Professional Documents
Culture Documents
Design of dynamic plant model and model-based controller for a heat recovery
system with a swirling flow incinerator
Jaeyoung Cho, Yongtae Kim, Jeongwoo Song, Tae Kyung Lee, Han Ho Song
PII: S0360-5442(17)32020-0
DOI: 10.1016/j.energy.2017.12.001
Reference: EGY 11950
Please cite this article as: Cho J, Kim Y, Song J, Lee TK, Song HH, Design of dynamic plant model and
model-based controller for a heat recovery system with a swirling flow incinerator, Energy (2018), doi:
10.1016/j.energy.2017.12.001.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Title
Design of Dynamic Plant Model and Model-based Controller for a Heat Recovery System with a Swirling Flow
Incinerator
PT
Authors and affiliation:
Jaeyoung Choa, Yongtae Kima, Jeongwoo Songa, Tae Kyung Leea and Han Ho Songa,*
RI
a
Department of Mechanical Engineering, Seoul National University, Gwanak-gu, Seoul, South Korea
*Corresponding author
SC
Corresponding author’s contact information:
Address: Bldg. 301, Rm. 214, Gwanak-ro 1, Gwanak-gu, Seoul, South Korea (Zip: 08826)
U
AN
Fax: +82-2-880-1652
1
ACCEPTED MANUSCRIPT
Abstract
We developed a plant model and model-based controller for a heat recovery system with a swirling flow incinerator
located in South Korea. First, a plant model was desinged to understand the dynamic characteristics of the system.
PT
The system was divided into three parts: incinerator, boiler, and steam drum. Then, each part was modeled with a
multi-zone, zero-dimensional modeling scheme. Consequently, we found that the plant model followed the dynamic
RI
behavior of the target system, particularly for the seven important variables: boiler inlet temperature, superheater
inlet temperature, boiler outlet temperature, oxygen mole fraction, steam production rate, steam pressure, and steam
SC
temperature. Then, we designed a multi-input multi-output controller for the system with a nonlinear control model.
In the controller design, input variables were selected as air flow rate, fuel supply rate and steam valve opening area;
U
while output variables were oxygen mole fraction of exhaust gas, steam production rate and steam pressure. A
nonlinear control model was constructed by reducing the plant model. Then a linear quadratic regulator was applied
AN
while evaluating control gain at each time step. Finally, we validated the controller on the plant model, which
adequately suppressed the disturbance on fuel composition and adjusted the steam production rate.
M
Keywords
D
Waste-to-energy, Solid refuse fuel, Waste incinerator, Model-based control, Multi-input multi-output controller,
TE
2
ACCEPTED MANUSCRIPT
Nomenclature
: Stoichiometric coefficient of the gas reactant of the kth reaction, especially for a solid-gas phase reaction, [-]
: Effective heat transfer surface area of the steam generator (superheater) in the ith zone, [m2]
PT
,
() : Mean specific heat of air (fuel) at constant pressure, [J kg–1 K–1]
(), : Specific heat of the metal part of the drum (superheater), [J kg–1 K–1]
RI
: Concentration of the jth gas species in the ith zone, [kmol m–3]
: Particle diameter, [m]
SC
: Hydraulic diameter of the ith zone, [m]
: Binary diffusion coefficient of the gas reactant of the kth reaction, [m2 s–1]
: Pipe diameter of the ith zone, [m]
U
AN
ℎ() : Specific enthalpy of liquid (vapor) water in the drum, [J kg–1]
(),,
(!") : Inlet (outlet) enthalpy of the jth gas (solid) species into (from) the ith zone, [J s–1]
#$!
D
[m s–1 for a heterogeneous reaction, (kmol m–3)x s–1 for homogenous reactions (x depends on reaction order)]
%'()*()+)
EP
: Log mean temperature difference between the exhaust gas in the ith zone and the steam generator
(superheater), [K]
,(),
: Mass of the jth gas (solid) species in the ith zone, [kg]
,
, : Inlet mass flow rate of total air: core air, top air, and shell air, [kg s–1]
AC
,(),,
: Mass production rate of the jth gas (solid) species by reaction in the ith zone, [kg s–1]
,(),,
(!") : Inlet (outlet) mass flow rate of the jth gas (solid) species into (from) the ith zone, [kg s–1]
,(), (!") : Inlet (outlet) mass flow rate of the liquid (vapor) water into (from) the drum, [kg s–1]
,"
!. : Steam production rate, which is equal to the outlet mass flow rate from the superheater, [kg s–1]
3
ACCEPTED MANUSCRIPT
,)/0, : Inlet mass flow rate or the supply rate of the solid refuse fuel, [kg s–1]
,"!
, : Inlet mass flow rate of the top air: second blower flow rate, [kg s–1]
'1 : Molecular weight of the gas reactant of the kth reaction, [kg kmol–1]
'1
PT
: Mean molecular mass of the jth species, [kg kmol–1]
2$
3
4: The number of char with diameter
in the ith zone, [-]
RI
5
: Pressure in the drum, [Pa]
SC
6$!
. ! : Convective heat transfer rate from the ith zone to the wall, [J s–1]
6!,78()9) : Heat loss rate from the waste incinerator (steam drum), [J s–1]
6:
U
: Convective heat transfer rate from the gas phase to the solid phase in the ith zone, [J s–1]
AN
6(),
;
: Heat release rate by the reaction in the ith zone in the gas (solid) phase, [J s–1]
6
.(),:
: Radiative heat transfer rate to the gas phase in the kth zone from the gas phase in the ith zone, [J s–1]
6
.() !
