You are on page 1of 16

Automatica 36 (2000) 363}378

Drum-boiler dynamicsq
K.J. As stroK m!,*, R.D. Bell"
!Department of Automatic Control, Lund Institute of Technology, Box 118, S-221 00 Lund, Sweden
"Department of Computing, School of Mathematics, Physics, Computing and Electronics, Macquarie University, New South Wales 2109, Australia
Received 2 October 1998; revised 7 March 1999; received in "nal form 8 June 1999

Abstract

A nonlinear dynamic model for natural circulation drum-boilers is presented. The model describes the complicated dynamics of the
drum, downcomer, and riser components. It is derived from "rst principles, and is characterized by a few physical parameters.
A strong e!ort has been made to strike a balance between "delity and simplicity. Results from validation of the model against unique
plant data are presented. The model describes the behavior of the system over a wide operating range. ( 2000 Elsevier Science Ltd.
All rights reserved.

1. Introduction complex and they do not capture dynamics. Black box


models are only valid for speci"c operating conditions.
There are dramatic changes in the power industry This paper presents a nonlinear model for steam gen-
because of deregulation. One consequence of this is that eration systems which are a crucial part of most power
the demands for rapid changes in power generation is plants. The goal is to develop moderately complex non-
increasing. This leads to more stringent requirements on linear models that capture the key dynamical properties
the control systems for the processes. It is required to over a wide operating range. The models are based on
keep the processes operating well for large changes in the physical principles and have a small number of para-
operating conditions. One way to achieve this is to incor- meters; most of which are determined from construction
porate more process knowledge into the systems. There data. Particular attention has been devoted to model
has also been a signi"cant development of methods for drum level dynamics well. Drum level control is an im-
model-based control, see Garcia, Prett and Morari portant problem for nuclear as well as conventional
(1989), Qin and Badgwell (1997) and Mayne, Rawlings plants, see Kwatny and Berg (1993) and Ambos, Duc and
and Rao (1999). Lack of good nonlinear process models Falinower (1996). In Parry, Petetrot and Vivien (1995) it
is a bottleneck for using model-based controllers. For is stated that about 30% of the emergency shutdowns in
many industrial processes there are good static models French PWR plants are caused by poor level control of
used for process design and steady-state operation. By the steam water level. One reason is that the control
using system identi"cation techniques it is possible to problem is di$cult because of the complicated shrink
obtain black box models of reasonable complexity that and swell dynamics. This creates a nonminimum phase
describe the system well in speci"c operating conditions. behavior which changes signi"cantly with the operating
Neither static models nor black box models are suitable conditions.
for model-based control. Static design models are quite Since boilers are so common there are many modeling
e!orts. There are complicated models in the form of large
simulation codes which are based on "nite element ap-
proximations to partial di!erential equations. Although
q
This paper was presented at IFAC 13th World Congress, San such models are important for plant design, simulators,
Francisco, CA, 1996. This paper was recommended for publication in
and commissioning, they are of little interest for control
revised form by Associate Editor T.A. Johansen under the direction of
Editor S. Skogestad. design because of their complexity. Among the early
* Corresponding author. Tel.: 00-46-46-222-8781; fax: 00-46-46-138-
work on models suitable for control we can mention
8118. Profos (1955, 1962), Chien, Ergin, Ling and Lee (1958),
E-mail address: kja@control.lth.se (K. J. As stroK m) de Mello (1963), Nicholson (1964), Thompson (1964),

0005-1098/00/$ - see front matter ( 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 5 - 1 0 9 8 ( 9 9 ) 0 0 1 7 1 - 5
364 K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378

Quazza (1968, 1970), Caseau and Godin (1969), Kwan


and Andersson (1970), McDonald and Kwatny (1970),
Speedy, Bell and Goodwin (1970), Dolezal and Varcop
(1970), McDonald, Kwatny and Spare (1971), Eklund
(1971), As stroK m and Eklund (1972), As stroK m (1972), Bell
(1973), Borsi (1974), Lindahl (1976), Tyss+, Brembo and
Lind (1976), Bell, Rees and Lee (1977) and Morton and
Price (1977). Boiler modeling is still of substantial inter-
est. Among more recent publications we can mention
Ma!ezzoni (1988, 1992, 1996), Klefenz (1986), Jarkovsky,
Fessl and Medulova (1988), Unbehauen and Kocaarslan
(1990), HoK ld (1990), Na and No (1992), Kwatny and Berg
(1993), Na (1995). Fig. 1. Schematic picture of the boiler.
The work presented in this paper is part of an ongoing
long-range research project that started with Eklund
(1971) and Bell (1973). The work has been a mixture of circulation in the riser-drum-downcomer loop. Feed-
physical modeling, system identi"cation and model sim- water, q , is supplied to the drum and saturated steam, q ,
& 4
pli"cation. It has been guided by plant experiments in is taken from the drum to the superheaters and the
Sweden and Australia. The unique measurements re- turbine. The presence of steam below the liquid level in
ported in Eklund (1971) have been particularly useful. the drum causes the shrink-and-swell phenomenon
A sequence of experiments with much excitation were which makes level control di$cult. In reality the system
performed on a boiler over a wide range of operating is much more complicated than shown in the "gure. The
conditions. Because of the excitation used, these system has a complicated geometry and there are many
measurements reveal much of the dynamics of interest for downcomer and riser tubes. The out#ow from the risers
control. Results of system identi"cation experiments in- passes through a separator to separate the steam from
dicated that the essential dynamics could in fact be cap- the water. In spite of the complexity of the system it turns
tured by simple models, see As stroK m and Eklund (1972). out that its gross behavior is well captured by global
However, it has not been easy to "nd "rst principles mass and energy balances.
models of the appropriate complexity. Many di!erent A key property of boilers is that there is a very e$cient
approaches have been used. We have searched for the heat transfer due to boiling and condensation. All parts
physical phenomena that yield models of the appropriate of the system which are in contact with the saturated
complexity. Over the years the models have changed in liquid}vapor mixture will be in thermal equilibrium. En-
complexity both increasing and decreasing; empirical co- ergy stored in steam and water is released or absorbed
e$cients have been replaced by physical parameters as very rapidly when the pressure changes. This mechanism
our understanding of the system has increased. The pa- is the key for understanding boiler dynamics. The rapid
pers As stroK m and Eklund (1972, 1975), As stroK m and Bell release of energy ensures that di!erent parts of the boiler
(1988, 1993) and Bell and As stroK m (1996) describe how the change their temperature in the same way. For this
models have evolved. The models have also been used for reason the dynamics can be captured by models of low
control design, see Miller, Bentsman, Drake, Fahkfahk, order. Drum pressure and power dynamics can, in fact,
Jolly, Pellegrinetti and Tse (1990), Pellegrinetti, Be- be represented very well with "rst-order dynamics as
ntsman and Polla (1991), and Cheng and Rees (1997). shown in As stroK m and Eklund (1972). At "rst it is surpris-
Models based on a similar structure have been used for ing that the distributed e!ects can be neglected for a sys-
simulation and control of deaerators, see Lu, Bell and tem with so large physical dimensions.
Rees (1997), and nuclear reactors, see Yeung and Chan Typical values of stored energy for two di!erent boilers
(1990), HoK ld (1990), Irving, Miossec and Tassart (1980), are given in Table 1. The P16-G16 plant is a 160 MW
Parry et al. (1995), Menon and Parlos (1992), Thomas, unit in Sweden and the Eraring plant is a 660 MW unit in
Harrison and Hollywell (1985), Schneider and Boyd Australia. The ratio of the energy stored in the metal to
(1985) and Kothare, Mettler, Morari, Bendotti and that stored in the water is approximately 1 for P16-G16
Falinower (1999). and 4 for the Eraring unit.
The numbers in Table 1 also give a measure of the
time it takes to deplete the stored energy at the gener-
2. Global mass and energy balances ated rate. Although the total normalized stored energy
is approximately the same for both plants the fraction
A schematic picture of a boiler system is shown in of the energy stored in water is much smaller for the
Fig. 1. The heat, Q, supplied to the risers causes boiling. larger plant. This results in larger variations in
Gravity forces the saturated steam to rise causing a water level for the larger plant under proportionally
K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378 365

