
the right is after. The retuning gave both better set point following and disturbance
rejection, but it had severe implications for the steam producers who could not handle the
large variations in demand.

If this is a good tuning depends on how well we can neglect the time
delay, but also on the other process parameters. Large T1, for example,
tends gives very small kc.

Remark
Trying to place any of the closed loop poles in í1/T1 gives infinite
controller gain. This matches what was observed with PI control.
Controller parameters Ti and Td will have finite values though.

5.7.5 Some concluding remarks

In this section, a few different tuning methods have been evaluated to see
how well they match the IPZ-tuning rule. A few of them give a reasonable
tuning in some region of process parameters but only modest performance
outside this region. The reason is that they do not fit well to the IPZ-
process, since they often are derived from another process structure. It is
always important to not only look at the process output but also the
control signal. Ziegler-Nichols tuning rule gives a very fast response but
also large variations in the control signal, which is particularly undesirable
for the steam distribution system, see Figure 5.30. How large variations
that are acceptable is a case-by-case matter, and it is therefore vital to
have a design parameter to adjust the tradeoff between robustness and
performance. Many other methods lack this design parameter. All this is
the motivation for a specific tuning rule for the IPZ process.

136
 Chapter 5. A Tuning Method for IPZ Models
2.5

2.0

1.5
ki

ZN step
1.0
ZN freq

0.5

IPZ Ź TL
ƅ AMIGO
0
0 1 2 3 4 5 6 7
kc

Figure 5.31 Stability region for the process in (5.17) controlled by a PI controller. The
different tuning rules examined in this section are also indicated, ż Ziegler-Nichols
frequency method, + Ziegler-Nichols step test method, Ź Tyreus-Luyben’s modified rule,
ƅ the AMIGO tuning rule, Ƒ Nelson-Gardner, and Ɣ IPZ-tuning (Ms = 1.1í1.4). The
dashed line shows the contour that the IPZ-tuning follows. For large Ms values, the tuning
method is moving towards the peak of the stability region, namely the maximum ki.

To see the difference between the evaluated tuning methods from another
perspective, the proposed controller settings by the different methods for
PI control of (5.17) are indicated in the stability region for the process, see
Figure 5.31. The two Ziegler-Nichols methods distinguish themselves by
grouping together away from the other methods. All four settings from the
IPZ-tuning are shown in the figure. It was previously concluded that the
AMIGO tuning rule and Tyreus-Luyben’s rule give satisfactory controller
settings for this set of process parameters. Therefore, it seems reasonable
that their controller settings are in the vicinity of the IPZ-tuning. The
example of Nelson-Gardner’s pole placement is also in that neighborhood
but closer to the origin, which gives a much more robust control.
Remember that it is only the IPZ tuning among the evaluated methods that
have a design parameter and that give good performance for a wide range
of different parameters of the IPZ model.
Finally, a robustness test is given in Figure 5.32, showing a simulation
of the different methods tuned for (5.17) but where the time delay is

137
Chapter 5. A Tuning Method for IPZ Models
7 12
x 10 x 10
2 2
ZN frequency ZN step
0
y 0 y
-2

-2 -4
0 50 100 150 200 0 50 100 150 200
2 2
TL AMIGO

y 1 y
1

0 0
0 50 100 150 200 0 50 100 150 200
2 2
NG IPZ Ms = 1.3
1
y y
1
0

-1
0
0 50 100 150 200 0 100 15050 200
Time (s) Time (s)
Figure 5.32 Robustness analysis of the tuning methods used in comparison with the IPZ
tuning. The time delay in (5.17) is changed from 1 to 2 seconds.

multiplied by a factor of two. Both Ziegler-Nichols methods become

unstable and Tyreus-Luyben is close to the point of instability. This
clearly shows the necessity to not only consider performance but also
robustness. Overall, the IPZ tuning appears to give the best result when
weighing together all the comparisons given in this section.

5.8 Summary

In this chapter, a tuning method for PI and PID control of the IPZ model
has been presented. It is based on optimized load disturbance rejection
subject to a robustness constraint. To calculate the controller parameters,
the method requires the four process parameters of the IPZ-model and a
design parameter that is defined by the user. The process parameters can
easily be obtained from an open loop step response. Since the dead time
has a significant effect on the method, good quality dead time estimation

138
 Chapter 5. A Tuning Method for IPZ Models
is essential. The advantage with the IPZ-tuning is that the equations are
simple and it is easy to use. From experiments on several paper machines
at different mills, the method has been validated and proved to perform
well.
A few other tuning methods found in the literature have been
compared to the proposed tuning method. The conclusion is that there is a
need for a new method, since neither of the other give good result when
they are evaluated for different process parameters.
Since the method is based on optimized load disturbance rejection it
can with advantage be combined with a set point filter. Examples of this
are the ȕ and Ȗ factors in (3.19), or the more general 2DOF-controller
described in Section 3.3.
The IPZ structure can be used in many other process areas, apart from
modeling the steam pressure in cylinders [Forsman, 2005]. Therefore the
tuning method has a much wider field of applications than solely drying
section control.

139
Chapter 5. A Tuning Method for IPZ Models

140
Part Modeling and control
of paper moisture in
2 the drying section

Papermaking in the early 1900s, painted by Thomas M. Dietrich. It shows

the dry end of the paper machine at the Fox River Paper mill in Appleton,
Wisconsin. The man is checking the feel and transparency of the sheet for
its thickness and formation. The paper has moved from right to left in this
painting, passing through the dryers, calender stack, and reel. By courtesy
of Fox Valley Corporation.
 Chapter 6. Enhanced Moisture Control Using the Air System

6
6.

Enhanced Moisture Control

Using the Air System

The drying section is enclosed inside a drying hood. The main purposes of
the hood are to create a controlled environment for the drying process,
improve energy utilization, and also to establish good working conditions
in the machine room. The exhaust air removes the evaporated water from
the paper web while preheated dry air is added to the hood by the supply
air. Traditionally, these two variables are used to control the humidity and
pressure inside the hood. This chapter proposes using the supply air, in
combination with the steam cylinders, to control the moisture in the sheet.
Trials have shown that the supply air has a fairly large and fast impact on
moisture. This suggests that using the supply air would give a closed loop
system with higher performance compared to conventional steam pressure
control. However, the drying capacity of the steam cylinder process is
much larger than the drying capacity of the air process and both processes
are therefore essential to obtain a well functioning control system. The
two manipulated variables are combined in a mid-ranging structure to
control the moisture.
The proposed control structure is evaluated by simulations of a
physical model of the drying section.

143
Chapter 6. Enhanced Moisture Control Using the Air System
6.1 A literature review of drying section models

Mathematical modeling of cylinder drying started with the pioneer work

by [Nissan and Kaye, 1955]. An extensive review of drying models up to
1980 with some 130 references is given in [McConnell, 1980]. Many
drying models with different approximations and objectives have been
proposed in the last decade. In [Ramesh, 1991], [Wilhelmsson, 1995], and
[Reardon, et al, 2000] models based on non-linear steady-state relations
are given. [Berrada, et al, 1997] develop a linear state-space model from
physical relations. In [Rao, et al, 1994] a simplified dynamic model is
developed, where the whole drying section is modeled as one or only a
few large cylinders. Some physical properties are then adjusted to fit the
assumption. A combination of statistical multivariate models and physical
models can be found in [Viitamäki, 2004]. In [Chen, 1995], [Menani, et
al, 1998], [Skoglund, et al, 2000], and [Sun, et al, 2000] different kinds of
black-box models are used. In [Sadeghi, 2003] and [Sadeghi and Douglas,
2004] a model that includes different transport phenomena within the
paper sheet is presented but only in steady-state. Both [Mori, et al, 2000]
and [Gaillemard and Johansson, 2004] give a physical model but omit the
steam system. [Perré, et al, 2004] shows a simulation model for drying of
coated paper that is implemented in Excel, but neglects the steam system
and [Videau and Lemaitre, 1982] develops a static simulator used for
analysis of paper production and energy consumption.
This chapter describes a physical simulation model of a drying section,
implemented in Simulink. The model contains approximately 105 states
and is capable of dynamically describing moisture gradients and other
properties inside the paper sheet. This model is used for simulations in
this chapter and also in Chapter 7.

6.2 The model

The model used for simulation and analysis in this chapter and Chapter 7
is developed at the Department of Chemical Engineering, Lund
University, see [Karlsson and Stenström, 2005a] and [Karlsson and
Stenström, 2005b]. In this thesis, the model is not described in detail but a
general overview is given. It is adapted to a board machine at a paper mill
in Sweden. It consists of 93 steam cylinders divided in 12 groups. It is
assumed that the machine speed is 430 m/min and the dry basis weight is
267 g/m2.

144
 Chapter 6. Enhanced Moisture Control Using the Air System

One cylinder

Output data
Input data

Time Time
Figure 6.1 Illustration of the simulation method, where one cylinder with a following free
draw is simulated at a time.

The model is built on basic physical relations, in terms of mass and energy
balances, and algebraic equations. The continuity equation forms a basis
for all the balances. Written as a partial differential equation (PDE) it is
given by

w)
’ ˜ J  k , (6.1)
wt

where ĭ is some extensive system property (properties that are strictly

additive, normally mass or energy), J is flow and k is net consumption or
generation. It states that the amount of property entering a infinitesimal
volume element either leaves that volume, accumulates within it, or is
consumed, see [Hangos and Cameron, 2001] or [Sparr and Sparr, 2000].
Algebraic relations describing transport for mass and energy in
combination with (6.1), gives balances for water, vapor, air, and fibre in
the web. This means that properties like gas pressure inside the paper web,
different transport mechanisms, and moisture gradients and shrinkage in
the thickness direction can be simulated. The cylinder shell and the
surrounding air are also included in the model, but not the dynamics for
the steam inside the cylinder. The model is calibrated to steady-state
measurements by adjusting two parameters, heat transfer coefficients for
the contact between the cylinder and the paper, and for the condensate
inside the cylinder.
The set of PDEs are converted into a set of ordinary differential
equations (ODEs) by discretization. There are 10 nodes in the thickness
direction and 2005 nodes in the machine direction. The model is
implemented in Simulink with the ODEs written in C-code to reduce the
simulation time. To reduce the requirements for internal memory storage
in the computer, one cylinder at a time is simulated, see Figure 6.1. The
result from a simulation of one cylinder is then used as input data for the

145
Chapter 6. Enhanced Moisture Control Using the Air System

pressure setpoint pressure

Ȉ PID IPZ

Figure 6.2 The model used to simulate the closed loop steam pressure dynamics.

simulation of the next cylinder, by following the flow of the paper. In this
way, the number of nodes in each simulation is greatly reduced, and the
memory requirement is reduced from gigabytes to megabytes. Simulations
shown in this chapter still take 1í2 days on a standard computer. The
simulation technique will have some implications for simulation of the
closed loop system, as it will be shown later.

Modeling of the steam pressure dynamics

Since the physical model does not include the steam pressure dynamics
inside the cylinders, the IPZ model given in Chapter 3 is used. It is
combined with a PID controller and a valve saturation to give a model of
the closed loop steam pressure system, see Figure 6.2. From a step
response in each group in the drying section of the real machine,
parameters of an IPZ model are estimated, see Figure 6.3 and Figure 6.4.
By including the PID parameters, used by the mill for each pressure
controller, a simulation model for each group is obtained that is linked to
the physical model described previously. This gives a complete simulation
model of the drying section, including both steam system, cylinder shell
and paper sheet.

Modeling of the air system

The air system is of special interest in this chapter and is therefore further
discussed. Figure 6.5 shows an example of how the blow boxes can be
configured in a drying section. A blow box is the unit through which the
main part of the supply air is distributed. The other part comes from
leakage air through the hood, which is roughly 20í30 % of the total air
flow in modern machines [Karlsson, 2000]. Apart from bringing dry
supply air close to the sheet, the blow box also improves runnability by
reducing sheet flutter. The areas where the most of the evaporation from
the sheet occurs are indicated in the figure. To model this it is assumed
that there is a specific volume of air around each cylinder that is involved

146
 Chapter 6. Enhanced Moisture Control Using the Air System
0.00178(97.37s+1) - 1.0 s Range y: (-100 - 450) 0.00350(44.69s+1) - 1.00 s Range y: (-100 - 450)
Steam group 1 G(s) = e Range u: (0 - 100) Steam group 2 G(s) = e Range u: (0 - 100)
s(7.97s+1) s(2.61s+1)
18 Loss function = 0.31596 96 Loss function = 0.38311

16 94

y 92
14 y

90
12
88
10
86
60 70
u 58 u 68
56 66
500 550 600 650 700 750 800 850 900 0 50 100 150 200 250 300
Time (s) Time (s)

0.00158(121.49s+1) - 1.0 s Range y: (-100-450) 0.00345(44.98s+1) - 1.0 s Range y: (-100 - 450)

Steam group 3 G(s) = e Range u: (0-100) Steam group 4 G(s) = e Range u: (0 - 100)
s(6.26s+1) s(3.70s+1)
Loss function=0.41249 Loss function = 0.21842
168
62

60 166

58 164
y 56 y
162
54

52 160

50 158

64 50
u 62 48
u
60 46
58 44
400 450 500 550 600 650 700 750 0 50 100 150 200 250 300 350
Time (s)
Time (s)

0.00112(55.60s+1) - 2.0 s Range y: (-100-450) 0.00228(90.10s+1) - 1.00 s Range y: (-100 - 450)

Steam group 5 G(s) = e Range u: (0-100) Steam group 6 G(s) = e Range u: (0 - 100)
s(14.01s+1) s(14.93s+1)
Loss function=0.20334 Loss function = 0.38462
98 282
97
278
96
274
y 95 y
94 270

93 266
92
262
91
50
48
u u 45
44
40 40
450 500 550 600 650 700 750 0 100 200 300 400 500
Time (s) Time (s)

0.00270(40.31s+1) - 1.0 s Range y: (-100 - 450) 0.00369(78.64s+1) - 1.96 s Range y: (-100 - 450)
Steam group 7 G(s) = e Range u: (0 - 100) Steam group 8 G(s) = e Range u: (0 - 100)
s(6.86s+1) s(13.23s+1)
Loss function = 0.23028 Loss function = 0.21070
226
198
224
196 222

y y 220
194
218

192 216

214
190
28
64
u u 26
62
60 24
700 800 900 1000 1100 1200 100 150 200 250 300 350 400 450
Time (s) Time (s)

Figure 6.3 Graphical output from the modeling tool [Wallén, 2000] used to identify the
steam pressure dynamics included in the simulation model. The solid lines are process
signals and dotted lines the obtained models (also given as transfer functions). The unit of y
is kPa (gauge) and the unit of uc is %. The figure shows group 1í8.

147
Chapter 6. Enhanced Moisture Control Using the Air System
0.00324(75.96s+1) - 2.42 s Range y: (-100-450) 0.00362(74.83s+1) - 1.89 s Range y: (-100 - 450)
Steam group 9 G(s) = e Range u: (0-100) Steam group 10 G(s) = e Range u: (0 - 100)
s(21.23s+1) s(15.73s+1)
Loss function=0.16867 Loss function = 0.30897
258

250
254

246
250
y y
242
246

238
242

234
238
42 42
u 38 u 38

34
34
50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800
Time (s) Time (s)

0.00139(110.35s+1) - 1.43 s Range y: (-100 - 450) 0.00138(108.94s+1) - 1.70 s Range y: (-100 - 450)
Steam group 11 G(s) = e Range u: (0 - 100) Steam group 12 G(s)= e Range u: (0 - 100)
s(24.95s+1) Loss function = 0.22603 s(24.26s+1) Loss function = 0.16917
326

324 342

322
340
y 320 y
338
318

316 336

314
334
58 52
u 54 u
48
50 44
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700
Time (s) Time (s)

Figure 6.4 Graphical output from the modeling tool [Wallén, 2000] used to identify the
steam pressure dynamics included in the simulation model. The solid lines are process
signals and dotted lines the obtained models (also given as transfer functions). The unit of y
is kPa (gauge) and the unit of uc is %. The figure shows group 9í12.

Figure 6.5 An example of blow box configuration in both a single-tier (above) and two-
tier (below) machine. The oval areas indicate the region where most evaporation occurs.
The active air volume is essentially bounded by machine equipment and fabrics. The large
arrow indicates the direction of the paper and a few blow boxes are indicated by smaller
arrows. By courtesy of Metso Paper.

148
 Chapter 6. Enhanced Moisture Control Using the Air System

Evaporation
Exhaust Air
System
boundary

Cylinder
Wet Air

Paper
Dry Air

Supply
Air Actuator
Figure 6.6 A picture of the air system used in the model.

in the dynamics, see Figure 6.6, and the volume was set to 10 m3. With a
nominal total supply air flow of 50 m3/s divided to 93 cylinders, this
corresponds to a residence time of 18.6 s. Dry air is mixed with
recirculated moist air and in this way, the dew point of the supply air is
manipulated while the flow rate is constant. The idea is to prevent
runnability problems due to web flutter in case of aggressive use of the
supply air.

6.3 A prestudy

To investigate the potential to use the supply air as an actuator, a step test
has been performed on a board machine, see Figure 6.7. The machine
speed is 724 m/min and the basis weight is 134 g/m2. The step test is
performed on the last third of the drying section (the after dryer), while
both the moisture and basis weight controller are put in manual mode (the
steam pressure controllers are still in automatic). The air flows are
manipulated by variable speed fan motors (some machines have air flow
dampers instead). Measurements prior to the experiment showed that
67 % control signal corresponds to 41.4 m3/s supply air and 70.3 m3/s
exhaust air. Both the supply air and exhaust air are changed equally
during the step test to maintain the air balance in the hood. Because of the
instrumentation at the particular mill, it is not possible to automatically
log the signal to the fans, therefore the control signal shown in the figure
has been recreated afterwards from notes taken during the experiment. In
addition, the change in supply air is done 5í10 s after the change in
exhaust air, since it is done manually by one person. The moisture and

149
Chapter 6. Enhanced Moisture Control Using the Air System

Sheet moisture (%) 9

65
exhaust air ( C)
o
Dew point

60

55

50

100
supply air (%)
Control signal

50

0
0 200
600 400 800 1000 1200
Time (s)
Figure 6.7 Results of the prestudy where it is shown that the supply air has a large impact
on the sheet moisture, measured at the reel-up. The dew point of the exhaust air is also
shown.

dew point are logged in two different systems and are synchronized at a
later stage by the time tags from the internal system clocks. Therefore, it
is considered somewhat uncertain to make a dynamic model from the
experiment result, but it still clearly gives an idea of the potential to use
the air system as an actuator for the moisture controller.
The experiment gives a promising result. The response is distinct, the
change in sheet moisture is rather significant and the time constant is
somewhere between 15í25 s if it is modeled as a first order process, see
Figure 6.7. This also agrees well with the residence time of the air system
in the model described in Section 6.2.

