Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
Modeling and Control of the
Paper Machine Drying Section
Ola Slätteke
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
SE221 00 LUND
Sweden
ISSN 0280í5316
ISRN LUTFD2/TFRT1075SE
© 2006 by Ola Slätteke. All rights reserved.
Printed in Sweden by MediaTryck
Lund 2006
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 blackbox 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 objectoriented 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.
Abstract
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 PhDstudies 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.
Acknowledgements
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 singleloop 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 DCSsystem 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.
Preface
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. Blackbox Models and Controller Structures .................................. 45
3.1 A blackbox 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 twopole 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
Contents
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 Midranging........................................................................... 150
6.5 Moisture control by midranging 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. ObjectOriented Modeling and Predictive Control of the Moisture
Content .............................................................................................. 183
8.1 The model.............................................................................. 184
8.2 Steadystate model validation ............................................... 194
8.3 Open loop simulations........................................................... 196
8.4 Control of moisture by midrange 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
13
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:
 Copy paper for printers, copying machines and writing
 Newsprint
 Cardboard
 Lightweight coated paper for magazines
 Wrapping and packaging paper
 Hygienic tissue paper
 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
1
Introduction
Chapter 1. Introduction
14
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
Chapter 1. Introduction
15
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.
99 % 80 % 50 % 5 %
Wire section Press section Drying section
140 m
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.
Drying section
~
Chapter 1. Introduction
16
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.
 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.
 With reduced variance the moisture set point can be increased
without changing the probability for an offspec 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.
 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.
 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
Chapter 1. Introduction
17
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.
 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.
 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.
5 6 7 8 9 10 11 12 13 14 15
0.0
0.2
0.4
0.6
0.8
Moisture (%)
P
r
o
b
a
b
i
l
i
t
y
d
e
n
s
i
t
y
Tolerance
limit
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 giveaway.
Chapter 1. Introduction
18
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 NicholasLouis 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.
Chapter 1. Introduction
19
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 statespace 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 minimumvariance controller [Åström,
1967] and the selftuning 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
 Analysis of different controller structures for the steam
pressure loop based on a previously proposed blackbox model.
 Presentation of a physical steam cylinder model with the same
structure as the blackbox model. The purpose is to gain deeper
understanding in the physics behind the process.
 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 tradeoff between performance and robustness. It is
compared to a few other design methods and tested on a real
paper machine.
 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 midranging of two IMC
controllers.
 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).
Chapter 1. Introduction
20
 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.
 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:
 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 7
th
New Available Technologies, pp. 70–73,
Stockholm, Sweden.
 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.
 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.
 Slätteke, O. (2003): Steam and condensate system control in paper
making, licentiate thesis, ISRN LUTFD2/TFRT3231SE,
Department of Automatic Control.
 Karlsson, M., O. Slätteke, B. Wittenmark, and S. Stenström
(2005): “Reducing moisture transients in the paper machine drying
section with the midranging control technique,” Nordic Pulp and
Paper Research Journal, 20(2), pp. 150í156.
 Slätteke, O., and K. J. Åström (2005): “Modeling of a steam
heated rotating cylinder í A greybox approach,” Proceedings
American Control Conference 2005, Portland, OR.
 Karlsson, M., O. Slätteke, T. Hägglund, and S. Stenström:
“Feedforward control in the paper machine drying section,”
submitted to American Control Conference 2006.
Chapter 1. Introduction
21
 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 blackbox 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 objectoriented 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.
Chapter 2. Fundamentals of the Paper Drying Process
22
2.
The most common way to evaporate water from the paper web is to use
the latent heat of vaporization in steam. A steamfilled 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
selfsupporting in terms of electricity. For chemical pulp mills, steam is
obtained simply as a byproduct 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 multicylinder drying. In this thesis, attention is only
given to the multicylinder 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.
2
Fundamentals of the Paper
Drying Process
Chapter 2. Fundamentals of the Paper Drying Process
23
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 socalled blowthrough 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 twotier configuration.
Chapter 2. Fundamentals of the Paper Drying Process
24
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 singletier
design (singlefelted) and the twotier (doublefelted). The twotier
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
Figure 2.2 A singletier configuration.
Chapter 2. Fundamentals of the Paper Drying Process
25
runnability problems at higher machine speeds, the singletier
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 SeyedYagoobi, 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
Steam header
of the group
Condensate
pipe
C
Y
L
I
N
D
E
R
1
C
Y
L
I
N
D
E
R
2
C
Y
L
I
N
D
E
R
3
C
Y
L
I
N
D
E
R
4
C
Y
L
I
N
D
E
R
5
C
Y
L
I
N
D
E
R
6
PC
Figure 2.3 Sketch of a dryer group consisting of 6 cylinders and one common pressure
gauge and controller.
Chapter 2. Fundamentals of the Paper Drying Process
26
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 blowthrough 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 makeup 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 PDCvalve. This minimum pressure difference depends on
both the operating point and machine specific properties. The differential
Dryer group B Dryer group A
LC1
PDC
PC1
PC2
To the condenser
LC2
Steam header
Tank A
Tank B
Figure 2.4 Part of a drying section with a cascade system. Sometimes the PDCvalve is
installed in the pipe between group A and tank A.
Chapter 2. Fundamentals of the Paper Drying Process
27
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 highpressure 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
Dryer group
LC
PC
To the condenser
Steam header
Tank
Tank B
PDC
1
1
2
2
Thermo
compressor
Figure 2.5 A steam group with a thermo compressor unit. Both the PC and PDC controller
work in splitrange and the numbers indicate in which order the actuators are manipulated.
Some care must be taken since noncondensibles might be recirculated and accumulated.
Chapter 2. Fundamentals of the Paper Drying Process
28
Figure 2.6 A typical piping and instrumentation diagram of the steam and condensate
system.
T
4
T
5
S
G
3
3
c
y
l
.
S
G
2
7
c
y
l
.
S
G
1
5
c
y
l
.
F
I
P
C
L
C
L
C
S
G
6
1
2
c
y
l
.
S
G
5
1
4
c
y
l
.
S
G
4
1
7
c
y
l
.
T
3
L
C
T
2
L
C
T
1
T
0
L
C
P
D
C
P
D
C
F
r
o
m
b
o
i
l
e
r
T o c o n d e n s e r
P
D
C
1
2
P
D
C
T
o
s
t
e
a
m
b
o
x
P
D
C
P
C
P
C
P
C
P
C
P
D
C
T
C
3
.
4
m
3
2
m
3
3
.
1
m
3
6
m
3
3
m
3
6
m
3
2
1
F
I
F
I
P
C
1
2
2
1
T
I
P
I
C
o
n
d
e
n
s
e
r
T
C
T
o
s
t
e
a
m
b
o
x
L
C
L
C
P
C
1
2
P
C
P
I
2
1
Chapter 2. Fundamentals of the Paper Drying Process
29
controller primarily uses the recirculated steam from the thermo
compressor but have the possibility to also use makeup steam from the
header, if necessary. In the same manner, PDC leads the blowthrough
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 casebycase 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 online 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.
Chapter 2. Fundamentals of the Paper Drying Process
30
[Kastanakis and Lizr, 1991]. In [Natarajan, et al, 1988] an algorithm is
developed, which uses least squares to estimate the CD component and
Kalmanfiltering for the MD component. It is then further developed in
[Dumont, et al, 1991], and [Chen, 1992]. A similar decomposing
algorithm based on KarhunenLoeve 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 [SeyedYagoobi, 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 labscale 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
overdried and then remoisturized. Of course, then the gain in higher
quality has to be weighted against the cost of higher energy use. This
~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.
Chapter 2. Fundamentals of the Paper Drying Process
31
thesis focuses solely on the MDcontrol. 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 “2sigma” or two standard
deviations of the controlled variables. All produced reels of paper leaves
the paper machine together with a “reelreport” that include statistics like
“2sigma MD”, “2sigma CD”, and “2sigma total” for both the moisture
Dryer
Group 1
Dryer
Group 2
Machine direction
Dryer
Group 3
Dryer
Group 4
Dryer
Group 5
Dryer
Group 6
S
c
a
n
n
e
r
Figure 2.10 Structure for the moisture control loop with one scanner device and six steam
groups.
0 500 1000 1500 2000 2500 3000 3500
4.25
4.30
4.35
4.40
M
o
i
s
t
u
r
e
(
%
)
0 500 1000 1500 2000 2500 3000 3500
79.4
79.6
79.8
80.0
80.2
t (s)
B
a
s
i
s
w
e
i
g
h
t
(
g
/
m
2
)
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/m
2
. The 2sigma values were 0.056% and 0.3 g/m
2
respectively.
Chapter 2. Fundamentals of the Paper Drying Process
32
and basis weight [Sell, 1995]. These are the average values for the whole
reel, but it is the short term 2sigma 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/m
2
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 multiinputsingleoutput (MISO)
system. This means that the drying process has many degrees of freedom
Group 1 Group 5 Group 4 Group 3 Group 2
S
t
e
a
m
p
r
e
s
s
u
r
e
ǻp
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.
Chapter 2. Fundamentals of the Paper Drying Process
33
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 , 1
or
0 , 1 0
where
¦
¹
¦
´
¦
s =
= s <
+ =
n n
n n
n n n
m k
m k
m r k f (2.1)
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
) (
1
r f
PC1
) (
2
r f
PC2
) (
3
r f
PC3
) (
4
r f
PC4 PC5
) (
6
r f
PC6
r
Figure 2.12 The set point r from the moisture controller is distributed to the steam
pressure controllers by passing it through a ratio/biasfunction except for the lead group,
in this case group 5.
Chapter 2. Fundamentals of the Paper Drying Process
34
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
deadtime 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 PIcontrollers in the inner loop. In Chapter 6, a mid
ranging control structure by combining two IMCcontrollers is investi
Moisture
Dry weight
Production speed
Layer distribution
Formation
Bulk
Web temperature
Retention
Filler
Refining
Freeness
Steam pressure
Process air
Leakage air
Blow through steam
Condensate flow
Exhaust air
Condition of fabrics
Web tension
Condensate distribution
Tuning of controllers
Dryer section
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.
IMC
Steam
system
Dryer
Setpoint
Moisture
Setpoint
steam pressure
Steam
pressure
PI
controller
E
–1
Moisture
Figure 2.13 A block diagram of the moisture control loop.
Chapter 2. Fundamentals of the Paper Drying Process
35
gated. In this way, two singleloop controllers form a quasimultivariable
controller. The same structure is also implemented by a true multi
variable controller but with different manipulated variables, see Chapter 8.
Other nonconventional 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.
 Production Speed: Affects the amount of steam needed, since high
production also involves higher vaporization.
 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.
 Inlet Moisture: The moisture content of the sheet after the press
section is a disturbance variable that normally is unknown.
 Degree of Refining: This parameter naturally affects both the
freeness (measure of the drainability) and the ability to dry the
sheet.
 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.
 Air Dew Point: A high dew point inhibits effective evaporation.
 Dryer Fabric Condition: An old fabric can be clogged and give a
higher evaporation resistance.
 Bulk: High bulk means that the water inside the web has a longer
transport distance to the surface and ambient air.
 Retention aids: It is easier to dry the web when the retention is
high since it then contains more filler.
 Web tension: High web tension increases the heat transfer
coefficient and the drying rate.
 Leakage air: The air from the machine room is cooler that the
preheated supply air and therefore impair the drying conditions.
Chapter 2. Fundamentals of the Paper Drying Process
36
 Ply loading: In paperboard, different layers consist of different
pulps, hence different physical properties. This influences the
drying.
 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.
0 200 400 600 800 1000 0 200 400 600 800
10
11
12
13
M
o
i
s
t
u
r
e
(
%
)
0 200 400 600 800 1000 0 200 400 600 800
137
138
139
140
141
142
time (s)
C
o
n
d
.
w
e
i
g
h
t
(
g
/
m
2
)
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.
Chapter 2. Fundamentals of the Paper Drying Process
37
0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
0.4
0.6
0.8
1.0
1.2
2o dry weight (g/m
2
)
2

o
m
o
i
s
t
u
r
e
(
%
)
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.
Stock
flow
Steam
pressure
Total weight
Moisture
a)
Drying
process
Paper sheet
process
Stock
flow
Steam
pressure
Dry weight
Moisture
b)
Drying
process
Paper sheet
process
Figure 2.17 There are crossconnections 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).
Chapter 2. Fundamentals of the Paper Drying Process
38
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 crosscoupling 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.
Chapter 2. Fundamentals of the Paper Drying Process
39
0 100 200 300 400 500 600 700 800 900 1000
4.2
4.3
4.4
Time (s)
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
0 100 200 300 400 500 600 700 800 900 1000
195
200
205
210
215
S
t
e
a
m
p
r
e
s
s
u
r
e
(
k
P
a
)
Figure 2.19 Time series for the steam pressure and moisture, used for spectrum estimations.
10
4
10
3
10
2
10
1
10
0
10
1
10
1
10
3
P
o
w
e
r
s
p
e
c
t
r
u
m
p
r
e
s
s
u
r
e
10
4
10
3
10
2
10
1
10
0
10
4
10
2
10
0
Frequency (Hz)
P
o
w
e
r
s
p
e
c
t
r
u
m
m
o
i
s
t
u
r
e
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.
Chapter 2. Fundamentals of the Paper Drying Process
40
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
ds wc
wc
m m
m
w
+
=
100
[%] (2.2)
where m
wc
is the mass of the water content in the sheet and m
ds
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
ds
wc
m
m
u = [kg/kg] (2.3)
Moisture ratio is often used in chemical engineering and literature on
physical modeling of paper drying. The relation between the two
quantities is
w
w
u
u
u
w
÷
=
+
=
100
,
1
100
(2.4)
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 blackbox 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.
Chapter 2. Fundamentals of the Paper Drying Process
41
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.
0 0.5 1.0 1.5 2.0 2.5 3
0
10
20
30
40
50
60
70
80
Moisture ratio (kg/kg)
M
o
i
s
t
u
r
e
c
o
n
t
e
n
t
(
%
)
Figure 2.20 Relation between moisture content and moisture ratio.
Chapter 2. Fundamentals of the Paper Drying Process
42
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.
Modeling and control
of the steam and
condensate system
Part
1
Chapter 3. Blackbox Models and Controller Structures
45
3.
The pulp and paper industry is a highly competitive and capitalintensive
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 “whatif”
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 firstprinciples versus blackbox (also known as statistical or
empirical), or steadystate versus dynamical, or linear versus nonlinear,
or continuous versus discrete, or lumped versus distributed.
In this thesis, preferably continuoustime dynamical models are used.
They will be either blackbox or firstprinciples, linear or nonlinear, and
lumped or distributed. It all depends on the context.
This chapter introduces a linear blackbox model structure for the
steam pressure in a drying cylinder that is based on step response data. It
3
Blackbox Models and
Controller Structures
Chapter 3. Blackbox Models and Controller Structures
46
will be used to analyze the closed loop system for a few different
controller structures. In the next chapter, a corresponding firstprinciples
model is presented.
3.1 A blackbox 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 IPZmodel. This model
structure has also been suggested in [Sell, 1995], and [Nelson and
Gardner, 1996]. The IPZmodel is represented by the transfer function
( )
( )
.
1
1
) (
2 1
2
1
T T e
sT s
sT
k s G
sL
v IPZ
>
+
+
=
÷
(3.1)
0 20 40 60 80 100 120 140 160 180 200
170
180
190
200
210
P
r
e
s
s
u
r
e
(
k
P
a
)
0 20 40 60 80 100 120 140 160 180 200
65
70
75
80
85
Time (s)
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Figure 3.1 Open loop step response, taken from a liner machine. The response can be
approximated by (3.1) with the process parameters k
v
=0.0027, T
1
=58, T
2
=3.1, and L=2.0.
Chapter 3. Blackbox Models and Controller Structures
47
Note that T
1
is always larger than T
2
, typically by a factor of 5 to 50. The
transfer function can then be regarded as an integrator in series with a
leadnetwork. 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/m
2
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 IPZprocess.
[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
0 100 200 300 400 500 600
500
505
510
515
520
525
P
r
e
s
s
u
r
e
(
k
P
a
)
0 100 200 300 400 500 600
26
27
28
29
30
31
Time (s)
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Figure 3.2 Open loop step response, taken from a board machine. The response can be
approximated by (3.1) with the process parameters k
v
=0.0020, T
1
=73, T
2
=21, and L=1.0.
Chapter 3. Blackbox Models and Controller Structures
48
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 buildup 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
0 50 150 250 350 450 550
660
670
680
690
700
710
P
r
e
s
s
u
r
e
(
k
P
a
)
0 50 150 250 350 450 550
64
66
68
70
72
74
Time (s)
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Figure 3.3 Open loop step response taken from a Yankee dryer. The model corresoinds
to (3.1) with the process parameters k
v
=0.0026, T
1
=269, T
2
=87, and L=2.0
Chapter 3. Blackbox Models and Controller Structures
49
[Karlsson, 2000]. This should be compared with multicylinder 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
IPZprocess 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, k
v
, T
1
, T
2
, and L from a simple open loop step
response.
Figure 3.4 shows a step response of an IPZmodel (3.1), where L is
chosen to zero for simplicity. Start by drawing two straight lines, the
20 0 20 40 60 80 100 120
0
0.5
1
1.5
2
2.5
3
3.5
Time (s)
O
u
t
p
u
t
(
k
P
a
)
(t
0
,y
0
)
q
2
(t)
q
1
(t)
Figure 3.4 Simulated unit step response of an IPZprocess with k
v
=0.01, T
1
=220, T
2
=20,
and L=0.
Chapter 3. Blackbox Models and Controller Structures
50
tangent to the step response at time t = 0 (called q
1
(t)) and the asymptote
as time tends to infinity (called q
2
(t)), both marked in the figure. Suppose
that the size of the step in the control signal is u
0
and that the slope of
q
1
(t) is k
1
and the slope of q
2
(t) is k
2
. Also, suppose that the two lines
intersect at the coordinates (t
0
,y
0
). Then we have
. , ,
0 2
2
0
1
0
2
t T
k
y
T
u
k
k
v
= = = (3.2)
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
( )
( )
0 0
2
1
0
0 2
0
2
1
1
lim ) ( lim
) (
lim u k u
sT
sT k
s
u
s G s
dt
t dy
k
v
v
s s t
=
+
+
= = =
÷ ÷ · ÷
(3.3)
By the initial value theorem, the initial derivative is
( )
( )
.
1
1
lim ) ( lim
) (
lim
) (
0
2
1
0
2
1 0
2
0
0
1
u
T
T k
u
sT
sT k
s
u
s G s
dt
t dy
dt
t dy
k
v v
s s t
t
=
+
+
= = = =
· ÷ · ÷ ÷
=
(3.4)
Thus the tangent of y(t) at t = 0 is
, ) (
2
0 1
1
t
T
u T k
t q
v
= (3.5)
since the response starts in the origin. By the inverse Laplace transform,
the step response can also be written as
( )( ) ( ), 1 ) (
2
/
2 1 0
t e T T u k t y
T t
v
+ ÷ ÷ =
÷
(3.6)
and to get the equation for the asymptote of the step response we observe
that
Chapter 3. Blackbox Models and Controller Structures
51
( ) ( ), ) ( lim
2 1 0 0
T T u k t u k t y
v v
t
÷ = ÷
· ÷
(3.7)
which is the value of q
2
(0). Consequently, we have
( ), ) (
2 1 0 2
T T t u k t q
v
÷ + = (3.8)
and the two lines q
1
(t) and q
2
(t) intersect at t=T
2
, since
. ) ( ) (
0 1 2 2 2 1
u T k T q T q
v
= = (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 k
v
T
1
and k
v
T
1
/T
2
are more or less constant. This is
probably due to difficulties with finding the slope of the asymptote q
2
, see
Figure 3.4. If the effect from the lowpass filter part with time constant T
2
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
. ) ( , ) (
2 2 2 1 1
y t k t q t k t q + = = (3.10)
The intersection of the two lines is the given by
, ,
2 0 2 0 0 1 0
y t k y t k y + = = (3.11)
which can be rewritten as
.
1
,
2
2 1
0 2
2 1
1
0
y
k k
t y
k k
k
y
÷
=
÷
= (3.12)
We also have
, ,
) (
,
2 1
2
2
2 2 1
2 1
1
0
2
k k
y
T
k k k
y k
T
u
k
k
v
÷
=
÷
= = (3.13)
Chapter 3. Blackbox Models and Controller Structures
52
and
, ,
) (
0
1
2
1
2
0 2 1
2 1
1 1
u
k
T
T k
E
u k k
y k
T k E
v
v
= =
÷
= = (3.14)
where E
1
and E
2
are the two expressions of interest. Assume an error in
k
2
, and differentiate T
1
with respect to k
2
.
,
) (
) 2 (
2
2
2
2 1
2 1 2 1
2
1
k k k
k k y k
dk
dT
÷
÷
÷ = (3.15)
and rewrite as a relative error
2
2
2 1
2 1
1
1
2
k
dk
k k
k k
T
dT
÷
÷
÷ = (3.16)
Similar calculations for the other parts of (3.13) gives
. 0 , ,
2
2
2
2
2 1
2
1
1
2
2
=
÷
= =
dk
dE
k
dk
k k
k
E
dE
k
dk
k
dk
v
v
(3.17)
The first thing to notice is that E
2
is independent of the slope k
2
, so it is
therefore obvious that k
v
T
1
/T
2
tends to be independent of different
experiments, if there is an uncertainty in k
2
. Also, since, k
1
is larger than
k
2
, the relative error of E
1
is smaller than the relative errors of k
v
and T
1
.
Consequently the observation is confirmed. It is also clear from (3.5) that
k
v
T
1
/T
2
is independent of line q
2
.
The product k
v
T
1
can be interpreted as follows. Rewrite (3.1) as
sL
v
v
T
IPZ
e T k
s
k
s G
÷
=

.

\

+ =
1
0
2
) ( (3.18)
The step response of this function is shown in Figure 3.5. A constant
product k
v
T
1
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 k
v
, this will give a small variation in T
1
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
Chapter 3. Blackbox Models and Controller Structures
53
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 autotuning, gain
scheduling, and loop performance analysis.
The PID controller, in ideal form, can be written as
( ) ( ) ( ) ) ( ) ( ) ( ) ( ) ( ) ( ) (
0
t y t r
dt
d
T k d y r
T
k
t y t r k t u
d c
t
i
c
c c
÷ + ÷ + ÷ =
)
¸ t t t 
(3.19)
where u
c
(t) is the control signal, y(t) is the process output, r(t) is the set
point, and k
c
, T
i
, T
d
, , 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 ,
1


.

\

+ + =


.

\

+ + = s T
s T
k C s T
s T
k C
d
i
c c d
i
c ff
¸  (3.20a)
k
v
T
1
k
v
Figure 3.5 A unit step response of the IPZ process, if the lowpass filtering part is
removed or equivalently T
2
Ł 0.
Chapter 3. Blackbox Models and Controller Structures
54
where C
ff
and C
c
are transfer functions, defined by the block diagram in
Figure 3.6. A lowpass filter added to the derivative part gives
,
1
1
1 ) ( ,
1
1
) (




.

\

+
+ + =




.

\

+
+ + =
s
N
T
s T
s T
k s C
s
N
T
s T
s T
k s C
d
d
i
c c
d
d
i
c ff
¸

(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
C
ff
(s)
C
c
(s) P(s) Ȉ Ȉ Ȉ
1
r
y
u
c
n d
Figure 3.6 Block diagram of the PIDcontroller used in this thesis, together with the
process. r is the set point, u
c
the control signal, y the measurement, d a load disturbance,
and n measurement noise.
Chapter 3. Blackbox Models and Controller Structures
55
controller be given by (3.20a). Now, assume that the process has an
integrator. Write the process transfer function as
), (
ˆ
1
) ( s P
s
s P = (3.21)
where
0 ) 0 (
ˆ
= P and , 0 ) (
ˆ
lim
0
=
÷
s P s
s
(3.22)
and since
( ) ), ( ) ( ) ( ) ( 1 ) ( ) ( ) ( s R s G s R s G s Y s R s E
re ry
= ÷ = ÷ = (3.23)
where G
ry
(s) is the transfer function from set point to process output and
G
re
(s) is the transfer function from set point to control error, we can write
the transfer function from set point to control error as
0 50 100 150 200 250 300
725
730
735
740
745
P
r
e
s
s
u
r
e
(
k
P
a
)
0 50 100 150 200 250 300
46
48
50
52
54
56
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Time (s)
Figure 3.7 Typical closed loop step response of the pressure loop when using a PID
controller, taken from a fluting machine.
Chapter 3. Blackbox Models and Controller Structures
56
( ) ( )
( )
.
) (
ˆ
) 1 (
) (
ˆ
) 1 ( 1
) (
ˆ
) 1 (
) (
ˆ
1 ) (
ˆ
1
) (
ˆ
1 1
1 1
) (
ˆ
1 1
) (
ˆ
1 1
1 1
) (
ˆ
1 1
1 1
) (
ˆ
1 1
1 ) ( 1 ) (
2 2
2
2 2
2 2 2
s P s T T s T k s T
s T k s P s T s T
s P s T T s T k s T
s P s T T s T k s P s T T s T k s T
s P
s
s T
s T
k
s P
s
s T
s T
k s P
s
s T
s T
k
s P
s
s T
s T
k
s P
s
s T
s T
k
s G s G
d i i c i
i c d i
d i i c i
d i i c d i i c i
d
i
c
d
i
c d
i
c
d
i
c
d
i
c
ry re
+ + +
÷ + ÷ +
=
+ + +
+ + ÷ + + +
=


.

\

+ + +


.

\

+ + ÷


.

\

+ + +
=


.

\

+ + +


.

\

+ +
÷ = ÷ =
¸ 
¸ 
¸ 
¸ 
(3.24)
Now let us study the integral of the controller error as time goes to
infinity and the set point is a step.
( )
( ) . 1
) (
ˆ
) 1 (
) (
ˆ
) 1 ( 1
lim
) (
1
lim ) (
2 2
0
0
2
0
i
d i i c i
i c d i
s
re
s
T
s P s T T s T k s T
T k s P s T s T
s G
s
s dt t e

¸ 
÷ =
+ + +
÷ + ÷ +
=
=
÷
·
÷
)
(3.25)
In the classical case, when  = 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  >1 will give an overshoot in the step response since the
integral has a negative value. The range of  is normally defined as
, 1 0 s s  (3.26)
Chapter 3. Blackbox Models and Controller Structures
57
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, k
c
, in (3.25).
Figure 3.8 shows two closed loop frequency plots for the process in
Figure 3.7. G
ry
is the response from set point r to output y. G
ny
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 G
ry
, 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.
10
4
10
3
10
2
10
1
10
0
10
1
0
0.5
1
1.5
G
r
y
10
4
10
3
10
2
10
1
10
0
10
1
0
0.2
0.4
0.6
0.8
1.0
1.2
G
n
y
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 PIDcontroller.
Chapter 3. Blackbox Models and Controller Structures
58
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 twodegreeoffreedom controller. The transfer function from
set point to process output is then
,
1 1
) (
PC
M PM
M
PC
M CM P
G
y u
y
u y
ry
+
÷
+ =
+
+
= (3.27)
where M
y
is the desired response from r to y. The condition for ideal
feedforward is
.
1
y u
M P M
÷
= (3.28)
If we require that M
u
is realizable, M
y
should have a time delay that is
equal to or larger than the time delay of the process and the pole excess of
M
y
should be at least as large as the pole excess of P. In addition, M
y
should have the same zeros in the right half plane as P.
M
y
M
u
C P
í1
Ȉ Ȉ
r
u
fb
y
u
ff
e
u
c
Figure 3.9 A twodegreeoffreedom controller (2DOF), where P is the process, C the
controller, and M
u
and M
y
are feedforward filters. The advantage of this configuration is
that the servo and regulation problems are separated.
Chapter 3. Blackbox Models and Controller Structures
59
Let P be given by the IPZ transfer function
,
) 1 (
1
2
1 sL
v
e
sT s
sT
k P
÷
+
+
= (3.29)
and choose the desired set point response to
,
1
1
sL
cl
y
e
sT
M
÷
+
= (3.30)
then
,
) 1 (
1
) ( ) 1 (
1
) (
) 1 )( 1 (
) 1 (
1 1 1
2 1
1
2
1
2
1
2
sT T T T k
T T
sT T T T k
T T
T T k
T
sT sT k
sT s
M
cl v cl cl cl v
cl
cl v
cl v
u
+ ÷
÷
÷
+ ÷
÷
+ =
+ +
+
=
(3.31)
and u
ff
can be generated as the output of the sum of a constant gain and
two low pass filters. This is necessary if the DCSsystems 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 u
fb
remains constant during a set point change and the purpose of C is simply
CONTR
MV
OUTP
SP
FF
M
u
M
y
y
r
u
c
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.
Chapter 3. Blackbox Models and Controller Structures
60
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 antiwindup, bumpless transfer, and other
functions. Observe that normally there is a pure time delay in M
y
, which
also should be implementable in the control system.
How should T
cl
be chosen? Looking at the shape of the control signal
gives a hint. The transfer function from set point to control signal is
,
) 1 (
) 1 (
) 1 (
2
1 cl v
u ru
sT
sT
sT k
s
M G
+
+
+
= =
(3.32)
and we have leadnetwork (lowpass filtered derivative) followed by a
zeropolenetwork. If T
cl
< T
2
, more lead action is added to the first lead
network, and T
cl
> T
2
gives a lowpass filter. Observe that T
1
is always
greater than T
2
.
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
1.0
1.2
P
r
o
c
e
s
s
o
u
t
p
u
t
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
1
Time (s)
C
o
n
t
r
o
l
s
i
g
n
a
l
Figure 3.11 Set point step response of the 2DOFcontroller. T
cl
= 5 (solid), T
cl
= 10
(dashed), and T
cl
= 20 (dashdotted). The dotted line is the set point.
Chapter 3. Blackbox Models and Controller Structures
61
If r is a unit step, the initial value theorem gives
. ) 0 (
1
2
T T k
T
u
cl v
c
= (3.33)
This can also be seen in the partial fraction expansion of M
u
in (3.31).
Figure 3.11 shows a simulation where the process is given by k
v
=0.01,
T
1
=200, T
2
=10, L=2. It clearly shows how the relation between T
cl
and T
2
affects the shape of u
c
. The selection of T
cl
will be a tradeoff between
performance and control action, and letting T
cl
T
2
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/T
1
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
0 50 100 150 200 250 300
0
0.2
0.4
0.6
0.8
1.0
1.2
P
r
o
c
e
s
s
o
u
t
p
u
t
0 50 100 150 200 250 300
0
0.2
0.4
0.6
0.8
1
Time (s)
C
o
n
t
r
o
l
s
i
g
n
a
l
Figure 3.12 Dashed í 2DOFcontroller T
cl
= 10, solid í PI with aggressive tuning, and
dashdotted í PI with robust tuning. The thin dashed curve is the set point. Note that the
2DOFcontroller 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.
Chapter 3. Blackbox Models and Controller Structures
62
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 steadystate, and the cylinder and paper
temperatures are still increasing.
Figure 3.12 shows how a welltuned PI controller behaves without the
feedforward compared to the 2DOFcontroller. 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 2DOFcontroller 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 2DOFcontroller structure. It is also an
advantage that this structure is easy to implement into most commercial
0 20 40 60 80 100 120 140 160 180
0
0.5
1.0
P
r
o
c
e
s
s
o
u
t
p
u
t
0 20 40 60 80 100 120 140 160 180
0
0.5
1.0
Time (s)
C
o
n
t
r
o
l
s
i
g
n
a
l
Figure 3.13 Robustness analysis to modeling errors of the 2DOFcontroller. Perfect
process model (solid), doubled time delay in process (dotted), and doubled process gain
(dashed). The dashdotted line is the set point.
Chapter 3. Blackbox Models and Controller Structures
63
DCSsystems, 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 M
u
. 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
.
) (
1
) 1 (
1
) (
1
2 1
2
2 2
1
2
1
a s s
b s b
T
s s
T
k
s
T
T k
sT s
sT
k s G
v v
v
+
+
=


.

\

+
+
=
+
+
= (3.34)
Assume that the model is written in observable canonical form
 
.
) ( 0 1 ) ( ) (
) ( ) (
0 0
1
) ( ) ( ) (
2
1 1
¦
¹
¦
´
¦
= =
(
¸
(
¸
+
(
¸
(
¸
÷
= + =
t x t Cx t y
t u
b
b
t x
a
t Bu t Ax t x
c c
(3.35)
Chapter 3. Blackbox Models and Controller Structures
64
By definition this system is observable. It is also controllable unless T
1
=
T
2
which implies that we have a pure integrator instead of an IPZ process.
Let the state feedback be given by
  ). ( ) ( ) ( ) ( ) (
2 1
t r l t x l l t r l t Lx t u
r r c
+ ÷ = + ÷ = (3.36)
The closed loop system then becomes
 
,
) ( 0 1 ) (
) ( ) (
1
) (
2
1
2 2 1 2
2 1 1 1 1
¦
¦
¹
¦
¦
´
¦
=
(
¸
(
¸
+
(
¸
(
¸
÷ ÷
÷ ÷ ÷
=
t x t y
t r
b
b
l t x
l b l b
l b l b a
t x
r
(3.37)
which in inputoutput form becomes
( )
.
) (
) (
1 2 2 2 1 2 2 1 1 1
2
2 1
l b l b a s l b l b a s
l b s b
s G
r
ry
+ + + + +
+
= (3.38)
By choosing
,
2 1 1
l a l l
r
+ = (3.39)
we get unit steadystate gain. If we choose to place the poles in Ȝ
1
and Ȝ
2
,
we get the equation system
.
2 1
1 2 1
2
1
2 1 2
2 1
(
¸
(
¸
÷ ÷ ÷
=
(
¸
(
¸
(
¸
(
¸
ì ì
ì ì a
l
l
b a b
b b
(3.40)
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.
Chapter 3. Blackbox Models and Controller Structures
65
A Simulation Example
Assume that we have the system
( )
( )
.
1 . 0
002 . 0 1 . 0
1 10
1 50
02 . 0
1
2
2 1
2
s a s
b s b
s s
s
s s
s
s G
+
+
=
+
+
=
+
+
= (3.41)
The open loop poles are in p
1
= 0 and p
2
= í0.1, and the zero is in
z = í0.02. To examine a cancellation of the process zero, place one pole
at Ȝ
1
= í0.2, and let the other one be at Ȝ
2
= 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
. 100
, 1125 5 . 12
, 5 . 12 25 . 1
2
2 2
2 1
ì
ì
ì
÷ =
÷ ÷ =
+ =
r
l
l
l
(3.42)
Figure 3.14 shows the simulation results. We can see that when both
poles are faster than the process zero we get an overshoot due to the lead
action of the zero (which is closer to the origin). When Ȝ
2
= 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 Ȝ
2
= 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 x
3
.
3
Cx r y r x ÷ = ÷ = (3.43)
The new system can then be written as
Chapter 3. Blackbox Models and Controller Structures
66
.
1
0
0
0 0 0 1
0 0 0
0 1
2
1
3
2
1 1
3
2
1
r u b
b
x
x
x a
x
x
x
c
(
(
(
¸
(
¸
+
(
(
(
¸
(
¸
+
(
(
(
¸
(
¸
(
(
(
¸
(
¸
÷
÷
=
(
(
(
¸
(
¸
(3.44)
This augmented system is also controllable from u
c
unless T
1
= T
2
. By
using the standard state feedback law
,
3 3
x l Lx u
c
÷ ÷ = (3.45)
we get the following closed loop system
,
1
0
0
0 0 1
1
3
2
1
3 2 2 2 1 2
3 1 2 1 1 1 1
3
2
1
r
x
x
x
l b l b l b
l b l b l b a
x
x
x
(
(
(
¸
(
¸
+
(
(
(
¸
(
¸
(
(
(
¸
(
¸
÷
÷ ÷ ÷
÷ ÷ ÷ ÷
=
(
(
(
¸
(
¸
(3.46)
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1.0
1.5
y
,
r
0 20 40 60 80 100 120 140 160 180 200
0
1
2
3
4
Time (s)
u
c
Figure 3.14 Simulation of the three different cases of state feedback. Set point (solid), Ȝ
2
= í0.04 (dashed), Ȝ
2
= í0.02 (dotted), and Ȝ
2
= í0.01 (dashdotted).
Chapter 3. Blackbox Models and Controller Structures
67
which has the transfer function
.
) ( ) (
) (
) (
3 2 3 1 1 2 2 2 1
2
2 2 1 1 1
3
1 2 3
l b s l b l b l b a s l b l b a s
s b b l
s G
ry
÷ ÷ + + + + +
+ ÷
= (3.47)
The new state, x
3
, can be added by having an additional feedback loop as
shown in Figure 3.15. The integrator gain, k
i
, is then given by
.
3
l k
i
÷ = (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
, 0 0008 . 0 048 . 0 42 . 0
2 3
= + + + s s s (3.49)
and get the following system of equations
,
0008 . 0
048 . 0
42 . 0
3 2
3 1 1 2 2 2 1
1 1 2 2 1
¦
¹
¦
´
¦
= ÷
= ÷ +
= + +
l b
l b l b l b a
l b l b a
(3.50)
which has the solution l
1
= 3, l
2
= 10 and l
3
= í0.4. Figure 3.16 shows the
simulation result. Since there is an extra pole in the closed loop system
Process
Ȉ
í L
s
k
i
Ȉ
í 1
r
y u
c
Figure 3.15 Adding integral action to a state feedback system
Chapter 3. Blackbox Models and Controller Structures
68
due to the integrator state, x
3
, 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
PIDcontroller 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.
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1.0
0
0.5
1.0
y
,
r
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1
1.5
2
Time (s)
u
c
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 (dashdotted).
Observe that the dynamics are different in the two cases since the number of closed loop
poles is different.
Chapter 3. Blackbox Models and Controller Structures
69
3.5 A twopole 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 1
1
) (
2 1 3
3 2
1
T T T e
sT sT
sT
k s G
sL
p PPZ
> >
+ +
+
=
÷
(3.51)
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 (polepolezero).
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
1600 1800 2000 2200 2400 2600 2800 3000 3200
220
240
260
280
P
r
e
s
s
u
r
e
(
k
P
a
)
1600 1800 2000 2200 2400 2600 2800 3000 3200
15
20
25
30
35
40
45
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Time (s)
Au
·
Figure 3.17 Closed loop step response. Taken from a liner machine.
Chapter 3. Blackbox Models and Controller Structures
70
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 predictionerror 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 PPZstructure 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 A
( ) , 0 ) ( lim ) ( lim
0 0
0
0
r G u u
s
r
s G s u t u u
ru ru
s
c
t
A = ÷

.

\

+
A
= ÷ ÷ A
÷ ÷·
·
(3.52)
where G
ru
(s) is the transfer function from the set point r to the control
signal u
c
. 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 1 1
1 1
1
1 1
1 1
1
) ( ) ( 1
) (
) (
1
2
2
2
2
2
2
1
sL
i d i c v i
i d i c
sL
d
i
c v
d
i
c
IPZ c
ff
ru
e sT sT T T s k k sT T s
sT T s T T s s k
e
sT s
sT
s T
s T
k k
T s
sT
k
s G s C
s C
s G
÷
÷
+ + + + +
+ + +
=
+
+


.

\

+ + +


.

\

+ +
=
+
=
 ¸
¸ 
(3.53)
The steady state value of the control signal is then
( ) , 0 0 = A = A
·
r G u
ru
(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 PPZtype, see (3.51), then the
steady state value of the control signal will become
Chapter 3. Blackbox Models and Controller Structures
71
( )
( ) ( )
( )( )( )
( )( ) ( )( )
p
s
sL
d i i p c i
d i i c
PPZ c
ff
k
r
e sT T T s sT k k sT sT sT
r sT sT T T s T s k
r
G C
C
A
=
+ + + + + +
A + + + +
=
A
+
=
÷
0
1
2
3 2
3 2
2
1 1 1 1
1 1 1
0 0 1
0
¸ 
(3.55)
and it depends of the process gain and the size of the step.
Normally, the difference between the PPZ and the IPZmodel 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
330
340
350
360
370
380
390
400
410
420
430
S
t
e
a
m
p
r
e
s
s
u
r
e
(
k
P
a
)
Identification
Validation
0 1000 2000 3000 4000 5000 6000
0
20
40
60
Time (s)
v
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Figure 3.18 Model validation showing the difference between the IPZ (dashed) and the
PPZ (dotted) process. Measurements taken from a board machine.
Chapter 3. Blackbox Models and Controller Structures
72
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
( )
( )( )
s
e
s s
s
s G
83 . 2
1
2 . 74 1 797 1
291 1
5 . 15 ) (
÷
+ +
+
= (3.56)
and the estimated IPZ model is
( )
( )
s
e
s s
s
s G
83 . 2
2
138 1
1896 1
00421 . 0 ) (
÷
+
+
= (3.57)
The difference between the two models is apparent. The IPZmodel does
not capture the low frequency component of the process as well as the
PPZmodel does. By manipulating the parameters in the IPZmodel, 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 crossover 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 openloop 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 IPZmodel is adequate and well suited
for control design usage. We also note that compared to the PPZmodel,
the IPZmodel in general has larger phase lag. Thus, we can, by using the
IPZmodel for design, be sure to have a stable controller even if the
process is of PPZtype.
Chapter 3. Blackbox Models and Controller Structures
73
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”.
10
5
10
4
10
3
10
2
10
1
10
0
10
1
10
3
10
0
10
3
M
a
g
n
i
t
u
d
e
(
a
b
s
)
10
5
10
4
10
3
10
2
10
1
10
0
10
1
100
80
60
40
20
0
Frequency (rad/s)
P
h
a
s
e
(
d
e
g
)
Figure 3.19 Bode plot for the IPZ (dashed) and PPZ (dotted) models.
Chapter 3. Blackbox Models and Controller Structures
74
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 crossconnections between
the pressure and differential pressure, since they are both acting on the
same physical unit. We then have a twoinputtwooutput (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
(
¸
(
¸
(
(
(
(
¸
(
¸
+
+
+
+
+
+
=
(
¸
(
¸
÷ ÷
÷
dP
P
s s
s
dP
P
u
u
e
s s
s
e
s s
s
e
s s
s
y
y
6 . 2 2
7 . 1
) 1 1 . 3 (
1 343
0029 . 0
) 1 3 . 2 (
1 122
0395 . 0
0
) 1 9 . 1 (
1 29
186 . 0
(3.58)
Dryer group
LC
PC
Flash steam
Steam header
1
2
3
PDC
Flash steam
To boiler house
Tank
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.
Chapter 3. Blackbox Models and Controller Structures
75
2200 2400 2600 2800 3000 3200 3400
220
240
260
280
300
y
P
(
k
P
a
)
2200 2400 2600 2800 3000 3200 3400
6
8
10
12
14
16
u
P
(
%
)
2200 2400 2600 2800 3000 3200 3400
40
50
60
70
80
y
d
P
(
k
P
a
)
2200 2400 2600 2800 3000 3200 3400
14
16
18
20
22
24
26
Time (s)
u
d
P
(
%
)
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.
Chapter 3. Blackbox Models and Controller Structures
76
2000 2200 2400 2600 2800 3000 3200 3400 3600
280
300
320
340
y
P
(
k
P
a
)
2000 2200 2400 2600 2800 3000 3200 3400 3600
4
6
8
10
12
14
u
P
(
%
)
2000 2200 2400 2600 2800 3000 3200 3400 3600
50
60
70
80
90
y
d
P
(
k
P
a
)
2000 2200 2400 2600 2800 3000 3200 3400 3600
2
4
6
8
Time (s)
u
d
P
(
%
)
Figure 3.22 A second step response model of a similar process as in Figure 3.20. Taken
from a fine paper machine.
Chapter 3. Blackbox Models and Controller Structures
77
where y
P
and u
P
are output and input of the steam pressure process,
respectively, and y
dP
and u
dP
are output and input of the differential
pressure process, respectively. Most conspicuous of this model is that the
differential pressure valve, u
dP
, has no influence on the pressure, y
P
. 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, y
dP
. 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 blackbox 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 twodegreeof
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.
Chapter 3. Blackbox Models and Controller Structures
78
Finally, a multivariable 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.
Chapter 4. A Physical Model of a Steam Heated Cylinder
79
4.
In Chapter 3, an empirical blackbox 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
differentialalgebraic 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 greybox 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 greybox 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.
4
A Physical Model of a Steam
Heated Cylinder
Chapter 4. A Physical Model of a Steam Heated Cylinder
80
4.1 The model
Let q
s
(kg/s) be the mass flow rate of steam into the cylinder, q
c
(kg/s) be
the condensation rate, and q
w
(kg/s) be the siphon flow rate. Also, let V
s
(m
3
) and V
w
(m
3
) be the volume of steam and water in the cylinder, and let
µ
s
(kg/m
3
) and µ
w
(kg/m
3
) be the densities of steam and water. The mass
balances for water and steam are then
( ) ,
c s s s
q q V
dt
d
÷ = µ (4.1a)
( ) ,
w c w w
q q V
dt
d
÷ = µ (4.1b)
where no blowthrough steam is assumed. Sometimes the blowthrough
steam is modeled as a fraction of q
s
[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
( ) ,
s c s s s s s
h q h q V u
dt
d
÷ = µ (4.2a)
( ) ,
m w w s c w w w
Q h q h q V u
dt
d
÷ ÷ = µ (4.2b)
( ) ,
, p m m m p
Q Q T mC
dt
d
÷ = (4.2c)
where Q
m
(W) is the power supplied from the water to the metal, Q
p
(W)
is the power supplied from the metal to the paper, h
s
(J/kg) is the steam
enthalpy, h
w
(J/kg) is the water enthalpy, m (kg) the mass of the cylinder
shell, C
p,m
(J/(kgK)) the specific heat capacity of the shell, T
m
(K) the
mean temperature of the metal, u
s
(J/kg) and u
w
(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
Chapter 4. A Physical Model of a Steam Heated Cylinder
81
( ),
m s cyl sc m
T T A Q ÷ =o (4.3)
where o
sc
(W/(m
2
K)) is the heat transfer coefficient from the steam
condensate interface to the centre of the cylinder shell, A
cyl
(m
2
) is the
inner cylinder area, and T
s
(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
( ),
p m cyl cp p
T T A Q ÷ = q o (4.4)
where o
cp
(W/(m
2
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 T
p
(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.
Steam
Paper web
Dryer shell
Condensate
T
s
Q
m
T
m
T
p
Q
p
Figure 4.1 A piece of the crosssection of a drying cylinder, visualizing the assumption on
the temperature profile and the energy flows.
Chapter 4. A Physical Model of a Steam Heated Cylinder
82
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,
T
p
, 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 T
p
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 V
s
+ V
w
= 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§ V
s
.
Summarizing, we find that the system can be described by the equations
( )
( )
( )
( ),
,
, 0
,
,
,
p m cyl cp p
m s cyl sc m
m w w s c
p m m m p
c s s
T T A Q
T T A Q
Q h q h q
Q Q T mC
dt
d
q q V
dt
d
÷ =
÷ =
÷ ÷ =
÷ =
÷ =
q o
o
µ
(4.5)
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 q
c
and Q
m
, the model becomes
Chapter 4. A Physical Model of a Steam Heated Cylinder
83
( ) ( )
( ) ( ) ( ).
,
, p m cyl cp m s cyl sc m m p
m s cyl sc w w s s s s
T T A T T A T mC
dt
d
T T A h q h q V
dt
d
h
÷ ÷ ÷ =
÷ ÷ ÷ =
q o o
o µ
(4.6)
Assuming that the steam in the cylinder is saturated, enthalpies h
s
, h
w
,
density µ
s
and the temperature T
s
, are all functions of the pressure p. The
model can thus be written as
( )
( ) ( ), ) (
, ) ( ) ( ) ( ) ( ) (
, p m cyl cp m s cyl sc
m
m p
m s cyl sc w w s s
s
s
T T A T p T A
dt
dT
mC
T p T A p h p q p h q
dt
dp
dp
d
V p h
÷ ÷ ÷ =
÷ ÷ ÷ =
q o o
o
µ
(4.7)
where the states are pressure p and mean metal temperature T
m
. The steam
inlet flow, q
s
, is the input. The equilibrium gives the relations
( )
( ) ( ). ) ( 0
, ) ( ) ( ) ( ) ( 0
0 0 0 0
0 0 0 0 0 0
p m cyl cp m s cyl sc
m s cyl sc w w s s
T T A T p T A
T p T A p h p q p h q
÷ ÷ ÷ =
÷ ÷ ÷ =
q o o
o
(4.8)
Hence
.
) ( ) (
0 0 0 0
0 0
q o
cyl cp
w w s s
m p
A
p h q p h q
T T
÷
÷ =
(4.9)
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
Chapter 4. A Physical Model of a Steam Heated Cylinder
84
,
, ) (
) ( ) (
) (
0
0
0
,
0
0 0 0
0
m cyl cp m cyl sc
p p
s
cyl sc
m
m p
s s m cyl sc
p p
s
cyl sc
w
w
w
w
s
s
p p
s
s
T A T A p
dp
dT
A
dt
T d
mC
q p h T A
p
dp
dT
A
dp
dq
p h
dp
dh
p q
dp
dh
q
dt
p d
dp
d
V p h
A ÷ A ÷ A =
A
A + A +
A


.

\

÷ ÷ ÷ =
A
=
=
=
q o o o
o
o
µ
(4.10)
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
, ) ( ) (
0 0 0
dp
dT
A
dp
dq
p h
dp
dh
p q
dp
dh
q
s
cyl sc
w
w
w
w
s
s
o << ÷ ÷ (4.11a)
and
,
sc cp
o q o << (4.11b)
the model becomes
.
, ) ( ) (
0
0 0
,
0 0
m cyl sc
p p
s
cyl sc
m
m p
s s m cyl sc
p p
s
cyl sc
p p
s
s
T A p
dp
dT
A
dt
T d
mC
q p h T A p
dp
dT
A
dt
p d
dp
d
V p h
A ÷ A =
A
A + A + A ÷ =
A
=
= =
o o
o o
µ
(4.12)
Assumption (4.11b) is another way of saying that Q
p
is varying much
more slowly than Q
m
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 o
cp
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
Chapter 4. A Physical Model of a Steam Heated Cylinder
85
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 statespace form, we find that
,
,
s
s
q D Cx y
q B Ax x
A + =
A + =
(4.13)
where  
T
m
T p x A A = and
  . 0 , 0 1
,
0
1
,
, ,
= =
(
(
(
(
(
(
¸
(
¸
=
(
(
(
(
(
(
(
(
¸
(
¸
÷
÷
=
D C
dp
d
V
B
mC
A
mC
dp
dT
A
dp
d
V h
A
dp
d
V h
dp
dT
A
A
s
m p
cyl
m p
s
cyl
s
s
cyl
s
s
s
cyl
sc
µ
µ µ
o
(4.14)
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
( ) ( )
, ) (
1
1
22 11
22
1

.

\

+
÷
+ =
+
+
=
÷ ÷
÷
=
ì
ì
ì ì s
z
s
z b
s s
z s
b
a a s s
a s
b s G (4.15)
where
Chapter 4. A Physical Model of a Steam Heated Cylinder
86
.
, ,
1
,
22 11
22 1




.

\

+ = ÷ ÷ =
= ÷ = =
m p
cyl
s
s
s
cyl
sc
p
cyl sc
s
mC
A
dp
d
V h
dp
dT
A
a a
mC
A
a z
dp
d
V
b
µ
o ì
o
µ
(4.16)
Note that the model (4.15) has an IPZ structure. The essential parameters
of the model are
 Cylinder volume V
 Cylinder mass m
 Specific heat capacity of metal C
p,m
 Area of the cylinder surface A
cyl
 Steam properties dp dT dp d h
s s s
, , µ
 Heat transfer coefficient o
sc
All parameters, except the heat transfer coefficient, o
sc
, 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
 No blowthrough steam
 The steam in the cylinder is saturated
 Paper temperature is constant
 The thermal dynamics of the condensate is fast compared to the
cylinder shell
 The condition (4.11)
The pole ì and the zero z are both proportional to the heat transfer
coefficient o
sc
. For large s the transfer function (4.15) is approximated by
Chapter 4. A Physical Model of a Steam Heated Cylinder
87
, ) (
1
s
b
s G ~ (4.17)
where b
1
does not depend on o
sc
. For small s the transfer function can be
approximated by
, ) (
1
s
z b
s G
ì
~ (4.18)
where
,
,
1
dp
d
Vh
dp
dT
mC
h z b
s
s
s
m p
s
µ
ì
+
= (4.19)
does not depend on o
sc
. 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 blackbox and greybox parameters are
.
, ,
,
,
2
,
1
,


.