Table 1
Energy stored in metal, water, and steam for two boilers operating at rated pressure and temperature
but at di!erent power generation conditions. The values are normalized with the power at the
operating conditions. The unit is J/W"s, the entries can thus be interpreted as time constants for
the di!erent storage mechanisms

Boiler Metal Water Steam Total

P16-G16 80 MW 641 739 64 1444


P16-G16 160 MW 320 333 37 690
Eraring 330 MW 1174 303 60 1537
Eraring 660 MW 587 137 35 759

similar operating condition changes. This implies that where < and < represent the total steam and water
45 85
the level control problem is more di$cult for large volumes, respectively. The total volume of the drum,
boilers. downcomer, and risers, < is
5
2.1. Balance equations < "< #< . (4)
5 45 85
Much of the behavior of the system is captured by The metal temperature t can be expressed as a func-
global mass and energy balances. Let the inputs to the .
tion of pressure by assuming that changes in t are
system be the heat #ow rate to the risers, Q, the feedwater .
strongly correlated to changes in the saturation temper-
mass #ow rate, q , and the steam mass #ow rate, q .
& 4 ature of steam t and thus also to changes in p. Simula-
4
Furthermore, let the outputs of the system be drum tions with models having a detailed representation of
pressure, p, and drum water level, l. This way of charac- the temperature distribution in the metal show that
terizing the system is convenient for modeling. For simu- the steady-state metal temperature is close to the satura-
lation and control it is necessary to account for the fact tion temperature and that the temperature di!erences
that mass #ow rate q depends on the pressure by
4 also are small dynamically. The right-hand side of Eq. (3)
modeling the turbine and the superheaters. represents the energy #ow to the system from fuel and
To write the equations, let < denote volume, . denotes feedwater and the energy #ow from the system via the
speci"c density, u speci"c internal energy, h speci"c en- steam.
thalpy, t temperature and q mass #ow rate. Furthermore,
let subscripts s, w, f and m refer to steam, water, feed-
water, and metal, respectively. Sometimes, for clari"ca- 2.1.1. A second-order model
tion, we need a notation for the system components. For Eqs. (1), (3), and (4) combined with saturated
this purpose we will use double subscripts where t de- steam tables yields a simple boiler model. Mathe-
notes total system, d drum and r risers. The total mass of matically, the model is a di!erential algebraic system.
the metal tubes and the drum is m and the speci"c heat Such systems can be entered directly in modeling
5
of the metal is C . languages such as Omola and Modelica and it can be
p
The global mass balance is simulated directly using Omsim, see Mattsson, Anders-
son and As stroK m (1993) or Dymola. In this way we avoid
d
[. < #. < ]"q !q (1) making manual operations which are time consuming
dt 4 45 8 85 & 4
and error prone.
and the global energy balance is We will, however, make manipulations of the model to
obtain a state model. This gives insight into the key
d physical mechanisms that a!ect the dynamic behavior of
[. u < #. u < #m C t ]"Q#q h !q h .
dt 4 4 45 8 8 85 5 p . & & 4 4 the system. There are many possible choices of state
(2) variables. Since all parts are in thermal equilibrium it is
natural to choose drum pressure p as one state variable.
Since the internal energy is u"h!p/. , the global This variable is also easy to measure. Using saturated
energy balance can be written as steam tables, the variables . , . , h , and h can then be
4 8 4 8
d expressed as functions of steam pressure. The second
[. h < #. h < !p< #m C t ] state variable can be chosen as the total volume of water
dt 4 4 45 8 8 85 5 5 p .
in the system, i.e. < . Using Eq. (4) and noting that < is
85 5
"Q#q h !q h , (3) constant, < can then be eliminated from Eqs. (1) and (3)
& & 4 4 45
366 K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378

to give the following state equations: get the following approximate model:
d< dp dp
e 85 #e "q !q e "Q!q (h !h )!q h , (7)
11 dt 12 dt & 4 1 dt & 8 & 4 #

d< dp where
e 85 #e "Q#q h !q h , (5)
21 dt 22 dt & & 4 4
L. Lh Lh
e "h < 4 #. < 4 #. < 8
where 1 # 45 Lp 4 45 Lp 8 85 Lp

e ". !. Lt
11 8 4 #m C 4 !< .
5 p Lp 5
L. L.
e "< 4 #< 8 The term < in e comes from the relation between
12 45 Lp 85 Lp 5 1
internal energy and enthalpy. This term is often neglected
e ". h !. h in modeling, see Denn (1987). The relative magnitudes of
21 8 8 4 4 the terms of e for two boilers are given in Table 2. The
1
A B
L. Lh terms containing Lh /Lp and Lt /Lp are the dominating
e "< h 4 #. 4 8 4
22 45 4 Lp 4 Lp terms in the expression for e . This implies that the
1
changes in energy content of the water and metal masses

A B
L. Lh
#< h 8 #. 8 are the physical phenomena that dominate the dynamics
85 8 Lp 8 Lp of drum pressure. A good approximation of e is
1
Lt Lh Lt
! < #m C 4 . (6) e +. < 8 #mC 4 .
5 5 p Lp 1 8 85 Lp p Lp
This model captures the gross behavior of the boiler Table 2 gives good insight into the physical mechanisms
quite well. In particular it describes the response of drum that govern the behavior of the system. Consider for
pressure to changes in input power, feedwater #ow rate, example the situation when the pressure changes. The
and steam #ow rate very well. The model does, however, change in stored energy for this pressure change will be
have one serious de"ciency. Although it describes the proportional to the numbers in the last two columns of
total water in the system it does not capture the behavior the table. The column for the steam (Lh /Lp) indicates
of the drum level because it does not describe the distri- 4
that energy changes in the steam are two orders of
bution of steam and water in the system. magnitude smaller than the energy changes in water and
metal. The balance of the change in energy is used in the
2.2. Further simplixcations boiling or condensation of steam. The condensation #ow
rate is
Additional simpli"cations can be made if we are only

A
h !h 1 dh dh
interested in the drum pressure. Multiplying (1) by q " 8 &q # . < 4 #. < 8
h and subtracting the result from (3) gives #5 h & h 4 45 dt 8 85 dt
8 # #

B
d dh dh dp dp dt
h (. < )#. < 4 #. < 8 !< !< #m C 4 . (8)
# dt 4 45 4 45 dt 8 85 dt 5 dt 5 dt 5 p dt

dt Model (7) captures the responses in drum pressure to


# m C 4 "Q!q (h !h )!q h , changes in heat #ow rate, feedwater #ow rate, feedwater
5 p dt & 8 & 4 #
temperature and steam #ow rate very well. An attractive
where h "h !h is the condensation enthalpy. feature is that all parameters are given by steam tables
# 4 8
If the drum level is controlled well the variations in the and construction data. The equation gives good insight
steam volume are small. Neglecting these variations we into the nonlinear characteristics of the pressure