M
: Radiative heat transfer rate to the wall from the solid phase in the ith zone, [J s–1]
6)*()+)
: Convective heat transfer rate from the exhaust gas in the ith zone to the steam generator (superheater), [J s–1]
D
<9800,
TE
: Diffusive resistance for the kth reaction in the ith zone, [s kg–1]
<=8>,
: Kinetical resistance for the kth reaction in the ith zone, [s kg–1]
( (), : Intake temperature of the air (fuel) into the incinerator, [K]
(?!
(!") :
C
Inlet (outlet) temperature of the exhaust gas into (from) the boiler, [K]
( : Temperature of superheated steam, which is equal to that of the superheater, [K]
AC
@",
: Terminal velocity of the jth species in the ith zone, [m s–1]
A()
: Internal energy of gas (solid) in the ith zone, [J]
&() : Volume of the liquid (vapor) phase of water in the drum, [m3]
4
ACCEPTED MANUSCRIPT
BCD ,.. : Mole fraction of oxygen in the exhaust gas (at steady state), [-]
E : Mass fraction of the jth component in the solid refuse fuel, [-]
PT
E,
: Mass fraction of the gas reactant of the kth reaction on the surface of char in the ith zone, [-]
RI
G : Specific heat ratio of the superheated vapor in the superheater, [-]
H)* : The portion of heat from the exhaust gas transferred to the steam generator, [-]
SC
H"!
,
" : The portion of top air participating in the reaction, [-]
H ? $ (I+J ) :
U
The portion of unburned char (methane), [-]
N,
O
: Diameter growing rate of the jth species in the ith zone, [m s–1]
D
TE
C EP
AC
5
ACCEPTED MANUSCRIPT
1. Introduction
Waste incinerators have been widely used in the past few decades to help recover energy from
waste as their use reduces dependency on fossil fuels and stress on landfill areas. In a typical heat recovery
PT
system, solid refuse fuel (SRF) made from the municipal waste is carried into an incinerator. Subsequently,
the hot exhaust gas from combustion enters a boiler and produces high temperature and pressure steam,
RI
which is then used for either electricity generation or district heating [1]. OECD countries, on average,
recover and convert 19% of municipal solid waste into heat or electricity using an incinerator [2]. To date,
SC
there are 137 facilities in South Korea that convert 24% of the total municipal waste into energy [2, 3].
Most incinerators are either stoker-type or fluidized bed-type. Researchers have recently
U
introduced other types of incinerators. In this work, the swirling flow incinerator (SFI) has been studied,
AN
which generates an intensive swirling flow between the flame and the inner wall. This flow acts as an air
curtain for reducing the heat loss and enhancing the mixing of fuel and air, thus increasing thermal
M
efficiency. Moreover, reduced clinker generation and lower wall temperature induced by an insulation
effect of the swirling flow enhances system durability. There are six heat recovery facilities in South
D
Korea with fourteen SFIs that burn approximately 220 thousand tons of SRF every year.
TE
However, waste incinerator systems are difficult to control because the composition of the SRF
is not well regulated. This can lead the system to exhibit unstable behavior. In South Korea, regulations on
EP
the heating value of SRF only suggest a minimum limit without a maximum [4], contrary to that the
regulations for coal, which specify both and divide the fuel into different grades according to their heating
value [5]. While this rather loose regulation encourages SRF production, it makes the combustion
C
In order to enhance the stability of the waste incinerator operation, some research groups have
designed a controller for different waste incinerators. Kozek and Voigt [6], Leskens et al. [7], and
Miyamoto et al. [8] have used a model-based controller to cover a wide range of nonlinearity; while other
research groups have used either neural networks [9] or the fuzzy method [10, 11] for this purpose, which
6
ACCEPTED MANUSCRIPT
are suitable when there is plenty of data for the system. Other researchers have focused on understanding
the dynamic behavior inside the incinerator. Maris et al. [12] have designed a dynamic model for a
fluidized bed incinerator with SRF by dividing the zones inside the incinerator and Rovaglio et al. [13]
PT
Our research focused on the controller and dynamic model for the system with a SFI, which has
RI
a shorter time constant and exhibits inhomogeneous combustion characteristics in comparison with other
types of incinerators. This makes it inherently difficult to simulate and control. In this study, we designed a
SC
multi-input multi-output (MIMO) controller with a nonlinear control model and a dynamic plant model
that simulates the behavior of the SFI with a boiler. A plant model was used in order to understand the
U
physics of the system on which the control model is based, and as a test platform to validate the controller.
Although we studied a particular type of incinerator installed in South Korea, this work models how to
AN
simplify a complicated combustion system, and how to apply that model to controller design.
M
The main body of this article consists of three sections. Section 2 provides the details of the
target system to be modeled and controlled. Section 3 is comprised of three subsections: 3.1 introduces the
D
assumptions and equations for the incinerator model, 3.2 models the boiler and the steam drum, and 3.3
validates the plant model. Section 4 describes the controller in two subsections: 4.1 deals with the
TE
nonlinear control model and its validation, and 4.2 describes the algorithm of the controller and the
7
ACCEPTED MANUSCRIPT
2. Target system
The waste incinerator and boiler built by ENERONE Inc., located at Ul-San-Si in South Korea,
was chosen as the target system in this study (Figure 1). The system utilizes fluff SRF as a fuel and has
PT
the capacity to incinerate 2 ton h–1 of waste and generate 20 ton h–1 of steam at about 400 °C and 40 bar,
RI
The schematics of the system are shown in Figure 2. As shown in Figure 2 (side view of the
system), SRF is fed from the bottom of the incinerator, and air is blown inside at three different locations.