\

+
=
=
+
=
dp
d
Vh
dp
dT
mC A
dp
d
V h mC
T
A
mC
T
dp
d
Vh
dp
dT
mC
h
k
s
s
s
m p cyl sc
s
s m p
cyl sc
m p
s
s
s
m p
s
v
µ
o
µ
o
µ
(4.20)
Note that the velocity gain, k
v
, above is not normalized by the measuring
or any actuator range.
As can be noticed, the greybox model does not explain the time delay
often seen in the blackbox 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 greybox model
that there are neglected dynamics in the model, which may possibly give
rise to an estimation of the deadtime that is larger than the true value.
Chapter 4. A Physical Model of a Steam Heated Cylinder
88
Remark 1
A unit analysis of the expressions in (4.20) shows that T
1
and T
2
are given
in seconds, while k
v
is given in Pa/kg. Since k
v
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
, T kA Q V ÷ = (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 buildup of heat at any point along the path of the heat flow, Q is
constant along the path. Consider a onedimensional and homogeneous
system, and integrate (4.21) from point “0” to point “1”
) )
÷ =
1
0
1
0
x
x
T
T
dT kA Qdx (4.22)
or equivalently
). (
1 0
0 1
1 0
T T A
x
k
x x
T T
kA Q ÷
A
=
÷
÷
= (4.23)
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 buildup. By combining the energy balance for the metal in (4.2c) with
), (
m s cyl sc m
T T A Q ÷ =o (4.3)
we get a firstorder 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).
Chapter 4. A Physical Model of a Steam Heated Cylinder
89
Remark 3
The steadystate solution to (4.1) and (4.2) is
p m
w s c m
w c s
Q Q
h h q Q
q q q
=
÷ =
= =
) ( (4.24)
As expected, both the mass flows and energy flows are equal. The energy
flow to the metal, Q
m
, 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/m
2
. The machine dependent values used for the
simulation are
 Cylinder volume: V = 12.6 m
3
 Cylinder mass: m = 7610 kg
 Cylinder area: A
cyl
= 37.2 m
2
 Heat capacity for cast iron: C
p,m
= 500 J/(kgK)
 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 o
sc
, 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 PIDcontroller, this
difference in gain and phase influences the controller parameters.
Otherwise it would be possible to tune the controller solely from cylinder
Chapter 4. A Physical Model of a Steam Heated Cylinder
90
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 o
sc
, 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
10
3
10
2
10
1
10
0
10
1
10
0
10
1
M
a
g
n
i
t
u
d
e
(
a
b
s
)
10
3
10
2
10
1
10
0
90
65
45
Frequency (rad/s)
P
h
a
s
e
(
d
e
g
)
Figure 4.2 Frequency properties for different values of the heat transfer coefficient o
sc
=
500 (dotted), 1000 (dashed), and 2000 (solid).
Chapter 4. A Physical Model of a Steam Heated Cylinder
91
from the steam properties, we also need an expression for the derivative of
siphon flow rate, q
w
, 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 greybox 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
 Cylinder volume: V = 18.4 m
3
 Cylinder mass: m = 8300 kg
 Cylinder area: A
cyl
= 45.5 m
2
0 10 20 30 40 50 60 70 80 90 100
0
2000
4000
6000
A
p
(
P
a
)
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
Time (s)
A
T
(
K
)
Figure 4.3 Step responses for different values of the heat transfer coefficient o
sc
= 500
(dotted), 1000 (dashed), and 2000 (solid). The upper graph shows the steam pressure and
the lower shows the cylinder temperature.
Chapter 4. A Physical Model of a Steam Heated Cylinder
92
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, u
c
, and the steam input flow, q
s
, namely
,
c v s
u d q = (4.25)
where d
v
is a valve constant which will be the second calibration
parameter together with the heat transfer coefficient, o
sc
. Using this valve
description, we keep the linearity and IPZstructure 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 steadystate pressure level can only be reached for a
closed valve). The calibration parameters obtained are
o
sc
= 1820 W/(m
2
K), d
v
= 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,
sc
o'
. From
[Karlsson, 2000] the relationship
,
1
1
cyl
cyl
sc
sc
ì
o
o
o
+
'
= (4.27)
is given, where o
cyl
is the distance into the cylinder where the temperature
is equal to the mean cylinder temperature, and ì
cyl
is the thermal
conductivity of the cylinder shell. In this example, the cylinder thickness
is 25 mm (the mean temperature, T
m
, occurs in the middle of the cylinder
shell), and the thermal conductivity is 50 W/(mK). The heat transfer
coefficient through the condensate is then
sc
o' = 3340 W/(m
2
K). (4.28)
Chapter 4. A Physical Model of a Steam Heated Cylinder
93
0 200 400 600 800 1000 1200 1400 1600 1800 2000
360
365
370
375
p
(
k
P
a
)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
35
40
45
50
55
Time (s)
u
c
(
%
)
Figure 4.5 Validation of model (dotted) against measured data (solid).
0 50 100 150 200 250 300 350 400 450 500 550
415
420
425
p
(
k
P
a
)
0 50 100 150 200 250 300 350 400 450 500 550
45
50
55
60
Time (s)
u
c
(
%
)
Figure 4.4 Calibrated model (dotted) and measured data (solid).
Chapter 4. A Physical Model of a Steam Heated Cylinder
94
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/(m
2
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 d
v
, 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 greybox model is compared with the corresponding
blackbox model, adjusted on the same data set. The velocity gain in the
models is normalized with the measuring range of the pressure gauge. The
blackbox 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).
( )
( )
( )
( )
s
black
s
grey
e
s s
s
s G
e
s s
s
s G
÷
÷
+
+
=
+
+
=
1 3 . 21
1 8 . 77
00176 . 0 ) (
1 4 . 20
1 1 . 50
00243 . 0 ) (
(4.29)
The greybox 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 IPZstructure 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 greybox
Chapter 4. A Physical Model of a Steam Heated Cylinder
95
model this can be accomplished by disregarding the assumption from
(4.11b), saying that
.
sc cp
o q o << (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
,
) (
, ,
(
(
(
(
(
(
(
(
¸
(
¸
+
÷
÷
= '
m p
cp sc cyl
m p
s
cyl sc
s
s
cyl sc
s
s
s
cyl sc
mC
A
mC
dp
dT
A
dp
d
V h
A
dp
d
V h
dp
dT
A
A
q o o
o
µ
o
µ
o
(4.30)
and the other matrices are unchanged. By examining time series where the
IPZstructure is sufficient with cases where it is not, an explanation to the
modeling problem is found. If o
sc
>>o
cp
then model (4.13) is adequate
and o
sc
§ o
cp
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
o
sc
>>o
cp
gives a system with one fast pole and one close to the origin.
When o
cp
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
, ) (
21 12 22 11 22 11
2
a a a a s a a s ÷ + + ÷ (4.31)
where a
11
, a
12
, a
21
, and a
22
are the elements of matrix A´. Identification of
the parameters gives
Chapter 4. A Physical Model of a Steam Heated Cylinder
96
, 0
, 0
) (
,
2
21 12 22 11
,
22 11
> = ÷
<
+
÷ ÷ = +
dp
d
VmC h
dp
dT
A
a a a a
mC
A
dp
d
V h
dp
dT
A
a a
s
m p s
s
cyl cp sc
m p
cp sc cyl
s
s
s
cyl sc
µ
o qo
q o o
µ
o
(4.32)
and since the coefficients of the polynomial (4.31) are all positive, the
system is stable. The roots of (4.31) are
,
4
) (
2
21 12
2
22 11 22 11
2 , 1
a a
a a a a
s +
÷
±
+
= (4.33)
and since both a
12
and a
21
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
) (
s
dp
d
V
s G
s
µ
~ (4.34)
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
.
) (
) ( lim
0
dp
dT
A
h
s G
s
cp cyl sc
cp sc s
s
qo o
q o o +
=
÷
(4.35)
Chapter 4. A Physical Model of a Steam Heated Cylinder
97
4.5 Summary
Chapter 3 introduced a blackbox 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 blackbox model is purely controller tuning and the main
purpose of the greybox model is to gain insight into the physical laws
behind the blackbox 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 greybox 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.
Chapter 5. A Tuning Method for IPZ Models
98
5.
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 firstorder 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 IPZtuning rule is tested and evaluated on an industrial
paper machine. It is also compared with a few other design methods.
5
A Tuning Method for IPZ
Models
Chapter 5. A Tuning Method for IPZ Models
99
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 nonconvex optimization is much more demanding due to less
effective numerics and the existence of several local optimal solutions.
The derivation of the IPZtuning 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 k
i
= k
c
/ T
i
, subject to a robustness
constraint. Due to the form of the constraint, the optimization problem is
nonconvex. 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
). (
) ( ) ( 1
) (
) (
) ( ) ( 1
) ( ) (
1
) (
) ( ) ( 1
) ( ) (
) (
) ( ) ( 1
) (
) (
) ( ) ( ) (
s D
s C s P
s P
s R
s C s P
s C s P
s R
s C s P
s C s P
s D
s C s P
s P
s R
s Y s R s E
c c
ff
c
ff
c
+
÷


.

\

+
÷ =


.

\

+
+
+
÷ =
÷ =
(5.1)
Assume a step unit load disturbance and that the set point is zero. Evaluate
the integral of the error by the final value theorem
Chapter 5. A Tuning Method for IPZ Models
100
.
) ( ) 1 (
) (
lim
1
1
) ( 1
) (
lim
1
) ( ) ( 1
) (
lim ) ( IE
2
0
2
0
0
0
c
i
d i i c i
i
s
i
d i i
c
s
c
s
k
T
s P s T T s T k s T
s P T
s
s T
s T T s T
k s P
s P
s s C s P
s P
dt t e
÷ =
+ + +
÷ =
+ +
+
÷ =
+
÷ = =
÷
÷
÷
·
)
(5.2)
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, R
0
, stability is guaranteed. The radius of
the circle will then be the design parameter, and the smaller R
0
is the more
aggressive the controller will be. This can also be expressed in terms of
the sensitivity function, defined as
Time
C
o
n
t
r
o
l
e
r
r
o
r
Negative area
Positive area
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.
Chapter 5. A Tuning Method for IPZ Models
101
,
) ( 1
1
) (
s L
s S
+
= (5.3)
where L(s) is the loop transfer function. The maximum sensitivity, M
s
, is
then given by
.
) ( 1
1
max ) ( max
e
e
e e
i L
i S M
s
+
= = (5.4)
Since 1+L(ie) 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/M
s
. Therefore, we get
,
1
0
R
M
s
= (5.5)
and we can then use M
s
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 M
s
= 1.2 means in practice. Simulations and
familiarization is one answer to that issue.
Im
Re
R
0
–1
Figure 5.2 The robustness constraint of the design method. By varying the radius of the
circle, the degree of robustness is changed.
Chapter 5. A Tuning Method for IPZ Models
102
The shape of the Nyquist curve is changed by varying the controller
parameters, in order to avoid the M
s
circle while achieving a high integral
gain. The derivative time T
d
, affects the curve close to the origin (high
frequencies), the integral time T
i
affects low frequencies, while the gain k
c
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 ít. At high frequencies the phase is ít/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  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 (M
s
= 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.
Figure 5.3 The complete Nyquist curve for the IPZprocess and a PID controller if the
dead time is assumed to be zero. Otherwise, there will be the typical circular appearance in
the origin.
Chapter 5. A Tuning Method for IPZ Models
103
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 longterm 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 M
s
values. The M
s
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 << T
2
to L > T
2
, T
1
from 50 to
800, and T
2
from 2 to 400. However, process parameters T
1
, T
2
, and L are
presumed to be positive. This implies that, e.g. nonminimum phase
systems are not included. Furthermore, T
1
is always larger than T
2
but
cases where T
2
is almost as large as T
1
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 M
s
= 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 k
v
, for example, is given in kPa/(%s) or
simply s
í1
, 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, k
v
and k
c
belongs together, in
Chapter 5. A Tuning Method for IPZ Models
104
combination with a parameter that is given in seconds (T
1
, T
2
, T
i
, or L). As
will be obvious shortly, it must be L because of the influence it has on the
controller gain. Moreover, larger T
1
compared to T
2
give more lead action
in the process, which must results in a smaller controller gain, given a
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
(T
2
+L/3)/T
1
k
v
k
c
L
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.5
0.6
0.7
0.8
0.9
1.0
T
2
/ (T
2
+L)
T
i
/
(
T
i
+
L
)
Figure 5.4 The linear relationship between the PI controller parameters and process
parameters for M
s
= 1.1. Each circle represents one set of process parameters from the
investigated batch and corresponding control parameters from the optimization routine.
Chapter 5. A Tuning Method for IPZ Models
105
certain robustness. Therefore the ratio between T
1
and T
2
(together with a
correction term L/3, that is found ad hoc) is related to k
c
. Also, it seems
natural to expect T
i
depend on the ‘time constant’ T
2
, 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 M
s
values are 1.1, 1.2, 1.3, and 1.4. The resulting
controller parameters are given by
.
85 12
15 88
,
3
28 . 0 : 4 . 1
,
94 11
17 100
,
3
23 . 0 : 3 . 1
,
105 9
4 23
5 ,
3
16 . 0 : 2 . 1
,
21
6
4 ,
3
09 . 0 : 1 . 1
2
2
1
2
2
2
1
2
2
2
1
2
2
2
1
2
L T
L T
L T
L k T
L
T
k M
L T
L T
L T
L k T
L
T
k M
L T
L T
L T
L k T
L
T
k M
L T
L T
L T
L k T
L
T
k M
i
v
c s
i
v
c s
i
v
c s
i
v
c s
+
+
=
+
= =
+
+
=
+
= =
+
+
=
+
= =
+
+
=
+
= =
(5.6)
By looking at the controller gain for the different M
s
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 T
2
of the IPZ model, for a specific time delay L. Here we can see
that the larger the value of the design parameter, M
s
, is the smaller will the
integral time be, for a given value of T
2
. As in the analysis of k
c
above,
this is also expected.
Chapter 5. A Tuning Method for IPZ Models
106
If the time delay in the IPZmodel 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 k
c
and T
i
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]
 Sensor noise
 Unmodeled process dynamics
 Actuator saturation
 Admissible control signal variations
 Sensor and actuator resolution
0 10 20 30 40 50
0
5
10
15
20
25
30
T
2
T
i
Solid: M
s
= 1.1
Dashed: M
s
= 1.2
Dotted: M
s
= 1.3
Dashdotted: M
s
= 1.4
Figure 5.5 The integral time, T
i
, as function of the time constant T
2
, for the PI tuning rule.
For small T
2
, the two variables are almost equal, otherwise T
i
is smaller that T
2
. The thin
solid line shows T
i
= T
2
, as a comparison. This is also suggested in a tuning method by
[Nelson and Gardner, 1996].
Chapter 5. A Tuning Method for IPZ Models
107
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 DCSsystems. Thus, this
should not impose large restrictions to the usefulness of the method.
In Figure 5.7 and Figure 5.8 simulations of an IPZtuned PI controller
for different values of M
s
and with ȕ = 1 or ȕ = 0. The process is given by
.
) 1 20 (
) 1 200 (
01 . 0 ) (
s
e
s s
s
s P
÷
+
+
= (5.7)
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 2DOFcontroller 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
counterintuitive phenomenon can also be seen when controlling a pure
integrating process with a PI or PID controller.
2
1
T
÷
1
1
T
÷
i
T
1
÷
Figure 5.6 The root locus for the IPZprocess in connection with a PI controller for the
case T
1
> T
i
> T
2
. Good tuning normally requires T
1
>> T
i
§ T
2
.
Chapter 5. A Tuning Method for IPZ Models
108
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1.0
r
,
y
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1
1.5
2
Time (s)
u
c
Figure 5.8 Simulation with the same process and controller as in Figure 5.7 but with ȕ = 0.
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1.0
1.5
r
,
y
0 20 40 60 80 100 120 140 160 180 200
0
1
2
3
4
Time (s)
u
c
Figure 5.7 Closed loop response of (5.7) and an IPZtuned PI controller for M
s
= 1.1
(solid), M
s
= 1.2 (dashed), M
s
= 1.3 (dotted), and M
s
= 1.4 (dashdotted). ȕ = 1.
Chapter 5. A Tuning Method for IPZ Models
109
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 / T
1
.
Remark 3
The feature of the IPZtuning, to give a nonrealizable 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 T
2
will generally give a short integral time, T
i
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 1
1 lim
2
1
2
1
T
T T k k
e
T i i
T i
i T
i T
k k
d v c
L i
d
i
v c
=
+
+


.

\

+ +
÷
· ÷
e
e
e e
e
e
e
(5.8)
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 M
s
= 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 seconddegree 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 IPZtuned PID
controller for different values of M
s
and with ȕ = 1 or ȕ = 0.
Chapter 5. A Tuning Method for IPZ Models
110
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
(T
2
+L/3)/T
1
k
v
k
c
L
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
T
2
/ (T
2
+L)
T
i
/
(
T
i
+
L
)
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.2
0.4
0.6
0.8
1.0
T
2
/ (T
2
+L)
T
d
/
T
2
Figure 5.9 The relationship between the PID controller parameters and process parameters
for M
s
= 1.2.
Chapter 5. A Tuning Method for IPZ Models
111
M
s
= 1.1:
10 ,
) 2 ( 1000
) 1359 264 7 (
) 354 529 28 (
) 146 471 472 (
22
3
2
2
2
2
2
2
2
2 2
2
2
2
2
2
2
2
2
1
2
=
+ +
+ +
=
+ +
+ +
=
+
=
N
L L T T
L L T T T
T
L L T T
L L T T L
T
L T k
L T
k
d
i
v
c
M
s
= 1.2:
10 ,
) 2 ( 1000
) 1400 361 (
) 238 375 39 ( 3
) 286 875 883 (
12
3
2
2
2
2
2
2
2
2 2
2
2
2
2
2
2
2
2
1
2
=
+ +
+ +
=
+ +
+ +
=
+
=
N
L L T T
L L T T T
T
L L T T
L L T T L
T
L T k
L T
k
d
i
v
c
(5.9)
M
s
= 1.3:
10 ,
) 2 ( 1000
) 736 176 3 ( 2
) 241 386 55 ( 3
) 277 842 835 (
26
) 3 ( 3
2
2
2
2
2
2
2
2 2
2
2
2
2
2
2
2
2
1
2
=
+ +
+ +
=
+ +
+ +
=
+
=
N
L L T T
L L T T T
T
L L T T
L L T T L
T
L T k
L T
k
d
i
v
c
M
s
= 1.4:
10 ,
) 2 ( 1000
) 1443 367 8 (
) 722 1149 214
) 278 851 786 (
20
) 3 ( 3
2
2
2
2
2
2
2
2 2
2
2
2
2
2
2
2
2
1
2
=
+ +
+ +
=
+ +
+ +
=
+
=
N
L L T T
L L T T T
T
L L T T
L L T T L
T
L T k
L T
k
d
i
v
c
Chapter 5. A Tuning Method for IPZ Models
112
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1.0
1.5
r
,
y
0 20 40 60 80 100 120 140 160 180 200
0
2
4
6
Time (s)
u
c
Figure 5.10 Closed loop response of (5.7) and an IPZtuned PID controller for M
s
= 1.1
(solid), M
s
= 1.2 (dashed), M
s
= 1.3 (dotted), and M
s
= 1.4 (dashdotted). ȕ = 1 and Ȗ = 0.
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1.0
r
,
y
0 20 40 60 80 100 120 140 160 180 200
0
1
2
3
Time (s)
u
c
Figure 5.11 Simulation with the same process and controller as Figure 5.10 but with ȕ = 0.
Chapter 5. A Tuning Method for IPZ Models
113
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 IPZtuning 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
( ), ) ( sin ) ( cos ) ( ) ( ) (
) (
e ¢ e ¢ e e e
e ¢
i r e r i G
i
+ = = (5.10)
and let the controller be given by
, ) ( s k
s
k
k s C
d
i
c
+ + = (5.11)
where k
d
= k
c
T
d
. The condition for oscillation is then given by Nyquist’s
theorem
( ) . 1 ) ( sin ) ( cos ) ( ) ( ) ( ÷ = 
.

\

+ ÷ + = e
e
e ¢ e ¢ e e e
d
i
c
ik
k
i k i r i C i G (5.12)
Identifying the real and imaginary parts, we find that the boundary of the
stability region can be represented by
.
) (
) ( sin
,
) (
) ( cos
2
e
e ¢ e
e
e
e ¢
r
k k
r
k
d i c
÷ = ÷ = (5.13)
For the IPZ process, we have
.
2
) arctan( ) arctan( ) (
,
) 1 (
) 1 (
) (
2 1
5 . 0 2
2
2
5 . 0 2
1
2
L T T
T
T
k r
v
e
t
e e e ¢
e e
e
e
÷ ÷ ÷ =
+
+
=
(5.14)
Figure 5.12 and Figure 5.13 show examples of stability regions and IPZ
tuning. The IPZtuning gives parameters rather distant from the boundary
Chapter 5. A Tuning Method for IPZ Models
114
0 5 10 15
0
2
4
6
8
10
k
c
k
i
0 5 10 15
0
2
4
6
8
10
k
c
k
i
0 5 10 15
0
2
4
6
8
10
k
c
k
i
0 5 10 15
0
2
4
6
8
10
k
c
k
i
M
s
= 1.1 M
s
= 1.2
M
s
= 1.3
M
s
= 1.4
Figure 5.13 Stability region for the IPZprocess controlled by a PID controller. The
process is given by k
v
= 0.005, T
1
= 50, T
2
= 5, L = 3. The stability region forms a three
dimensional surface in the k
c
ík
i
ík
d
plane. Here, it is visualized in two dimensional plots
by looking at the k
c
ík
i
plane for the optimal k
d
(constant in each plot) given by the IPZ
tuning rule.
0 5 10 15
0
1
2
3
4
k
c
k
i
0 5 10 15
0
1
2
3
4
k
c
k
i
Figure 5.12 Stability regions for the IPZprocess controlled by a PI controller. Left:
k
v
= 0.005, T
1
= 50, T
2
= 5, L = 3; Right: k
v
= 0.001, T
1
= 200, T
2
= 1, L = 2. The points
indicate the position of the IPZtuning for the different M
s
values in (5.6). The indications
are moving from the origin to larger k
c
and k
i
with larger M
s
.
Chapter 5. A Tuning Method for IPZ Models
115
of stability. With higher robustness (lower M
s
values), the tuning moves
closer to the origin. With lower robustness, the proposed tuning moves in
direction to the maximum k
i
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
IPZprocess is zero, the stability region fills the whole first quadrant, and
any and k
i
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 M
s
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
( )
( )
.
1 79 . 7
1 6 . 51
0196 . 0 ) (
30 . 1 s
e
s s
s
s P
÷
+
+
= (5.15)
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)
The step responses in the figures and the table show very well the
difference between various values of the tuning parameter M
s
. 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.
K
c
T
i
M
s
=1.1 0.56 7.1
M
s
=1.2 1.0 5.8
M
s
=1.3 1.4 5.0
M
s
=1.4 1.8 4.5
Chapter 5. A Tuning Method for IPZ Models
116
0 100 200 300 400 500 600 700 800 900 1000
185
190
195
200
P
r
e
s
s
u
r
e
(
k
P
a
)
0 100 200 300 400 500 600 700 800 900 1000
35
40
45
50
55
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Time (s)
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800
185
190
195
200
P
r
e
s
s
u
r
e
(
k
P
a
)
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800
35
40
45
50
55
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Time (s)
Figure 5.14 Experiment on a real paper machine showing a closed loop step response,
using IPZtuning for a PI controller. M
s
= 1.1 (above) and M
s
= 1.2 (below). Controller
parameters ȕ = 1.
Chapter 5. A Tuning Method for IPZ Models
117
0 100 200 300 400 500 600 700 800 900 1000
185
190
195
200
P
r
e
s
s
u
r
e
(
k
P
a
)
0 100 200 300 400 500 600 700 800 900 1000
35
40
45
50
55
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Time (s)
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800
185
190
195
200
P
r
e
s
s
u
r
e
(
k
P
a
)
800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800
35
40
45
50
55
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Time (s)
Figure 5.15 Experiment on a real paper machine showing a closed loop step response,
using IPZtuning for a PI controller. M
s
= 1.3 (above) and M
s
= 1.4 (below). Controller
parameter ȕ = 1.
Chapter 5. A Tuning Method for IPZ Models
118
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 M
s
= 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 IPZprocess. It is a wellknown
fact that for a firstorder 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 IPZprocess the integral error (IE) is calculated
for a few different processes and M
s
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 derivativeterm 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.
M
s
k
v
T
1
T
2
L k
c
T
i
k
c
T
i
T
d
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
Chapter 5. A Tuning Method for IPZ Models
119
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
.
) 15 1 (
50 1
01 . 0 ) (
3s
e
s s
s
s P
÷
+
+
= (5.16)
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 G
ry
. However, this can be adjusted by the feedforward
parameters ȕ and Ȗ of the PID controller, or by a more general set point
0 50 100 150 200 250 300 350 400
0
0.5
1.0
y
0 50 100 150 200 250 300 350 400
2
0
2
4
Time (s)
u
c
Figure 5.16 Comparison between a PI controller (solid) and a PID controller (dotted)
regulating (5.16). Both controllers are tuned for M
s
= 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.
Chapter 5. A Tuning Method for IPZ Models
120
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
2.0 1.5 1.0 0.5 0 0.5
1.0
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1.0
2.0 1.5 1.0 0.5 0 0.5
1.0
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1.0
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 M
s
circle in two points
but only once for the PI controller. This can also be seen in the sensitivity function in
Figure 5.17.
10
4
10
3
10
2
10
1
10
0
0
0.5
1.0
1.5
G
r
y
10
4
10
3
10
2
10
1
10
0
0
0.5
1.0
1.5
G
n
y
Frequency (Hz)
Figure 5.17 Amplitude diagram for the transfer functions from set point r to process
output y, G
ry
, and from noise n to process output y, G
ny
. G
ny
is equal to the sensitivity
function and G
ry
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.
Chapter 5. A Tuning Method for IPZ Models
121
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 IPZtuning, 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
twodegreeoffreedom 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 IPZprocess 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 overshoot 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
crosscouplings 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 IPZprocess. Many different tuning methods have been
presented previously, but they mostly suppose a first or secondorder
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
IPZprocess is not a selfregulating process, it is most natural to use the
latter. However, it is not obvious how to match the IPZprocess to a pure
integrator and these methods do not always give a satisfactory result.
Nevertheless, there a few examples, which are modifications of the
ZieglerNichols 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 IPZtuning. The purpose
is not to discredit these methods but to demonstrate the need for a new
Chapter 5. A Tuning Method for IPZ Models
122
tuning method for the steam pressure process. All examples are based on
the process model
.
) 20 1 (
100 1
05 . 0 ) (
s
e
s s
s
s P
÷
+
+
= (5.17)
The simulations of IPZtuning use M
s
= 1.2 and M
s
= 1.3, while the
obtained M
s
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 IPZtuning. The resulting parameters for the
IPZtuning are
M
s
= 1.2 PI: k
c
= 0.67, T
i
= 8.1,
M
s
= 1.3 PI: k
c
= 0.92, T
i
= 6.4,
(5.18)
M
s
= 1.2 PID: k
c
= 1.0, T
i
= 5.3, T
d
= 0.41,
M
s
= 1.3 PID: k
c
= 1.4, T
i
= 3.9, T
d
= 0.49.
5.7.1 ZieglerNichols two tuning methods
The ZieglerNichols 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 k
c
P(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, k
0
, and the oscillation
period is T
0
. The controller settings are then
Chapter 5. A Tuning Method for IPZ Models
123
0 50 100 150
0
0.5
1
y
0 50 100 150
0
0.5
1
1.5
2
2.5
Time (s)
u
c
Figure 5.20 Evaluation of ZieglerNichols frequency response method for PID control
(solid line) given the process in (5.17). Also shown is the IPZtuning for M
s
= 1.2 (dotted)
and M
s
= 1.3 (dashed).
0 50 100 150
0
0.5
1
y
0 50 100 150
0
0.5
1
1.5
2
2.5
Time (s)
u
c
Figure 5.19 Evaluation of ZieglerNichols frequency response method for PI control (solid
line) given the process in (5.17). Also shown is the IPZtuning for M
s
= 1.2 (dotted) and
M
s
= 1.3 (dashed).
Chapter 5. A Tuning Method for IPZ Models
124
0 50 100 150
0
0.5
1
y
0 50 100 150
0
0.5
1
1.5
2
2.5
Time (s)
u
c
Figure 5.22 Evaluation of ZieglerNichols step response method for PID control (solid
line) given the process in (5.17). Also shown is the IPZtuning for M
s
= 1.2 (dotted) and
M
s
= 1.3 (dashed).
0 50 100 150
0
0.5
1
y
0 50 100 150
0
0.5
1
1.5
2
Time (s)
u
c
Figure 5.21 Evaluation of ZieglerNichols step response method for PI control (solid line)
given the process in (5.17). Also shown is the IPZtuning for M
s
= 1.2 (dotted) and
M
s
= 1.3 (dashed).
Chapter 5. A Tuning Method for IPZ Models
125
PI: ,
2 . 1
, 45 . 0
0
0
T
T k k
i c
= = (5.19)
and
PID: .
8
,
2
, 6 . 0
0 0
0
T
T
T
T k k
d i c
= = = (5.20)
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
) 1 (
1
2
1
÷ =
+
+
÷sL
c
e
sT s
sT
k (5.21)
By letting s = ie, and equating the complex and real parts, the following
controller settings and M
s
values are obtained
PI: k
c
= 2.87, T
i
= 3.28, M
s
= 2.7, (5.22)
and
PID: k
c
= 3.83, T
i
= 1.97, T
d
= 0.49, M
s
= 2.55. (5.23)
As mentioned before, the ZieglerNichols 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 IPZtuning, even though they are not completely comparable.
This is since the simulated IPZtuning has different M
s
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
Chapter 5. A Tuning Method for IPZ Models
126
PI: , 33 . 3 ,
9 . 0
L T
aL
u
k
i
c
c
=
A
= (5.24)
and
PID: , 5 . 0 , 2 ,
2 . 1
L T L T
aL
u
k
d i
c
c
= =
A
= (5.25)
where ǻu
c
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: k
c
= 3.6, T
i
= 3.33, M
s
= 3.8 (5.26)
and
PID: k
c
= 4.8, T
i
= 2, T
d
= 0.5, M
s
= 4.3. (5.27)
The controller settings are somewhat more aggressive than in the
frequency response method. This is also confirmed in the simulations.
a
L
Figure 5.23 The two parameters used in the ZieglerNichols step response method.
Chapter 5. A Tuning Method for IPZ Models
127
5.7.2 TyreusLuyben’s modified ZN tuning rule
The ZieglerNichols 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
PI: , 2 . 2 ,
22 . 3
0
0
T T
k
k
i c
= = (5.28)
where k
0
and T
0
is defined as in Section 5.7.1. This gives the following
values for the process in (5.17)
PI: k
c
= 1.97, T
i
= 8.66, M
s
= 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 M
s
) and the control signal is much
more aggressive. As a comparison, Figure 5.25 shows IPZtuning for
M
s
= 1.65. The disturbance rejection is then better for IPZtuning than
TyreusLuyben’s tuning rule. The time to reach a new set point is similar
for the two methods even though the size of the overshoot is different.
Remember that the overshoot should be dealt with the feedforward part
of the controller, if necessary.
As for the case of ZieglerNichols, the TyreusLuyben method lacks a
tuning parameter.
5.7.3 The AMIGO tuning rule
The AMIGOrule [Åström and Hägglund, 2005] is developed in a similar
way as the IPZrule, with M
s
= 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 firstorder
system or an integrator. Given the system
, ) (
sL
e
s
k
s P
÷
= (5.30)
Chapter 5. A Tuning Method for IPZ Models
128
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1.0
1.5
y
0 20 40 60 80 100 120 140 160 180 200
0
1
2
3
Time (s)
u
c
Figure 5.25 Evaluation of TyreusLuyben’s method for PI control (solid line) given the
process in (5.17). Also shown is the IPZtuning for M
s
= 1.65 (dotted) (k
c
= 1.57,
T
i
= 4.09).
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1.0
y
0 20 40 60 80 100 120 140 160 180 200
0
1
2
3
Time (s)
u
c
Figure 5.24 Evaluation of TyreusLuyben’s method for PI control (solid line) given the
process in (5.17). Also shown is the IPZtuning for M
s
= 1.2 (dotted) and M
s
= 1.3
(dashed).
Chapter 5. A Tuning Method for IPZ Models
129
0 50 100 150
0
0.5
1
y
0 50 100 150
0
0.5
1
1.5
2
2.5
Time (s)
u
c
Figure 5.27 Evaluation of the AMIGO tuning rule for PID control (solid line) given the
process in (5.17). Also shown is the IPZtuning for M
s
= 1.2 (dotted) and M
s
= 1.3
(dashed).
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1
y
0 20 40 60 80 100 120 140 160 180 200
0
1
2
3
Time (s)
u
c
Figure 5.26 Evaluation of the AMIGO tuning rule for PI control (solid line) given the
process in (5.17). Also shown is the IPZtuning for M
s
= 1.2 (dotted) and M
s
= 1.3
(dashed).
Chapter 5. A Tuning Method for IPZ Models
130
the proposed tuning rule is
PI: , 4 . 13 ,
35 . 0
L T
kL
k
i c
= = (5.31)
PID: . 5 . 0 , 8 ,
45 . 0
L T L T
kL
k
d i c
= = = (5.32)
The next step is to match the IPZprocess to a pure integrator with time
delay. By letting k be equal to the maximum slope of a unit step response
of the IPZprocess, (3.5) gives
.
2
1
T
T k
k
v
= (5.33)
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: k
c
= 1.4, T
i
= 13.4, M
s
= 1.4, (5.34)
and
PID: k
c
= 1.8, T
i
= 8, T
d
= 0.5, M
s
= 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
overshoot in the set point response, compared to the IPZtuning 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
s
e
s s
s
s P
3
) 10 1 (
100 1
01 . 0 ) (
÷
+
+
= (5.36)
we get the tuning
PI: k
c
= 1.17, T
i
= 40.2, M
s
= 1.3, (5.37)
Chapter 5. A Tuning Method for IPZ Models
131
and
PID: k
c
= 1.5, T
i
= 24, T
d
= 1.5, M
s
= 1.4. (5.38)
Figure 5.28 shows the simulation for (5.37) compared with one case of
IPZtuning. 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 closedloop
system with a specified characteristic polynomial. In Chapter 3, this was
introduced by statefeedback 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 firstorder process, and a
PID controller handles the same thing for a secondorder 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.
0 50 100 150 200 250 300
0
0.5
1.0
y
0 50 100 150 200 250 300
0
0.5
1.0
1.5
Time (s)
u
c
Figure 5.28 Evaluation of the AMIGO tuning rule for PI control (solid line) given the
process in (5.36). Also shown is the IPZtuning for M
s
= 1.3 (dashed).
Chapter 5. A Tuning Method for IPZ Models
132
NelsonGardner’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 timedelay is disregarded. Let the process be
given by
,
) 1 (
) 1 )( 1 (
) 1 (
) 1 (
) (
2
1
2
1
sT s
sL sT
k e
sT s
sT
k s P
v
sL
v
+
÷ +
~
+
+
=
÷
(5.39)
where the time delay have been approximated by the firstorder Taylor
series. The controller is given by
.
1
) (
i
i
c
sT
sT
k s C
+
= (5.40)
Now, let T
i
= T
2
. This implies that the process pole is canceled by the zero
of the controller. This can be compared with the IPZtuning rule, where
the integral time follows the parameter T
2
in the sense that larger values of
T
2
gives larger values of T
i
, and for small T
2
the two are almost equal, see
Figure 5.5. The closed loop system is now given by
.
) ( ) (
) 1 )( 1 (
) (
1
2
1 2
1
c v c v c v c v
c v
ry
k k s L k k T k k s L T k k T
sL sT k k
s G
+ ÷ + ÷
÷ +
= (5.41)
The two zeros of the denominators are
.
) ( 2
4 )) ( ( ) (
1 2
2
2
1 1
L T k k T
T k k T L k k L T k k
s
c v
c v c v c v
÷
÷ + ± ÷
÷ = (5.42)
By the controller gain k
c
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
0 4 )) ( (
2
2
1
= ÷ + T k k T L k k
c v c v
(5.43)
Chapter 5. A Tuning Method for IPZ Models
133
is fulfilled. This gives us the tuning rule
. ,
) (
4
2
2
1
2
T T
T L k
T
k
i
v
c
=
+
= (5.44)
and the double pole is then positioned in
L T ÷
÷
1
2
(5.45)
Assuming the process given in (5.17), the controller settings are
PI: k
c
= 0.16, T
i
= 20, M
s
= 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 M
s
value is fairly small. Further investigations
show that this method gives a quite robust controller with M
s
values
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1
y
0 20 40 60 80 100 120 140 160 180 200
0
1
2
3
Time (s)
u
c
Figure 5.29 Evaluation of NelsonGardner’s pole placement rule (5.44) for PI control
(solid line) given the process in (5.17). Also shown is the IPZtuning for M
s
= 1.2 (dotted)
and M
s
= 1.3 (dashed).
Chapter 5. A Tuning Method for IPZ Models
134
below 1.05 for short time delays (L § 1). Larger time delays or shorter T
2
give M
s
§ 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 IPZprocess fits into this model structure.
Then the process is given by
,
) 1 (
) 1 (
) (
2
1
sT s
sT
k s P
v
+
+
= (5.47)
and the controller is
.
1
) (
2
i
d i i
c
sT
s T T s T
k s C
+ +
= (5.48)
The closed loop system then becomes
.
) ( ) 1 ( ) (
) 1 )( 1 (
) (
1
2
1
3
2 1
1
2
c v i c v d c v c v i d c v i
d i i c v
ry
k k s T T k k s T k k T k k T s T T T k k T
s T s T T s T k k
s G
+ + + + + + +
+ + +
=
=
(5.49)
A suitable characteristic polynomial for a thirdorder system is
) 2 )( (
2
0 0
2
1
e çe e + + + s s s (5.50)
By equating coefficients of equal power in s, we get the following system
of equations
Chapter 5. A Tuning Method for IPZ Models
135
¦
¦
¦
¦
¹
¦
¦
¦
¦
´
¦
+ =
+
+ +
+ =
+
+
=
+
1 0
2 1
1
2
0 1 0
2 1
1
2
0 1
2 1
2
) 1 (
2
) (
e ç e
e ç e e
e e
i d i c v
d c v c v i
i d i c v
i c v
i d i c v
c v
T T T T T k k
T k k T k k T
T T T T T k k
T T k k
T T T T T k k
k k
(5.51)
The solution is given by
.
) 2 2 (
1 2 2
,
2
,
) 1 2 2 (
) 2 2 (
1 0
2
1 0 1 1 1 1 0 2 1 0 2 1 2 0
0 2 1 2
2
0 2 1 1 0 2 1 1
2
0 2
2
1
1 0
1 0 1 0 1
2
0
2
1
2
1 1 0 0 1 1 1
2
0 1
3
1
1 0
2
1 0 2 1 2 0 1 1 0 2 1 1 1 0
e e e ç e e e e ç e e
ç e e e ç e e e e
e e
e e e ç e
e ç e e ç e e e e
e e e ç e e e e ç e e
T T T T T T T
T T T T T T T T
T
T
T
T T T T T k
T T T T T T T
k
d
i
v
c
+ ÷ ÷ ÷ +
÷ + + ÷ ÷
=
÷ +
=
÷ ÷ ÷ + +
÷ ÷ ÷ + +
=
(5.52)
Even though any e
0
, e
1
, 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
.
4
3
,
4
3
,
2
3
2
1
1
0
= = = ç e e
T T
(5.53)
The solution is then
.
) ( 9
,
3
4
,
) 4 3 (
) ( 9
2 1
2
1
2
2 1
2
1
2 1 2
T T
T
T T T
T T T k
T T T
k
d i
v
c
÷
= =
÷
÷
= (5.54)
Chapter 5. A Tuning Method for IPZ Models
136
If this is a good tuning depends on how well we can neglect the time
delay, but also on the other process parameters. Large T
1
, for example,
tends gives very small k
c
.
Remark
Trying to place any of the closed loop poles in í1/T
1
gives infinite
controller gain. This matches what was observed with PI control.
Controller parameters T
i
and T
d
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 IPZtuning 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. ZieglerNichols 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 casebycase 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.
0 1000 2000 3000
15
16
17
18
19
20
21
Time (s)
S
t
e
a
m
f
l
o
w
(
k
g
/
s
)
0 1000 2000 3000
15
16
17
18
19
20
21
Time (s)
S
t
e
a
m
f
l
o
w
(
k
g
/
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.
Chapter 5. A Tuning Method for IPZ Models
137
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 ZieglerNichols methods distinguish themselves by
grouping together away from the other methods. All four settings from the
IPZtuning are shown in the figure. It was previously concluded that the
AMIGO tuning rule and TyreusLuyben’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 IPZtuning. The
example of NelsonGardner’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
0 1 2 3 4 5 6 7
0
0.5
1.0
1.5
2.0
2.5
k
c
k
i
ZN step
ZN freq
TL
AMIGO
IPZ
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 ZieglerNichols
frequency method, + ZieglerNichols step test method, Ź TyreusLuyben’s modified rule,
ƅ the AMIGO tuning rule, Ƒ NelsonGardner, and Ɣ IPZtuning (M
s
= 1.1í1.4). The
dashed line shows the contour that the IPZtuning follows. For large M
s
values, the tuning
method is moving towards the peak of the stability region, namely the maximum k
i
.
Ź
ƅ
Chapter 5. A Tuning Method for IPZ Models
138
multiplied by a factor of two. Both ZieglerNichols methods become
unstable and TyreusLuyben 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 IPZmodel 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
0 50 100 150 200
2
0
2
x 10
7
ZN frequency
y
0 50 100 150 200
0
1
2
TL
y
0 50 100 150 200
4
2
0
2
x 10
12
ZN step
y
0 50 100 150 200
0
1
2
AMIGO
y
0 50 100 150 200
1
0
1
2
NG
Time (s)
y
0 50 100 150 200
0
1
2
IPZ M
s
= 1.3
Time (s)
y
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.
Chapter 5. A Tuning Method for IPZ Models
139
is essential. The advantage with the IPZtuning 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 2DOFcontroller
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.
Chapter 5. A Tuning Method for IPZ Models
140
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.
Modeling and control
of paper moisture in
the drying section
Part
2
Chapter 6. Enhanced Moisture Control Using the Air System
143
6.
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 midranging structure to
control the moisture.
The proposed control structure is evaluated by simulations of a
physical model of the drying section.
6
Enhanced Moisture Control
Using the Air System
Chapter 6. Enhanced Moisture Control Using the Air System
144
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 nonlinear steadystate relations
are given. [Berrada, et al, 1997] develop a linear statespace 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
blackbox 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 steadystate. 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 10
5
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/m
2
.
Chapter 6. Enhanced Moisture Control Using the Air System
145
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
, k J
t
+ ÷V =
c
u c
(6.1)
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 steadystate
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 Ccode 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
Time Time
I
n
p
u
t
d
a
t
a
O
u
t
p
u
t
d
a
t
a One cylinder
Figure 6.1 Illustration of the simulation method, where one cylinder with a following free
draw is simulated at a time.
Chapter 6. Enhanced Moisture Control Using the Air System
146
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
IPZ PID
Ȉ
pressure setpoint
pressure
1
Figure 6.2 The model used to simulate the closed loop steam pressure dynamics.
Chapter 6. Enhanced Moisture Control Using the Air System
147
10
12
14
16
18
500 550 600 650 700 750 800 850 900
56
58
60
u
Time (s)
G(s) =
e
 1.0 s
0.00178(97.37s+1)
s(7.97s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.31596
Steam group 1
y
86
88
90
92
94
96
y
0 50 100 150 200 250 300
66
68
70
u
Time (s)
G(s) = e
 1.00 s
0.00350(44.69s+1)
s(2.61s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.38311
Steam group 2
50
52
54
56
58
60
62
y
400 450 500 550 600 650 700 750
58
60
62
64
u
Time (s)
G(s) = e
 1.0 s
0.00158(121.49s+1)
s(6.26s+1)
Range y: (100450)
Range u: (0100)
Loss function=0.41249
Steam group 3
158
160
162
164
166
168
y
0 50 100 150 200 250 300 350
44
46
48
50
u
Time (s)
G(s) = e
 1.0 s
0.00345(44.98s+1)
s(3.70s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.21842
Steam group 4
91
92
93
94
95
96
97
98
y
450 500 550 600 650 700 750
40
44
48
u
Time (s)
G(s) = e
 2.0 s
0.00112(55.60s+1)
s(14.01s+1)
Range y: (100450)
Range u: (0100)
Loss function=0.20334
Steam group 5
262
266
270
274
278
282
y
0 100 200 300 400 500
40
45
50
u
Time (s)
G(s) = e
 1.00 s
0.00228(90.10s+1)
s(14.93s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.38462
Steam group 6
190
192
194
196
198
y
700 800 900 1000 1100 1200
60
62
64
u
G(s) = e
 1.0 s
0.00270(40.31s+1)
s(6.86s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.23028
Time (s)
Steam group 7
214
216
218
220
222
224
226
y
100 150 200 250 300 350 400 450
24
26
28
u
Time (s)
G(s) = e
 1.96 s
0.00369(78.64s+1)
s(13.23s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.21070
Steam group 8
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 u
c
is %. The figure shows group 1í8.
Chapter 6. Enhanced Moisture Control Using the Air System
148
Figure 6.5 An example of blow box configuration in both a singletier (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.
234
238
242
246
250
y
50 100 150 200 250 300 350 400 450 500
34
38
42
u
Time (s)
G(s) = e
 2.42 s
0.00324(75.96s+1)
s(21.23s+1)
Range y: (100450)
Range u: (0100)
Loss function=0.16867
Steam group 9
238
242
246
250
254
258
y
550 600 650 700 750 800
34
38
42
u
Time (s)
G(s) = e
 1.89 s
0.00362(74.83s+1)
s(15.73s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.30897
Steam group 10
314
316
318
320
322
324
326
y
0 50 100 150 200 250 300 350 400 450
50
54
58
u
Time (s)
G(s) = e
 1.43 s
0.00139(110.35s+1)
s(24.95s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.22603
Steam group 11
334
336
338
340
342
y
500 550 600 650 700
44
48
52
u
Time (s)
G(s)= e
 1.70 s
0.00138(108.94s+1)
s(24.26s+1)
Range y: (100  450)
Range u: (0  100)
Loss function = 0.16917
Steam group 12
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 u
c
is %. The figure shows group 9í12.
Chapter 6. Enhanced Moisture Control Using the Air System
149
in the dynamics, see Figure 6.6, and the volume was set to 10 m
3
. With a
nominal total supply air flow of 50 m
3
/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/m
2
. 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 m
3
/s supply air and 70.3 m
3
/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
Evaporation
Supply
Air
Exhaust Air
Cylinder
Paper
Wet Air
Dry Air
Actuator
System
boundary
Figure 6.6 A picture of the air system used in the model.
Chapter 6. Enhanced Moisture Control Using the Air System
150
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 Midranging
As described previously, the steam and air system is combined by mid
ranging control. Generally, midranging refers to control problems where
there is one process output and two or more manipulated inputs, see
7
8
9
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
50
55
60
65
D
e
w
p
o
i
n
t
e
x
h
a
u
s
t
a
i
r
(
o
C
)
0 200 400 600 800 1000 1200
0
50
100
Time (s)
C
o
n
t
r
o
l
s
i
g
n
a
l
s
u
p
p
l
y
a
i
r
(
%
)
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 reelup. The dew point of the exhaust air is also
shown.
Chapter 6. Enhanced Moisture Control Using the Air System
151
[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 midranging. 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 midranging structures
There exist several different alternatives to implement a midranging
controller, and four different structures are described below. It is assumed
that P
1
is a fast process with a small range, and P
2
is a slow process with a
large range. The signals u
1
and u
2
are control signals for P
1
and P
2
,
respectively, and r
1
is set point for y and r
2
set point for u
1
. The last three
midranging structures are evaluated in [Allison and Isaksson, 1998]. The
Pulp tower
W
a
t
e
r
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.
Chapter 6. Enhanced Moisture Control Using the Air System
152
conclusion is that the MPC solution is superior compared to the other two,
and the valve position controller comes in second hand.
Simple Midranging
This is probably the simplest way to implement midranging 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 u
1
is a
P controller, whereas the controller manipulating u
2
is a controller with
integral action. When the integral action in C removes the control error,
the output u
1
is the offset of the P controller. By adjusting the offset, the
target value for u
1
is set. The disadvantage with this solution is that P
1
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 steadystate error which is removed by C through a much slower
process P
2
.
Valve Position Controller
The valve position controller in Figure 6.10 is probably the most common
implementation of midranging found in industry today, see [Shinskey,
1978] and [Allison and Ogawa, 2003]. Controller C
1
controls the output y
with input u
1
, while C
2
controls u
1
with u
2
. In the literature describing the
valve position controller, the decoupling filter C
D
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 C
2
to control u
1
is through the error r
1
íy,
consequently introducing disturbances into y. The decoupling should
ideally be chosen as
C(s)
k
c
Ȉ Ȉ
r
1
u
1
u
2
y
1
P
1
(s)
P
2
(s)
Ȉ
r
2
Figure 6.9 The structure of the simple midranging controller. The set point r
2
is the bias of
the Pcontroller. In practice, C(s) is often chosen as a PI controller
Chapter 6. Enhanced Moisture Control Using the Air System
153
,
) (
) (
) (
1
2
s P
s P
s C
D
÷ = (6.2)
which is realizable as long as the time delay of P
1
is shorter or equal to the
time delay of P
2
, and P
1
and P
2
have the same unstable zeros. In addition,
the pole excess of P
2
should be at least as large as the pole excess of P
1
. It
is assumed that both P
1
and P
2
have stable poles or integrators. With (6.2),
changes in u
2
do not affect y, and the transfer function from u
2
to r
2
í u
1
is
equal to íC
D
. If the ideal decoupling is not realizable, approximations
have to be used.
This implementation does not require antiwindup protection in C
1
since u
1
is a controlled variable. Regardless if u
1
saturates or not during an
upset, controller C
2
brings u
1
back to r
2
in steadystate. However,
depending on the speed of controller C
2
, antiwindup protection in C
1
might be beneficial, see [Haugwitz, et al, 2005].
Hybrid Midranging
An approach, originally used to control pressure of a steam header, is the
hybrid midranging [Love, 1994]. It consists of one PI controller, two
saturations, a low pass filter, and three gains, see Figure 6.11. The
parameter R
m
, should be chosen as
,
100
2 1
1
k k
k
R
m
+
= (6.3)
where k
1
and k
2
are the steadystate gains of P
1
and P
2
, respectively. The
filter in combination with the direct term creates the midranging function.
The transfer function from the output of the PI controller to the
P
2
(s)
P
1
(s)
C
2
(s)
C
1
(s)
C
D
(s)
Ȉ
Ȉ Ȉ Ȉ
r
1
r
2
u
1
u
2
y
1
1
Figure 6.10 The structure of the valve position controller.
Chapter 6. Enhanced Moisture Control Using the Air System
154
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
u
1
when u
2
is reaching its limit. By this, the controller combines mid
ranging (normal operation) with splitrange (saturated u
2
) in a nice way.
The purpose of the other saturation block is to immediately engage control
signal u
2
when u
1
reaches its limit.
The PI controller can, without losing the midranging function, be
replaced by any SISO controller.
Midrange MPC
In [Allison and Isaksson, 1998] a midranging controller is implemented
in the MPC (model predictive control) approach, see Figure 6.12. By
letting the minimized cost function be given by
,
) (
) (
) ( ) (
)  ( ˆ ) (
) (
1
0
2
2
1
1
0
2
1 2
1
_ _
÷
=
÷
=
(
¸
(
¸
+ A
+ A
+
(
¸
(
¸
+ ÷ +
+ ÷ +
=
u
p
H
i
R
H
i
Q
i k u
i k u
i k u i k r
k i k y i k r
k J (6.4)
and the weighting matrices chosen as
,
large 0
0 small
,
small 0
0 large
(
¸
(
¸
=
(
¸
(
¸
= R Q (6.5)
PI Filter Ȉ
1
Ȉ
Ȉ
m
R
100
100
m
R
m
R ÷ 100
100
Ȉ
1
Ȉ
1
u
2
u
1
r
1
r
2
y
P
1
(s)
P
2
(s)
Figure 6.11 The structure of the hybrid midranging controller.
Chapter 6. Enhanced Moisture Control Using the Air System
155
the desired behavior is achieved. Since the MPC formulation is inherently
discretetime, the cost function is given as a summation. Notation
ǔ(k + i  k) denotes the i stepahead prediction. H
p
is the prediction
horizon, H
u
is the control horizon, and ǻ is the difference operator
(ǻ = 1 í z
í1
, where z
í1
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 H
p
, the control horizon H
u
, and the
weights Q and R. It is possible to omit the weight on the control
increments, R, in (6.4), without losing the midranging 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 midranging the air system
A midranging 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 midranging 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 u
1
manipulates the air
flow actuator while u
2
is a set point for the steam pressure. The motivation
for not manipulating the steam valves directly is both to preserve the
Ȉ
y
MPC
r
1
r
2
u
1
u
2
P
2
(s)
P
1
(s)
Figure 6.12 The structure of the midrange MPC.
Chapter 6. Enhanced Moisture Control Using the Air System
156
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 C
1
and C
2
are based on the IMC concept, which
is well known as an effective deadtime compensator for a stable process
with long timedelays [Morari and Zafiriou, 1989]. The transfer function
of an IMC can be written as
,
) ( ) ( ) ( 1
) ( ) (
) (
s P s P s H
s P s H
s C
f
f
+
+
÷
= (6.6)
where P is the process model, P
+
is the realizable inverse of the process
model, and H
f
is a low pass filter with steadystate gain one that
determines the closed loop performance.
The process transfer functions are obtained from openloop step
responses, see Figure 6.14 and Figure 6.15, and are modeled as
,
1 66 . 35 3 . 617
04887 . 0 2548 . 0
) (
10
2
1
s
e
s s
s
s P
÷
+ +
+
÷ = (6.7)
and
.
1 5 . 61
1365 . 0
) (
40
2
s
e
s
s P
÷
+
÷ = (6.8)
’moisture’
’set point steam pressure’
’air flow actuator’
’set point
moisture’
’set point
air flow actuator’
C
2
(s)
C
1
(s)
C
D
(s)
Ȉ
Ȉ Ȉ Ȉ
r
1
r
2
u
1
u
2
y
1
1
P
2
(s)
P
1
(s) Ȉ
n
’moisture
disturbance’
Figure 6.13 The midranging structure used in the simulations where the different
variables are also given. Note that u
1
manipulates the air flow actuator while u
2
is the set
point to the steam pressure controllers.
Chapter 6. Enhanced Moisture Control Using the Air System
157
6
8
10
12
14
M
o
i
s
t
u
r
e
(
%
)
50 0 50 100 150 200 250 300 350 400 450 500
400
420
440
460
P
r
e
s
s
u
r
e
s
e
t
p
o
i
n
t
(
k
P
a
)
Figure 6.15 Step response of the steam pressure í moisture process (solid) and the
obtained model (dotted), also given in (6.8).
7.5
8.0
8.5
9.0
9.5
M
o
i
s
t
u
r
e
(
%
)
50 0 50 100 150 200 250 300 350 400
20
30
40
50
60
Time (s)
A
i
r
f
l
o
w
a
c
t
u
a
t
o
r
(
%
)
Figure 6.14 Step response of the air flow í moisture process (solid) and the obtained
model (dotted), also given in (6.7). The dashdotted line indicates the final value.
Chapter 6. Enhanced Moisture Control Using the Air System
158
The model from air flow actuator to moisture, P
1
, 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 airpaper 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 openloop simulation in Figure 6.14. The moisture in Figure 6.7
reaches a new steadystate 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 overshoot in the response because of disturbances.
P
1
has both faster dynamics and shorter time delay compared to P
2
.
The design parameter for the IMC is the filter H
f
. For C
1
it is chosen as
,
1
1
) (
2
1
1

.