Table 2
Numerical values of the terms of the coe$cient e at normal operating pressure
1
Boiler h < /. 4 . < /h4 . < /h8 m C /t4 <
# 45 /p 4 45 /p 8 85 /p 5 p /p 5
P16-G16 80 MW 360 !40 2080 1410 85
P16-G16 160 MW 420 !40 1870 1410 85
Eraring 330 MW 700 !270 2240 4620 169
Eraring 660 MW 810 !270 2020 4620 169
K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378 367

response, since both e and the enthalpies on the Let a denote the mass fraction of steam in the #ow, i.e.
1 .
right-hand side of the equation depend on the operating the quality of the mixture, and let h and h denote the
4 8
pressure. To obtain a complete model for simulating speci"c enthalpies of saturated steam and water. The spe-
the drum pressure a model for the steam valve has to be ci"c internal energy of the mixture of steam and water is
supplied.
h"a h #(1!a )h "h #a h . (9)
The pressure model given by Eq. (7) is similar to the . 4 . 8 8 . #
models in de Mello (1963), Quazza (1968), As stroK m and In steady state we get
Eklund (1972), and Ma!ezzoni (1988). Models similar to
(7) are included in most boiler models. Since model (7) is Lq
"0,
based on global mass and energy balances it cannot Lz
capture phenomena that are related to the distribution of
Lqh La QA
steam and water in the boiler. Therefore it cannot model "qh . "
Lz # Lz <
the drum level.
and it then follows from Eq. (9) that
QA
3. Distribution of steam in risers and drum a " z.
. qh <
#
To obtain a model which can describe the behavior of Let m be a normalized length coordinate along the risers
the drum level we must account for the distribution and let a be the steam quality at the riser outlet. The
of steam and water in the system. The redistribution of 3
steam fraction along the tube is
steam and water in the system causes the shrink-and-
swell e!ect which causes the nonminimum-phase behav- a (m)"a m, 04m41. (10)
. 3
ior of level dynamics, see Kwatny and Berg (1993). One A slight re"nement of the model is to assume that
manifestation is that the level will increase when the boiling starts at a distance x from the bottom of the
steam valve is opened because the drum pressure will 0
risers. In this case the steam distribution will be charac-
drop, causing a swelling of the steam bubbles below the terized by two variables a and x instead of just a . For
drum level. 3 0 3
the experimental data in this paper it adds very little to
The behavior of two phase #ow is very complicated the prediction power of the model. For this reason we use
and is typically modeled by partial di!erential equations, the simpler model although the modi"cation may be
see Kutateladze (1959) and Heusser (1996). A key contri- important for other boilers. There is actually a slip be-
bution of this paper is that it is possible to derive relative- tween water and steam in the risers. To take this into
ly simple lumped parameter models that agree well with account requires much more complicated models. The
experimental data. justi"cation for neglecting this is that it does not have
a major in#uence on the "t to experimental data.
3.1. Saturated mixture quality in a heated tube The volume and mass fractions of steam are related
through a "f (a ), where
7 .
We will start by discussing the dynamics of water and
. a
steam in a heated tube. Consider a vertical tube with f (a )" 8 . . (11)
uniform heating. Let . be the density of the steam}water . . #(. !. )a
4 8 4 .
mixture. Furthermore let q be the mass #ow rate, A the It has been veri"ed that the simple model which uses
area of the cross section of the tube, < the volume, h the a linear steam-mass fraction given by Eq. (10) and
speci"c enthalpy, and Q the heat supplied to the tube. All a steam-volume fraction given by Eq. (11) describes quite
quantities are distributed in time, t, and space, z. Assume well what happens in a typical riser tube. This is illus-
for simplicity that all quantities are the same in a cross trated in Fig. 2 which compares the steam distribution in
section of the tube. The spatial distribution can then be a tube computed from Eqs. (10) and (11) with computa-
captured by one coordinate z and all variables are then tions from a detailed computational #uid dynamics code
functions of z and time t. for a riser tube in a nuclear reactor. The complex code
The mass and energy balances for a heated section of also takes into account that there is a slip between the
the tube are #ow of steam and water. It is interesting to see that the
simple model captures the steam distribution quite well.
L. Lq
A # "0,
Lt Lz 3.2. Average steam volume ratio

L. h 1 Lqh Q To model drum level it is essential to describe the total


# " .
Lt A Lz < amount of steam in the risers. This is governed by the
368 K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378

the #ow rate is instead driven by the density gradients in


the risers and the downcomers. The momentum balance
for the downcomer riser loop is
dq k q2
( # ) $# "(. !. )a6 < g! $# ,
3 $# dt 8 4 7 3 2. A
8 $#
where k is a dimensionless friction coe$cient, and
3
are lengths and A is the area. This is a "rst-order
$# $#
system with the time constant
( # )A .
" 3 $# $# 8 .
kq
$#
Fig. 2. Comparison of steady-state steam volume distribution cal- With typical numerical values we "nd that the time
culated from Eqs. (10) and (11) (full lines) with results of numerical constant is about a second. This is short in comparison
solutions of detailed partial di!erential equation models (circles).
with the sampling period of our experimental data which
is 10 s and we will therefore use the steady-state relation
average volume fraction in the risers. Assume that the 1kq2 ". A (. !. )ga6 < . (15)
2 $# 8 $# 8 4 7 3
mass fraction is linear along the riser as expressed by
Eq. (10) we "nd that the average volume fraction a6 is 3.5. Distribution of steam in the drum
7
given by
The physical phenomena in the drum are complicated.

P P
1 1 a3 .
a6 " a (m)" f (m) dm" 8 Steam enters from many riser tubes, feedwater enters
7 7 a . !. through a complex arrangement, water leaves through
0 3 0 8 4
the downcomer tubes and steam through the steam valve.
A A BB
. . !.
] 1! 4 ln 1# 8 4a . (12) Geometry and #ow patterns are complex. The basic
(. !. )a . 3
8 4 3 4 mechanisms are separation of water and steam and con-
densation.
3.3. A lumped parameter model Let < and < be the volume of steam and water
4$ 8$
under the liquid level and let the steam #ow rate through
Since we do not want to use partial di!erential equa- the liquid surface in the drum be q . Recall that q is the
tions they will be approximated using the Galerkin 4$ 3
#ow rate out of the risers, q the feedwater #ow rate and
method. To do this it will be assumed that the steam- &
q the downcomer #ow rate. The mass balance for the
mass quality distribution is linear, i.e. Eq. (10) holds, also $#
steam under the liquid level is
under dynamic conditions.
The transfer of mass and energy between steam and d
(. < )"a q !q !q , (16)
water by condensation and evaporation is a key element dt 4 4$ 3 3 4$ #$
in the modeling. When modeling the phases separately
where q is the condensation #ow which is given by
the transfer must be accounted for explicitly. This can be #$

A
avoided by writing joint balance equations for water and h !h 1 dh dh
q " 8 &q # . < 4 #. < 8
steam. The global mass balance for the riser section is #$ h & h 4 4$ dt 8 8$ dt
# #

B
d dp dt
(. a6 < #. (1!a6 )< )"q !q , (13) !(< #< ) #m C 4 . (17)
dt 4 7 3 8 7 3 $# 3 4$ 8$ dt $ p dt
where q is the total mass #ow rate out of the risers and The #ow q is driven by the density di!erences of water
3 4$
q is the total mass #ow rate into the risers. The global and steam, and the momentum of the #ow entering the
$#
energy balance for the riser section is drum. Several models of di!erent complexity have been
d attempted. Good "t to the experimental data have been
(. h a6 < #. h (1!a6 )< !p< #m C t ) obtained with the following empirical model:
dt 4 4 7 3 8 8 7 3 3 3 p 4
.
" Q#q h !(a h #h )q . (14) q " 4 (< !<0 )#a q #a b(q !q ). (18)
$# 8 3 # 8 3 4$ 4$ 4$ 3 $# 3 $# 3
$
3.4. Circulation yow Here <0 denotes the volume of steam in the drum in the
4$
hypothetical situation when there is no condensation of
For a forced circulation boiler downcomer #ow rate, steam in the drum and is the residence time of the
$
q is a control variable. For a natural circulation boiler steam in the drum.
$#
K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378 369