6.4 Mid-ranging

As described previously, the steam and air system is combined by mid-

ranging control. Generally, mid-ranging refers to control problems where
there is one process output and two or more manipulated inputs, see

150
 Chapter 6. Enhanced Moisture Control Using the Air System

Pulp tower

Water
QC

Figure 6.8 The consistency in the pulp tower is typically 8í12 %. Dilution to pumpable
suspension takes place at the bottom of the tower by injection of dilution water. An
agitator helps the mixing.

[Shinskey, 1978], [Allison and Isaksson, 1998] and [Allison and Ogawa,
2003]. A common example in the pulp and paper industry is consistency
control of the pulp that is pumped from the storage tower, see Figure 6.8.
The controller has two actuators, a large dilution valve and a smaller valve
for fine adjustments. The main purpose of the water through the large
valve is to make the pulp pumpable. Nevertheless, it is also important to
avoid letting too much water being injected at the bottom of the tower to
prevent the small valve from being closed, and vice versa, since the small
valve often has a smaller operating range and finer resolution. One
solution is to let the small valve return to its midpoint or target value in
steady state. This is known as mid-ranging. In general, the effect of the
inputs differs significantly in both range and speed, and sometimes in
cost. When there is a cost tied to the inputs, the optimal target value for
the expensive input is often not the midpoint.

Four types of mid-ranging structures

There exist several different alternatives to implement a mid-ranging
controller, and four different structures are described below. It is assumed
that P1 is a fast process with a small range, and P2 is a slow process with a
large range. The signals u1 and u2 are control signals for P1 and P2,
respectively, and r1 is set point for y and r2 set point for u1. The last three
mid-ranging structures are evaluated in [Allison and Isaksson, 1998]. The

151
Chapter 6. Enhanced Moisture Control Using the Air System
u2
C(s) P2(s)
r2
r1 u1 y
Ȉ kc Ȉ P1(s) Ȉ

1
Figure 6.9 The structure of the simple mid-ranging controller. The set point r2 is the bias of
the P-controller. In practice, C(s) is often chosen as a PI controller

conclusion is that the MPC solution is superior compared to the other two,
and the valve position controller comes in second hand.

Simple Mid-ranging
This is probably the simplest way to implement mid-ranging and it
consists of only one feedback loop but two controllers in parallel with the
same set point, see Figure 6.9. The controller manipulating u1 is a
P controller, whereas the controller manipulating u2 is a controller with
integral action. When the integral action in C removes the control error,
the output u1 is the offset of the P controller. By adjusting the offset, the
target value for u1 is set. The disadvantage with this solution is that P1 is
limited to be controlled by a P controller. This gives a slow response to
both set point changes and disturbance rejections, since the P controller
gives a steady-state error which is removed by C through a much slower
process P2.

Valve Position Controller

The valve position controller in Figure 6.10 is probably the most common
implementation of mid-ranging found in industry today, see [Shinskey,
1978] and [Allison and Ogawa, 2003]. Controller C1 controls the output y
with input u1, while C2 controls u1 with u2. In the literature describing the
valve position controller, the decoupling filter CD is seldom included. An
exception is [Allison and Isaksson, 1998], where it is briefly discussed
without any details. The filter is not necessary for the function of the
control system but it improves the performance. If the decoupling filter is
removed, the only way for C2 to control u1 is through the error r1 í y,
consequently introducing disturbances into y. The decoupling should
ideally be chosen as

152
 Chapter 6. Enhanced Moisture Control Using the Air System
r2 u2
Ȉ C2(s) P2(s)

CD(s)
1
r1 y
Ȉ C1(s) Ȉ
u1 P1(s) Ȉ

1
Figure 6.10 The structure of the valve position controller.

P2 ( s )
CD ( s)  , (6.2)
P1 ( s )

which is realizable as long as the time delay of P1 is shorter or equal to the

time delay of P2, and P1 and P2 have the same unstable zeros. In addition,
the pole excess of P2 should be at least as large as the pole excess of P1. It
is assumed that both P1 and P2 have stable poles or integrators. With (6.2),
changes in u2 do not affect y, and the transfer function from u2 to r2 í u1 is
equal to íCD. If the ideal decoupling is not realizable, approximations
have to be used.
This implementation does not require anti-windup protection in C1
since u1 is a controlled variable. Regardless if u1 saturates or not during an
upset, controller C2 brings u1 back to r2 in steady-state. However,
depending on the speed of controller C2, anti-windup protection in C1
might be beneficial, see [Haugwitz, et al, 2005].

Hybrid Mid-ranging
An approach, originally used to control pressure of a steam header, is the
hybrid mid-ranging [Love, 1994]. It consists of one PI controller, two
saturations, a low pass filter, and three gains, see Figure 6.11. The
parameter Rm, should be chosen as

100k1
Rm , (6.3)
k1  k 2

where k1 and k2 are the steady-state gains of P1 and P2, respectively. The
filter in combination with the direct term creates the mid-ranging function.
The transfer function from the output of the PI controller to the
153
Chapter 6. Enhanced Moisture Control Using the Air System
r2
r1 100
Ȉ PI Filter 1 Ȉ
Rm
Ȉ P1(s)
u1
Rm
100
1
100 u2 y
Ȉ
100 Rm
P2(s) Ȉ

1
Figure 6.11 The structure of the hybrid mid-ranging controller.

summation after the filter is in effect a low pass filtered derivative. The
purpose of the saturation block preceding the filter is to allow an offset in
u1 when u2 is reaching its limit. By this, the controller combines mid-
ranging (normal operation) with split-range (saturated u2) in a nice way.
The purpose of the other saturation block is to immediately engage control
signal u2 when u1 reaches its limit.
The PI controller can, without losing the mid-ranging function, be
replaced by any SISO controller.

Mid-range MPC
In [Allison and Isaksson, 1998] a mid-ranging controller is implemented
in the MPC (model predictive control) approach, see Figure 6.12. By
letting the minimized cost function be given by

H p 1 2 H u 1 2
ª r1 (k  i )  yˆ (k  i | k ) º ª 'u1 (k  i ) º
J (k ) ¦
i 0
« r (k  i)  u (k  i) »
¬ 2 1 ¼
 ¦
i 0
« 'u (k  i ) » ,
¬ 2 ¼ R
(6.4)
Q

and the weighting matrices chosen as

ªlarge 0 º ªsmall 0 º
Q « 0 , R , (6.5)
¬ small»¼ « 0
¬ large»¼

154
 Chapter 6. Enhanced Moisture Control Using the Air System

r1 u1
P1(s)
r2
MPC
u2 y
P2(s) Ȉ

Figure 6.12 The structure of the mid-range MPC.

the desired behavior is achieved. Since the MPC formulation is inherently

discrete-time, the cost function is given as a summation. Notation
ǔ(k + i | k) denotes the i step-ahead prediction. Hp is the prediction
horizon, Hu is the control horizon, and ǻ is the difference operator
(ǻ = 1 í zí1, where zí1 is the backward shift operator). The main
advantage with the MPC formulation is that it takes constraints into
account explicitly and integrator windup is no issue. The main tuning
parameters are the prediction horizon Hp, the control horizon Hu, and the
weights Q and R. It is possible to omit the weight on the control
increments, R, in (6.4), without losing the mid-ranging function. However,
the formulation is less flexible since there in practice is only one
weighting parameter (one of the parameters in Q can be set to e.g. one).

6.5 Moisture control by mid-ranging the air system

A mid-ranging strategy to control the sheet moisture by a combination of

both the blow boxes and steam cylinders, is evaluated by simulations of
disturbance rejection, using the model introduced in Section 6.2. The
results are compared to conventional moisture control, where only the
steam pressure in the cylinders is used.

6.5.1 Modeling and control design

The valve position controller is chosen as the mid-ranging structure to be
used, both for its flexibility and for being the most common approach in
industry, see Figure 6.13. The process consists of one controlled variable,
the sheet moisture, and two manipulated variables, the steam pressure set
point and the air flow actuator. The control signal u1 manipulates the air
flow actuator while u2 is a set point for the steam pressure. The motivation
for not manipulating the steam valves directly is both to preserve the

155
Chapter 6. Enhanced Moisture Control Using the Air System
’set point
air flow actuator’ ’set point steam pressure’

r2 u2
Ȉ C2(s) P2(s) ’moisture
disturbance’
CD (s)
’moisture’
1 n
r1 y
Ȉ C1(s) Ȉ
u1 P1(s) Ȉ Ȉ

’set point ’air flow actuator’

moisture’
1
Figure 6.13 The mid-ranging structure used in the simulations where the different
variables are also given. Note that u1 manipulates the air flow actuator while u2 is the set
point to the steam pressure controllers.

function of the cascade system in the condensate system and for paper
quality reasons, as described in Section 2.3.
Since the paper drying process includes a significant transport dead-
time, the two controllers C1 and C2 are based on the IMC concept, which
is well known as an effective dead-time compensator for a stable process
with long time-delays [Morari and Zafiriou, 1989]. The transfer function
of an IMC can be written as

H f ( s) P  ( s)
C ( s) , (6.6)
1  H f ( s) P  ( s) P(s)

where P is the process model, P + is the realizable inverse of the process

model, and Hf is a low pass filter with steady-state gain one that
determines the closed loop performance.
The process transfer functions are obtained from open-loop step
responses, see Figure 6.14 and Figure 6.15, and are modeled as

0.2548s  0.04887 10 s

P1 ( s )  e , (6.7)
617.3s 2  35.66 s  1
and

0.1365  40 s
P2 ( s )  e . (6.8)
61.5s  1

156
 Chapter 6. Enhanced Moisture Control Using the Air System
9.5

Moisture (%) 9.0

8.5

8.0

7.5

60
Air flow actuator (%)

50

40

30

20
-50 0 50 100 150 200 250 300 350 400
Time (s)
Figure 6.14 Step response of the air flow í moisture process (solid) and the obtained
model (dotted), also given in (6.7). The dash-dotted line indicates the final value.

14

12
Moisture (%)

10

460
Pressure set point (kPa)

440

420

400

-50 0 50 100 150 200 250 300 350 400 450 500

Figure 6.15 Step response of the steam pressure í moisture process (solid) and the
obtained model (dotted), also given in (6.8).

157
Chapter 6. Enhanced Moisture Control Using the Air System
The model from air flow actuator to moisture, P1, has two complex
conjugated poles to capture the dynamics. The physical explanation for
the overshoot is a combination of both dynamics in the paper and
cylinder. When the dry air flow is decreased, the amount of moisture in
the air around the sheet is increased. This reduces the evaporation of water
from the paper and the paper moisture increases. This also decreases the
paper temperature which makes the energy flow to the paper to increase,
and this in turn will increase the evaporation. However, the dynamics in
the cylinder is much slower than the air-paper process and the increased
evaporation due to increased energy flow to the paper lags behind the
increase in sheet moisture due to reduced air flow, which gives an over-
shoot in the paper moisture.
Notice the resemblance between the prestudy experiment in Figure 6.7
and the open-loop simulation in Figure 6.14. The moisture in Figure 6.7
reaches a new steady-state promptly in a manner that does not appear as a
process with only one or two real poles. However, it is difficult to
distinguish the over-shoot in the response because of disturbances.
P1 has both faster dynamics and shorter time delay compared to P2.
The design parameter for the IMC is the filter Hf. For C1 it is chosen as

1
H f 1 (s) 2
, (6.9)
§ W1 ·
¨1  s ¸
© Z ¹

where IJ1 = 1 (nominal value) implies that the closed-loop poles are real
and placed at the same distance from the origin as the open-loop poles, see
Figure 6.16. A value of IJ1 greater than one makes the closed-loop system
slower, and vice versa. Apart from the two poles, P1 also has a zero.
However, it has been observed that a simpler model, with only two
complex conjugated poles and no zero, is sometimes sufficient for the
airflow í moisture process. Therefore, a second order filter is chosen in
(6.9), to make the controller C1 realizable in both cases.
Ideally, the only term affecting C2 is CD and the controller is tuned for
that process. This can be seen by inserting C2 into the structure in
Figure 6.13, assuming perfect process models. The purpose of C2 is to
slowly restore the signal u1 to its desired value. In practice, it can not be
assumed to have an ideal decoupling filter, and therefore it is important
that C1 and C2 do not interfere with each other. The closed-loop time
constant of C2 is therefore related to the fast loop and the IMC filter is
chosen as
158
 Chapter 6. Enhanced Moisture Control Using the Air System
1
H f 2 ( s) . (6.10)
W
1 2 s
Z

The value of W2 is set to be significantly larger (at least five to ten times)
thanW1, to separate the controllers C1 and C2 in frequency.
The mid-ranging controller is compared to the case when only the
steam pressure in the cylinders are used to control the sheet moisture, here
denoted as steam pressure control. It is assume that the steam pressure
control is based on an IMC tuned for the process in (6.8) and the
corresponding filter is chosen as

1
H f 3 (s) , (6.11)
1  61.5W 3 s

where W3 = 1.5. To obtain an adequate comparison between the two

moisture control systems, the mid-ranging controller is tuned to have the
same maximum value of the sensitivity function, Ms, as the steam pressure
control, see Figure 6.17. In this way they have the same robustness to
modeling errors and are in that sense comparable. The value of Ms is
chosen to 1.3 and this gives W1 = 1.2. Also given in the figure is the
frequency response from moisture set point to sheet moisture, which in
this case is also equal to the complementary sensitivity function. Observe

Figure 6.16 The open-loop poles () and the closed-loop double pole (*) of the air flow í
sheet moisture process. The parameter Z is the distance between the open loop poles and
the origin.

159
Chapter 6. Enhanced Moisture Control Using the Air System
1
Compl. sens. function
0.8

0.6

0.4

0.2

0 -4 -3 -2 -1
10 10 10 10

1.5
Sensitivity function

0.5

0 -4 -3 -2 -1
10 10 10 10
Frequency (Hz)
Figure 6.17 Frequency plot for the mid-ranging system (dotted line) and steam pressure
control (solid line). The sensitivity function represents the transfer function from
disturbance n to sheet moisture y, see Figure 6.13. The complementary sensitivity
function is, for this control structure, equal to the transfer function from set point r to
output y.

that ideally, both the sensitivity and complementary sensitivity functions

are independent of controller C2 and consequently also W2, because of the
decoupling filter CD. The bandwidth of the closed loop system with the
mid-ranging controller is more than twice as large, compared to steam
pressure control. This is a good indication that taking advantage of the air
system together with the steam cylinders gives a higher performance than
solely using the steam system. There is a region in the sensitivity plot
where the mid-ranging controller has a higher amplification of
disturbances compared to steam pressure control. Compared to the
estimated level of noise in moisture given in Figure 2.18, there are no
severe variations in that region and the amplification of noise in that
frequency region by the mid-ranging is therefore not a problem for that
specific example. However, the noise distribution should be regarded
before implementing the mid-ranging controller on a drying section.

160
 Chapter 6. Enhanced Moisture Control Using the Air System
Remark
The air flow í moisture process can also be modeled with two real poles
and slow zero. The overshoot is then due to the lead action of the zero. In
this work, the process was identified by using System Identification
Toolbox in Matlab which gave a model with two complex conjugated
poles.

6.5.2 Simulation results

Because of the simulation technique, described in Section 6.2, where one
cylinder at a time is simulated in the physical model of the drying section,
it is not possible to attach a continuous control system to it. During a
simulation, the resulting change in sheet moisture is not known until the
last cylinder is simulated, which is when the whole simulation is
performed. Therefore, continuous feedback control is not achievable and
the control system is discretized. A sample time of 5 s is chosen and when
the controller puts out a new control signal, the paper machine is
simulated for 5 s and the moisture after the last cylinder is fed back to the
controller. Obviously, this solution is not a disadvantage since it imitates
the procedure of a control system in reality.
All simulations show the response to a disturbance in inlet moisture to
the drying section. This can be interpreted as changed conditions in either
the wire section or press section of the paper machine. The inlet moisture
is changed from 62.12 to 62.82 % (this corresponds to adding 50 grams of
extra water to each kg of dry solids). The size of the step disturbance and
nominal supply air flow are chosen so that the air flow actuator is not
saturated. This means that only the linear part of the control system is
analyzed. However, windup protection and saturations in mid-ranging
control is further discussed in Chapter 8.
In Figure 6.18, the response in sheet moisture due to the change in inlet
moisture is shown. The mid-ranging has a significantly better disturbance
rejection than the steam pressure control. There is a slight fluctuation in
sheet moisture, for the mid-ranging case, after t = 300 s. This is because of
imperfect models in the decoupling filter, CD, and it becomes more
prominent when pushing the performance level for the controller
(W1 = 0.7). When u2 mid-ranges u1 by increasing the steam pressure, CD
reduces the air flow accordingly but the compensation is not perfect which
affects the process output. This issue would probably benefit from
changing model (6.8) to a two-pole model, see also Figure 6.15.

161
Chapter 6. Enhanced Moisture Control Using the Air System
10.5
Mid-ranging, W1 = 0.7, W2 = 5.0
Mid-ranging, W1 = 1.2, W2 = 5.0
10.0 Steam pressure control

9.5
Sheet moisture (%)

9.0

8.5

8.0

7.5
0 100 200 300 400 500 600
Time (min)
Figure 6.18 Comparison between the mid-ranging control system (W2 = 5.0) and steam
pressure control. The steam pressure control should be compared to mid-ranging with
W1 = 1.2.

10.5
Mid-ranging, W1 = 1.2, W2 = 5.0
Mid-ranging, W1 = 1.2, W2 = 10.0
10.0 Steam pressure control
Sheet moisture (%)

9.5

9.0

8.5

8.0

7.5
0 100 200 300 400 500 600
Time (s)
Figure 6.19 Moisture in paper for two different W2.

162
 Chapter 6. Enhanced Moisture Control Using the Air System
Figure 6.19 shows that the response in sheet moisture is practically
independent ofW2. However, there is a slight difference at the second half
of the simulation. Larger values of W2 reduce the fluctuation since the
steam system is less aggressive in its attempt to ‘mid-range’ the air
system. Figure 6.20 and Figure 6.21 show the corresponding dew point of
the air and steam pressure in the lead group. It is evident that both the
steam and air process becomes less aggressive with larger W2. However,
the larger W2 is, the more the control system becomes ‘single-loop’ and the
advantages of mid-ranging are reduced.
Figure 6.21 also shows the steam pressure in the lead group for steam
pressure control. The steam pressure is not increased as rapidly by the
mid-ranging control as by steam pressure control. In the mid-ranging case,
the fast moisture transients are handled by the air system and the steam
system is only used to restore the air system in steady-state. Less
variations in steam pressure is advantageous since it reduces the injection
of disturbances in the steam and condensate system, which has negative
effect on both steam production and other steam users, see also
Section 2.4.

Remark 1
One drawback of using the dry air flow (and thus the dew point) to control
the sheet moisture, is the risk of reduced efficiency in the heat recovery. A
well optimized drying section ventilation has a dew point close to the
allowed maximum. For the proposed control technique to function well,
the average dew point then needs to be decreased which leads to higher
energy costs. This should be taken into account by weighing the gain of
reduced variability in sheet moisture against increased energy usage
[Lindell and Stenström, 2004], when evaluating the control principle for a
specific drying section.