\

+
=
s
s H
f
e
t
(6.9)
where Ĳ
1
= 1 (nominal value) implies that the closedloop poles are real
and placed at the same distance from the origin as the openloop poles, see
Figure 6.16. A value of Ĳ
1
greater than one makes the closedloop system
slower, and vice versa. Apart from the two poles, P
1
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 C
1
realizable in both cases.
Ideally, the only term affecting C
2
is C
D
and the controller is tuned for
that process. This can be seen by inserting C
2
into the structure in
Figure 6.13, assuming perfect process models. The purpose of C
2
is to
slowly restore the signal u
1
to its desired value. In practice, it can not be
assumed to have an ideal decoupling filter, and therefore it is important
that C
1
and C
2
do not interfere with each other. The closedloop time
constant of C
2
is therefore related to the fast loop and the IMC filter is
chosen as
Chapter 6. Enhanced Moisture Control Using the Air System
159
.
1
1
) (
2
2
s
s H
f
e
t
+
= (6.10)
The value of t
2
is set to be significantly larger (at least five to ten times)
thant
1
, to separate the controllers C
1
and C
2
in frequency.
The midranging 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
,
5 . 61 1
1
) (
3
3
s
s H
f
t +
= (6.11)
where t
3
= 1.5. To obtain an adequate comparison between the two
moisture control systems, the midranging controller is tuned to have the
same maximum value of the sensitivity function, M
s
, 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 M
s
is
chosen to 1.3 and this gives t
1
= 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
Ȧ
Figure 6.16 The openloop poles (+) and the closedloop double pole (
*
) of the air flow í
sheet moisture process. The parameter e is the distance between the open loop poles and
the origin.
Chapter 6. Enhanced Moisture Control Using the Air System
160
that ideally, both the sensitivity and complementary sensitivity functions
are independent of controller C
2
and consequently also t
2
, because of the
decoupling filter C
D
. The bandwidth of the closed loop system with the
midranging 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 midranging 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 midranging is therefore not a problem for that
specific example. However, the noise distribution should be regarded
before implementing the midranging controller on a drying section.
10
4
10
3
10
2
10
1
0
0.2
0.4
0.6
0.8
1
C
o
m
p
l
.
s
e
n
s
.
f
u
n
c
t
i
o
n
10
4
10
3
10
2
10
1
0
0.5
1
1.5
Frequency (Hz)
S
e
n
s
i
t
i
v
i
t
y
f
u
n
c
t
i
o
n
Figure 6.17 Frequency plot for the midranging 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.
Chapter 6. Enhanced Moisture Control Using the Air System
161
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 midranging
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 midranging has a significantly better disturbance
rejection than the steam pressure control. There is a slight fluctuation in
sheet moisture, for the midranging case, after t = 300 s. This is because of
imperfect models in the decoupling filter, C
D
, and it becomes more
prominent when pushing the performance level for the controller
(t
1
= 0.7). When u
2
midranges u
1
by increasing the steam pressure, C
D
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 twopole model, see also Figure 6.15.
Chapter 6. Enhanced Moisture Control Using the Air System
162
0 100 200 300 400 500 600
7.5
8.0
8.5
9.0
9.5
10.0
10.5
Time (s)
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
Midranging, t
1
= 1.2, t
2
= 5.0
Midranging, t
1
= 1.2, t
2
= 10.0
Steam pressure control
Figure 6.19 Moisture in paper for two different t
2
.
0 100 200 300 400 500 600
7.5
8.0
8.5
9.0
9.5
10.0
10.5
Time (min)
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
Midranging, t
1
= 0.7, t
2
= 5.0
Midranging, t
1
= 1.2, t
2
= 5.0
Steam pressure control
Figure 6.18 Comparison between the midranging control system (t
2
= 5.0) and steam
pressure control. The steam pressure control should be compared to midranging with
t
1
= 1.2.
Chapter 6. Enhanced Moisture Control Using the Air System
163
Figure 6.19 shows that the response in sheet moisture is practically
independent oft
2
. However, there is a slight difference at the second half
of the simulation. Larger values of t
2
reduce the fluctuation since the
steam system is less aggressive in its attempt to ‘midrange’ 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 t
2
. However,
the larger t
2
is, the more the control system becomes ‘singleloop’ and the
advantages of midranging 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
midranging control as by steam pressure control. In the midranging case,
the fast moisture transients are handled by the air system and the steam
system is only used to restore the air system in steadystate. 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
Chapter 6. Enhanced Moisture Control Using the Air System
164
0 100 200 300 400 500 600
435
440
445
450
455
460
465
Time (s)
S
t
e
a
m
P
r
e
s
s
u
r
e
s
e
t
p
o
i
n
t
(
k
P
a
)
Midranging, t
1
= 1.2, t
2
= 5.0
Midranging, t
1
= 1.2, t
2
= 10.0
Steam pressure control
Figure 6.21 Steam pressure in the last steam group (lead group) when comparison steam
pressure control and midranging.
0 100 200 300 400 500 600
58
59
60
61
62
63
Time (s)
D
e
w
p
o
i
n
t
(
°
C
)
t
1
= 1.2, t
2
= 5.0
t
1
= 1.2, t
2
= 10.0
Figure 6.20 Change in dew point (midranging).
Chapter 6. Enhanced Moisture Control Using the Air System
165
(different pulp layers are separated) problems in board machines. Since
the midranging 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 C
1
then gives
a set point to the dew point controller, which is included in P
1
, 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 Zerolevel and dew point control
There are two important variables to control in the air inside the dryer
hood, the dew point and zerolevel, 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 zerolevel, [Karlsson, 2000].
Chapter 6. Enhanced Moisture Control Using the Air System
166
air is used to control the zerolevel, 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 zerolevel. 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 midranging
structure, is shown in Figure 6.23. By adding an extra controller, C
3
, that
controls the dew point in the exhaust air through a selector that is shared
with set point r
2
, the air flow actuator set point is increased if the dew
point exceeds set point r
3
.
6.6 Summary
A new control strategy has been developed for the multicylinder 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.
C
2
(s)
C
1
(s)
C
D
(s)
Ȉ
Ȉ Ȉ Ȉ
r
1
r
2
u
1
u
2
y
1
1
>
C
3
(s) Ȉ
1
r
3
P
2
(s)
P
2
(s)
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 C
3
to avoid windup.
Chapter 6. Enhanced Moisture Control Using the Air System
167
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 midranging 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 midranging structure,
t
1
and t
2
. The performance of the closed loop system is determined by t
1
and t
2
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 t
2
5t
1
. Letting
t
1
= 1, means that the closed loop poles have the same distance to the
origin as the open loop poles.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
168
7.
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 midranging. The midranging 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 highcapacity 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.
7
Feedforward from a Paper
Surface Temperature
Measurement
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
169
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
0 10 20 30 40 50 60 70 80 90 100
0
10
20
30
40
50
60
70
Cylinder no.
M
o
i
s
t
u
r
e
c
o
n
t
e
n
t
(
%
)
Inlet moisture content: 61.39 %
Inlet moisture content: 62.12 %
Node 2
Node 1
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 %.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
170
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 steadystate 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 IRcamera which gives the surface temperature
in a twodimensional window, temperature values along a line in the
0 20 40 60 80 100
50
60
70
80
90
100
110
120
130
Cylinder no.
P
a
p
e
r
t
e
m
p
e
r
a
t
u
r
e
(
°
C
)
Simulated
Measured
15 20 25 30 35 40 45 50
75
80
85
90
95
100
105
110
115
Cylinder no.
Simulated
Measured
Figure 7.2 Comparison between measured and simulated paper surface temperature. The
most important section for the feedforward structure in enlarged and the 100°Clevel is
indicated.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
171
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.
p
1
Different surface
temperatures.
Cylinder shell
Condensate
Steam
p
1
(a)
(b)
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.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
172
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
Suggested measuring
position
(a) (b)
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.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
173
5 10 15 20 25 30 35 40 45
1
0
1
2
3
4
5
Cylinder no.
F
r
e
e
d
r
a
w
t
e
m
p
e
r
a
t
u
r
e
d
i
f
f
e
r
e
n
c
e
(
°
C
)
w
in
= 62.12 %
w
in
= 61.39 %
Changed
peak position
Figure 7.6 A temperature curve calculated by taking the difference between two adjacent
temperature measurements, as indicated in Figure 7.4.
5 10 15 20 25 30 35 40 45 50
50
60
70
80
90
100
110
120
130
140
150
P
a
p
e
r
s
u
r
f
a
c
e
t
e
m
p
e
r
a
t
u
r
e
(
°
C
)
5 10 15 20 25 30 35 40 45 50
100
150
200
250
300
Cylinder no.
S
t
e
a
m
p
r
e
s
s
u
r
e
(
k
P
a
)
Group 2
Group 3
Group 4
Sudden increase
in temperature
Constant
temperature
zone
Constant
temperature
zone
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.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
174
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 driedout 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 heatup
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.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
175
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, H
ff
is a feedforward block and the
definition of the other parts are given in Chapter 6, see Figure 6.13. Since
0 50 100 150 200 250 300
7
8
9
10
11
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
0 50 100 150 200 250 300
25.0
25.5
26.0
26.5
Time (s)
P
e
a
k
p
o
s
i
t
i
o
n
(
c
y
l
.
)
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).
P
2
(z)
P
1
(z)
C
2
(z)
C
1
(z)
C
D
(z)
Ȉ
Ȉ Ȉ Ȉ
r
1
r
2
u
1
u
2
y
1
1
Ȉ
H(z)
Ȉ
”peak position”
H
ff
(z)
Figure 7.7 A simplified block diagram of the control structure used in the simulations.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
176
the simulation model is very large (~10
5
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
.
599 . 0 528 . 1 1
185 . 0
) (
2 1
6
÷ ÷
÷
+ ÷
=
z z
z
z H (7.1)
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
P
1,3
Ȉ
u
1
y
Ȉ
H
Ȉ
”Temperature
peak position”
H
ff,3
P
1,4
P
1,5
Ȉ
Ȉ
H
ff,4
H
ff,5
Ȉ
Figure 7.9 A block diagram of the feedforward structure used in the simulations.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
177
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
P
1
. Since the impact of the disturbance at y is to be eliminated, the
condition
, 0
5 , 5 , 1 4 , 4 , 1 3 , 3 , 1
= + + +
ff ff ff
H P H P H P H (7.2)
holds. For simplicity the feedforward filters were selected as
.
3
,
3
,
3
5 , 1
5 ,
4 , 1
4 ,
3 , 1
3 ,
P
H
H
P
H
H
P
H
H
ff ff ff
÷ = ÷ = ÷ = (7.3)
The control signals of the last three air groups now might be different and
therefore C
2
was set to control the actuator signal of the first two groups
(u
1
). Another possible solution is to let C
2
control the mean value of all
8.0
8.5
9.0
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
0 50 100 150 200 250 300 350 400
20
30
40
50
Time (s)
A
i
r
f
l
o
w
a
c
t
u
a
t
o
r
(
%
)
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.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
178
five air flow actuators or to introduce multivariable control in the steam
system so that each air group is midranged 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
.
639 . 0 537 . 1 1
00214 . 0 00537 . 0
) (
,
590 . 0 490 . 1 1
00224 . 0 00476 . 0
) (
,
685 . 0 609 . 1 1
00198 . 0 00411 . 0
) (
1
2 1
1
5 , 1
4
2 1
1
4 , 1
6
2 1
1
3 , 1
÷
÷ ÷
÷
÷
÷ ÷
÷
÷
÷ ÷
÷
+ ÷
+ ÷
=
+ ÷
+ ÷
=
+ ÷
+ ÷
=
z
z z
z
z P
z
z z
z
z P
z
z z
z
z P
(7.4)
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 midranging
control system in Chapter 6. The sheet moisture is shown in Figure 7.11.
0 100 200 300 400 500 600
7.5
8.0
8.5
9.0
9.5
10.0
10.5
11.0
Time (s)
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
No control  open loop
Steam pressure control
Midranging only
Feedforward + midranging
Figure 7.11 Comparison of different control systems.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
179
0 100 200 300 400 500 600
58.5
59
59.5
60
60.5
61
61.5
62
62.5
63
Time (s)
D
e
w
p
o
i
n
t
(
o
C
)
Midranging only
Feedforward + midranging
Figure 7.13 Change in dew point.
25.0
25.5
26.0
26.5
P
e
a
k
p
o
s
i
t
i
o
n
(
c
y
l
.
)
0 100 200 300 400 500 600
7.5
8.0
8.5
7.5
8.0
8.5
Time (s)
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
Figure 7.12 Peak position and sheet moisture during the closed loop simulation.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
180
10
3
10
2
10
1
10
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Frequency (rad/s)
A
m
p
l
i
t
u
d
e
Steam pressure control
Midranging only
Feedforward + midranging
Figure 7.15 Frequency plot, showing how the disturbances in inlet moisture propagate, for
the different control systems.
0 100 200 300 400 500 600
435
440
445
450
455
460
465
Time (s)
S
t
e
a
m
p
r
e
s
s
u
r
e
(
k
P
a
)
Steam pressure control
Midranging only
Feedforward + midranging
Figure 7.14 Comparison of the steam pressure in the lead group by the different control
system.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
181
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 u
1
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 P
1,3
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 midranging 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 reelup 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 10
í3
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 midranging
control system. It might as well be combined with any other moisture
control structure. The midranging controller is chosen here to give
natural connection to Chapter 6.
Chapter 7. Feedforward from a Paper Surface Temperature Measurement
182
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 reelup.
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.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
183
8.
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 10
5
states which make simulations fairly tedious.
It also lacks a physical model for the steam system, where it instead uses a
blackbox model. Therefore a less complex model is developed that is
based purely on physical relations. The objective is control of the moisture
at the reelup and not to accurately describe microscale 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 objectoriented modeling language
Modelica, see [Mattsson and Elmqvist, 1997], [Mattsson, et al, 1998], and
[Fritzson, 2004]. Like any objectoriented 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 noncausal
equation structure (ii) it is possible to create model components that
8
ObjectOriented Modeling
and Predictive Control of the
Moisture Content
Chapter 8. ObjectOriented Modeling and Control of the Moisture
184
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 midranging 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 q
s
be the mass flow rate of steam into the cylinder, q
c
be the
condensation rate, q
bt
the blow through steam, and q
w
be the siphon flow
rate. Also, let V
s
and V
w
be the volume of steam and water in the cylinder,
and let µ
s
and µ
w
be the densities. The mass balances for water and steam
are then
( )
( ) .
,
w c w w
bt c s s s
q q V
dt
d
q q q V
dt
d
÷ =
÷ ÷ =
µ
µ
(8.1)
Chapter 8. ObjectOriented Modeling and Control of the Moisture
185
The energy balances for steam, water, and metal are
( )
( )
( ) ,
,
, ) (
, p m m m p
m w w s c w w w
s c s bt s s s s
Q Q T mC
dt
d
Q h q h q V u
dt
d
h q h q q V u
dt
d
÷ =
÷ ÷ =
÷ ÷ =
µ
µ
(8.2)
where Q
m
is the power supplied from the water to the metal, Q
p
is the
power supplied from the metal to the paper, h
s
is the steam enthalpy, h
w
is
the water enthalpy, m the mass of the cylinder shell, C
p,m
the specific heat
capacity of the shell, T
m
the mean temperature of the metal, u
s
and u
w
are
the specific internal energies of steam and water. The steam and water
volumes add up to the total cylinder volume,
.
w s
V V V + = (8.3)
The power flow to the metal is given by
( ),
m s cyl sc m
T T A Q ÷ =o (8.4)
where o
sc
is the heat transfer coefficient from the steamcondensate
interface to the centre of the cylinder shell, A
cyl
is the inner cylinder area,
and T
s
the steam temperature. The power flow to the paper is
( ),
p m cyl cp p
T T A Q ÷ = q o (8.5)
where T
p
is the paper temperature, o
cp
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, o
sc
is used to calibrate the model
against plant data. Here, both o
cp
and o
sc
are possible candidates for that
purpose. Experiments have shown that o
sc
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 o
sc
is used as a free variable
to calibrate the model with. Empirical models for o
cp
have been
Chapter 8. ObjectOriented Modeling and Control of the Moisture
186
developed. From [Wilhelmsson, 1995] a linear relation with moisture ratio
u is given
, 955 ) 0 ( ) ( u u
cp cp
+ =o o (8.6)
is obtained, where o
cp
(0) varies between 200 and 500 W/(m
2
K). It is well
known that o
cp
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
, ) ( ) (
s sh v v v s
p p x f C q µ ÷ = (8.7)
where C
v
(m
2
) is the valve conductance, x
v
is the position of the valve
stem and the function f
v
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 p
sh
. We use equal
0 2 4 6 8 10 12 14 16
0
500
1000
1500
2000
2500
3000
Condensate thickness (mm)
H
e
a
t
t
r
a
n
s
f
e
r
c
o
e
f
f
i
c
i
e
n
t
(
W
/
m
2
K
)
15 spoiler bars
25 spoiler bars
30 spoiler bars
No spoiler bars
Figure 8.1 Steadystate 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.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
187
percentage trim, since it is the most common characteristic in the process
industry [Hägglund, 1991], see Figure 8.2. It is given by
. ) (
1 ÷
=
v
x
v v v
R x f (8.8)
R
v
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
. 1141 ln 43 . 52 ) (ln 792 . 6 ) (ln 3136 . 0
, 10 ] 26 . 64 005048 . 0 [
, 10 ] 5 . 748 ln 200 ) (ln 77 . 18 ) (ln 8842 . 0 [
, 10 ] 1824 ln 260
) (ln 58 . 39 ) (ln 887 . 2 ) (ln 07402 . 0 [
, 5 . 124 ln 71 . 37 ) (ln 388 . 3 ) (ln 1723 . 0
2 3
3
3 2 3
3
2 3 4
2 3
+ ÷ + ÷ =
+ =
÷ + ÷ =
+ +
÷ + ÷ =
+ + ÷ =
÷
p p p
p
p p p h
p
p p p h
p p p T
w
s
w
s
s
µ
µ
(8.9)
Equations (8.1) í (8.9) are a crude nonlinear model for the steamcylinder
process. By choosing p, V
w
, and T
m
as state variables and using partial
0 0.2 0.4 0.6 0.8 1.0
0
0.2
0.4
0.6
0.8
1.0
x
v
f
v
(
x
v
)
R
v
= 200
R
v
= 50
R
v
= 10
Figure 8.2 Equal percentage valve characteristic.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
188
derivatives, the system can be rewritten into a third order state equation
(most steps are omitted here), where f
1
, f
2
, and f
3
are defined as the right
hand side of the equations.
, ) ( ) 955 ) 0 ( ( ) (
, ) (
) ( ) (
, ) ( ) (
3 33
2
22 21
1 12 11
f T T A u T T A
dt
dT
e
f T T A h q h q
h p p x f C
dt
dV
e
dt
dp
e
f q q p p x f C
dt
dV
e
dt
dp
e
p m cyl cp m s cyl sc
m
m s cyl sc w w s bt
s s sh v v v
w
bt w s sh v v v
w
= ÷ + ÷ ÷ =
= ÷ ÷ ÷ ÷
÷ = +
= ÷ ÷ ÷ = +
q o o
o
µ
µ
(8.10)
where
.
,
, ) ( ) (
,
, ) (
33
22
21
12
11
p
s s w w
w
w w
w
w w
s
w s
s
w s
s w
w
w
s
w
mC e
h h e
V
dp
dh
V
dp
d
V h
dp
dh
V V
dp
d
V V h e
e
dp
d
V
dp
d
V V e
=
÷ =
÷ + + ÷ + ÷ =
÷ =
+ ÷ =
µ µ
µ
µ
µ
µ
µ µ
µ µ
(8.11)
In the rewritings of the energy balances above the specific internal energy
has been eliminated by the definitions u
s
= h
s
÷ p/ȡ
s
and u
w
= h
w
÷ p/ȡ
w
.
Using f
1
, f
2
, and f
3
the system can be further rewritten into an explicit state
form.
.
,
) (
,
) (
33
3
22 11 12 21 33
1 22 33 2 33 12
22 11 12 21 33
2 33 11 1 21 33
e
f
dt
dT
e e e e e
f e e f e e
dt
dp
e e e e e
f e e f e e
dt
dV
w
=
÷
÷
=
÷
÷
=
(8.12)
Chapter 8. ObjectOriented Modeling and Control of the Moisture
189
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.
Figure 8.3 A Simulink model of (8.12). The submodel f
2
is also opened, which in turn
contains submodels.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
190
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
, ln
,
,


.

\

÷
÷
=
p v tot
a v tot
p g
w tot
evap
p p
p p
T R
KM p
q (8.13)
where q
evap
is the evaporation rate (kg/m
2
s), K is the mass transfer
coefficient (m/s), M
w
is the molecular weight of water (kg/mole), p
tot
the
total pressure (Pa), p
v,a
the partial pressure for water vapor in the air (Pa),
p
v,p
the partial pressure for the water vapor at the paper surface, R
g
the gas
constant (J/mole·K), and T
p
the paper temperature (K). The partial
pressure p
v,a
is given by the moisture content of air, x (kg water vapor/kg
dry air), and the total pressure,
.
62 . 0
, tot a v
p
x
x
p
+
= (8.14)
The vapor partial pressure at the paper surface is
,
0 , v p v
p p ¢ = (8.15)
where p
v0
is the partial vapor pressure for free water. This is given by
Antoine’s equation
Chapter 8. ObjectOriented Modeling and Control of the Moisture
191
. 10
15 . 43
1690
127 . 10
0


.

\

÷
÷
=
p
T
v
p (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
), ) 273 ( 10085 . 0 58 . 47 exp( 1
0585 . 1 877 . 1
u T u
p
÷ ÷ ÷ ÷ = ¢ (8.17)
where u is the moisture ratio (kg moisture/kg fiber). Also, let v
x
be the
speed of the paper web (m/s), d
y
the width of the paper web (m), A
xy
the
area of the dryer surface covered by paper (m
2
), and g the dry basis weight
(kg/m
2
). Then the mass balance of moisture in the paper web can be
written as
.
) (
gu v d q A gu v d
dt
ugA d
x y evap xy in x y
xy
÷ ÷ = (8.18)
Figure 8.6 shows a schematic picture of the mass flows in the model. To
model the energy balance, introduce
0 5 10 15 20 25
0
0.5
1.0
Moisture content (%)
S
o
r
p
t
i
o
n
i
s
o
t
h
e
r
m
0 5 10 15 20 25
0
500
1000
1500
Moisture content (%)
H
e
a
t
o
f
s
o
r
p
t
i
o
n
(
k
J
/
k
g
)
30
o
C
60
o
C
90
o
C
30
o
C
60
o
C
90
o
C
Figure 8.5 Sorption isotherm, ĳ, and heat of sorption, ǻH
s
given by (8.17) and (8.21).
Chapter 8. ObjectOriented Modeling and Control of the Moisture
192
,
1
, ,
,
u
uC C
C
w p fiber p
p p
+
+
= (8.19)
where C
p,p
, C
p,fiber
, and C
p,w
is the specific heat capacity for the paper, fiber
and water, respectively (J/kg·K). As we can see, C
p,p
is a weighted sum of
the heat capacities of the parts. From [Wilhelmsson, 1995] we have C
p,fiber
= 1256 J/(kg·K).
Also, let T
p
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 ǻH
s
(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 ClausiusClapeyron
,
) / 1 (
) (ln
(
(
¸
(
¸
÷ = A
p w
g
s
T d
d
M
R
H
¢
(8.20)
and by applying this on (8.17), we get
.
1
10085 . 0
2 0585 . 1
¢
¢
w
g p s
M
R T u H
÷
= A (8.21)
The amount of energy required to evaporate the water from the surface of
the web is the given by
,
s vap
H H H A + A = A (8.22)
evap xy
q A
in x y
gu v d
Paper web
gu v d
x y
x
v
Figure 8.6 The mass balance for moisture in the paper web. The shaded area is the
cylinder wall
Chapter 8. ObjectOriented Modeling and Control of the Moisture
193
where ΔH
vap
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
, ) 1 ( ) (
) 1 (
) ) 1 ( (
,
, ,
,
p p p x y s vap evap xy
in p p p in x y p
p p p xy
T C u g v d H H q A
T C u g v d Q
dt
T C A u g d
+ − Δ + Δ −
+ + =
+
(8.23)
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,
) (
s vap evap xy
H H q A Δ + Δ
in p p p in x y
T C u g v d
, ,
) 1 ( +
p p p x y
T C u g v d
,
) 1 ( +
p
Q
Paper web
Figure 8.7 The energy balance of the paper web. The shaded area is the cylinder wall.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
194
where the outflow of paper of one component becomes the inflow of the
next component.
8.2 Steadystate 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 steadystate 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/m
2
. 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 Į
sc
is used as fitting parameter with the same value in all
cylinders and 1100 W/(m
2
·K) is found to give the best fit by visual
inspection. The agreement between model and measurements is good even
5 10 15 20 25 30 35
0
10
20
30
40
50
60
Cylinder number
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
Figure 8.8 Validation by comparing steadystate simulations with moisture
measurements on a fine paper machine.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
195
though this is rough validation without using all possible degrees of
freedom, e.g. by assuming different Į
sc
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.
Pope
Moisture controller
Press
HeaderPres...
k={1000e3}
P
i P
I
D
F
PID
Header pressure
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.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
196
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/m
2
. 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 steamcondensate interface
is much larger than the heat transfer coefficient of the cylinderpaper
interface, o
sc
>> o
cp
, 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 o
cp
. This implies that in this analysis o
cp
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
450
500
550
S
t
e
a
m
p
r
e
s
s
u
r
e
(
k
P
a
)
130
140
150
S
t
e
a
m
p
r
e
s
s
u
r
e
(
k
P
a
)
100 150 200
55
60
65
70
75
Time (s)
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
o
cp
= 500
o
cp
= 4000
Figure 8.10 Pressure response for a step in valve position for two different values of o
cp
.
Heat transfer coefficient o
sc
= 4000.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
197
o
cp
= 500 W/(m
2
·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 o
cp
= 4000
W/(m
2
·K) an IPZ model is not sufficient. Note that since the heat transfer
coefficients are different in the two figures, the steadystate solutions are
350
400
450
Q
p
,
Q
m
(
k
W
)
100 150 200 250
50
0
50
100
150
200
250
Time (s)
E
n
e
r
g
y
d
e
r
i
v
a
t
i
v
e
s
(
k
J
/
s
)
Q
m
Q
p
d( µ
s
u
s
V
s
)/dt
d( mC
p,m
T
m
)/dt
d( µ
w
u
w
V
w
)/dt
Figure 8.12 Open loop step response in pressure with o
sc
= 4000 W/(m
2
·K). Heat transfer
coefficient o
cp
is set to a constant value 500 W/(m
2
·K).
0.16
0.20
0.24
M
a
s
s
f
l
o
w
i
n
/
o
u
t
(
k
g
/
s
)
90 100 110 120 130 140 150 160 170 180 190 200
460
470
480
490
500
510
Time (s)
S
t
e
a
m
p
r
e
s
s
u
r
e
(
k
P
a
)
q
s
q
c
+ q
bt
Figure 8.11 Open loop step response in pressure with o
sc
= 4000 W/(m
2
·K). Heat transfer
coefficient o
cp
is set to a constant value 500 W/(m
2
·K).
Chapter 8. ObjectOriented Modeling and Control of the Moisture
198
also different and the two simulations start from different steadystate
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 buildup 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.
8.5
9.0
9.5
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
300
305
310
315
320
325
S
t
e
a
m
p
r
e
s
s
u
r
e
(
k
P
a
)
0 100 200 300 400 500 600 700 800 900 1000
0
20
40
60
80
Time (s)
V
a
l
v
e
p
o
s
i
t
i
o
n
(
%
)
Figure 8.13 Simulation of set point change in steam pressure and a web break.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
199
Figure 8.12 shows that this assumption is valid. Since the cylinder shell
acts as a thermal filter Q
p
is increasing more slowly that Q
m
.
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 M
s
= 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 overshoot 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
K W/m 2000
1 2 . 46
109 . 0
) (
2 1 . 16
=
+
÷ =
÷
sc
s
e
s
s P o (8.24)
and
K W/m 4000
1 2 . 21
0917 . 0
) (
2 6 . 16
=
+
÷ =
÷
sc
s
e
s
s P o (8.25)
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 midrange 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
Chapter 8. ObjectOriented Modeling and Control of the Moisture
200
MPC
Steam
system
Dryer
Setpoint
Moisture
Setpoint
steam pressure
Steam
pressure
PID
controller
Moisture
y
u
c
r
Figure 8.15 Moisture control by single loop MPC.
7.8
8.0
8.2
8.4
8.6
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
5.6
5.8
6.0
6.2
S
h
e
e
t
m
o
i
s
t
u
r
e
(
%
)
o
sc
= 4000 W/m
2
K
o
sc
= 2000 W/m
2
K
450 500 550 600 650 700 750 800
370
375
380
385
Time (s)
S
t
e
a
m
p
r
e
s
s
u
r
e
s
e
t
p
o
i
n
t
(
k
P
a
)
Figure 8.14 Response in moisture to step change in steam pressure set point for two
different values of heat transfer coefficient o
cp
.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
201
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 deadtime 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 highorder nonlinear
physical model, a simple blackbox model is achieved. In the Laplace
domain, it is given by
,
1 48
010 . 0
1 48
098 . 0
2
5
1
14
c
s
c
s
U e
s
U e
s
Y
÷ ÷
+
÷
+
÷ = (8.26)
where y is the moisture content, u
c1
is the steam pressure set point to the
first 50 cylinders, and u
c2
the set point to the last 10 cylinders. The PID
controllers in the inner loops are tuned according to the IPZtuning
MPC
Dryer
Setpoint
moisture
Moisture
PID
controller
PID
controller
u
c1
u
c2
Steam system
Last part
Steam system
First part
y
r
1
Figure 8.16 The proposed moisture control loop.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
202
derived in Chapter 5. The response from u
c2
has a significantly shorter
time delay but also smaller process gain. The advantage of the high gain
from u
c1
and fast dynamics from u
c2
is utilized in a midranging MPC
structure.
The simplified model for the singeloop case in Figure 8.15, obtained
correspondingly to (8.26), is given by
,
1 48
11 . 0
13
c
s
U e
s
Y
÷
+
÷ = (8.27)
The cost function being minimized in the midranging MPC is
,
) (
) (
) ( ) (
)  ( ˆ ) (
) (
3
0
2
2
1
49
0
2
2 2
1
_ _
= =
(
¸
(
¸
+ A
+ A
+
(
¸
(
¸
+ ÷ +
+ ÷ +
=
i
R
c
c
i
Q
c
i k u
i k u
i k u i k r
k i k y i k r
k J (8.28)
where r
1
is the set point for the moisture, and r
2
the set point for u
c2
.
Notation ǔ(k + i  k) denotes the i stepahead 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 u
c2
take
care of the variations in paper moisture and let u
c1
be positioned at a level
where u
c2
has an adequate control range in both directions in steadystate.
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
.
10 0
0 100
,
10 0
0 500
4 3 (
¸
(
¸
=
(
¸
(
¸
=
÷ ÷
R Q (8.29)
There is obviously a larger weight on deviations from r
1
than from r
2
,
since moisture control is the main objective. The weight on ǻu
c1
is larger
than on ǻu
c2
making the controller to primarily use signal u
c2
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 openloop system (to assure that the prediction ‘sees’ the
Chapter 8. ObjectOriented Modeling and Control of the Moisture
203
7.0
7.2
7.4
7.6
y
(
%
)
0 200 400 600 800 1000 1200
430
440
450
Time (s)
u
c
(
k
P
a
)
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.
7.0
7.2
7.4
7.6
y
(
%
)
430
440
450
u
c
1
(
k
P
a
)
0 200 400 600 800 1000 1200
400
500
600
Time (s)
u
c
2
(
k
P
a
)
Figure 8.17 Simulation of midrange 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
Chapter 8. ObjectOriented Modeling and Control of the Moisture
204
full response of a change in ǻu
c1
and ǻu
c2
) and a small control horizon to
limit the computational effort for the MPC. There is also a rate constraint
for the control signals
. kPa 50 , kPa 10
2 1
s A s A
c c
u u (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 midrange MPC and it clearly
visualizes the thought of midranging. The signal u
c2
is used to quickly
react to set point changes and disturbances while u
c1
is used to push u
c2
back to its set point in steadystate. Both during the set point change and
disturbance, the rate constraint for u
c2
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 u
c2
, 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 singleloop 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 singleloop MPC is given by
10
4
10
3
10
2
10
1
0
0.5
1
1.5
Frequency (Hz)
M
a
g
n
i
t
u
d
e
(
a
b
s
)
Figure 8.19 The sensitivity function for single loop MPC (dotted) and midrange 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.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
205
  , ) ( )  ( ˆ ) ( ) (
3
0
2
50
1
2
_ _
= =
+ A + + ÷ + =
i
R
c
i
Q
i k u k i k y i k r k J (8.31)
and the chosen weights that give the sensitivity in Figure 8.19, and
constraint are
kPa 10 , 1 , 10 s A = =
c
u R Q (8.32)
Figure 8.19 also indicates the difference in performance between the two
controllers, at least in the nonconstrained case. The midrange 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 midrange MPC has both better set point
following and disturbance rejection. However, the transient steam
6.8
7.0
7.2
7.4
7.6
M
o
i
s
t
u
u
r
e
(
%
)
0 200 400 600 800 1000 1200
15
16
17
18
19
20
Time (s)
S
t
e
a
m
c
o
n
s
u
m
p
t
i
o
n
(
k
g
/
s
)
Figure 8.20 Comparison between midrange MPC (solid) and single loop MPC (dotted),
showing the moisture content and steam consumption.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
206
consumption is twice as large for the midrange 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 midrange 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 midrange 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
6.8
7.0
7.2
7.4
7.6
M
o
i
s
t
u
r
e
(
%
)
0 200 400 600 800 1000 1200
10
15
20
25
Time (s)
S
t
e
a
m
c
o
n
s
u
m
p
t
i
o
n
(
k
g
/
s
)
Figure 8.21 Comparison between midrange MPC (solid) and single loop MPC (dotted),
showing moisture content and steam consumption when tuned to give similar performance
in disturbance rejection.
Chapter 8. ObjectOriented Modeling and Control of the Moisture
207
receiving group. This guarantees that the steam flows in the intended
direction.
This midranging MPC has two essential advantages compared to the
midranging structure used in Chapter 6, see Figure 6.10. Firstly, the MPC
does not require any decoupling filter, C
D
, 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 midrange MPC
compared to e.g. the IMC or Dahlin controller used by many mills today.
8.5 Summary
A physical model, implemented in the objectoriented 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 draganddrop
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.
10
4
10
3
10
2
10
1
0
0.5
1
1.5
Frequency (Hz)
M
a
g
n
i
t
u
d
e
(
a
b
s
)
Figure 8.22 The sensitivity function for single loop MPC (dotted) and midrange MPC
(solid), when tuned to give similar performance
Chapter 8. ObjectOriented Modeling and Control of the Moisture
208
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 multivariable process with a midrange structure, the performance of
the closed loop system is greatly improved. The midranging MPC is
compared with singleloop MPC and the evaluation is done by comparing
performance in terms of sensitivity function, disturbance rejection and
steam consumption. An important advantage of the midranging 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.
Chapter 9. Conclusions
209
9.
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 blackbox 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
9
Conclusions
Chapter 9. Conclusions
210
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 closedloop frequency
response from set point to process output, and this is accomplished by a
setpoint feedforward controller (twodegreeoffreedom controller) and a
state feedback controller. The two types of structures solve the objective
differently but they roughly give the same performance. However, the
twodegreeoffreedom 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 setpoint feedforward design gives the desired response
from setpoint 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
blackbox model can be derived from a few simple mass and energy
balances. This means that the process parameters of the blackbox 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 midranging controllers.
The first uses the air supply system in combination with the steam system.
Chapter 9. Conclusions
211
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 midranging structure utilizes similar
advantages by combining the last drying group with the remaining part of
the dryers to improve performance. Two different of midranging
controllers are used. The first midranging 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 midranging 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 objectoriented modeling language
Modelica. By draganddrop 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 midranging 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
fullscale 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, crosscouplings
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.
Chapter 9. Conclusions
212
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, Smithpredictor, IMC, PID, deadbeat control,
minimumvariance, 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.
Appendix A. Glossary
213
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/m
2
, 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.
Appendix A
Glossary
Appendix A. Glossary
214
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.
Chemithermomechanical pulp (CTMP): Chemimechanical 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.
Appendix A. Glossary
215
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.
Appendix A. Glossary
216
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 rollfeed 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 oilbased inks (this process is
carried out at the size press, which is about twothirds 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 waterbased 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/m
2
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/m
2
. 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
Appendix A. Glossary
217
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.
Appendix B. Conservation Balance for Compartmental Models
218
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
q
in
h
in
q
out
h
out
( )
i
U
dt
d
Appendix B
Conservation Balance for
Energy in Compartmental
Models
Figure A.1 The general energy balance for steam and condensate, used in this thesis. q is
mass flow, h specific enthalpy, and U
i
internal energy. The change in internal energy
depends on inflow and outflow of enthalpy.
Appendix B. Conservation Balance for Compartmental Models
219
B.1 The energy balance
Starting from a general energy balance, the total energy of a system
comprises of three principal components
 internal energy U
i
 kinetic energy E
k
 potential energy E
p
Observe that we here use internal energy U
i
[J] and not specific internal
energy u
i
[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 Q
c
[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: ) (
, , , in p in k in i in
e e u q + +
Convective outflow: ) (
, , , out p out k out i out
e e u q + +
Appendix B. Conservation Balance for Compartmental Models
220
where q
in
and q
out
are mass flows [kg/s], and e
k
and e
p
are specific energies
[J/kg].
Mechanical work
This is denoted as W
m
[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:
in in in
v p q
Work by outflow:
out out out
v p q
where p
in
and p
out
are pressures [Pa] at inlet and outlet, and v
in
and v
out
are
specific volumes [m
3
/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 crosssectional area of the inlet be
A [m
2
] Then the energy required to push the element across the boundary
is given by the force times length,
in in in in
v p l A p = × ) (
Similarly it can be shown that the flow work at the outlet is given by
out out
v p
Putting it all together
The principle of conservation of energy gives us
out m in m out c in c out out out in in in
out p out k out i out in p in k in i in p k i
W W Q Q v p q v p q
e e u q e e u q E E U
dt
d
, , , ,
, , , , , ,
) ( ) ( ) (
÷ + ÷ + ÷ +
+ + ÷ + + = + +
Appendix B. Conservation Balance for Compartmental Models
221
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
out m in m out c in c
out p out k out out in p in k in in p k i
W W Q Q
e e h q e e h q E E U
dt
d
, , , ,
, , , ,
) ( ) ( ) (
÷ + ÷ +
+ + ÷ + + = + +
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
out c in c out out in in i
Q Q h q h q V u
dt
d
, ,
) ( ÷ + ÷ = µ
where µ is density [kg/m
3
] and V volume [m
3
].
l
inlet
control volume
p
Figure A.2 A small section at inlet to the system. There is an expenditure in energy to
push the fluid across the boundary.
Appendix B. Conservation Balance for Compartmental Models
222
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
out c in c out out in in
Q Q h q h q hV
dt
d
, ,
) ( ÷ + ÷ = µ
This is the case when considering the water phase (moisture) in the paper
web.
Appendix C. Solution to the One Dimensional Heat Equation
223
The general heat equation is a partial differential equation, describing heat
conduction in solids in a given region over time, of the following form
, ) ( k T
t
T
c = V V ÷
c
c
ì µ (C.1)
where ȡ (kg/m
3
) is the density, Ȝ (W/m·K) is the thermal conductivity, c
(J/kg·K) the specific heat capacity, and k (W/m
3
) the added heat per unit
volume and time. Assuming homogeneous material, no added or
subtracted heat inside the given region, and a onedimensional problem, it
can be written as
, 0
2
2
=
c
c
÷
c
c
x
T
a
t
T
(C.2)
where a (m
2
/s) is the thermal diffusivity, defined as
.
c
a
µ
ì
= (C.3)
Appendix C
Solution to the One
Dimensional Heat Equation
Appendix C. Solution to the One Dimensional Heat Equation
224
Heat transfer through a cylinder wall with a step change
in temperature
A steam filled cylinder with wall thickness L, has the temperature T
1
on
the steam side and temperature T
0
on the paper side, in steadystate. At
t = 0 the steam temperature is suddenly changed to T
2
, see Figure C.1.
Also, the heat flow to the paper is assumed to be Q (W/m
2
) . Thus, the
system is given by
¦
¦
¦
¦
¹
¦
¦
¦
¦
´
¦
=
c
c
÷
> =
÷ + =
> < < =
c
c
÷
c
c
Q
x
t L T
t T t T
T T
L
x
T x T
t L x
x
T
a
t
T
) , (
0 , ) , 0 (
) ( ) 0 , (
0 , 0 , 0
2
1 0 1
2
2
ì
(C.4)
From theory of Hilbert Spaces and eigenfunctions the solution to (C.4)
can be written as [Sparr and Sparr, 2000]
L
T
1
T
0
T
2
Figure C.1 At t = 0, the temperature on the steam side is changed from T
1
to T
2
. The two
steadystate solutions are shown in the figure.
Steam side Paper side
T
x
Appendix C. Solution to the One Dimensional Heat Equation
225
.
2
1
sin
) 1 2 (
) ( 4
) , (
1
2
1
2 1
2
2
2
2
_
·
=

.

\

÷ ÷

.