3.6. Drum level 4.3. Riser dynamics

Having accounted for the distribution of the steam The mass and energy balances for the risers are given
below the drum level, we can now model the drum level. by Eqs. (13) and (14). Eliminating the #ow rate out of the
The volume of water in the drum is risers, q , by multiplying Eq. (13) by !(h #a h ) and
3 8 3 #
adding to Eq. (14) gives,
< "< !< !(1!a6 )< . (19)
8$ 85 $# 7 3
d d
The drum has a complicated geometry. The linearized (. h a6 < )!(h #a h ) (. a6 < )
behavior can be described by the wet surface A at the dt 4 4 7 3 8 3 # dt 4 7 3
$
operating level. The deviation of the drum level l mea-
d
sured from its normal operating level is # (. h (1!a6 )< )
dt 8 8 7 3
< #<
l" 8$ 4$ "l #l . (20)
A 8 4 d
$ !(h #a h ) (. (1!a6 )< )
8 3 # dt 8 7 3
The term l represents level variations caused by
8
changes of the amount of water in the drum and the dp dt
term l represents variations caused by the steam in the !< #m C 4 "Q!a h q .
4 3 dt 3 p dt 3 # $#
drum.
This can be simpli"ed to

4. The model d dh
h (1!a ) (. a6 < )#. (1!a6 )< 8
# 3 dt 4 7 3 8 7 3 dt
Combining the results of Sections 2 and 3 we can now
obtain a model that gives a good description of the boiler d dh
including the drum level. The model is given by the !a h (. (1!a6 )< )#. a6 < 4
3 # dt 8 7 3 4 7 3 dt
di!erential equations (1), (3), (13), (14), and (16). In addi-
tion there are a number of algebraic equations. The dp dt
circulation #ow rate q is given by the static momentum !< #m C 4 "Q!a h q . (21)
$# 3 dt 3 p dt 3 # $#
balance (15), the steam #ow rate through the liquid
surface of the drum q by (18), and the drum level l by
4$ If the state variables p and a are known the riser
Eq. (20). The volumes are related through Eqs. (4) and 3
#ow rate q can be computed from Eq. (13). This
(19). The model is a di!erential algebraic system, see 3
gives
Hairer, Lubich and Roche (1989). Since most available
simulation software requires state equations we will also
d d
derive a state model. q "q ! (. a6 < )! (. (1!a6 )< )
3 $# dt 4 7 3 dt 8 7 3

4.1. Selection of state variables d


"q !< ((1!a6 ). #a6 . )
$# 3 dt 7 8 7 4
State variables can be chosen in many di!erent ways. It
is convenient to choose states as variables with good
d
physical interpretation that describe storage of mass, "q !< (. !a6 (. !. ))
$# 3 dt 8 7 8 4
energy, and momentum. The accumulation of water is
represented by the total water volume < . The total
85 L dp
energy is represented by the drum pressure p and the "q !< ((1!a6 ). #a6 . )
distribution of steam and water is captured by the $# 3 Lp 7 8 7 4 dt
steam-mass fraction in the risers a and the steam volume
3 La6 da
in the drum < . #< (. !. ) 7 3 .
4$ 3 8 4 La dt
(22)
3
4.2. Pressure and water dynamics
4.4. Drum dynamics
State equations for pressure p and the total amount of
water < in the systems were obtained from the global The dynamics for the steam in the drum is obtained
85
mass and energy balances, Eqs. (1) and (3). These equa- from the mass balance (16). Introducing expression (22)
tions can be written as (5). for q , expression (17) for q and expression (18) for
3 #$
370 K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378

A B
q into this equation we "nd L. Lh
4$ # (1!a )h 4 #. 4 a6 <
3 # Lp 4 Lp 7 3

A
d< d. 1 dh dh
. 4$ #< 4# . < 4 #. < 8
4 dt 4$ dt h 4 4$ dt 8 8$ dt La6
# #(. #(. !. )a ) h < 7
4 8 4 3 # 3 Lp

B
dp dt
!(< #< ) #m C 4
4$ 8$ dt $ p dt Lt
!< #m C 4 , (25)
3 3 p Lp
d
#a (1#b)< ((1!a6 ). #a6 . )
3 3 dt 7 8 7 4 La6
e "((1!a ). #a . )h < 7
33 3 4 3 8 # 3 La
. h !h 3
" 4 (<0 !< )# & 8q . (23)
A
4$ 4$ h & L. 1 Lh Lh

$ # e "< 4# . < 4 #. < 8 !<
42 4$ Lp h 4 4$ Lp 8 8$ Lp 4$
Many of the complex phenomena in the drum are cap- #

B
tured by this equation. Lt
!< #m C 4 #a (1#b)<
8$ $ p Lp 3 3
4.5. Summary

A B
L. L. La6
a6 4 #(1!a6 ) 8 #(. !. ) 7 ,
The state variables are: drum pressure p, total water 7 Lp 7 Lp 4 8 Lp
volume < , steam quality at the riser outlet a , and vol-
85 3 La6
ume of steam under the liquid level in the drum < . The
4$ e "a (1#b)(o !o )< 7 ,
time derivatives of these variables are given by Eqs. (5), 43 3 4 8 3 La
3
(21), and (23). Straightforward but tedious calculations
e ". .
show that these equations can be written as 44 4
In addition steam tables are required to evaluate
d< dp
e 85 #e "q !q , h , h , o , o , Lo /Lp, Lo /Lp, Lh /Lp, Lh /Lp, t , and Lt /Lp
11 dt 12 dt & 4 4 8 4 8 4 8 4 8 4 4
at the saturation pressure p. The results in Sections 5 and
6 are based on approximations of steam tables with
d< dp
e 85 #e "Q#q h !q h , quadratic functions. More elaborated approximations
21 dt 22 dt & & 4 4
with table look-up and interpolation have been tried but
dp da the di!erences in the dynamic responses are not signi"-
e #e 3 "Q!a h q , (24) cant.
32 dt 33 dt 3 # $#
The steam volume fraction a6 is given by Eq. (12), the
7
volume of water in drum < by Eq. (19), the drum level
e
dp da
3 #e
d<
4$ 8$
42 dt
#e
43 dt 44 dt l by Eq. (20), and the downcomer mass #ow rate q by
$#
Eq. (15).
. h !h The partial derivatives of the steam volume fraction
" 4 (<0 !< )# & 8q , with respect to pressure and mass fraction are obtained
4$ 4$ h &
$ # by di!erentiating Eq. (12). We get
where h "h !h and
# 4 8
A B
La6 1 Lo Lo
7" o 4 !o 8
e ". !. , Lp (o !o )2 8 Lp 4 Lp
11 8 4 8 4

A B
L. L. o 1 o #o
e "< 8 #< 4, 1# 8 ! 4 8 ln(1#g) (26)
12 85 Lp 45 Lp o 1#g go
4 4

A B
e ". h !. h , La6 . 1 1
21 8 8 4 4 7 " 8 ln(1#g)! ,
La . g g 1#g
3 4
A B
L. Lh
e "< h 8 #. 8
22 85 8 Lp 8 Lp where g"a (. !. )/. .
3 8 4 4
It is also of interest to know the total condensation

A B
L. Lh Lt #ow rate q and the #ow rate out of the risers q . These
#< h 4 #. 4 !< #m C 4 , #5 3
45 4 Lp 4 Lp 5 5 p Lp #ows are given by Eqs. (8) and (22), hence

A B A A BB
Lh L. . . . !.
e " . 8 !a h 8 (1!a6 )< a6 " 8 1! 4 ln 1# 8 4a ,
32 8 Lp 3 # Lp 7 3 7 . !. (. !. )a . 3
8 4 8 4 3 4
K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378 371