SC
First, core air is blown along the central axis of incinerator. Then, the flame meets the top air that enters
the incinerator at the bottom wall. The inlet nozzle for top air is tilted toward the inner wall of the
U
incinerator, thus the top air makes intensive swirling flow as it enters the chamber. Finally, the shell air
AN
supplies additional oxygen to the flame. The same blower provides the core and shell air, which are
As the exhaust gas from the incinerator enters the boiler, it first passes through the steam generator 1
and then through the superheater. Next, the exhaust gas passes steam generator 2, and exits to the
D
economizer (not shown in Figure 2), which constitutes the saturated liquid supply to steam drum. The
TE
walls of the boiler are covered with tubes connected to the steam drum, as depicted with blue lines in
Figure 2.
C EP
AC
8
ACCEPTED MANUSCRIPT
In this section, the development of the plant model, which simulated the real plant, has been
described. In a preliminary study, 3D CFD simulations were performed on the SFI using Star-CCM+ with
PT
800,000 spatial grids and four simplified reaction equations. The model, however, takes about two weeks
in an ideal computation environment (CPU clock speed 2.7 GHz, 12 cores, and 24 threads) to simulate
RI
only a few seconds. Therefore, it was deemed an inappropriate tool to design a controller that would
require validation for several hours of operation. Instead, the computational cost was minimized by
SC
adopting a multi-zone, zero-dimensional model with reasonable assumptions. The model was developed in
the MATLAB-Simulink environment and designed in a manner to follow the dynamic characteristics of
U
the target system for a wide range of real plant operations.
AN
3.1. Plant model – swirling flow incinerator
M
The typical temperature and velocity field in the incinerator derived from the 3D CFD
simulation are shown in Figure 3. The flame is initiated with the core air and continues with the top
D
and shell air, while the swirling flow makes an air curtain around the flame. Based on this structure,
TE
the thermodynamic state of the incinerator was simplified into three zones, as shown in Figure 4. The
first zone, also called as the primary combustion zone (PCZ), is the region where SRF is devolatilized
and reacts with the core air. The products from the PCZ flow into second combustion zone (SCZ) and
EP
react with the shell air and top air. The last zone is the air curtain zone (ACZ), which is a characteristic
region of the SFI between the flame and the inner wall.
C
,,
= ,,,
− ,,,!"
+ ,,,
(T)
P
AC
,,!"
where ,,,!"
= ,,
,
A
= Y ,,
− Y ,,!"
+ 6,
;
− 6$!
− Y 6
.(),:
− 6:
([)
P . !
Z
9
ACCEPTED MANUSCRIPT
,,
5 & 3'1
/( 4
Y = (])
P < P
Each zone was split into the solid and gas phases, and mass conservation, energy conservation,
PT
and ideal gas laws were applied for the gas phase in Eqs. (1)-(3). Using the velocity field from the 3D
simulation results shown in Figure 3, the mass transfer taking place between each zone, defined as
,,,
(!") ,
RI
was simplified. The diffusive mass transfer was neglected as the convection velocity is as
fast as 5–20 m s–1. Furthermore, only the convective mass transfer from PCZ and ACZ to SCZ was
SC
considered because other convections are relatively small. Therefore, the inlet mass flow rate into
ACZ, ,,,
^I_
, only comprises the shell air and top air supply rates. Similarly, ,,,
`I_
describes the
U
denotes the convective mass transfers from PCZ and ACZ.
AN
Meanwhile, the pressure inside the incinerator was assumed to remain unchanged (atmospheric
pressure), even though in a real system it is slightly negative in order to prevent a blowout.
The species conversion and heat release by gas phase reaction was modeled as ,,,
and
M
6,
;
. The ACZ experiences convective heat loss to the wall, calculated as 6$!
. ! . Furthermore,
D
there is radiative heat transfer amongst the zones because the product gas contains gray gases such as
CO2 and H2O, and the amount of heat transfer, ∑Z 6
.(),:
,was evaluated based on the
TE
configuration factor and the total emittance from Hottel’s charts. Both the solid and gas phases exist
EP
within the PCZ and SCZ, and the convective heat transfer between these phases was calculated as
6:
e
c,d fce
. Subsequently, the remaining (2 + 2) unknowns, ,,!"
" "
, , and , were calculated by
g,,
3
4 gN,
3
4,,
3
4
= ,,,
3
4 − ,,,!"
3
4 + ,,,
3
4 − (h)
gP g
2
where N,
3
4 = ,,,
(
)
K$
i
j 2$
(
)
10
ACCEPTED MANUSCRIPT
A
= Y ,,
− Y ,,!"
+ 6,
;
− 6
.() !
+ 6:
(k)
P
Equations (4) and (5) define the mass conservation and energy conservation for the solid phase,
respectively. As in the case of the gas phase, the mass transfers from the other zones to the ACZ, from
PT
SCZ to other zones, and from ACZ to PCZ were neglected. Furthermore, the diameter growing term,
e e
lmn,d 3M 4n,d 3M 4
, was included in order to account for the mass change of the given diameter, which
RI
lM
affects the overall surface-to-volume ratio of the solid particles and leads to a change in the reaction
SC
rate [14]. Here, N3
4 is the rate of the diameter change, which is negative when the diameter
decreases.