\

÷
÷
÷
+ ÷ =
k
t
L
k a
x
L
k e
k
T T
x
Q
T t x T
t
t ì
t
(C.5)
The solution can be thought of as the sum of the steadystate temperature
distribution (given by T
2
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 steadystate 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 firstorder linear model.
In Chapter 4, the heat equation (C.1) is compared to a simplified static
relation
), (
2 1
T T A Q ÷ =o (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.
Appendix C. Solution to the One Dimensional Heat Equation
226
0 0.5 1.0 1.5 2.0 2.5 3.0
114
116
118
120
122
124
126
128
130
Shell thickness (cm)
T
e
m
p
e
r
a
t
u
r
e
(
o
C
)
t = 0
t = 1
t = 5
t = 10
t = 20
t = 50
t = 100
t = 500
Figure C.3 The temperature as a function of position x for different points of time.
Figure C.2 The temperature as a function of both position x and time t (given in seconds).
T
1
= 120°C, T
2
= 130°C, L = 0.03 m, c = 500 J/(kg°C), Ȝ = 50 W/(m
2
°C), µ = 7000 kg/m
3
,
Q = 10 kW.
Appendix C. Solution to the One Dimensional Heat Equation
227
0 20 40 60 80 100 120 140 160 180 200
117
118
119
120
121
122
123
124
125
126
127
128
Time (s)
M
e
a
n
t
e
m
p
e
r
a
t
u
r
e
(
o
C
)
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 firstorder system.
0 20 40 60 80 100 120 140 160 180 200
114
115
116
117
118
119
120
121
122
123
124
125
Time (s)
T
e
m
p
e
r
a
t
u
r
e
(
o
C
)
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.
Appendix C. Solution to the One Dimensional Heat Equation
228
Heat conduction in an infinitely long rod
Regard the heat equation for an infinitely long rod
¦
¹
¦
´
¦
=
> 9 e =
c
c
÷
c
c
) ( ) 0 , (
0 , , 0
2
2
x g x T
t x
x
T
a
t
T
(C.7)
This equation can be solved by the Fourier or Laplace transform and the
solution is given by the convolution formula
, ) ( ) ( ) , ( ) , (
)
·
· ÷
÷ =  = o o o d g x G t x g G t x T (C.8)
where
.
4
1
) , (
4
2
t a x
e
t a
t x G
÷
=
t
(C.9)
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
.
4
1
) ( ) ( ) , (
4
2
t a x
e
t a
d x G G t x T
÷
·
· ÷
= ÷ =  =
)
t
o o o o o (C.10)
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
, 1
4
1
4
2
=
)
·
· ÷
÷
dx e
t a
t a x
t
(C.11)
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
Appendix C. Solution to the One Dimensional Heat Equation
229
(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 nonphysical 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.
References
230
Åkesson, J., Operator interaction and optimization in control systems,
Licentiate thesis, ISRN LUTFD2/TFRT3234SE, 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): “Greybox
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
midranging 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. SeyedYagoobi (1992): “Theoretical drying study
of singletier versus conventional twotiered dryer configurations.”
Tappi Journal, 75(10), pp. 203í211.
References
References
231
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 nonconvex
optimization.” Automatica, 34(5), pp. 585681.
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 ZieglerNichols
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): Computercontrolled systems í
theory and design, 3
rd
ed., PrenticeHall, 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.
References
232
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
selftuning regulator.” In 4
th
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
zdirection.” Preprints 7
th
New Available Technologies, pp. 60í64,
Stockholm, Sweden.
Brown, T., and D. Millard (1993): “New modelbased machine direction
control algorithm.” In Proceedings TAPPI Process Control
Conference, pp. 15í20, Nashville, TN.
References
233
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 sheetforming processes.” AIChE
Journal, 46(8), pp. 1581í1592.
Chen, S.C. (1992): “Fullwidth 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): “Crossmachine direction
response modeling with twodimensional 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/deadtime 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 papermaking machine, its invention,
evolution and development, Pergamon Press.
References
234
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
20040214.
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.
References
235
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 53
rd
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): Objectoriented 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 midranging control.” Proceedings of 44
th
IEEE
Conference on Decision and Control and European Control
conference, Seville, Spain.
References
236
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 nonfourier 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 wellknown 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. 4145.
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.
References
237
Karlsson, I. (2002): ”Förfarande vid torkning av ett banformigt material.”
Swedish patent, 00048512 (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 multicylinder 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
crossdirectional behaviour in web production: techniques and
experience.” Control Engineering Practice, 3(1), pp. 21í29.
Krumenacker, R. (ed) (1997): Paper machine steam and condensate
systems, 5
th
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.
References
238
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 14
th
International Drying Sympos
ium, Sao Paulo, Brazil.
Ljung, L. (1999): System identification  Theory for the user, 2
nd
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 MDcontrol.” 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 laboratoryscale 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.” 7
th
IFAC
symp. on computer aided control systems design, Gent, Belgium.
References
239
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 nonsquare 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 CoastIslands 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.
References
240
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 10
th
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
pilotscale 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 14
th
International Drying Symposium (IDS’2004),
vol. B, pp. 1263í1270, São Paulo, Brazil.
Persson, H. (1998): Dynamic modelling and simulation of multicylinder
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.
References
241
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 lightweight
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 14
th
International Drying Symposium
(IDS’2004), vol. A, pp. 444í451, São Paulo, Brazil.
Schmidt, E. (1969): Properties of water and steam in SIunits, 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.
References
242
Sell, N. J. (ed) (1995): Process control fundamentals  for the pulp and
paper industry, Tappi Press.
SeyedYagoobi, J., S. J. Sikirica, and K. M. Counts (2001):
“Heating/drying of paper sheet with gasfired 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 greybox 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 7
th
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.
References
243
Stenström, S., B. Wilhelmsson, L. Nilsson, R. Krook, and R.
Wimmerstedt (1994): “Measurement of reference experimental
drying data for the multicylinder paper dryer.” Proceedings of the 9
th
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 3
rd
World Congress on
Intelligent Control and Automation, Heifei, P. R. China, pp.
3724í3728.
Thomas, P. (1999): Simulation of industrial processes í for control
engineers, ButterworthHeinemann, 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,
PhDthesis, 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.
References
244
Wallén, A. (2000): Tools for autonomous process control, PhDthesis,
Department of Automatic Control, Lund Institute of Technology,
Sweden.
Wang, H. (1996): “Applying neurofuzzy 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.
List of Symbols
245
A
cyl
m
2
Inner cylinder area
A
xy
m
2
Area of paper covering the cylinder
C
p,m
J/(kg·K) Specific heat capacity of cylinder shell
C
p,p
J/(kg·K) Specific heat capacity of paper
C
v
m
2
Valve conductance
C(s) í Controller transfer function
d í Load disturbance
d
v
kg/(s·%) Valve constant
d
y
m Width of paper sheet
g kg/m
2
Dry basis weight of paper
G(s) í General transfer function
h
s
J/kg Enthalpy of steam
h
w
J/kg Enthalpy of condensate
ǻH
s
J/kg Heat of sorption
ǻH
vap
J/kg Latent heat of vaporization
ǻH J/kg Energy required to evaporate water from the
paper surface
J í Cost function in MPC
k
c
í Gain of the PID controller
K m/s Mass transfer coefficient for paper sheet
m kg Mass of cylinder shell
M
s
í Maximum value of sensitivity function
M
w
kg/mole Molecular weight of water
List of Symbols
List of Symbols
246
n í Noise
p Pa Steam pressure inside cylinder
p
sh
Pa Steam pressure in header
p
tot
Pa Standard pressure (101.325 kPa)
p
v0
Pa Partial vapor pressure for free water
p
v,a
Pa Partial pressure for water vapor in the air
p
v,p
Pa Partial pressure for water vapor at paper surface
P(s) í Process transfer function
q
bt
kg/s Blow through steam
q
c
kg/s Condensation rate
q
evap
kg/(m
2
·s) Evaporation rate
q
s
kg/s Inflow of steam to cylinder
q
w
kg/s Outflow of condensate
Q í Weighting matrix for MPC
Q
m
W Energy flow to cylinder shell
Q
p
W Energy flow to paper
r í Set point
R í Weighting matrix for MPC
R
g
J/(mole·K) Gas constant ( § 8.31)
R
v
í Valve rangeability
T
d
s Derivative time of the PID controller
T
i
s Integral time of the PID controller
T
m
K Temperature of cylinder shell
T
p
K Temperature of paper
u kg/kg Moisture ratio
u
c
í Controller output
u
s
J/kg Internal energy of steam
u
w
J/kg Internal energy of condensate
List of Symbols
247
V m
3
Volume inside cylinder
V
s
m
3
Volume of steam
V
w
m
3
Volume of condensate
v
x
m/s Speed of paper sheet
w % Moisture content
x kg/kg Water content in air
x
v
í Valve opening
y í Process output
Į
sc
W/(m
2
·K) Heat transfer coefficient steam í cylinder
Į
cp
W/(m
2
·K) Heat transfer coefficient cylinder í paper
į
cyl
m Cylinder thickness / 2
ĳ í Sorption isotherm
Ȝ
cyl
W/(m·K) Thermal conductivity of cylinder shell
Ș í Fraction of dryer surface covered by paper
ȡ
s
kg/m
3
Density of steam
ȡ
w
kg/m
3
Density of condensate
Ĳ
1
and Ĳ
2
í Tuning parameters of midranging controller
Department of Automatic Control
ISSN 02805316
ISRN LUTFD2/TFRT1075SE
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
All rights reserved.To Kristin Department of Automatic Control Lund University Box 118 SE221 00 LUND Sweden ISSN 0280 5316 ISRN LUTFD2/TFRT1075SE 2006 by Ola Slätteke. Printed in Sweden by MediaTryck Lund 2006 .
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. Two new approaches to control the moisture by feedback are evaluated. It also creates opportunity for increased production rate. . Studies have shown that the drying section uses of the total energy requirement in paper making. The first part deals with the control of the steam pressure inside the cylinders. A tuning rule for both PI and PID control is derived and various other controller structures are investigated. There exist several commercial incentives to focus on the performance of the moisture control. Both a blackbox model and a physical model are given for the steam pressure process. The second part of the thesis treats the moisture controller.Abstract The topic of this thesis is modeling and control of the last part of the paper machine – the drying section. while the first part is put at an appropriate level. 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. The physical model from the first part is expanded with a model for the paper. feedforward of a surface temperature signal is examined. Reduced variations in moisture gives opportunity for target shifts (changed set point) which reduces the amount of raw material and steam requirement. The moisture in the paper is controlled by adjusting the set point of the steam pressure controllers. The thesis is divided in two parts. Finally. This gives a complete simulation model for the drying section that is implemented in the objectoriented modeling language Modelica. Many of the results are verified by experiments on paper machines at different paper mills. The time to perform a grade change is often limited by the moisture and shorter grade change time is directly correlated to economic profit.
.
Stefan Ericsson and Lars Jonhed at AssiDomän Frövi. Magnus really deserves an extra salute for the work we have done together. Guy Dumont. and Erik Baggerud. During the last year of my PhDstudies I had the privilege to work for a month at the Pulp and Paper Centre. under direction of Prof. I would like to thank all my colleagues for the years we have had together. 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. Tore Hägglund. It all started as a minor discussion and ended up as a major piece of work. Jonas Warnqvist. Jonas Berggren. I have been given a large amount of independence in my research which is something I have appreciated. . It has also been a pleasure getting acquainted with Stefan Snygg at Stora Enso Hylte. This was an instructive and very interesting time for me. Working at the Department of Automatic Control in Lund is an honor and it is a great atmosphere to operate in. and Krister Forsman. Stig Stenström. At the same time. I have had the opportunity to work with a few people at the Department of Chemical Engineering in Lund. I would also like to acknowledge some of the people at ABB. and Alf Isaksson. Per Sandström. University of British Columbia. Bernt Nilsson. I would particularly like to mention all of my old colleagues at Stora Enso Nymölla. 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. Magnus Karlsson. I will miss you. Vancouver. Our regular meetings have been very constructive and fruitful. 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. and this thesis would not have been possible without their outstanding support. I have learnt a lot from him. First of all I would like to thank my advisors Björn Wittenmark.Acknowledgements There are a number of people who have contributed to this thesis.
.
I noticed that the level was too low but the controller only opened the valve by 40% and it was increasing slowly. He then explained to me the concepts of dynamics. manufactured by Fisher & Porter. 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. 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). One day we were replacing a malfunctioning flow gauge at the pulp dryer and I was watching a level controller at the instrument panel. There were two important control loops to keep an eye on. Both were controlled by singleloop controllers. why the valve was not fully opened.Preface I first encountered process control in the summer of 1990. The next summer I was working at the same site but this year at the instrument department. I learned a lot. I thought that was the appropriate thing for the controller to do if the level was low. the level control and the consistency control. . if I remember it correctly. Luckily I managed to do fine through the night and I was placed there the following nights too. overshoot and stability. 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 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. For the first time in my life I heard words like set point and control signal. After my degree I worked there for a few years more before I went back to the university to become a PhD student. programming the DCSsystem and understanding different control structures. every thing from repairing old pneumatic controllers with liquid solvent. trying to understand how it worked. I got a two minute crash course in control theory by one of the operators. things that are still useful for me today. I remember that I did not understand much of it at that time. I promised the operator to not reach that point and hoped that I was right. I asked the maintenance guy who was dismounting the flow meter.
.
......Contents 1........................ 79 4..5 A twopole model of the steam pressure ........... 69 3.....................1 The model........................ 80 4........................... 58 3..... Modeling and Control of the Steam and Condensate System 3........................................... 38 2........................................................... A Physical Model of a Steam Heated Cylinder................3 Improved set point response by feedforward ......... 91 4............5 A note on the choice of units.......... 19 2............2 The steam and condensate system...................................... 40 PART 1................ 53 3......................................................................... Fundamentals of the Paper Drying Process ..................... 45 3..... 94 4.. 46 3...................... 23 2.......................................................................................... 63 3.......................4 Disturbances in the drying section .................... 89 4..........1 A blackbox model structure the IPZ transfer function ........ 29 2...........6 The differential pressure loop ......4 A modified model ............................................................................................2 PID control of the steam pressure ................................................................ Introduction ..........................4 A state feedback controller..............3 The moisture control loop ..................................... 74 3.. 13 1................................. 77 4............................. 25 2...2 Outline and contribution of the thesis ........................................................ 97 ......3 Comparisons with plant data ........ 22 2..... Blackbox Models and Controller Structures ..........2 Time and frequency domain analysis...................................................7 Summary ..1 Cylinder configurations in the drying section ..........................5 Summary .......1 Introduction and motivation ..................... 13 1.............
.... 196 8...........................................1 Summary .... 144 6.................................................................................................. 115 5.5....................................... 103 5....................................6 Comparison between PI and PID control ................................. Modeling and Control of Paper Moisture in the Drying Section 6.........5 Summary ......... 213 B.....................5 Moisture control by midranging the air system ............... 113 5........ 155 6.......................................... 109 5....................................4 Midranging......... 209 9. 166 7..... Enhanced Moisture Control Using the Air System ...........2 The IPZ tuning rule for PI control.......................................................... 118 5...... Conclusions ..............7 Comparison to other design methods ............... 182 8.......................................................... 144 6..............................................................4 Stability regions.2 Steadystate model validation ................4 Control of moisture by midrange MPC.........................2 The model.............................................................................................. 143 6..................4 Summary ............................... Feedforward from a Paper Surface Temperature Measurement 168 7..................................3 The IPZ tuning rule for PID control........ 211 A............................6 Summary .................... 223 References ......................................... 207 9.......................................... Glossary .................................................. 169 7........................................................................................... 98 5.................... 194 8................. 149 6.. 175 7..........................3 A prestudy ........ 218 C....1 The model............................................................................... 199 8................................ 230 List of Symbols........................................3 Open loop simulations..2 Future work ........................................ 121 5............................................................. ObjectOriented Modeling and Predictive Control of the Moisture Content .................. 138 PART 2............ 245 ..............1 The peak position the position of a dry surface ........5 Industrial verification of the tuning rule....................1 A literature review of drying section models ..................................2 Design of a feedforward controller ....................................3 Simulations. Solution to the One Dimensional Heat Equation .................................... 184 8.................. 209 9.................................................. 183 8..........................................8 Summary ........................ 150 6. A Tuning Method for IPZ Models ..................................................................................................................................1 A design method based on optimization ............ Conservation Balances for Energy in Compartmental Models ........................ 178 7. 99 5.............
Some common types of paper qualities include the following: Copy paper for printers. paper has become a basic material.1 Introduction and motivation Paper is used for printing and writing. copying machines and writing Newsprint Cardboard Lightweight coated paper for magazines Wrapping and packaging paper Hygienic tissue paper Currency paper In modern times. Just try to imagine a day without paper in your life. 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 newspaper in the morning. for wrapping and packaging.Chapter 1. It simply comes in an enormous variety of qualities. Introduction 1 Introduction 1. commonly found in almost all parts of the world. No notepad to write your shopping list on before you go 13 . and for a variety of other applications ranging from kitchen towels to the manufacture of building materials.
An empty mailbox each day you come home from work. Number of companies Number of paper machines Employment Turnover (million euros) Capacity (1000 tonnes) Consumption (1000 tonnes) 1991 1 042 2 181 362 100 39 263 72 343 62 140 2001 918 1 863 288 700 77 028 100 713 83 306 2002 901 1 811 285 000 74 235 103 489 85 674 2003 884 1 815 279 400 71 866 104 978 86 186 14 . As a result the forest industry companies have grown by size and the industry has become more consolidated. 2004].Chapter 1.1. and pharmaceutical. the total production of paper in 2003 was 95 million tonnes with a turnover of €72 billion [CEPI. 2003]. less sensitivity to economic fluctuations. Compared to other industries such as food. better terms of delivery. and other positive synergy effects. The world simply became a better place to live in with the advent of paper some 2000 years ago. reduced labor costs.g. expected to have a rapid growth in coming years. Therefore the production of paper requires constant attention on process efficiency. Companies have realized that it might be cheaper (and certainly quicker) to buy production capacity rather than building it. In Europe. At the same time there is a steady overcapacity in the world. Customers are demanding lower costs. 2004]. The main objectives behind the mergers and acquisitions are lower production costs. Introduction to the super market. reduced transportation costs.1 Figures illustrating the consolidation trend in Europe with less number of companies and paper machines. the paper industry has delivered a relatively low return on capital employed (ROCE). and yet a higher capacity [CEPI. the industry is facing increasing environmental requirements and there is an increased competition from other industries as alternatives to fiber products appear [Dumont. Table 1. increasing productivity. chemical. 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. 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. and lower costs. 2004]. 1988]. The pulp and paper industry is a highly competitive and capitalintensive market that is under increasing price pressure. and higher product quality. No thesis to hold in your hand right now. see Table 1. The plastic packaging demand is e.
This implies that the drying section is the most expensive part of the paper machine in terms of energy use per kg removed water. Introduction The function of a paper machine is to form the paper sheet and remove the water from the sheet. the dry solids content is around 20 %. 1998]. the drying section affects a lot of the important physical properties of the final product. After a few press nips the web enters the drying section with a solid content of approximately about 50 %. and curl. and the drying section. By adding different types of fillers the paper surface obtains different properties. ~ 140 m Wire section 99 % Press section 80 % 50 % Drying section Drying section 5% Figure 1. 15 . These are large hollow metal cylinders. In the press section. Typical values of moisture content are indicated. such as paper sheet elasticity. Although the drying section is only responsible for removing less than 1 % of the water volume in the original stock to the head box. heated internally with steam. Finally.Chapter 1. As it travels on the wire. The water is separated from the original stock which is smoothened out to a thin and endless paper sheet. the press section. When the stock enters the head box in the wire section. which dry the paper as it passes them. twist. the paper is wound up on a big roll and removed from the paper machine.1 The principle of paper production is simple. consumes most energy. see Figure 1. this is the part of the paper machine that. it contains roughly 1 % of fibers or less. Moreover. A paper machine is divided into three main parts. by far. the wire section. It now encounters the dryer cylinders. much of the water drains away by gravitational forces or is pulled away by suction from underneath. the newly formed sheet is pressed between rotating steel rolls and water is displaced into a press felt. Studies have shown that the drying section uses around of the total energy requirement in paper making [Fellers and Norman. As the water disappears.1. When the paper web leaves the wire section and enters the press section. The moisture content is now roughly 5 10 %. This low viscous mix is dispensed through a long slice onto the wire. By courtesy of Skogsindustrierna. the cellulose fibres start to adhere to one another by hydrogen bonds and form a paper web.
1 % corresponds to 365 tons of raw material per year. the moisture feedback loop is often turned off during a grade change and the process is run in open loop (feedforward).2. A modern paper machine makes around 1000 tons of paper per day. see below. Then the drying section often becomes a bottle neck by lacking the required capacity. Introduction For a paper mill. moisture content is therefore measured and monitored online. the paper mill is selling more water at an excessive price (paper is sold according to weight). this in turn means a large economical saving for the mill. A high production rate and capacity is therefore essential to achieve a high return on the investment. In plain language. Below are a few reasons of why a well tuned moisture control system provides economic yield. An obvious way to increase production is to increase the machine speed. An increase in moisture also gives a reduction in energy use (steam consumption). 2004]. With reduced variance the moisture set point can be increased without changing the probability for an offspec product. If the specific paper machine is dryer limited this also gives an opportunity to increase the machine speed. With a production cost for pulp roughly around €500 per ton [Dagens Industri. Large variations in moisture can adversely affect post processing units like calendering. and even for a group of companies. In practice. the converting or packaging line. erecting a new paper machine is a large investment. and the paper product is rejected if it deviates outside the specified limits. Maximum production is achieved by operating at maximum speed while remaining within the control constraints. Due 16 . During production. A well tuned moisture control system will reduce the time to carry out a grade change (state transition). One of the most important quality variables in paper manufacturing is moisture content. Reduced variations in moisture then implies that the speed can be increased without reaching the maximum available steam pressure. A stable and uniform moisture content during normal operation guarantees low reject and consequently high production rates.Chapter 1. or even the customer’s printing press (worsen printability). A reduction of moisture by 0. see Figure 1.
When the paper web is put back. the time it takes to get the paper sheet back on the reel can be reduced. The solid curve represents a condition where the standard deviation has been reduced by 50 % compared to the dashed curve. 0.8 0.2 0. 1972]. Many paper properties depend on moisture content. this thesis is focused on both modeling and control of the drying section of a paper machine. curl. it is possible to increase the set point from 10 to 11 %.6 Probability density Tolerance limit 0. strength and stiffness [Gavelin. As the title reflects.Chapter 1. The difference between the tolerance limit and mean value is sometimes called the giveaway. 17 . 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. These are some of the reasons why the drying section plays a vital role in paper manufacturing. picking and new web breaks easily occurs.4 0. By having an optimized steam control system during a web break.g. e. stretch. Hence.0 5 6 7 8 9 10 11 Moisture (%) 12 13 14 15 Figure 1. Hence.2 The reduction of moisture variation makes it possible to increase the set point (target shift). A very common problem is that the cylinders become overheated since there is no longer any cooling paper around them. The moisture control loop is indirectly involved during a web break by the steam pressure in the steam cylinders. tear. Introduction to model errors in the feedforward loop the moisture will deviate from the set point when the feedback is turned on again.
By 1810. the newly formed paper transferred to a moving web of woolen cloth (the felt). From the wire. 1967]. He was backed in England by the Fourdrinier brothers. their engineer. also known as the Fourdrinier paper machine [Clapperton. the Fourdrinier brothers found themselves in bankruptcy and Bryan Donkin. who built and sold the first paper machines. Soon he was successfully manufacturing a machine that mechanized the process of making paper. while working for the French paper mill owned by the Didot family. It was invented in 1798 by NicholasLouis Robert. His machine used a belt of wire screen to produce a continuous web of paper. vibrating web of woven wire cloth.3 A drawing of the first paper machine from 1808. before being dried. A water and pulp mixture flowed across a moving. Introduction Figure 1. forming a wet mat of interlocking fibers. continued to improve the basic design.Chapter 1. 18 .
see Figure 1. It is compared to a few other design methods and tested on a real paper machine. The method has one tuning parameter that adjusts the tradeoff between performance and robustness. 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 Analysis of different controller structures for the steam pressure loop based on a previously proposed blackbox model.2 Outline and contribution of the thesis Although the fundamental principle of producing paper has not changed since the invention of the paper machine. a large majority of all paper machines in the world are computer controlled.Chapter 1. A new approach to control the moisture content in the sheet by manipulating the last steam group independently of the others. very much has happened since then in terms of quality and efficiency. Introduction 1. Kalman filtering. This control challenge is solved by using midranging of two IMCcontrollers. 1986] and [Bialkowski. and optimal control were introduced. new concepts like statespace theory. 1967] and the selftuning controller [Borisson and Wittenmark. 19 . A new tuning method for both PI and PID controllers based on optimization of disturbance rejection.g. subject to a robustness constraint. Presentation of a physical steam cylinder model with the same structure as the blackbox model. Some of the major breakthroughs in advanced control theory have been tested first in the pulp and paper industry. see also [Dumont. Today. The purpose is to gain deeper understanding in the physics behind the process.3. 2000]. In those days there did not exist much automatic control in a paper mill. Since then. e. It is solved by a model predictive controller (MPC). 1974]. This was in a time when major changes occurred in the area of control theory. the minimumvariance controller [Åström. The pulp and paper industry first initiated computer applications to process control in the early 1960’s. 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.
M.. and S.Chapter 1. M.. Wittenmark. and K. and K. M. O. Karlsson. and S. and S.” Preprints 7th New Available Technologies. Department of Automatic Control.” Proceedings Control Systems 2002. Forsman. Slätteke. Karlsson. pp. Slätteke. O. licentiate thesis. 20 ..” Proceedings American Control Conference 2005. Wittenmark. Stenström: “Feedforward control in the paper machine drying section. B. 20(2). 298 302. M. Slätteke. T. Karlsson.. Wittenmark. Slätteke. Slätteke. Stenström (2003): “Evaluation of models for the steam supply system. Stenström (2005): “Reducing moisture transients in the paper machine drying section with the midranging control technique. B. Åström (2005): “Modeling of a steam heated rotating cylinder A greybox approach. O. Wittenmark (2002): “On identification and control tuning of cylinder dryers. O. pp. IL. and B. T. J. Sweden. Slätteke. OR. Portland. 150 156. O. Design of a model library in Modelica® including components for a drying section. Hägglund. Stockholm. (2003): Steam and condensate system control in paper making. K.. with possibility to easily build a dynamic simulation model of a whole drying section.. Forsman (2002): “Productivity increase from a better understanding of dynamic processes and control of the paper dryer.” Tappi Spring Technical Conference & Trade Fare. S. Publications The thesis is based on the following publications: Stenström. Sweden. pp. O. Chicago.” submitted to American Control Conference 2006. Karlsson. The signal is based on the temperature profile in the machine direction. Introduction Introduction of a new signal which can be used in feedforward to improve control performance. Hägglund. Slätteke.” Nordic Pulp and Paper Research Journal. O. ISRN LUTFD2/TFRT3231SE. Stockholm. B. 70–73.
and the second focuses on modeling and control of paper moisture. It is also validated against plant data. Introduction Slätteke. Chapter 7 introduces a new feedforward signal that is based on temperature measurements of the paper surface. Much of the nomenclature used in other chapters is introduced here. Chapter 4 derives a physical model for the steam pressure process that is compared with the black box model. Chapter 3 presents a blackbox model for the steam pressure process and a few different controller structures are investigated. The first treats modeling and control of the steam and condensate system.Chapter 1. O. The different chapters are organized as follows. Chapter 9 gives a conclusion and suggests possible future work. Chapter 5 presents a tuning rule for the steam pressure controller.: “Object oriented modeling and predictive control of the moisture content in paper production. 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 8 presents a physical model of the drying section that is implemented in the objectoriented modeling language Modelica.” submitted to American Control Conference 2006. 21 . It assumes PI or PID control and is compared to other tuning rules found in the literature. Chapter 2 gives the fundamentals of the paper drying process. 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. Outline This thesis is divided into two main parts.
1972]. 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 glossary can be found in Appendix A.3. The moist paper can be led around a single large steam heated cylinder. called multicylinder drying. Two thorough textbooks about paper drying are [Karlsson. 22 . In this thesis. Since most of the heat content of steam is stored as latent heat. For chemical pulp mills. 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). and high heat capacity. attention is only given to the multicylinder dryer but most of the theory can also be applied to the Yankee cylinder.Chapter 2. the energy can be extracted as mechanical work through a turbine which makes many mills more or less selfsupporting in terms of electricity. Other advantages are low toxicity. A steamfilled dryer is a cost effective method to transfer heat into the sheet. which is a useful attribute in paper drying and many other heating applications. Also. and [Gavelin. large quantities of heat can be transferred efficiently at a constant temperature. see Figure 3. steam is obtained simply as a byproduct in the chemical recovery process line. easy of transportability. 2000]. Fundamentals of the Paper Drying Process 2. The energy in steam has proven to cost less than a quarter of any other available method [Pauksta. 1998].
Chapter 2. It is mainly old board machine running at these speeds. At some machines. Noncondensable gases reduce the partial pressure of the steam in the cylinders and lower the condensation temperature at a given total pressure. the condensate is removed on both sides. it is desirable to let some steam pass through the siphon together with the condensate. 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.1 A twotier configuration. This socalled blowthrough steam ensures removal of condensate. Fundamentals of the Paper Drying Process 2. On slow machines (< 300 400 m/min) the condensate forms a pool at the bottom of the cylinder. As the speed increases. 23 . Effective condensate removal is important for the heat transfer of the dryer cylinder.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. and other noncondensable gases from the cylinder. 1972]. This condensate is drawn off by suction with a siphon and fed back to the boiler house. and the condensate is evacuated on the front side (called the operator side) or the backside. The steam is typically fed to the cylinders on the backside of the machine (called the drive side). In addition. the condensate starts cascading Figure 2. air. 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. Therefore. especially on wide machines.
The dryer fabric is also used to press the web onto the cylinders to provide good thermal contact between the two surfaces. 1997] show that the condensate film thickness can be greatly reduced. The water present in the web is moving directly as vapor through the fabric into the air. this technique has not yet been installed or verified on a real paper machine. To prevent these 24 .2 A singletier configuration.1. the singletier design (singlefelted) and the twotier (doublefelted). Almost every dryer in a modern paper machine has dryer bars on the inside of the cylinder shell. by exchanging the siphon by a rubber scraper. the rim will break down again but at a much lower speed than where it was first formed (hysteresis). These provide higher and more uniform heat transfer from the steam to the cylinder by increasing the turbulent behavior of the condensate. The condensate is then mechanically removed from the cylinder. dryer fabrics are utilized. To support and transport the paper web through the drying section. et al. If the speed is reduced. The dryer fabrics are woven with synthetic yarns and do not absorb any water. and suddenly a rim of condensate forms around the circumference. which is the older one of the two.Chapter 2. one is used on the top cylinders and the other on the bottom cylinders (marked as a dashed line in the figure). The twotier configuration. is shown in Figure 2. To the author’s knowledge. [Peng. Wet paper is transferred unsupported from one dryer to the next. and nearly eliminated. as one might think. and this can cause problems like wrinkles and sheet breaks. Fundamentals of the Paper Drying Process Figure 2. There are mainly two types of dryer arrangements today. They are also called turbulent bars or spoiler bars. Here two separate fabrics are used. One immediate effect of this is that less energy is required to rotate the cylinder and the dryer load drops.
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. Fundamentals of the Paper Drying Process runnability problems at higher machine speeds. normally between five and ten groups. a single fabric is supporting the web on both the top and the bottom cylinders.3 Sketch of a dryer group consisting of 6 cylinders and one common pressure gauge and controller. 2. The steam pressure in the different dryer groups can then be controlled individually to obtain the desired pressure profile through the drying section. dryer steam pressure is increased gradually for drying Main steam header PC Steam header of the group Figure 2. see Figure 2. the bottom row of cylinders is therefore replaced by smaller vacuum rolls to increase the runnability even more [Asensio and SeyedYagoobi. from the first group to the last one.2. CYLINDER 1 CYLINDER 2 CYLINDER 3 CYLINDER 4 Condensate pipe CYLINDER 5 CYLINDER 6 25 .3. there is a direct correlation between the steam pressure and steam temperature and you could also talk about a temperature profile. 1992]. Since the steam inside the cylinder can be regarded as saturated because of the continuous condensation at the cylinder wall. For most paper grades. Since the fabric is between the web and the cylinders in the bottom row. The cylinders in a drying section are divided in separate dryer groups. see Figure 2. In modern machines. Using this technique. no significant drying occurs there. 1989]). as well as in the passage between them.Chapter 2. the singletier configuration was invented (in 1975 at Stora Enso Hylte mill [Carlberg.
1997] give a good review of the steam and condensate system. 1993]. However. This minimum pressure difference depends on both the operating point and machine specific properties. from simple troubleshooting to advanced control schemes. Controller PC2 then adjusts its valve. and adds some extra makeup steam from the header. which operates at a lower pressure. In Figure 2. The simplest. 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 PDCvalve.4 Part of a drying section with a cascade system. The blow through steam from one dryer group at higher pressure is reused in a group operating at lower pressure. and [Krumenacker. Sometimes the PDCvalve is installed in the pipe between group A and tank A.Chapter 2. The differential 26 . which is an efficient arrangement from an energy usage perspective. [Hill. capacity and runnability reasons. it vaporizes and forms new steam) from Tank A is piped to Group B. Fundamentals of the Paper Drying Process Steam header PC2 PC1 Dryer group B Dryer group A To the condenser Tank B LC2 PDC Tank A LC1 Figure 2. way to supply the steam to the different steam groups would be to let them all take steam from the header and dump the blowthrough to the condenser (a heat exchanger unit that heats process water by the left over steam). the fact that the dryer groups operate at different pressures can be utilized and this is done in the cascade system. et al. 1991]. [Perrault. The blow through steam from Group A and flash steam (when some of the condensate meets the lower pressure in the condensate tank. to maintain the desired pressure in Group B. but least energy efficient.4 we see a simple example of a system with two dryer groups.
5 A steam group with a thermo compressor unit. pressure over Group A is preserved by controller PDC to guarantee a satisfactory condensate evacuation. Group B has the lowest pressure in the machine and therefore Tank B must dump its steam to the condenser. In this example. material recycling can also severely affect the overall dynamics and in most cases leads to positive feedback [Morud and Skogestad. 1996]. The disadvantage of the interconnections from the cascade configuration can be resolved by a thermo compressor unit. The PC27 . In this way. A thermo compressor is a device that uses highpressure steam from the steam header to compress blow through steam to a desired pressure. the cascade system is an inconvenience. the energy perspective has higher priority and the cascade system is the dominating configuration. and provides extra pathways for disturbance distribution through the system.Chapter 2. since it introduces additional interconnections between the different control loops. the blow through steam can be recirculated to the same steam group. Fundamentals of the Paper Drying Process Steam header PC 2 1 Thermo compressor PDC 2 1 Dryer group To the condenser Tank B LC Tank Figure 2. Naturally.5. 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. see Figure 2. From a control perspective. Both the PC and PDC controller work in splitrange and the numbers indicate in which order the actuators are manipulated. Some care must be taken since noncondensibles might be recirculated and accumulated. making the different steam groups independent. In general.
To condenser 28 FI TC PC 2 2 1 PC PC PC 1 TI PI 1 2 PC Condenser PI PC 1 TC PC PC 2 To steam box LC SG1 SG2 7 cyl.1 m3 Figure 2. 2 PDC SG3 3 cyl. SG4 PDC SG5 14 cyl. 5 cyl.4 m3 T1 FI Chapter 2. PDC 1 PDC PDC To steam box 1 2 LC 3. FI . 17 cyl.6 A typical piping and instrumentation diagram of the steam and condensate system. PDC SG6 12 cyl. Fundamentals of the Paper Drying Process From boiler LC LC 6 m3 2 m3 LC T0 T2 T3 LC T4 6 m3 LC T5 3 m3 3.
Figure 2. the measurements form a zigzag pattern on the paper sheet. The steam and condensate system almost always uses a casebycase design.5 are purely simplified cases to illustrate some basic ideas.Chapter 2.4 and Figure 2.7 The scanner platform moves the measuring sensor back and forth across the sheet. web breaks or grades with low drying demand. By courtesy of ABB Ltd. 2. ash content. 29 . Due to the MD movement of the paper. there are many other ways to structure a steam. and brightness are measured online in a paper machine. Sometimes the thermo compressor is used in cascading configurations too. indicating both the complexity and the control engineering challenge. PDC leads the blowthrough steam to the compressor and dumps the steam to the condenser only in exceptional cases e. as shown in Figure 2. The sensor is mounted in a scanner platform.8. The conventional technique is to measure the MD and CD signals by scanning the sheet with a single sensor. P&ID of a drying section. moisture. This implies that the MD and CD variations are mixed together by the measuring principle and the two signals must be separated Figure 2.g. The quality control system (QCS) is divided in two separate dimensions. but still somewhat simplified.3 The moisture control loop The measuring principle To control something. where it moves back and forth in the cross direction. fibre orientation. if necessary. Fundamentals of the Paper Drying Process controller primarily uses the recirculated steam from the thermo compressor but have the possibility to also use makeup steam from the header.6 shows a typical. Quality parameters. see Figure 2. Naturally. and Figure 2. color.7.and condensate system. the machine direction control (MD) and the cross direction control (CD). In the same manner. you must be able to measure or estimate it. such as basis weight. caliper.
The primary mechanism today for the control of the moisture MD variations is the dryer steam pressure. Other methods have been proposed. Of course. 1995]. steam boxes (a device that improves the vaporization in the paper by adding superheated steam directly onto the sheet). 2001].Chapter 2. et al. [Chang. or by infrared heating boxes located at intervals across the machine’s width [Dumont. therefore it occurs that the paper is overdried and then remoisturized. et al. 1993]. and [Martinez. 1988] an algorithm is developed. The CD profile. The large arrow points out the direction of machine speed. and [Chen. Notice the different length scale in the machine direction and cross direction. Most of these methods have been tested in labscale for many years but have not yet found acceptance in industry for various reasons [Crotogino. 1999]. [Kastanakis and Lizr. and Condebelt drying [Lehtinen. 2001]. As stated above. which has been running since 1996. and [Chen and Subbarayan. on the other hand. the ultimate objective of these measurements is control. and [SeyedYagoobi. 1991]. A similar decomposing algorithm based on KarhunenLoeve expansion is [Rigopoulos. It is then further developed in [Dumont. 1995]. et al. This 30 . is controlled either by remoisturizing showers. impulse drying [Orloff and Crouse. and [Retulainen. 2000] proposes an elliptic sensor trajectory by variable scanning speed or the use of two scanners traveling in opposite direction to improve the MD/CDestimations. 1992]. et al. 1999]. An even moisture content in the CD is easiest to achieve if it is low. In [Natarajan. then the gain in higher quality has to be weighted against the cost of higher energy use. which uses least squares to estimate the CD component and Kalmanfiltering for the MD component. 1997]. et al. The exception is Condebelt who has one installation in Finland.8 The path of the scanning sensor. like infrared drying [Kuang. 2001]. et al. Fundamentals of the Paper Drying Process ~1000 m ~10 m Figure 2. 1991]. et al. et al. 2001].
10 Structure for the moisture control loop with one scanner device and six steam groups. “2sigma CD”.3 g/m2 respectively. [Heaven.9 Moisture content and basis weight measurements taken from a fine paper machine. thesis focuses solely on the MDcontrol.3% and the basis weight set point was 80 g/m2. and [Kjaer. 1994]. More details about CD estimation and control can be found in [Stewart.2 80.40 Moisture (%) 4. The 2sigma values were 0. The performance of the control system has. historically been described in “2sigma” or two standard deviations of the controlled variables.4 0 500 1000 1500 2000 2500 3000 3500 t (s) Figure 2. All produced reels of paper leaves the paper machine together with a “reelreport” that include statistics like “2sigma MD”. et al. in the pulp and paper industry.0 79. 2003].8 79. et al. 1995]. and “2sigma total” for both the moisture Scanner Dryer Group 1 Dryer Group 2 Dryer Group 3 Dryer Group 4 Dryer Group 5 Dryer Group 6 Machine direction Figure 2. et al.25 0 500 1000 1500 2000 2500 3000 3500 Basis weight (g/m ) 2 80. Fundamentals of the Paper Drying Process 4.056% and 0.6 79. 31 .30 4.35 4.Chapter 2. The set point for moisture in this case was 4.
An example of scanner measurements. For grade change control.10. the moisture in the sheet is controlled by the steam pressure in the cylinder groups. and the set points of the quality variables are constant during long periods and only altered at grade changes. see Section 2.Chapter 2. when the measuring head is positioned at one of the ends in the CD. in machine direction.5 for a comment on moisture units. et al. At 1500 s. This means that the drying process has many degrees of freedom 32 . [Kuusisto. These are the average values for the whole reel. Also.11 Example of feasible steam pressure distribution of the drying section in Figure 2. and basis weight [Sell. 1995]. and [Viitamäki. this is a multiinputsingleoutput (MISO) system. Since the drying section is divided in separately controlled groups. A typical value of p is 50 kPa. 2002]. there is a short period of time when the measurements are not updated. To focus on the variability in this way makes sense since a consistent and uniform product is an important objective. The minimum pressure difference. as pointed out in Chapter 1. most distinct in the basis weight. p. between cascade groups depends on machine speed. but it is the short term 2sigma values that are used to make the decision if the product meets the quality requirements.6 and Figure 2. taken from a machine producing 80 g/m2 of high quality copy paper. Fundamentals of the Paper Drying Process Steam pressure p Group 1 Group 2 Group 3 Group 4 Group 5 Group 6 Figure 2. siphon types. The paper moisture loop As explained previously. 1999]. This is due to the automatic calibration of the scanner. see [Murphy and Chen. during a normal run are shown in Figure 2.9. 2004]. and steam and condensate pipe size. performed at constant intervals.
A combination of the two function alternatives (ratio/difference) in (2. The set points of the other groups are then calculated from that value.11. and secondly it is also important for both runnability and the quality of the paper. 33 . m n 0 (2.Chapter 2.12 The set point r from the moisture controller is distributed to the steam pressure controllers by passing it through a ratio/biasfunction except for the lead group. also called measuring frame. yielding a SISO system for the moisture controller. Some machines use two scanners. the constant relation between the pressure in the groups gives good conditions for the function of the cascade system. and the six dryer groups in Figure 2. m n 0 or .12 are given by 0 kn fn knr mn where kn 1. to improve the control. Figure 2. Firstly. one in the middle of the drying section and one at the end.11 and Figure 2. either as a ratio or a difference. 1.12.10 shows how this can be arranged with one scanner device. The functions f in Figure 2.6. Fundamentals of the Paper Drying Process in terms of control. The moisture controller then manipulates the steam pressure set point of one dryer group and the others follow that one. These expressions can be used to achieve pressure differentials between the groups as in Figure 2. The purpose of this is twofold.1) is of course possible but not common.1) where index n refers to group number. 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. Dryer group 5 (called lead group) operates at the highest steam pressure and receives the control signal from the moisture controller. in this case group 5. see Figure 2. this has been solved by letting all steam pressure controllers follow the same signal. Traditionally.
In Chapter 6. As indicated above. This is in general accomplished by a PI. Fundamentals of the Paper Drying Process Setpoint Moisture IMC Setpoint steam pressure PIcontroller Steam system Steam pressure Dryer Moisture –1 Figure 2. In the outer loop there is in general a model based deadtime compensating controller. it looks like in Figure 2. 1989] or based on the Dahlin type [Dahlin. by giving set point values to the PIcontrollers in the inner loop.Chapter 2. et al. 1995]. 34 .13 A block diagram of the moisture control loop. a midranging control structure by combining two IMCcontrollers is investiSteam pressure Process air Leakage air Moisture Dry weight Production speed Layer distribution Formation Bulk Web temperature Retention Filler Refining Freeness Condition of fabrics Web tension Condensate distribution Tuning of controllers Dryer section Blow through steam Condensate flow Exhaust air Figure 2. 1968] (which is a subset of IMC). A short description of some of the terms can be found in Appendix A. The middle scanner can of course also be used for feedforward control. The inner loop controls the steam pressure in the dryer groups. The performance of these controllers are evaluated in [Bialkowski.13.or PID controller. the scanner at the end of the machine controls the second part. typically of the internal model control (IMC) concept [Morari and Zafiriou.14 A list of variables that affect the final moisture in the paper during the drying process. Drawn as a block diagram. The IMC controls the moisture in the paper sheet. 1996] and [Makkonen. the moisture control loop is a cascade loop.
14. Bulk: High bulk means that the water inside the web has a longer transport distance to the surface and ambient air. two singleloop controllers form a quasimultivariable controller. Apart from the steam pressure in the cylinders.Chapter 2.16. In this way. Degree of Refining: This parameter naturally affects both the freeness (measure of the drainability) and the ability to dry the sheet. see Chapter 8. 1993]. 35 . [Rudd and Schweiger. Leakage air: The air from the machine room is cooler that the preheated supply air and therefore impair the drying conditions. Production Speed: Affects the amount of steam needed. 1994]. 1993]. The same structure is also implemented by a true multivariable controller but with different manipulated variables. see Figure 2. Broke Quotient: This is defined as the amount of broke being blended into the pulp. 1967]. Web tension: High web tension increases the heat transfer coefficient and the drying rate. Other nonconventional moisture control schemes can also be found in [Åström. 1999]. 1996]. To indicate the complexity of the problem some of them are listed below and given in Figure 2. [Brown and Millard. Inlet Moisture: The moisture content of the sheet after the press section is a disturbance variable that normally is unknown. [Wang. Dryer Fabric Condition: An old fabric can be clogged and give a higher evaporation resistance. Retention aids: It is easier to dry the web when the retention is high since it then contains more filler. there are a large number of variables that determine the moisture in the paper sheet. [Murphy.15 and Figure 2. Fundamentals of the Paper Drying Process gated. et al. The broke pulp (if dried before) can be more easily dried than the new pulp. [Xia. et al. 1996] and [Wells. since high production also involves higher vaporization. Air Dew Point: A high dew point inhibits effective evaporation. Dry Weight: A thick sheet is more difficult to dry than a thin sheet at the same production speed.
Fundamentals of the Paper Drying Process Ply loading: In paperboard. weight (g/m ) 2 200 400 600 800 1000 0 200 400 600 800 141 140 139 138 137 0 200 400 600 800 1000 time (s) 0 200 400 600 800 Figure 2. Some of these disturbances are controlled variables and can therefore be regarded as known. hence different physical properties. 13 Moisture (%) 12 11 10 0 142 Cond. the moisture is also affected. leakage air. The set points are indicated with dotted lines. Variations in dryer performance due to conditions of dryer fabrics can be considered as constant since this is a slowly degrading process. 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. unless it is unevenly distributed on the fabric.Chapter 2. or amount of condensate in the cylinders can be very difficult to measure.15 A case study from a fluting machine. Other variations like inlet moisture. This has a large influence on the heat transfer to the paper (and the load on the drives). This influences the drying. different layers consist of different pulps. 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. Blow through steam: In case of improper amount of blow through steam. This opens for possibilities of feedforward. the cylinder may be flooded. and can only be reduced by feedback. which often is the case for production speed and dry weight (see below). 36 .
8 2.moisture (%) 1.17 There are crossconnections between both steam pressure total weight and stock flow moisture (a).15.dry weight (g/m2) 1.8 0. Fundamentals of the Paper Drying Process 1.16 Each ‘*’ corresponds to the mean value of a 30 minute sample from the machine in Figure 2.0 Figure 2. but one of these is easily eliminated by using dry weight as a controlled variable (b).Chapter 2.8 1. a) Steam pressure Drying process Moisture Stock flow b) Steam pressure Paper sheet process Total weight Drying process Moisture Stock flow Paper sheet process Dry weight Figure 2. 37 .4 0.2 2.6 2.4 1.2 1.0 1.6 0. It clearly shows the strong correlation between variations in weight and moisture.0 0.6 0.
Figure 2. The cut off frequency for the moisture is a decade lower (around 0. Fundamentals of the Paper Drying Process Apart from moisture.19. 1999]. and the system was an IBM 1710 with a CPU running at 100 kHz and 80 kB of memory.01 Hz) and there is small frequency peak at 0. The history of process control in relation to the development of computers is overviewed in [Balchen. a part of the corresponding time series is given. By estimating an ARMAX model for closed loop data. there exist others where also the machine speed is included).1 Hz.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. The total weight is naturally affected by the amount of water in the web. The system had a special real time operating system. 1967] and [Åström. basis weight is also measured at the reel and controlled by using the stock flow as the manipulated variable (this is the conventional configuration. the digital computer provided set points to the analog system and it was based on stochastic control theory. written as a part of the installation project.003 Hz. However. This was in the middle of the 1960’s. see Figure 2.17. 2. the amount of water is easy to deduct from the total weight measurement. 2000a]. 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. All control was done in a supervisory mode.Chapter 2. Figure 2. As a matter of curiosity. The steam pressure shows low frequency variations and a frequency peak at 0.18 illustrates an example of the power spectrum for the noise in steam pressure and moisture. where the effect of the controller is removed.18 will be used in different chapters throughout the thesis when analyzing different aspects of the control of both steam pressure and sheet moisture. In Figure 2. Since the moisture is measured. this coupling is eliminated by instead controlling the dry weight and leaving only the crosscoupling stockmoisture behind. a noise model is obtained together with a process model. It can be seen that there is an apparent difference in frequency content between the two variables. 38 . taken from a fine paper machine.
from a fine paper machine. Fundamentals of the Paper Drying Process 10 Power spectrum pressure 3 10 1 10 1 10 10 Power spectrum moisture 0 4 10 3 10 2 10 1 10 0 10 2 10 4 10 4 10 3 10 Frequency (Hz) 2 10 1 10 0 Figure 2.3 4.Chapter 2.2 0 100 200 300 400 500 Time (s) 600 700 800 900 1000 Figure 2.4 Sheet moisture (%) 4. The dotted lines indicate the 95 % confidence interval.19 Time series for the steam pressure and moisture.18 Power spectrum for steam pressure (above) and sheet moisture (below). 39 . used for spectrum estimations. 215 Steam pressure (kPa) 210 205 200 195 0 100 200 300 400 500 600 700 800 900 1000 4.