Remark 2
Physically, the dew point is a driving force for the evaporation of water in
the sheet. A low dew point implies a low vapor partial pressure in the air,
and high difference in vapor pressure between sheet and air. It can be
interpreted as a low dew point pulls out the moisture in the sheet. This is
the opposite of an increase in steam pressure which increases the
evaporation by increasing the vapor pressure in the sheet, and pushes out
the moisture in the sheet. In [Karlsson and Stenström, 2005b] it is shown
that a high vapor pressure inside the sheet can cause delamination

163
Chapter 6. Enhanced Moisture Control Using the Air System
63

62
Dew point (°C)

61

60

59

W1 = 1.2, W2 = 5.0
W1 = 1.2, W2 = 10.0
58
0 100 200 300 400 500 600
Time (s)
Figure 6.20 Change in dew point (mid-ranging).

465

460
Steam Pressure set point (kPa)

455

450

445

440 Mid-ranging, W1 = 1.2, W2 = 5.0

Mid-ranging, W1 = 1.2, W2 = 10.0
Steam pressure control
435
0 100 200 300 400 500 600
Time (s)
Figure 6.21 Steam pressure in the last steam group (lead group) when comparison steam
pressure control and mid-ranging.

164
 Chapter 6. Enhanced Moisture Control Using the Air System
(different pulp layers are separated) problems in board machines. Since
the mid-ranging control presented here, gives smaller variations in steam
pressure it should reduce this problem.

Remark 3
An alternative control configuration is to let the dew point be the inner
part of a cascade loop with the moisture control. Controller C1 then gives
a set point to the dew point controller, which is included in P1, compare
with Figure 6.13. However, the dew point is normally measured in the
exhaust air, far from the sheet. Experiments in [Forsman and Birgerson,
1999] show that the time constant in the process from air flow actuator to
dew point is around 1.5 minutes due to the large volume of air in the
hood. This would give a slow moisture control loop, unless the dew point
is measured close to the paper (in the pocket). The experiment in
Figure 6.7 gives a time constant for the actuator í dew point process
around 30 seconds. Also in this case the dew point needs to be measured
closer to the sheet to be useful in a cascade control configuration.

6.5.3 Zero-level and dew point control

There are two important variables to control in the air inside the dryer
hood, the dew point and zero-level, see Figure 6.22. Normally the supply

2 2.5 m

Figure 6.22 The air balance in the hood. The large unfilled arrows indicate the exhaust air.
To provide uniform air flows around the cylinders, there is a false ceiling. The smaller
arrows indicate the pressure inside the hood. Below the doors of the hood, there is an under
pressure to prevent moist air to leak into the machine room. The height where the air
pressure inside the hood equals the outside pressure, is called zero-level, [Karlsson, 2000].

165
Chapter 6. Enhanced Moisture Control Using the Air System

C3(s) Ȉ
r3
r2 u2
> Ȉ C2(s) P2(s)

CD (s)
1
1
r1 y
Ȉ C1(s) Ȉ
u1 P2(s) Ȉ

1
Figure 6.23 By allowing the dew point to adjust the air flow actuator set point, it is
possible to set an upper constraint on the dew point. The output from the selector is fed
back to the integral mode of C3 to avoid windup.

air is used to control the zero-level, and the exhaust air to control the dew
point [Forsman and Birgerson, 1999]. The dew point is often measured in
the exhaust air channel.
The proposed control strategy in this chapter uses the supply air to
control the sheet moisture. That means that the exhaust air must be used to
control the zero-level. It is vital to have an upper constraint on the dew
point to prevent condensation that might cause dripping on the sheet and
corrosion on machine units. A possible solution, in the mid-ranging
structure, is shown in Figure 6.23. By adding an extra controller, C3, that
controls the dew point in the exhaust air through a selector that is shared
with set point r2, the air flow actuator set point is increased if the dew
point exceeds set point r3.

6.6 Summary

A new control strategy has been developed for the multi-cylinder drying
section, to reduce sheet moisture variability. The air system in the blow
boxes are used to control sheet moisture, in combination with the steam
system. A previously developed physical model for the drying section is
used in simulations, where the performance of the control system is
compared with a conventional moisture control.

166
 Chapter 6. Enhanced Moisture Control Using the Air System
The result is that by combining the conventional steam control loop with
the supply air system, it is possible to significantly improve the
disturbance rejection of the sheet moisture feedback loop, without
changing the level of robustness. This is shown in both in simulations and
by observing that the bandwidth for the disturbance rejection is more than
twice as large for the mid-ranging control. The physical reason for this is
the fast response of the dewatering rate to changes in air dew point.
There are two design parameters in the selected mid-ranging structure,
W1 and W2. The performance of the closed loop system is determined by W1
and W2 determines how fast the steam system should restore the air system
to its original level. Ideally, these two parameters can be chosen
independently but in practice it is recommended to let W2 • 5W1. Letting
W1 = 1, means that the closed loop poles have the same distance to the
origin as the open loop poles.

167
Chapter 7. Feedforward from a Paper Surface Temperature Measurement

7
7.

Feedforward from a Paper

Surface Temperature
Measurement

The paper machine is often run with constant set points for all quality
variables for a long period of time, typically several hours or even days.
The primary function of the control system is then to maintain the desired
properties in the presence of disturbances. In Chapter 6, it is shown that
the air in the surrounding hood can beneficially be integrated with the
steam pressure control system by mid-ranging. The mid-ranging control
system proved to be far more efficient than using only the steam pressure
as manipulated variable. By using a combination of both the fast but low-
capacity air system and the slow but high-capacity steam system, the mid-
ranging control system was able to significantly reduce the effect of
moisture disturbances.
Due to the long time constants and time delays in the process, there is a
large potential gain in disturbance rejection by introducing feedforward to
the control system. It is essential that the disturbance signal is measured
early in the process to make the feedforward effective. Installation of a
moisture scanner between the press and drying section is possible and the
moisture measurement can be used for feedforward control. However, the
space between the press and drying section is often very small and a
scanner is also a fairly large investment for the mill.

168
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
Therefore, a new signal to be used for feedforward control of the paper
moisture content is presented. The signal is based on the paper surface
temperature in a few positions in the machine direction. The drying
process generates large moisture gradients in the thickness direction of the
paper and the surface becomes dry before the center of the sheet. When
the paper surface becomes dry the surface temperature quickly increases
to a value above 100°C. The general idea is therefore to estimate the
position where the paper surface becomes dry by measuring the surface
temperature. The possibility to measure the signal is verified by
measurements on a real paper machine. The control system is evaluated
by simulations of the drying section model presented in Chapter 6. Similar
work for the wire section, where new feedforward signals have been
proposed, can be found in [Larsson and Gustafsson, 1998] and [Li, et al,
2001].

7.1 The peak position í the position of a dry surface

An important physical aspect in paper drying are the concepts of free

water and bound water. The amount of energy required to evaporate the

70
Inlet moisture content: 61.39 %
Inlet moisture content: 62.12 %
60

50
Moisture content (%)

40
Node 2

30

20

Node 1
10

0
0 10 20
50 30 40
60 70 80 90 100
Cylinder no.
Figure 7.1 Moisture content at the first two nodes in the thickness direction. Note that
when the first node reaches a moisture content of 10 %, the second node is still at 60 %.

169
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
free water is equal to the latent heat of vaporization. The bound water,
however, is tightly bound to the fibres by hydrogen bonds, and requires
more energy to evaporate compared to the free water. This means that the
paper is more difficult to dry in presence of only bound water. The
proposed feedforward structure uses this indirectly by measuring the
surface temperature.
As described in Section 6.2, the discretization of the paper sheet model
in the thickness direction is set to 10 nodes. A steady-state simulation of
the moisture content at the two first nodes seen from the lower surface, is
shown in Figure 7.1. The first layer reaches the bound water region at
cylinder number 25. This can be seen since the evaporation rate changes
significantly. The same event happens to the second fibre layer at cylinder
number 43. The paper surface then contains a low moisture content at the
same time as a lot of free water is still present in the middle of the paper
structure.
To validate the model in terms of simulated paper surface temperature,
the temperature has been measured on the paper machine the model is
adapted for. By using an IR-camera which gives the surface temperature
in a two-dimensional window, temperature values along a line in the
130

120

110
Paper temperature (°C)

100

90

80 115

70 110

60 105 Simulated
100 Measured
50
0 20 40 60 80 100
Cylinder no. 95
90
85
80
Simulated
75
Measured
15 20 25 30 35 40 45 50
Cylinder no.
Figure 7.2 Comparison between measured and simulated paper surface temperature. The
most important section for the feedforward structure in enlarged and the 100°C-level is
indicated.

170
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
Steam
Condensate (a) (b)
p1 p1

Cylinder shell

Different surface
temperatures.
Figure 7.3 Schematic picture that shows the changed properties as the paper surface dries
out. A wet surface (a) will act differently compared to a dry surface (b), given the same
external conditions.
machine direction are singled out and the results are shown in Figure 7.2.
The simulations show both the temperature for the free draw and the part
where the paper is in contact with the cylinder. Since the paper in the
contact zone is not visible due to the dryer fabrics, the temperature is only
measured in the free draw. The simulated values show very good
agreement with the measured values. Observe that this is a pure validation
without any adjustment of the model, at an operating point different from
the one where the model was fitted. Paper temperatures exceeding 100°C
are first observed in the free draw section after cylinder number 27 and
this phenomenon appears more frequently after cylinder number 35.
The layer of surface fibres without any surrounding free water show
totally different properties when it comes to both evaporation and heat
transfer mechanisms, compared to fibres surrounded by free water. Since
the evaporation requires a large amount of energy, the absence of free
water will cause the paper temperature to increase. This can also be
considered from another viewpoint. If the surface temperature is above
100°C, there can not exist any free water on the surface since it would
instantly evaporate to the surrounding air. The important event is thus the
formation of a dry fibre layer since the paper surface temperature will
change as the free water is removed. An illustration of the phenomenon is
shown in Figure 7.3.

171
Chapter 7. Feedforward from a Paper Surface Temperature Measurement

(a) (b)

Suggested measuring
position

Figure 7.4 Description of the suggested measuring position. It is important to measure on

the same side of the paper surface since the drying often is uneven in the thickness
direction.
The paper surface temperature measurements shown in Figure 7.2 indicate
that the paper surface moisture content is at a low level at cylinder number
27, otherwise the observed high temperatures are not possible. The
measurements thus confirm the large moisture gradients in the thickness
direction, suggested by the model. The paper surface temperature signal is
reliable and showed little variation during the time of the measurements.
The appealing aspect here is that variations in the the inlet moisture
content to the drying section affects the position where the paper surface
reaches the bound water region. It can therefore serve as an indirect
moisture measurement. The paper surface temperature itself is not
interesting but the point where there is a sudden increase in temperature
indicates that the surface is dry. By measuring the temperature in the
middle of a free draw in a number of positions a temperature curve is
obtained, see Figure 7.4 and Figure 7.5.
The paper surface temperature obviously also changes when the paper
sheet reaches a new steam group (with a different steam pressure inside its
cylinders) and it is then not possible to distinguish the occurrence of a dry
surface. Measurement of the feedforward signal therefore requires that the
paper surface dries out in the middle of a steam group. The steam
pressures are also shown in Figure 7.5 to mark the different steam groups.
Within each group, the paper surface temperature reaches a constant level
but in group 3 and 4 the paper temperature suddenly increases in the
middle of the steam group, which corresponds to the first and second node
reaching the bound water region. The differences in temperature between

172
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
150 300

140 Group 4

130
Constant 250
Paper surface temperature (°C)

120 temperature
zone

Steam pressure (kPa)

110
Group 3
100 200

90

80
Sudden increase 150
Group 2
70 in temperature
Constant
temperature
60 zone
50 100
5 10 15 20 25 30 35 40 45 50
Cylinder no.
Figure 7.5 The temperature curve obtained by measuring the surface temperature as
indicated in Figure 7.4. The steam pressures in the different groups are also shown.

w in = 62.12 %
5
w in = 61.39 %
Free draw temperature difference (°C)

Changed
-1 peak position

5 10 15 20 25 30 35 40 45
Cylinder no.
Figure 7.6 A temperature curve calculated by taking the difference between two adjacent
temperature measurements, as indicated in Figure 7.4.

173
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
two adjacent measurements are shown in Figure 7.6. All the differences in
free draw paper temperature that deviate from zero are either due to a new
steam group or because the paper surface has dried out. The paper surface
temperature differences were fitted to cubic splines and the position the
maximum of these splines (originated from a dried-out surface) will
hereafter be called “peak position”.
Two different cases with different inlet paper moisture ratios are
shown in Figure 7.6. The paper temperature is affected differently at the
two positions where the two first fiber layers dry out but shows the same
behavior for the positions of all other peaks (representing the heat-up
periods). The peak at cylinder number 43 is a reaction of the free water
removal in the second node and does not show large variation in position
(cylinder number).
Note that there might well be a difference in peak position and the
position where the paper surface actually dries out. The maximum of the
temperature difference shows where the dry surface causes maximum
effect and not the exact location of a dry surface. For instance, if position
(a) in Figure 7.4 show 1 °C higher temperature than the previous
measuring position and position (b) show 6 °C higher temperature than
(a), the surface may still have dried out just before (a). The effect of the
dry surface is however radically increased as the paper reaches (b). The
physical meaning of the signals, however, is of minor importance as long
as it indicates changes in paper moisture. Both the peak position and the
sheet moisture at the reel up will be affected by a disturbance in inlet
moisture content and the peak position can therefore be used in
feedforward control.
To measure the peak position, an IR camera should be positioned in
the free draws around the position where the surface becomes dry. It is
crucial to measure at the same position in each free draw to obtain
relevant comparisons of the temperatures. Here, the chosen position is the
middle of the free draw. The peak position may however be difficult to
measure online with the current available measuring technique. The
measuring devices must e.g. withstand the humid environment inside the
hood and problems with fibre coating on the device must be dealt with.
Also, the measuring depth could cause problems if several fibre layers are
visible by the measuring equipment. During the experiments it was also
observed that both the viewing angle and distance to the paper is
important for the result.

174
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
r2 u2
Ȉ C2(z) P2(z) ”peak position”

CD (z) Hff (z)

H(z)
1
r1 y
Ȉ C1(z) Ȉ Ȉ P1(z) Ȉ Ȉ
u1
1
Figure 7.7 A simplified block diagram of the control structure used in the simulations.

7.2 Design of a feedforward controller

Figure 7.7 shows a somewhat simplified block diagram of the mid-

ranging control system together with the feedforward. H is the process
from peak position to moisture content, Hff is a feedforward block and the
definition of the other parts are given in Chapter 6, see Figure 6.13. Since
11
Sheet moisture (%)

10

7
0 50 100 150 200 250 300

26.5
Peak position (cyl.)

26.0

25.5

25.0
0 50 100 150 200 250 300
Time (s)
Figure 7.8 The experiment used for identification of H. The solid lines are the result of the
simulation and the dotted line is the obtained simplified model given in (7.1).

175
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
”Temperature
peak position”
Hff,3

Ȉ P1,3
H
Hff,4
u1 y
Ȉ P1,4 Ȉ Ȉ Ȉ

Hff,5

Ȉ P1,5

Figure 7.9 A block diagram of the feedforward structure used in the simulations.

the simulation model is very large (~105 states) it needs to be simplified

before it can be used for control design. This is done by system
identification of a low order model. To excite the peak position a step
change in inlet moisture content to the drying section is performed. The
identification of the process and the design of the control system is done
in discrete time, where the sample time is 5 seconds. This differs from the
procedure in Chapter 6 where the modeling and design are done in
continuous time and the discretized right before the implementation. To
identify the process H, the moisture controller is put in manual mode
(while the steam pressure control is in automatic mode), and a step
disturbance is introduced in the moisture content at the inlet of the drying
section. The result is shown in Figure 7.8 and the obtained discrete model
of the process is

0.185 z 6
H ( z) . (7.1)
1  1.528 z 1  0.599 z  2

The feedforward can be implemented as described above but it is slightly

modified. The air system is divided into five groups so that the
disturbance compensation from the feedforward “follows” the paper
through the machine. More specifically, when there is an indication of a
moisture disturbance in the peak position (which is at the first part of the
drying section) it is unfavorable to compensate it directly at the end of the
drying section. From step responses it is found that the time delay from
the first and second air flow actuator to moisture is longer than the time

176
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
9.0

Sheet moisture (%)

8.5

8.0
Air flow actuator (%)

50

40

30

20
0 50 100 150 200 250 300 350 400
Time (s)
Figure 7.10 Step responses from the five air flow actuators to moisture content at the reel
up. The two dashed lines are the first two air groups which are not used for the
feedforward control.

delay of H. Consequently, feedforward is not realizable for those groups

and only the three last air groups were used. This means that when a
disturbance is measured, it is more effective to use the last three air
groups. Figure 7.9 shows the configuration, with the three utilized parts of
P1. Since the impact of the disturbance at y is to be eliminated, the
condition

H  P1,3 H ff ,3  P1, 4 H ff , 4  P1,5 H ff ,5 0, (7.2)

holds. For simplicity the feedforward filters were selected as

H H H
H ff ,3  , H ff , 4  , H ff ,5  . (7.3)
3P1,3 3P1, 4 3P1,5

The control signals of the last three air groups now might be different and
therefore C2 was set to control the actuator signal of the first two groups
(u1). Another possible solution is to let C2 control the mean value of all

177
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
five air flow actuators or to introduce multivariable control in the steam
system so that each air group is mid-ranged by a steam group.
Just as in the case of H, the air groups are identified from step responses,
shown in Figure 7.10, and are given by

 0.00411  0.00198 z 1  6
P1,3 ( z ) z ,
1  1.609 z 1  0.685 z  2
 0.00476  0.00224 z 1  4
P1, 4 ( z ) z , (7.4)
1  1.490 z 1  0.590 z  2
 0.00537  0.00214 z 1 1
P1,5 ( z ) z .
1  1.537 z 1  0.639 z  2

7.3 Simulations

The feedforward system in (7.3) is simulated with a step disturbance in

inlet moisture ratio. It is then compared to simulations with conventional
steam pressure control and with the previously developed mid-ranging
control system in Chapter 6. The sheet moisture is shown in Figure 7.11.

11.0

10.5

10.0
No control - open loop
Steam pressure control
Sheet moisture (%)

Mid-ranging only
9.5 Feedforward + mid-ranging

9.0

8.5

8.0

7.5
0 100 200 300 400 500 600
Time (s)
Figure 7.11 Comparison of different control systems.

178
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
26.5

Peak position (cyl.) 26.0

25.5

25.0

8.5
Sheet moisture (%)

8.0

7.5

8.5

8.0

7.5
0 100 200 300 400 500 600
Time (s)
Figure 7.12 Peak position and sheet moisture during the closed loop simulation.