U
In this study, we assumed that all the particles are spherical, and the diameter of the incoming
AN
SRF assumes a Gaussian distribution, as shown in Eq. (6). The maximum diameter for the fluff
SRF,
,; , is 50 mm by law. However, this value was modified to 15 mm, considering 0.3 of
M
,,)/0,
`I_
3
4 1 ttt
4
3
−
j
= exp s− u (v)
D
∑M ,,)/0,
`I_
3
4 2LM
j
p2iLjM
TE
ttt
= M,wxy, L = M,wxy
where
jM z
EP
Equation (7) represents the outlet mass flow rate of solid particles. As shown, the particle outlet
velocity was assumed as the difference between the gas bulk velocity and the terminal velocity of the
C
particle. This assumption was applied because the residence time of the particles is about ten times
AC
longer than the time required for equilibrating the gravitational force and drag force under a typical
e
n,d
condition. Thus, (2 + 1) unknowns, , and (2 + 1) equations remain for each zone and
fne
" "
and
11
ACCEPTED MANUSCRIPT
,,
3
4
(@ − @",
(
)), @",
(
) < @
,,,!"
3
4 = { & ()
0, @",
(
) ≥ @
For the oxidation reaction of SRF with air, the reaction pathways were assumed as depicted in
PT
Figure 5. The SRF is dried right after it enters the incinerator and undergoes the devolatilization
process where char and the volatile matter separate. Then, the char is burned to form carbon monoxide
RI
and ash remains, where the former finally reacts with oxygen to carbon dioxide. The volatile matter
reacts with oxygen to complete combustion products, i.e. carbon dioxide and water vapor. The detailed
SC
descriptions and assumptions for each process are given below.
As the temperature in the incinerator is much higher than the boiling point of water, the drying
U
process was modeled as an instantaneous reaction. The rate of devolatilization was then evaluated by
AN
using Eq. (8) with the kinetic parameter given by Rui Sun et al. [15]. During this process, the particle
diameter was assumed to be invariant, and therefore only the internal structure of particle changed
M
[16]. For further simplification, it was assumed that the volatile matter consisted only of CH4 and CO2,
which react with oxygen with the rate coefficients described in Table 1. This simplification can lead to
D
errors in evaluating the reaction rate of volatile matter, but it hardly affects the performance of model
TE
since the time constant of volatile matter combustion is a hundred times shorter than that of other
(∗
,,)/0, (
) = − exp − ,,)/0 3
4 ()
EP
(
The char burnout process was simulated as a one film model [17], while the structure of the
AC
boundary condition around the char particle was simplified. It was first assumed that there is no
homogeneous gas reaction inside the boundary layer. Next, the reaction of the char with oxygen and
carbon dioxide occur independently, although each reaction can affect the rate of the other reaction by
changing composition inside the boundary layer. With these assumptions, the rate of char burnout was
12
ACCEPTED MANUSCRIPT
diffused resistance, respectively, when the char burnout process is analyzed with circuit analogy [17].
The ash was separated from the char particles in an amount proportional to the burn rate of the fixed
carbon component. Table 1 shows the relevant parameters of the chemical reactions of the SRF. Here,
PT
it is worth noting that all reaction rates were calculated by assuming that the particles are spherical.
The effect of the particle shape on the reaction rate is above the scope of this study.
RI
j
E; $
?! + E E
,,$
,
3
4 = − 2$
3
4 Y ()
E; $
?! <=8>, 3
4 + <9800,
3
4
SC
E
,,,
= − Y ,,$
,
(
) (T)
E; $
?! + E
M
where
U
AN
'1
< (
'1$
<=8>,
3
4 = ,
i
j '1
# 5
M
'1
+ E,
3
4
'1$
<9800,
3
4 = , and
2i
K
D
E
E,
3
4 = <=8>,
3
4
<=8>,
3
4 + <9800,
3
4
TE
The boiler part was divided into three zones, as shown in Figure 6, and simulated as a zero-
dimensional model. In the first zone, steam generator zone 1 (SGZ1), heat transfer takes place
C
between the exhaust gas and the saturated liquid water from the steam drum in the wall tube. In the
AC
second zone, the superheater zone (SHZ), the outlet gas from the SGZ1 transfers heat to the
superheater. In the third zone, steam generator zone 2 (SGZ2), the exhaust gas heats the saturated
The thermodynamic states of each zone were calculated by using the mass conservation, energy
conservation, and ideal gas laws, Eqs. (11)-(13). The state of the exhaust gas at the boiler inlet was
13
ACCEPTED MANUSCRIPT
obtained directly from the incinerator model. The pressure inside each zone was also set to be constant
at atmospheric pressure and the solid particles in the exhaust gas were removed before it entered the
,
= ,,
− ,,!"
(TT)
PT
P
A
= Y ,
− Y ,!"
− 6)*(!
)+)
(T[)
RI
P
SC
Y = (T])
P < P
U
Equation (14) shows the rate of heat transfer between the exhaust gas and the steam generator
AN
(or superheater). As shown, the log mean temperature difference (LMTD) was used as the
representative temperature difference. For each zone, the intake temperature of the exhaust gas was set
M
to be equal to the temperature of the previous zone and the outlet temperature was made equal to the
present zone. The convective heat transfer coefficient was calculated as a function of the Reynolds
D
number and the Prandtl number. As there is only wall tubing in SGZ1, we applied the correlation for
fully developed turbulent internal flow [18]. On the other hand, we used the correlation for turbulent
TE
external flow over a bundle of pipes for the steam pipes across the zones in SHZ and SGZ2 [19].