2) where mwc is the mass of the water content in the sheet and mds is the mass of dry solids. A change in moisture content by 98 % 99 % corresponds to a change of 50 kg water.Chapter 2. expected target group for thesis. by the author.012 kg of water. Throughout this thesis. Which one is used depends on the context.3) Moisture ratio is often used in chemical engineering and literature on physical modeling of paper drying. since this simplifies them slightly. This is the quantity used by most control system vendors and also by staff at the mills.5 A note on the choice of units Moisture There exist two alternatives to express the amount of moisture in the sheet. see also Figure 2. defined as w 100mwc mwc mds [%] (2. Fundamentals of the Paper Drying Process 2. while a change in moisture content by 8 % 9 % corresponds to a change of 0.4) The advantage of the moisture ratio is the linearity and it better reflects variations of water content. Books and articles treating control of the sheet moisture use moisture content. the amount of moisture can also be expressed as moisture ratio. 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. moisture content is used in all figures and also in blackbox models regarding sheet moisture. u 1 u w 100 w (2. 40 . Alternatively. The reason for this choice is that the. The relation between the two quantities is w 100u .20. defined as u m wc m ds [kg/kg] (2.
20 Relation between moisture content and moisture ratio.0 Moisture ratio (kg/kg) 2. temperatures are shown in °C for simple interpretation.5 2.0 1. Temperature Temperatures in formulas are given in K. Fundamentals of the Paper Drying Process 80 70 60 Moisture content (%) 50 40 30 20 10 0 0 0.Chapter 2. it is clearly notified in those equations and the corresponding figures still show moisture content. Pressure All pressures are given in Pa. In figures. It is assumed that the absolute value is 101. 41 .325 kPa above the gauge pressure. However.5 3 Figure 2. It is then clearly notified.5 1. The value is always given in absolute pressure apart from a few exceptions where it is given in gauge pressure.
Fundamentals of the Paper Drying Process 42 .Chapter 2.
painted by Thomas M. Dietrich.Part 1 Modeling and control of the steam and condensate system Papermaking in the early 1900s. This painting illustrates the forerunner of the head box where stock is sprayed onto the wire. By courtesy of Fox Valley Corporation. Wisconsin. . This man is adjusting valves in order to adjust the amount of stock sprayed. It shows the wet end of the paper machine at the Fox River Paper mill in Appleton.
.
It all depends on the context. In this thesis. or lumped versus distributed.Chapter 3. or (iv) improved control system design. It 45 . better terms of delivery. There are different classes of models and which class is best suited depends on the problem. Some examples are (i) improved process understanding from experiments with “whatif” scenarios. The word model is derived from the Latin modus. 1998]. much effort is spent on process modeling [Foss. Blackbox Models and Controller Structures 3. This chapter introduces a linear blackbox model structure for the steam pressure in a drying cylinder that is based on step response data. or linear versus nonlinear. To meet these requirements. linear or nonlinear. et al. preferably continuoustime dynamical models are used. or continuous versus discrete. which means a measure [Bequette. They will be either blackbox or firstprinciples. The purpose of the models is varying. 1998]. and higher product quality. or steadystate versus dynamical. 3 Blackbox Models and Controller Structures The pulp and paper industry is a highly competitive and capitalintensive market that is under increasing cost pressure. and lumped or distributed. (iii) creating process simulators for operator training. (ii) identify the bottlenecks in a process and suggest modifications. Models can be divided into firstprinciples versus blackbox (also known as statistical or empirical). Customers are demanding lower costs.
This model structure has also been suggested in [Sell.0.1) with the process parameters kv=0.1 A blackbox model structure function the IPZ transfer From experiments on a large number of different industrial paper machines.Chapter 3. (3. a corresponding firstprinciples model is presented.1 Open loop step response. and [Nelson and Gardner. and one zero.1) 60 80 100 120 140 160 180 200 60 80 100 Time (s) 120 140 160 180 200 Figure 3. producing a whole range of different paper qualities. Blackbox Models and Controller Structures will be used to analyze the closed loop system for a few different controller structures. T2=3. one pole. T1=58. it has been found that a linear process model can describe the dynamics from the steam valve to the steam pressure. and L=2. 1996]. The IPZmodel is represented by the transfer function G IPZ ( s ) 210 Pressure (kPa) 200 190 180 170 0 85 Valve position (%) 80 75 70 65 0 20 40 20 40 kv 1 sT1 e s 1 sT2 sL T1 T2 . The response can be approximated by (3.1.0027. 3. In the next chapter. This model has an integrator. taken from a liner machine. therefore it is called the IPZmodel. 46 . 1995].
0020. possibly with a time delay.0. and L=1. Blackbox Models and Controller Structures Note that T1 is always larger than T2. The response can be approximated by (3. [Stenström et al.Chapter 3. Figure 3. T1=73. taken from a board machine. typically by a factor of 5 to 50. The transfer function can then be regarded as an integrator in series with a leadnetwork. This gives a characteristic open loop step response. the increasing condensation 525 Pressure (kPa) 520 515 510 505 500 0 100 200 300 400 500 600 31 Valve position (%) 30 29 28 27 26 0 100 200 300 Time (s) 400 500 600 Figure 3. the steam pressure makes a rapid decrease after the first step in the valve position. It is performed on the second of totally nine cylinder groups. However.1) with the process parameters kv=0. 2002] indicates the physics behind this phenomenon. Initially. 47 . and then there is a distinct breakpoint in the curve with a significant change in pressure decrease rate. This is the characteristic appearance of the IPZprocess. The basis weight is 246 g/m2 and the moisture content set point is 7. The same thing occurs at the second step in the control signal. It is different from most processes normally encountered in the process industry. and consequently the saturation temperature of the steam and the condensation rate. When the steam valve position is increased it will increase the pressure inside the cylinder. T2=21.2 Open loop step response.1 shows a step response from a liner machine. which are often approximated by first order systems or pure integrators.5 %. which consisted of six dryer cylinders.
the integrator in (3. The valve is first opened a small amount and then equally closed. and L=2. a step response has been done on a Yankee dryer. In Figure 3.3 Open loop step response taken from a Yankee dryer. This is the same behavior as for the level in a tank when the influent or effluent is manipulated.Chapter 3. As will be shown later. The model corresoinds to (3. In closed loop control. When the control signal is put back to the original position.5 m 710 Pressure (kPa) 700 690 680 670 660 0 50 150 250 350 450 550 Valve position (%) 74 72 70 68 66 64 0 50 150 250 Time (s) 350 450 550 Figure 3.1) with the process parameters kv=0. This and many other physical properties in the drying section will be further analyzed in Chapter 8.0026. there will be a fast initial buildup in steam pressure. Blackbox Models and Controller Structures rate lags behind the increasing steam inlet flow as the condensate layer heats up to the new steady state temperature. These are mainly used for drying of tissue and have diameters up to 5.0 48 . before the steam consumption has reached its new value.2. T2=87. this implies that the controller output always returns to the original level in steady state. in absence of disturbances acting on the process. a new level of steady state pressure is reached. the effect of the integrator in the model structure is particularly apparent. Therefore. To examine if the dimension of the cylinder has any impact on the model structure.1) is an approximation of a real pole close to the origin. T1=269.
5 0 20 0 20 40 Time (s) 60 80 100 120 Figure 3.5 or 1. Exceptions are [Ljung. and L from a simple open loop step response. 2000].4 shows a step response of an IPZmodel (3. where L is chosen to zero for simplicity.5 3 q 2(t) 2. The result is shown in Figure 3. T2. kv.Chapter 3.4 Simulated unit step response of an IPZprocess with kv=0. Figure 3. and L=0.y0) Output (kPa) 2 q 1(t) 1.01.3. 49 . 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. we will investigate how physical properties of the steam and dimensions of the cylinder will affect the parameters of the linear process model.1). 2004] and [Wallén. Blackbox Models and Controller Structures [Karlsson. T1=220.5 (t0 . which normally are 1.5 1 0.8 m in diameter. suited for their specific control system). Start by drawing two straight lines. This should be compared with multicylinder dryers. 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. Therefore a graphical identification procedure for the IPZ model is helpful and it will now be illustrated how to obtain the four parameters. these are seldom systematized for identification of the IPZprocess or work preferably in discrete time. T1. the 3. In Chapter 4. 2000]. T2=20.
6) and to get the equation for the asymptote of the step response we observe that 50 . Also.y0).3) By the initial value theorem. To derive the expressions in (3.2) 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. (3. we use the final value theorem k2 t lim dy (t ) dt lim s 2 G ( s ) s 0 u0 s lim s 0 k v 1 sT1 u0 1 sT2 kvu0 (3. the step response can also be written as y (t ) k v u 0 T1 T2 1 e t / T2 t. T1 u0 y0 . T2 (3. By the inverse Laplace transform.Chapter 3. Then we have kv k2 . (3. the initial derivative is k1 dy (t ) dt t lim 0 t 0 dy (t ) dt s lim s 2G ( s ) u0 s s lim kv 1 sT1 u0 1 sT2 kvT1 u0 .4) Thus the tangent of y(t) at t = 0 is q1 ( t ) k v T1u 0 t. 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. start by denoting the step response by y(t).2). Blackbox 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)). T2 k2 t0 . T2 (3. both marked in the figure. suppose that the two lines intersect at the coordinates (t0. To get the final slope of the step response (which also is the steady state value of the impulse response).5) since the response starts in the origin.
11) y2 . T2 (k1 k 2 )k 2 y2 k1 k 2 . see Figure 3. given in (3.7) which is the value of q2(0). (3. Let the two lines in Figure 3. the final slope is hard to acquire and the identified value will depend the length of the response. (3.8) and the two lines q1(t) and q2(t) intersect at t=T2.12) We also have kv k2 . y0 k 2t0 y2 .2). the observation can be verified analytically. 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.4 be given by q1 (t ) k1t . q2 (t ) k 2t The intersection of the two lines is the given by y0 which can be rewritten as y0 k1 k1 k 2 y2 . (3. (3. Blackbox Models and Controller Structures t lim y (t ) kvu0t k v u 0 T1 T2 .4. T1 u0 k1 y2 . Consequently. (3. since q1 (T2 ) q 2 (T2 ) k vT1u 0 . By analyzing the sensitivity of the parameter estimation. This is probably due to difficulties with finding the slope of the asymptote q2.9) A note on the identification procedure When doing graphical identification of a step response of a steam cylinder process. we have q 2 (t ) kvu0 t T1 T2 .13) 51 . (3. t0 1 k1 k 2 y2 . (3.10) k1t0 .Chapter 3. If the effect from the lowpass filter part with time constant T2 has not vanished from the response.
since. Consequently the observation is confirmed.5) that kvT1/T2 is independent of line q2. Also. k2 dE1 E1 k2 k1 dk 2 . As we will see later. dT1 dk 2 and rewrite as a relative error dT1 T1 k1 2k 2 dk 2 k1 k 2 k 2 (3.Chapter 3.14) where E1 and E2 are the two expressions of interest. so it is therefore obvious that kvT1/T2 tends to be independent of different experiments.17) The first thing to notice is that E2 is independent of the slope k2. 2 (k1 k 2 ) 2 k 2 (3. and differentiate T1 with respect to k2.5. if there is an uncertainty in k2. u0 (3. Rewrite (3. k1 is larger than k2. the relative error of E1 is smaller than the relative errors of kv and T1.15) Similar calculations for the other parts of (3. If the identification procedure tends to get a good estimate of the height but a slightly altered kv. A constant product kvT1 is equal to a constant initial height in the response. k2 k2 dE2 dk 2 0.18) The step response of this function is shown in Figure 3.16) k1 y2 (k1 2k 2 ) . long durations have the same effect due to the 52 . The product kvT1 can be interpreted as follows. It is not only short durations of the step response that can cause this difficulty. this will give a small variation in T1 too. Assume an error in k2. E2 (k1 k 2 )u0 kvT1 T2 k1 . (3.13) gives dk v kv dk 2 . Blackbox Models and Controller Structures and E1 kvT1 k1 y2 . It is also clear from (3.1) as G IPZ ( s ) T 2 0 kv s k vT1 e sL (3.
19) where uc(t) is the control signal. Often the set point is not included in the derivative part ( = 0) and there is also a first order filter on the derivative part. The advantage of PID control is that its function is easy to understand. (3. C c kc 1 1 Ti s Td s .20a) 53 . Written in the frequency domain using the Laplace operator. y(t) is the process output. Td. Ti. approximation done when the slow time constant is assumed to be an integrator. more than 95% of the control loops in process industry are of PID type [Åström and Hägglund. 2005]. can be written as uc (t ) kc r (t ) y (t ) kc Ti t 0 r( ) y( ) d kcTd d dt r (t ) y (t ) (3.5 A unit step response of the IPZ process. gain scheduling.Chapter 3. It is also possible to tune it. sometimes with more advanced features like autotuning. r(t) is the set point. 3. even for people without knowledge in control theory. without having any process model (you need to know the direction of the process response though). (3. at least with an acceptable performance. In general. if the lowpass filtering part is removed or equivalently T2 0. and loop performance analysis. in ideal form. The PID controller. Blackbox Models and Controller Structures kv kvT1 Figure 3.2 PID control of the steam pressure It is probably fair to say that in today’s paper machine control systems.19) becomes C ff kc 1 Ti s Td s . and are controller parameters. and kc. the standard PI or PID controller is the most frequent controller element used to regulate the steam pressure. The PID controller can be found in all major DCS systems. .
To see this. and n measurement noise.Chapter 3. The closed loop behavior The pressure loop is in cascade control with the moisture control. which turns out to be caused by the integral action in the controller. 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. r is the set point.6 Block diagram of the PIDcontroller used in this thesis. let the 54 . C c (s) Td 1 s N kc 1 1 Ti s Td s . A lowpass filter added to the derivative part gives C ff ( s ) kc 1 Ti s Td s . like the other parameters in the controller [Isaksson and Graebe.20b) where N = 2 20 [Åström and Hägglund. to avoid sheet picking. and other runnability problems caused by too rapid heating of the sheet. as noted before. d a load disturbance. together with the process. But it is not unusual that the steam pressure in the first dryer group is run with a constant set point. uc the control signal. The step response has an overshoot.6.7 shows a typical behavior of the pressure when there is a step in the reference value. Figure 3. where Cff and Cc are transfer functions. given by the operator. 2005]. 2002]. fibre rising. and therefore its set point varies more or less continuously during normal operation. Td 1 s N (3. defined by the block diagram in Figure 3. y the measurement. Blackbox Models and Controller Structures r Cff (s) d n uc y Cc (s) P(s) 1 Figure 3. dusting.
Chapter 3. s (3. s 0 1 ˆ P( s ). (3.22) and since E (s) R( s ) Y ( s ) 1 G ry ( s ) R ( s ) G re ( s ) R ( s ).21) (3. assume that the process has an integrator.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. Write the process transfer function as P( s) where ˆ ˆ P(0) 0 and lim sP( s ) 0. controller be given by (3. we can write the transfer function from set point to control error as 55 . taken from a fluting machine. Blackbox Models and Controller Structures 745 Pressure (kPa) 740 735 730 725 0 50 100 150 200 250 300 56 Valve position (%) 54 52 50 48 46 0 50 100 150 Time (s) 200 250 300 Figure 3.7 Typical closed loop step response of the pressure loop when using a PID controller.20a). Now.
This implies that there must be an overshoot in the set point step response to compensate the initial positive area in the time plot.24) Now let us study the integral of the controller error as time goes to infinity and the set point is a step. Blackbox Models and Controller Structures kc Gre ( s ) 1 Gry ( s ) 1 1 kc 1 1 Ti s 1 Ti s Td s Td s 1 Ti s 1 ˆ P( s) s 1 ˆ P( s) s Td s 1 ˆ P( s) s 1 kc 1 1 1 ˆ Td s P ( s ) kc Ti s s 1 kc 1 1 Ti s Td s 1 ˆ P(s) s Ti s 2 ˆ ˆ k c Ti s 1 TiTd s 2 P ( s ) k c Ti s 1 TiTd s 2 P ( s ) ˆ Ti s 2 k c (Ti s 1 TiTd s 2 ) P ( s ) ˆ Ti s 2 1 (1 )Td s P ( s ) kcTi s . ˆ Ts 1 (1 )Td s P( s )kcTi lim i 2 ˆ s 0 Ts kc (Ti s 1 TiTd s 2 ) P ( s ) i (3. when = 1. ˆ Ti s 2 k c (Ti s 1 TiTd s 2 ) P ( s ) (3. The range of is normally defined as 0 1.Chapter 3.25) In the classical case.26) 56 . (3. e(t )dt 0 lim s s 0 1 Gre ( s ) s2 1 Ti . We can also conclude that >1 will give an overshoot in the step response since the integral has a negative value. the integral of the control error is zero.
Chapter 3.7. possibly because of this response amplification.8 G ny 0. Figure 3.5 1 G ry 0.4 0. An interesting observation is the resonance peak in Gry.25). Gny is the response from noise n to output y (the sensitivity function) and has implications for the robustness as will be discussed later. Blackbox Models and Controller Structures 1.2 1. We will now show how this peak can be avoided by having controllers with more degrees of freedom than the standard PID. 57 .5 0 4 10 1. In practice it has been observed that often the pressure variations in the steam and condensate system are in that specific frequency region.8 shows two closed loop frequency plots for the process in Figure 3. and extensive simulation results have indicated that there will be an overshoot in this case too.0 0.7.6 0. Note that the integral error is independent of the controller gain. kc. in (3. This is obviously an undesired feature and problems might occur if all steam pressure loops amplify r in the same frequency region. when governed by a PIDcontroller.2 0 10 4 10 3 10 2 10 1 10 0 10 1 10 3 10 10 Frequency (1/s) 2 1 10 0 10 1 Figure 3. showing how some frequencies are amplified by the controller. Gry is the response from set point r to output y.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.
where P is the process. 2005].2 with the set point weights and . However.28) If we require that Mu is realizable. The condition for ideal feedforward is Mu P 1M y . 1997] and [Åström and Hägglund. In fact. In addition. this has already been introduced for the PID controller in Section 3.3 Improved set point response by feedforward Normally feedforward is introduced as a method to reduce the impact of measurable disturbances. Blackbox Models and Controller Structures 3. My should have the same zeros in the right half plane as P. The transfer function from set point to process output is then G ry P(CM y Mu) My PM u My .27) 1 PC 1 PC where My is the desired response from r to y. we will have a more systematic treatment for a general controller but assume that the process is given by (3. see [Åström and Wittenmark. it can also be used to improve set point responses.9. (3. Assume that we have a controller structure as in Figure 3. 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. (3. also known as a twodegreeoffreedom controller.1). Here. The advantage of this configuration is that the servo and regulation problems are separated. C the controller.9 A twodegreeoffreedom controller (2DOF).Chapter 3. and Mu and My are feedforward filters. r Mu e My C ufb uff uc P y 1 Figure 3. 58 .
59 . the control signal ufb remains constant during a set point change and the purpose of C is simply CONTR y r My Mu MV OUTP SP FF uc Figure 3.30) (3. Under ideal conditions. Blackbox Models and Controller Structures Let P be given by the IPZ transfer function P kv 1 sT1 e s (1 sT2 ) sL . k v T1 (T1 Tcl ) (1 sT1 ) 1 e 1 sTcl sL . This is necessary if the DCSsystems have no component for high order filters or no possibility to implement the pure derivative.31) and uff can be generated as the output of the sum of a constant gain and two low pass filters. the objects should be grouped together and presented as one controller element in the operator station. Observe that the calculations made above are independent of the feedback controller C.Chapter 3. as indicated by the dashed line.9 in a DCS system. (3. To let the operator manipulate the correct set point.10 Implementation of the feedforward structure in Figure 3. (3.29) and choose the desired set point response to My then Mu s (1 sT2 ) k v (1 sT1 )(1 sTcl ) T2 k v Tcl T1 Tcl T2 1 k v Tcl (T1 Tcl ) (1 sTcl ) T1 T2 1 .
8 0.2 0 0 10 20 30 40 50 Time (s) 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100 Figure 3. 60 . and Tcl = 20 (dashdotted). k v (1 sT1 ) (1 sTcl ) s (3.4 0. bumpless transfer. Figure 3. Observe that normally there is a pure time delay in My. 1. The dotted line is the set point. and Tcl > T2 gives a lowpass filter.10 shows how the feedforward part could be implemented in an industrial control system to make use of its antiwindup. Blackbox Models and Controller Structures to give good disturbance rejection and robustness. Observe that T1 is always greater than T2. more lead action is added to the first leadnetwork.6 0. The transfer function from set point to control signal is G ru Mu (1 sT2 ) .8 0.11 Set point step response of the 2DOFcontroller.32) and we have leadnetwork (lowpass filtered derivative) followed by a zeropolenetwork. Tcl = 5 (solid). and other functions.0 Process output 0. How should Tcl be chosen? Looking at the shape of the control signal gives a hint.2 0 0 1 Control signal 0. If Tcl < T2. In reality.6 0.Chapter 3.2 1.4 0. Tcl = 10 (dashed). 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. which also should be implementable in the control system.
Figure 3.6 0.2 0 0 1 Control signal 0. T1=200. solid PI with aggressive tuning. the control signal will always go to zero in absence of disturbances. The selection of Tcl will be a tradeoff between performance and control action.2 1.6 0. kvTclT1 (3.31).4 0. However. the initial value theorem gives u c ( 0) T2 .8 0. Blackbox Models and Controller Structures If r is a unit step. and letting Tcl T2 is a good choice. It clearly shows how the relation between Tcl and T2 affects the shape of uc.2 0 0 50 100 50 100 150 200 250 300 150 200 250 300 Time (s) Figure 3.Chapter 3.11 shows a simulation where the process is given by kv=0. The thin dashed curve is the set point.8 0. 61 .32).4 0. L=2.01. due to the cancellation of the slow process zero in 1/T1 by (3. the control signal is slowly brought back to zero long after we have reached the set point. T2=10.0 Process output 0.33) This can also be seen in the partial fraction expansion of Mu in (3. and dashdotted PI with robust tuning. But the control signal must behave like this to maintain the output at the 1.12 Dashed 2DOFcontroller Tcl = 10. Since there is an integrator in the process. Note that the 2DOFcontroller cancels process dynamics from r y but not from load disturbance d. therefore there is a slight overshoot when the disturbance acts on the system.
5 0 0 20 40 60 80 100 Time (s) 120 140 160 180 Figure 3. Since the transfer function from r to y is a first order system. The two settings are chosen to give both a faster and slower response compared to the 2DOFcontroller. 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. here denoted as aggressive and robust.5 0 0 20 40 60 80 100 120 140 160 180 1. Therefore.Chapter 3. Blackbox Models and Controller Structures 1. What is meant by well tuned is obviously relative. Figure 3. The method has one user parameter that determines the robustness of the loop and two different settings are used in the figure. and doubled process gain (dashed).0 Process output 0. Perfect process model (solid). It is also an advantage that this structure is easy to implement into most commercial 62 .0 Control signal 0. desired value. even though the set point has been reached the system has not reached steadystate.12 shows how a welltuned PI controller behaves without the feedforward compared to the 2DOFcontroller. and the cylinder and paper temperatures are still increasing. To compensate for this the controller must add extra steam to the cylinder long after the set point has been reached.13 Robustness analysis to modeling errors of the 2DOFcontroller. the frequency response does not have a peak as in Figure 3.8. doubled time delay in process (dotted). The dashdotted line is the set point. The 2DOFcontroller gives a smoother performance in set point response. which is a nice feature of the 2DOFcontroller structure. The tuning method used for the PI controller is introduced in Chapter 5.
3. which is not the case for the controller structure presented in the following section. The zeros of the closed loop systems can not be positioned with state feedback but the poles can be located arbitrarily. This is also possible to obtain by a state feedback controller as will be shown here.13 shows closed loop simulations where the time delay and gain in the process is increased by a factor two. which is sufficient to avoid the overshoot. the control system proves to handle model errors well. In practice. the delay is assumed to be zero. Figure 3. the sensitivity to modeling errors is investigated. In the previous section.4 A state feedback controller As we have shown. (3. Finally. and the second term in (3. 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. 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 purpose here is merely to show the idea. Since a time delay in continuous time can not be represented by a finite number of poles and zeros. However. s ( s a1 ) (3. a feedforward filter was introduced which gives a closed loop response without this overshoot and peak. Blackbox Models and Controller Structures DCSsystems.35) y (t ) Cx (t ) 63 .34) Assume that the model is written in observable canonical form x(t ) Ax(t ) Buc (t ) 1 0 x(t ) a1 1 0 0 x(t ) b1 b2 uc (t ) . Let the process be given by k v T1 s T2 s s kv T2 1 T2 G(s) k v sT1 1 s ( sT2 1) b1 s b2 .Chapter 3. 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.27) is not canceled by Mu.
which practically is unlikely.Chapter 3. since an observer will not affect the dynamics of the closed loop system from set point r to output y.40) Note that we have assumed that the two states of the system are measurable. 64 . (3. It is also controllable unless T1 = T2 which implies that we have a pure integrator instead of an IPZ process. What is also needed then is a Kalman filter to estimate the states. However. If we choose to place the poles in we get the equation system b1 b2 b2 a1b2 l1 l2 1 2 1 2 a1 .38) (3. Blackbox Models and Controller Structures By definition this system is observable. except for initial transients in the estimation.37) which in inputoutput form becomes G ry ( s ) By choosing lr l1 a1l 2 . 1 (b1 s s 2 b2 )l r a1b2 l 2 b2 l1 a1 b1l1 b2 l 2 s . (3. it is adequate to consider simply the state feedback in this analysis. 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 ). we get unit steadystate gain. (3.36) The closed loop system then becomes x(t ) y (t ) a1 b1l1 1 b1l 2 b2 l1 1 0 x(t ) b2 l 2 x(t ) lr b1 b2 r (t ) . (3.39) and 2. which is feasible since the system is observable.
14 has one drawback.02. and let the other one be at 2 = 0. The controller parameters can then be calculated as l1 1. and finally one case where one pole is slower than the zero. When 2 = 0.04. called x3 x3 r y r Cx.42) Figure 3. there will be a steady state error in the output.14 shows the simulation results.1s s2 0. and 0. (3. To examine a cancellation of the process zero. 12. 0.43) The new system can then be written as 65 .5 1125 (3. In this case. we let 2 = 0. Adding Integral Action The feedback loop in Figure 3. This can be accomplished by augmenting the system with an extra state. and the closed loop system has one fast and one slow pole.25 12.5 l2 lr 100 2. (3.2. the performance of the system is given only by the position of the other pole. We can see that when both poles are faster than the process zero we get an overshoot due to the lead action of the zero (which is closer to the origin).41) The open loop poles are in p1 = 0 and p2 = 0. 2. In other words.Chapter 3.02. place one pole at 1 = 0. In presence of disturbances. The slow pole is clearly visible in the step response. and the zero is in z = 0.02. we have one case where both closed loop poles are faster than the process zero. one case where one pole cancels the zero.02 50s 1 s 10s 1 0. Finally.01 (in three different simulations). 2.1. By introducing integral action. we cancel the zero and get a first order response.01. the asymptotic error becomes zero. Blackbox Models and Controller Structures A Simulation Example Assume that we have the system Gs 0.1s b1 s b2 s2 a1 s .002 0.
0 0. 2 = 0. Blackbox Models and Controller Structures 1. 2 x1 x2 x3 a1 0 1 1 0 0 0 0 0 x1 x2 x3 b1 b2 u c 0 0 0 r.5 y. r 1. Set point (solid). = 0. and 2 = 0.02 (dotted).45) we get the following closed loop system x1 x2 x3 a1 b1l1 1 b1l 2 b2 l1 1 b2 l 2 0 b1l 3 b2 l 3 0 x1 x2 x3 0 0 r.14 Simulation of the three different cases of state feedback.46) 66 . By using the standard state feedback law uc Lx l 3 x3 . 1 (3.Chapter 3.01 (dashdotted). 1 (3.04 (dashed).5 0 0 20 40 60 80 100 120 140 160 180 200 4 3 uc 2 1 0 0 20 40 60 80 100 Time (s) 120 140 160 180 200 Figure 3. (3.44) This augmented system is also controllable from uc unless T1 = T2.
(3.0008 0. Blackbox Models and Controller Structures r ki s uc y Process L 1 Figure 3.48) A Simulation Example.50) which has the solution l1 = 3. l2 = 10 and l3 = 0.42 0.49) and get the following system of equations a1 a1b2 l 2 b2 l 2 b2 l1 b1l1 b1l 3 b2 l 3 0.Chapter 3.2 .15. can be added by having an additional feedback loop as shown in Figure 3.16 shows the simulation result. Figure 3. The integrator gain. 0.15 Adding integral action to a state feedback system which has the transfer function Gry ( s ) l3 (b2 b1s ) . Since there is an extra pole in the closed loop system 67 . (3.048 .4. (3. Cont’d We add an integrator to the state feedback controller and let one closed loop pole be placed in = 0. we want the characteristic equation to be given by s3 0.47) b2l2 ) s 2 (a1b2l2 b2l1 b1l3 ) s b2l3 s 3 (a1 b1l1 The new state. is then given by ki l3 .42 s 2 0.02 (cancellation) and a double pole in = 0.048s 0. ki. x3.0008 (3. In other words.
the dynamics of the set point step response is different from the system in (3. 68 . and with integral action (dashdotted). The practical disadvantage with state feedback is that most commercial DCS do not have any support for this controller type. as the standard PIDcontroller introduces. it must be implemented by the user together with features like windup protection. is eliminated by canceling process dynamics by feedforward.0 0. x3. Instead. Without integral action in the controller (dotted same as the dotted curve in Figure 3.5 0 0 20 40 60 80 100 Time (s) 120 140 160 180 200 Figure 3.3 the overshoot in set point step response.5 0 0 20 40 60 80 100 120 140 160 180 200 2 1. Here we have shown how the process zero can be cancelled by one of the closed loop poles introduced by feedback. due to the integrator state. A Remark In Section 3.5 uc 1 0. r 0 1.14). we avoid both overshoot in the step response and resonance in the frequency response. Observe that the dynamics are different in the two cases since the number of closed loop poles is different. bumpless transfer.42). Blackbox Models and Controller Structures 1.5 y. and bumpless parameter change. By letting the remaining two poles (in the case of integral action in the controller) be real.Chapter 3.16 A simulation where a disturbance acts on the controller output for the case when there is a cancellation of the process zero.0 0.
Since it in most practical cases is not desirable to let the steam pressure in an open loop step response reach steady state. we get G PPZ ( s ) kp 1 sT1 e 1 sT2 1 sT3 sL . Taken from a liner machine. However. Here the integrator is replaced by a new pole close to the origin. the integrator in (3.1) is more likely to be a long time constant. With an extra pole.51) instead of (3. 69 . (3. because of the influence on the paper qualities.5 A twopole model of the steam pressure Since the steam pressure in a cylinder group finally must reach some steady state value. T3 T1 T2 .Chapter 3. the purpose of the model is not simulation and the question is if its complexity is sufficient for a control design usage. This implies that the model does not capture the low frequency behavior well and this has been one of the criticisms against it.1). Blackbox Models and Controller Structures 3. this slow pole can be difficult to identify graphically 280 Pressure (kPa) 260 240 220 1600 45 Valve position (%) 40 35 30 25 20 15 1600 1800 2000 2200 2400 2600 2800 3000 3200 u 1800 2000 2200 2400 2600 2800 3000 3200 Time (s) Figure 3.17 Closed loop step response. and instead we call the model structure PPZ (polepolezero).
Blackbox Models and Controller Structures from a single step. and assume a step in the set point with size r u lim uc (t ) u0 t lim s Gru ( s ) s 0 r s u0 u0 Gru 0 r . In Figure 3. exposes the PPZstructure of the process. see (3.1).Chapter 3. Then we have Gru ( s ) C ff ( s ) 1 Cc ( s )GIPZ ( s ) kc s s 2 TiTd kc 1 kv kc 1 1 sTi s Td sL sT1 1 1 Td s e Ti s s sT2 1 . (3. Assume that the process dynamics are described by the IPZmodel given in (3.52) where Gru(s) is the transfer function from the set point r to the control signal uc.51). sL (3.20a). Start by defining the change in control signal as time goes to infinity. If we instead let the process be of PPZtype. Also.17 a closed loop step response from a steam group in a liner machine. assume that the controller is given by (3. such as the predictionerror methods. though.17 should have returned to its original steady state value (to the level it had before the step in the set point). (3.53) s Ti 1 sT2 1 2 s Ti sT2 1 2 kv kc s TiTd sTi 1 sT1 1 e The steady state value of the control signal is then u Gru 0 r 0. It can be identified from well excited closed or open loop data with system identification techniques for linear parametric models. This is obviously because of the integrator in the process and it means that the control signal in Figure 3. consisting of 13 cylinders. 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. then the steady state value of the control signal will become 70 . is shown.54) and we can see that the final value of the control signal must be equal to zero.
71 .55) s 2TiTd 1 sT1 e and it depends of the process gain and the size of the step. Measurements taken from a board machine. Figure 3. Normally.18 Model validation showing the difference between the IPZ (dashed) and the PPZ (dotted) process.18 shows an example from a paper board machine (a group consisting of 12 cylinders) where almost two 430 420 410 Steam pressure (kPa) 400 390 380 370 360 350 Identification 340 330 60 valve position (%) Validation 40 20 0 0 1000 2000 3000 Time (s) 4000 5000 6000 Figure 3. Blackbox Models and Controller Structures C ff 0 1 Cc 0 GPPZ 0 r s 2 TiTd 1 sT2 1 sT3 kc k p 1 sTi r sL s 0 kc 1 s Ti sTi 1 sT2 1 sT3 r kp (3.Chapter 3. the difference between the PPZ.and the IPZmodel is only exposed in long time series.
56) and (3.2 s 2. By manipulating the parameters in the IPZmodel. The first part of the time series is used for identification (marked in the figure) and the second part is a pure simulation. to the other part’s disadvantage. The IPZmodel does not capture the low frequency component of the process as well as the PPZmodel does. The estimated PPZ model is G1 ( s ) 15. by using the IPZmodel for design. we can. This means that the IPZmodel is adequate and well suited for control design usage.56) and the estimated IPZ model is G2 ( s ) 0.83 s (3.19 shows the openloop Bode plot for the models in equation (3.00421 1 1896 s e s 1 138 s 2. as long as it is reliable in certain frequency ranges.83 s (3. the IPZmodel in general has larger phase lag. the model has to be reliable around the crossover frequency ( the bandwidth of the closed loop system).” Figure 3. there is a large discrepancy for low frequencies but the disparity between the models is negligible in the high frequency region. 72 .Chapter 3. 1999]: “Feedback control is both forgiving and demanding in the sense that we can have good control even with a mediocre model. 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. be sure to have a stable controller even if the process is of PPZtype. the two different process models have similar appearance. the graphical fit can be improved in the validation part of the figure. and it may be bad where the closed loop sensitivity function is small.57) The difference between the two models is apparent.5 1 291s e 1 797 s 1 74. 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).57). We also note that compared to the PPZmodel. Both in the magnitude and the phase plot. Blackbox Models and Controller Structures hours of logged data is used. Loosely speaking. Thus. Also around the crossover frequency.
can portray a good model in an unpromising manner. the levels of the simulated and the measured output may drift apart. 73 . In [Ljung. From an open loop step response it is simple to graphically obtain the parameters of the transfer function. even for a model that is quite good (at least for control purposes). Practically it is more difficult to get a good PPZ model since it has more parameters to identify. Blackbox Models and Controller Structures 10 Magnitude (abs) 3 10 0 10 5 10 0 20 Phase (deg) 40 60 80 100 5 10 3 10 4 10 3 10 2 10 1 10 0 10 1 10 4 10 3 10 10 Frequency (rad/s) 2 1 10 0 10 1 Figure 3.19 Bode plot for the IPZ (dashed) and PPZ (dotted) models.Chapter 3.18 in terms of a pure simulation. “If the model is unstable. We have previously referred to the identification issue and pointed out the advantage with the IPZ model. or has integration or very slow time constants. Let us finally regard another quotation that agrees with the conclusions we have already made. 2004] it is indicated that a model validation as in Figure 3. It is then a good idea to evaluate the model’s predicted output rather than the simulated one”.
21. since they are both acting on the same physical unit. The pressure controller PC primarily uses flash steam (from another group operating at higher pressure).2. the differential pressure.Chapter 3. secondly steam from the compressor. But there is a third controller loop described in Section 2. one loop at a time. shown in Figure 2. The processes are excited by step responses in the control valves. see Figure 3.0395 e 2s s (2.20. and a paper process with a moisture controller. an experiment has been performed on a fine paper machine. 74 .3s 1) 0. The model is found to be 29 s 1 e 1. Blackbox Models and Controller Structures 3.1s 1) 2.6 s yP ydP uP udP (3.13.6 The differential pressure loop The main objective in this thesis is to give a systematic treatment of the moisture control loop. and thirdly live steam from the header. It is easy to realize that there ought to be crossconnections between the pressure and differential pressure.9s 1) 122 s 1 0.0029 e s (3. That involves a steam pressure process with a steam pressure controller. see Figure 3. To examine this.186 0 343s 1 0.7 s s (1. which here will be shortly analyzed. We then have a twoinputtwooutput (TITO) system.58) Steam header 2 PC 3 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.
Chapter 3.21 The step response experiment of the pressure and differential pressure loops. Blackbox Models and Controller Structures 300 yP (kPa) 280 260 240 220 2200 16 14 u P (%) 12 10 8 6 2200 2400 2600 2800 3000 3200 3400 2400 2600 2800 3000 3200 3400 80 ydP (kPa) 70 60 50 40 2200 26 24 u dP (%) 22 20 18 16 14 2200 2400 2600 2800 Time (s) 3000 3200 3400 2400 2600 2800 3000 3200 3400 Figure 3. Taken from a fine paper machine. 75 .58). The identified model is given in (3.
22 A second step response model of a similar process as in Figure 3.Chapter 3. Blackbox Models and Controller Structures 340 yP (kPa) 320 300 280 2000 14 12 u P (%) 10 8 6 4 2000 90 80 ydP (kPa) 70 60 50 2000 2200 2400 2600 2800 3000 3200 3400 3600 2200 2400 2600 2800 3000 3200 3400 3600 2200 2400 2600 2800 3000 3200 3400 3600 8 u dP (%) 6 4 2 2000 2200 2400 2600 2800 Time (s) 3000 3200 3400 3600 Figure 3. Taken from a fine paper machine. 76 .20.
ydP. First. Figure 3.Chapter 3. udP. however. It is then followed by two different controller structures.3 this is accomplished by canceling process dynamics by feedforward in terms of a twodegreeoffreedom controller. but on a different dryer group. However. This is also very clear from the figure. The conclusion is. and ydP and udP are output and input of the differential pressure process. A different type of model for the steam pressure has also been investigated where the integrator is replaced by a pole. This experiment gives equivalent results. however. that the integrator model is sufficient for controller tuning purposes and therefore the natural choice.21. the closed loop properties when using a PID controller are examined. In Section 3.8. The main purpose of this is to avoid resonances in the frequency response. Most conspicuous of this model is that the differential pressure valve. However. it should be noted that in some cases it has shown sufficient to model it as a first order system. see Figure 3. 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.22 shows a similar experiment as in Figure 3. Both are acting on the same system and must. It is not a big surprise that also the differential pressure can be modeled as an IPZ process. In Section 3. respectively. just like the pressure. This interconnection has to be regarded when tuning the PDC controller. is significantly affected by the pressure valve. The differential pressure. respectively. and from here on only the moisture cascade loop will be treated 3. This is a nice property since the pressure controller can then be tuned without regarding the differential pressure. 77 . Deep analysis of the differential pressure loop is not covered in this thesis. as feedback with a simple PID gives. The 2DOF controller is best suited for implementation since the existing DCS today seldom support state feedback controllers.7 Summary In this chapter a blackbox model structure has been presented and some controller structures have been analyzed. good regulation of this is more important and the performance of the PDC loop is of minor importance. has no influence on the pressure. most easily seen in Figure 3. be governed by the same physical equations. since the PC loop has a larger effect on the drying. r y. to some extent.21. yP.4 dynamics are instead cancelled by state feedback. Blackbox Models and Controller Structures where yP and uP are output and input of the steam pressure process.
78 . a multivariable model for the pressure and differential pressure loop has been given.Chapter 3. Blackbox Models and Controller Structures Finally. The main conclusion of this is that the pressure loop can be tuned without regarding the differential pressure.
A nomenclature can be found at the end of the thesis. 1994]. 1999]. since both these variables are possible to measure. it will have the same structure as the IPZ model. but by algebraic manipulations and a linearization. The primary model is a nonlinear differentialalgebraic equation set. 1995] and [Bohlin. The nonlinear model will be further examined in Chapter 8.Chapter 4. Two comprehensive books on physical modeling and model analysis is [Hangos and Cameron. where it is expanded with dynamics for the paper web to give a complete simulation model for a whole drying section. The foundation of the model is simple mass and energy balances. [Bohlin and Graebe. 2001] and [Thomas. Here we will present a first principles model. 2000b]. This class of models is adequate for controller tuning purposes but does not tell anything about the physics that generate the dynamic behavior. initially proposed by [Åström. see [Allison et al. 79 . 2003]. 4 A Physical Model of a Steam Heated Cylinder In Chapter 3. 1997]. A similar approach for a drum boiler has been presented in [Åström and Bell. A Physical Model of a Steam Heated Cylinder 4. an empirical blackbox model structure for the steam pressure in a dryer cylinder is presented. One of the main purposes of this greybox model is to gain insight into which physical characteristics and mechanisms have key effect on the parameters in the IPZ model. For more on greybox modeling. The steam pressure and the cylinder shell temperature are chosen as state variables.
m Tm dt Qm Qp .2a) (4. where Qm (W) is the power supplied from the water to the metal. us (J/kg) and uw (J/kg) are the specific internal energies of steam and water.2c) d mC p .Chapter 4. The energy balances for steam. Also. and qw (kg/s) be the siphon flow rate. A Physical Model of a Steam Heated Cylinder 4. Due to higher resistance for heat transfer. qw hw Qm . m (kg) the mass of the cylinder shell. 2002]. only the steady state gain. [Janson and Nordgren. 1958] found that the temperature variation during one revolution is negligible (less that 0.2b) (4. Qp (W) is the power supplied from the metal to the paper.1 The model Let qs (kg/s) be the mass flow rate of steam into the cylinder. water and metal are d dt d dt s u sVs wu wVw qs hs qc hs qc hs . the energy loss to the air is only a fraction of the energy flow to the paper or fabric. Cp.1b) where no blowthrough steam is assumed. The energy flow to the metal is given by 80 .3°C) and the metal temperature is therefore assumed to be independent of the cylinder rotation. Tm (K) the mean temperature of the metal. (4. let Vs (m3) and Vw (m3) be the volume of steam and water in the cylinder. and in contact with the surrounding air the other part. hs (J/kg) is the steam enthalpy. However.m (J/(kg K)) the specific heat capacity of the shell. For the period of one revolution. The mass balances for water and steam are then d dt d dt sVs qs qc . the cylinder is in contact with the paper or dryer fabric one part of the time.1a) wVw qc qw . (4. hw (J/kg) is the water enthalpy. qc (kg/s) be the condensation rate. (4. Sometimes the blowthrough steam is modeled as a fraction of qs [Karlsson et al. This does not affect the dynamics of the system. and let 3 3 s (kg/m ) and w (kg/m ) be the densities of steam and water.
visualizing the assumption on the temperature profile and the energy flows.5 and 0. For simplicity. The outer surface area of the cylinder is assumed to be equal to its inside area. Fraction is between 0. Acyl (m2) is the inner cylinder area. cylinder shell.1. as illustrated in Figure 4. and Ts (K) the steam temperature. (4. (unitless) is the fraction of dryer surface covered by the paper web.3) and (4.4). Qm sc Acyl Ts Tm . and Tp (K) is the paper temperature. The energy flow to the paper is given by Qp cp Acyl Tm T p . (4. A Physical Model of a Steam Heated Cylinder Paper web Ts Tm Qm Tp Qp Steam Condensate Dryer shell Figure 4. The error is negligible (less than 5%). because the thickness of the cylinder shell is much smaller than the outer cylinder diameter. and paper web. all steam within the cylinder cavity is assumed to be homogeneous with the same pressure and temperature. we make the assumption of a temperature gradient in the condensate layer. 81 .1 A piece of the crosssection of a drying cylinder. From (4.7.Chapter 4.3) where sc (W/(m2 K)) is the heat transfer coefficient from the steamcondensate interface to the centre of the cylinder shell.4) where cp (W/(m2 K)) is the heat transfer coefficient from the center of the cylinder shell to the centre of the paper sheet.
ii. so we replace it by a static model. Therefore. iv. the temperature of the paper. which are a mass balance for the steam. mTm Qm Q p . Observing that the volumes are constrained by Vs + Vw = V.1). A Physical Model of a Steam Heated Cylinder Equations (4. Since the water volume is small we also have V Vs . Tp. Therefore. we find that the system can be described by the equations d dt sV qs qc . and an algebraic equation for the energy flow. Eliminating the variables qc and Qm.5) d mC p .3) are a crude nonlinear model for the steam and condensate system in the cylinder cavity.1) can be eliminated. where V is the total cylinder volume. an energy balance for the metal. the second mass balance in (4. is likely to vary slowly compared to the steam. we assume that Tp is constant (otherwise we would also need an energy balance for the paper web). This because there is a continuous condensation occurring at the cylinder wall. dt 0 qc hs qw hw Qm . we leave out the energy balance (4. (4. To obtain a linear secondorder model.2). i. we make a few simplifications. and (4. the model becomes 82 . First assume that the steam in the cylinder is saturated.Chapter 4.and cylinder dynamics. Summarizing. 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.2a). a static energy balance for the water. In addition. iii. (4. When the inflow of steam is varied. due to the low pass effect of the cylinder shell. sc Acyl Ts Qp cp Acyl Tm Tp . the thermal dynamics of the water is very fast compared to the cylinder. Qm Tm .
qs. is the input. density s and the temperature Ts. A Physical Model of a Steam Heated Cylinder hs d dt sV qs hs qw hw Ts Tm sc Acyl Ts Tm .m m dt qs hs ( p ) qw ( p )hw ( p ) sc Acyl Ts ( p ) Tm sc Acyl Ts ( p ) Tm . cp Acyl Tm (4. cp Acyl Tm where the states are pressure p and mean metal temperature Tm. hw. Linearizing around the equilibrium gives 83 . are all functions of the pressure p.6) Tm Tp . cp Acyl (4. d mC p .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. (4. The steam inlet flow. (4. The model can thus be written as hs ( p )V d s dp dp dt dT mC p . mTm dt sc Acyl cp Acyl Assuming that the steam in the cylinder is saturated.7) Tp . The equilibrium gives the relations 0 0 0 qs hs ( p 0 ) qw ( p 0 )hw ( p 0 ) 0 0 sc Acyl Ts ( p ) Tm sc Acyl 0 Ts ( p 0 ) Tm . 0 T p0 .8) Hence T p0 0 Tm 0 0 q s hs ( p 0 ) q w hw ( p 0 ) .9) The numerator in (4.Chapter 4. enthalpies hs.
2000]. A Physical Model of a Steam Heated Cylinder hs ( p 0 )V d s dp 0 qs p p0 d p dt qw ( p 0 ) dhw dp hw ( p 0 ) dqw dp sc Acyl dhs dp dTs dp p p p0 (4. dTs dp p p p0 sc Acyl mC p .m d Tm dt sc Acyl Tm cp Acyl Tm .11b) the model becomes hs ( p 0 )V d s dp p p0 d p dt sc Acyl dTs dp p p p0 sc Acyl Tm hs ( p 0 ) qs .10) sc Acyl Tm hs ( p 0 ) qs .12).Chapter 4.11b). However. m d Tm dt sc Acyl dTs dp p p p 0 sc Acyl Tm .8) have been used to simplify the expression. dp (4. (4.11a) and cp sc . mC p . where the states are expressed in terms of deviations and the equilibrium point in (4. it is probably unsatisfying to use model (4. For the lower region of that range. Assuming that 0 qs dhs dp qw ( p 0 ) dhw dp hw ( p 0 ) dqw dp sc Acyl dTs .12) Assumption (4. The interface between the cylinder and paper then acts as a large heat transfer barrier. (4. it is found that the right hand side is 2 20 times larger than the left hand side in (4. experiments show that large values of cp only occur for high moisture contents ( > 40 %) and high fabric tensions. From experimental values in [Karlsson. It is therefore the belief of the author that this is an uncommon situation and the right hand 84 .11b) is another way of saying that Qp is varying much more slowly than Qm when the inlet steam flow is changed.
(4. (4. The steam properties are here assumed to in given in their equilibrium values.m 1 d V s dp .11b) is generally much larger that the left hand side. and also in Chapter 8.14) D 0. we find that x Ax B qs . 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 G ( s ) b1 s ss a 22 a11 a 22 b1 s z ss b1 z s z s . Acyl d hsV s dp . The inequality in (4.4 and Chapter 8. Writing the system in standard statespace form. where x p Tm T and A sc dTs dp d hsV s dp dTs Acyl dp mC p . B Acyl mC p .15) where 85 .m Acyl C 1 0.13) y Cx D qs . This is further discussed in Section 4. 0 (4.11a) will be commented and examined later in the simulations. A Physical Model of a Steam Heated Cylinder side of (4.Chapter 4.
either by machine specifications or from a physics handbook which includes a steam table and heat capacities. Note that the model (4. d s dp. dTs dp Heat transfer coefficient sc All parameters. Therefore it is used to fit the model to the measured data.15) has an IPZ structure. except the heat transfer coefficient. The following assumptions have been made in the development of the model No blowthrough steam The steam in the cylinder is saturated Paper temperature is constant The thermal dynamics of the condensate is fast compared to the cylinder shell The condition (4. z d s V dp a22 sc Acyl mC p . and is very difficult to predict. sc.m Area of the cylinder surface Acyl Steam properties hs .Chapter 4. The heat transfer coefficient depends on both amount of condensate and its degree of turbulence. For large s the transfer function (4. A Physical Model of a Steam Heated Cylinder b1 1 . a11 a22 sc dTs dp d hsV s dp Acyl (4.15) is approximated by .m . Note that it is only the last two items in the parameter list that depend on the operating point.11) The pole coefficient 86 and the zero z are both proportional to the heat transfer sc. are known beforehand. The essential parameters of the model are Cylinder volume V Cylinder mass m Specific heat capacity of metal Cp.16) Acyl mC p .
20) Note that the velocity gain. Finally.Chapter 4. A Physical Model of a Steam Heated Cylinder G ( s) where b1 does not depend on approximated by sc. Therefore neither the initial part of a step response nor the slope of the asymptote depend on the heat transfer coefficient. b1 . T1 . dTs d s sc Acyl mC p . in practice. As can be noticed. m hs . that the time delay of the system often is close to the sampling time. dTs d Vhs s dp dp (4.19) does not depend on sc. s (4. Moreover. kv.18) hs . which may possibly give rise to an estimation of the deadtime that is larger than the true value. above is not normalized by the measuring or any actuator range. the relations between the blackbox and greybox parameters are kv T2 mC p . dTs d s Vhs sc Acyl mC p . the greybox model does not explain the time delay often seen in the blackbox model. m dp dp (4. m b1 z . we have seen from the derivation of the greybox model that there are neglected dynamics in the model. m hsV s dp . s (4. m Vhs dp dp d mC p . 87 . It is important to remember that we often are dealing with sampled measurements and it has been observed.17) For small s the transfer function can be G(s) where b1 z mC p .
and Q the energy flow. Since kv is a velocity gain. Q is constant along the path. 88 . one would expect its unit to include seconds. which also corresponds to the solution given by the distributed heat equation (with constant energy flow to the paper as boundary condition).Chapter 4.23) Writing the quotient k/ x as a heat transfer coefficient . x (4. while kv is given in Pa/kg.3) can be derived as the solution of Fourier’s law of heat conduction Q kA T . Remark 2 The energy flow to the metal (4. By combining the energy balance for the metal in (4.21) where k is the thermal conductivity.21) from point “0” to point “1” x1 T1 Qdx x0 kA dT T0 (4.3).20) shows that T1 and T2 are given in seconds. we have the relation in (4. T the temperature. see Appendix C (Figure C. A is the area.2c) with Qm sc Acyl (Ts Tm ).22) or equivalently Q kA T0 T1 x1 x0 k A(T0 T1 ).5). However.3) we get a firstorder system from steam temperature to mean metal temperature. Assuming there is no buildup of heat at any point along the path of the heat flow. Consider a onedimensional and homogeneous system. A Physical Model of a Steam Heated Cylinder Remark 1 A unit analysis of the expressions in (4. and integrate (4. the inflow of steam is given in kg/s and the time unit is therefore canceled. 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 buildup. (4. The negative sign indicates that the temperature gradient is in the opposite direction of the energy flow. (4.
13). both the mass flows and energy flows are equal.Chapter 4. given by (4.1) and (4.18). The heat transfer coefficient has a considerable influence on both the gain and phase in the midfrequency range. The energy flow to the metal. powered by the latent heat of vaporization of the steam.24) qc (hs As expected.6 m3 Cylinder mass: m = 7610 kg Cylinder area: Acyl = 37. The drying is.17) and (4. For the purpose of designing a PIDcontroller. this difference in gain and phase influences the controller parameters. 4. is given by the condensation rate times the difference in enthalpy due to condensation. Otherwise it would be possible to tune the controller solely from cylinder 89 .25 kg/s.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. Qm. we will look at the Bode plot and a step response. A Physical Model of a Steam Heated Cylinder Remark 3 The steadystate solution to (4. both at high and low frequencies. Figure 4. This value is obtained by simply dividing a measurement of the total machine steam consumption by the number of cylinders. in other words. as shown in (4.m = 500 J/(kg K) Steam properties for the given operating point The nominal steam mass flow rate to each cylinder is approximately 0. running at an operating point with a steam pressure of 90 kPa (gauge pressure).2 m2 Heat capacity for cast iron: Cp.2) is qs Qm Qm qc Qp qw hw ) (4. The machine dependent values used for the simulation are Cylinder volume: V = 12. The amplitude gain is independent of sc. The machine dependent parameters are taken from a steam group of a fluting machine. a nominal speed of 450 600 m/min and a basis weight between 110 200 g/m2.2 Time and frequency domain analysis To investigate the dynamic behavior of the linearized model.
we can examine its justification. at any time index. a higher heat transfer coefficient gives a lower steam pressure.2 Frequency properties for different values of the heat transfer coefficient 500 (dotted). It can also be seen that.3. in basic terms. due to a larger heat transfer to the cylinder shell.Chapter 4. Using values from this example. This has also been pointed out in [Nelson and Gardner. since there is a larger heat transfer to the cylinder and a higher condensation rate. at a given time instance. shown in Figure 4. As the figures illustrate. dominates over the effect of a lower steam temperature. In (4. The effect of a larger sc. A Physical Model of a Steam Heated Cylinder 10 Magnitude (abs) 1 10 0 10 1 10 3 10 2 10 1 10 0 45 Phase (deg) 65 90 3 10 10 2 10 Frequency (rad/s) 1 10 0 Figure 4. sc = dimensions and a steam table. 1996].11a) an inequality that depends on the operating point and cylinder dimensions was utilized to make a significant simplification. A greater energy flow through the cylinder shell gives a higher cylinder temperature even though the steam has a lower temperature. and the cylinder temperature depends on both. and 2000 (solid). Apart 90 . a higher heat transfer coefficient yields a lower steam pressure gain. The linearized model has also been simulated with a step in the input signal. 1000 (dashed). The heat transfer coefficient varies between the simulations and the steam temperature is changing with time. It is essential to remember that there are two effects taking place in this pressure temperature process.
11a) is 10 to 20 times larger than the left hand side.Chapter 4.3 Comparisons with plant data To evaluate the accuracy of the greybox model it has been calibrated and validated against measurements from a steam.5 m2 91 .4 m3 Cylinder mass: m = 8300 kg Cylinder area: Acyl = 45. we also need an expression for the derivative of siphon flow rate.and condensate system. The cylinder data is Cylinder volume: V = 18.4 0. qw. and 2000 (solid). A Physical Model of a Steam Heated Cylinder 6000 p (Pa) 4000 2000 0 0 10 20 30 40 50 60 70 80 90 100 0. with respect to the cylinder pressure. 4.2 0 0 10 20 30 40 50 Time (s) 60 70 80 90 100 Figure 4. 1991] and using those. The experiments have been carried out on a paperboard machine and signals have been measured with a sampling time of 1 s. The next section shows that the model has a good fit to experimental data. from the steam properties. The upper graph shows the steam pressure and the lower shows the cylinder temperature.6 T (K) 0.3 Step responses for different values of the heat transfer coefficient sc = 500 (dotted). Some experimental values are given in [Stenström and Svanquist. 1000 (dashed). we find that the right hand side of (4.
were used to find the optimal calibration parameters.25) where dv is a valve constant which will be the second calibration parameter together with the heat transfer coefficient.Chapter 4. cyl cyl (4.m and pem. The heat transfer coefficient through the condensate is then sc = 3340 W/(m2 K). 2000] the relationship 1 sc 1 sc . the functions idgrey.4 shows an open loop response together with the calibrated model.00308 kg/(s %). The optimization method is based on minimizing the prediction error. From [Karlsson. sc. Figure 4.13). Using this valve description. (4.28) 92 . Tm. see [Ljung. (4. a model for a steam valve has to be added to (4. sc . neither measured. dv = 0. 1999]. uc. occurs in the middle of the cylinder shell). namely qs d v uc .m in System Identification Toolbox for Matlab. we need a heat transfer coefficient through only the condensate film. we keep the linearity and IPZstructure in the model. given in (4.27) is given. and the thermal conductivity is 50 W/(m K). and the steam input flow.26) To compare the result with nominal values cited in literature. The calibration parameters obtained are sc = 1820 W/(m2 K).13). A Physical Model of a Steam Heated Cylinder Since the input signal (steam input flow) is not manipulated directly. the cylinder thickness is 25 mm (the mean temperature. qs. To calibrate the model. where cyl is the distance into the cylinder where the temperature is equal to the mean cylinder temperature. (4. A simple approach is to assume a static linear relationship between the controller signal. a steadystate pressure level can only be reached for a closed valve). and cyl is the thermal conductivity of the cylinder shell. In this example. where the control signal and model output are bias corrected (by the linear model.
Chapter 4. 375 p (kPa) 370 365 360 0 55 50 u c (%) 45 40 35 0 200 400 600 800 1000 Time (s) 1200 1400 1600 1800 2000 200 400 600 800 1000 1200 1400 1600 1800 2000 Figure 4.5 Validation of model (dotted) against measured data (solid). A Physical Model of a Steam Heated Cylinder 425 p (kPa) 420 415 0 60 50 100 150 200 250 300 350 400 450 500 550 u c (%) 55 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). 93 .