< "< !< !(1!a6 )< , 4.7. Parameters


8$ 85 $# 7 3
< #<
l" 8$ 4$ , An interesting feature of the model is that it requires
A
$ only nine parameters:
o <0
" 4 4$ , f drum volume < ,
$
$ q
4$ f riser volume < ,
3
2. A (. !. )ga6 < f downcomer volume < ,
q2 " 8 $# 8 4 7 3 $#
$# f drum area A at normal operating level,
k $
f total metal mass m ,
5
A
h !h 1 Lh Lh f total riser mass m ,
q " 8 &q # . < 4 #. < 8 3
#5 h & h 4 st Lp 8 85 Lp f friction coe$cient in downcomer-riser loop k,
# #
f residence time of steam in drum,
$
B
Lt dp f parameter b in the empirical equation (18).
!< #m C 4 ,
5 5 p Lp dt
A convenient way to "nd the parameter k is to compute it

A
L. L.
q "q !< a6 4 #(1!a6 ) 8 from the circulation #ow rate. Perturbation studies have
3 $# 3 7 Lp 7 Lp shown that the behavior of the system is not very sensi-
tive to the parameters.
B
La6 dp La6 da
#(o !o ) 7 #(o !o )< 7 3 . The parameters used in this paper were based on
8 4 Lp dt 8 4 3 La dt
3 construction data. Some of them were quite crude.
Gray-box identi"cation, Bohlin (1991) was used in
4.6. Structure of the equations
a comprehensive investigation in Eborn and S+rlie (1997)
and S+rlie and Eborn (1999). Parameters were estimated
Note that Eq. (24) has an interesting lower triangular
and hypothesis testing was used to compare several
structure where the state variables can be grouped as
model structures. The results showed that pressure dy-
(((< , p),a ), < ) where the variables inside each paren-
85 3 4$ namics can be improved signi"cantly by increasing the
thesis can be computed independently. The model can
metal masses. Signi"cant improvements can also be ob-
thus be regarded as a nesting of a second-, a third-, and
tained by adjusting the coe$cients in the calibration
a fourth-order model. The second-order model describes
formula for the sensors.
drum pressure and total volume of water in the system.
In S+rlie and Eborn (1999) it was shown that the
The equations are global mass and energy balances.
friction coe$cient is not identi"able from the data. There
There is a very weak coupling between these equations
is in fact a relation between the initial steam quality and
which is caused by the condensation #ow. The third-
friction. A consequence of this is that it is highly desirable
order model captures the steam dynamics in the risers
for accurate modeling to measure the circulation #ow.
and the fourth-order model also describes the dynamics
This could also be an important signal to use in a level
of steam below the water surface in the drum. The third
control system.
equation is a combination of mass and energy balances
In Eborn and S+rlie (1997) hypothesis testing was
for the riser, and the fourth equation is a mass balance for
applied to the models As stroK m and Bell (1988, 1993) (third
steam under the water level in the drum.
order) and Bell and As stroK m (1996) (fourth order). These
Linearizing Eq. (24) shows that the system has
studies showed conclusively that the improvements ob-
a double pole at the origin and poles at !h q /e and
# $# 33 tained with the fourth-order model are signi"cant. Com-
!1/ . One pole at the origin is associated with water
$ putations on "fth-order models with a more detailed
dynamics and the other with pressure dynamics. The pole
representation of the drum showed that the increased
associated with pressure dynamics is at the origin be-
complexity could not be justi"ed.
cause the steam #ow is chosen as a control variable. The
pole moves into the left half-plane when the drum is
connected to the turbines. The poles !h q /e and 4.8. Equilibrium values
# $# 33
!1/ are associated with dynamics of steam in the
$ The steady-state solution of Eq. (24) is given by
risers and the drum.
The nested structure re#ects how the model was de- q "q ,
veloped. The third-order model is an improved version of & 4
the models in As stroK m and Bell (1988). In Bell and As stroK m Q"q h !q h ,
4 4 & &
(1996) we added drum dynamics as a fourth-order
model. The model in this paper is a re"ned version of that Q"q a h ,
$# 3 #
model. Di!erent models of higher order have also been (h !h )
< "<0 ! $ 8 & q,
developed. 4$ 4$ . h &
4 #
372 K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378

where q is given by Eq. (15), i.e. Dymola, Elmqvist (1978), Omola, Mattsson et al. (1993),
$#
or Modelica, Elmqvist, Mattsson and Otter (1998). In

S
2. A (. !. )ga6 < these languages the model is described in its most basic
q " 8 $# 8 4 7 3.
$# k form in terms of di!erential algebraic equations. In our
case this means that the basic mass and energy balances,
A convenient way to "nd the initial values is to "rst such as (1) and (2), are entered into the system together
specify steam #ow rate q and steam pressure p. The
4 with the algebraic equations, such as (4). The software
feedwater #ow rate q and the input power Q are then
& then makes algebraic manipulations symbolically to sim-
given by the "rst two equations and the steam volume in plify the equations for e$cient simulation.
the drum is given by the last equation. The steam quality
a is obtained by solving the nonlinear equations
3

S
2. A (. !. )ga6 < 5. Step responses
Q"a h 8 $# 8 4 7 3
3 # k
To illustrate the dynamic behavior of the model we

A A BB
. . . !. will simulate responses to step changes in the inputs.
a6 " 8 1! 4 ln 1# 8 4a .
7 . !. (. !. )a . 3 Since there are many inputs and many interesting vari-
8 4 8 4 3 4
ables we will focus on a few selected responses. One input
(27)
was changed and the others were kept constant. The
The steam volume in the drum can then be computed magnitudes of the changes were about 10% of the nom-
directly. inal values of the signals. To compare responses at di!er-
Eq. (27) de"nes the steam volume ratio a6 as a function ent load conditions the same amplitudes were used at
7
of the input power Q. This function which is shown in high and medium load.
Fig. 3 gives important insight into the shrink and swell
phenomena. The curve shows that a given change in 5.1. Plant parameters
input power gives a larger variation in average steam
volume ratio at low power. This explains why the shrink The parameters used were those from the Swedish
and swell e!ects are larger at low power than at high power plant. The values are < "40 m3, < "37 m3,
$ 3
power. < " 1 1 m 3, A " 2 0 m 2, m " 3 0 0 , 0 0 0 k g , m "
$# $ 5 3
160,000 kg, k"25, b"0.3, and "12 s. The steam
$
4.9. Impact of modeling languages tables were approximated by quadratic functions.

Development of physical models is a tedious iterative 5.2. Fuel yow changes at medium load
process. Di!erent physical assumptions are made,
a model is developed and compared with experiments by Fig. 4 shows the responses of the state variables, the
simulation, parameters may be "tted. Detailed investiga- circulation #ow rate q , the riser #ow rate q , and the
$# 3
tion of the results gives ideas for improvements and total condensation #ow rate q to a step increase in fuel
#5
modi"cations. It is a signi"cant e!ort to transform the #ow rate equivalent to 10 MW. Pressure increases at
equations to state space form because of the algebra approximately constant rate. The reason for this is that
involved. This is re#ected in the manipulations resulting steam #ow out of the drum is constant. Total water
in Eq. (24). Many intermediate steps have actually been volume < increases due to the condensation that oc-
85
omitted in the paper. The modeling e!ort can be reduced curs due to the increasing pressure. Steam quality at the
substantially by using modeling languages such as riser outlet a "rst increases rapidly and then more grad-
3
ually. The volume of steam in the drum "rst increases
a little and it then decreases. The rapid initial increase in
steam volume is due to the fast increase in steam from the
risers. The decrease is due to the increased pressure which
causes condensation of the steam. At the onset of the step
there is a rapid increase in the outlet #ow rate from the
risers. The #ow then decreases to match the downcomer
#ow rate. The #ow rates are equal after about 30 s. The
condensation #ow changes in a step-like manner.