where
C
14
ACCEPTED MANUSCRIPT
PT
Both steam generators 1 and 2 are connected to a steam drum, as depicted in Figure 7, where
the saturated liquid-vapor equilibrium is maintained in order to reduce the fluctuation of the
RI
thermodynamic state of the steam output during the plant operation. Equations (15)-(18) define the
drum and superheater model, which is based on Astrom and Bell’s study [20]. In the first equation,
SC
is the saturated liquid water feed rate from
the economizer and ,,!" is the outlet mass flow rate of the saturated vapor from the steam drum to
U
the superheater. The second equation is that of the energy conservation in the steam drum, where the
6!,)9 . Equations (17) and (18) denote the mass and energy balance for the superheater, where the
M
superheated vapor is assumed as an ideal gas. The coefficients on the left hand side of Eqs. (15)-(18),
to , were calculated from the thermodynamic database for the saturated liquid and vapor
D
area, ¨ , which relates to the desired steam production rate in the real plant operation. and
15
ACCEPTED MANUSCRIPT
where
gK gK
= K − K , j = & + & ,
g5
g5
= 0, = 1,
PT
j = K ℎ − K ℎ,
jj
RI
gK gℎ gK gℎ
= & ℎ + K + & ℎ + K − (& + & )
g5
g5
g5
g5
SC
g(" (5
)
+ ,
,
, ,
g5
j = 0, j = ℎ ,
U
AN
= 0, j = 0, = & , = −1,
and ,"
!. = ∗ ¨ + ©
EP
The plant model was compared with the results from the operation of the real plant. The
C
validation data was collected for approximately 2500 seconds with variations in the top air flow rate
AC
and the SRF feed rate. The PID controller regulated the pressure inside the drum, causing values for
¨ to vary. In the plant operation, it was nearly impossible to regulate the composition of the SRF.
Therefore, the representative composition in Table 2, which was measured from a sample of SRF, was
used.
The SRF feed rate into the incinerator was not measured directly. Instead, the electrical power
16
ACCEPTED MANUSCRIPT
input to the inverter and motor, 5 " , which controls SRF feed rate in the fuel hopper, was used to
estimate the rate. As shown in Eq. (19), the feed rate was based on 5 " in terms of an ordinary
differential equation, where the first order was assumed considering the time delay between the fuel
hopper control and the actual feed into the incinerator chamber. Here, ¬)/0 represents the dynamic
PT
response of the fuel supply system, which had been calibrated during the plant model validation. The
parameters " and © " represented the static response and were based on the data
RI
provided by ENERONE Inc..
SC
"
"
Table 3 lists the important variables regarding plant model validation. First, the inputs into the
U
system included the 1st blower flow rate, the 2nd blower flow rate, the SRF feed rate, and the steam
AN
valve opening area. The 1st blower controlled the mass flow rate of core air and shell air, while the 2nd
Secondly, the seven major outcomes, which are listed on the second column of Table 3, were
compared. The first three variables, namely, the temperature of the exhaust gas at the boiler inlet,
D
superheater inlet, and the boiler outlet, is deemed to be important parameters as they directly affect the
TE
steam production rate. The fourth variable is the remaining oxygen mole fraction in the exhaust gas,
which is used as a representative indicator of the emission characteristics in the plant operation. Then,
EP
we chose steam production rate, steam pressure, and steam temperature; which are directly related to
Finally, the third column in Table 3 lists the model parameters that were calibrated during
AC
validation. First, we adjusted the proportion of top air, which does not participate in the reaction. As
shown in Figure 4, as the top air is blown into the incinerator along the wall, only a portion of it
participates in the oxidation reaction in the flame. The second parameter, i.e., the effective surface area
of the steam generator and the superheater was adopted in order to evaluate the dependency of heat
transfer on the structure of the heat exchanger, which is difficult to capture directly using the current
17
ACCEPTED MANUSCRIPT
zero-dimensional model. The third and fourth parameters, and © , relate the steam
production rate with the valve opening area. Finally, ¬)/0 is the last parameter to be considered for
the characteristic time constant of the fuel supply system. The values of the calibration variables are
listed in Table 4.
PT
Figure 8 shows the validation results of the plant model (blue) against the real plant operation
RI
(red). The left five figures describe the change of the input variables, and the right five depict that of
the output variables listed in Table 3. The computation time for the simulated 2500-sec operation
SC
using a typical lab computer with a CPU clock speed 3.40 GHz is less than 800 seconds, which is fast
enough to design and validate the controller. As shown, this model simulates the behavior of the plant
U
quite well, with the exception of some fluctuation on the oxygen mole fraction. This might be mainly
due to the limitation of the zero-dimensional model, which lumps the spatial distribution of the
AN
variables even though the sensor measures the local oxygen mole fraction. Moreover, the fuel
composition could have been varied during the measurement, which was not simulated in the plant
M
model, as it was only possible to adopt the representative fuel composition without further real-time
measurements.
D
TE
C EP
AC
18
ACCEPTED MANUSCRIPT
4. Controller
The controller was designed on the basis of the experience of the plant operator. In other words,
the input variables were chosen as the ones that the operator most frequently manipulated, namely, the
PT
second blower flow rate (top air flow rate), SRF feed rate, and the steam valve opening area. Then, we set
the controlled output variables for the following states: oxygen mole fraction, steam production rate, and
RI
steam pressure, which we would aim to regulate in this step. As an observer had not been built during the
controller development, which is beyond the scope of this research, we set the state variables equal to the
SC
output variables. In order to describe the relationship between the input and output variables, a non-linear
control model was first developed and then linearized around the nominal operating condition, since the
U
output states in the target system vary slowly and slightly over time with respect to the time scale of the
physical phenomena in the system. Subsequently, the controller was designed based on the linear quadratic
AN
regulator (LQR) scheme, one of the fundamental multi-input multi-output (MIMO) controllers.
M
In this section, the non-linear control model with three input variables and three state variables
D
has been described. The control model was built for simplifying the plant model with 7 input variables
TE
and 60 state variables (in Section 3). For this, both the incinerator and the boiler were combined into a
single zone and assigned some simple first order differential equations to the dynamic behavior. The
EP
assumptions and the reasons or justifications for these are summarized below.