There is also a difference in the parameters of the transfer functions. in (4. The steam flow is likely to vary a great deal between different drying groups but by comparing the two values.8s Gblack ( s ) 0. The velocity gain in the models is normalized with the measuring range of the pressure gauge. In the greybox 94 . The time delay is equal to the sampling time and comes from the identification procedure (pem. In (4. Nevertheless.254 kg/s. 4.1s s 20. The machine has 93 cylinders. where 2000 is a nominal value. This can be resolved by changing the integrator to a real pole. and the average valve opening gives the steam flow to the particular cylinder in the model. we know that also the second calibration parameter is realistic. 50. so the average steam flow per cylinder is 0. They depend strongly on the condensate thickness and turbulence. the fact that our estimated value from the model is within that range. and can vary between 1000 and 4000 W/(m2 K). which is not shown in the figure to keep it clear. namely 0.5 shows such an evaluation. A Physical Model of a Steam Heated Cylinder In [Karlsson. gives support for the legitimacy of the model. The model output is then compared with the measured steam pressure. during the experiment. it was discussed that in some cases the IPZstructure is not sufficient to describe the pressure dynamics in a steam cylinder.3s G grey ( s ) 0.154 kg/s. since it has more degrees of freedom (three parameters to adjust instead of two).m in System Identification Toolbox for Matlab). The parameter dv. The total mass flow rate of steam.00176 s 21. typical values of the heat transfer coefficient are given. adjusted on the same data set.00243 1 e s 1 1 e s 1 (4. The excitation in the control signal is generated by a series of steps in the set point (closed loop). Figure 4.4 A modified model In Chapter 3.4s 77. The blackbox model is not shown in the figures but it gives a slightly better fit to the data.Chapter 4. 2000].26).29) The greybox model has also been validated graphically by using the control signal values to simulate an output.29) the greybox model is compared with the corresponding blackbox model. to the drying section is 85 ton/h.
This is nice since it would be unsatisfying to have a model that is unstable for some combinations of physical parameters.30) is s 2 (a11 a22 ) s a11a22 a12 a21 . By examining time series where the IPZstructure is sufficient with cases where it is not. Identification of the parameters gives 95 . The characteristic equation of (4. A Physical Model of a Steam Heated Cylinder model this can be accomplished by disregarding the assumption from (4. knowing that all factors in the elements of (4. a21. ) (4. and a22 are the elements of matrix A´. an explanation to the modeling problem is found.m and the other matrices are unchanged.Chapter 4. (4. Closer examination of (4. This has also been verified by simulation of the primary DAE system (4. a12. when it is clear that the process is stable in reality.3) in Modelica. When cp is increased.14) now becomes sc Acyl dTs dp sc Acyl A d s dp dTs sc Acyl dp mC p . The system matrix in (4.30) shows that sc >> cp gives a system with one fast pole and one close to the origin. this assumption might not be valid.11b).30) is a better structure.30) are positive. It can be shown. (4. see Chapter 8.1) (4. that the eigenvalues of A´ are real and negative.31) where a11.m hsV hsV Acyl ( d s dp sc cp .13) is adequate and sc cp means that (4.11b) It was previously discussed that for high moisture contents and high fabric tensions. and it can then no longer be regarded as an integrator. saying that cp sc .30) mC p . the slow pole moves along the real negative axis towards the other pole. If sc >> cp then model (4.
the solution has no complex parts.34) The low frequency properties will be different for the two models however.16). and since the coefficients of the polynomial (4. The relation between the position of the poles and the physical parameters is a bit more complicated than in (4. since (4. m dTs dp d s dp 0.35) 96 .32) a11a22 a12 a21 hsVmC p . 2 a11 a22 2 (a11 a22 ) 2 4 a12 a21 . d s s V dp (4.31) are all positive. For small s the modified model becomes lim G ( s ) s 0 hs ( sc cp cp sc Acyl ) .Chapter 4. the system is stable. (4. For large s we have G(s) 1 . dTs dp (4. m 0. The roots of (4.33) and since both a12 and a21 are positive. It can also be shown that the initial dynamics of the modified model in (4.31) are s1.13) contains an integrator and has no steady state gain. (4.17).30) are equal to (4. A Physical Model of a Steam Heated Cylinder sc Acyl a11 a22 hsV dTs dp Acyl ( sc cp ) d s dp sc 2 cp Acyl mC p .
The relative value is a useful parameter though. Model errors will therefore be included in the calibration parameter and its absolute value might be uncertain. An important thing to remember here is that the model is an approximation of the real process. dryer bars. cylinder diameter. 97 .5 Summary Chapter 3 introduced a blackbox model structure for the steam pressure dynamics. where it is combined with a paper web model. such as the siphon shape and form. Further examination of the model can be found in Chapter 8. This might have effect on the mechanical design of the drying section. Issues like how the heat transfer coefficient. A Physical Model of a Steam Heated Cylinder 4. The intention of the blackbox model is purely controller tuning and the main purpose of the greybox model is to gain insight into the physical laws behind the blackbox model. to acquire an accurate estimate.Chapter 4. The greybox model has been validated by measurement from a paper machine with good results. It would then be beneficial to have a separate pressure meter and mass flow meter installed at the drying cylinder of interest. selection of materials or steam pressure affect the dynamical performance of the system can then be answered. a greybox model. Both have the same structure but different purposes. There is also a potential to make a recursive identification of the heat transfer coefficient for fault detection with respect to condensate evacuation. cylinder dimensions etc. condensate removal. In this chapter an additional model has been presented.
it has been seen that there can be significant disturbances in the drying section. which makes disturbance attenuation an important issue. The tuning rule is based on the four process parameters of the IPZ transfer function and is therefore denoted as IPZtuning. 98 . a new method is necessary. To give the user the option to balance between robustness and performance. the tuning rule has a design parameter. Since this does not fit very well to the IPZprocess.Chapter 5. The idea is to provide operators and control engineers at the paper mill with a simple tuning tool. Many tuning methods for PI and PID control have been proposed previously but most of them exclusively suppose a firstorder system with dead time or an integrating process with dead time. a simple tuning rule for both PI and PID control is presented. It is also compared with a few other design methods. In this section. A Tuning Method for IPZ Models 5. The IPZtuning rule is tested and evaluated on an industrial paper machine. 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. which can be used without any complex optimization calculations. which is a nice property since it is then independent of unit selections. This parameter is derived from the Nyquist stability theory and is dimensionless. 5 A Tuning Method for IPZ Models From practical experience. The design goal is to obtain good load disturbance response.
The formulation of the problem also affects how easily it is solved. The derivation of the IPZtuning rule is based on a design proposal in [Åström. much effort is put on finding a simple relation between the process parameters and the optimal solution of controller parameters. the optimal controller might not even be stable. Due to the form of the constraint. subject to one or a few constraints. so that the user is relieved from the optimization issue. 1 P ( s )C c ( s ) (5. the optimization problem is nonconvex. Many control methods are based on optimization of some vital criteria. the absolute value of the integrated error (IE) of a step load disturbance is minimized.1) Assume a step unit load disturbance and that the set point is zero. 2004]. 1998] and [Åström and Hägglund. since this obviously very much affects the result and it is not necessarily a good controller that comes out of an optimal solution. Optimization methods are an important tool in nearly all engineering domains. However. defined as ki = kc / Ti. Instead. et al. Evaluate the integral of the error by the final value theorem 99 . This can easily be seen by using the nomenclature in Figure 3.1 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). A Tuning Method for IPZ Models 5. Therefore.6 and writing the error as E (s) R( s ) 1 R( s) Y ( s) P( s )C ff ( s ) P( s) D( s) R( s ) 1 P ( s )C c ( s ) 1 P ( s )C c ( s ) P ( s )C ff ( s ) 1 P ( s )C c ( s ) R( s) P( s) D( s ). The idea is to maximize the integral gain. subject to a robustness constraint.Chapter 5. he or she will get an approximate solution from just a few button pushes on a pocket calculator. There exist very powerful numerical tools that solve linear or quadratic programming problems. while nonconvex optimization is much more demanding due to less effective numerics and the existence of several local optimal solutions. If the optimization problem is not correctly formulated. it is important to consider how the optimization problem is set up. By maximizing the integral gain. and the control field is indeed no exception.
see Figure 5.1. This can also be expressed in terms of the sensitivity function. to only maximize the integral gain is not sufficient since there is no guarantee that the control loop is stable. since equally large positive and negative areas cancel each other by the IE criteria. By avoiding the point of instability with a certain distance.2) lim s 0 1 P( s) 2 1 Ti s TiTd s s 1 P( s) kc Ti s Ti P ( s ) Ti s kc (1 Ti s TiTd s 2 ) P ( s ) Ti .Chapter 5. an additional constraint is needed. stability is guaranteed. Therefore. and the smaller R0 is the more aggressive the controller will be. However. R0.1 A characteristic response to a load disturbance on the IPZ process. 100 . A Tuning Method for IPZ Models IE 0 e(t )dt lim s 0 1 P( s) 1 P ( s )Cc ( s ) s (5. The robustness constraint used here is characterized by a circle with its centre at the point 1 in the Nyquist diagram. kc lim s 0 By maximizing the integral gain. see Figure 5. It is crucial add a constraint to the optimization problem to avoid an oscillatory solution. controlled by a PID controller. defined as Positive area Control error Negative area Time Figure 5. The radius of the circle will then be the design parameter.2. this integrated error is minimized if the sign is disregarded.
101 . A common question. 1 L(i ) (5.3) where L(s) is the loop transfer function.. we get Ms 1 . Ms. 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. A Tuning Method for IPZ Models Im R0 Re –1 Figure 5. can be what Ms = 1. 1 L( s ) (5.4) Since 1+L(i ) is the distance from a point on the Nyquist curve to the critical point 1.g. S (s) 1 .2 The robustness constraint of the design method. the shortest distance from the Nyquist curve to the point 1 is thus 1/Ms. is then given by Ms max S (i ) max 1 .2 means in practice. By varying the radius of the circle. The maximum sensitivity. Simulations and familiarization is one answer to that issue. the degree of robustness is changed. e. Therefore. R0 (5. It is also dimensionless which is a nice property. The disadvantage is that it can be difficult to relate to for the unfamiliar user.5) and we can then use Ms as our design parameter.Chapter 5.
The derivative time Td. in order to avoid the Mscircle while achieving a high integral gain. Note that this design method does not give any suggestion about the feedforward parameters or of the controller. affects the curve close to the origin (high frequencies). the integral time Ti affects low frequencies.19). even if the resulting controllers from the two criteria prove to be close. At high frequencies the phase is /2. Otherwise. if we assume no dead time. The complete Nyquist plot then appears as in Figure 5. A Tuning Method for IPZ Models Figure 5. However. 2002] it was shown that for a certain value of the maximum sensitivity (Ms = 1. maximizing the integral gain or maximizing the bandwidth of the closed loop system (from set point to measurement) gave essentially the same controller parameters.3.3 The complete Nyquist curve for the IPZprocess and a PID controller if the dead time is assumed to be zero. In [Slätteke. while the gain kc affects all frequencies equally. there will be the typical circular appearance in the origin. The shape of the Nyquist curve is changed by varying the controller parameters.Chapter 5. Since both the IPZprocess and a PID controller contain one integrator each. the Nyquist curve starts at the phase lag . et al. This does not always hold though. since these parameters have no influence on the sensitivity function or the disturbance rejection. see (3. The design method puts emphasis on disturbance rejection. 102 .2). a few properties of the Nyquist curve are independent of the controller parameters. That means that emphasis is put on both the regulator and servo problem.
This implies that. The result for Ms = 1. The purpose of this is to cover the whole range of models found in the industry. kv and kc belongs together. there is also a structure in the selection. the integrated absolute error (IAE). where each circle represents one set of process parameters and the corresponding optimal controller parameters. T2.1). for example. producing different grades and qualities. in 103 . The advantage with dimensionless parameters is that they are independent of the selection of units. Due to their units. a few other minimization criteria have been proposed in different contexts. since it also is computationally effortless. from L << T2 to L > T2. However. integrated square error (ISE). IE and IAE give similar results. Even though there is some experimental work in finding combinations that give a clear relationship between the parameters. see (3. T1 from 50 to 800. and L are presumed to be positive. A Tuning Method for IPZ Models Remark Apart from IE. T1 is always larger than T2 but cases where T2 is almost as large as T1 are also included in the batch of models. does not affect the plot in Figure 5. it can be modeled as an IPZ process. and integrated absolute time weighted error (IATE). IAE have the advantage to avoid oscillatory responses.g. 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. e. e.2 The IPZ tuning rule for PI control A large number of different IPZ models.g. Whether the unit of the velocity gain kv. This is also noted in [Åström. 5.4. 1998]. Furthermore. In the plot. et al. However. is given in kPa/(% s) or simply s 1. with respect to different Msvalues. if the system is well damped.1 is plotted in Figure 5. if the error is positive. The Msvalues 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. the controller gain and integral time have been drawn in combination with the process parameters to give a linear dimensionless relationship. and T2 from 2 to 400. process parameters T1. Keep in mind that IE = IAE. A disadvantage with ISE is that it penalizes large errors and gives a small but longterm error.Chapter 5. IE is therefore chosen as the optimization criterion.4. The models cover many different cases. nonminimum phase systems are not included. have been tuned according to the method described in the previous section.
8 T i / (T i+L ) 0.5 0.04 0.2 0.7 1.8 0. T2.5 0.4 (T 2 +L /3)/T 1 0.01 0 0 0.9 0.6 0.2 0.3 0.6 0.4 0.07 0. As will be obvious shortly.03 0. Each circle represents one set of process parameters from the investigated batch and corresponding control parameters from the optimization routine. which must results in a smaller controller gain. larger T1 compared to T2 give more lead action in the process. given a 104 .4 The linear relationship between the PI controller parameters and process parameters for Ms = 1.Chapter 5.7 0.0 0. it must be L because of the influence it has on the controller gain.05 kv kc L 0.5 0.06 0.7 T 2 / (T 2 + L ) 0. combination with a parameter that is given in seconds (T1.3 0.9 1 Figure 5.02 0.1 0. Ti.6 0.1. or L).4 0. Moreover. A Tuning Method for IPZ Models 0.
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 Msvalues are 1.1, 1.2, 1.3, and 1.4. The resulting controller parameters are given by L 3, T 0.09 i T1 k v L T2 L 3, T 0.16 i T1 k v L T2 L T2 3, T 0.23 i T1 k v L L 3, T 0.28 i T1 k v L T2
Ms
1.1 : k c
4L
6T2 L , T2 21L 23T2 4 L , 9T2 105 L
Ms
1.2 : k c
5L
(5.6)
Ms
1.3 : k c
L
100T2 17 L , 11T2 94 L 88T2 12T2 15 L . 85 L
Ms
1.4 : k c
L
By looking at the controller gain for the different Msvalues, 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 s 25 Dotted: M = 1.3 s Dashdotted: M = 1.4 s 20 T
i
15 10 5 0 0
10
20
T
30
2
40
50
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 IPZmodel 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] Sensor noise Unmodeled process dynamics Actuator saturation Admissible control signal variations 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 DCSsystems. Thus, this should not impose large restrictions to the usefulness of the method. In Figure 5.7 and Figure 5.8 simulations of an IPZtuned PI controller for different values of Ms and with = 1 or = 0. The process is given by P ( s ) 0.01 (200 s 1) s e . s (20 s 1) (5.7)
With = 0 the control signal smoother and the overshoot smaller for the step in set point compared to the case with = 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 2DOFcontroller 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 counterintuitive phenomenon can also be seen when controlling a pure integrating process with a PI or PID controller.
1 T2
1 Ti
1 T1
Figure 5.6 The root locus for the IPZprocess 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 4 3 uc 2 1 0 0 20 40 60 80 100 120 Time (s) 140 160 180 200
20
40
60
80
100
120
140
160
180
200
Figure 5.7 Closed loop response of (5.7) and an IPZtuned PI controller for Ms = 1.1 (solid), Ms = 1.2 (dashed), Ms = 1.3 (dotted), and Ms = 1.4 (dashdotted). = 1.
1.0 r, y
0.5
0
0 2
20
40
60
80
100
120
140
160
180
200
1.5 uc 1 0.5 0 0 20 40 60 80 100 120 Time (s) 140 160 180 200
Figure 5.8 Simulation with the same process and controller as in Figure 5.7 but with = 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 1 / T1. Remark 3 The feature of the IPZtuning, to give a nonrealizable 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 IPZtuning. 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
lim kc kv 1
1 Ti i
Td i
1 i T1 e i (1 i T2 )
i L
kc kvT1Td . T2
(5.8)
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 seconddegree 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 IPZtuned PID controller for different values of Ms and with = 1 or = 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 (T 2+L/3)/T 1
0.5
0.6
0.7
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 T 2 / (T 2+L)
0.7
0.8
0.9
1.0
1.0
0.8
0.6 Td / T2 0.4 0.2 0 0.2
0.3
0.4
0.5 0.6 T 2 / (T 2+L)
0.7
0.8
0.9
1
Figure 5.9 The relationship between the PID controller parameters and process parameters for Ms = 1.2.
110
N 10 367T2 L 1443L2 ) 1000(T22 111 .9) 361T2 L 1400 L2 ) 1000(T22 3(3T2 L) 26k v T1 L L(835T22 3(55T22 2T2 (3T22 kc Ms = 1. N 10 1000(T22 3(3T2 L) 20k v T1 L L(786T22 214T22 T2 (8T22 kc Ms = 1.1: Ti Td 1000(T22 3T2 L 12k v T1 L L(883T22 3(39T22 T2 (T22 kc Ms = 1. N 10 264T2 L 1359 L2 ) 2T2 L Ms = 1. N 10 (5.Chapter 5.4: Ti Td 851T2 L 1149T2 L 2T2 L 278L2 ) 722 L2 ) L2 ) .2: Ti Td 875T2 L 375T2 L 2T2 L 286 L2 ) 238L2 ) L2 ) . A Tuning Method for IPZ Models kc 3T2 L 22k v T1 L L(472T22 (28T22 T2 (7T22 471T2 L 146 L2 ) 529T2 L 354 L2 ) L2 ) .3: Ti Td 842T2 L 386T2 L 176T2 L 2T2 L 277 L2 ) 241L2 ) 736 L2 ) L2 ) .
5 0 0 20 40 60 80 100 120 140 160 180 200 6 4 uc 2 0 0 20 40 60 80 100 120 Time (s) 140 160 180 200 Figure 5.5 0 0 3 20 40 60 80 100 120 140 160 180 200 2 uc 1 0 0 20 40 60 80 100 120 Time (s) 140 160 180 200 Figure 5.0 r.10 Closed loop response of (5.4 (dashdotted).11 Simulation with the same process and controller as Figure 5. y 0. 112 . 1.7) and an IPZtuned PID controller for Ms = 1. and Ms = 1. = 1 and = 0. Ms = 1.5 1.3 (dotted).2 (dashed).10 but with = 0.0 r. Ms = 1.Chapter 5.1 (solid). y 0. A Tuning Method for IPZ Models 1.
11) where kd = kcTd.14) arctan( T1 ) arctan( T2 ) Figure 5.5 T2 ) 2 L. (5. (5. we have r( ) ( ) kv (1 (1 2 T12 ) 0. It will be used here to relate the parameters suggested by the IPZtuning and the stability boundary. The condition for oscillation is then given by Nyquist’s theorem G (i )C (i ) r ( ) cos ( ) i sin ( ) k c i ki ik d 1. It will later be applied in Section 5.10) and let the controller be given by C ( s ) kc ki s k d s.13) For the IPZ process. (5.12) Identifying the real and imaginary parts. ki r( ) 2 kd sin ( ) . we find that the boundary of the stability region can be represented by kc cos ( ) . when comparing other tuning methods.5 .Chapter 5.12 and Figure 5. A Tuning Method for IPZ Models 5.13 show examples of stability regions and IPZtuning. 2 2 0. 2001].4 Stability regions A nice tool to graphically examine stability is stability regions [Åström and Hägglund. staying inside this boundary does not automatically imply good tuning. r( ) (5. Obviously. (5. The IPZtuning gives parameters rather distant from the boundary 113 .7 also. only that the closed loop is stable. Start by writing the process as G (i ) r ( )ei ( ) r ( ) cos ( ) i sin ( ) .
Chapter 5. A Tuning Method for IPZ Models
4 3 2 1 0 0 5 kc 10 15 ki ki 4 3 2 1 0 0 5 kc 10 15
Figure 5.12 Stability regions for the IPZprocess 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 IPZtuning for the different Msvalues in (5.6). The indications are moving from the origin to larger kc and ki with larger Ms.
10 8 6 ki 4 2 0 0 5 kc 10 8 6 ki 4 2 0 0 5 kc 10 15 ki M s = 1.3 10 15 ki M s = 1.1
10 8 6 4 2 0 0 5 kc 10 8 6 4 2 0 0 5 kc 10 15 M s = 1.4 10 15 M s = 1.2
Figure 5.13 Stability region for the IPZprocess 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 kc ki kd plane. Here, it is visualized in two dimensional plots by looking at the kc ki plane for the optimal kd (constant in each plot) given by the IPZtuning rule.
114
Chapter 5. A Tuning Method for IPZ Models of stability. With higher robustness (lower Msvalues), the tuning moves closer to the origin. With lower robustness, the proposed tuning moves in direction to the maximum kivalue 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 IPZprocess 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 Msvalues. 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 s 7.79 s 1
1.30 s
.
(5.15)
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)
Ms=1.1 Ms=1.2 Ms=1.3 Ms=1.4
Kc 0.56 1.0 1.4 1.8
Ti 7.1 5.8 5.0 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
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 Time (s) 600 700 800 900 1000
200 Pressure (kPa)
195
190
185 800 55 Valve position (%) 50 45 40 35 800
900
1000
1100
1200
1300
1400
1500
1600
1700
1800
900
1000
1100
1200
1300 Time (s)
1400
1500
1600
1700
1800
Figure 5.14 Experiment on a real paper machine showing a closed loop step response, using IPZtuning for a PI controller. Ms = 1.1 (above) and Ms = 1.2 (below). Controller parameters = 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 Time (s)
200 Pressure (kPa)
600
700
800
900
1000
195
190
185 800 55 Valve position (%) 50 45 40 35 800
900
1000
1100
1200
1300
1400
1500
1600
1700
1800
900
1000
1100
1200
1300 Time (s)
1400
1500
1600
1700
1800
Figure 5.15 Experiment on a real paper machine showing a closed loop step response, using IPZtuning for a PI controller. Ms = 1.3 (above) and Ms = 1.4 (below). Controller parameter = 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 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 IPZprocess. It is a wellknown fact that for a firstorder 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 IPZprocess the integral error (IE) is calculated for a few different processes and Msvalues, 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 IPZprocess. The integrated area is more or less reduced by a factor two by derivative action. What happens is that the derivativeterm 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 1.2 1.2 1.2 1.4 1.4 1.4 1.4 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 400 100 50 25 400 100 50 25 40 8 15 2 40 8 15 2 4 1 3 1 4 1 3 1 0.42 1.35 1.74 1.47 0.72 2.29 2.91 2.51 22.9 5.0 13.7 1.9 17.0 3.8 10.5 1.7 0.64 2.05 2.63 2.33 1.10 3.52 4.51 3.92 15.89 3.57 10.15 1.68 10.45 2.42 6.98 1.33 1.88 0.48 1.37 0.45 1.97 0.50 1.45 0.49 54.52 3.70 7.87 1.29 23.61 1.66 3.61 0.68 24.83 1.74 3.86 0.72 9.50 0.69 1.55 0.34
118
Figure 5. acting on the process P ( s ) 0. (5.2.16 Comparison between a PI controller (solid) and a PID controller (dotted) regulating (5. Both controllers are tuned for Ms = 1. set point following is also an important property for the closed loop system.16). However. However. Apart from the disturbance rejection.01 1 50s e s (1 15s ) 3s . in both those cases the control signal is slightly more aggressively used by the PID.17 shows the gain of the transfer function from r to y and also the sensitivity function. A Tuning Method for IPZ Models 1. see Table 5.16) 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.5 0 0 50 100 150 200 250 300 350 400 4 2 uc 0 2 0 50 100 150 200 Time (s) 250 300 350 400 Figure 5. N = 10. The process also reacts faster to a set point change with the PID controller. and a load disturbance starting at t = 250 s.16 shows both a set point change and a load disturbance.0 y 0. Figure 5. Figure 5. or by a more general set point 119 .2.18 shows the corresponding Nyquist curves. There is a set point change at t = 10.Chapter 5. see (3. = 0.20b). which might be a disadvantage due to cross couplings to adjacent steam groups and increased variations in the steam demand to the boiler house. The PID controller gives a somewhat larger peak in the amplitude curve for Gry. The other controller parameters are chosen as = 1. this can be adjusted by the feedforward parameters and of the PID controller.
so this is actually not an issue of the feedback loop. PI solid.18 Nyquist curves for the marked controllers in Table 5.0 0.5 4 10 3 10 2 10 1 10 0 1. Note that the maximum sensitivity is equal for the to controllers as the tuning prescribes.2 0.5 1.2 0 0.0 2. Gny. PID right.0 2.0 1.17 Amplitude diagram for the transfer functions from set point r to process output y.8 1.Chapter 5.4 0.16).4 0.5 1.0 0. if = 1 and = 1.4 0.8 0. filter.17.6 0.5 0 10 1. it can be seen that this extra peak in the sensitivity function 120 .5 1. An often occurring difference between the two optimized Nyquist curves is that for the PID controller the curve touches the Mscircle in two points but only once for the PI controller. and PID dotted.6 0.0 G ny 0.8 1.4 0.8 0.0 0.0 1. Gny is equal to the sensitivity function and Gry is equal to the complementary sensitivity function.2 0. The sensitivity function shows that some disturbances are amplified more by the PID controller.5 1.6 0. and from noise n to process output y.5 0 0.2 and the process in (5.0 G ry 0.2 0 0.5 0 10 4 10 3 10 Frequency (Hz) 2 10 1 10 0 Figure 5.6 0. PI left. Gry.18.0 0. This can also be seen in the sensitivity function in Figure 5. A Tuning Method for IPZ Models 1.5 Figure 5. By looking at the spectrum of pressure disturbances in Figure 2. 1.5 0 0.
see e.g. 1999]. but they mostly suppose a first or secondorder process.g. Nevertheless. and also a bit more overshoot in the set point response for the standard PID structure. 2001]. A Tuning Method for IPZ Models coincides with the variations found in that specific case study. 1993]. et al.7 Comparison to other design methods This chapter has presented a tuning method for a PI and PID controller regulating an IPZprocess. However. and [Poulin and Pomerleau. 2003]. It is therefore difficult to dogmatically say that the one or the other is better. there a few examples. It simply depends on how sensitive the steam system is to disturbances and how severe the crosscouplings between the different steam groups are.g. [Chien. in Chapter 3 a twodegreeoffreedom controller structure with set point feedforward is given that can be combined with the PI or PID controller. et al. or an integrating process. presented in this chapter. and [Skogestad. e. 1996]. Many different tuning methods have been presented previously. and does not give any recommendations on the feedforward settings. The conclusion is that PID control of the IPZprocess is more effective than PI control but it will give a controller with slightly more aggressive use of the control signal. 1992]. if possible). Then the regulation problem and servo problem are separated in a nice way and a design procedure is given for both. [Poulin and Pomerleau. 1952]. [Tyreus and Luyben. [Wang and Cai. However. The IPZtuning. [Zhuang and Atherton. 1953]. 2003]. 1999]. a few design methods found in the literature are examined to see how well they are compared to IPZtuning. [Visioli. 5. Since the IPZprocess is not a selfregulating process. it is not obvious how to match the IPZprocess to a pure integrator and these methods do not always give a satisfactory result. 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. Therefore. [Cohen and Coon. The purpose is not to discredit these methods but to demonstrate the need for a new 121 . and [Chidambaram and Sree. [Ho. In this section. see e. it is most natural to use the latter. is based on stability of the closed loop system and disturbance rejection. which are modifications of the ZieglerNichols method that give reasonable tuning. 2001].Chapter 5. but these methods have no design parameter which is a deficiency.
One is based on frequency response data (a closed loop test) and the other one is a step test method (an open loop test). The frequency method The frequency method uses information of one point in the Nyquist curve. Ti = 6. and the oscillation period is T0.3. This gain is called the ultimate gain. Simply disable any integral or derivative action and increase the controller gain until a stable oscillation is achieved.0. Ti = 8.3 PID: kc = 1.17) The simulations of IPZtuning use Ms = 1.92.2 and Ms = 1. Td = 0. they are used here for comparison due to their widespread use.4. Ms = 1.41. s (1 20s ) (5. Nevertheless. It was presented in 1942 and consists of two different methods.1 ZieglerNichols two tuning methods The ZieglerNichols rules are the most famous tuning rule for the PID controller [Ziegler and Nichols. This point can be found without knowing the transfer function P(s) by a straightforward closed loop experiment. k0. Ms = 1.3.Chapter 5. 1942]. The resulting parameters for the IPZtuning are Ms = 1. Ti = 3. A Tuning Method for IPZ Models tuning method for the steam pressure process.4.2 PI: kc = 0.2 PID: kc = 1. Td = 0.67. while the obtained Msvalue for the other design methods are also given for a comparison. All examples are based on the process model P ( s) 0. 5.3 PI: kc = 0. (5.05 1 100s e s. Remember that these other methods are sometimes based on a different criterion than the IPZtuning.1.7. The controller settings are then 122 .18) Ms = 1. namely the frequency where kcP(i ) pass through the point 1.9. Both these methods are known to give only reasonable performance.49. Ti = 5.
3 (dashed).5 2 1.Chapter 5.3 (dashed).5 0 0 50 Time (s) 100 150 Figure 5. A Tuning Method for IPZ Models 1 y 0. Also shown is the IPZtuning for Ms = 1.5 uc 1 0.5 0 0 50 Time (s) 100 150 Figure 5.2 (dotted) and Ms = 1.5 0 0 50 100 150 2. Also shown is the IPZtuning for Ms = 1.5 uc 1 0.17).5 0 0 50 100 150 2.20 Evaluation of ZieglerNichols frequency response method for PID control (solid line) given the process in (5.19 Evaluation of ZieglerNichols frequency response method for PI control (solid line) given the process in (5. 123 .2 (dotted) and Ms = 1. 1 y 0.5 2 1.17).
1 y 0.3 (dashed).2 (dotted) and Ms = 1.17).5 0 0 50 Time (s) 100 150 Figure 5. 124 .5 0 0 50 100 150 2. A Tuning Method for IPZ Models 1 y 0.Chapter 5.21 Evaluation of ZieglerNichols step response method for PI control (solid line) given the process in (5. Also shown is the IPZtuning for Ms = 1.17).5 0 0 50 Time (s) 100 150 Figure 5. Also shown is the IPZtuning for Ms = 1.5 2 1.5 uc 1 0.5 uc 1 0.3 (dashed).5 0 0 50 100 150 2 1.2 (dotted) and Ms = 1.22 Evaluation of ZieglerNichols step response method for PID control (solid line) given the process in (5.
21 and Figure 5. Figure 5.83.49. Figure 5.21) By letting s = i . Ti = 3. Ms = 2. Ti T0 . (5.7. This is since the simulated IPZtuning has different Msvalues. it is related to IPZtuning.23) (5. Ti T0 .19 shows simulation results for PI control and Figure 5. Ms = 2. and equating the complex and real parts. 1 . which is the point where the response has the maximum derivative. the ZieglerNichols method is known to give oscillatory results and it is no surprise that the maximum sensitivity functions are very high.87. Td 2 T0 . which is presented next. the following controller settings and Msvalues are obtained PI: kc = 2. 8 (5. a and L. (5. see Figure 5.28. The step test method The step test method is based on two parameters of an open loop step test.2 (5. The tangent through this point gives the two parameters of interest.22) As mentioned before. Assume the IPZ process and a P controller. They are achieved by finding the point of inflection. Td = 0.23.55. Ti = 1. In both cases.22 show the corresponding simulations for the step response method. The method now suggests the controller settings as 125 .19) The ultimate gain and oscillation period can also be calculated directly from the transfer function. A Tuning Method for IPZ Models PI: kc and PID: kc 0.97.20 shows results for PID control.6k 0 . The condition for the point of instability is kc 1 sT1 e s (1 sT2 ) sL 1.45k0 .Chapter 5.20) 0. and PID: kc = 3. even though they are not completely comparable.
(5.9 uc .33L.6.17) is examined by a simulation.27) (5. Ti = 2. Ms = 4.5. PI: and PID: kc kc 0. (5. (5. The obtained settings are PI: kc = 3.25) where uc is the size of the step.23 The two parameters used in the ZieglerNichols step response method.26) The controller settings are somewhat more aggressive than in the frequency response method.8 and PID: kc = 4. (5.8. Ti aL 2 L. Like in the case of the frequency method. Ti aL 3.Chapter 5. A Tuning Method for IPZ Models L a Figure 5. 126 . Td 0 . Td = 0.33. This is also confirmed in the simulations. Ms = 3.5 L . Ti = 3.24) 1 .3.2 u c .
65 (5. a test batch of nine different process structures have been used (IPZ not included). 2005] is developed in a similar way as the IPZrule. (5. For many process control system this is too aggressive. if necessary.97. 5.7.7.2 TyreusLuyben’s modified ZN tuning rule The ZieglerNichols tuning rules are derived to give decay ratio of ¼. Ti 3. As a comparison. However. Both the set point response and disturbance rejection is faster than the IPZtuning.7.29) Figure 5. The method is obtained by maximizing the closed loop resonant frequency given a maximum complementary sensitivity function of 2 dB ( 1. However. but it is also less robust (larger Ms) and the control signal is much more aggressive. and the method presumes that the process can be approximated by either a firstorder system or an integrator. Ti = 8. The disturbance rejection is then better for IPZtuning than TyreusLuyben’s tuning rule. Given the system P( s ) k e s sL . with Ms = 1.17) PI: kc = 1.Chapter 5. A Tuning Method for IPZ Models 5. This gives the following values for the process in (5. the TyreusLuyben method lacks a tuning parameter. (5.65. The result is given by PI: kc k0 .26 ). As for the case of ZieglerNichols.1.24 shows the simulation results and the method gives a reasonable good tuning. Ms = 1. Figure 5.22 2.30) 127 . 1992] to derive a more conservative PI tuning rule. Remember that the overshoot should be dealt with the feedforward part of the controller.25 shows IPZtuning for Ms = 1. the maximum complementary sensitivity function is different (in this case 1.2T0 .66.4. The time to reach a new set point is similar for the two methods even though the size of the overshoot is different.3 The AMIGO tuning rule The AMIGOrule [Åström and Hägglund.28) where k0 and T0 is defined as in Section 5. since it is derived for a different process. which led [Tyreus and Luyben.16) from the value prescribed by the method.
2 (dotted) and Ms = 1.24 Evaluation of TyreusLuyben’s method for PI control (solid line) given the process in (5.65 (dotted) (kc = 1.57.0 y 0.5 y 1.17). A Tuning Method for IPZ Models 1.25 Evaluation of TyreusLuyben’s method for PI control (solid line) given the process in (5.0 0.17). 128 . Also shown is the IPZtuning for Ms = 1. Also shown is the IPZtuning for Ms = 1.5 0 0 3 2 20 40 60 80 100 120 140 160 180 200 uc 1 0 0 20 40 60 80 100 120 Time (s) 140 160 180 200 Figure 5. 1.3 (dashed).5 0 0 3 2 uc 1 0 0 20 40 60 80 100 120 Time (s) 140 160 180 200 20 40 60 80 100 120 140 160 180 200 Figure 5.Chapter 5. Ti = 4.09).
3 (dashed). Also shown is the IPZtuning for Ms = 1. Also shown is the IPZtuning for Ms = 1.Chapter 5.17).2 (dotted) and Ms = 1.27 Evaluation of the AMIGO tuning rule for PID control (solid line) given the process in (5.5 2 1.17).26 Evaluation of the AMIGO tuning rule for PI control (solid line) given the process in (5.3 (dashed).5 uc 1 0.5 0 0 50 Time (s) 100 150 Figure 5.5 0 0 50 100 150 2. 1 y 0.5 0 0 20 40 60 80 100 120 140 160 180 200 3 2 uc 1 0 0 20 40 60 80 100 120 Time (s) 140 160 180 200 Figure 5.2 (dotted) and Ms = 1. 129 . A Tuning Method for IPZ Models 1 y 0.
35 .5.8. the controller settings and obtained maximum sensitivity are PI: kc = 1. Td 0. Td = 0. (5.3.36) 130 . We observe the AMIGO rule gives a pleasant performance for this process. Ti kL 0. Ti = 40. compared to the IPZtuning and the controllers also bring back the process output to the set point nicely at a load disturbance.4 L.17. Ti kL 13. Ti = 8.4. T2 (5. L.37) 1 100s e s (1 10 s ) 3s (5. Assuming (5. is equal to the time delay of the true process. and PID: kc = 1. (5.35) (5.27 show the simulations.Chapter 5. By letting k be equal to the maximum slope of a unit step response of the IPZprocess.01 we get the tuning PI: kc = 1.5L.33) The time delay. Ms = 1. (3.31) PID: kc 8L.4.4.17). Ms = 1. However. Ti = 13. There is less overshoot in the set point response.26 and Figure 5. A Tuning Method for IPZ Models the proposed tuning rule is PI: kc 0.34) Figure 5.5) gives k k vT1 . (5. (5.32) The next step is to match the IPZprocess to a pure integrator with time delay. Ms = 1. if we instead let the process be given by P ( s ) 0.2.4.45 .
4. Also shown is the IPZtuning for Ms = 1.0 y 0. (5. It is clear that it is possible to find a good controller with such a general tuning method as pole placement.5.Chapter 5. Td = 1. In Chapter 3. The design method is shown by a few examples. and a PID controller handles the same thing for a secondorder process. The difficulty can be to know where to place the poles to obtain a satisfactory feedback loop. this was introduced by statefeedback and all closed loop poles could be placed arbitrarily. 5.4 Pole placement The idea of pole placement is to find a controller that gives a closedloop system with a specified characteristic polynomial.7. a PI controller can give the characteristic polynomial arbitrary values for a firstorder process.36). 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.28 shows the simulation for (5. 131 .5 1. Ti = 24.37) compared with one case of IPZtuning. A Tuning Method for IPZ Models 1.0 uc 0. Ms = 1.5 0 0 50 100 150 200 250 300 1.3 (dashed).5. and PID: kc = 1. This is not always possible when the process is controlled by a PI or PID controller.28 Evaluation of the AMIGO tuning rule for PI control (solid line) given the process in (5.38) Figure 5.5 0 0 50 100 150 Time (s) 200 250 300 Figure 5. In the general case.
and the equation (k v k c ( L T1 )) 2 4 k v k c T2 0 (5. 1996] a tuning rule for the IPZ process is given. By choosing a double pole. written by Bill Bialkowski.42) By the controller gain kc we have some liberty to choose the position of the closed loop poles.39) where the time delay have been approximated by the firstorder Taylor series. and for small T2 the two are almost equal. It can also be found in one of the exercises in [Sell.43) 132 . let Ti = T2. 1995]. The closed loop system is now given by Gry ( s ) kv kc (1 sT1 )(1 sL) . where the integral time follows the parameter T2 in the sense that larger values of T2 gives larger values of Ti. This implies that the process pole is canceled by the zero of the controller.41) (T2 The two zeros of the denominators are s kv kc (T1 L) (kv kc ( L T1 )) 2 4kv kcT2 . that is derived from simple pole placement. The controller is given by C ( s) k c 1 sTi . s (1 sT2 ) (5. Let the process be given by P(s) k v (1 sT1 ) e s (1 sT2 ) sL kv (1 sT1 )(1 sL) . the contribution from the square root is zero. sTi (5.5.40) Now. This can be compared with the IPZtuning rule. A Tuning Method for IPZ Models NelsonGardner’s pole placement rule for PI control In [Nelson and Gardner. In both references the idea is given in words without carrying out the calculations and the timedelay is disregarded. but not as detailed. 2(T2 kv kcT1L) (5. kv kcT1L) s 2 (kv kcT1 kv kc L) s kv kc (5. see Figure 5.Chapter 5.
This gives us the tuning rule kc 4T2 k v ( L T1 ) 2 .Chapter 5.29 and we can immediately see that this is not a satisfying controller.16. A Tuning Method for IPZ Models 1 y 0.17). Ti T2 . since the Msvalue is fairly small.29 Evaluation of NelsonGardner’s pole placement rule (5.04. Also shown is the IPZtuning for Ms = 1.3 (dashed).46) The simulation result is shown in Figure 5.44) and the double pole is then positioned in 2 T1 L (5. We can also see that it is a rather robust controller. the controller settings are PI: kc = 0.5 0 0 20 40 60 80 100 120 140 160 180 200 3 2 uc 1 0 0 20 40 60 80 100 120 Time (s) 140 160 180 200 Figure 5.17).2 (dotted) and Ms = 1. Ti = 20. (5. Ms = 1.45) Assuming the process given in (5.44) for PI control (solid line) given the process in (5. is fulfilled. (5. Further investigations show that this method gives a quite robust controller with Msvalues 133 .
49) A suitable characteristic polynomial for a thirdorder system is (s 1 )( s 2 2 0s 2 0) (5.48) Gry ( s ) Ti (kv kcT1Td kv kc (1 Ti s TiTd s 2 )(1 T1s ) .1 1. T2 ) s 3 Ti (1 kv kcT1 kv kcTd ) s 2 kv kc (T1 Ti ) s kv kc (5. The disadvantage with this method is the lack of any design parameter but it gives a reasonable tuning.05 for short time delays (L 1). we get the following system of equations 134 . By assuming that the time delay can be neglected. Then the process is given by P( s) kv (1 sT1 ) . the IPZprocess fits into this model structure. Larger time delays or shorter T2 give Ms 1. A Tuning Method for IPZ Models below 1.2. sTi The closed loop system then becomes (5. PID control The poles of a second order process with one zero.47) and the controller is 1 Ti s Ti Td s 2 C ( s) k c .50) By equating coefficients of equal power in s. and relatively fine simulation results. can be placed independently of one another by a PID controller. s (1 sT2 ) (5.Chapter 5.
(5. 4 0 1 (5. E. and might make the system unstable due to neglected process dynamics. Even though any 0. we select to place the poles in 3 .Chapter 5. to give a controller with satisfactory performance. 4T2 3 . From (5.52) the controller settings can easily be calculated for a given closed loop response. and can be realized there should be some relation between the process and the desired closed loop poles.g. This injects much noise into the loop. 0 1 0 T12 T2 0 ( 2T2 2 0 1 2T1T2 T2 0 T1T2 1 2 0 1 T2 1 2T2 0 T1T2 2T1 T1 0 T12 1 0 1) . Td 3 T12 9(T1 T2 ) .53) The solution is then kc 9T2 (T1 k v T12 (3T1 T2 ) 4T2 ) . Ti 4 T2 . can be damaging to actuators. A Tuning Method for IPZ Models kv kc k v k c T1Ti Td 1 2 0 T2 Ti 2 k v k c (T1 Ti ) k v k c T1Ti Td T2 Ti 0 1 2 0 (5. As an example. 1.54) 135 .51) Ti (1 k v k c T1 k v k c Td ) k v k c T1Ti Td T2 Ti The solution is given by 2 0 1 kc Ti Td 2 0 ( 2T1 1 1 0 0 1 1 T1T2 2 0 0 1 1 1 T1 2T1 0 0 2T2 2 0 1 T2 2 1 T1 0 T12 2 0 0 1) k v (T13 1 T1 0 T12 1) .52) T1 . (5. requesting a bandwidth that is too high for the closed loop system gives very high controller gain. 2T1 3 .
Chapter 5. If this is a good tuning depends on how well we can neglect the time delay. Many other methods lack this design parameter. a few different tuning methods have been evaluated to see how well they match the IPZtuning rule. but it had severe implications for the steam producers who could not handle the large variations in demand. Remark Trying to place any of the closed loop poles in 1/T1 gives infinite controller gain. see Figure 5. The retuning gave both better set point following and disturbance rejection. ZieglerNichols tuning rule gives a very fast response but also large variations in the control signal. A Tuning Method for IPZ Models 21 Steam flow (kg/s) Steam flow (kg/s) 0 1000 2000 Time (s) 3000 20 19 18 17 16 15 21 20 19 18 17 16 15 0 1000 2000 Time (s) 3000 Figure 5. Both figures show the total steam usage by the drying section. which is particularly undesirable for the steam distribution system. How large variations that are acceptable is a casebycase matter.5 Some concluding remarks In this section. Large T1. 136 .30 A case study before (left) and after (right) an aggressive retuning of the steam pressure controllers in a drying section. It is always important to not only look at the process output but also the control signal. This matches what was observed with PI control. A few of them give a reasonable tuning in some region of process parameters but only modest performance outside this region. but also on the other process parameters.30. The reason is that they do not fit well to the IPZprocess. All this is the motivation for a specific tuning rule for the IPZ process. and it is therefore vital to have a design parameter to adjust the tradeoff between robustness and performance.7. for example. Controller parameters Ti and Td will have finite values though. tends gives very small kc. since they often are derived from another process structure. To the left is before retuning of the steam pressure controllers and to the right is after. 5.
the proposed controller settings by the different methods for PI control of (5.1 1. A Tuning Method for IPZ Models 2.32. a robustness test is given in Figure 5. showing a simulation of the different methods tuned for (5.5 2. All four settings from the IPZtuning are shown in the figure. the AMIGO tuning rule. the tuning method is moving towards the peak of the stability region. + ZieglerNichols step test method.Chapter 5.0 1. The two ZieglerNichols methods distinguish themselves by grouping together away from the other methods. it seems reasonable that their controller settings are in the vicinity of the IPZtuning.17) controlled by a PI controller. TyreusLuyben’s modified rule.4). Finally. Therefore. The different tuning rules examined in this section are also indicated.5 IPZ 0 0 1 2 TL AMIGO 3 kc 4 5 6 7 Figure 5. ZieglerNichols frequency method. The dashed line shows the contour that the IPZtuning follows. It was previously concluded that the AMIGO tuning rule and TyreusLuyben’s rule give satisfactory controller settings for this set of process parameters. For large Ms values. which gives a much more robust control.31.31 Stability region for the process in (5.5 ki 1. NelsonGardner. see Figure 5. To see the difference between the evaluated tuning methods from another perspective.17) but where the time delay is 137 .0 ZN freq ZN step 0. 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.17) are indicated in the stability region for the process. The example of NelsonGardner’s pole placement is also in that neighborhood but closer to the origin. and IPZtuning (Ms = 1. namely the maximum ki.
the IPZ tuning appears to give the best result when weighing together all the comparisons given in this section.17) is changed from 1 to 2 seconds. To calculate the controller parameters. The process parameters can easily be obtained from an open loop step response. Since the dead time has a significant effect on the method.8 Summary In this chapter.32 Robustness analysis of the tuning methods used in comparison with the IPZ tuning. a tuning method for PI and PID control of the IPZ model has been presented. 5. A Tuning Method for IPZ Models 2 x 10 7 ZN frequency y 2 0 2 x 10 12 ZN step y 0 2 2 0 50 100 TL 150 200 4 2 y 0 50 100 AMIGO 150 200 y 1 1 0 2 0 50 100 NG 150 200 0 2 y 0 50 100 IPZ Ms = 1. 0 50 100 Time (s) 150 200 0 0 50 multiplied by a factor of two. It is based on optimized load disturbance rejection subject to a robustness constraint. good quality dead time estimation 138 .3 150 200 1 y 0 1 1 100 150 200 Time (s) Figure 5.Chapter 5. Both ZieglerNichols methods become unstable and TyreusLuyben is close to the point of instability. This clearly shows the necessity to not only consider performance but also robustness. the method requires the four process parameters of the IPZmodel and a design parameter that is defined by the user. Overall. The time delay in (5.
apart from modeling the steam pressure in cylinders [Forsman. or the more general 2DOFcontroller described in Section 3.3. The IPZ structure can be used in many other process areas. A Tuning Method for IPZ Models is essential. since neither of the other give good result when they are evaluated for different process parameters. the method has been validated and proved to perform well. 139 . Examples of this are the and factors in (3. Since the method is based on optimized load disturbance rejection it can with advantage be combined with a set point filter. Therefore the tuning method has a much wider field of applications than solely drying section control. The advantage with the IPZtuning is that the equations are simple and it is easy to use. 2005].Chapter 5. A few other tuning methods found in the literature have been compared to the proposed tuning method.19). From experiments on several paper machines at different mills. The conclusion is that there is a need for a new method.
A Tuning Method for IPZ Models 140 .Chapter 5.
The paper has moved from right to left in this painting. It shows the dry end of the paper machine at the Fox River Paper mill in Appleton. Dietrich. calender stack. . By courtesy of Fox Valley Corporation. painted by Thomas M. passing through the dryers. and reel. The man is checking the feel and transparency of the sheet for its thickness and formation. Wisconsin.Part 2 Modeling and control of paper moisture in the drying section Papermaking in the early 1900s.
.
143 . However. in combination with the steam cylinders. The exhaust air removes the evaporated water from the paper web while preheated dry air is added to the hood by the supply air.Chapter 6. Traditionally. The two manipulated variables are combined in a midranging structure to control the moisture. The main purposes of the hood are to create a controlled environment for the drying process. and also to establish good working conditions in the machine room. This chapter proposes using the supply air. The proposed control structure is evaluated by simulations of a physical model of the drying section. 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. Trials have shown that the supply air has a fairly large and fast impact on moisture. 6 Enhanced Moisture Control Using the Air System The drying section is enclosed inside a drying hood. improve energy utilization. This suggests that using the supply air would give a closed loop system with higher performance compared to conventional steam pressure control. Enhanced Moisture Control Using the Air System 6. to control the moisture in the sheet. these two variables are used to control the humidity and pressure inside the hood.
Both [Mori. Enhanced Moisture Control Using the Air System 6. and [Reardon. 2000] and [Gaillemard and Johansson.2 The model The model used for simulation and analysis in this chapter and Chapter 7 is developed at the Department of Chemical Engineering. It is assumed that the machine speed is 430 m/min and the dry basis weight is 267 g/m2. et al. 2004] a model that includes different transport phenomena within the paper sheet is presented but only in steadystate. et al. 2000].Chapter 6. In [Ramesh. et al. [Berrada. implemented in Simulink. The model contains approximately 105 states and is capable of dynamically describing moisture gradients and other properties inside the paper sheet. 1997] develop a linear statespace model from physical relations. 1982] develops a static simulator used for analysis of paper production and energy consumption. It is adapted to a board machine at a paper mill in Sweden. In this thesis. 2005b]. 2004] shows a simulation model for drying of coated paper that is implemented in Excel. Many drying models with different approximations and objectives have been proposed in the last decade. An extensive review of drying models up to 1980 with some 130 references is given in [McConnell. A combination of statistical multivariate models and physical models can be found in [Viitamäki. 1980]. 2004]. 1991]. This model is used for simulations in this chapter and also in Chapter 7. It consists of 93 steam cylinders divided in 12 groups. et al. the model is not described in detail but a general overview is given. Lund University. This chapter describes a physical simulation model of a drying section. et al. [Menani. and [Sun. [Skoglund. 1995]. In [Chen. et al. In [Rao. 1994] a simplified dynamic model is developed. Some physical properties are then adjusted to fit the assumption. 1995]. 1998]. [Wilhelmsson. where the whole drying section is modeled as one or only a few large cylinders. 2004] give a physical model but omit the steam system. et al. In [Sadeghi. [Perré. 1955].1 A literature review of drying section models Mathematical modeling of cylinder drying started with the pioneer work by [Nissan and Kaye. 144 . 2005a] and [Karlsson and Stenström. 2000] models based on nonlinear steadystate relations are given. et al. see [Karlsson and Stenström. but neglects the steam system and [Videau and Lemaitre. 2003] and [Sadeghi and Douglas. 2000] different kinds of blackbox models are used. 6.
The model is built on basic physical relations. accumulates within it. Enhanced Moisture Control Using the Air System Output data Input data One cylinder Time Time Figure 6. normally mass or energy). one cylinder at a time is simulated.1. Written as a partial differential equation (PDE) it is given by J k. There are 10 nodes in the thickness direction and 2005 nodes in the machine direction. vapor. in terms of mass and energy balances. t (6. air. and moisture gradients and shrinkage in the thickness direction can be simulated. or is consumed. where one cylinder with a following free draw is simulated at a time. but not the dynamics for the steam inside the cylinder. 2001] or [Sparr and Sparr.1 Illustration of the simulation method. To reduce the requirements for internal memory storage in the computer. The result from a simulation of one cylinder is then used as input data for the 145 . gives balances for water. The continuity equation forms a basis for all the balances.1) where is some extensive system property (properties that are strictly additive. The model is calibrated to steadystate measurements by adjusting two parameters. The model is implemented in Simulink with the ODEs written in Ccode to reduce the simulation time. different transport mechanisms. It states that the amount of property entering a infinitesimal volume element either leaves that volume.1). heat transfer coefficients for the contact between the cylinder and the paper. see [Hangos and Cameron. and fibre in the web. and algebraic equations. see Figure 6.Chapter 6. The cylinder shell and the surrounding air are also included in the model. J is flow and k is net consumption or generation. and for the condensate inside the cylinder. Algebraic relations describing transport for mass and energy in combination with (6. 2000]. The set of PDEs are converted into a set of ordinary differential equations (ODEs) by discretization. This means that properties like gas pressure inside the paper web.
2. From a step response in each group in the drying section of the real machine.2 The model used to simulate the closed loop steam pressure dynamics. parameters of an IPZ model are estimated.3 and Figure 6.4. simulation of the next cylinder. 2000]. see Figure 6. Apart from bringing dry supply air close to the sheet. the IPZ model given in Chapter 3 is used. The areas where the most of the evaporation from the sheet occurs are indicated in the figure. the number of nodes in each simulation is greatly reduced. Modeling of the steam pressure dynamics Since the physical model does not include the steam pressure dynamics inside the cylinders.5 shows an example of how the blow boxes can be configured in a drying section. Modeling of the air system The air system is of special interest in this chapter and is therefore further discussed. as it will be shown later. by following the flow of the paper.Chapter 6. cylinder shell and paper sheet. the blow box also improves runnability by reducing sheet flutter. A blow box is the unit through which the main part of the supply air is distributed. Simulations shown in this chapter still take 1 2 days on a standard computer. a simulation model for each group is obtained that is linked to the physical model described previously. The simulation technique will have some implications for simulation of the closed loop system. It is combined with a PID controller and a valve saturation to give a model of the closed loop steam pressure system. Enhanced Moisture Control Using the Air System pressure setpoint pressure PID IPZ 1 Figure 6. Figure 6. By including the PID parameters. including both steam system. In this way. and the memory requirement is reduced from gigabytes to megabytes. which is roughly 20 30 % of the total air flow in modern machines [Karlsson. This gives a complete simulation model of the drying section. The other part comes from leakage air through the hood. see Figure 6. To model this it is assumed that there is a specific volume of air around each cylinder that is involved 146 . used by the mill for each pressure controller.
450) Range u: (0 .00350(44.100) Loss function = 0.10s+1) .0 s e s(6.98s+1) . The figure shows group 1 8.00 s e s(2.96 s e s(13.00369(78.70s+1) Range y: (100 .23028 Steam group 8 226 224 G(s) = 0.00112(55.37s+1) .0 s e s(7.450) Range u: (0 .3 Graphical output from the modeling tool [Wallén. Enhanced Moisture Control Using the Air System Steam group 1 18 G(s) = 0.1.450) Range u: (0 .1.450) Range u: (0 .21070 196 y 222 y 220 218 194 192 216 214 190 64 u 62 60 700 800 900 Time (s) 1000 1100 1200 28 u 26 24 100 150 200 250 300 Time (s) 350 400 450 Figure 6.1. 147 . 2000] used to identify the steam pressure dynamics included in the simulation model.Chapter 6.01s+1) Range y: (100450) Range u: (0100) Loss function=0.100) Loss function = 0.00270(40.60s+1) . The solid lines are process signals and dotted lines the obtained models (also given as transfer functions).100) Loss function = 0.38311 16 y 86 70 u 68 66 0 50 100 Time (s) 150 Time (s) 200 250 300 Steam group 3 62 60 58 y 56 54 52 50 64 u 62 60 58 400 450 G(s) = 0.31596 Steam group 2 96 94 92 14 y 90 12 88 10 60 u 58 56 500 550 600 650 700 750 800 850 900 G(s) = 0.69s+1) .1.38462 91 48 u 44 40 450 500 550 600 Time (s) 650 700 750 50 u 45 40 0 100 200 300 Time (s) 400 500 Steam group 7 198 G(s) = 0.00178(97.100) Loss function = 0.0 s e s(14.86s+1) Range y: (100 .00345(44.450) Range u: (0 .97s+1) Range y: (100 .00 s e s(14.93s+1) Range y: (100 .100) Loss function = 0.49s+1) .26s+1) Range y: (100450) Range u: (0100) Loss function=0.64s+1) .61s+1) Range y: (100 .31s+1) .20334 Steam group 6 282 278 274 y 270 266 262 G(s) = 0.450) Range u: (0 .0 s e s(3.1.23s+1) Range y: (100 .0 s e s(6.1. The unit of y is kPa (gauge) and the unit of uc is %.00228(90.00158(121.100) Loss function = 0.21842 u 500 550 600 Time (s) 650 700 750 100 150 200 Time (s) 250 300 350 Steam group 5 98 97 96 y 95 94 93 92 G(s) = 0.2.1.41249 Steam group 4 168 166 164 y 162 160 158 50 48 46 44 0 50 G(s) = 0.
42 s e s(21. The solid lines are process signals and dotted lines the obtained models (also given as transfer functions). Figure 6.23s+1) Range y: (100450) Range u: (0100) Loss function=0.100) Loss function = 0.00139(110. The figure shows group 9 12.00324(75.450) Range u: (0 .16867 Steam group 10 258 254 246 y 242 238 234 42 u 38 34 50 100 150 200 250 300 Time (s) 350 400 450 500 y 246 242 238 42 u 38 34 550 600 650 Time (s) 700 750 800 250 G(s) = 0. By courtesy of Metso Paper. The large arrow indicates the direction of the paper and a few blow boxes are indicated by smaller arrows.73s+1) Range y: (100 .4 Graphical output from the modeling tool [Wallén.1.70 s e s(24. The oval areas indicate the region where most evaporation occurs.450) Range u: (0 .94s+1) .00138(108.26s+1) Range y: (100 .5 An example of blow box configuration in both a singletier (above) and twotier (below) machine.2.96s+1) .89 s e s(15.95s+1) Range y: (100 . The active air volume is essentially bounded by machine equipment and fabrics.100) Loss function = 0.450) Range u: (0 .100) Loss function = 0.Chapter 6.1. 148 .35s+1) .30897 250 Steam group 11 326 324 322 y 320 318 316 314 G(s) = 0.16917 58 u 54 50 0 50 100 150 200 250 Time (s) 300 350 400 450 u 52 48 44 500 550 600 Time (s) 650 700 Figure 6.83s+1) .22603 342 340 y 338 336 334 Steam group 12 G(s)= 0.1. Enhanced Moisture Control Using the Air System Steam group 9 G(s) = 0. 2000] used to identify the steam pressure dynamics included in the simulation model. The unit of y is kPa (gauge) and the unit of uc is %.43 s e s(24.00362(74.