63

62.5

62

61.5

61
Dew point ( C)
o

60.5

60

59.5

59
Mid-ranging only
Feedforward + mid-ranging
58.5
0 100 200 300 400 500 600
Time (s)
Figure 7.13 Change in dew point.

179
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
465

460
Steam pressure (kPa)

455

450

445

440
Steam pressure control
Mid-ranging only
Feedforward + mid-ranging
435
0 100 200 300 400 500 600
Time (s)

Figure 7.14 Comparison of the steam pressure in the lead group by the different control
system.
0.8
Steam pressure control
Mid-ranging only
0.7 Feedforward + mid-ranging

0.6

0.5
Amplitude

0.4

0.3

0.2

0.1

0 -3 -2 -1 0
10 10 10 10
Frequency (rad/s)
Figure 7.15 Frequency plot, showing how the disturbances in inlet moisture propagate, for
the different control systems.

180
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
The most apparent result is that the disturbance is greatly attenuated. After
150 seconds the decreasing rate of the moisture content restoration is
reduced for the feedforward system. The reason for this can be seen in
Figure 7.12, where the peak position is shown for the same simulation.
When the control system compensates for the disturbance by increasing
the control signal, it also slightly affects the peak position. The
feedforward signal is thus affected but this dynamic relation from u1 to the
peak position in Figure 7.7 is neglected. By considering this in (7.2) the
feedforward could probably be slightly improved. For comparative
reasons Figure 7.11 also shows the moisture deviation when all control
systems are switched off (open loop).
The dew points in P1,3 are shown in Figure 7.13 and the steam pressures
are shown in Figure 7.14. Notable for the two feedforward control
systems is that the air flow actuators are not restored to their original
positions by the mid-ranging part. This was also noted in Section 7.2.
Multivariable control of the air and steam groups could be introduced to
give maximum manoeuvring range to the individual air flow actuators, if
desired.
The different control systems are also compared by evaluating their
disturbance rejection in the frequency domain, see Figure 7.15. The figure
is generated by identifying the process from inlet moisture content to
moisture content at the reel-up and the figure simply shows the Bode plot
of the identified model. The feedforward control shows to be superior.
Due to the low pass characteristics of the process, all systems attenuate
high frequencies similarly. For low frequencies the difference between the
systems is considerable. Since there are both slow dynamics and long time
delays in the process, the feedback systems can only attenuate extremely
slow disturbances (note that 10í3 rad/s is a variation with a period time of
almost two hours). The feedforward can beneficially use its prior
information about the disturbances acting on the system.
The feedforward assumes linear systems while the system is nonlinear.
With the presented technique to obtain the control parameters it is
therefore not possible to completely eliminate the effect of disturbances.

Remark
The implementation of the feedforward does not require a mid-ranging
control system. It might as well be combined with any other moisture
control structure. The mid-ranging controller is chosen here to give
natural connection to Chapter 6.

181
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
7.4 Summary

A new measured signal, called the peak position, has been presented that
can be used for feedforward in moisture control. The signal is based on
measurements of paper surface temperature in several free draw zones. By
observing the position at which the paper dries out on the surface it is
possible to compensate for moisture disturbances before they are
measured by the moisture scanner at the reel-up.
With the addition of the peak position in combination with a process
model, it is possible to significantly reduce the transients from inlet
moisture disturbances. From the studied step disturbance it is clear that the
feedforward signal greatly improves the performance of the paper
moisture control system.
The peak position may be difficult to measure with the current
available technique. The potential increase in performance is, however,
large and the signal may be replaced by any other suitable signal that
indicates disturbances in moisture content. Equipment that measure the
surface moisture content are being developed [Boström, 2002]. This
signal would be very suitable for feedforward control. What type of signal
that is finally chosen is of minor importance, as long as it can be used as a
disturbance indication.

182
 Chapter 8. Object-Oriented Modeling and Control of the Moisture

88.
Object-Oriented Modeling
and Predictive Control of the
Moisture Content

The model used in Chapter 6 and Chapter 7 is capable of dynamically

describing moisture gradients and other properties inside the paper sheet.
This is necessary for the simulation of feedforward from surface
temperature measurements (the peak position). However, the model
contains approximately 105 states which make simulations fairly tedious.
It also lacks a physical model for the steam system, where it instead uses a
black-box model. Therefore a less complex model is developed that is
based purely on physical relations. The objective is control of the moisture
at the reel-up and not to accurately describe micro-scale moisture
variations inside the sheet. The model is built on heat and mass balances
for steam, cylinder, and paper. The core of the model is based on work by
[Wilhelmsson, 1994], [Persson, 1998], and [Slätteke and Åström, 2005].
The model is implemented in the object-oriented modeling language
Modelica, see [Mattsson and Elmqvist, 1997], [Mattsson, et al, 1998], and
[Fritzson, 2004]. Like any object-oriented programming language,
Modelica provides the notions of classes and objects, also called
instances, as the fundamental tool. Properties like inheritance and abstract
classes provide a structured approach to implement equations. The
advantages of such a modeling tool are (i) it is built on a non-causal
equation structure (ii) it is possible to create model components that
183
Chapter 8. Object-Oriented Modeling and Control of the Moisture
correspond to physical objects in the real world, in contrast to modeling
techniques that require conversion to signal blocks (iii) it permits mixing
of physics with empirical models (iv) it is easy to go from simple models
to high fidelity models by drag and drop features (v) it is easy to build and
exchange model libraries and (vi) it is suited for modeling in several
engineering domains.
The obtained model is used to investigate how the heat transfer
coefficient affects some dynamic properties and this is compared to some
of the results presented in Chapter 4. It is also validated against
measurements on a real paper machine. Finally, the model is used to
evaluate a new MPC strategy to control the moisture that has similar ideas
as the mid-ranging structure in Chapter 6.

8.1 The model

This chapter describes a physical simulation model of a drying section,

implemented in an object oriented modeling language. A similar attempt
is [Bergström and Dumont, 1998] and [Bortolin, et al, 2004], where the
object oriented modeling technique is demonstrated by modeling the stock
preparation (the process section that precedes the paper machine) and the
wet end, and the commercial process simulator APROS [Silvennoinen, et
al, 1989], [Niemenmaa, et al, 1996]. The objective is to develop a non-
linear model that captures the key dynamical properties for a wide
operating range. The equations for the steam and cylinder process are
taken from Chapter 4, apart from a few exceptions. For completeness,
they are all shown here.

The steam and cylinder process

Let qs be the mass flow rate of steam into the cylinder, qc be the
condensation rate, qbt the blow through steam, and qw be the siphon flow
rate. Also, let Vs and Vw be the volume of steam and water in the cylinder,
and let Us and Uw be the densities. The mass balances for water and steam
are then

d
U sV s  q s  q c  q bt ,
dt (8.1)
d
 U wV w  q c  q w .
dt

184
 Chapter 8. Object-Oriented Modeling and Control of the Moisture
The energy balances for steam, water, and metal are

d
U s u sV s  (q s  q bt )hs  q c hs ,
dt
d
 U w u wV w  q c h s  q w h w  Q m , (8.2)
dt
d
dt
 
mC p , m Tm Q m  Q p ,

where Qm is the power supplied from the water to the metal, Qp is the
power supplied from the metal to the paper, hs is the steam enthalpy, hw is
the water enthalpy, m the mass of the cylinder shell, Cp,m the specific heat
capacity of the shell, Tm the mean temperature of the metal, us and uw are
the specific internal energies of steam and water. The steam and water
volumes add up to the total cylinder volume,

V Vs  Vw . (8.3)

The power flow to the metal is given by

Qm D sc Acyl Ts  Tm  , (8.4)

where Dsc is the heat transfer coefficient from the steam-condensate

interface to the centre of the cylinder shell, Acyl is the inner cylinder area,
and Ts the steam temperature. The power flow to the paper is

Qp D cp AcylK Tm  T p  , (8.5)

where Tp is the paper temperature, Dcp the heat transfer coefficient from
the cylinder shell to the paper, and Ș is the fraction of dryer surface
covered by the paper web. In Chapter 4, Dsc is used to calibrate the model
against plant data. Here, both Dcp and Dsc are possible candidates for that
purpose. Experiments have shown that Dsc depends on both condensate
thickness, machine speed, and the number of spoiler bars, see [Pulkowski
and Wedel, 1988] and [Wilhelmsson, 1995]. However, the condensate has
a turbulent behavior and the heat transfer coefficient has proven to be
difficult to model, see Figure 8.1. Therefore Dsc is used as a free variable
to calibrate the model with. Empirical models for Dcp have been

185
Chapter 8. Object-Oriented Modeling and Control of the Moisture

15 spoiler bars
3000
25 spoiler bars
30 spoiler bars
No spoiler bars
2500
Heat transfer coefficient (W/m K)
2

2000

1500

1000

500

0
0 2 4 6 8 10 12 14 16
Condensate thickness (mm)
Figure 8.1 Steady-state measurements of how the number of spoiler bars affect the heat
transfer coefficient for the condensate [Pulkowski and Wedel, 1988]. It also depends on
machine speed and bar size, and is difficult to model due to the turbulent behavior.

developed. From [Wilhelmsson, 1995] a linear relation with moisture ratio

u is given

D cp (u ) D cp (0)  955u, (8.6)

is obtained, where Dcp(0) varies between 200 and 500 W/(m2K). It is well
known that Dcp depends on other things, e.g. the web tension, and surface
smoothness of both paper and cylinder, but this is omitted here.
Since the steam flow to the cylinder cannot be manipulated directly, a
valve model is also needed. From [Thomas, 1999] we have

qs Cv f v ( xv ) ( psh  p) U s , (8.7)

where Cv (m2) is the valve conductance, xv is the position of the valve

stem and the function fv is the valve characteristics called valve trim. The
valve stem varies from 0 (minimum valve opening) to 1 (maximum valve
opening). The supply pressure at the steam header is psh. We use equal

186
 Chapter 8. Object-Oriented Modeling and Control of the Moisture
1.0
R v = 200
R v = 50
0.8
R = 10
v
fv ( xv ) 0.6

0.4

0.2

0
0 0.2 0.4 xv 0.6 0.8 1.0
Figure 8.2 Equal percentage valve characteristic.

percentage trim, since it is the most common characteristic in the process

industry [Hägglund, 1991], see Figure 8.2. It is given by

x v 1
f v ( xv ) Rv . (8.8)

Rv is a constant known as the “rangeability” since it is the ratio between

the maximum and minimum valve opening.
For simplicity, all steam within the cylinder cavity is assumed to be
homogeneous with the same pressure and temperature. We also assume
that the steam in the cylinder is saturated. This means that the enthalpy,
density, and temperature are functions of the pressure only. Fitting
polynomials to the tabulated values for saturated steam in [Schmidt 1969],
gives

Ts 0.1723(ln p )3  3.388(ln p ) 2  37.71ln p  124.5,

hs [0.07402(ln p) 4  2.887(ln p)3  39.58(ln p) 2
 260 ln p  1824] ˜103 ,
(8.9)
hw [0.8842(ln p )3  18.77(ln p ) 2  200 ln p  748.5] ˜103 ,
U s [0.005048 p  64.26] ˜10 3 ,
Uw 0.3136(ln p)3  6.792(ln p ) 2  52.43 ln p  1141.

Equations (8.1) í (8.9) are a crude nonlinear model for the steam-cylinder
process. By choosing p, Vw, and Tm as state variables and using partial

187
Chapter 8. Object-Oriented Modeling and Control of the Moisture
derivatives, the system can be rewritten into a third order state equation
(most steps are omitted here), where f1, f2, and f3 are defined as the right
hand side of the equations.

dp dV
e11  e12 w Cv f v ( xv ) ( psh  p) U s  qw  qbt f1 ,
dt dt
dp dVw
e21  e22 Cv f v ( xv ) ( psh  p ) U s hs (8.10)
dt dt
 qbt hs  qw hw  D sc Acyl (Ts  Tm ) f 2 ,
dTm
e33 D sc Acyl (Ts  Tm )  (D cp (0)  955u ) AcylK (Tm  T p ) f3 ,
dt

where

dU s dU
e11 (V  Vw )  Vw w ,
dp dp
e12 Uw  Us ,
dU s dh dU dh
e21 hs (V  Vw )  U s (V  Vw ) s  hwVw w  U wVw w  V ,
dp dp dp dp
e22 U w hw  U s hs ,
e33 mC p .
(8.11)

In the rewritings of the energy balances above the specific internal energy
has been eliminated by the definitions us = hs  p/ȡs and uw = hw  p/ȡw.
Using f1, f2, and f3 the system can be further rewritten into an explicit state
form.

dVw e33 e21 f1  e11e33 f 2

,
dt e33 (e21e12  e11e22 )
dp e12 e33 f 2  e33 e22 f 1
, (8.12)
dt e33 (e21e12  e11e22 )
dT f3
.
dt e33

188
 Chapter 8. Object-Oriented Modeling and Control of the Moisture

Figure 8.3 A Simulink model of (8.12). The submodel f2 is also opened, which in turn
contains submodels.
model Cylinder

equation
Ms = rhos*Vs;
der(Ms) = qs – qc - qbt; % (8.1)
Mw = rhow*Vw;
der(Mw) = qc - qw;

Es = rhos*us*Vs;
us = hs – p/rhos;
der(Es) = (qs – qbt)*hs – qc*hs;
Ew = rhow*uw*Vw; % (8.2)
uw = hw – p/rhow;
der(Ew) = qc*hs – qw*hw – Qm;
Em = m*Cp*Tm;
der(Em) = Qm - Qp;

V = Vs + Vw; % (8.3)
Qm = alpha_sc*Acyl*(Ts - Tm); % (8.4)
Qp = alpha_cp*Acyl*eta*(Tm - Tp); % (8.5)

end Cylinder;

Figure 8.4 Code segment of the Modelica model of the steam cylinder. In addition,
equations for steam properties are required to give a complete simulation model.

189
Chapter 8. Object-Oriented Modeling and Control of the Moisture
In this form the model can be directly implemented and simulated in e.g.
Simulink, see Figure 8.3. By using Modelica instead the tedious and error
prone procedure of transforming the system to explicit form is avoided
and we let the simulation environment decide the state realization. Since
the transformation of equations is automated, it is also easier to change the
model at a later stage. Equations (8.1) í (8.5) are put into the simulation
environment as they are, see Figure 8.4.

The paper web process

We will now expand the model to also include dynamics for the paper
sheet. To describe the moisture in the paper we need a mass balance and
to describe the paper temperature we need an energy balance. Starting
with the mass balance, we describe how much water is evaporating from
the paper surface to the air. From [Persson, 1998] we get the Stefan
equation

ptot KM w §¨ ptot  pv , a ·¸
qevap ln , (8.13)
Rg T p ¨ ptot  pv , p ¸
© ¹

where qevap is the evaporation rate (kg/m2s), K is the mass transfer

coefficient (m/s), Mw is the molecular weight of water (kg/mole), ptot the
total pressure (Pa), pv,a the partial pressure for water vapor in the air (Pa),
pv,p the partial pressure for the water vapor at the paper surface, Rg the gas
constant (J/mole·K), and Tp the paper temperature (K). The partial
pressure pv,a is given by the moisture content of air, x (kg water vapor/kg
dry air), and the total pressure,

x
p v,a p tot . (8.14)
x  0.62

The vapor partial pressure at the paper surface is

p v, p M pv0 , (8.15)

where pv0 is the partial vapor pressure for free water. This is given by
Antoine’s equation

190
 Chapter 8. Object-Oriented Modeling and Control of the Moisture
1500
1.0 o

Heat of sorption (kJ/kg)

30 C
o
Sorption isotherm
60 C
o
1000 90 C

0.5

30 C
o 500
o
60 C
o
90 C
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Moisture content (%) Moisture content (%)
Figure 8.5 Sorption isotherm, ij, and heat of sorption, ǻHs given by (8.17) and (8.21).

§ ·
¨ 10.127  1690 ¸
¨ T  43.15 ¸
© p ¹
p v 0 10 . (8.16)

As long as capillary transport can bring new water to the paper surface,
the vapor partial pressure at the paper surface is equal to the partial
pressure for free water. When the paper becomes dryer a correction factor
called sorption isotherm, ij, is invoked which has a value between zero
and one, see Figure 8.5. The sorption isotherm of a paper web depends on
its composition and temperature. It is not very well investigated when
compared to other materials [Pettersson and Stenström, 2000], but
[Heikkilä, 1993] gives an empirical expression for paper pulp, namely

M 1  exp( 47.58u 1.877  0.10085(T p  273) u 1.0585 ), (8.17)

where u is the moisture ratio (kg moisture/kg fiber). Also, let vx be the
speed of the paper web (m/s), dy the width of the paper web (m), Axy the
area of the dryer surface covered by paper (m2), and g the dry basis weight
(kg/m2). Then the mass balance of moisture in the paper web can be
written as

d (ugAxy )
d y v x guin  Axy qevap  d y v x gu. (8.18)
dt

Figure 8.6 shows a schematic picture of the mass flows in the model. To
model the energy balance, introduce

191
Chapter 8. Object-Oriented Modeling and Control of the Moisture
Axy qevap
vx
d y v x guin d y v x gu
Paper web

Figure 8.6 The mass balance for moisture in the paper web. The shaded area is the
cylinder wall

C p , fiber  uC p , w
C p, p , (8.19)
1 u

where Cp,p, Cp,fiber, and Cp,w is the specific heat capacity for the paper, fiber
and water, respectively (J/kg·K). As we can see, Cp,p is a weighted sum of
the heat capacities of the parts. From [Wilhelmsson, 1995] we have Cp,fiber
= 1256 J/(kg·K).
Also, let Tp be the paper temperature and ǻH be the amount of energy
needed to evaporate the water. Analogously to the discussion about the
mass balance, if the web is wet enough this energy is equal to the latent
heat of vaporization for free water. When the paper becomes dryer an
extra amount of energy ǻHs (the heat of sorption) is necessary besides the
latent heat of vaporization for free water. The heat of sorption can be
derived from the sorption isotherm by thermodynamic theory and this
relation is known as the law of Clausius-Clapeyron

R g ª d (ln M ) º
'H s  « », (8.20)
M w «¬ d (1 / T p ) »¼

and by applying this on (8.17), we get

M 1
'H s 0.10085u 1.0585T p2 R g . (8.21)
M wM

The amount of energy required to evaporate the water from the surface of
the web is the given by

'H 'H vap  'H s , (8.22)

192
 Chapter 8. Object-Oriented Modeling and Control of the Moisture
Axy qevap ( ΔH vap + ΔH s )

d y v x g (1 + u in ) C p , p T p ,in d y v x g (1 + u )C p , pT p

Paper web

Qp
Figure 8.7 The energy balance of the paper web. The shaded area is the cylinder wall.

where ΔHvap is the latent heat of vaporization, equal to 2260 kJ/kg (at
atmospheric pressure).
Reference [Pettersson and Stenström, 2000] investigates some sorption
isotherms found in the literature. Many of those give a heat of sorption
that goes to infinity as u goes to zero. This is physically unrealistic since
the bond energy between the last fraction of water and a cellulose fiber
must be finite. From [Heikkilä, 1993], a finite heat of sorption at the
origin which matches the hydrogen bond energy between water−fiber is
given and is therefore found to be most appropriate. The heat loss in the
paper due to mass evaporation is dominating the heat conduction and
radiation, which therefore can be neglected. In addition, since water is an
incompressible medium there is no pressure volume work on the
surroundings and we write the energy balance as a change in enthalpy.
The energy balance of the paper web is thus modeled as

d ( g (u + 1) Axy C p , p T p )
= Q p + d y v x g (1 + u in )C p , p T p ,in
dt (8.23)
− Axy q evap (ΔH vap + ΔH s ) − d y v x g (1 + u )C p , p T p ,

see Figure 8.7. In addition, we let the heat transfer coefficient from the
cylinder to the paper web depend on the moisture content in the web.
To summarize the complete drying section model, it is given by the
balances in (8.1), (8.2), (8.18), and (8.23), together with the algebraic
relations given in (8.3)−(8.9), (8.13)−(8.17), (8.19), (8.21), and (8.22).
Equation (8.1) and (8.2) define the dynamics of a cylinder, and one set of
these equations are needed for each cylinder in the drying section. The
dynamics of a lumped paper web model is given by (8.18) and (8.23). By
connecting a series of these equations a discretized model is obtained,

193
Chapter 8. Object-Oriented Modeling and Control of the Moisture
60

50

40
Sheet moisture (%)

30

20

10

0
5 10 15 20 25 30 35
Cylinder number
Figure 8.8 Validation by comparing steady-state simulations with moisture
measurements on a fine paper machine.

where the outflow of paper of one component becomes the inflow of the
next component.