5.3. Steam yow changes at medium load

Fig. 3. Steady-state relation between steam volume ratio a6 and input Fig. 5 shows the responses to a step increase of 10 kg/s
7
power Q. The dashed curve shows the steam mass ratio a . in steam #ow rate at medium load. Pressure decreases
3
K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378 373

Fig. 4. Responses to a step in fuel #ow rate of 10 MW at medium load. Fig. 6. Responses to a step in fuel #ow rate of 10 MW at medium (solid)
and high (dashed) loads.

Fig. 5. Responses to a step in steam #ow rate of 10 kg/s at medium


load. Fig. 7. Responses to a step in steam #ow rate of 10 kg/s at medium
(solid) and high (dashed) loads.

practically linearly because of the increased steam #ow. 5.4. Drum level responses
Total water volume also decreases because of increased
evaporation due to the decreasing pressure. Steam qual- Since the behavior of the drum level is of particular
ity at the riser outlet "rst increases rapidly due to the interest we will show step responses that give good in-
pressure decrease and it then decreases due to the in- sight into drum level dynamics for di!erent operating
creased circulation #ow rate. The volume of the steam in conditions. It follows from Eq. (20) that drum level is the
the drum increases due to the decreased pressure. There sum of l "< /A and l "< /A , which depend on
8 8$ $ 4 4$ $
is a very rapid increase of #ow out of the riser due to the the volumes of water and steam in the drum. Drum
pressure drop. After this initial transient the riser #ow pressure, steam mass, and volume fractions will also be
rate then decreases to match the downcomer #ow rate. shown. Responses for medium and high load will be given
There is a steady increase in both due to the decreased to illustrate the dependence on operating conditions.
pressure. The condensation #ow drops in an almost Fig. 6 shows the response to a step in fuel #ow corre-
step-like fashion because the pressure decreases at sponding to 10 MW. The response in drum level is com-
constant rate. plicated and depends on a combination of the dynamics
374 K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378

of water and steam in the drum. The initial part of the 6.1. Experiments at medium load
swell is due to the rapid initial response of steam that was
also seen in Fig. 4. The response in level is a combination The results of experiments at medium load will be
of two competing mechanisms. The water volume in the described "rst. This is the operating condition where the
drum increases due to increased condensation caused by shrink and swell phenomenon is most pronounced.
the increasing pressure. The volume of the steam in the
drum "rst increases a little and it then decreases because 6.1.1. Fuel yow rate changes
of the increasing pressure. Note that there are signi"cant Fig. 8 shows responses in drum pressure and drum level
changes in steam quality and steam volume ratio for the for perturbations in fuel #ow rate. There is very good
di!erent operating conditions. Compare with Fig. 3. agreement between the model and the experimental data
Fig. 7 shows the response to a step increase in steam for drum pressure and drum water level. Note in particu-
#ow rate of 10 kg/s. There is a strong shrink and swell lar that there is an overshoot in the drum level response
e!ect in this case too. The contributions from the vol- for step changes in fuel #ow. This is caused by the
umes of steam and water have the same sign initially. The interaction between the two state variables that describe
water volume will, however, decrease because of the the dynamics of steam under the liquid level in the drum.
steam #ow.
6.1.2. Changes in feedwater yow rate
Fig. 9 shows the responses to changes in the feedwater
#ow rate. The general character of the responses agrees
6. Comparisons with plant data
well. There are some deviations in the pressure responses
and some of the "ner details of the drum level are exag-
Much of the model development was based on plant
gerated. The pressure does deviate in the 500}1500 s
experiments performed with the P16-G16 unit at OG re-
region, but since the changes in pressure are small the
sundsverket in MalmoK , Sweden in collaboration with
results are considered adequate.
Sydkraft AB. The experiments are described in Eklund
(1971) and As stroK m and Eklund (1972, 1975). They were 6.1.3. Changes in the steam valve
carried out in open loop with the normal regulators Fig. 10 shows the responses to changes in the steam
removed. The signals were "ltered and sampled at a rate valve. There is very good agreement between the model
of 0.1 Hz. To ensure a good excitation of the process and the experimental data. Note that there is a signi"cant
PRBS-like perturbations were introduced in fuel #ow shrink and swell e!ect which is captured very well by the
rate, feedwater #ow rate, and steam #ow rate. The inten- model.
tion was to change one input signal in each experiment.
To ensure that critical variables such as drum water level 6.2. Experiments at high load
did not go outside safe limit we made manual correction
occasionally. This means that several inputs were 6.2.1. Changes in fuel yow rate
changed in each experiment. The steam #ow rate Fig. 11 shows responses in drum pressure and drum
changed in response to pressure changes in all experi- level for perturbations in fuel #ow rate. There is good
ments because we were unable to control it tightly.
A large number of signals were logged during the
experiment. This proved to be very valuable because it
has been possible to use the data for a very large number
of investigations. The experiments were performed both
at high and medium load. In this paper we have used data
where three variables, fuel #ow rate, feedwater #ow rate,
and steam #ow rate were changed. This gives a total of
six experiments which can be used to validate the model.
There were problems with the calibration of the #ow
rate measurement transducers and also some uncertainty
in the e!ective energy content of the oil. The approach
used was to start with the nominal calibration values and
then make small corrections so that the long-term track-
ing between the plant data and the model was as close as
possible. The change in all cases was well within the
transducers accuracy limits. Apart from that, no "ddling
with the coe$cients was made. When showing the results
in the following, we present the primary input variable Fig. 8. Comparison of model (solid line) and plant data (dots) for
and the responses in drum pressure and drum level. perturbations in fuel #ow rate at medium load.
K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378 375

Fig. 9. Comparison of model (solid line) and plant data (dots) for Fig. 11. Comparison of model (solid line) and plant data (dots) for
perturbations in feedwater #ow rate at medium load. perturbations in fuel #ow rate at high load.

Fig. 12. Comparison of model (solid line) and plant data (dots) for
Fig. 10. Comparison of model (solid line) and plant data (dots) for perturbations in feedwater #ow rate at high load.
perturbations in steam #ow rate at medium load.

6.2.3. Changes in the steam valve


Fig. 13 shows the responses to changes in the steam
agreement between the model and the experimental data. #ow rate. There is very good agreement between the
A comparison with Fig. 8 shows that the shrink and swell model and the experimental data. Note that the model
e!ect is much smaller at high load. It is interesting to see captures the drum level variations, particularly the swell
that the model captures this. and shrink e!ect very well. A comparison with Fig. 10
shows that the shrink and swell e!ect is smaller at high
6.2.2. Changes in feedwater yow rate loads. This is well captured by the model.
Fig. 12 shows the responses to changes in feedwater
#ow rate. There are discrepancies in pressure from time 6.3. Comparison of behavior at high and medium load
400 to 1400. The total changes in pressure are small and
minor variations in feedwater conditions can easily cause The experiments have indicated that there are signi"-
variations of this magnitude. The model exaggerates the cant di!erences in behavior at high and medium loads
level changes for rapid variations. Compare for example that are well predicted by the model. We will now look
data in the interval 1500}2000. closer at these di!erences.
376 K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378

Fig. 13. Comparison of model (solid line) and plant data (dots) for Fig. 15. Comparison of behavior of drum water level at medium and
perturbations in steam #ow rate at high load. high loads for perturbations in fuel #ow rate. The model response is
shown in solid lines and plant data is indicated by dots.

Fig. 14. Comparison of behavior of drum water level at medium and


high loads for perturbations in steam #ow rate. The model response is Fig. 16. Comparison of behavior of drum water level at medium and
shown in solid lines and plant data is indicated by dots. high loads for perturbations in fuel #ow rate. The model response is
shown in solid lines and plant data is indicated by dots.