7. therefore the control signal shown in the figure has been recreated afterwards from notes taken during the experiment. The air flows are manipulated by variable speed fan motors (some machines have air flow dampers instead). see Figure 6. in the dynamics. this corresponds to a residence time of 18. Because of the instrumentation at the particular mill. while both the moisture and basis weight controller are put in manual mode (the steam pressure controllers are still in automatic). The idea is to prevent runnability problems due to web flutter in case of aggressive use of the supply air. With a nominal total supply air flow of 50 m3/s divided to 93 cylinders.4 m3/s supply air and 70. Both the supply air and exhaust air are changed equally during the step test to maintain the air balance in the hood. The machine speed is 724 m/min and the basis weight is 134 g/m2.Chapter 6. it is not possible to automatically log the signal to the fans. The moisture and 149 . Measurements prior to the experiment showed that 67 % control signal corresponds to 41. a step test has been performed on a board machine. since it is done manually by one person.6. the change in supply air is done 5 10 s after the change in exhaust air.6 A picture of the air system used in the model. Enhanced Moisture Control Using the Air System Evaporation System boundary Cylinder Wet Air Paper Supply Air Exhaust Air Dry Air Actuator Figure 6. In addition.3 A prestudy To investigate the potential to use the supply air as an actuator. 6. The step test is performed on the last third of the drying section (the after dryer).6 s. the dew point of the supply air is manipulated while the flow rate is constant. and the volume was set to 10 m3. see Figure 6.3 m3/s exhaust air. Dry air is mixed with recirculated moist air and in this way.
but it still clearly gives an idea of the potential to use the air system as an actuator for the moisture controller. the steam and air system is combined by midranging control.Chapter 6. 0 0 200 400 dew point are logged in two different systems and are synchronized at a later stage by the time tags from the internal system clocks.2.7 Results of the prestudy where it is shown that the supply air has a large impact on the sheet moisture. The dew point of the exhaust air is also shown. see Figure 6. Therefore. The experiment gives a promising result. The response is distinct. Enhanced Moisture Control Using the Air System Sheet moisture (%) Dew point o exhaust air ( C) Control signal supply air (%) 9 8 7 65 60 55 50 100 50 600 800 1000 1200 Time (s) Figure 6.7.4 Midranging As described previously. it is considered somewhat uncertain to make a dynamic model from the experiment result. midranging refers to control problems where there is one process output and two or more manipulated inputs. see 150 . the change in sheet moisture is rather significant and the time constant is somewhere between 15 25 s if it is modeled as a first order process. Generally. 6. measured at the reelup. This also agrees well with the residence time of the air system in the model described in Section 6.
The last three midranging structures are evaluated in [Allison and Isaksson. [Shinskey.Chapter 6. The 151 . 1998] and [Allison and Ogawa. One solution is to let the small valve return to its midpoint or target value in steady state. the optimal target value for the expensive input is often not the midpoint. A common example in the pulp and paper industry is consistency control of the pulp that is pumped from the storage tower. respectively. 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. When there is a cost tied to the inputs. 1978]. 1998]. This is known as midranging. [Allison and Isaksson. The controller has two actuators.8 The consistency in the pulp tower is typically 8 12 %. 2003]. An agitator helps the mixing. In general. and sometimes in cost. Four types of midranging structures There exist several different alternatives to implement a midranging controller. and r1 is set point for y and r2 set point for u1. and vice versa. Enhanced Moisture Control Using the Air System Pulp tower Water QC Figure 6. The main purpose of the water through the large valve is to make the pulp pumpable. since the small valve often has a smaller operating range and finer resolution. The signals u1 and u2 are control signals for P1 and P2.8. and P2 is a slow process with a large range. Dilution to pumpable suspension takes place at the bottom of the tower by injection of dilution water. the effect of the inputs differs significantly in both range and speed. It is assumed that P1 is a fast process with a small range. and four different structures are described below. a large dilution valve and a smaller valve for fine adjustments. see Figure 6.
The controller manipulating u1 is a P controller. 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. whereas the controller manipulating u2 is a controller with integral action. 2003]. When the integral action in C removes the control error. the decoupling filter CD is seldom included. An exception is [Allison and Isaksson. see Figure 6. The set point r2 is the bias of the Pcontroller. In the literature describing the valve position controller. the only way for C2 to control u1 is through the error r1 y. since the P controller gives a steadystate error which is removed by C through a much slower process P2. Controller C1 controls the output y with input u1. while C2 controls u1 with u2.10 is probably the most common implementation of midranging found in industry today. In practice. The disadvantage with this solution is that P1 is limited to be controlled by a P controller.9. 1978] and [Allison and Ogawa. The filter is not necessary for the function of the control system but it improves the performance. 1998]. the output u1 is the offset of the P controller. Simple Midranging This is probably the simplest way to implement midranging and it consists of only one feedback loop but two controllers in parallel with the same set point. where it is briefly discussed without any details. If the decoupling filter is removed. The decoupling should ideally be chosen as 152 . Valve Position Controller The valve position controller in Figure 6. Enhanced Moisture Control Using the Air System C(s) r2 r1 kc 1 u2 P2(s) y u1 P1(s) Figure 6. the target value for u1 is set. see [Shinskey.Chapter 6. consequently introducing disturbances into y.9 The structure of the simple midranging controller. By adjusting the offset. This gives a slow response to both set point changes and disturbance rejections.
2005]. k1 k 2 (6. Regardless if u1 saturates or not during an upset. a low pass filter. and the transfer function from u2 to r2 u1 is equal to CD.11.3) where k1 and k2 are the steadystate gains of P1 and P2.2) which is realizable as long as the time delay of P1 is shorter or equal to the time delay of P2. In addition. the pole excess of P2 should be at least as large as the pole excess of P1. and three gains. controller C2 brings u1 back to r2 in steadystate. If the ideal decoupling is not realizable.2). The parameter Rm. Enhanced Moisture Control Using the Air System r2 C2(s) u2 CD(s) P2(s) 1 r1 C1(s) u1 1 Figure 6. antiwindup protection in C1 might be beneficial.Chapter 6. is the hybrid midranging [Love. respectively. P ( s) 1 (6. The filter in combination with the direct term creates the midranging function. 1994]. see [Haugwitz. depending on the speed of controller C2. and P1 and P2 have the same unstable zeros. P1(s) y CD ( s) P2 ( s ) . approximations have to be used. However. Hybrid Midranging An approach. This implementation does not require antiwindup protection in C1 since u1 is a controlled variable. two saturations. changes in u2 do not affect y. With (6. The transfer function from the output of the PI controller to the 153 .10 The structure of the valve position controller. et al. It consists of one PI controller. should be chosen as Rm 100k1 . It is assumed that both P1 and P2 have stable poles or integrators. originally used to control pressure of a steam header. see Figure 6.
the controller combines midranging (normal operation) with splitrange (saturated u2) in a nice way. R 0 small small 0 . The purpose of the other saturation block is to immediately engage control signal u2 when u1 reaches its limit.Chapter 6. Enhanced Moisture Control Using the Air System r2 r1 PI Filter 1 Rm 100 100 Rm P1(s) u1 1 100 100 Rm u2 P2(s) y 1 Figure 6. The purpose of the saturation block preceding the filter is to allow an offset in u1 when u2 is reaching its limit. By letting the minimized cost function be given by Hp 1 J (k ) i 0 ˆ r1 (k i ) y (k i  k ) r2 (k i ) u1 (k i ) 2 Hu 1 i 0 Q u1 (k i ) u2 (k i) 2 . be replaced by any SISO controller. The PI controller can.12. 1998] a midranging controller is implemented in the MPC (model predictive control) approach. see Figure 6. without losing the midranging function. R (6.4) and the weighting matrices chosen as Q large 0 . Midrange MPC In [Allison and Isaksson. summation after the filter is in effect a low pass filtered derivative. 0 large (6. By this.5) 154 .11 The structure of the hybrid midranging controller.
The results are compared to conventional moisture control. where z 1 is the backward shift operator).5. in (6. The control signal u1 manipulates the air flow actuator while u2 is a set point for the steam pressure.2. using the model introduced in Section 6.5 Moisture control by midranging the air system A midranging strategy to control the sheet moisture by a combination of both the blow boxes and steam cylinders. the sheet moisture. and the weights Q and R. Enhanced Moisture Control Using the Air System r1 r2 MPC u2 P2(s) y u1 P1(s) Figure 6. Notation (k + i  k) denotes the i stepahead prediction. both for its flexibility and for being the most common approach in industry. one). the control horizon Hu. and two manipulated variables.13. R. and ( = 1 z 1.4). The motivation for not manipulating the steam valves directly is both to preserve the 155 . Hu is the control horizon. The process consists of one controlled variable. where only the steam pressure in the cylinders is used.12 The structure of the midrange MPC.g. Hp is the prediction is the difference operator horizon. The main tuning parameters are the prediction horizon Hp. without losing the midranging function. However. 6. the steam pressure set point and the air flow actuator. see Figure 6.Chapter 6. It is possible to omit the weight on the control increments. The main advantage with the MPC formulation is that it takes constraints into account explicitly and integrator windup is no issue. is evaluated by simulations of disturbance rejection. Since the MPC formulation is inherently discretetime. 6. 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.1 Modeling and control design The valve position controller is chosen as the midranging structure to be used. the cost function is given as a summation. the desired behavior is achieved.
Enhanced Moisture Control Using the Air System ’set point air flow actuator’ ’set point steam pressure’ r2 C2(s) u2 CD (s) P2(s) ’moisture disturbance’ ’moisture’ 1 r1 ’set point moisture’ n u1 1 P1(s) ’air flow actuator’ C1(s) y Figure 6. (6. Since the paper drying process includes a significant transport deadtime. Note that u1 manipulates the air flow actuator while u2 is the set point to the steam pressure controllers.13 The midranging structure used in the simulations where the different variables are also given.14 and Figure 6. (6.7) .04887 e 617. and Hf is a low pass filter with steadystate gain one that determines the closed loop performance.3s 2 35. and are modeled as P1 ( s ) and P2 ( s ) 0.15. The transfer function of an IMC can be written as C ( s) H f ( s) P ( s) 1 H f ( s) P ( s) P(s) . the two controllers C1 and C2 are based on the IMC concept. which is well known as an effective deadtime compensator for a stable process with long timedelays [Morari and Zafiriou.66 s 1 10 s . (6.6) where P is the process model.5s 1 40 s 0. P + is the realizable inverse of the process model. see Figure 6.3.Chapter 6.2548s 0. as described in Section 2. The process transfer functions are obtained from openloop step responses.8) 156 . 1989].1365 e 61. function of the cascade system in the condensate system and for paper quality reasons.
0 7.5 9.8).15 Step response of the steam pressure obtained model (dotted).7). Enhanced Moisture Control Using the Air System 9.5 8. also given in (6. 14 Moisture (%) 12 10 8 6 460 Pressure set point (kPa) 440 420 400 50 0 50 100 150 200 250 300 350 400 450 500 Figure 6.5 Moisture (%) Air flow actuator (%) 60 50 40 30 20 50 0 50 100 150 200 Time (s) 250 300 350 400 Figure 6. also given in (6. The dashdotted line indicates the final value.Chapter 6.0 8. moisture process (solid) and the 157 .14 Step response of the air flow moisture process (solid) and the obtained model (dotted).
This also decreases the paper temperature which makes the energy flow to the paper to increase.14.16. P1 has both faster dynamics and shorter time delay compared to P2. This reduces the evaporation of water from the paper and the paper moisture increases. is sometimes sufficient for the airflow moisture process.9) s where 1 = 1 (nominal value) implies that the closedloop poles are real and placed at the same distance from the origin as the openloop poles. see Figure 6. it has been observed that a simpler model. However. However. (6. a second order filter is chosen in (6. P1 also has a zero.9). The moisture in Figure 6. Therefore. the only term affecting C2 is CD and the controller is tuned for that process. the dynamics in the cylinder is much slower than the airpaper 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.7 and the openloop simulation in Figure 6. Enhanced Moisture Control Using the Air System The model from air flow actuator to moisture. Notice the resemblance between the prestudy experiment in Figure 6. has two complex conjugated poles to capture the dynamics.7 reaches a new steadystate promptly in a manner that does not appear as a process with only one or two real poles. However. P1. the amount of moisture in the air around the sheet is increased. The physical explanation for the overshoot is a combination of both dynamics in the paper and cylinder. and this in turn will increase the evaporation. The closedloop time constant of C2 is therefore related to the fast loop and the IMC filter is chosen as 158 . The purpose of C2 is to slowly restore the signal u1 to its desired value. with only two complex conjugated poles and no zero. Ideally. it is difficult to distinguish the overshoot in the response because of disturbances. In practice. assuming perfect process models. The design parameter for the IMC is the filter Hf. and vice versa. which gives an overshoot in the paper moisture. This can be seen by inserting C2 into the structure in Figure 6.13. it can not be assumed to have an ideal decoupling filter. When the dry air flow is decreased.Chapter 6. and therefore it is important that C1 and C2 do not interfere with each other. For C1 it is chosen as H f 1 (s) 1 1 2 1 . Apart from the two poles. A value of 1 greater than one makes the closedloop system slower. to make the controller C1 realizable in both cases.
Chapter 6. Enhanced Moisture Control Using the Air System
H f 2 ( s) 1 1
2
. s
(6.10)
The value of 2 is set to be significantly larger (at least five to ten times) than 1, to separate the controllers C1 and C2 in frequency. The midranging 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
H f 3 (s) 1 , 1 61.5 3 s
(6.11)
where 3 = 1.5. To obtain an adequate comparison between the two moisture control systems, the midranging 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 1 = 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 openloop poles ( ) and the closedloop double pole (*) of the air flow sheet moisture process. The parameter 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 10 1.5 Sensitivity function
3 2 1
10
10
10
1
0.5
0 4 10
10
3
10 Frequency (Hz)
2
10
1
Figure 6.17 Frequency plot for the midranging 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 2, because of the decoupling filter CD. The bandwidth of the closed loop system with the midranging 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 midranging 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 midranging is therefore not a problem for that specific example. However, the noise distribution should be regarded before implementing the midranging controller on a drying section.
160
Chapter 6. Enhanced Moisture Control Using the Air System Remark The air flow 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 midranging 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 midranging has a significantly better disturbance rejection than the steam pressure control. There is a slight fluctuation in sheet moisture, for the midranging 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 ( 1 = 0.7). When u2 midranges 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 twopole model, see also Figure 6.15.
161
Chapter 6. Enhanced Moisture Control Using the Air System
10.5 Midranging, Midranging, 10.0
1 1
= 0.7, = 1.2,
2 2
= 5.0 = 5.0
Steam pressure control
Sheet moisture (%)
9.5
9.0
8.5
8.0
7.5
0
100
200
300 Time (min)
400
500
600
Figure 6.18 Comparison between the midranging control system ( 2 = 5.0) and steam pressure control. The steam pressure control should be compared to midranging with 1 = 1.2.
10.5 Midranging, Midranging, 10.0 = 1.2, = 1.2, 1
1
= 5.0 = 10.0 2
2
Steam pressure control
Sheet moisture (%)
9.5
9.0
8.5
8.0
7.5
0
100
200
300 Time (s)
400
500
2.
600
Figure 6.19 Moisture in paper for two different
162
Chapter 6. Enhanced Moisture Control Using the Air System Figure 6.19 shows that the response in sheet moisture is practically independent of 2. However, there is a slight difference at the second half of the simulation. Larger values of 2 reduce the fluctuation since the steam system is less aggressive in its attempt to ‘midrange’ 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 2. However, the larger 2 is, the more the control system becomes ‘singleloop’ and the advantages of midranging 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 midranging control as by steam pressure control. In the midranging case, the fast moisture transients are handled by the air system and the steam system is only used to restore the air system in steadystate. 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. 0 100 200 300 Time (s) 400 = 1. Enhanced Moisture Control Using the Air System 63 62 Dew point (°C) 61 60 59 = 1.2. 465 460 Steam Pressure set point (kPa) 455 450 445 440 Midranging.2.2.0 2 2 Steam pressure control 435 500 600 Figure 6.0 = 10.0 = 10. 164 . Midranging. = 1. 1 1 = 5. 1 1 = 5.21 Steam pressure in the last steam group (lead group) when comparison steam pressure control and midranging. = 1.0 2 2 58 0 100 200 300 Time (s) 400 500 600 Figure 6.20 Change in dew point (midranging).2.
there is an under pressure to prevent moist air to leak into the machine room. Experiments in [Forsman and Birgerson. 2000]. the dew point and zerolevel. see Figure 6.22 The air balance in the hood. Also in this case the dew point needs to be measured closer to the sheet to be useful in a cascade control configuration.13. there is a false ceiling. The experiment in Figure 6. Controller C1 then gives a set point to the dew point controller. Remark 3 An alternative control configuration is to let the dew point be the inner part of a cascade loop with the moisture control. compare with Figure 6.5 minutes due to the large volume of air in the hood. is called zerolevel.7 gives a time constant for the actuator dew point process around 30 seconds. unless the dew point is measured close to the paper (in the pocket). To provide uniform air flows around the cylinders.5. 6. The height where the air pressure inside the hood equals the outside pressure. The large unfilled arrows indicate the exhaust air.5 m Figure 6. far from the sheet. The smaller arrows indicate the pressure inside the hood. This would give a slow moisture control loop.Chapter 6. Since the midranging control presented here. gives smaller variations in steam pressure it should reduce this problem. 1999] show that the time constant in the process from air flow actuator to dew point is around 1. 165 .22. However. Normally the supply 2 2. [Karlsson. the dew point is normally measured in the exhaust air. which is included in P1. Enhanced Moisture Control Using the Air System (different pulp layers are separated) problems in board machines.3 Zerolevel and dew point control There are two important variables to control in the air inside the dryer hood. Below the doors of the hood.
to reduce sheet moisture variability. that controls the dew point in the exhaust air through a selector that is shared with set point r2. Enhanced Moisture Control Using the Air System C3(s) r2 > r3 u2 CD (s) P2(s) 1 P2(s) y C2(s) 1 r1 C1(s) u1 1 Figure 6. is shown in Figure 6. A previously developed physical model for the drying section is used in simulations. and the exhaust air to control the dew point [Forsman and Birgerson. 1999]. By adding an extra controller. the air flow actuator set point is increased if the dew point exceeds set point r3. The air system in the blow boxes are used to control sheet moisture.23. A possible solution.6 Summary A new control strategy has been developed for the multicylinder drying section. The output from the selector is fed back to the integral mode of C3 to avoid windup. That means that the exhaust air must be used to control the zerolevel. 6. The dew point is often measured in the exhaust air channel. in the midranging structure. 166 . The proposed control strategy in this chapter uses the supply air to control the sheet moisture. 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. C3. air is used to control the zerolevel. where the performance of the control system is compared with a conventional moisture control. it is possible to set an upper constraint on the dew point.23 By allowing the dew point to adjust the air flow actuator set point.Chapter 6. in combination with the steam 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 midranging control. The physical reason for this is the fast response of the dewatering rate to changes in air dew point. Enhanced Moisture Control Using the Air System The result is that by combining the conventional steam control loop with the supply air system.Chapter 6. Letting 1 = 1. means that the closed loop poles have the same distance to the origin as the open loop poles. these two parameters can be chosen independently but in practice it is recommended to let 2 5 1. Ideally. The performance of the closed loop system is determined by 1 and 2 determines how fast the steam system should restore the air system to its original level. 1 and 2. There are two design parameters in the selected midranging structure. 167 .
Chapter 7. typically several hours or even days. However. the midranging control system was able to significantly reduce the effect of moisture disturbances. The primary function of the control system is then to maintain the desired properties in the presence of disturbances. Feedforward from a Paper Surface Temperature Measurement 7. Due to the long time constants and time delays in the process. The midranging control system proved to be far more efficient than using only the steam pressure as manipulated variable. the space between the press and drying section is often very small and a scanner is also a fairly large investment for the mill. 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. By using a combination of both the fast but lowcapacity air system and the slow but highcapacity steam system. 168 . In Chapter 6. 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. it is shown that the air in the surrounding hood can beneficially be integrated with the steam pressure control system by midranging. Installation of a moisture scanner between the press and drying section is possible and the moisture measurement can be used for feedforward control.
Note that when the first node reaches a moisture content of 10 %. where new feedforward signals have been proposed. The signal is based on the paper surface temperature in a few positions in the machine direction. The general idea is therefore to estimate the position where the paper surface becomes dry by measuring the surface temperature.1 The peak position the position of a dry surface An important physical aspect in paper drying are the concepts of free water and bound water. The control system is evaluated by simulations of the drying section model presented in Chapter 6.Chapter 7. et al. 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. Feedforward from a Paper Surface Temperature Measurement Therefore. The amount of energy required to evaporate the 70 Inlet moisture content: 61.39 % Inlet moisture content: 62. When the paper surface becomes dry the surface temperature quickly increases to a value above 100°C. The possibility to measure the signal is verified by measurements on a real paper machine. Similar work for the wire section. Figure 7. the second node is still at 60 %.1 Moisture content at the first two nodes in the thickness direction. a new signal to be used for feedforward control of the paper moisture content is presented.12 % 60 50 Moisture content (%) 40 Node 2 30 20 Node 1 10 50 60 70 80 90 100 Cylinder no. 2001]. 0 0 10 20 30 40 169 . can be found in [Larsson and Gustafsson. 1998] and [Li. 7.
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. the discretization of the paper sheet model in the thickness direction is set to 10 nodes. the temperature has been measured on the paper machine the model is adapted for. Feedforward from a Paper Surface Temperature Measurement free water is equal to the latent heat of vaporization. and requires more energy to evaporate compared to the free water. 40 45 50 Figure 7.2. A steadystate simulation of the moisture content at the two first nodes seen from the lower surface. By using an IRcamera which gives the surface temperature in a twodimensional window. The first layer reaches the bound water region at cylinder number 25. As described in Section 6. temperature values along a line in the 130 120 Paper temperature (°C) 110 100 90 80 70 60 50 0 20 40 115 110 105 100 Simulated Measured 80 100 60 Cylinder no.2 Comparison between measured and simulated paper surface temperature. is tightly bound to the fibres by hydrogen bonds.Chapter 7. This can be seen since the evaporation rate changes significantly. To validate the model in terms of simulated paper surface temperature. The proposed feedforward structure uses this indirectly by measuring the surface temperature.1. is shown in Figure 7. This means that the paper is more difficult to dry in presence of only bound water. however. The most important section for the feedforward structure in enlarged and the 100°Clevel is indicated. The bound water. 95 90 85 80 75 Simulated Measured 15 20 25 30 35 Cylinder no. The same event happens to the second fibre layer at cylinder number 43. 170 .
there can not exist any free water on the surface since it would instantly evaporate to the surrounding air.2. This can also be considered from another viewpoint. the absence of free water will cause the paper temperature to increase. The layer of surface fibres without any surrounding free water show totally different properties when it comes to both evaporation and heat transfer mechanisms. Observe that this is a pure validation without any adjustment of the model. given the same external conditions. Since the evaporation requires a large amount of energy.3 Schematic picture that shows the changed properties as the paper surface dries out. machine direction are singled out and the results are shown in Figure 7. the temperature is only measured in the free draw. An illustration of the phenomenon is shown in Figure 7. The simulated values show very good agreement with the measured values.Chapter 7. Since the paper in the contact zone is not visible due to the dryer fabrics. at an operating point different from the one where the model was fitted. compared to fibres surrounded by free water. If the surface temperature is above 100°C. 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. 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. Feedforward from a Paper Surface Temperature Measurement Steam Condensate (a) (b) p1 p1 Cylinder shell Different surface temperatures.3. Figure 7. The simulations show both the temperature for the free draw and the part where the paper is in contact with the cylinder. 171 . A wet surface (a) will act differently compared to a dry surface (b).
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. The steam pressures are also shown in Figure 7. otherwise the observed high temperatures are not possible. Feedforward from a Paper Surface Temperature Measurement (a) (b) Suggested measuring position Figure 7. 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. The paper surface temperature signal is reliable and showed little variation during the time of the measurements. which corresponds to the first and second node reaching the bound water region. suggested by the model. The differences in temperature between 172 .4 and Figure 7.4 Description of the suggested measuring position. By measuring the temperature in the middle of a free draw in a number of positions a temperature curve is obtained. It can therefore serve as an indirect moisture measurement. 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. see Figure 7.5.Chapter 7. Measurement of the feedforward signal therefore requires that the paper surface dries out in the middle of a steam group. 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. The measurements thus confirm the large moisture gradients in the thickness direction. Within each group. It is important to measure on the same side of the paper surface since the drying often is uneven in the thickness direction.2 indicate that the paper surface moisture content is at a low level at cylinder number 27. The paper surface temperature measurements shown in Figure 7.5 to mark the different steam groups.
39 % Free draw temperature difference (°C) 4 3 2 1 0 Changed peak position 1 5 10 15 20 25 30 35 40 45 Cylinder no.6 A temperature curve calculated by taking the difference between two adjacent temperature measurements. The steam pressures in the different groups are also shown.Chapter 7.4. 5 w in = 62. Feedforward from a Paper Surface Temperature Measurement 150 140 130 Paper surface temperature (°C) 120 110 100 90 80 70 60 50 5 10 15 Group 2 Constant temperature zone 20 Sudden increase in temperature 150 Group 3 200 Constant temperature zone 250 Steam pressure (kPa) Group 4 300 25 30 Cylinder no. 35 40 45 100 50 Figure 7.4.5 The temperature curve obtained by measuring the surface temperature as indicated in Figure 7. 173 . as indicated in Figure 7.12 % w in = 61. Figure 7.
the measuring depth could cause problems if several fibre layers are visible by the measuring equipment. the surface may still have dried out just before (a). 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). 174 . The paper surface temperature differences were fitted to cubic splines and the position the maximum of these splines (originated from a driedout surface) will hereafter be called “peak position”. The measuring devices must e. 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. withstand the humid environment inside the hood and problems with fibre coating on the device must be dealt with. To measure the peak position.Chapter 7. 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 heatup periods). The maximum of the temperature difference shows where the dry surface causes maximum effect and not the exact location of a dry surface. 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. Two different cases with different inlet paper moisture ratios are shown in Figure 7. The effect of the dry surface is however radically increased as the paper reaches (b). Feedforward from a Paper Surface Temperature Measurement two adjacent measurements are shown in Figure 7. During the experiments it was also observed that both the viewing angle and distance to the paper is important for the result. Here. The peak position may however be difficult to measure online with the current available measuring technique. For instance. an IR camera should be positioned in the free draws around the position where the surface becomes dry. the chosen position is the middle of the free draw.6.g. however. is of minor importance as long as it indicates changes in paper moisture. It is crucial to measure at the same position in each free draw to obtain relevant comparisons of the temperatures. 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). Also.6. Note that there might well be a difference in peak position and the position where the paper surface actually dries out. The physical meaning of the signals.
0 0 50 100 150 Time (s) 200 250 300 Figure 7.13. Since 11 Sheet moisture (%) 10 9 8 7 0 50 100 150 200 250 300 26.7 shows a somewhat simplified block diagram of the midranging control system together with the feedforward. 7. 175 .0 25.1).8 The experiment used for identification of H. H is the process from peak position to moisture content.) 26. see Figure 6. Feedforward from a Paper Surface Temperature Measurement r2 C2(z) u2 CD (z) P2(z) Hff (z) ”peak position” 1 r1 C1(z) 1 u1 P1(z) H(z) y Figure 7.Chapter 7.2 Design of a feedforward controller Figure 7.5 25. The solid lines are the result of the simulation and the dotted line is the obtained simplified model given in (7. Hff is a feedforward block and the definition of the other parts are given in Chapter 6.5 Peak position (cyl.7 A simplified block diagram of the control structure used in the simulations.
To excite the peak position a step change in inlet moisture content to the drying section is performed. Feedforward from a Paper Surface Temperature Measurement ”Temperature peak position” Hff. The result is shown in Figure 7. The air system is divided into five groups so that the disturbance compensation from the feedforward “follows” the paper through the machine. and a step disturbance is introduced in the moisture content at the inlet of the drying section. the moisture controller is put in manual mode (while the steam pressure control is in automatic mode). 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.9 A block diagram of the feedforward structure used in the simulations.4 the simulation model is very large (~105 states) it needs to be simplified before it can be used for control design.599 z 2 . H y P1. More specifically.1) The feedforward can be implemented as described above but it is slightly modified.5 Figure 7. The identification of the process and the design of the control system is done in discrete time.Chapter 7.3 P1.185 z 6 1 1. 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 .3 Hff. To identify the process H. 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.4 u1 Hff. This is done by system identification of a low order model.8 and the obtained discrete model of the process is H ( z) 0.5 P1. (7. where the sample time is 5 seconds.528 z 1 0.
the condition H P .0 Sheet moisture (%) Air flow actuator (%) 8.9 shows the configuration. 4 3P . (7.5 3P .0 50 40 30 20 0 50 100 150 200 Time (s) 250 300 350 400 Figure 7. H ff .2) holds. The two dashed lines are the first two air groups which are not used for the feedforward control.3) 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). 3 1 H . H ff . 4 1 H . feedforward is not realizable for those groups and only the three last air groups were used. 4 H ff .5 1 0. Feedforward from a Paper Surface Temperature Measurement 9. 3P . 5 1 (7. delay of H. Consequently. it is more effective to use the last three air groups.5 8. with the three utilized parts of P1. 4 1 P .3 H ff .3 1 P . Since the impact of the disturbance at y is to be eliminated.10 Step responses from the five air flow actuators to moisture content at the reel up.Chapter 7. Figure 7.3 H . For simplicity the feedforward filters were selected as H ff .5 H ff . This means that when a disturbance is measured. Another possible solution is to let C2 control the mean value of all 177 .
639 z 2 (7.4) 7.3) is simulated with a step disturbance in inlet moisture ratio.0 7. the air groups are identified from step responses. The sheet moisture is shown in Figure 7. 1 1.0 Sheet moisture (%) 9.4 ( z) 1 P .5 0 100 200 300 Time (s) 400 500 600 Figure 7. Just as in the case of H.5 8. shown in Figure 7.3 Simulations The feedforward system in (7.11 Comparison of different control systems. 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 midranged by a steam group. and are given by P .3 ( z ) 1 P .685 z 2 0.11. 11.00214 z 1 1 z .490 z 1 0.00411 0.00224 z 1 4 z .609 z 1 0.Chapter 7.5 No control . 178 . 1 1.00476 0.00537 0.590 z 2 0.00198 z 1 6 z .open loop Steam pressure control Midranging only Feedforward + midranging 9. It is then compared to simulations with conventional steam pressure control and with the previously developed midranging control system in Chapter 6.537 z 1 0. 1 1.0 10.0 8.5 ( z ) 1 0.5 10.10.
5 Dew point ( C) 61 60.5 8.5 59 58.5 60 59. Feedforward from a Paper Surface Temperature Measurement 26.0 8.5 Peak position (cyl.Chapter 7.0 7.5 0 100 200 300 Time (s) 400 500 600 Figure 7.5 8.5 Midranging only Feedforward + midranging 0 100 200 300 Time (s) 400 500 600 o Figure 7.5 Sheet moisture (%) 8.) 26.12 Peak position and sheet moisture during the closed loop simulation. 63 62.0 25.13 Change in dew point. 179 .5 62 61.0 7.5 25.
7 0. Feedforward from a Paper Surface Temperature Measurement 465 460 Steam pressure (kPa) 455 450 445 440 Steam pressure control Midranging only Feedforward + midranging 0 100 200 300 Time (s) 400 500 600 435 Figure 7. for the different control systems.6 0.4 0.1 0 3 10 Steam pressure control Midranging only Feedforward + midranging 10 2 10 Frequency (rad/s) 1 10 0 Figure 7. 0. showing how the disturbances in inlet moisture propagate.15 Frequency plot.14 Comparison of the steam pressure in the lead group by the different control system.5 Amplitude 0.8 0. 180 .3 0.2 0.Chapter 7.
2. After 150 seconds the decreasing rate of the moisture content restoration is reduced for the feedforward system. where the peak position is shown for the same simulation. Remark The implementation of the feedforward does not require a midranging control system. 181 .15.13 and the steam pressures are shown in Figure 7.11 also shows the moisture deviation when all control systems are switched off (open loop). For low frequencies the difference between the systems is considerable. Notable for the two feedforward control systems is that the air flow actuators are not restored to their original positions by the midranging part. This was also noted in Section 7.14. The midranging controller is chosen here to give natural connection to Chapter 6. Feedforward from a Paper Surface Temperature Measurement The most apparent result is that the disturbance is greatly attenuated. it also slightly affects the peak position. all systems attenuate high frequencies similarly.12. The feedforward control shows to be superior. see Figure 7. if desired. For comparative reasons Figure 7. the feedback systems can only attenuate extremely slow disturbances (note that 10 3 rad/s is a variation with a period time of almost two hours). When the control system compensates for the disturbance by increasing the control signal.Chapter 7. The figure is generated by identifying the process from inlet moisture content to moisture content at the reelup and the figure simply shows the Bode plot of the identified model. The different control systems are also compared by evaluating their disturbance rejection in the frequency domain. It might as well be combined with any other moisture control structure. The reason for this can be seen in Figure 7. Due to the low pass characteristics of the process. Multivariable control of the air and steam groups could be introduced to give maximum manoeuvring range to the individual air flow actuators. With the presented technique to obtain the control parameters it is therefore not possible to completely eliminate the effect of disturbances. The dew points in P1. The feedforward signal is thus affected but this dynamic relation from u1 to the peak position in Figure 7. Since there are both slow dynamics and long time delays in the process.7 is neglected.2) the feedforward could probably be slightly improved. The feedforward assumes linear systems while the system is nonlinear. The feedforward can beneficially use its prior information about the disturbances acting on the system.3 are shown in Figure 7. By considering this in (7.
With the addition of the peak position in combination with a process model. 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 reelup. 2002]. The signal is based on measurements of paper surface temperature in several free draw zones. as long as it can be used as a disturbance indication. 182 . The potential increase in performance is. large and the signal may be replaced by any other suitable signal that indicates disturbances in moisture content. This signal would be very suitable for feedforward control. From the studied step disturbance it is clear that the feedforward signal greatly improves the performance of the paper moisture control system. has been presented that can be used for feedforward in moisture control. Feedforward from a Paper Surface Temperature Measurement 7. it is possible to significantly reduce the transients from inlet moisture disturbances. What type of signal that is finally chosen is of minor importance. however.Chapter 7.4 Summary A new measured signal. called the peak position. Equipment that measure the surface moisture content are being developed [Boström. The peak position may be difficult to measure with the current available technique.
and [Fritzson. also called instances. Like any objectoriented programming language. This is necessary for the simulation of feedforward from surface temperature measurements (the peak position). Modelica provides the notions of classes and objects. The core of the model is based on work by [Wilhelmsson. cylinder.Chapter 8. 1998]. 1997]. 8 ObjectOriented 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. and paper. The objective is control of the moisture at the reelup and not to accurately describe microscale moisture variations inside the sheet. 1998]. and [Slätteke and Åström. 2005]. as the fundamental tool. The advantages of such a modeling tool are (i) it is built on a noncausal equation structure (ii) it is possible to create model components that 183 . Therefore a less complex model is developed that is based purely on physical relations. [Mattsson. 1994]. 2004]. It also lacks a physical model for the steam system. where it instead uses a blackbox model. [Persson. The model is built on heat and mass balances for steam. the model contains approximately 105 states which make simulations fairly tedious. The model is implemented in the objectoriented modeling language Modelica. ObjectOriented Modeling and Control of the Moisture 8. However. see [Mattsson and Elmqvist. et al. Properties like inheritance and abstract classes provide a structured approach to implement equations.
(8. qbt the blow through steam. q bt . let Vs and Vw be the volume of steam and water in the cylinder. et al. ObjectOriented Modeling and Control of the Moisture correspond to physical objects in the real world.1) 184 . apart from a few exceptions. 8. A similar attempt is [Bergström and Dumont. Also. et al. The equations for the steam and cylinder process are taken from Chapter 4. 1998] and [Bortolin. The steam and cylinder process Let qs be the mass flow rate of steam into the cylinder. and the commercial process simulator APROS [Silvennoinen. The mass balances for water and steam are then d dt d dt sV s wV w qs qc qc qw . they are all shown here. Finally. 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.1 The model This chapter describes a physical simulation model of a drying section. It is also validated against measurements on a real paper machine. the model is used to evaluate a new MPC strategy to control the moisture that has similar ideas as the midranging structure in Chapter 6. The objective is to develop a nonlinear model that captures the key dynamical properties for a wide operating range. 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. 2004]. [Niemenmaa. et al. 1996]. implemented in an object oriented modeling language. 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. and let s and w be the densities.Chapter 8. For completeness. and qw be the siphon flow rate. qc be the condensation rate. 1989].
Acyl is the inner cylinder area. us and uw are the specific internal energies of steam and water. m Tm dt (q s q c hs Qm q bt )hs q w hw Qp . and is the fraction of dryer surface covered by the paper web. and Ts the steam temperature. and metal are d s u sV s dt d w u wV w dt d mC p . Qp is the power supplied from the metal to the paper. Cp. (8.Chapter 8. water. Empirical models for cp have been 185 .m the specific heat capacity of the shell. (8. the condensate has a turbulent behavior and the heat transfer coefficient has proven to be difficult to model. (8. q c hs . Tm the mean temperature of the metal. In Chapter 4. Therefore sc is used as a free variable to calibrate the model with. (8. and the number of spoiler bars. m the mass of the cylinder shell. V Vs Vw . machine speed. Qm . sc is used to calibrate the model against plant data. The steam and water volumes add up to the total cylinder volume.3) The power flow to the metal is given by Qm sc Acyl Ts Tm .5) where Tp is the paper temperature.1. 1988] and [Wilhelmsson. 1995].2) where Qm is the power supplied from the water to the metal. both cp and sc are possible candidates for that purpose.4) where sc is the heat transfer coefficient from the steamcondensate interface to the centre of the cylinder shell. see [Pulkowski and Wedel. However. The power flow to the paper is Qp cp Acyl Tm Tp . hw is the water enthalpy. Here. cp the heat transfer coefficient from the cylinder shell to the paper. ObjectOriented Modeling and Control of the Moisture The energy balances for steam. Experiments have shown that sc depends on both condensate thickness. see Figure 8. hs is the steam enthalpy.
Chapter 8. and surface smoothness of both paper and cylinder.1 Steadystate measurements of how the number of spoiler bars affect the heat transfer coefficient for the condensate [Pulkowski and Wedel. ObjectOriented Modeling and Control of the Moisture 3000 15 spoiler bars 25 spoiler bars 30 spoiler bars No spoiler bars Heat transfer coefficient (W/m K) 2500 2 2000 1500 1000 500 0 0 2 4 6 8 10 Condensate thickness (mm) 12 14 16 Figure 8.7) where Cv (m2) is the valve conductance. (8. xv is the position of the valve stem and the function fv is the valve characteristics called valve trim.g. The valve stem varies from 0 (minimum valve opening) to 1 (maximum valve opening). 1999] we have qs Cv f v ( xv ) ( psh p) s . developed. It is well known that cp depends on other things. a valve model is also needed. where cp(0) varies between 200 and 500 W/(m2K). It also depends on machine speed and bar size. e. From [Thomas. From [Wilhelmsson. We use equal 186 . and is difficult to model due to the turbulent behavior. Since the steam flow to the cylinder cannot be manipulated directly.6) is obtained. but this is omitted here. The supply pressure at the steam header is psh. 1995] a linear relation with moisture ratio u is given cp (u ) cp (0) 955u. (8. 1988]. the web tension.
5] 103 . (8.388(ln p ) 2 37.2. Fitting polynomials to the tabulated values for saturated steam in [Schmidt 1969].Chapter 8.5. since it is the most common characteristic in the process industry [Hägglund. and Tm as state variables and using partial 187 .2 0 0 0.2 Equal percentage valve characteristic. 0. [0.9) are a crude nonlinear model for the steamcylinder process. For simplicity. (8. This means that the enthalpy. We also assume that the steam in the cylinder is saturated. density.07402(ln p) 4 2.4 percentage trim. gives Ts hs hw s w 0.8 fv ( xv ) R v = 200 R v = 50 R = 10 v 0.2 0.8 1.0 0.8842(ln p )3 18.71ln p 124.0 xv 0.58(ln p) 2 200 ln p 748. see Figure 8.26] 10 3 . 1991].6 Figure 8.1723(ln p )3 3.43 ln p 1141.8) Rv is a constant known as the “rangeability” since it is the ratio between the maximum and minimum valve opening.9) 260 ln p 1824] 103 . ObjectOriented Modeling and Control of the Moisture 1.6 0. Vw.005048 p 64.792(ln p ) 2 52.77(ln p ) 2 [0. 0. all steam within the cylinder cavity is assumed to be homogeneous with the same pressure and temperature. [ 0. It is given by f v ( xv ) Rv xv 1 .887(ln p)3 39.4 0.1) (8. and temperature are functions of the pressure only.3136(ln p)3 6. Equations (8. By choosing p.
10) Tm ) ( cp (0) 955u ) Acyl (Tm Tp ) f3 . where f1. and f3 the system can be further rewritten into an explicit state form.Chapter 8. dp hs (V Vw ) w hw s (V Vw ) hwVw wVw mC p .12) 188 . Using f1. f2. f2. ObjectOriented Modeling and Control of the Moisture derivatives. e33 (e21e12 e11e22 ) e12 e33 f 2 e33 e22 f 1 . where e11 e12 e21 e22 e33 (V Vw ) w s. e33 (e21e12 e11e22 ) f3 . Vw d w . (8. the system can be rewritten into a third order state equation (most steps are omitted here). e11 dV dp e12 w Cv f v ( xv ) ( psh p) s qw dt dt dVw dp e21 e22 Cv f v ( xv ) ( psh p ) s hs dt dt qbt hs qw hw Tm ) f 2 . dVw dt dp dt dT dt e33 e21 f1 e11e33 f 2 . d s dp d s dp s hs . and f3 are defined as the right hand side of the equations. sc Acyl (Ts e33 dTm dt sc Acyl (Ts qbt f1 .11) In the rewritings of the energy balances above the specific internal energy has been eliminated by the definitions us = hs p/ s and uw = hw p/ w. e33 (8. dp dhs dp d w dp dhw V. (8.
qw.1) % (8. der(Em) = Qm .4 Code segment of the Modelica model of the steam cylinder. equations for steam properties are required to give a complete simulation model. end Cylinder.12). der(Es) = (qs – qbt)*hs – qc*hs.Tp). % (8.3) % (8. The submodel f2 is also opened. der(Mw) = qc . der(Ms) = qs – qc .4) % (8. Mw = rhow*Vw. uw = hw – p/rhow.Tm).5) Figure 8.Chapter 8. V = Vs + Vw. model Cylinder equation Ms = rhos*Vs. 189 . ObjectOriented Modeling and Control of the Moisture Figure 8. Es = rhos*us*Vs. Qm = alpha_sc*Acyl*(Ts . Em = m*Cp*Tm. der(Ew) = qc*hs – qw*hw – Qm.Qp. Ew = rhow*uw*Vw. which in turn contains submodels.qbt.3 A Simulink model of (8. In addition. us = hs – p/rhos. Qp = alpha_cp*Acyl*eta*(Tm .2) % (8.
pv. see Figure 8. The partial pressure pv. we describe how much water is evaporating from the paper surface to the air. Equations (8. pv .13) where qevap is the evaporation rate (kg/m2s).a x p tot . pv. To describe the moisture in the paper we need a mass balance and to describe the paper temperature we need an energy balance. Mw is the molecular weight of water (kg/mole). it is also easier to change the model at a later stage.3.1) (8. ptot the total pressure (Pa). and Tp the paper temperature (K).p the partial pressure for the water vapor at the paper surface. From [Persson. (8. The paper web process We will now expand the model to also include dynamics for the paper sheet. This is given by Antoine’s equation 190 .14) The vapor partial pressure at the paper surface is p v. x 0. 1998] we get the Stefan equation qevap p ptot KM w ln tot Rg T p ptot pv . p (8.g. Starting with the mass balance. p v. and the total pressure.Chapter 8. ObjectOriented Modeling and Control of the Moisture In this form the model can be directly implemented and simulated in e.a the partial pressure for water vapor in the air (Pa). K is the mass transfer coefficient (m/s). Simulink.15) where pv0 is the partial vapor pressure for free water. Rg the gas constant (J/mole·K). p pv0 .4. 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. see Figure 8.a is given by the moisture content of air.5) are put into the simulation environment as they are. x (kg water vapor/kg dry air).62 (8. a .
.16) As long as capillary transport can bring new water to the paper surface. When the paper becomes dryer a correction factor called sorption isotherm.21).15 10.127 p v 0 10 . introduce 191 . The sorption isotherm of a paper web depends on its composition and temperature. (8. see Figure 8.17) where u is the moisture ratio (kg moisture/kg fiber). Hs given by (8.58u 1.18) Figure 8.877 0. . (8. 1993] gives an empirical expression for paper pulp. namely 1 exp( 47.5 Sorption isotherm. It is not very well investigated when compared to other materials [Pettersson and Stenström.0585 ). and g the dry basis weight (kg/m2).10085(T p 273) u 1.Chapter 8. Axy the area of the dryer surface covered by paper (m2). Also. dy the width of the paper web (m).17) and (8.6 shows a schematic picture of the mass flows in the model. Then the mass balance of moisture in the paper web can be written as d (ugAxy ) dt d y v x guin Axy qevap d y v x gu.0 1500 30 C o 60 C o 90 C o Sorption isotherm 1000 0.5 30 C o 60 C o 90 C 0 0 5 10 15 20 25 o 500 0 0 5 10 15 20 25 Moisture content (%) Moisture content (%) Figure 8. 2000]. 1690 T p 43. the vapor partial pressure at the paper surface is equal to the partial pressure for free water. To model the energy balance. and heat of sorption. let vx be the speed of the paper web (m/s). (8. is invoked which has a value between zero and one. but [Heikkilä. ObjectOriented Modeling and Control of the Moisture Heat of sorption (kJ/kg) 1.5.
Cp. 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 Hs (the heat of sorption) is necessary besides the latent heat of vaporization for free water. (8. and Cp. (8.p.w is the specific heat capacity for the paper. Analogously to the discussion about the mass balance. respectively (J/kg·K).p is a weighted sum of the heat capacities of the parts.17). (8. fiber and water.22) 192 . 1995] we have Cp.0585T p2 R g 1 Mw . d (1 / T p ) (8.fiber = 1256 J/(kg·K). Cp.20) and by applying this on (8. Also.6 The mass balance for moisture in the paper web. As we can see.19) where Cp. ObjectOriented Modeling and Control of the Moisture vx d y v x guin Axy qevap d y v x gu Paper web Figure 8.fiber. From [Wilhelmsson. let Tp be the paper temperature and H be the amount of energy needed to evaporate the water. w 1 u . The shaded area is the cylinder wall C p. we get Hs 0. The heat of sorption can be derived from the sorption isotherm by thermodynamic theory and this relation is known as the law of ClausiusClapeyron Hs Rg Mw d (ln ) .Chapter 8.10085u 1. p C p . fiber uC p .21) The amount of energy required to evaporate the water from the surface of the web is the given by H H vap Hs.
18) and (8. The shaded area is the cylinder wall.1). In addition. (8. The energy balance of the paper web is thus modeled as d ( g (u + 1) Axy C p . (8.23).in Paper web d y v x g (1 + u )C p . ObjectOriented Modeling and Control of the Moisture Axy qevap ( ΔH vap + ΔH s ) d y v x g (1 + u in ) C p .Chapter 8. 193 . p T p . we let the heat transfer coefficient from the cylinder to the paper web depend on the moisture content in the web. where ΔHvap is the latent heat of vaporization.2) define the dynamics of a cylinder. equal to 2260 kJ/kg (at atmospheric pressure). (8. it is given by the balances in (8.17). This is physically unrealistic since the bond energy between the last fraction of water and a cellulose fiber must be finite. Equation (8. together with the algebraic relations given in (8. 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. (8. see Figure 8. 2000] investigates some sorption isotherms found in the literature. p T p ) dt = Q p + d y v x g (1 + u in )C p .19).23). and (8. and (8. In addition.13)−(8.9).7 The energy balance of the paper web. which therefore can be neglected.2). By connecting a series of these equations a discretized model is obtained. Reference [Pettersson and Stenström.in (8. p T p . To summarize the complete drying section model. The dynamics of a lumped paper web model is given by (8. p T p .7.18). and one set of these equations are needed for each cylinder in the drying section.3)−(8.23) − Axy q evap (ΔH vap + ΔH s ) − d y v x g (1 + u )C p . The heat loss in the paper due to mass evaporation is dominating the heat conduction and radiation. pT p Qp Figure 8. 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. Many of those give a heat of sorption that goes to infinity as u goes to zero. 1993].22).21). From [Heikkilä. (8.1) and (8.
ObjectOriented Modeling and Control of the Moisture 60 50 Sheet moisture (%) 40 30 20 10 0 5 10 15 20 25 30 35 Cylinder number Figure 8. 2005] and [Persson.2 Steadystate model validation The two partial models in Section 8.8 Validation by comparing steadystate simulations with moisture measurements on a fine paper machine. which can be seen in Figure 8. The heat transfer coefficient sc 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 moisture content is measured in 11 different positions.8 together with the simulation result. 8. 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. 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. 1994].Chapter 8.012 kg/kg. are validated separately in [Slätteke and Åström. The agreement between model and measurements is good even 194 . 1998].1. where the outflow of paper of one component becomes the inflow of the next component. To assure that the combined model is satisfying it has been validated against steadystate data taken from a paper machine producing fine paper [Stenström. et al.
ObjectOriented Modeling and Control of the Moisture Pres s Pope Mois ture controller Header pressure HeaderPres.. 6 30). though this is rough validation without using all possible degrees of freedom. 195 . and the falling drying rate phase [Karlsson..g. showing the cylinders. hence modeled as one control volume.9 The Modelica model of the drying section used in the simulations. The component of the second dryer group is expanded. In the constant rate phase (cyl. steam valve. It is assumed that the temperature and moisture are constant at a single cylinder due to the high machine speed. 1 5) most steam energy is used to heat the web and the evaporation rate is low. F P k={1000e3} D I P PID i Figure 8. In the falling rate phase (cyl. 31 37). drying rate begins to decrease due to the hygroscopic nature of the fibres.Chapter 8. by assuming different sc in different cylinders. and PID controller. the heating phase. 2000]. e. the constant drying rate phase. The model also captures the three zones in the drying process. In the heating phase (cyl. energy to the web is equal to the energy consumed for water vaporization.
10 that for the case with Steam pressure (kPa) 550 500 cp = 500 450 150 Steam pressure (kPa) 140 cp = 4000 130 75 70 65 60 55 100 150 Time (s) 200 cp.6) to define it independently of u.10 response in steam pressure during a step in valve position is shown for two different cases of heat transfer coefficient cp. the machine width is 7. and the dry weight of the paper sheet is 77 g/m2.3 Open loop simulations A simulation model of a fictitious drying section is used in simulations to investigate different properties. 196 Valve position (%) . The machine speed is chosen to 1100 m/min. Figure 8.15 m. It is clear from Figure 8. This is also the drying section model used in Section 8. In Figure 8.10 Pressure response for a step in valve position for two different values of Heat transfer coefficient sc = 4000. see (4.4). ObjectOriented Modeling and Control of the Moisture 8. see Figure 8. It consists of 60 drying cylinders divided in 7 drying groups.Chapter 8.9. sc >> cp.5 to evaluate MPC control of the moisture. In Chapter 4 the IPZ model is deduced from physical relations.3) and (4. It is concluded that a slow pole in the process dynamics can be regarded as an integrator if the heat transfer coefficient of the steamcondensate interface is much larger than the heat transfer coefficient of the cylinderpaper interface. This implies that in this analysis cp is set to a constant value instead of the linear relationship (8.
11 Open loop step response in pressure with coefficient cp is set to a constant value 500 W/(m2·K). However.Chapter 8. ObjectOriented Modeling and Control of the Moisture Mass flow in/out (kg/s) 0. sc = 4000 W/(m2·K). Heat transfer Q p.16 510 Steam pressure (kPa) 500 490 480 470 460 90 100 110 120 130 140 150 Time (s) 160 170 180 190 200 Figure 8. cp = 4000 W/(m2·K).24 0.mT m )/dt d( wu wV w )/dt 250 Figure 8. 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). Note that since the heat transfer coefficients are different in the two figures. Q m (kW) 450 Qm Qp 400 350 Energy derivatives (kJ/s) 250 200 150 100 50 0 50 100 150 Time (s) 200 sc d( su sV s )/dt d( mC p. Heat transfer = 500 W/(m2·K). the steadystate solutions are 197 .20 qs q c+ q bt 0. for the response with cp = 4000 W/(m2·K) an IPZ model is not sufficient.12 Open loop step response in pressure with coefficient cp is set to a constant value 500 W/(m2·K).
0 8. the increasing condensation rate lags behind the increasing steam inlet flow as the condensate layer heats up to the new steady state temperature. Figure 8. it will increase the steam pressure inside the cylinder and consequently the saturation temperature of the steam and the condensation rate.13 Simulation of set point change in steam pressure and a web break.11 gives an explanation for the characteristic appearance of the open loop step response of the steam pressure. 9. and also the energy derivatives. In the derivation of the linear model in Chapter 4. The energy balance is therefore replaced by a static model by defining its derivative as zero. This is also noted in Chapter 3.5 325 320 315 310 305 300 80 60 40 20 0 0 100 200 300 400 500 600 Time (s) 700 800 900 1000 Figure 8. The conclusion is unaffected if the two responses start at the same pressure.5 Sheet moisture (%) Steam pressure (kPa) Valve position (%) 9.12 shows the energy flows to the cylinder and paper. 198 . it is simplified by assuming that the thermal dynamics of the water is fast compared to the dynamics of steam and metal. When the steam valve position is increased. Figure 8. before the steam consumption has reached its new value. ObjectOriented Modeling and Control of the Moisture also different and the two simulations start from different steadystate steam pressures. However.Chapter 8. there will be a fast initial buildup in steam pressure. Therefore.
109 e 46. The pressure is controlled by a PI controller tuned according to the method in Chapter 5 with Ms = 1.25) The difference in dynamics between the two processes is significant and will greatly affect the performance of the closed loop system.2s 1 16.13 shows a simulation where the steam pressure controllers are in closed loop while the moisture is uncontrolled. where the simulation time is counted in days. ObjectOriented Modeling and Control of the Moisture Figure 8.Chapter 8. and the curve is out of the scale in the figure.24) 4000 W/m 2 K (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. This should be compared to the more complex model used in Chapter 6 and Chapter 7. the control valve is almost closed to keep the desired steam pressure.12 shows that this assumption is valid. It has been described as one of the most 199 . Since the cylinder shell acts as a thermal filter Qp is increasing more slowly that Qm. 8. Note that the characteristic overshoot in steam pressure does not result in an overshoot in the moisture. Since the main energy sink is lost for the cylinders. Figure 8. The simulation time for the figures in this section is a few minutes. This clearly shows that effective condensate removal and the cylinder design in terms of spoiler bars are important. 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. The valve position returns to its initial position when the set point is reached due to the integral action in the process.6 s sc 0. Figure 8.2 s 1 16.4 Control of moisture by midrange MPC In the last decade. they are given by P(s) and P( s) 0.1s sc 2000 W/m 2 K (8. Modeling the two responses as a first order process. MPC (model predictive control) has received large attention in the process industry.2. At t = 100 s there is a steam pressure set point change.0917 e 21.