8.2 Steady-state model validation

The two partial models in Section 8.1, are validated separately in [Slätteke
and Åström, 2005] and [Persson, 1998]. To assure that the combined
model is satisfying it has been validated against steady-state data taken
from a paper machine producing fine paper [Stenström, et al, 1994]. The
machine is running at 708 m/min with a basis weight of 80 g/m2. The
paper is over dried to a final moisture content of only 0.012 kg/kg, since
this is a predryer which is immediately followed by a size press (a unit
where starch is applied to the surface to obtain strength and water
resistance) where a certain amount of rewetting occurs.
The moisture content is measured in 11 different positions, which can
be seen in Figure 8.8 together with the simulation result. The heat transfer
coefficient Įsc is used as fitting parameter with the same value in all
cylinders and 1100 W/(m2·K) is found to give the best fit by visual
inspection. The agreement between model and measurements is good even

194
 Chapter 8. Object-Oriented Modeling and Control of the Moisture

Pres s
Pope

Mois ture controller

Header pressure
HeaderPres...
F

P
k={1000e3}

ID
PID i P

Figure 8.9 The Modelica model of the drying section used in the simulations. The
component of the second dryer group is expanded, showing the cylinders, steam valve, and
PID controller. It is assumed that the temperature and moisture are constant at a single
cylinder due to the high machine speed, hence modeled as one control volume.

though this is rough validation without using all possible degrees of

freedom, e.g. by assuming different Įsc in different cylinders. The model
also captures the three zones in the drying process, the heating phase, the
constant drying rate phase, and the falling drying rate phase [Karlsson,
2000]. In the heating phase (cyl. 1í5) most steam energy is used to heat
the web and the evaporation rate is low. In the constant rate phase (cyl.
6í30), energy to the web is equal to the energy consumed for water
vaporization. In the falling rate phase (cyl. 31í37), drying rate begins to
decrease due to the hygroscopic nature of the fibres.

195
Chapter 8. Object-Oriented Modeling and Control of the Moisture
8.3 Open loop simulations

A simulation model of a fictitious drying section is used in simulations to

investigate different properties. It consists of 60 drying cylinders divided
in 7 drying groups, see Figure 8.9. The machine speed is chosen to
1100 m/min, the machine width is 7.15 m, and the dry weight of the paper
sheet is 77 g/m2. This is also the drying section model used in Section 8.5
to evaluate MPC control of the moisture.
In Chapter 4 the IPZ model is deduced from physical relations. It is
concluded that a slow pole in the process dynamics can be regarded as an
integrator if the heat transfer coefficient of the steam-condensate interface
is much larger than the heat transfer coefficient of the cylinder-paper
interface, Dsc >> Dcp, see (4.3) and (4.4). In Figure 8.10 response in steam
pressure during a step in valve position is shown for two different cases of
heat transfer coefficient Dcp. This implies that in this analysis Dcp is set to a
constant value instead of the linear relationship (8.6) to define it
independently of u. It is clear from Figure 8.10 that for the case with
Steam pressure (kPa)

550

500
D cp = 500

450
Steam pressure (kPa)

150

140
D cp = 4000

130

75
Valve position (%)

70
65
60
55
100 150 200
Time (s)
Figure 8.10 Pressure response for a step in valve position for two different values of Dcp.
Heat transfer coefficient Dsc = 4000.

196
 Chapter 8. Object-Oriented Modeling and Control of the Moisture

Mass flow in/out (kg/s)

0.24

0.20
qs
q c+ q bt
0.16

510
Steam pressure (kPa)

500

490

480

470

460
90 100 110 120 130 140 150 160 170 180 190 200
Time (s)

Figure 8.11 Open loop step response in pressure with Dsc = 4000 W/(m2·K). Heat transfer
coefficient Dcp is set to a constant value 500 W/(m2·K).

450
Q p, Q m (kW)

Qm
400 Qp

350

250
Energy derivatives (kJ/s)

d( U su sV s )/dt
200
d( mC p,mT m )/dt
150 d( U wu wV w )/dt
100
50
0
-50
100 150 200 250
Time (s)
Figure 8.12 Open loop step response in pressure with Dsc = 4000 W/(m2·K). Heat transfer
coefficient Dcp is set to a constant value 500 W/(m2·K).

Dcp = 500 W/(m2·K), the process can be modeled with an integrator plus
some dynamics for the given time frame (in reality we know that the
integrator is a slow pole). However, for the response with Dcp = 4000
W/(m2·K) an IPZ model is not sufficient. Note that since the heat transfer
coefficients are different in the two figures, the steady-state solutions are

197
Chapter 8. Object-Oriented Modeling and Control of the Moisture
also different and the two simulations start from different steady-state
steam pressures. The conclusion is unaffected if the two responses start at
the same pressure.
Figure 8.11 gives an explanation for the characteristic appearance of
the open loop step response of the steam pressure. When the steam valve
position is increased, it will increase the steam pressure inside the cylinder
and consequently the saturation temperature of the steam and the
condensation rate. However, the increasing condensation rate lags behind
the increasing steam inlet flow as the condensate layer heats up to the new
steady state temperature. Therefore, there will be a fast initial build-up in
steam pressure, before the steam consumption has reached its new value.
This is also noted in Chapter 3.
Figure 8.12 shows the energy flows to the cylinder and paper, and also
the energy derivatives. In the derivation of the linear model in Chapter 4,
it is simplified by assuming that the thermal dynamics of the water is fast
compared to the dynamics of steam and metal. The energy balance is
therefore replaced by a static model by defining its derivative as zero.
9.5
Sheet moisture (%)

9.0

8.5

325
Steam pressure (kPa)

320
315
310
305
300

80
Valve position (%)

60

40

20

0
0 100 200 300 400 500 600 700 800 900 1000
Time (s)
Figure 8.13 Simulation of set point change in steam pressure and a web break.

198
 Chapter 8. Object-Oriented Modeling and Control of the Moisture
Figure 8.12 shows that this assumption is valid. Since the cylinder shell
acts as a thermal filter Qp is increasing more slowly that Qm.
Figure 8.13 shows a simulation where the steam pressure controllers
are in closed loop while the moisture is uncontrolled. The pressure is
controlled by a PI controller tuned according to the method in Chapter 5
with Ms = 1.2. At t = 100 s there is a steam pressure set point change. The
valve position returns to its initial position when the set point is reached
due to the integral action in the process. Note that the characteristic over-
shoot in steam pressure does not result in an over-shoot in the moisture.
At t = 500 s there is a web break and the moisture quickly reaches a level
around 60 % since the sheet has lost the energy source for its evaporation,
and the curve is out of the scale in the figure. Since the main energy sink
is lost for the cylinders, the control valve is almost closed to keep the
desired steam pressure.
Figure 8.14 shows two simulations of moisture response for a set point
change in steam pressure: One with a medium heat transfer coefficient
and one with a high heat transfer coefficient. Modeling the two responses
as a first order process, they are given by

0.109 16.1s
P(s)  e D sc 2000 W/m 2 K (8.24)
46.2 s  1

and

0.0917 16.6 s
P( s)  e D sc 4000 W/m 2 K (8.25)
21.2s  1

The difference in dynamics between the two processes is significant and

will greatly affect the performance of the closed loop system. This clearly
shows that effective condensate removal and the cylinder design in terms
of spoiler bars are important.
The simulation time for the figures in this section is a few minutes.
This should be compared to the more complex model used in Chapter 6
and Chapter 7, where the simulation time is counted in days.

8.4 Control of moisture by mid-range MPC

In the last decade, MPC (model predictive control) has received large
attention in the process industry. It has been described as one of the most

199
Chapter 8. Object-Oriented Modeling and Control of the Moisture
8.6

Sheet moisture (%) 8.4

8.2

8.0
Dsc = 2000 W/m2K
7.8

6.2
Sheet moisture (%)

6.0

5.8

5.6 Dsc = 4000 W/m2K

Steam pressure set point (kPa)

385

380

375

370
450 500 550 600 650 700 750 800
Time (s)
Figure 8.14 Response in moisture to step change in steam pressure set point for two
different values of heat transfer coefficient Dcp.

Setpoint Setpoint Steam

Moisture steam pressure pressure
r uc
MPC PID- Steam Moisture
controller system
Dryer
y

Figure 8.15 Moisture control by single loop MPC.

200
 Chapter 8. Object-Oriented Modeling and Control of the Moisture
Setpoint
moisture PID- Steam system
uc1 controller First part
r1
MPC Dryer
Moisture
y

PID- Steam system

uc2 controller Last part

Figure 8.16 The proposed moisture control loop.

significant developments in process control [Doyle, 1999] and the only

methodology to handle constraints in a systematic way [García, et al,
1989]. In this section, a new strategy to control the moisture in paper
production is evaluated. It is implemented in a MPC structure and the
analysis is done by simulations of the paper machine model shown in
Figure 8.9. A Matlab toolbox for MPC [Åkesson, 2003] is linked to the
Modelica environment. In this way, the advantages of the simple
modeling technique in Modelica and the rich family of toolboxes in
Matlab are used.
Figure 8.15 shows a block diagram where the moisture is controlled by
MPC. The steam pressure set point is distributed to the dryer groups as
described in Chapter 2, see Figure 2.12. The performance of this closed
loop system is limited by the long transport dead-time in the drying
section. By manipulating the steam pressure in the cylinders of the last
part of the machine independently of the first part, a more effective
moisture control system can be derived, see Figure 8.16. The objective is
still to control the moisture in the sheet with the steam pressure in the
cylinders but now the process has two inputs (and the same output as
before) and this extra degree of freedom can be taken advantage of.
By identification of step responses on the high-order nonlinear
physical model, a simple black-box model is achieved. In the Laplace
domain, it is given by

0.098 14 s 0.010 5 s

Y  e U c1  e U c2 , (8.26)
48s  1 48s  1

where y is the moisture content, uc1 is the steam pressure set point to the
first 50 cylinders, and uc2 the set point to the last 10 cylinders. The PID-
controllers in the inner loops are tuned according to the IPZ-tuning

201
Chapter 8. Object-Oriented Modeling and Control of the Moisture
derived in Chapter 5. The response from uc2 has a significantly shorter
time delay but also smaller process gain. The advantage of the high gain
from uc1 and fast dynamics from uc2 is utilized in a mid-ranging MPC
structure.
The simplified model for the singe-loop case in Figure 8.15, obtained
correspondingly to (8.26), is given by

0.11 13s
Y  e Uc , (8.27)
48s  1

The cost function being minimized in the mid-ranging MPC is

2 2
49 3
ª r1 (k  i )  yˆ (k  i | k )º ª 'uc1 (k  i ) º
J (k ) ¦
i 0
« r (k  i)  u (k  i) »
¬ 2 c2 ¼
 ¦
i 0
« 'u (k  i )» , (8.28)
¬ c2 ¼ R
Q

where r1 is the set point for the moisture, and r2 the set point for uc2.
Notation ǔ(k + i | k) denotes the i step-ahead prediction and ǻ is the
difference operator. Since the MPC formulation is inherently discrete, the
cost function is given as a summation. The idea of (8.28) is to let uc2 take
care of the variations in paper moisture and let uc1 be positioned at a level
where uc2 has an adequate control range in both directions in steady-state.
In this way, the first part of the drying section serves as the base level of
the drying while the last part controls the moisture. This is done by
choosing appropriate weighting matrices, Q and R. For the simulations in
this chapter they are chosen as

ª500 0 º ª100 0 º
Q « 0 10 3 », R « 0 10  4 » . (8.29)
¬ ¼ ¬ ¼

There is obviously a larger weight on deviations from r1 than from r2,

since moisture control is the main objective. The weight on ǻuc1 is larger
than on ǻuc2 making the controller to primarily use signal uc2 when acting
on disturbances in moisture or on set point changes. The MPC settings in
(8.29) are found from a combination of these ‘rules of thumb’ and
evaluation of performance by simulations. The sample time is 5 s, the
prediction horizon is chosen to 50 and the control horizon is chosen to 4,
see (8.28). The prediction horizon is set to approximately match five time
constants of the open-loop system (to assure that the prediction ‘sees’ the

202
 Chapter 8. Object-Oriented Modeling and Control of the Moisture
7.6

y (%) 7.4

7.2

7.0

450
u c1 (kPa)

440

430

600
u c2 (kPa)

500

400
0 200 400 600 800 1000 1200
Time (s)
Figure 8.17 Simulation of mid-range MPC. There is a set point change at t = 100 s and a
disturbance in inlet moisture content to the drying section from 60.0 % to 60.5 % at

7.6

7.4
y (%)

7.2

7.0

450
uc (kPa)

440

430
0 200 400 600 800 1000 1200
Time (s)
Figure 8.18 Simulation of single loop MPC. A set point change occurs at t = 100 s and a
disturbance in inlet moisture content from 60.0 % to 60.5 % at t = 600 s.

203
Chapter 8. Object-Oriented Modeling and Control of the Moisture
1.5
Magnitude (abs)

0.5

0 -4 -3 -2 -1
10 10 10 10
Frequency (Hz)
Figure 8.19 The sensitivity function for single loop MPC (dotted) and mid-range MPC
(solid), using the linearized model (8.26). The maximum sensitivity is chosen to 1.25. The
difference in limit frequency for the two systems is around a factor two.

full response of a change in ǻuc1 and ǻuc2) and a small control horizon to
limit the computational effort for the MPC. There is also a rate constraint
for the control signals

'uc1 d 10 kPa , 'uc 2 d 50 kPa. (8.30)

The purpose of this is to avoid severe injections of disturbances into the

steam system by allowing large variations in steam usage, which has
negative effect on both steam production and other steam users.
Figure 8.17 shows a simulation of the mid-range MPC and it clearly
visualizes the thought of mid-ranging. The signal uc2 is used to quickly
react to set point changes and disturbances while uc1 is used to push uc2
back to its set point in steady-state. Both during the set point change and
disturbance, the rate constraint for uc2 is initially active. This reduces the
performance of the controller slightly but is important for the steam
consumption, as described above. Simulations show that leaving out the
constraints gives a more aggressive use of uc2, but it is of course not ‘free’
to use since there still is a cost from (8.28).
To make an evaluation of the proposed control structure, the mid-
ranging MPC is compared to single-loop MPC, see Figure 8.18. The
tuning is chosen so that the two control systems have the same maximum
value of the sensitivity function, see Figure 8.19. This implies that they, in
some sense, have the same degree of robustness. However, due to the
constraint handling, MPC is nonlinear and the comparison only serves as
guidance. The cost function for the single-loop MPC is given by

204
 Chapter 8. Object-Oriented Modeling and Control of the Moisture
7.6

Moistuure (%)
7.4

7.2

7.0

6.8

20
Steam consumption (kg/s)

19

18

17

16

15
0 200 400 600 800 1000 1200
Time (s)
Figure 8.20 Comparison between mid-range MPC (solid) and single loop MPC (dotted),
showing the moisture content and steam consumption.
50 3

¦ >r (k  i)  yˆ (k  i | k )@ Q  ¦
2 2
J (k ) 'uc (k  i ) R , (8.31)
i 1 i 0

and the chosen weights that give the sensitivity in Figure 8.19, and
constraint are

Q 10, R 1, 'uc d 10 kPa (8.32)

Figure 8.19 also indicates the difference in performance between the two
controllers, at least in the non-constrained case. The mid-range MPC has a
limit frequency almost twice as large as the limit frequency of the single
loop MPC. Since paper machines often are run from several hours to days
with the same set point, disturbance rejection is important and it therefore
makes sense to look at this property.
Figure 8.20 shows the moisture response for the two different
controllers together with the steam flow in the header. The difference in
performance is apparent. The mid-range MPC has both better set point
following and disturbance rejection. However, the transient steam

205
Chapter 8. Object-Oriented Modeling and Control of the Moisture
7.6
Moisture (%)
7.4

7.2

7.0

6.8

25
Steam consumption (kg/s)

20

15

10
0 200 400 600 800 1000 1200
Time (s)
Figure 8.21 Comparison between mid-range MPC (solid) and single loop MPC (dotted),
showing moisture content and steam consumption when tuned to give similar performance
in disturbance rejection.

consumption is twice as large for the mid-range MPC. This is the price
being paid for the extra performance.
In Figure 8.21, the single loop MPC has been tuned to give similar
performance in disturbance rejection as the mid-range MPC, shown in
Figure 8.17 and Figure 8.20. The output of the single loop MPC is in this
case much more aggressive than it was in Figure 8.18 and therefore the
transient steam consumption is much larger. This to be able to match the
performance of the mid-range MPC. The maximum value of the
sensitivity is also larger, hence the control system is less robust, see
Figure 8.22.
The advantage with the proposed control structure is that, in general, it
does not require any rebuild of the drying section. The cylinders are
normally divided into different groups even though all groups follow the
same set point. Therefore, it is simply a matter of changing the controller
software and a majority of the main system vendors have the possibility to
include a MPC package into their DCS system. However, it is an
advantage if the flash steam from the last group is recirculated through a
thermo compressor, see Figure 2.5. If the flash steam is reused by another
group it is important to let the steam pressure of the last group be
constrained so that its pressure never falls below the steam pressure of the

206
 Chapter 8. Object-Oriented Modeling and Control of the Moisture

1.5

Magnitude (abs)
1

0.5

0 -4 -3 -2 -1
10 10 10 10
Frequency (Hz)
Figure 8.22 The sensitivity function for single loop MPC (dotted) and mid-range MPC
(solid), when tuned to give similar performance

receiving group. This guarantees that the steam flows in the intended
direction.
This mid-ranging MPC has two essential advantages compared to the
mid-ranging structure used in Chapter 6, see Figure 6.10. Firstly, the MPC
does not require any decoupling filter, CD, since the complete multi-
variable process model is given in the cost function. The MPC therefore
automatically balances the two process inputs to get a desired output and
there is no need for the extra ‘fix’ with a decoupling filter between the two
process inputs as in Figure 6.10. Secondly, the MPC has a systematic
implicit way to treat constraints. For the structure in Chapter 6, this has to
be taken care of explicitly and not as a part of the tuning phase. A
disadvantage can be that it is a bit more difficult to tune a mid-range MPC
compared to e.g. the IMC or Dahlin controller used by many mills today.