1. Reaction kinetics are much faster than other system dynamics. Thus, it was assumed
C
that all the reactants change to the designated products right after being supplied into
AC
incinerator.
2. When the combustion zone is fuel-rich, the volatile matter first reacts with oxygen and
3. The drum level is well regulated with an intrinsic PID controller, and can therefore be
assumed to be constant.
19
ACCEPTED MANUSCRIPT
4. It was also assumed that the proportion of heat transfers for the steam generator and
As shown in Figure 3, the typical temperature inside the flame is over 1500 K, which is high
PT
enough to assume an instantaneous reaction. Furthermore, the oxygen consumption rate from the
combustion of volatile matter is about a hundred times faster than that of char particles in most cases
RI
according to the plant model. In the real plant operation, there is a PID controller for regulating the
drum level by adjusting the water inlet mass flow rate by considering the inverse response
SC
characteristics [20], thus its values would fluctuate only slightly during operation (±3%). Finally, the
results from the plant model tests show that the proportion of heat transfer to the steam generator and
U
superheater is typically between 80:20 and 81:19.
AN
¬I®,CD B CD + BCD = BCD ,.. (20)
= ∗ , Case 2
,
, + ,)/0, 31 − E − E; $
?! ∗ H ?
$
4 '1CD
TE
Based on these simplifying assumptions, the first state, i.e. the oxygen mole fraction in the
exhaust gas, was modeled as a first order simple differential equation in Eq. (20). Here,
¬I®,CD represents the dynamic response of the oxygen mole fraction in the exhaust gas to the oxygen
C
AC
mole fraction input change by controlling the blower or the SRF feed rate into the incinerator. This
parameter was evaluated by comparing it with the plant model. Here, BCD ,.. is the static response of
the oxygen mole fraction, depending on the different combustion cases as follows. Case 1 occurs when
the sufficient air is provided in PCZ and SCZ for complete combustion of the fuel. Case 2 occurs
when there is enough air present in PCZ and SCZ to burn methane, however, insufficient to burn char
20
ACCEPTED MANUSCRIPT
particles completely. Finally, Case 3 occurs when there is insufficient air that cannot even burn
methane completely.
5
= 6)*
)*_
+ 6)*
)*_j
− 6!,)9 − ,"
!. (ℎ − ℎ ) ([T)
P
PT
where
RI
gK gℎ gℎ
= & (ℎ − ℎ ) + K & + K & − (& + & )
g5
g5
g5
SC
g(" (5
)
+ ,
,
, ,
g5
6)*
)*_
= H)* Y
, ,,!" ((?!
− (?!
!" ),
U
AN
6)+
)+_
= (1 − H)* ) Y
, ,,!" ((?!
− (?!
!" ), and
M
Y
, ,,!" (?!
D
,
(
, ,
, +
,)/0 ()/0, ,)/0,
+ %&I+J ,)/0, EI+J + %&; $
?! ,)/0, E; $
?! − 6!,78 , Case 1
TE
,
(
, ,
, +
,)/0 ()/0, ,)/0,
=
+ %&I+J ,)/0, EI+J + %&; $
?! ,)/0, E; $
?! (1 − H ?
$
) − 6!,78 , Case 2
,
(
, ,
, +
,)/0 ()/0, ,,
EP
For the second state of steam production rate, the relationship between ¨ and steam
C
production rate in the control model was used, which had also been introduced in the plant model
AC
design (Section 3.2). We evaluated the last state, the steam pressure, in Eq. (21) [20], which is the
reduced steam drum equation for the case when the level is well-regulated. The amount of heat
temperature, which was evaluated for different cases from Eq. (20). Here, the boiler outlet temperature
21
ACCEPTED MANUSCRIPT
The control model was validated by comparing the output state variables against those of the
plant model. We used the same test case shown in Figure 8. As the input variables were varied, the
relevant output variables were plotted together in Figure 9. The control model predicted the output
PT
4.2. Controller development and application
RI
This sub-section introduces the control algorithm that was used to calculate the input variables
based on the developed control model, and the validation of the controller has been demonstrated by
SC
applying it to the plant model in two different cases.
In developing the controller, we applied the LQR algorithm using a linearized and discretized
U
control model. As described in Figure 10, the nominal state, B> , and nominal input, @> , were
AN
renewed at each time step. Then, the control model was linearized and discretized. By using the
derived system matrix, , and the input matrix, ©, control gain was calculated to minimize the cost
function in Eq. (22) with pre-determined matrices 6, <, and 2. Next, the input variables were
M
evaluated, which were fed back into the plant system model. This process was repeated at every time
D
step.
TE
In the following paragraphs, we will perform two representative test cases for the controller
application that are useful in real plant operation. As previously mentioned, the composition of SRF
C
could be varied during the plant operation in a somewhat arbitrary and unknown fashion. This would
AC
lead to an unstable operation of the system, depending on the significance of the fluctuation. In the
first test case, it was demonstrated that the controller could effectively suppress this disturbance of fuel
composition variation. Here, the SRF composition was changed periodically by mixing two extremely
different SRFs, as shown in Table 5. While SRF-1 is the fuel that was used for the plant model
validation in Table 2, SRF-2 is another arbitrary choice of SRF with a higher volatile matter and
22
ACCEPTED MANUSCRIPT
heating value, which is also produced in South Korea [22]. The ratio of two fuels, SRF-1:SRF-2, was
varied from 0:100 to 100:0 over a period of 200 s. In the meantime, the output variables from the plant
model were monitored both without the controller for the first 1000 seconds, and with the controller
for the following 2000 seconds. While the controller operated, the SRF composition in the control
PT
model was fixed to be equal to that of SRF-1, i.e., a nominal value, as there was no real-time sensor
(or model) that could measure (or predict) the composition of SRF in the real plant. This is a realistic
RI
situation because the SRF composition is generally measured only twice a year, and workers mix one
truck-load of the new fuel with the old fuel stock in the yard every day. Thus, the controller was
SC
supplied with some fixed, nominal fuel composition values.