200 .4 8.6 = 4000 W/m2K sc Steam pressure set point (kPa) 385 380 375 370 450 500 550 600 Time (s) 650 700 750 800 Figure 8.Chapter 8.8 5. Setpoint Moisture Setpoint steam pressure Steam pressure Steam system Moisture r MPC uc PIDcontroller Dryer y Figure 8.8 6.2 8.0 sc = 2000 W/m2K 7.0 5.2 Sheet moisture (%) 6.6 Sheet moisture (%) 8.14 Response in moisture to step change in steam pressure set point for two different values of heat transfer coefficient cp. ObjectOriented Modeling and Control of the Moisture 8.15 Moisture control by single loop MPC.
16. It is implemented in a MPC structure and the analysis is done by simulations of the paper machine model shown in Figure 8. In the Laplace domain. see Figure 2. uc1 is the steam pressure set point to the first 50 cylinders. a more effective moisture control system can be derived. Figure 8.26) where y is the moisture content.15 shows a block diagram where the moisture is controlled by MPC. it is given by Y 0. 2003] is linked to the Modelica environment.16 The proposed moisture control loop. In this way. ObjectOriented Modeling and Control of the Moisture Setpoint moisture uc1 MPC PIDcontroller Steam system First part r1 Dryer PIDcontroller Steam system Last part Moisture y uc2 Figure 8.9. see Figure 8. In this section. 1999] and the only methodology to handle constraints in a systematic way [García. The performance of this closed loop system is limited by the long transport deadtime in the drying section.Chapter 8. The steam pressure set point is distributed to the dryer groups as described in Chapter 2. 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. a new strategy to control the moisture in paper production is evaluated. A Matlab toolbox for MPC [Åkesson. The PIDcontrollers in the inner loops are tuned according to the IPZtuning 201 . et al. By identification of step responses on the highorder nonlinear physical model. significant developments in process control [Doyle. 1989].098 e 48s 1 14 s U c1 0.12. the advantages of the simple modeling technique in Modelica and the rich family of toolboxes in Matlab are used.010 5 s e U c2 . and uc2 the set point to the last 10 cylinders. 48s 1 (8. a simple blackbox model is achieved. By manipulating the steam pressure in the cylinders of the last part of the machine independently of the first part.
The weight on uc1 is larger than on uc2 making the controller to primarily use signal uc2 when acting on disturbances in moisture or on set point changes. In this way. the prediction horizon is chosen to 50 and the control horizon is chosen to 4. obtained correspondingly to (8.Chapter 8.11 e 48s 1 13 s Uc . ObjectOriented Modeling and Control of the Moisture derived in Chapter 5.28) R Q where r1 is the set point for the moisture. since moisture control is the main objective. Q and R.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 steadystate. The prediction horizon is set to approximately match five time constants of the openloop system (to assure that the prediction ‘sees’ the 202 .29) are found from a combination of these ‘rules of thumb’ and evaluation of performance by simulations. R 100 0 0 10 4 .29) There is obviously a larger weight on deviations from r1 than from r2. and r2 the set point for uc2. The response from uc2 has a significantly shorter time delay but also smaller process gain. see (8. This is done by choosing appropriate weighting matrices. For the simulations in this chapter they are chosen as Q 500 0 0 10 3 . the first part of the drying section serves as the base level of the drying while the last part controls the moisture. Notation (k + i  k) denotes the i stepahead prediction and is the difference operator. The MPC settings in (8.27) The cost function being minimized in the midranging MPC is 49 J (k ) i 0 ˆ r1 (k i ) y (k i  k ) r2 ( k i ) uc 2 (k i ) 2 3 i 0 uc1 (k i ) uc 2 ( k i ) 2 . The sample time is 5 s. is given by Y 0. The advantage of the high gain from uc1 and fast dynamics from uc2 is utilized in a midranging MPC structure. Since the MPC formulation is inherently discrete. (8. The simplified model for the singeloop case in Figure 8. (8.26).15. (8. The idea of (8. the cost function is given as a summation.28).
0 % to 60.17 Simulation of midrange MPC.18 Simulation of single loop MPC.2 7.0 450 440 430 0 200 400 600 Time (s) 800 1000 1200 Figure 8.5 % at t = 600 s.0 y (%) u c1 (kPa) u c2 (kPa) 450 440 430 600 500 400 0 200 400 600 800 1000 1200 Time (s) Figure 8. ObjectOriented Modeling and Control of the Moisture 7. 203 .6 7. A set point change occurs at t = 100 s and a disturbance in inlet moisture content from 60.6 7. There is a set point change at t = 100 s and a disturbance in inlet moisture content to the drying section from 60.Chapter 8.4 y (%) uc (kPa) 7.0 % to 60.4 7.2 7.5 % at 7.
26). in some sense.28). To make an evaluation of the proposed control structure.Chapter 8. but it is of course not ‘free’ to use since there still is a cost from (8. see Figure 8. Simulations show that leaving out the constraints gives a more aggressive use of uc2.19. The difference in limit frequency for the two systems is around a factor two. as described above. full response of a change in uc1 and uc2) and a small control horizon to limit the computational effort for the MPC. However.19 The sensitivity function for single loop MPC (dotted) and midrange MPC (solid). have the same degree of robustness. uc 2 50 kPa. ObjectOriented Modeling and Control of the Moisture 1. MPC is nonlinear and the comparison only serves as guidance. (8.18. using the linearized model (8.5 Magnitude (abs) 1 0. due to the constraint handling. The tuning is chosen so that the two control systems have the same maximum value of the sensitivity function. This reduces the performance of the controller slightly but is important for the steam consumption. The cost function for the singleloop MPC is given by 204 . 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 steadystate. Both during the set point change and disturbance.17 shows a simulation of the midrange MPC and it clearly visualizes the thought of midranging.30) The purpose of this is to avoid severe injections of disturbances into the steam system by allowing large variations in steam usage. the rate constraint for uc2 is initially active.25. The maximum sensitivity is chosen to 1. which has negative effect on both steam production and other steam users.5 0 4 10 10 3 10 Frequency (Hz) 2 10 1 Figure 8. see Figure 8. the midranging MPC is compared to singleloop MPC. Figure 8. This implies that they. There is also a rate constraint for the control signals uc1 10 kPa .
20 shows the moisture response for the two different controllers together with the steam flow in the header. However. The midrange MPC has both better set point following and disturbance rejection.19. the transient steam 205 .Chapter 8. and constraint are Q 10.8 20 19 18 17 16 15 0 200 400 600 Time (s) 800 1000 1200 Figure 8. uc 10 kPa (8.6 Moistuure (%) Steam consumption (kg/s) 7.20 Comparison between midrange MPC (solid) and single loop MPC (dotted). 50 J (k ) i 1 ˆ r (k i) y (k i  k ) 2 Q 3 i 0 uc ( k i ) R . Since paper machines often are run from several hours to days with the same set point.0 6.32) Figure 8. ObjectOriented Modeling and Control of the Moisture 7. Figure 8. The midrange MPC has a limit frequency almost twice as large as the limit frequency of the single loop MPC. R 1. 2 (8. at least in the nonconstrained case.31) and the chosen weights that give the sensitivity in Figure 8.4 7.19 also indicates the difference in performance between the two controllers. showing the moisture content and steam consumption.2 7. disturbance rejection is important and it therefore makes sense to look at this property. The difference in performance is apparent.
17 and Figure 8. Therefore.18 and therefore the transient steam consumption is much larger.21 Comparison between midrange MPC (solid) and single loop MPC (dotted). it is an advantage if the flash steam from the last group is recirculated through a thermo compressor.22. it does not require any rebuild of the drying section. The output of the single loop MPC is in this case much more aggressive than it was in Figure 8. the single loop MPC has been tuned to give similar performance in disturbance rejection as the midrange MPC. consumption is twice as large for the midrange MPC.5.21. In Figure 8.Chapter 8. The cylinders are normally divided into different groups even though all groups follow the same set point. see Figure 2.6 Moisture (%) Steam consumption (kg/s) 7. The maximum value of the sensitivity is also larger. in general.4 7. showing moisture content and steam consumption when tuned to give similar performance in disturbance rejection.2 7. 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. 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 . shown in Figure 8. This to be able to match the performance of the midrange MPC.0 6. ObjectOriented Modeling and Control of the Moisture 7. However. hence the control system is less robust.20. see Figure 8. The advantage with the proposed control structure is that.8 25 20 15 10 0 200 400 600 Time (s) 800 1000 1200 Figure 8. This is the price being paid for the extra performance.
5 Magnitude (abs) 1 0. and algebraic constraints.22 The sensitivity function for single loop MPC (dotted) and midrange MPC (solid). the MPC has a systematic implicit way to treat constraints. By simply adjusting one parameter. The model has been validated against measurements on a real paper machine. see Figure 6. 207 . paper web. This midranging MPC has two essential advantages compared to the midranging structure used in Chapter 6. ObjectOriented Modeling and Control of the Moisture 1. the IMC or Dahlin controller used by many mills today. Firstly. Secondly. and different moisture controllers have been developed and collected in a model library. a good fit is obtained. implemented in the objectoriented modeling language Modelica. when tuned to give similar performance receiving group. This guarantees that the steam flows in the intended direction. CD. By draganddrop features it is easy to build a simulation model of virtually any existing drying section. 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. Components for steam cylinder. 8. the heat transfer coefficient for the condensate. It is also easy to expand the model library with components for press and wire section. A disadvantage can be that it is a bit more difficult to tune a midrange MPC compared to e. For the structure in Chapter 6. since the complete multivariable process model is given in the cost function. for a drying section has been given.Chapter 8. the MPC does not require any decoupling filter. All equations are based on mass and energy balances. control valve.5 Summary A physical model. this has to be taken care of explicitly and not as a part of the tuning phase.5 0 4 10 10 3 10 Frequency (Hz) 2 10 1 Figure 8.10.g.
208 . The strategy utilizes the possibility to divide the drying section into two parts. For the purpose of the feedback control investigated in this thesis. ObjectOriented 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. disturbance rejection and steam consumption.Chapter 8. this assumption sufficient. The midranging MPC is compared with singleloop MPC and the evaluation is done by comparing performance in terms of sensitivity function. the feedforward control in Chapter 7 requires a model that takes differences in the thickness direction into account to observe the peak position. and moisture gradients. which is a few minutes instead of a few days. However. By controlling this multivariable process with a midrange structure. The model used in the previous two chapters is capable of simulating microscopic phenomena like water and vapor transport. inside the paper sheet. An important advantage of the midranging MPC is that it does not require any rebuild of the physical process. the performance of the closed loop system is greatly improved. The advantage of this model compared to the one used in Chapter 6 and Chapter 7 is the simulation time. The model described in this chapter neglects this and assumes a homogeneous paper in the thickness direction.
Conclusions 9. it consists of a pole. To give the user the 209 .1 Summary In this thesis.Chapter 9. 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. 9 Conclusions 9. It is based on optimized load disturbance rejection subject to a robustness constraint and is referred to as IPZ tuning. an integrator. From the blackbox model. A simple graphical identification procedure is given that estimates the four model parameters from two asymptotes of an open loop step response. and moisture control. Written as a transfer function. control of the steam and condensate system. 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 blackbox model. The treatment is divided into two main parts. a tuning rule for PI and PID control is derived. and a time delay. a zero. The model structure is referred to as the IPZ model. modeling and control of the paper machine drying section is presented.
The setpoint feedforward design gives the desired response from setpoint to output. As this controller structure separates the servo and regulation problem it also fits well to the IPZ tuning. A new measurement signal. The main objective is to avoid resonance peaks in the closedloop frequency response from set point to process output. Therefore the tuning method has a much wider field of applications than solely drying section control. The two types of structures solve the objective differently but they roughly give the same performance. This means that the process parameters of the blackbox model are directly related to physical quantities as cylinder dimension. the twodegreeoffreedom controller is recommended since it is best suited for implementation in today’s industrial control systems. called the peak position is introduced. and steam properties. Moisture control Traditionally the moisture in the paper sheet is controlled by manipulating the pressure in the steam cylinders. The feedback is carried out by two types of midranging controllers. However. It is shown that peak position can significantly improve the control performance by feedforward. By making some assumptions and a linearization. apart from modeling the steam pressure in cylinders. and this is accomplished by a setpoint feedforward controller (twodegreeoffreedom controller) and a state feedback controller. 210 .Chapter 9. Conclusions option to balance between robustness and performance the tuning rule has a design parameter. The first uses the air supply system in combination with the steam system. The IPZ structure can be used in many other process areas. 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. To improve the performance of the closed loop moisture control system both feedforward and feedback is investigated. It is based on measurements of the paper surface temperature and indicates the position where the surface becomes dry. The IPZ model is used to examine some properties of the closed loop system and to investigate two different control structures. valve characteristics. 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 shown that the blackbox model can be derived from a few simple mass and energy balances. while a PID controller based on IPZ tuning takes care of the disturbance rejection in the feedback loop. It is successfully verified from experiments on real paper machine at different mills.
The modeling package would also benefit from more validation with real plant data. By draganddrop features. The next step is to evaluate them experimentally on a pilot or fullscale paper machine. The advantages of the midranging MPC are that the decoupling filter is not needed and that constraints are explicitly taken into account. 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. have to this point only been verified by simulations. and hood. together with a decoupling filter and a method to tune the system is given. The modeling package can also be used for optimization. different controller structures. with respect to given rate constraints. Two different of midranging controllers are used. The second midranging structure utilizes similar advantages by combining the last drying group with the remaining part of the dryers to improve performance. and normal run. However. and steam pressure. startups. grade changes. can be investigated. it lacks models for the pipings. two midranging controllers and a feedforward structure. caused by the cascade configurations in the condensate system. 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 second structure is implemented in a MPC. From mass and energy balances.Chapter 9. It can be used to investigate effects of a rebuild. a simulation model of practically any drying section can easily be built.2 Future work The proposed solutions for moisture control. By adding components for these objects. condensate tanks. The first midranging structure combines two singleloop controllers. 211 . and minimization of the time for a grade change by finding the optimal trajectories for machine speed. in this case IMC. and different ways to operate the machine. 9. a dynamic simulation package of a drying section is implemented in the objectoriented modeling language Modelica. The simulation package in Modelica contains components for cylinder and paper sheet. basis weight. crosscouplings between cylinders. Some examples of this are minimization of the total energy usage by finding the optimal steam distribution between the dryer groups. All this can be done for web breaks.
IMC. are compared would be valuable.Chapter 9. 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. Conclusions The differential pressure loop has only been treated briefly. deadbeat control. Smithpredictor. PID. Every control system vendor use their control structure but no thorough comparison have been published in the literature. 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. This loop will probably benefit from a feedforward structure since it is greatly affected by the operation of the pressure loop. minimumvariance. or LQG. are desired. A deeper analysis and a tuning rule. 212 . like the one for the pressure loop. A systematic analysis of moisture control where different control paradigms as MPC.
folding cartons) and graphic applications.afandpa. www.internationalpaper. for certain grades even higher. widely used for packaging (e. inverse of density of the paper.org.instituteofpaper. www.org. Board: Generic term for stiff paper usually made in several layers with a substance normally varying from 160 to 500 g/m2.paperonline. Ash content: The amount of residue when a sample of paper is burned under controlled conditions so that all ignitable matter is removed. 213 . It is usually repulped.Appendix A. Broke: Papermakers own waste paper created during papermaking process. concentrated by evaporation and burned in the recovery boiler to regenerate the cooking chemicals and generate energy. also called grammage. recovered during pulp washing. Glossary Appendix A Glossary This glossary contains explanations from the following websites: www.com. The ash derives from the mineral loading in the paper. Basis weight: Weight in grams of one square metre of paper or board. Black liquor: Mixture of cooking chemicals and dissolved wood material remaining after sulphate cooking.g.com. Bulk: Specific volume. and www.
Appendix A. Chemical pulp: Pulp in which wood fibres have been separated by chemical. Fine paper: A broad term including printing. Furnish: The ingredients or constituents of which the paper is made. which is suitable for the reproduction of fine screen artwork. Formation: The arrangement of fibers in a sheet of paper. Freeness: A measure of the drainability of an aqueous suspension of stock. Curl: The degree of curvature. Usually measured in nanometers. Filler: A material or substance that is added to the furnish to smooth out the spaces between fibers. writing. Conditioned basis weight: dsd Corrugated board: Formed fluting that is faced with a liner on both sides. as distinguished from wrapping papers and paper not generally used for printing purposes. Also known as thickness or bulk. in the course of beating or refining. Fibrous furnish concerns only the types and proportions of the fibres present. enhancing the printing properties of the paper. Fabric tension: The tension of the fabric creates contact pressure between the wet web and cylinder. Higher contact pressure gives higher heat transfer. and cover papers. which are generally referred to as coarse papers. rather than mechanical. Fines: Small fragments of fibres produced. The coating is applied in separate coaters or in the paper machine. Coated paper: The uniform application of a coating yields a more even and more closed surface of printing papers. for example. Fluting: Paper that has been formed into the flutes that make up the ridged part of the corrugated board between the liners. can be seen by holding it up to a light source. means. determined and expressed as specified in a standard method of test. Chemithermomechanical pulp (CTMP): Chemimechanical pulp produced by treating wood chips with chemicals (usually sodium sulphite) and steam before mechanical defibration. measured when paper is positioned flat. Glossary Caliper: The thickness of a sheet of paper or board. 214 . Dry weight: Mass of paper sheet excluding the moisture. Dry end: Final part of the paper machine from the drying section onwards.
engineered to be bright and opaque for the good print contrast needed by newspapers. directories. weight. gummed sealing tape. Newsprint also contains special tensile strength for repeated folding. tire wraps. flour. Glossary Grade: (1) A class or level of quality of a paper or pulp which is ranked. Kraft paper: A paper made essentially from wood pulp produced by a modified sulfate pulping process.Appendix A. It is a comparatively coarse paper particularly noted for its strength.) differing from another only in size. as well as specialty bags and sacks. or calendered. It does not includes printing papers of types generally used for purposes other than newspapers such as groundwood printing papers for catalogs. manufacturing history. and multiwall shipping sacks used for shipping such products as cement. appearance. striped. chemicals and animal food. others are commonly recognized but lack official definition.v. e. envelopes. coated paper. quality.g. made mainly from mechanical wood pulp. shopping and merchandise bags. an offset book paper cut grain long is not the same grade as the same paper cut grain short. etc. (2) With reference to one particular quality. and it has an acceptable surface for printing. In addition to its use as a wrapping paper. Fiber orientation: The direction of the fibers in paper. potato chips. ice cream.. and in unbleached grades is primarily used as a wrapper or packaging material. Mechanical pulp: Pulp consisting of fibres separated entirely by mechanical rather than chemical means. It can be watermarked. butcher wraps. sugar. cream tints. asphalted papers. multiwall sacks. or grain. Newsprint: A lightweight paper. Liner: A paper that is used as the facing material in the production of corrugated and solid fibre shipping containers. Packaging papers: These papers are used to wrap or package consumer and industrial products such as grocer's bags and sacks. it is converted into such products as: grocery bags. or a combination of these factors. Opacity: The ability of a sheet of paper to prevent light transmission through it. or distinguished from other papers or pulps. waxed paper. Some grades have been officially identified and described. 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. one item (q. "Specialty" packaging papers are used for cookies. and white. Opacity prevents print that is on one side of a sheet of paper from showing through to the other side. raw materials. and similar products. 215 . on the basis of its use.
Retention: Proportion of fibre and filler retained on the paper machine wire. which is about twothirds of the way down the dry end). chemicals and dyes that will be pumped onto the paper machine to form paper. Recovery boiler: In wood pulping. Glossary Paper Stock: A mix of pulp fibers. usually equipped with discs or with a cone and plug. Stock: The wet pulp before it is fed onto a paper making machine. Other examples of tissue papers are decorative and laminated tissue papers and crepe papers. and special sanitary papers. Smoothness: It is a measured property of paper that describes or rates the flatness and evenness of a sheet's surface.Appendix A. towels. or during the paper making processes before it becomes a sheet of paper. and durability. 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 oilbased inks (this process is carried out at the size press. napkin. Picking: Fibers in the paper which tend to pull away from the surface during the drying process. 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. Stock preparation: Collective term for all treatment necessary for the preparation of the stock before it reaches the making machine. Desirable characteristics here are appearance. Wet end: First part of the paper machine up to the drying section. Refiner: A machine. contains around 99% water and 1% fibre. water. 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 waterbased inks into the sheet. such as paper when it is on the paper machine or on rollfeed equipment. The extremely thin Japanese tissue papers are sometimes produced in grammages as small as 6 to 8 g/m2. White water: The filtrate from the wet end of the paper machine 216 . often used in gift wrapping and to decorate. 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. 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. additives. strength. wipes. Sheet (or Web): A continuous length of paper.
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.Appendix A. Woodfree paper: Paper consisting of chemical pulp fibres. 217 . It does not contain any mechanical pulp beyond a permissible content of 5 % by mass.
It will be shown why change in internal energy depends on in.1 The general energy balance for steam and condensate.and outflow of enthalpy. q is mass flow. control volume qin hin d Ui dt qout hout Figure A.1. and Ui internal energy. The balance equations come from the fundamental conservation principle in physics. 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. see Figure A. together with a few constitutive relations. h specific enthalpy. 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. In this appendix the energy balance for the steam and condensate in the cylinder model is discussed.Appendix B. used in this thesis. particularly not for gas systems. The change in internal energy depends on inflow and outflow of enthalpy. 218 .
Convective energy flow This is the energy carried by means of the mass flow into and out of the control volume. Convective inflow: qin (ui . the total energy of a system comprises of three principal components internal energy Ui kinetic energy Ek potential energy Ep Observe that we here use internal energy Ui [J] and not specific internal energy ui [J/kg]. and potential energy. It consists of internal.Appendix B. 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.out e p . Conservation Balance for Compartmental Models B.in ek . We call this Qc [J/s].out ) Convective outflow: qout (ui .out 219 .1 The energy balance Starting from a general energy balance. kinetic.in ) e p .in ek .
To see this.Appendix B.in ) qout (ui .out Qc .out ) qin pin vin qout pout vout Wm . a pump rotor).out e p .2.in ek . and ek and ep are specific energies [J/kg]. Expansion work accounts for work done in expansion or contraction of the control volume.in e p . Let the element have a length l [m/kg] and the crosssectional 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.out 220 . and vin and vout are specific volumes [m3/kg] at inlet and outlet. 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.in Qc .out ek . consider a unit volume of fluid across the boundary at the input. Shaft work is the work done by the fluid or surrounding through a moving part (e. Work done on the system is positive work. ( pin Ain ) 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 dt Ek Ep ) qin (ui .g. Flow work This term relates to the work done on the fluid as it moves into and out of the control volume. Mechanical work This is denoted as Wm [J/s] and can be both shaft work and expansion work. Conservation Balance for Compartmental Models where qin and qout are mass flows [kg/s]. This term is often not present especially not when the control volume is fixed.in Wm . see Figure A.
in e p . out ) Qc . in Wm . It is then simplified to d ( uiV ) qin hin dt qout hout Qc . and is given by the symbol h (h = u + pv). Conservation Balance for Compartmental Models The sum of specific internal energy and the pv term is called specific enthalpy. control volume inlet p l Figure A. out When the steam cylinder model is developed in Chapter 4. 221 . out where is density [kg/m3] and V volume [m3]. which is applicable to most systems in the process industry.in ) qout (hout ek . in Qc . out Wm . 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. the kinetic and potential energies are neglected. We can therefore write the energy balance as d (U i dt Ek E p ) qin (hin ek . There is an expenditure in energy to push the fluid across the boundary. out e p . in Qc .Appendix B. Also.2 A small section at inlet to the system.
and the energy balance can be further simplified to d ( hV ) qin hin dt qout hout Qc . out This is the case when considering the water phase (moisture) in the paper web. there is no pressurevolume work done on the system.Appendix B. Conservation Balance for Compartmental Models If we are dealing with an incompressible liquid. in Qc . 222 .
Assuming homogeneous material. it can be written as T t a T x2 2 0. and a onedimensional problem. (C. describing heat conduction in solids in a given region over time. of the following form c T t ( T ) k.1) where (kg/m3) is the density.2) where a (m2/s) is the thermal diffusivity.Appendix C. (C. c (J/kg·K) the specific heat capacity.3) 223 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. (C. defined as a . no added or subtracted heat inside the given region. and k (W/m3) the added heat per unit volume and time. (W/m·K) is the thermal conductivity.
1. the heat flow to the paper is assumed to be Q (W/m2) .1 At t = 0. t 0 T ( x. Thus.4) can be written as [Sparr and Sparr. T ( L.Appendix C. in steadystate. The two steadystate solutions are shown in the figure.0) T1 T (0. see Figure C. t ) x (C. t ) T2 .4) t 0 Q From theory of Hilbert Spaces and eigenfunctions the solution to (C. Solution to the One Dimensional Heat Equation T2 T1 Steam side T0 T x L Paper side Figure C. At t = 0 the steam temperature is suddenly changed to T2. the temperature on the steam side is changed from T1 to T2. the system is given by T t a T x2 2 0. Heat transfer through a cylinder wall with a step change in temperature A steam filled cylinder with wall thickness L. x (T0 L 0 T1 ) x L. has the temperature T1 on the steam side and temperature T0 on the paper side. 2000] 224 . Also.
t ) T2 Q x k 4(T1 T2 ) e 1) 1 ( 2k a k 1 2 2 2 L 2 t sin k 1 x. it is important to remember that the heat equation is also an approximation.5) 2 L The solution can be thought of as the sum of the steadystate temperature distribution (given by T2 Qx/ ) which is independent of time. (C.5. shown in figure C. the heat equation (C. The response is similar to a step response of a system with order higher than one.2 and Figure C.4 shows how the temperature at the paper side varies as a consequence of the step in steam temperature.Appendix C. given by (C. Since we have assumed constant energy flow Q. the slope of the initial temperature curve and the steadystate temperature curve are equal.6) where the heat equation is assumed to be the correct description of heat transfer. Figure C. However. the mean temperature of the cylider shell. 225 . 2000]. has more resemblance to the step response of a firstorder system. Solution to the One Dimensional Heat Equation T ( x . Figure C. The heat equation is a parabolic PDE and this is a typical character of this type of equation [Sparr and Sparr. the initial discontinuity in temperature is quickly smoothened out. Also.5).3 shows the temperature distribution. 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 firstorder linear model. (C. In Chapter 4. and a transient solution (given by the series) that tends to zero as t . However. To show this it will be solved it for an infinitely long rod.1) is compared to a simplified static relation Q A(T1 T2 ).
T1 = 120°C.5 1.5 Shell thickness (cm) 2.0 1.3 The temperature as a function of position x for different points of time.03 m.2 The temperature as a function of both position x and time t (given in seconds).Appendix C. = 7000 kg/m3. L = 0. 226 .5 3. Q = 10 kW. T2 = 130°C. = 50 W/(m2°C).0 2.0 Figure C. 130 128 126 Temperature ( C) 124 122 120 t=0 t=1 t=5 t = 20 t = 500 t = 100 t = 50 o t = 10 118 116 114 0 0. Solution to the One Dimensional Heat Equation Figure C. c = 500 J/(kg°C).
5 The mean temperature of the cylinder shell as a function of time. The response is similar to a step response of a firstorder system. 128 127 126 125 Mean temperature ( C) 124 123 122 121 120 119 118 117 0 20 40 60 80 100 120 Time (s) 140 160 180 200 o Figure C.Appendix C. o 227 .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. Solution to the One Dimensional Heat Equation 125 124 123 122 Temperature ( C) 121 120 119 118 117 116 115 114 0 20 40 60 80 100 120 Time (s) 140 160 180 200 Figure C.
In other words. the solution to (C.8) where G ( x. If we assume that the initial condition is given by the Dirac pulse. and the model 228 .7) is T ( x.Appendix C.7) T ( x . t ) G( x ) g ( )d .11) independently of t and a. This because no energy leaves the rod and the Dirac pulse has unit area. x .0 ) This equation can be solved by the Fourier or Laplace transform and the solution is given by the convolution formula T ( x. The integral of (C.10) is that the energy impulse in the origin affects the temperature in each point x for all t > 0. The implication of (C.9) G is called the Green function for the heat equation.t 0 (C. (C. and evaluates to 1 e 4 at x2 4at dx 1. (C.10) over the whole space is proportional to the thermal energy in the rod. (C. t ) G G( x ) ( )d 1 e 4 at x2 4at . (C. t ) 1 e 4 at x2 4at . all points in space will sense the impulse instantaneously.6 shows this solution as a function of both time and space. t ) G g ( x.10) Figure C. Solution to the One Dimensional Heat Equation Heat conduction in an infinitely long rod Regard the heat equation for an infinitely long rod T t a T x2 g ( x) 2 0.
and is a field of research still today [Herwig and Beckert. 2000]. for the normalized case a = 1. Solution to the One Dimensional Heat Equation (C. Figure C. first explicitly solved by [Cattaneo. and [Belevich.Appendix C. the heat equation does not admit any discontinuous solutions.1) leads to an infinite propagation velocity for the heat at the starting point. The theory has become known as extended thermodynamics. This is due to the averaging effect of the integral in (C. 2004]. 229 .6 Temperature as a function of both position x and time t (given in seconds).1) involving a derivative of second order in time. 1948]. This nonphysical property of the heat equation has been extensively discussed in the literature. [Pulko. 2002]. 1989] gives a good overview of the field. Loosely speaking.8). et al. [Joseph and Preziosi. It then becomes a hyperbolic function which can be regarded as a “wave equation for heat” (with a certain speed of the heat wave). Often the drawback of infinite velocity is overcome by adding an extra term to (C.
. J. B. 2003. A.lth. Download can be found at www. 98(4). Department of Automatic Control. Allison. 25(1).” Tappi Journal. M.” Journal of Process Control. Karlström (1997): “Greybox identification of a TMP refiner. Allison. J. Licentiate thesis. 469 474. Allison.” Transactions of the Institute of Measurements and Control. and A. SeyedYagoobi (1992): “Theoretical drying study of singletier versus conventional twotiered dryer configurations. Asensio.. pp. B. 3 16. pp. Lund Institute of Technology. Operator interaction and optimization in control systems.. J. J. Isaksson (1998): “Design and performance of midranging controllers. 230 .se/user/johan.. and S. C. B. 8(5 6). Isaksson. ISRN LUTFD2/TFRT3234SE.akesson/mpctools. and A. J.. pp. pp.” Pulp and Paper Canada. 203 211. 123 126.References References Åkesson. and J.control. 75(10). Ogawa (2003): “Design and tuning of valve position controllers with industrial applications. J.
6(1). Åström. 3rd ed. pp.” Automatica.” Journal of Process Control. Åström. J.” Automatica.. (2000b): “Limitation on control system performance. Hägglund (2005): Advanced PID control. PrenticeHall. K. Balchen. pp. and B. Hägglund (2001): ”The future of PID control. J. J. 9(2). pp.” Journal of Physics A: Mathematical and General.. Bell (2000): ”Drum boiler dynamics. pp. (1998): Process dynamics modeling. (2000a): “Control problems in paper making – revisited. G. (2004): “Causal description of heat and mass transfer. 1163 1175. K. (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. J. K. and simulation. analysis.” European Journal of Control. Upper Saddle River. pp. 14(6). Åström. (1998): “Design of PI controllers based on nonconvex optimization. B. Åström. M. pp. NC. pp. 34(5). (2003): Personal communication.” Control Engineering Practice. Åström. 389 405. J. Prentice Hall. Instrument Society of America. Wittenmark (1997): Computercontrolled systems theory and design. 37. J. 3053 3069. Åström. 101 108. Canada. J. K. NJ. J.” IBM Journal of Research and Development. Upper Saddle River.. Belevich. 363 378. pp.. Åström. pp. Åström. 129 136..References Åström. J. Hägglund (2004): ”Revisiting the ZieglerNichols step response method for PID control. Åström. 36(3). and R. and T. (1967): “Computer control of a paper machine an application of linear stochastic control theory. 11(4). K. and T. K. NJ. K. Research Triangle Park. 9. K.” In Preprints Control Systems 2000. K. 231 . J. and T. Bequette. K. W. J. 585681. 635 650. 2 20. D.
. 30(2).” Preprints 7th New Available Technologies. and S. (1994): “A case study of grey box identification. L. (2000). IEEE Industry Applications. and D. Borisson. W. The Entech Report. 63 69. vol. 76 87. 3. 61 67. pp. 15 20. 65. pp. no. and P. 12. pp. 579 586.” Automatica. Radziszewski (1997): ”A state model for the drying paper in the paper product industry. 9(6). Bortolin. pp. Berrada. 1 8. Gutman (2004): ”Modeling of the wet end part of a paper mill with Dymola.” In 4th IFAC/IFIP Symposium on Digital Computer Applications to Process Control. and P. Tarasiewicz. vol.References Bergström. T. pp. U. 307 318. 31 38. Borg. pp. pp. 465 490. Nashville. Millard (1993): “New modelbased machine direction control algorithm. Stockholm. Dumont (1998): ”An object oriented framework for developing dynamic models of a paper machine. pp. Zürich. on Industrial Electronics. Elkadiri. (1996): “Effectiveness of deadtime compensators for basis weight and moisture control. no.” International Journal of Adaptive Control and Signal Processing. Halifax.” Mathematics and Computers in Simulation. Canada. pp. “Advanced process control: Is it a cure for all process control needs”. 60 64.” Dynamic Modeling Control Applications for Industry Workshop. and B. and G. Bialkowski. Sweden. Switzerland. M. M. J. O. Boström. B (2002): “Measurement of dynamic moisture gradients in the zdirection.” In Proceedings TAPPI Process Control Conference. Bialkowski. 232 . W. G.” IEEE Trans.” In Proceedings Control Systems 1996. T. 4. pp. S.. Brown.. Wittenmark (1974): “An industrial application of a selftuning regulator. TN. S. Graebe (1995): “Issues in nonlinear stochastic GreyBox identification. Bohlin.. L. T. Bohlin. 44.
Chang.cepi. 46(8).” Transactions of American Society of Mechanical Engineers. (ed) (1989): Teknisk uppkäftighet. Chen. W. 84(2). Canada. San Diego.” Pulp and Paper Canada. 247. Hrones. pp. Sree (2003): “A simple method of tuning PID controllers for integrator/deadtime processes. A. pp.” Confederation of European Paper Industries. pp.” Tappi Journal. 431 433.” Computers and Chemical Engineering.C. D. Choi. C. and I. and J. R. pp. Wambsganss (2001): “A novel multiport cylinder dryer. J. 3082 3086. Chen. France. 27(2). (1967): The papermaking machine. M.C. Reswick (1952): “On the automatic control of generalized passive systems. Chidambaram. S. its invention. C. Chien.. CA. Chen. R. 47. 1581 1592. D. 17 21. 233 . Whistler. W. Download can be found at www. pp. and M. M. P. Chien (2000): “Design sensor trajectory for control: application to sheetforming processes. S. Yu. CEPI (2004): “European pulp and paper industry Annual statistics. M. 175 185. 96(1).org.References Carlberg. pp. Pergamon Press.L.” in Proceedings of American Control Conference.. (1995): “Modelling of paper machines for control: theory and practice.M. and R. pp. Clapperton.C.” AIChE Journal. B. H. Yu. Skogsindustrierna and Bokförlaget Bra Böcker..C. L.. 2004. (1992): “Fullwidth sheet property estimation from scanning measurements.” Comptes Rendus Academie Science. S. 74. evolution and development. (1958): ”A form of heat conduction equation which eliminates the paradox of instantaneous propagation. Cattaneo.” in Proceedings Control Systems ’92. Subbarayan (1999): “Crossmachine direction response modeling with twodimensional sheet variations measurements.. S. K. 211 215. A.
19(10). 25(1). Y. 2559 2575. K. C. J.” Instruments and Control Systems. (1988): “Challenges and opportunities in pulp and paper process control.” Proceedings IEEE Control and Decision Conference. R. Luck versus good planning. Norman (1998): Pappersteknik. pp. 234 . vol. A. (2003): “Editorial. Atlanta. E. M. pp. H. Ordubadi. S. Institutionen för pappersteknik. and K.References Cohen. pp. 101 113. K. and G. Dumont. 1(2). Kristinsson. Fu. pp. H. Dumont. Davies. C.” Dagens Industri (a swedish daily paper). Jonsson (1991): ”An improved algorithm for estimating paper machine moisture profiles using scanned data. pp. M. pp. 827 834. (2001): “The mysteries of technology transfer. pp. Davies.” AIChE Symposium Series No. S. Natarajan (1993): “Estimation of moisture variations on paper machines. 322. 75. and L. 95. F. B. Doyle.. Dumont. G. published 20040214. Dahlin. Brighton.. F. (1999): “Computational issues in the application of modelbased control to the pulp and paper industry.” Transactions of the Institute of Measurements and Control. G. Coon (1953): “Theoretical consideration of retarded control.” Automatica. 1857 1862. Dumont. 22(2). 1 2.” Proceedings of the 1988 American Control Conference. pp. A (1986): “Applications of advanced control methods in the pulp and paper industry a survey.” IEEE Transactions on Control Systems Technology. 77 83. KTH. England. Crotogino. A.. GA.. 165 172. A. Fellers. pp. and B. Natarajan.” Drying Technology..” Transactions of American Society of Mechanical Engineers. S. 1959 1964. Dagens Industri (2004): “Skogsindustrierna tappar mark i tuff konkurrens. Lindeborg. G. G. Duncan. (1968): ”Designing and tuning digital controllers. A. 143 153. 41(6). G.
25. Canada. Lockwood Publishing.References FFIF (2004): “Paper and wood. New York.) (1972): Drying of paper and paperboard. 3. G. B.” Proceedings of the 53rd Appita annual conference. Gavelin. and W. and A. 335 348.” Proceedings of 44th IEEE Conference on Decision and Control and European Control conference. Johansson (2004): “Modeling of the moisture content of the paper in the drying section of a paper mill. Marquardt (1998): “A field study of the industrial modeling process. Vancouver. 183 187. 8.” Automatica. (1998): Paper machine design and operation. 5. and P. Seville. Fritzson. 325 338. Gaillemard. Hagander (2005): “Antiwindup in midranging control. pp. Velut. and I.. (ed. Angus Wilde Publications. Forsman. (2004): Objectoriented modeling and simulation with Modelica 2. Download can be found at www. no. García. Gavelin. no. 235 . and J. Studentlitteratur. S. Cameron (2001): Process modelling and model analysis. of Control Systems 2004. USA.. M.. Karlsson. E.forestindustries. Canada..” In Proc. Foss.. Lund. New Zealand.” Finnish Forest Industries Federation. vol. pp. and M. B. M. pp.” Journal of Process Control. C.fi. John Wiley & sons. Forsman. statistics 2003. D. K. Morari (1989): “Model predictive control: theory and practice a survey. K. Rotura. yearbook 2004. Prett. Haugwitz. K (2005): Reglerteknik för processindustrin.1. Quebec. S. Hangos. New York. Lohmann. G. Academic Press. P. vol. C. Sweden. Spain. Birgersson (1999): “Modelling and control of the process air in a paper machine hood.
Jonsson. and E.” AIChE Journal. H. Jansson. E. Joseph. 35 46.” IEEE Transactions on Control Systems Technology. 4(4). N. C. (1991): Process control in practice. C.” Svensk Papperstidning. Heikkilä. 61(1). Graebe (2002): “Derivative filter is an integral part of PID design. Ho. T. pp. D. 31(3). A. Lund. and R. 604 618. S. B. Manness. K. F.. Hill..References Heaven. Schwaber (1995): “Habituating control strategies for process control.” IEE Proceesings of Control Theory and Applications. M. Department of Chemical Engineering. pp. S. M.” Automatica. pp. Gan. PhD thesis.” Journal of Heat Transfer.. C. K. A.” IEEE Control System Magazine. and L. Kean. Herwig. pp. Ho. 497 502. J. 363 365. M. Isaksson. P. P. E. A.. 76(6). M. 41 73. 105 114. (1993): A study on the drying process of pigment coated paper webs. Hägglund. 14(5). D. (1993): “Analyzing the dryer section’s steam and condensate system. pp. Tay. 41(3). Ogunnaike. 473 477. and J. pp. Nordgren (1958): “Om torkning av papper på filttäckta torkcylindrar. Beckert (2000): “Fourier versus nonfourier heat conduction in materials with a nonhomogeneous inner structure. J.. W. M. T. Henson. and B. and L.. pp. Studentlitteratur. B. I.” Reviews of Modern Physics.” Tappi Journal. Åbo Akademi. and K.. 236 . 834 843. 149(1). W. L. A. Preziosi (1989): “Heat waves. 61(19). K. Sweden. 122(2). Vyse (1994): “Recent advances in cross machine profile control. 4145. Finland. Cao (1995): “Tuning of controllers based on gain and phase margain specifications. Hang. and S. Ang (1996): “Performance and gain and phase margains of wellknown PID tuning formulas. pp. O. pp. Åbo.
pp. pp. M. E. R. M. part I: theoretical model. pp.. pp. M. Slätteke. and T. 5th edition. 21 29. 237 . 23(1 2). 260 263. Kuang. Kastanakis. 143 163. 74(2). B. S.” Tappi Journal. 129 137. R.” Control Engineering Practice. 78(7). (ed) (2000): Paper making part 2. and P. Wellstead (1995): “Identification of crossdirectional behaviour in web production: techniques and experience.” Tappi Journal. Stenström. 3(1). and B. Karlsson. Baggerud (2002): ”Dynamic simulation of the steam supply system for a multicylinder dryer” Nordic Pulp and Paper Research Journal. Tappi Press.. Wittenmark. and E. (2002): ”Förfarande vid torkning av ett banformigt material. and S. Chicago. drying. pp.D. Stenström (2005a): “Static and dynamic modelling of cardboard drying. Thibault. J. pp.. Kosonen. Grandjean (1995): “Pilot scale investigation of infrared drying of paper. and S. 66 70. 17(1). Chen.” Drying Technology. Stenström (2003): ”Evaluation of models for the steam supply system. 103(10). Karlsson. I. part II: simulation and experimental results. G.” Tappi Spring Technical Conference & Trade Fare.. Kuusisto. Stenström (2005b): “Static and dynamic modelling of cardboard drying. IL. 23(1 2). Lizr (1991): “Interaction between the MD and CD control processes in paper making and plastics machines. Shakespeare. P. and A.. H. Heath. Tappi Press. R. Karlsson.” Pulp and Paper Canada. A. O. J. 165 186. pp. M..References Karlsson. 77 83. (ed) (1997): Paper machine steam and condensate systems. M.. Karlsson.” Drying Technology. Krumenacker. M. W. Huhtelin (2002): “Multivariable control of paper machine grade changes. Kjaer..” Swedish patent. and S. Karlsson. 00048512 (SE518049). P.
Lehtinen.” Proceedings of the American Control Conference. 1471 1474. Belgium. M. K. MA. (1994): “Innovative control technique improves control rangeability and resolution in paper mill applications.Theory for the user. L. NY. user’s guide version 6. pp. A. Jussila. Ljung.. pp. M. W. 13(8). J. PA. 19(10). 238 . and T.. C. (1995): “Condebelt drying of paper and paperboard for optimizing quality and production for many grades. and A. E. Sao Paulo. Philadelphia. Lai. (1999): System identification . Rantanen. P. 2049 2068. 2nd edition. (2004): System identification toolbox Mathworks. and T.” Drying Technology. and H. 2435 2450.References Larsson. J. 3(10). R. T. Martinez. Kaukovirta. 195 199.” 7th IFAC symp. Stenström (2004): “Assessment of different paper drying processes to reduce the total energy costs from a mill perspective. Koivisto.” Control Engineering Practice. Ramaswamy (2001): “Preemptive control of multiply actuated processes: application to moisture content control in paper manufacturing using surrogate measurements. Lindell. R. S. Y. P. 2997 3001. Prentice Hall. Ljung.” Drying technology. pp. J. Makkonen. New York. Matin (2001): “Dryness and physical properties of paper webs in laboratoryscale impulse processing. Gent.” Tappi Journal. J. Elmqvist (1997): “Modelica An international effort to design the next generation modeling language. H. D. and S. Gustafsson (1998): “Paper machine dry line position control during grade changes. H. Brazil. E. pp. Love.. Li. Lieslehto. Drotz. A. Koivo. 77(2). L. Bjegovic. Mattsson. R. Natick. N...” Proceedings of 2001 ASME International Mechanical Engineering Congress and Exposition. and S.. Huhtelin (1995): “Three control schemes for paper machine MDcontrol. pp. on computer aided control systems design.” Proceedings of the 14th International Drying Symposium.
” Proc.. PRC. H. A. McConnell. M. pp. Englewood Cliffs. pp. Elmqvist. Finland pp. Mori. T. K.” Journal of Process Control..C. Kuusisto (1998): “Dynamic modelling of paper machine from grade change data. Dumont. NY. 1278 1283. Victoria. Dearborn. and E. N. Practice. pp. H. Drying ‘80. Chen (1999): “Transition control of papermaking processes: paper grade change. A.” Proceedings of the 1996 IEEE International Conference on Control Applications. Dearborn. R.” Proceedings of the 1996 IEEE International Conference on Control Applications. Hawaii. Murphy. H. 239 . 501 510. J. S.” In Proceedings of Control Systems. G. F. A Henson (1996): “Habituating control for nonsquare nonlinear processes. E. Canada. B. Zafiriou (1989): Robust process control.. (1980): A literature review of drying research in the pulp and paper industry. 231 234. Beijing. and S. S. T. and R. and S. Morud. 6. PrenticeHall. Menani. Hemisphere Publishing. 273 278. vol. T. Y. and K. Otter (1998): “Physical system modeling with Modelica. Porvoo. Kurtz. Huhtelin.. Control Systems 2000.” Proceedings of IEEE International Conference on Control Applications. pp. pp.” IFAC/IFORS Symposium. and M. 79 86.. J. McLain. R. M.C. MI. J. Yurkovich. Ed.References Mattsson.. Mujumdar. 330 337. S. Skogestad (1996): ”Dynamic behaviour of integrated plants. Davies (1988): “An algorithm for estimating cross and machine direction moisture profiles for paper machines. 145 156. Morari. and M. Michigan. pp. pp. and M. F. Kohala CoastIslands of Hawaii. S. Murphy. Natarajan. Shimizu. Chen (1996): “Intelligent control for paper machine moisture control. 826 833.. Takao (2000): “Development of a new automatic grade change control system for paper machine. pp.. 1091 1096.” Control Eng. and S. Koivo. 6.
Lund Institute of Technology. S. K. Brocard (2004): “Modelling of paper drying: a comprehensive computational model used by the industry. pp. 8. GA. 7. W. K.. (1991): “Troubleshooting design and operation of the dryer section.” in TAPPI Practical Aspects of Pressing & Drying Short Course. and I. W. and S. 81(7).” Proceedings of the 14th International Drying Symposium (IDS’2004).. D.” Experimental Thermal and Fluid Science. Mizukami.” Proceedings of the 10th International Drying Symposium (IDS’96). Niemenmaa. P. 82(9). Kaye (1955): “An analytical approach to the problem of drying thin fibrous sheets on multicylinder machines. Gardner (1996): “Optimizing paper machine control a case study. pp. Y. no. Laukkanen (1996): “Dynamic modelling and simulation of paper machine drying section.. Takeba (1997): ”An experimental study of condensation heat transfer in a horizontally rotating cylinder with a scraper. pp 205 212. 97(11). 385 398. Lappalainen.References Nelson. D. A. 240 . Perrault.” Tappi Journal. (1998): Dynamic modelling and simulation of multicylinder paper dryers.” Tappi Journal. and J. D. J.” Tappi Journal. R. A. no. Licentiate thesis. H. B. B. Perré. M.1157 1164. Juslin.. M. pp. pp. vol. Kraków.. W. N. Lescanne. São Paulo. pp.. 38. vol. and W. 46 52. 143 149. pp. Petterson. Stenström (2000): “Experimental evaluation of electric infrared dryers. vol.” Tappi Journal. T. I. (1998): “Paper machines of tomorrow. Orloff. Atlanta. Nissan.. Liu. Persson. Sebe. Poland. 43 49. 14.” Pulp and Paper Canada. Peng. Crouse (1999): “Impulse drying: status of the pilotscale research program. and Y. 249 277. P. Department of Chemical Engineering. 1263 1270. Pauksta. drying to speed. vol. and T. Brazil. pp. 83.
and A.References Poulin É. 9. Col. 83. Douglas (2004): “From tissue to linerboard: validation of a microscale simulator for single technique and hybrid dryers. Wilkinson. S. 303 315.” Pulp and Paper Magazine of Canada.. Germany. Sadeghi. Ying (1994): Modeling and advanced control for process industries: application to paper making processes. (2003): Modeling and simulation of transport phenomena in paper drying.” Tappi Journal. E. 61 66. M. New York. Springer Verlag. and W. 444 451.. (2001): “Key development phases of Condebelt: long journey from idea to commercial product. and G. pp. pp. 15. 201 208. vol. M. 89(8). M. Reardon. L. J. Montreal.. Rudd (1994): “Prediction and control of paper machine parameters using adaptive technologies in process modelling. Retulainen. 2451 2467. H. Xia.” Drying Technology. (1969): Properties of water and steam in SIunits. Schmidt. M. J. E. Wedel (1988): “The effect of spoiler bars on dryer heat transfer. Saidane (2002): “TLM representation of the hyperbolic heat conduction equation. and A. A. pp.. A. Schweiger. S. Davis. Canada.” International Journal of Numerical Modelling. Q. Pomerleau (1999): “PI settings for integrating processes based on ultimate cycle information. Brazil. M. Berlin. McGill University.. Sadeghi. PhD thesis. no. PhD thesis. 77(11). pp.” IEEE Transactions on Control Systems Technology. São Paulo. (1991): Modeling of multicyliner drying of lightweight paper. A. Rao.. SpringerVerlag. of Environmental Science and Forestry. 7(4). Pulkowski. C. J. and J. R.” Tappi Journal. B. 241 . 509 511. vol. pp. State University of New York. Pulko. Doe (2000): “Computational modelling of paper drying machines. 19(10). H. and P.. and Y. Ramesh. pp.” Proceedings of the 14th International Drying Symposium (IDS’2004).
Research report 618. Porkholm (1989): The APROS software for process simulation and model development. Stenström. pp. Skogestad. Slätteke. (1978): “Control systems can save energy. pp. M. Karlsson. vol. G. pp. Oregon. Counts (2001): “Heating/drying of paper sheet with gasfired infrared emitters pilot machine trials. S.291 309. Shinskey. Tappi Journal. Wittenmark (2002): ”Productivity increase from a better understanding of dynamic processes and control of the paper dryer. J. Tiihonen. Wittenmark (2002). J. Brundin. and K. S.for the pulp and paper industry. A. Finland. O. Hänninen.” Nordic Pulp and Paper Research Journal. Stenström. “On identification and control tuning of cylinder dryers” In Proceedings Control Systems 2002. Sweden. and T. S. Juslin.” Drying Technology.” American Control Conference 2005. pp. 157 162. Studentlitteratur. 43 46... Portland. 70 73. pp. 183 188. (2003): “Simple analytic rules for model reduction and PID controller tuning. Sparr (2000): Kontinuerliga System. T. Slätteke.. 13. Lund. O. M. Silvennoinen. J.. and A.” Chemical Engineering Progress. Hägglund. 15. Sweden Sparr. 242 . and B. M. S.” Journal of Process control. Svanquist (1991): “A general model for calculating pressure drop in siphons and siphon riser tubes Part 2: experimental results”. and K. Sikirica. E. K. N.. O. O. Skoglund. 298 302.” Preprints 7th New Available Technologies. Stockholm. J. Åström (2005): “Modeling of a steam heated rotating cylinder – a greybox approach. Espoo.. G. Stockholm. and B. Tappi Press. Sweden. pp. J. and C.. F.F. pp. K. Mandenius (2000): “A multivariate process model for grade change in a paperboard machine. Forsman.References Sell. Technical Research Centre of Finland. SeyedYagoobi. K. 43. 74(12). and K. no. 19(3&4). 3. A. 639 651. Slätteke. Forsman. Kurki. (ed) (1995): Process control fundamentals .
Wilhelmsson. pp. Dumont (2003): “Feedback controller design for a spatially distributed system: the paper machine problem. Lemaitre (1982) An improved model of a paper machine multicylinder drying section. Krook. 3724 3728. 2625 2628. Tyreus. Australia. China. pp. P. pp. L. and R. Sun. G. 612 628.. E. 31. X. and G. pp. Drying ‘82. Oxford.. Thomas. M. 243 . Luyben (1992): “Tuning PI controllers for integrator/dead time processes.” IEE Proceedings of Control Theory Applications. D. Hemisphere Publishing.. Visioli. PhDthesis. Wimmerstedt (1994): “Measurement of reference experimental drying data for the multicylinder paper dryer. Great Britain. NY. Videau. D. Stewart. pp. (1999): Simulation of industrial processes for control engineers.References Stenström. Ed. J. 408 417. R. Braatz (2000): “Fast model predictive control of sheet and film processes. 1179 1186.” Proceedings 3rd World Congress on Intelligent Control and Automation. pp. R. B. and W. L. Heifei.. S. 8(3). 180 184.” Proceedings of the 9th International Drying Symposium (IDS’94). G. Gorinevsky. B.. ButterworthHeinemann. Helsinki University of Thechnology. VanAntwerp. P. and A.. P. A. Sun (2000): “The modeling and control of basis weight and moisture content. Viitamäki.” IEEE Transactions on Control Systems Technology. 148(2). Mujumdar. (2001): “Optimal tuning of PID controllers for integral and unstable processes. A. 11(5).” IEEE Transaction on Control Systems Technology. Department of Automation and Systems Technology. A. K. pp.” Industrial and Engineering Chemistry Research. Nilsson. 129 138. Gold Coast. (2004): Hybrid modeling of paper machine grade changes. Finland. and R. J. D. and Y. L. Yi.
Control Theory and Applications. Xia. M. Nichols (1942): “Optimum settings for automatic controllers.. Cai (2001): “PID tuning for integrating processes with sensitivity specification. Wilhelmsson. pp.” Transactions of American Society of Mechanical Engineers.” Proceedings Conference on Decision and Control.. (1999): “Robust multivariable IMC control for paper machines. pp. M. P. H. Wells. 159 168. 203 208. Lund Institute of Technology. Stenström (1994): “An experimental study of contact coefficients in paper drying. Department of Chemical Engineering. 244 . 140(3). Y..” Proceedings of IEEE International Conference on Control Applications.J. 216 224. Ziegler. and D.. pp. Q. G. H. Vancouver. 229 234. B. Wang. 82(7). (1995): An experimental and theoretical study of multicylinder paper drying. (1996): “Applying neurofuzzy modelling and control to MD moisture content systems in paper machines. Sweden. L.” IEE Proceedings D. B. pp. Wang. B. Bristish Columbia. pp.” Proceedings of the IEEE Conference on Control Applications. and J. 4087 4091. Dearborn. Qian (1993): “Model algorithmic control for paper machines. FL. Rao.” Tappi Journal. Atherton (1993): “Automatic tuning of optimum PID controllers. J. Fagerholm. C.” Tappi Journal. Lund Institute of Technology. MI. 64. (2000): Tools for autonomous process control. PhD thesis.G. Wilhelmsson. Orlando. L. A. and S. 759 768. pp. Zhuang.References Wallén. 77(5). pp. Nilsson. W. and N. Department of Automatic Control. 140 144. PhDthesis.
m Cp.p Cv C(s) d dv dy g G(s) hs hw Hs Hvap H J kc K m Ms Mw kg/mole m/s kg J/kg J/kg J/kg J/kg J/kg kg/(s·%) m kg/m 2 m2 m2 J/(kg·K) J/(kg·K) m2 Inner cylinder area Area of paper covering the cylinder Specific heat capacity of cylinder shell Specific heat capacity of paper Valve conductance Controller transfer function Load disturbance Valve constant Width of paper sheet Dry basis weight of paper General transfer function Enthalpy of steam Enthalpy of condensate Heat of sorption Latent heat of vaporization Energy required to evaporate water from the paper surface Cost function in MPC Gain of the PID controller Mass transfer coefficient for paper sheet Mass of cylinder shell Maximum value of sensitivity function Molecular weight of water 245 .List of Symbols List of Symbols Acyl Axy Cp.
List of Symbols n p psh ptot pv0 pv.325 kPa) Partial vapor pressure for free water Partial pressure for water vapor in the air Partial pressure for water vapor at paper surface Process transfer function Blow through steam Condensation rate Evaporation rate Inflow of steam to cylinder Outflow of condensate Weighting matrix for MPC Energy flow to cylinder shell Energy flow to paper Set point Weighting matrix for MPC J/(mole·K) Gas constant ( 8.a pv.31) Valve rangeability Derivative time of the PID controller Integral time of the PID controller Temperature of cylinder shell Temperature of paper Moisture ratio Controller output Internal energy of steam Internal energy of condensate .p P(s) qbt qc qevap qs qw Q Qm Qp r R Rg Rv Td Ti Tm Tp u uc us uw 246 J/kg J/kg s s K K kg/kg W W kg/s kg/s kg/(m2·s) kg/s kg/s Pa Pa Pa Pa Pa Pa Noise Steam pressure inside cylinder Steam pressure in header Standard pressure (101.
List of Symbols V Vs Vw vx w x xv y W/(m2·K) W/(m2·K) m W/(m·K) kg/m3 kg/m3 and 2 m3 m3 m 3 Volume inside cylinder Volume of steam Volume of condensate Speed of paper sheet Moisture content Water content in air Valve opening Process output Heat transfer coefficient steam Heat transfer coefficient cylinder Cylinder thickness / 2 Sorption isotherm Thermal conductivity of cylinder shell Fraction of dryer surface covered by paper Density of steam Density of condensate Tuning parameters of midranging controller cylinder paper m/s % kg/kg sc cp cyl cyl s w 1 247 .
.
.
Department of Automatic Control ISSN 02805316 ISRN LUTFD2/TFRT1075SE .
This action might not be possible to undo. Are you sure you want to continue?
Use one of your book credits to continue reading from where you left off, or restart the preview.