8.5 Summary

A physical model, implemented in the object-oriented modeling language

Modelica, for a drying section has been given. Components for steam
cylinder, control valve, paper web, and different moisture controllers have
been developed and collected in a model library. All equations are based
on mass and energy balances, and algebraic constraints. By drag-and-drop
features it is easy to build a simulation model of virtually any existing
drying section. It is also easy to expand the model library with
components for press and wire section.
The model has been validated against measurements on a real paper
machine. By simply adjusting one parameter, the heat transfer coefficient
for the condensate, a good fit is obtained.

207
Chapter 8. Object-Oriented Modeling and Control of the Moisture
The model is linked to a MPC toolbox in Matlab to evaluate a new
strategy to control the moisture in the drying section. The strategy utilizes
the possibility to divide the drying section into two parts. By controlling
this multi-variable process with a mid-range structure, the performance of
the closed loop system is greatly improved. The mid-ranging MPC is
compared with single-loop MPC and the evaluation is done by comparing
performance in terms of sensitivity function, disturbance rejection and
steam consumption. An important advantage of the mid-ranging MPC is
that it does not require any rebuild of the physical process.
The advantage of this model compared to the one used in Chapter 6
and Chapter 7 is the simulation time, which is a few minutes instead of a
few days. The model used in the previous two chapters is capable of
simulating microscopic phenomena like water and vapor transport, and
moisture gradients, inside the paper sheet. The model described in this
chapter neglects this and assumes a homogeneous paper in the thickness
direction. For the purpose of the feedback control investigated in this
thesis, this assumption sufficient. However, the feedforward control in
Chapter 7 requires a model that takes differences in the thickness direction
into account to observe the peak position.

208
 Chapter 9. Conclusions

9.
9
Conclusions

9.1 Summary

In this thesis, modeling and control of the paper machine drying section is
presented. The treatment is divided into two main parts, control of the
steam and condensate system, and moisture control.

The steam and condensate system

The main task for the steam and condensate system is to deliver sufficient
amount of energy to the evaporation process in the paper sheet. It is
shown that the process response from the steam valve to steam pressure
inside the drying cylinder can be described by a linear second order black-
box model. Written as a transfer function, it consists of a pole, a zero, an
integrator, and a time delay. The model structure is referred to as the IPZ
model. A simple graphical identification procedure is given that estimates
the four model parameters from two asymptotes of an open loop step
response.
From the black-box model, a tuning rule for PI and PID control is
derived. It is based on optimized load disturbance rejection subject to a
robustness constraint and is referred to as IPZ tuning. To give the user the

209
Chapter 9. Conclusions
option to balance between robustness and performance the tuning rule has
a design parameter. A nice property of the tuning rule is that the controller
settings are easily calculated from the process parameters of the IPZ
model. It is successfully verified from experiments on real paper machine
at different mills. The IPZ structure can be used in many other process
areas, apart from modeling the steam pressure in cylinders. Therefore the
tuning method has a much wider field of applications than solely drying
section control.
The IPZ model is used to examine some properties of the closed loop
system and to investigate two different control structures. The main
objective is to avoid resonance peaks in the closed-loop frequency
response from set point to process output, and this is accomplished by a
set-point feedforward controller (two-degree-of-freedom controller) and a
state feedback controller. The two types of structures solve the objective
differently but they roughly give the same performance. However, the
two-degree-of-freedom controller is recommended since it is best suited
for implementation in today’s industrial control systems. As this controller
structure separates the servo and regulation problem it also fits well to the
IPZ tuning. The set-point feedforward design gives the desired response
from set-point to output, while a PID controller based on IPZ tuning takes
care of the disturbance rejection in the feedback loop.
By making some assumptions and a linearization, it is shown that the
black-box model can be derived from a few simple mass and energy
balances. This means that the process parameters of the black-box model
are directly related to physical quantities as cylinder dimension, valve
characteristics, and steam properties. The model is validated against
measurements from a real paper machine and the heat transfer coefficient
from steam to cylinder is used as calibration parameter.

Moisture control
Traditionally the moisture in the paper sheet is controlled by manipulating
the pressure in the steam cylinders. To improve the performance of the
closed loop moisture control system both feedforward and feedback is
investigated. A new measurement signal, called the peak position is
introduced. It is based on measurements of the paper surface temperature
and indicates the position where the surface becomes dry. It is shown that
peak position can significantly improve the control performance by
feedforward.
The feedback is carried out by two types of mid-ranging controllers.
The first uses the air supply system in combination with the steam system.

210
 Chapter 9. Conclusions
The advantage of the fast impact from the supply air and the high capacity
of the steam system are combined to enhance the performance of the
closed loop system. The second mid-ranging structure utilizes similar
advantages by combining the last drying group with the remaining part of
the dryers to improve performance. Two different of mid-ranging
controllers are used. The first mid-ranging structure combines two single-
loop controllers, in this case IMC, together with a decoupling filter and a
method to tune the system is given. The second structure is implemented
in a MPC. The advantages of the mid-ranging MPC are that the
decoupling filter is not needed and that constraints are explicitly taken
into account.
From mass and energy balances, a dynamic simulation package of a
drying section is implemented in the object-oriented modeling language
Modelica. By drag-and-drop features, a simulation model of practically
any drying section can easily be built. It can be used to investigate effects
of a rebuild, different controller structures, and different ways to operate
the machine. All this can be done for web breaks, grade changes, start-
ups, and normal run.

9.2 Future work

The proposed solutions for moisture control, two mid-ranging controllers

and a feedforward structure, have to this point only been verified by
simulations. The next step is to evaluate them experimentally on a pilot or
full-scale paper machine.
The simulation package in Modelica contains components for cylinder
and paper sheet. However, it lacks models for the pipings, condensate
tanks, and hood. By adding components for these objects, cross-couplings
between cylinders, caused by the cascade configurations in the condensate
system, can be investigated. The modeling package would also benefit
from more validation with real plant data. It is desirable to validate the
model for a number of different operating points to assure that it is valid
in a wide operating range.
The modeling package can also be used for optimization. Some
examples of this are minimization of the total energy usage by finding the
optimal steam distribution between the dryer groups; and minimization of
the time for a grade change by finding the optimal trajectories for machine
speed, basis weight, and steam pressure, with respect to given rate
constraints.

211
Chapter 9. Conclusions
The differential pressure loop has only been treated briefly. A deeper
analysis and a tuning rule, like the one for the pressure loop, are desired.
This loop will probably benefit from a feedforward structure since it is
greatly affected by the operation of the pressure loop.
From the linear model of the cylinder dynamics there is a potential of
making a recursive identification of the heat transfer coefficient for fault
detection with respect to condensate evacuation. This would be very
beneficial since the heat transfer coefficient have a large impact on the
dynamics of the moisture process and consequently its control
performance.
A systematic analysis of moisture control where different control
paradigms as MPC, Smith-predictor, IMC, PID, deadbeat control,
minimum-variance, or LQG, are compared would be valuable. Every
control system vendor use their control structure but no thorough
comparison have been published in the literature.

212
 Appendix A. Glossary

Appendix A
Glossary

This glossary contains explanations from the following websites:

www.afandpa.org, www.internationalpaper.com,
www.instituteofpaper.com, and www.paperonline.org.

Ash content: The amount of residue when a sample of paper is burned

under controlled conditions so that all ignitable matter is removed. The
ash derives from the mineral loading in the paper.
Basis weight: Weight in grams of one square metre of paper or board;
also called grammage.
Black liquor: Mixture of cooking chemicals and dissolved wood
material remaining after sulphate cooking; recovered during pulp
washing, concentrated by evaporation and burned in the recovery boiler
to regenerate the cooking chemicals and generate energy.
Board: Generic term for stiff paper usually made in several layers
with a substance normally varying from 160 to 500 g/m2, for certain
grades even higher; widely used for packaging (e.g. folding cartons) and
graphic applications.
Broke: Papermakers own waste paper created during papermaking
process. It is usually repulped.
Bulk: Specific volume, inverse of density of the paper.

213
Appendix A. Glossary
Caliper: The thickness of a sheet of paper or board. Also known as
thickness or bulk. Usually measured in nanometers.
Chemical pulp: Pulp in which wood fibres have been separated by
chemical, rather than mechanical, means.
Chemi-thermomechanical pulp (CTMP): Chemi-mechanical pulp
produced by treating wood chips with chemicals (usually sodium
sulphite) and steam before mechanical defibration.
Coated paper: The uniform application of a coating yields a more
even and more closed surface of printing papers, which is suitable for the
reproduction of fine screen artwork. The coating is applied in separate
coaters or in the paper machine.
Conditioned basis weight: dsd
Corrugated board: Formed fluting that is faced with a liner on both
sides.
Curl: The degree of curvature, measured when paper is positioned
flat.
Dry end: Final part of the paper machine from the drying section
onwards.
Dry weight: Mass of paper sheet excluding the moisture.
Fabric tension: The tension of the fabric creates contact pressure
between the wet web and cylinder. Higher contact pressure gives higher
heat transfer.
Filler: A material or substance that is added to the furnish to smooth
out the spaces between fibers, enhancing the printing properties of the
paper.
Fine paper: A broad term including printing, writing, and cover
papers, as distinguished from wrapping papers and paper not generally
used for printing purposes, which are generally referred to as coarse
papers.
Fines: Small fragments of fibres produced, for example, in the course
of beating or refining.
Fluting: Paper that has been formed into the flutes that make up the
ridged part of the corrugated board between the liners.
Formation: The arrangement of fibers in a sheet of paper; can be seen
by holding it up to a light source.
Freeness: A measure of the drainability of an aqueous suspension of
stock, determined and expressed as specified in a standard method of test.
Furnish: The ingredients or constituents of which the paper is made.
Fibrous furnish concerns only the types and proportions of the fibres
present.

214
 Appendix A. Glossary
Grade: (1) A class or level of quality of a paper or pulp which is
ranked, or distinguished from other papers or pulps, on the basis of its
use, appearance, quality, manufacturing history, raw materials, or a
combination of these factors. Some grades have been officially identified
and described; others are commonly recognized but lack official
definition. (2) With reference to one particular quality, one item (q.v.)
differing from another only in size, weight, or grain; e.g., an offset book
paper cut grain long is not the same grade as the same paper cut grain
short.
Fiber orientation: The direction of the fibers in paper.
Kraft paper: A paper made essentially from wood pulp produced by
a modified sulfate pulping process. It is a comparatively coarse paper
particularly noted for its strength, and in unbleached grades is primarily
used as a wrapper or packaging material. It can be watermarked, striped,
or calendered, and it has an acceptable surface for printing. Its natural
unbleached color is brown but by the use of semibleached or fully
bleached sulfate pulps it can be produced in lighter shades of brown,
cream tints, and white. In addition to its use as a wrapping paper, it is
converted into such products as: grocery bags, envelopes, gummed
sealing tape, asphalted papers, multiwall sacks, tire wraps, butcher wraps,
waxed paper, coated paper, as well as specialty bags and sacks.
Liner: A paper that is used as the facing material in the production of
corrugated and solid fibre shipping containers.
Mechanical pulp: Pulp consisting of fibres separated entirely by
mechanical rather than chemical means.
Newsprint: A lightweight paper, made mainly from mechanical wood
pulp, engineered to be bright and opaque for the good print contrast
needed by newspapers. Newsprint also contains special tensile strength
for repeated folding. It does not includes printing papers of types
generally used for purposes other than newspapers such as groundwood
printing papers for catalogs, directories, etc.
Opacity: The ability of a sheet of paper to prevent light transmission
through it. Opacity prevents print that is on one side of a sheet of paper
from showing through to the other side.
Packaging papers: These papers are used to wrap or package
consumer and industrial products such as grocer's bags and sacks,
shopping and merchandise bags, and multiwall shipping sacks used for
shipping such products as cement, flour, sugar, chemicals and animal
food. "Specialty" packaging papers are used for cookies, potato chips, ice
cream, and similar products.

215
Appendix A. Glossary
Paper Stock: A mix of pulp fibers, water, additives, chemicals and
dyes that will be pumped onto the paper machine to form paper.
Picking: Fibers in the paper which tend to pull away from the surface
during the drying process.
Recovery boiler: In wood pulping, a unit for concentrating black
liquor to a stage where the residual carbon is then burned out and the
inorganic sodium salts melted and recovered.
Refiner: A machine, usually equipped with discs or with a cone and
plug, intended for the treatment of fibrous materials in an aqueous
medium to give them some of the properties needed for the manufacture
of pulp or paper with the necessary characteristics.
Retention: Proportion of fibre and filler retained on the paper
machine wire.
Sheet (or Web): A continuous length of paper, such as paper when it
is on the paper machine or on roll-feed equipment.
Sizing: This process can either be applied on the surface of the sheet
or in the sheet: in the first case starch is applied to the surface to increase
its strength and to resist the penetration of oil-based inks (this process is
carried out at the size press, which is about two-thirds of the way down
the dry end); in the second case chemicals are added to the stock at the
pulping stage before the sheet is formed: this is called internal or engine
sizing and its purpose is to stop penetration of water-based inks into the
sheet.
Smoothness: It is a measured property of paper that describes or rates
the flatness and evenness of a sheet's surface.
Stock: The wet pulp before it is fed onto a paper making machine, or
during the paper making processes before it becomes a sheet of paper;
contains around 99% water and 1% fibre.
Stock preparation: Collective term for all treatment necessary for the
preparation of the stock before it reaches the making machine.
Tissue: Collective term for papers of a grammage of less than 30 g/m2
that differ in application and composition but have the common feature of
being thin. Examples of different types of tissue papers include sanitary
grades such as toilet, facial, napkin, towels, wipes, and special sanitary
papers. The extremely thin Japanese tissue papers are sometimes
produced in grammages as small as 6 to 8 g/m2. Other examples of tissue
papers are decorative and laminated tissue papers and crepe papers, often
used in gift wrapping and to decorate. Desirable characteristics here are
appearance, strength, and durability.
Wet end: First part of the paper machine up to the drying section.
White water: The filtrate from the wet end of the paper machine
216
 Appendix A. Glossary
Winding: An operation in which the paper or board from the paper
machine is slit and wound into the roll widths ordered by the customer.
Woodfree paper: Paper consisting of chemical pulp fibres. It does
not contain any mechanical pulp beyond a permissible content of 5 % by
mass.

217
Appendix B. Conservation Balance for Compartmental Models

Appendix B
Conservation Balance for
Energy in Compartmental
Models

The models developed in Chapter 4 and Chapter 8 are based on mass and
energy balances, together with a few constitutive relations. The balance
equations come from the fundamental conservation principle in physics,
that mass and energy are neither destroyed nor created but simply change
their form. Putting up a mass balance is rather straightforward but the
energy balance is not always so evident, particularly not for gas systems.
In this appendix the energy balance for the steam and condensate in the
cylinder model is discussed. It will be shown why change in internal
energy depends on in- and outflow of enthalpy, see Figure A.1.

control volume

d
qin hin U i  qout hout
dt

Figure A.1 The general energy balance for steam and condensate, used in this thesis. q is
mass flow, h specific enthalpy, and Ui internal energy. The change in internal energy
depends on inflow and outflow of enthalpy.
218
 Appendix B. Conservation Balance for Compartmental Models
B.1 The energy balance

Starting from a general energy balance, the total energy of a system

comprises of three principal components

x internal energy Ui
x kinetic energy Ek
x potential energy Ep

Observe that we here use internal energy Ui [J] and not specific internal
energy ui [J/kg]. Energy flows into the system in terms of

 heat input (conductive energy flow)

 work input (mechanical work)
 energy brought into the control volume by the incoming fluid
(convective energy flow)
 work done on the control volume by the incoming fluid (flow
work)

and energy flows out of the system in terms of

 heat output
 work output
 energy leaving the control volume with the outgoing fluid
 work done by the outgoing fluid

Conductive energy flow

This is energy that flows in and out of the system through conductive heat
transfer mechanisms. We call this Qc [J/s].

Convective energy flow

This is the energy carried by means of the mass flow into and out of the
control volume. It consists of internal, kinetic, and potential energy.

Convective inflow: qin (ui ,in  ek ,in  e p ,in )

Convective outflow: qout (ui ,out  ek ,out  e p ,out )

219
Appendix B. Conservation Balance for Compartmental Models
where qin and qout are mass flows [kg/s], and ek and ep are specific energies
[J/kg].

Mechanical work
This is denoted as Wm [J/s] and can be both shaft work and expansion
work. Shaft work is the work done by the fluid or surrounding through a
moving part (e.g. a pump rotor). Expansion work accounts for work done
in expansion or contraction of the control volume. This term is often not
present especially not when the control volume is fixed. Work done on the
system is positive work.

Flow work
This term relates to the work done on the fluid as it moves into and out of
the control volume. This is expressed as

Work by inflow: qin pin vin

Work by outflow: qout pout vout

where pin and pout are pressures [Pa] at inlet and outlet, and vin and vout are
specific volumes [m3/kg] at inlet and outlet. To see this, consider a unit
volume of fluid across the boundary at the input, see Figure A.2. Let the
element have a length l [m/kg] and the cross-sectional area of the inlet be
A [m2] Then the energy required to push the element across the boundary
is given by the force times length,

( pin Ain ) u l pin vin

Similarly it can be shown that the flow work at the outlet is given by

pout vout

Putting it all together

The principle of conservation of energy gives us

d
(U i  Ek  E p ) qin (ui ,in  ek ,in  e p ,in )  qout (ui ,out  ek ,out  e p ,out )
dt
 qin pin vin  qout pout vout  Qc ,in  Qc ,out  Wm ,in  Wm ,out

220
 Appendix B. Conservation Balance for Compartmental Models
The sum of specific internal energy and the pv term is called specific
enthalpy, and is given by the symbol h (h = u + pv). We can therefore
write the energy balance as

d
(U i  Ek  E p ) qin (hin  ek , in  e p ,in )  qout (hout  ek , out  e p , out )
dt
 Qc , in  Qc , out  Wm , in  Wm , out

When the steam cylinder model is developed in Chapter 4, the kinetic and
potential energies are neglected, which is applicable to most systems in
the process industry. Also, there are evidently no mechanical work terms
acting on the system and the internal energy of the control volume is
written as specific internal energy times mass. It is then simplified to

d
( U uiV ) qin hin  qout hout  Qc , in  Qc , out
dt

where U is density [kg/m3] and V volume [m3].

control volume
inlet

Figure A.2 A small section at inlet to the system. There is an expenditure in energy to
push the fluid across the boundary.