Fig. 14 compares the responses in drum level to steam are substantial rapid variations. Again we note the excel-
#ow changes at high and medium loads. The steam valve lent agreement between model and experiments and we
changes were almost the same in both experiments (as is also note the signi"cant di!erence between the behaviors
shown in Figs. 10 and 13) but they were not identical. at medium and high loads.
Because of the integrators in the model there are natural
di!erences in the levels of the signals. Apart from this
level shift the model matches the experiments very well. 7. Conclusions
Fig. 14 also shows that the shrink and swell e!ect is
larger at medium load. It is even larger at low loads. A nonlinear physical model with a complexity that is
Fig. 15 compares responses to changes in fuel #ow. suitable for model-based control has been presented. The
Note the good agreement for dynamic responses between model is based on physical parameters for the plant and
the model and experiments. In this case there is a pro- can be easily scaled to represent any drum power station.
nounced di!erence between the behaviors for medium The model has four states; two account for storage of
and high load. Fig. 16 compares responses to changes in total energy and total mass, one characterizes steam
fuel #ow. We have taken a section of the data where there distribution in the risers and another the steam distribu-
K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378 377

tion in the drum. The model can be characterized by As stroK m, K. J., & Eklund, K. (1972). A simpli"ed non-linear model for
steam tables and a few physical parameters. a drum boiler * Turbine unit. International Journal of Control, 16,
The model is nonlinear and agrees well with experi- 145}169.
As stroK m, K. J., & Eklund, K. (1975). A simple non-linear drum-boiler
mental data. In particular, the complex shrink and swell model. International Journal of Control, 22, 739}740.
phenomena associated with the drum water level are well Bell, R. D. (1973). The on-line optimal control of constrained non-
captured by the model. The model has a triangular struc- linear processes and its application to steam generators. Ph.D. thesis,
ture that can be described as (((< , p),a ), < ), where the University of N.S.W., Australia.
85 3 4$ Bell, R. D., & As stroK m, K. J. (1996). A fourth order non-linear model for
states in each bracket can be determined sequentially.
drum-boiler dynamics. In IFAC '96, Preprints 13th World Congress
The linearized model has two poles at the origin and two of IFAC, vol. O, San Francisco, CA (pp. 31}36).
real stable poles. Bell, R. D., Rees, N.W., & Lee, K.B. (1977). Models of large boiler-
The model has been validated against plant data with turbine plant. In IFAC symposium on automatic control & protection
very rich excitation that covers a wide operating range. of electric power systems (pp. 469}475). Ins. of Engineers, Aust.
These experiments have given much insight into the be- National Conf. Publ.
Bohlin, T. (1991). Interactive system identixcation: Prospects and pitfalls.
havior of the system and they have guided the modeling Berlin: Springer.
e!ort. We believe that the approach used in this work can Borsi, L. (1974). Extended linear mathematical model of a power station
be applied to other con"gurations of steam generators. unit with a once through boiler. Siemens Forschungs und Entwicklin-
Model (24) can be simpli"ed by keeping only the domi- gsberichte, 3(5), 274}280.
nant terms in expressions (25) for the coe$cients e . This Caseau, P., & Godin, P. (1969). Mathematical modelling of power
ij plants. IE(Australia) Electronic Engineering Transactions.
could be useful for applications to model-based control. Cheng, C. M., & Rees, N. W. (1997). Fuzzy model based control of
Preliminary investigations indicate that several terms can steam generation in drum-boiler power plant. In Proceedings of
be neglected without sacri"cing the "t to experimental IFAC/CIGRE symposium on control of power systems and power plants
data. A comprehensive study of this is outside the scope CPSPP+97, Beijing, China (pp. 175}181). International Academic
of this paper. The model can also be re"ned in several Publishers.
Chien, K. L., Ergin, E. I., Ling, C., & Lee, A. (1958). Dynamic analysis of
ways. This will, however, require new measurements with a boiler. Transactions of ASME, 80, 1809}1819.
faster sampling rates. de Mello, F. P. (1963). Plant dynamics and control analysis. IEEE
Transactions on Power Apparatus and Systems, S82, 664}678. Paper
63-1401.
Denn, M. (1987). Process modelling. New York: Wiley.
Acknowledgements Dolezal, R., & Varcop, L. (1970). Process dynamics. Amstersam: Elsevier
Publ. Co.
The research has been supported by the Sydkraft Re- Eborn, J., & S+rlie, J. (1997). Parameter optimization of a non-linear
search Foundation and the Swedish National Board for boiler model. In Sydow, 15th IMACS world congress, vol. 5
Industrial and Technical Development under contract (pp. 725}730). Berlin, Germany: W & T Verlag.
Eklund, K. (1971). Linear drum boiler-turbine models. Ph.D. thesis
97-04573. This support is gratefully acknowledged. We TFRT-1001, Department of Automatic Control, Lund Institute of
would also like to express our sincere gratitude to Syd- Technology, Lund, Sweden.
kraft AB for their willingness to perform experiments on Elmqvist, H. (1978). A structured model language for large continuous
plants to increase our understanding of their behavior. systems. Ph.D. thesis TFRT-1015, Department of Automatic
Useful comments on several versions of the manuscript Control, Lund Institute of Technology, Lund, Sweden.
Elmqvist, H., Mattsson, S. E., & Otter, M. (1998). Modelica * the
have been given by our colleagues J. Eborn, H. Tummes- new object-oriented modeling language. In Proceedings of the 12th
cheit, and A. Glattfelder. We would also like to express european simulation multiconference (ESM+98). Manchester, UK:
our gratitude to the reviewers for constructive criticism. SCS, The Society for Computer Simulation.
Garcia, C. E., Prett, D. M., & Morari, M. (1989). Model predictive
control: Theory and practice * A survey. Automatica, 25(3), 335}348.
Hairer, E., Lubich, C., & Roche, M. (1989). The numerical solution
References of diwerential-algebraic systems by Runge}Kutta methods. Lecture
Notes in Mathematics, vol. 1409. Berlin: Springer.
Ambos, P., Duc, G., & Falinower, C.-M. (1996). Loop shaping Heusser, P. A. (1996). Modelling and simulation of boiling channels with
H design applied to the steam generator level control in edf a general front tracking approach. San Diego: Society for Computer
=
nuclear power plants. In Proceedings of xfth IEEE conference on Simulation, Inc.
control applications (pp. 751}756). Dearborn, Michigan: IEEE. HoK ld, A. (1990). UTSG * 2 a theoretical model describing the transient
As stroK m, K. J. (1972). Modelling and identi"cation of power system behavior of a pressurized water reactor natural-circulation U-tube
components. In Handschin, Real-time control of electric power sys- steam generator. Nuclear Technology, 90, 98}118.
tems, Proceedings of Symposium on real-time control of electric power Irving, E., Miossec, C., & Tassart, J. (1980). Towards e$cient full
system, Baden, Switzerland, 1971 (pp. 1}28). Amsterdam: Elsevier. automatic operation of the PWR steam generator with water level
As stroK m, K. J., & Bell, R. (1988). Simple drum-boiler models. In IFAC adaptive control. In Boiler dynamics and control in nuclear power
international symposium on power systems, modelling and control (pp. 309}329).
applications. Brussels, Belgium. Jarkovsky, J., Fessl, J., & Medulova, V. (1988). A steam generator
As stroK m, K. J., & Bell, R. D. (1993). A nonlinear model for steam dynamic mathematical modelling and its using for adaptive control
generation process. In Preprints IFAC 12th world congress. Sydney, systems testing. In Preprints IFAC symposium on power systems
Australia. modelling and control applications. Brussels, Belgium.
378 K.J. Asstro( m, R.D. Bell / Automatica 36 (2000) 363}378