U
As shown in Figure 11, all the output variables were found to converge into the desired values
with the application of the controller. The fluctuation in the oxygen mole fraction was effectively
AN
suppressed from ±1.2% point (±15.4%) to ±0.3% point (±3.9%), and the steam pressure was
M
stabilized to the desired value (while it diverged without the controller in the beginning). The steam
production rate started to fluctuate with the controlling process, but the amplitude was lower than 0.07
D
kg s–1 (1.3%).
TE
For the second test case, the steam production rate was controlled to the desired value as there is
a variation in steam demand depending on the steam turbine’s electric generation requirement or other
EP
steam related processes. In this test, the desired value of steam production rate was varied arbitrarily.
As in the real plant operation, the desired values of the oxygen mole fraction and steam pressure were
C
assigned to ensure pollutant emission control and product quality of superheated steam, respectively.
AC
As shown in Figure 12, the steam production rate followed the desired values well with the
maximum settling time of 400 seconds after the controller was applied at the zero second mark in the
figure. In a real plant operation, the typical time for steam production rate variation is much longer
than 400 seconds. At the same time, the other output variables met the target values with a minimal
fluctuation of ±0.6 bar (±1.4%) for steam pressure and ±0.14% point (±1.8%) for the oxygen mole
23
ACCEPTED MANUSCRIPT
fraction, respectively.
PT
RI
U SC
AN
M
D
TE
C EP
AC
24
ACCEPTED MANUSCRIPT
5. Conclusions
In this study, we controlled the steam generation system for the swirling-flow waste incinerator,
while suppressing various real-world disturbances. Initially, a plant model was developed in order to
PT
understand the physics of the system, and then a model-based controller was designed for a specific SFI
and boiler system installed in South Korea. A summary of the findings is given below:
RI
1. A plant model was built as a multi-zone, zero-dimensional model with three separate parts:
an incinerator, a boiler, and a steam drum. The thermodynamic state and combustion
SC
phenomena taking place inside the incinerator were simplified into three zones based on the
observation from in-house, 3D CFD results. We defined the boiler by the type of heat
U
exchangers, and incorporated a widely accepted model for the steam drum. The integrated
AN
model was then validated against the real-plant operation data for seven important state
variables: boiler inlet temperature, superheater inlet temperature, boiler outlet temperature,
M
oxygen mole fraction, steam production rate, steam pressure, and steam temperature. It was
found that the plant model could predict the behavior of the real system accurately with a
D
relatively small computing resource. Therefore, this model was deemed to be well-suited for
2. A reduced, non-linear control model for the three input variables, namely, second blow flow
EP
rate, fuel supply rate, and steam valve opening area, and three output variables, namely,
oxygen mole fraction, steam product rate, and steam pressure, was designed and these
variables were then validated against the developed plant model. This control model was
C
applied to a linear quadratic regulator after being linearized and discretized. Subsequently,
AC
the controller was designed and tests were performed on the plant model. The developed
controller could efficiently suppress the real-world disturbance from the fuel composition
variation as well as adjust the steam production rate as desired, while maintaining a proper
25
ACCEPTED MANUSCRIPT
6. Acknowledgement
This research was supported by Basic Science Research Program through the National Research
PT
RI
U SC
AN
M
D
TE
C EP
AC
26
ACCEPTED MANUSCRIPT
References
PT
In: Choi Y, editor. Production, Consumption and Import of Solid Refused Fuel in South Korea: Korea
Environment Corporation; 2015.
RI
[4] Environment Mo. ENFORCEMENT RULE OF THE ACT ON THE PROMOTION OF SAVING AND
RECYCLING OF RESOURCES. In: Legislation MoG, editor. 6842016.
[5] Energy MoTI. ENFORCEMENT RULE OF THE COAL INDUSTRY ACT. In: Legislation MoG, editor.
SC
2402014.
[6] Kozek M, Voigt A. Model based decoupling control for waste incineration plants. Conference
Model based decoupling control for waste incineration plants. IEEE, p. 629-34.
U
[7] Leskens M, van Kessel L, Bosgra O. Model predictive control as a tool for improving the process
AN
operation of MSW combustion plants. Waste Management. 2005;25(8):788-98.
[8] Miyamoto Y, Kurosaki Y, Fujiyama H, Nanbu E. Dynamic characteristic analysis and combustion
control for a fluidized bed incinerator. Control Engineering Practice. 1998;6(9):1159-68.
M
[9] Carrasco R, Sanchez EN, Ruiz-Cruz R, Cadet C. Neural control for a solid waste incinerator.
Conference Neural control for a solid waste incinerator. IEEE, p. 3289-94.
[10] Sugeno M, Kang G. Fuzzy modelling and control of multilayer incinerator. Fuzzy sets and systems.
D
1986;18(3):329-45.
TE
[11] Chen W, Chang N-B, Chen J-C. GA-based fuzzy neural controller design for municipal incinerators.
Fuzzy Sets and Systems. 2002;129(3):343-69.
[12] Marias F, Puiggali J, Flamant G. Modeling for simulation of fluidized‐bed incineration process.
EP
1998;53(15):2727-42.
[14] Levenspiel O, Kunii D, Fitzgerald T. The processing of solids of changing size in bubbling fluidized
AC
27
ACCEPTED MANUSCRIPT
Combustion.543.