221
Appendix B. Conservation Balance for Compartmental Models
If we are dealing with an incompressible liquid, there is no pressure-
volume work done on the system, and the energy balance can be further
simplified to

d
( U hV ) qin hin  qout hout  Qc , in  Qc , out
dt

This is the case when considering the water phase (moisture) in the paper
web.

222
 Appendix C. Solution to the One Dimensional Heat Equation

Appendix C
Solution to the One
Dimensional Heat Equation

The general heat equation is a partial differential equation, describing heat

conduction in solids in a given region over time, of the following form

wT
Uc  ’ ˜ ( O ’T ) k , (C.1)
wt

where ȡ (kg/m3) is the density, Ȝ (W/m·K) is the thermal conductivity, c

(J/kg·K) the specific heat capacity, and k (W/m3) the added heat per unit
volume and time. Assuming homogeneous material, no added or
subtracted heat inside the given region, and a one-dimensional problem, it
can be written as

wT w 2T
a 2 0, (C.2)
wt wx

where a (m2/s) is the thermal diffusivity, defined as

O
a . (C.3)
Uc

223
Appendix C. Solution to the One Dimensional Heat Equation

T2

T1
Steam side Paper side
T0
T

L
x

Figure C.1 At t = 0, the temperature on the steam side is changed from T1 to T2. The two
steady-state solutions are shown in the figure.

Heat transfer through a cylinder wall with a step change

in temperature
A steam filled cylinder with wall thickness L, has the temperature T1 on
the steam side and temperature T0 on the paper side, in steady-state. At
t = 0 the steam temperature is suddenly changed to T2, see Figure C.1.
Also, the heat flow to the paper is assumed to be Q (W/m2) . Thus, the
system is given by

­ wT w 2T
°  a 0, 0  x  L, t ! 0
° wt wx 2
°T ( x,0) T  x
° 1 (T0  T1 )
L (C.4)
®
°T (0, t ) T , t !0
2
°
° wT ( L, t )
° O Q
¯ wx

From theory of Hilbert Spaces and eigenfunctions the solution to (C.4)

can be written as [Sparr and Sparr, 2000]

224
 Appendix C. Solution to the One Dimensional Heat Equation
2
f § 1· S2
Q 4(T1  T2 )  a ¨© k  2 ¸¹ L2
t
§ 1·S
T ( x , t ) T2 
O
x ¦
k 1 ( 2k  1)S
e sin ¨ k  ¸ x. (C.5)
© 2¹ L

The solution can be thought of as the sum of the steady-state temperature

distribution (given by T2 í Qx/Ȝ) which is independent of time, and a
transient solution (given by the series) that tends to zero as t ĺ ’.
Figure C.2 and Figure C.3 shows the temperature distribution, given by
(C.5). Since we have assumed constant energy flow Q, the slope of the
initial temperature curve and the steady-state temperature curve are equal.
Also, the initial discontinuity in temperature is quickly smoothened out.
The heat equation is a parabolic PDE and this is a typical character of this
type of equation [Sparr and Sparr, 2000]. Figure C.4 shows how the
temperature at the paper side varies as a consequence of the step in steam
temperature. The response is similar to a step response of a system with
order higher than one. However, the mean temperature of the cylider shell,
shown in figure C.5, has more resemblance to the step response of a first-
order system. The mean temperature of the cylinder is what is used in
Chapter 4 when putting up a lumped model this system and it is then
described by a first-order linear model.
In Chapter 4, the heat equation (C.1) is compared to a simplified static
relation

Q DA(T1  T2 ), (C.6)

where the heat equation is assumed to be the correct description of heat

transfer. However, it is important to remember that the heat equation is
also an approximation. To show this it will be solved it for an infinitely
long rod.

225
Appendix C. Solution to the One Dimensional Heat Equation

Figure C.2 The temperature as a function of both position x and time t (given in seconds).
T1 = 120°C, T2 = 130°C, L = 0.03 m, c = 500 J/(kg°C), Ȝ = 50 W/(m2°C), U = 7000 kg/m3,
Q = 10 kW.

130

128

126
t = 500
Temperature ( C)

124 t = 100
o

t = 50
122 t=5

t = 20
t=1
120
t=0 t = 10
118

116

114
0 0.5 1.0 1.5 2.0 2.5 3.0
Shell thickness (cm)
Figure C.3 The temperature as a function of position x for different points of time.

226
 Appendix C. Solution to the One Dimensional Heat Equation
125

124

123

122

121
Temperature ( C)
o

120

119

118

117

116

115

114
0 20 40 60 80 100 120 140 160 180 200
Time (s)
Figure C.4 The temperature at the cylinder surface (paper side) as a function of time. The
response is similar to a step response of a higher order system.

128

127

126

125
Mean temperature ( C)

124
o

123

122

121

120

119

118

117
0 20 40 60 80 100 120 140 160 180 200
Time (s)

Figure C.5 The mean temperature of the cylinder shell as a function of time. The response
is similar to a step response of a first-order system.

227
Appendix C. Solution to the One Dimensional Heat Equation
Heat conduction in an infinitely long rod
Regard the heat equation for an infinitely long rod

­ wT w 2T
°  a 2 0, x  ƒ, t ! 0
® wt wx (C.7)
°
¯T ( x,0) g ( x)

This equation can be solved by the Fourier or Laplace transform and the
solution is given by the convolution formula

f
T ( x, t ) G g ( x, t ) ³ G( x  D ) g (D )dD ,
f
(C.8)

where

1 2
G ( x, t ) e x 4at
. (C.9)
4S at

G is called the Green function for the heat equation. If we assume that the
initial condition is given by the Dirac pulse, the solution to (C.7) is

f
1 2
T ( x, t ) G G ³ G( x  D )G (D )dD e x 4at
. (C.10)
f
4S at

Figure C.6 shows this solution as a function of both time and space. The
integral of (C.10) over the whole space is proportional to the thermal
energy in the rod, and evaluates to

f
1 2

³ e x 4at
dx 1, (C.11)
f
4S at

independently of t and a. This because no energy leaves the rod and the
Dirac pulse has unit area.
The implication of (C.10) is that the energy impulse in the origin
affects the temperature in each point x for all t > 0. In other words, all
points in space will sense the impulse instantaneously, and the model
228
 Appendix C. Solution to the One Dimensional Heat Equation
(C.1) leads to an infinite propagation velocity for the heat at the starting
point. This is due to the averaging effect of the integral in (C.8). Loosely
speaking, the heat equation does not admit any discontinuous solutions.
This non-physical property of the heat equation has been extensively
discussed in the literature, first explicitly solved by [Cattaneo, 1948]. The
theory has become known as extended thermodynamics, and is a field of
research still today [Herwig and Beckert, 2000], [Pulko, et al, 2002], and
[Belevich, 2004]. [Joseph and Preziosi, 1989] gives a good overview of
the field. Often the drawback of infinite velocity is overcome by adding
an extra term to (C.1) involving a derivative of second order in time. It
then becomes a hyperbolic function which can be regarded as a “wave
equation for heat” (with a certain speed of the heat wave).

Figure C.6 Temperature as a function of both position x and time t (given in seconds), for
the normalized case a = 1.

229
References

References

Åkesson, J., Operator interaction and optimization in control systems,

Licentiate thesis, ISRN LUTFD2/TFRT--3234--SE, Department of
Automatic Control, Lund Institute of Technology, 2003. Download
can be found at www.control.lth.se/user/johan.akesson/mpctools.

Allison, B. J., A. J. Isaksson, and A. Karlström (1997): “Grey-box

identification of a TMP refiner.” Pulp and Paper Canada, 98(4), pp.
123í126.

Allison, B. J., and A. J. Isaksson (1998): “Design and performance of

mid-ranging controllers.” Journal of Process Control, 8(5í6), pp.
469í474.

Allison, B. J., and S. Ogawa (2003): “Design and tuning of valve position
controllers with industrial applications.” Transactions of the Institute
of Measurements and Control, 25(1), pp. 3í16.

Asensio, M. C., and J. Seyed-Yagoobi (1992): “Theoretical drying study

of single-tier versus conventional two-tiered dryer configurations.”
Tappi Journal, 75(10), pp. 203í211.

230
 References
Åström, K. J. (1967): “Computer control of a paper machine í an
application of linear stochastic control theory.” IBM Journal of
Research and Development, 11(4), pp. 389í405.

Åström, K. J. (1998): “Design of PI controllers based on non-convex

optimization.” Automatica, 34(5), pp. 585-681.

Åström, K. J. (2000a): “Control problems in paper making – revisited.”

In Preprints Control Systems 2000, pp. 129í136, Canada.

Åström, K. J. (2000b): “Limitation on control system performance.”

European Journal of Control, 6(1), pp. 2í20.

Åström, K. J. (2003): Personal communication.

Åström, K. J., and R. D. Bell (2000): ”Drum boiler dynamics.”

Automatica, 36(3), pp. 363í378.

Åström, K. J., and T. Hägglund (2001): ”The future of PID control.”

Control Engineering Practice, 9, pp. 1163í1175.

Åström, K. J., and T. Hägglund (2004): ”Revisiting the Ziegler-Nichols

step response method for PID control.” Journal of Process Control,
14(6), pp. 635í650.

Åström, K. J., and T. Hägglund (2005): Advanced PID control,

Instrument Society of America, Research Triangle Park, NC.

Åström, K. J. and B. Wittenmark (1997): Computer-controlled systems í

theory and design, 3rd ed., Prentice-Hall, Upper Saddle River, NJ.

Balchen, J. G. (1999): “How have we arrived at the present state of

knowledge in process control? Is there a lesson to be learned?”
Journal of Process Control, 9(2), pp. 101í108.

Belevich, M. (2004): “Causal description of heat and mass transfer.”

Journal of Physics A: Mathematical and General, 37, pp. 3053í3069.

Bequette, B. W. (1998): Process dynamics í modeling, analysis, and

simulation, Prentice Hall, Upper Saddle River, NJ.

231
References
Bergström, J., and G. Dumont (1998): ”An object oriented framework for
developing dynamic models of a paper machine.” Dynamic Modeling
Control Applications for Industry Workshop, IEEE Industry
Applications, pp. 63í69.

Berrada, M., S. Tarasiewicz, M. Elkadiri, and P. Radziszewski (1997):

”A state model for the drying paper in the paper product industry.”
IEEE Trans. on Industrial Electronics, vol. 44, no. 4, pp. 579í586.

Bialkowski, W. L. (1996): “Effectiveness of deadtime compensators for

basis weight and moisture control.” In Proceedings Control Systems
1996, pp. 61í67, Halifax, Canada.

Bialkowski, W. L. (2000), “Advanced process control: Is it a cure for all

process control needs”, The Entech Report, vol. 12, no. 3, pp. 1í8.

Bohlin, T. (1994): “A case study of grey box identification.” Automatica,

30(2), pp. 307í318.

Bohlin, T. and S. Graebe (1995): “Issues in nonlinear stochastic Grey-

Box identification.” International Journal of Adaptive Control and
Signal Processing, 9(6), pp. 465í490.

Borisson, U. and B. Wittenmark (1974): “An industrial application of a

self-tuning regulator.” In 4th IFAC/IFIP Symposium on Digital
Computer Applications to Process Control, pp. 76í87, Zürich,
Switzerland.

Bortolin, G., S. Borg, and P. O. Gutman (2004): ”Modeling of the wet

end part of a paper mill with Dymola.” Mathematics and Computers
in Simulation, 65, pp. 31í38.

Boström, B (2002): “Measurement of dynamic moisture gradients in the

z-direction.” Preprints 7th New Available Technologies, pp. 60í64,
Stockholm, Sweden.

Brown, T., and D. Millard (1993): “New model-based machine direction

control algorithm.” In Proceedings TAPPI Process Control
Conference, pp. 15í20, Nashville, TN.

232
 References
Carlberg, M. (ed) (1989): Teknisk uppkäftighet, Skogsindustrierna and
Bokförlaget Bra Böcker.

Cattaneo, C. A. (1958): ”A form of heat conduction equation which

eliminates the paradox of instantaneous propagation.” Comptes
Rendus Academie Science, 247, pp. 431í433.

CEPI (2004): “European pulp and paper industry í Annual statistics.”

Confederation of European Paper Industries, 2004, Download can be
found at www.cepi.org.

Chang, D.-M., C.-C. Yu, and I.-L. Chien (2000): “Design sensor
trajectory for control: application to sheet-forming processes.” AIChE
Journal, 46(8), pp. 1581í1592.

Chen, S.-C. (1992): “Full-width sheet property estimation from scanning

measurements.” in Proceedings Control Systems ’92, Whistler,
Canada.

Chen, S.-C. (1995): “Modelling of paper machines for control: theory and
practice.” Pulp and Paper Canada, 96(1), pp. 17í21.

Chen, S.-C., and R. Subbarayan (1999): “Cross-machine direction

response modeling with two-dimensional sheet variations
measurements.” in Proceedings of American Control Conference, pp.
3082í3086, San Diego, CA.

Chidambaram, M., R. P. Sree (2003): “A simple method of tuning PID

controllers for integrator/dead-time processes.” Computers and
Chemical Engineering, 27(2), pp. 211í215.

Chien, K. L., J. A. Hrones, and J. B. Reswick (1952): “On the automatic

control of generalized passive systems.” Transactions of American
Society of Mechanical Engineers, 74, pp. 175í185.

Choi, S., W. Yu, D. M. France, and M. W. Wambsganss (2001): “A novel

multiport cylinder dryer.” Tappi Journal, 84(2), pp. 47.

Clapperton, R. H. (1967): The paper-making machine, its invention,

evolution and development, Pergamon Press.

233
References
Cohen, G. H., and G. A. Coon (1953): “Theoretical consideration of
retarded control.” Transactions of American Society of Mechanical
Engineers, 75, pp. 827í834.

Crotogino, R. H. (2001): “The mysteries of technology transfer. Luck

versus good planning.” Drying Technology, 19(10), pp. 2559í2575.

Dagens Industri (2004): “Skogsindustrierna tappar mark i tuff

konkurrens.” Dagens Industri (a swedish daily paper), published
2004-02-14.

Dahlin, E. B. (1968): ”Designing and tuning digital controllers.”

Instruments and Control Systems, 41(6), pp. 77í83.

Doyle, F. J., (1999): “Computational issues in the application of model-

based control to the pulp and paper industry.” AIChE Symposium
Series No. 322, vol. 95, pp. 165í172.

Dumont, G. A (1986): “Applications of advanced control methods in the

pulp and paper industry í a survey.” Automatica, 22(2), pp. 143í153.

Dumont, G. A. (1988): “Challenges and opportunities in pulp and paper

process control.” Proceedings of the 1988 American Control
Conference, pp. 1959í1964, Atlanta, GA.

Dumont, G. A., M. S. Davies, and K. Natarajan (1993): “Estimation of

moisture variations on paper machines.” IEEE Transactions on
Control Systems Technology, 1(2), pp. 101í113.

Dumont, G. A., M. S. Davies, K. Natarajan, C. Lindeborg, F. Ordubadi,

Y. Fu, K. Kristinsson, and L. Jonsson (1991): ”An improved
algorithm for estimating paper machine moisture profiles using
scanned data.” Proceedings IEEE Control and Decision Conference,
pp. 1857í1862, Brighton, England.

Duncan, S. (2003): “Editorial.” Transactions of the Institute of

Measurements and Control, 25(1), pp. 1í2.

Fellers, C., and B. Norman (1998): Pappersteknik, Institutionen för

pappersteknik, KTH.

234
 References
FFIF (2004): “Paper and wood, yearbook 2004, statistics 2003.” Finnish
Forest Industries Federation, Download can be found at
www.forestindustries.fi.

Forsman, K (2005): Reglerteknik för processindustrin, Studentlitteratur,

Lund, Sweden.

Forsman, K, and J. Birgersson (1999): “Modelling and control of the

process air in a paper machine hood.” Proceedings of the 53rd Appita
annual conference, Rotura, New Zealand.

Foss, B., B. Lohmann, and W. Marquardt (1998): “A field study of the

industrial modeling process.” Journal of Process Control, vol. 8,
no. 5, pp. 325í338.

Fritzson, P. (2004): Object-oriented modeling and simulation with

Modelica 2.1, John Wiley & sons, New York, USA.

Gaillemard, C., and A. Johansson (2004): “Modeling of the moisture

content of the paper in the drying section of a paper mill.” In Proc. of
Control Systems 2004, Quebec, Canada, pp. 183í187.

García, C. E., D. M. Prett, and M. Morari (1989): “Model predictive

control: theory and practice í a survey.” Automatica, vol. 25, no. 3,
pp. 335í348.

Gavelin, G. (ed.) (1972): Drying of paper and paperboard, Lockwood

Publishing, New York.

Gavelin, G. (1998): Paper machine design and operation, Angus Wilde

Publications, Vancouver, Canada.

Hangos, K., and I. Cameron (2001): Process modelling and model

analysis, Academic Press.

Haugwitz, S., M. Karlsson, S. Velut, and P. Hagander (2005): “Anti-

windup in mid-ranging control.” Proceedings of 44th IEEE
Conference on Decision and Control and European Control
conference, Seville, Spain.

235
References
Heaven, E. M., I. M. Jonsson, T. M. Kean, M. A. Manness, and R. N.
Vyse (1994): “Recent advances in cross machine profile control.”
IEEE Control System Magazine, 14(5), pp. 35í46.

Heikkilä, P. (1993): A study on the drying process of pigment coated

paper webs, PhD thesis, Department of Chemical Engineering, Åbo
Akademi, Åbo, Finland.

Henson, M. A., B. A. Ogunnaike, and J. S. Schwaber (1995):

“Habituating control strategies for process control.” AIChE Journal,
41(3), pp. 604í618.

Herwig, H., and K. Beckert (2000): “Fourier versus non-fourier heat

conduction in materials with a nonhomogeneous inner structure.”
Journal of Heat Transfer, 122(2), pp. 363í365.

Hill, K. C. (1993): “Analyzing the dryer section’s steam and condensate

system.” Tappi Journal, 76(6), pp. 105í114.

Ho, W. K., O. P. Gan, E. B. Tay, and E. L. Ang (1996): “Performance

and gain and phase margains of well-known PID tuning formulas.”
IEEE Transactions on Control Systems Technology, 4(4), pp.
473í477.