Klefenz, G. (1986). Automatic control of steam power plants (3rd ed.). Quazza, G. (1968). Sui modelli analitici delle caldaie a corpo cilindrico.
Bibliographisches Institut. Parte I: Dinamica del vaporizzatore. Automazione e Strumentazione,
Kothare, M. V., Mettler, B., Morari, M., Bendotti, P., & Falinower, C. 506}537.
M. (1999). Level control in the steam generator of a nuclear power Quazza, G. (1970). Automatic control in electric power systems. Auto-
plant. IEEE Transactions on Control Systems Technology, to appear. matica, 6, 123}150.
Kutateladze, S. S. (1959). Heat transfer in condensation and boiling. Schneider, W. G., & Boyd, J. T. (1985). Steam generator level controlla-
Technical Report AEC-tr-3770. United States Atomic Energy bility. In Boiler dynamics and control in nuclear power station (pp.
Commission. 97}117). Proceedings Power Station 3.
Kwan, H. W., & Andersson, J. H. (1970). A mathematical model of a S+rlie, J., & Eborn, J. (1999). A grey-box identixcation case study: The
200 MW boiler. International Journal of Control, 12, 977}998. Asstro( m-Bell Drum}boiler Model. Technical Report. Department of
Kwatny, H. G., & Berg, J. (1993). Drum level regulation at all loads. Automatic Control, Lund Institute of Technology, Lund, Sweden,
In Preprints IFAC 12th world congress, vol. 3, Sydney, Australia to appear.
(pp. 405}408). Speedy, C. B., Bell, R. D., & Goodwin, G. C. (1970). Dynamic modelling
Lindahl, S. (1976). Design and simulation of a coordinated drum of a steam generator using least squares analysis. In Proceedings of
boiler-turbine controller. Lic Tech thesis TFRT-3143, Department of JACC (pp. 365}372). Atlanta, Georgia.
Automatic Control, Lund Institute of Technology, Lund, Sweden. Thomas, P. J., Harrison, T. A., & Hollywell, P. D. (1985). Analysis of
Lu, C. X., Bell, R. D., & Rees, N. W. (1997). Scheduling control of limit cycling on a boiler feedwater control system. In Boiler dynam-
deaerator plant. In Proceedings of IFAC/CIGRE symposium on ics and control in nuclear power station (pp. 89}96). Proceedings of
control of power systems and power plants CPSPP+97, Beijing, China Power Station 3.
(pp. 219}226). International Academic Publishers. Thompson, F. T. (1964). A dynamic model for control of a drum type
Ma!ezzoni, C. (1988). Dinamica dei generatori di vapore. Milano: boiler system. Ph.D. thesis, University of Pittsburg, USA.
Hartman and Braun. Tyss+, A., Brembo, J. C., & Lind, K. (1976). The design of multivariable
Ma!ezzoni, C. (1992). Issues in modeling and simulation of power control system for a ship boiler. Automatica, 12, 211}224.
plants. In Proceedings of IFAC symposium on control of power plants Unbehauen, H., & Kocaarslan, I. (1990). Experimental modelling and
and power systems, vol. 1 (pp. 19}27). adaptive power control of a 750 MV once-through boiler. In Pre-
Ma!ezzoni, C. (1996). Boiler-turbine dynamics in power plant control. prints IFAC 11th world congress on automatic control, vol. 11, Tallinn,
In IFAC 13th triennial world congress, San Francisco, USA. Estonia (pp. 32}37).
Mattsson, S. E., Andersson, M., & As stroK m, K. J. (1993). Object-oriented Yeung, M. R., & Chan, P. L. (1990). Development and validation of
modelling and simulation. In Linkens, CAD for control systems a steam generator simulation model. Nuclear Technology, 92,
(pp. 31}69). New York: Marcel Dekker, Inc. 309}314.
Mayne, D. Q., Rawlings, J. B., & Rao, C. V. (1999). Model predictive
control: A review. Automatica, to appear.
Karl J. As stroK m is Professor and Head of
McDonald, J. P., & Kwatny, H. G. (1970). A mathematical model for
the Department of Automatic Control at
reheat boiler}turbine}generator systems. In Proceedings of IEEE. Lund University since 1965. He has broad
PES winter power meeting, New York. Paper 70 CP221-PWR. interests in automatic control including,
McDonald, J. P., Kwatny, G. H., & Spare, J. H. (1971). A nonlinear stochastic control, system identi"cation,
model for reheat boiler-turbine generation systems. Proceedings adaptive control, computer control and
JACC (pp. 227}236). computer-aided control engineering. He
Menon, N. N., & Parlos, N. N. (1992). Gain-scheduled non-linear has supervised 44 Ph.D. students, written
control of u-tube steam generator water level. Nuclear Science and six books and more than 100 papers in
Engineering, 111, 294}308. archival journals. He is a member of the
Royal Swedish Academy of Engineering
Miller, N., Bentsman, J., Drake, D., Fakhfakh, J., Jolly, T., Pellegrinetti,
Sciences (IVA) and the Royal Swedish
G., & Tse, J. (1990). Control of steam generation processes. In Academy of Sciences (KVA) and a foreign member of the US National
Proceedings of ISA 90, New Orleans, Louisiana (pp. 1265}1279). Academy of Engineering and the Russian Academy of Sciences. As stroK m
Morton, A. J., & Price, P. H. (1977). The controllability of steam output, has received many honors including three honorary doctorates, the
pressure and water level in drum boilers. Proceedings of the Institu- Callender Silver Medal, the Quazza Medal from IFAC, the Rufus
tion of Mechanical Engineers, 75}84. Oldenburger Medal from ASME, the IEEE Control Systems Science
Na, M. G. (1995). Design of a stem generator water level controller via Award and the IEEE Medal of Honor.
the estimation of the #ow errors. Annals of Nuclear Science and
Engineering, 22(6), 367}376.
Rod Bell is an Honorary Senior Research
Na, M. G., & No, H. C. (1992). Design of an adaptive observer-based
Fellow, a position he has held since Febru-
controller for the water level of steam generators. Nuclear Engineer- ary 1998. His previous position was Head
ing and Design, 135, 379}394. of the Computing Department which he
Nicholson, H. (1964). Dynamic optimisation of a boiler. Proceedings of held since the beginning of 1994. He ob-
IEE, 111, 1479}1499. tained his Ph.D. (Application of Optimal
Parry, A., Petetrot, J. F., & Vivier, M. J. (1995). Recent progress in sg Control Theory to Industrial Processes) in
level control in french pwr plants. In Proceedings of International 1972 from the University of New South
Conference on boiler dynamics and control in nuclear powerstations Wales. He joined the academic sta! at
(pp. 81}88). British Nuclear Energy Society. Macquarie University in 1972. He became
interested in computers (hardware and
Pellegrinetti, G., Bentsman, J., & Polla, K. (1991). Control of nonlinear
software) in 1958 when he worked on one
steam generation processes using h= design. In Proceedings of 1991 of the "rst digital computers in Australia, UTECOM at the University
ACC, Boston, Massachusetts (pp. 1292}1297). of New South Wales. His current research interests lie mainly in the
Profos, P. (1955). Dynamics of pressure and combustion control in Application of Control Theory to areas such as Industrial Processes
steam generators. Sulzer Technical Review 4, V37, 1}15. (Power Stations in particular), National Economies, Management and
Profos, P. (1962). Die Regelung von Dampfanlagen. Berlin: Springer. Robotics. He is a member of IEEE. Among his personal interests are
Qin, S. J., & Badgwell, T. A. (1997). An overview of industrial model horse riding, growing waratahs, bush walking and restoring Jaguar
predictive control technology. AIChE Symposium Series, vol. 97. cars.

You might also like