[18] McAdams WH. Heat Transmission, (1942). McGraw Hill Book Company, Inc.
[19] Kays W, Lo R. Basic Heat Transfer and Flow Friction Design Data for Gas Flow Normal to Banks of
Staggered Tubes: Use of a Transient Technique1952.
[20] Astrom KJ, Bell RD. Drum-boiler dynamics Automatica. 2000;36(3):363-78.
PT
[21] Reynolds WC. Thermodynamic properties in SI : graphs, tables, and computational equations for
forty substances. Stanford [Calif.] :: Dept. of Mechanical Engineering, Stanford University; 1979.
RI
[22] Shin DH. Study on the characteristics of RDF and RPF in South Korea. Suwon Universitry: Suwon
Universitry, 2008.
[23] Caram HS, Amundson NR. Diffusion and Reaction in a Stagnant Boundary Layer about a Carbon
SC
Particle. Industrial & Engineering Chemistry Fundamentals. 1977;16(2):171-81.
[24] Dryer F, Glassman I. High-temperature oxidation of CO and CH4. Conference High-temperature
oxidation of CO and CH4, vol. 14. Elsevier, p. 987-1003.
U
AN
M
D
TE
C EP
AC
28
ACCEPTED MANUSCRIPT
Figure 1 Steam generation system with a waste incinerator built by ENERONE Inc. at Ul-San-Si in
South Korea (Single column)
Figure 2 Schematic of the incinerator and boiler of ENERONE Inc. (top view, side view) (Single
column)
PT
Figure 3 Temperature and velocity field in the incinerator (from 3D CFD simulation) (Single
column)
RI
Figure 4 Division of zones in the incinerator (side view) (Single column)
SC
Figure 5 Reaction pathway of the combustion of SRF (1.5 column)
U
Figure 7 Schematic diagram of the steam drum (Single column)
AN
Figure 8 Plant model validation (1.5 column)
Figure 11 Application of the controller on the plant model - suppressing disturbance caused by
D
Figure 12 Application of the controller on the plant model adjusting the steam production rate
(1.5 column)
EP
Table 5 Characteristics of SRF for testing the fuel composition variation (Single column)
29
ACCEPTED MANUSCRIPT
PT
RI
SC
Figure 1 Steam generation system with a waste incinerator built by ENERONE Inc. at Ul-San-Si in
South Korea
U
AN
M
D
TE
C EP
AC
30
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
M
D
TE
C EP
Figure 2 Schematic of the incinerator and boiler of ENERONE Inc. (top view, side view)
AC
31
ACCEPTED MANUSCRIPT
PT
RI
SC
Figure 3 Temperature and velocity field in the incinerator
U
(from 3D CFD simulation)
AN
M
D
TE
EP
C
AC
32
ACCEPTED MANUSCRIPT
PT
RI
U SC
Figure 5 Reaction pathway of the combustion of SRF
AN
M
D
TE
EP
C
AC
33
ACCEPTED MANUSCRIPT
PT
RI
SC
Figure 7 Schematic diagram of the steam drum
U
AN
M
D
TE
EP
C
AC
34
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
M
D
TE
EP
C
AC
35
ACCEPTED MANUSCRIPT
PT
RI
Figure 9 Control model validation
U SC
AN
M
D
TE
EP
C
36
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
M
D
TE
EP
Figure 11 Application of the controller on the plant model - suppressing disturbance caused by
fuel composition variations
C
AC
37
ACCEPTED MANUSCRIPT
PT
RI
U SC
AN
M
D
TE
EP
Figure 12 Application of the controller on the plant model adjusting the steam production rate
C
AC
38
ACCEPTED MANUSCRIPT
Char burnout
PT
No. Reaction equations Ref.
17966
2C + Oj → 2CO 3.007 ∗ 10z exp − 2 ∗ 10: 1.12 ∗ 10
(
1 [23]
29790
C + COj → 2CO 4.1 ∗ 10Æ exp − 2 ∗ 10: −1.7 ∗ 10
(
2 [23]
RI
Gas phase reaction
No. Reaction equations ¸¹¸ [(¸ºÁÂ/º] )T:º:Ç /»] m n ¾¿¸ [À/¸ºÁÂ] Ref.
1 15098 .z
SC
CO + Oj → COj 1.3 ∗ 10 exp − +D C 2.82 ∗ 10
2 (
3 1 0.5 [23]
24358
CH + 2Oj → COj + 2Hj O 10 exp − 5.00 ∗ 10Ê
(
4 0.7 0.8 [24]
U
AN
Table 2 Characteristics of the SRF
M
H2O 8.5
Ash 10.0
LHV 30.8
C EP
AC
39
ACCEPTED MANUSCRIPT
Ë,ÌÁÍÎ,¹Ç +ºË,»ÒÎÂÂ,¹Ç )
(ºÏÐÑ
1st blower flow rate
((Ü!
, ( )*_ , ( )*_j )
ÓÐÑ
Boiler gas temperature
H"!
,
PT
"
(B!)*_j
Mole fraction of oxygen
RI
(,"
!. )
Steam production rate
, ©
(ºÏÐÑ
SRF feed rate
SC
»,Ö×Ø,¹Ç )
(5
)
Steam pressure
¬)/0
(ÙÚÛÂÚÎ ) (( )
Steam valve opening area Steam temperature
U
AN
Table 4 Value of calibration variables
M
SRF-1 SRF-2
SRF-1 SRF-2
40
ACCEPTED MANUSCRIPT
Highlights
Plant model was built to simulate heat recovery system with a waste incinerator.
The controller could efficiently suppress disturbance and adjusts load variation.
PT
RI
U SC
AN
M
D
TE
C EP
AC