Ho, W. K., C. C. Hang, and L. S. Cao (1995): “Tuning of controllers

based on gain and phase margain specifications.” Automatica, 31(3),
pp. 497í502.

Hägglund, T. (1991): Process control in practice, Studentlitteratur, Lund,

Sweden.

Isaksson, A. J., and S. F. Graebe (2002): “Derivative filter is an integral

part of PID design.” IEE Proceesings of Control Theory and
Applications, 149(1), pp. 41-45.

Jansson, J. and B. Nordgren (1958): “Om torkning av papper på filttäckta

torkcylindrar.” Svensk Papperstidning, 61(19), pp. 834í843.

Joseph, D. D., and L. Preziosi (1989): “Heat waves.” Reviews of Modern

Physics, 61(1), pp. 41í73.

236
 References
Karlsson, I. (2002): ”Förfarande vid torkning av ett banformigt material.”
Swedish patent, 0004851-2 (SE518049).

Karlsson, M. (ed) (2000): Paper making part 2, drying, Tappi Press.

Karlsson, M., O. Slätteke, B. Wittenmark, and S. Stenström (2003):

”Evaluation of models for the steam supply system.” Tappi Spring
Technical Conference & Trade Fare, Chicago, IL.

Karlsson, M., and S. Stenström (2005a): “Static and dynamic modelling

of cardboard drying, part I: theoretical model.” Drying Technology,
23(1í2), pp. 143í163.

Karlsson, M., and S. Stenström (2005b): “Static and dynamic modelling

of cardboard drying, part II: simulation and experimental results.”
Drying Technology, 23(1í2), pp. 165í186.

Karlsson, M., S. Stenström, and E. Baggerud (2002): ”Dynamic

simulation of the steam supply system for a multi-cylinder dryer”
Nordic Pulp and Paper Research Journal, 17(1), pp. 66í70.

Kastanakis, G., and A. Lizr (1991): “Interaction between the MD and CD

control processes in paper making and plastics machines.” Tappi
Journal, 74(2), pp. 77í83.

Kjaer, A. P., W. P. Heath, and P. E. Wellstead (1995): “Identification of

cross-directional behaviour in web production: techniques and
experience.” Control Engineering Practice, 3(1), pp. 21í29.

Krumenacker, R. (ed) (1997): Paper machine steam and condensate

systems, 5th edition, Tappi Press.

Kuang, H.-D., J. Thibault, R. Chen, and B. Grandjean (1995): “Pilot scale

investigation of infrared drying of paper.” Tappi Journal, 78(7), pp.
129í137.

Kuusisto, R., M. Kosonen, J. Shakespeare, and T. Huhtelin (2002):

“Multivariable control of paper machine grade changes.” Pulp and
Paper Canada, 103(10), pp. 260í263.

237
References
Larsson, J. E., and T. Gustafsson (1998): “Paper machine dry line
position control during grade changes.” Proceedings of the American
Control Conference, pp. 2997í3001, Philadelphia, PA.

Lehtinen, J. (1995): “Condebelt drying of paper and paperboard for

optimizing quality and production for many grades.” Drying
technology, 13(8), pp. 2049í2068.

Li, P. Y., P. J. Bjegovic, and S. Ramaswamy (2001): “Preemptive control

of multiply actuated processes: application to moisture content
control in paper manufacturing using surrogate measurements.”
Proceedings of 2001 ASME International Mechanical Engineering
Congress and Exposition, New York, NY.

Lindell, K., and S. Stenström (2004): “Assessment of different paper

drying processes to reduce the total energy costs from a mill
perspective.” Proceedings of the 14th International Drying Sympos-
ium, Sao Paulo, Brazil.

Ljung, L. (1999): System identification - Theory for the user, 2nd edition,
Prentice Hall.

Ljung, L. (2004): System identification toolbox í user’s guide version 6,

Mathworks, Natick, MA.

Love, W. C. (1994): “Innovative control technique improves control

rangeability and resolution in paper mill applications.” Tappi Journal,
77(2), pp. 195í199.

Makkonen, A., R. Rantanen, A. Kaukovirta, H. Koivisto, J. Lieslehto, T.

Jussila, H. N. Koivo, and T. Huhtelin (1995): “Three control schemes
for paper machine MD-control.” Control Engineering Practice,
3(10), pp. 1471í1474.

Martinez, D. M., M. Drotz, R. Lai, and A. R. Matin (2001): “Dryness and

physical properties of paper webs in laboratory-scale impulse
processing.” Drying Technology, 19(10), pp. 2435í2450.

Mattsson, S. E., and H. Elmqvist (1997): “Modelica í An international

effort to design the next generation modeling language.” 7th IFAC
symp. on computer aided control systems design, Gent, Belgium.

238
 References
Mattsson, S. E., H. Elmqvist, and M. Otter (1998): “Physical system
modeling with Modelica.” Control Eng. Practice, vol. 6, pp.
501í510.

McConnell, R. (1980): A literature review of drying research in the pulp

and paper industry, Drying ‘80, Ed. A. Mujumdar, Hemisphere
Publishing, NY, pp. 330í337.

McLain, R. B, M. J. Kurtz, and M. A Henson (1996): “Habituating

control for non-square nonlinear processes.” Proceedings of IEEE
International Conference on Control Applications, pp. 273í278,
Dearborn, MI.

Menani, S., H. Koivo, T. Huhtelin, and R. Kuusisto (1998): “Dynamic

modelling of paper machine from grade change data.” In Proceedings
of Control Systems, Porvoo, Finland pp. 79í86.

Morari, M., and E. Zafiriou (1989): Robust process control, Prentice-

Hall, Englewood Cliffs, N. J.

Mori, Y., H. Shimizu, and K. Takao (2000): “Development of a new

automatic grade change control system for paper machine.” Proc.
Control Systems 2000, pp. 231í234, Victoria, Canada.

Murphy, T. F., and S.-C. Chen (1999): “Transition control of paper-

making processes: paper grade change.” Proceedings of the 1996
IEEE International Conference on Control Applications,
pp. 1278í1283, Kohala Coast-Islands of Hawaii, Hawaii.

Murphy, T. F., S. Yurkovich, and S.-C. Chen (1996): “Intelligent control

for paper machine moisture control.” Proceedings of the 1996 IEEE
International Conference on Control Applications, pp. 826í833,
Dearborn, Michigan.

Morud, J., and S. Skogestad (1996): ”Dynamic behaviour of integrated

plants.” Journal of Process Control, 6, pp. 145í156.

Natarajan, K., G. A. Dumont, and M. S. Davies (1988): “An algorithm

for estimating cross and machine direction moisture profiles for paper
machines.” IFAC/IFORS Symposium, pp. 1091í1096, Beijing, PRC.

239
References
Nelson, D., and T. Gardner (1996): “Optimizing paper machine control -
a case study.” Pulp and Paper Canada, 97(11), pp. 43í49.

Niemenmaa, A., J. Lappalainen, K. Juslin, and I. Laukkanen (1996):

“Dynamic modelling and simulation of paper machine drying
section.” Proceedings of the 10th International Drying Symposium
(IDS’96), vol. B, pp.1157í1164, Kraków, Poland.

Nissan, A., and W. Kaye (1955): “An analytical approach to the problem
of drying thin fibrous sheets on multicylinder machines.” Tappi
Journal, vol. 38, no. 7, pp. 385í398.

Orloff, D. I., and J. W. Crouse (1999): “Impulse drying: status of the

pilot-scale research program.” Tappi Journal, 82(9), pp. 143í149.

Pauksta, P. M. (1998): “Paper machines of tomorrow, drying to speed.”

Tappi Journal, 81(7), pp. 46í52.

Peng, S. W., K. Mizukami, W. Liu, N. Sebe, T. Takeba (1997): ”An

experimental study of condensation heat transfer in a horizontally
rotating cylinder with a scraper.” Experimental Thermal and Fluid
Science, 14, pp 205í212.

Perrault, R. D. (1991): “Troubleshooting design and operation of the

dryer section.” in TAPPI Practical Aspects of Pressing & Drying
Short Course, pp. 249í277, Atlanta, GA.

Perré, P., Y. Lescanne, and Y. Brocard (2004): “Modelling of paper

drying: a comprehensive computational model used by the industry.”
Proceedings of the 14th International Drying Symposium (IDS’2004),
vol. B, pp. 1263í1270, São Paulo, Brazil.

Persson, H. (1998): Dynamic modelling and simulation of multi-cylinder

paper dryers, Licentiate thesis, Department of Chemical Engineering,
Lund Institute of Technology.

Petterson, M., and S. Stenström (2000): “Experimental evaluation of

electric infrared dryers.” Tappi Journal, vol. 83, no. 8.

240
 References
Poulin É., and A. Pomerleau (1999): “PI settings for integrating processes
based on ultimate cycle information.” IEEE Transactions on Control
Systems Technology, 7(4), pp. 509í511.

Pulko, S. H., A. J. Wilkinson, and A. Saidane (2002): “TLM

representation of the hyperbolic heat conduction equation.”
International Journal of Numerical Modelling, 15, pp. 303í315.

Pulkowski, J. H., and G. L. Wedel (1988): “The effect of spoiler bars on

dryer heat transfer.” Pulp and Paper Magazine of Canada, 89(8), pp.
61í66.

Ramesh, R. (1991): Modeling of multicyliner drying of light-weight

paper, PhD thesis, State University of New York, Col. of
Environmental Science and Forestry.

Rao, M., Q. Xia, and Y. Ying (1994): Modeling and advanced control for
process industries: application to paper making processes, Springer-
Verlag, New York.

Reardon, S., M. Davis, and P. Doe (2000): “Computational modelling of

paper drying machines.” Tappi Journal, vol. 83, no. 9.

Retulainen, E. (2001): “Key development phases of Condebelt: long

journey from idea to commercial product.” Drying Technology,
19(10), pp. 2451í2467.

Sadeghi, M. (2003): Modeling and simulation of transport phenomena in

paper drying, PhD thesis, McGill University, Montreal, Canada.

Sadeghi, M., and W. J. M. Douglas (2004): “From tissue to linerboard:

validation of a microscale simulator for single technique and hybrid
dryers.” Proceedings of the 14th International Drying Symposium
(IDS’2004), vol. A, pp. 444í451, São Paulo, Brazil.

Schmidt, E. (1969): Properties of water and steam in SI-units, Springer

Verlag, Berlin, Germany.

Schweiger, C. A., and J. B. Rudd (1994): “Prediction and control of paper

machine parameters using adaptive technologies in process
modelling.” Tappi Journal, 77(11), pp. 201í208.

241
References
Sell, N. J. (ed) (1995): Process control fundamentals - for the pulp and
paper industry, Tappi Press.

Seyed-Yagoobi, J., S. J. Sikirica, and K. M. Counts (2001):

“Heating/drying of paper sheet with gas-fired infrared emitters í pilot
machine trials.” Drying Technology, 19(3&4), pp. 639í651.

Shinskey, F. G. (1978): “Control systems can save energy.” Chemical

Engineering Progress, 43, pp. 43í46.

Silvennoinen, E., K. Juslin, M. Hänninen, O. Tiihonen, J. Kurki, and K.

Porkholm (1989): The APROS software for process simulation and
model development, Technical Research Centre of Finland, Research
report 618, Espoo, Finland.

Skogestad, S. (2003): “Simple analytic rules for model reduction and PID
controller tuning.” Journal of Process control, 13, pp.291í309.

Skoglund, A., A. Brundin, and C.-F. Mandenius (2000): “A multivariate

process model for grade change in a paperboard machine.” Nordic
Pulp and Paper Research Journal, vol. 15, no. 3, pp. 183í188.

Slätteke, O., and K. J. Åström (2005): “Modeling of a steam heated

rotating cylinder – a grey-box approach.” American Control
Conference 2005, Portland, Oregon.

Slätteke, O., K. Forsman, T. Hägglund, and B. Wittenmark (2002). “On

identification and control tuning of cylinder dryers” In Proceedings
Control Systems 2002, pp. 298í302, Stockholm, Sweden

Sparr, G. and A. Sparr (2000): Kontinuerliga System, Studentlitteratur,

Lund, Sweden.

Stenström, S., M. Karlsson, O. Slätteke, K. Forsman, and B. Wittenmark

(2002): ”Productivity increase from a better understanding of
dynamic processes and control of the paper dryer.” Preprints 7th New
Available Technologies, pp. 70í73, Stockholm, Sweden.

Stenström, S., and T. Svanquist (1991): “A general model for calculating

pressure drop in siphons and siphon riser tubes í Part 2: experimental
results”, Tappi Journal, 74(12), pp. 157í162.

242
 References
Stenström, S., B. Wilhelmsson, L. Nilsson, R. Krook, and R.
Wimmerstedt (1994): “Measurement of reference experimental
drying data for the multi-cylinder paper dryer.” Proceedings of the 9th
International Drying Symposium (IDS’94), Gold Coast, Australia, pp.
1179í1186.

Stewart, G. E., D. M. Gorinevsky, and G. A. Dumont (2003): “Feedback

controller design for a spatially distributed system: the paper machine
problem.” IEEE Transaction on Control Systems Technology, 11(5),
pp. 612í628.

Sun, X., K. Yi, and Y. Sun (2000): “The modeling and control of basis
weight and moisture content.” Proceedings 3rd World Congress on
Intelligent Control and Automation, Heifei, P. R. China, pp.
3724í3728.

Thomas, P. (1999): Simulation of industrial processes í for control

engineers, Butterworth-Heinemann, Oxford, Great Britain.

Tyreus, B. D., and W. L.. Luyben (1992): “Tuning PI controllers for

integrator/dead time processes.” Industrial and Engineering
Chemistry Research, 31, pp. 2625í2628.

VanAntwerp, J. G, and R. D. Braatz (2000): “Fast model predictive

control of sheet and film processes.” IEEE Transactions on Control
Systems Technology, 8(3), pp. 408í417.

Videau, J. L., and A. Lemaitre (1982) An improved model of a paper

machine multicylinder drying section, Drying ‘82, Ed. A. Mujumdar,
Hemisphere Publishing, NY, pp. 129í138.

Viitamäki, P. (2004): Hybrid modeling of paper machine grade changes,

PhD-thesis, Department of Automation and Systems Technology,
Helsinki University of Thechnology, Finland.

Visioli, A. (2001): “Optimal tuning of PID controllers for integral and

unstable processes.” IEE Proceedings of Control Theory Appli-
cations, 148(2), pp. 180í184.

243
References
Wallén, A. (2000): Tools for autonomous process control, PhD-thesis,
Department of Automatic Control, Lund Institute of Technology,
Sweden.

Wang, H. (1996): “Applying neuro-fuzzy modelling and control to MD

moisture content systems in paper machines.” Proceedings of IEEE
International Conference on Control Applications, pp. 229í234,
Dearborn, MI.

Wang, Y.-G., W.-J. Cai (2001): “PID tuning for integrating processes
with sensitivity specification.” Proceedings Conference on Decision
and Control, pp. 4087í4091, Orlando, FL.

Wells, C. H. (1999): “Robust multivariable IMC control for paper

machines.” Tappi Journal, 82(7), pp. 140í144.

Wilhelmsson, B. (1995): An experimental and theoretical study of multi-

cylinder paper drying, PhD thesis, Department of Chemical Eng-
ineering, Lund Institute of Technology.

Wilhelmsson, B., L. Fagerholm, L. Nilsson, and S. Stenström (1994):

“An experimental study of contact coefficients in paper drying.”
Tappi Journal, 77(5), pp. 159í168.

Xia, Q, M. Rao, and J. Qian (1993): “Model algorithmic control for paper
machines.” Proceedings of the IEEE Conference on Control
Applications, pp. 203í208, Vancouver, Bristish Columbia.

Zhuang, M., and D. P. Atherton (1993): “Automatic tuning of optimum

PID controllers.” IEE Proceedings D. Control Theory and
Applications, 140(3), pp. 216í224.

Ziegler, J. G., and N. B. Nichols (1942): “Optimum settings for automatic

controllers.” Transactions of American Society of Mechanical
Engineers, 64, pp. 759í768.

244
 List of Symbols

List of Symbols

Acyl m2 Inner cylinder area

Axy m2 Area of paper covering the cylinder
Cp,m J/(kg·K) Specific heat capacity of cylinder shell
Cp,p J/(kg·K) Specific heat capacity of paper
Cv m2 Valve conductance
C(s) í Controller transfer function
d í Load disturbance
dv kg/(s·%) Valve constant
dy m Width of paper sheet
2
g kg/m Dry basis weight of paper
G(s) í General transfer function
hs J/kg Enthalpy of steam
hw J/kg Enthalpy of condensate
ǻHs J/kg Heat of sorption
ǻHvap J/kg Latent heat of vaporization
ǻH J/kg Energy required to evaporate water from the
paper surface
J í Cost function in MPC
kc í Gain of the PID controller
K m/s Mass transfer coefficient for paper sheet
m kg Mass of cylinder shell
Ms í Maximum value of sensitivity function
Mw kg/mole Molecular weight of water

245
List of Symbols
n í Noise
p Pa Steam pressure inside cylinder
psh Pa Steam pressure in header
ptot Pa Standard pressure (101.325 kPa)
pv0 Pa Partial vapor pressure for free water
pv,a Pa Partial pressure for water vapor in the air
pv,p Pa Partial pressure for water vapor at paper surface
P(s) í Process transfer function
qbt kg/s Blow through steam
qc kg/s Condensation rate
qevap kg/(m2·s) Evaporation rate
qs kg/s Inflow of steam to cylinder
qw kg/s Outflow of condensate
Q í Weighting matrix for MPC
Qm W Energy flow to cylinder shell
Qp W Energy flow to paper
r í Set point
R í Weighting matrix for MPC
Rg J/(mole·K) Gas constant ( § 8.31)
Rv í Valve rangeability
Td s Derivative time of the PID controller
Ti s Integral time of the PID controller
Tm K Temperature of cylinder shell
Tp K Temperature of paper
u kg/kg Moisture ratio
uc í Controller output
us J/kg Internal energy of steam
uw J/kg Internal energy of condensate

246
 List of Symbols
V m3 Volume inside cylinder
Vs m3 Volume of steam
3
Vw m Volume of condensate
vx m/s Speed of paper sheet
w % Moisture content
x kg/kg Water content in air
xv í Valve opening
y í Process output

Įsc W/(m2·K) Heat transfer coefficient steam í cylinder

Įcp W/(m2·K) Heat transfer coefficient cylinder í paper
įcyl m Cylinder thickness / 2
ij í Sorption isotherm
Ȝcyl W/(m·K) Thermal conductivity of cylinder shell
Ș í Fraction of dryer surface covered by paper
ȡs kg/m3 Density of steam
ȡw kg/m3 Density of condensate
IJ1 and IJ2 í Tuning parameters of mid-ranging controller

247
Department of Automatic Control

ISSN 0280-5316
ISRN LUTFD2/TFRT--1075--SE