Sei sulla pagina 1di 18

Applied Mathematical Modelling 22 (1998) 823840

CFD modelling of waste heat recovery boiler


M. Manickam
a

a,b,*

, M.P. Schwarz a, J. Perry

a,b

CSIRO, Division of Minerals, Institute of Minerals Energy and Construction, Clayton, Victoria, Australia
b
CRC, New Technologies for Power Generation From Low-Rank Coal, Mulgrave, Victoria, Australia
Received 21 February 1997; received in revised form 17 November 1997; accepted 18 December 1997

Abstract
A model has been developed of a waste heat recovery boiler utilising typical plant o-gas consisting of both gaseous
and particulate combustibles. The model allows the calculation of temperatures of gas and particles within the boiler
and hence the likelihood of deposition onto the boiler walls. The model was applied to a typical waste heat boiler
geometry, and a typical o-gas composition including a mixture of combustibles (char) and non-combustible particulates. Mixing in the burner region, char burnout and char particle temperature were analysed using the model.
Combustion stability was also studied using a simple Eddy break-up model which accounts for combustion kinetics and
the results compared with a Mixed-is-burnt model. 1998 Elsevier Science Inc. All rights reserved.
Keywords: CFD modelling; Gas combustion; Particle tracking; Char combustion; Eddy break-up modelling

1. Introduction
In the highly competitive metal production and power generation industries, the dierence
between energy recovered and energy wasted often determines the dierence between prot and
loss. A waste heat boiler is one of the devices generally employed to recover energy from high
temperature furnace gases, which in some applications, contain unburnt combustibles that are
burnt to complete combustion, extracting a signicant fraction of waste energy.
The design and operation of a waste heat boiler employed to recover energy under such operating conditions depends critically on achieving stable combustion. Low caloric value of the
furnace gases when combined with the presence of inert gases makes it harder to maintain a stable
ame and necessitates the designing of an ecient combustion process. Computational Fluid
Dynamics (CFD) can provide an eective platform where various design options can be tested
and an optimal design can be determined at a relatively low cost.
To aid in the design of boiler, the Cooperative Centre for New Technologies for Power
Generation from Low-Rank Coal (CRC) has developed a CFD model. This paper describes the
model formulation and the application of the model to the simulation of a typical waste heat
boiler geometry and conditions similar to those of a waste heat boiler simulated by the CRC prior
to construction at a primary metal reduction plant. In this case, a boiler consisting of radiant and
convective passes was to be constructed onto the existing after-burner with exhaust gas tied back
*

Corresponding author. Tel.: +61 3 9545 8687; fax: +61 3 9562 8919.

0307-904X/98/$ see front matter 1998 Elsevier Science Inc. All rights reserved.
PII: S 0 3 0 7 - 9 0 4 X ( 9 8 ) 1 0 0 2 0 - 3

824

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

into the existing duct work upstream of the scrubber (Fig. 1). Air required for combustion was to
be injected through sets of slots in the radiant pass in three stages, referred to here as primary,
secondary and tertiary air. At each level air enters through nine injection ports, ve on one side of
the boiler and the rest on the opposite side. The plant o-gas carries char particles and dust
particles which are not chemically reactive. The plant o-gas is composed of combustible components such as CO and H2 , some moisture and a signicant percentage of inert gases like N2 and
CO2 .
This paper illustrates how the model can be used to estimate temperatures of char and dust
within the boiler and hence to determine the likelihood of deposition onto walls. Deposition will
be minimised if the particle temperature is lower than the temperature at which the particle would
become sticky.
Numerical modelling of the waste heat recovery boiler was carried out using dierent combustion models and Eulerian scheme was used for the continuum gas phase and a Lagrangian
technique was adopted for tracking the char/primary concentrate particles carried by the gas.
Gaseous combustion was modelled using Mixed-is-burnt combustion model which assumes instantaneous combustion of the fuel as soon as it comes in contact with the oxidant. As plant ogas has low caloric value it will burn at a lower temperature and a slower rate than higher
caloric value fuels. To determine the magnitude of this eect the combustion was also modelled
using the `Eddy break-up' combustion model which accounts for the reaction kinetics.

Fig. 1. Overall geometry and isothermal ow predictions.

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

825

2. Solution of uid ow equations


Gas ow through the boiler was solved using a nite volume method on a non-staggered grid,
with the RhieChow algorithm used to prevent oscillations in the pressure eld. The turbulent
ow eld was solved using the Reynold's time averaging for the uctuating turbulent quantities,
which allows splitting of velocity, pressure and other scalars into mean and uctuating components.
The instantaneous value of a turbulent quantity, say a velocity component, U 0 , can be split into
U 0 U u;

where U and u are the mean and the uctuating components, respectively.
The governing Reynolds averaged equations of the mass and momentum for a steady state and
incompressible ow are given by:
r  U 0;

r  qU
U r  r qu
u;

where r is the stress tensor


r pd lrU rUT :

Here q is the uid density, U U ; V ; W the uid velocity, p the pressure and l the molecular
viscosity. The terms qu
u are the turbulent ux terms known as Reynolds stresses.
The Reynolds-averaged energy equation is given by
r  qUH r  krT quh:

In order to close the momentum transport equations the Reynolds stress terms qu
u are determined, through the ke turbulence model which utilises the Boussinesq's assumption that
turbulent stresses can be linked to the mean velocity gradients through a term called eddy or
turbulent viscosity.
qu
u 23qkd lT rU rUT :

The eddy viscosity is dened using turbulent kinetic energy and rate of eddy dissipation
k2
:
e
Separate transport equations are used for k and e:

 
l
r  qUk r 
l T rk P qe;
rk

 
lT
e
e2
r  qUe r 
l
re C1 P C2 q ;
re
k
k
where P is the shear production of turbulent kinetic energy dened as
lT Cl q

P leff rU  rU rUT 23r  Uleff r  U qk;

8
9

10

leff is the eective viscosity given as the sum of molecular and eddy viscosity, rk ; re ; Cl ; C1 ; C2 are
turbulence model constants which usually have values of 1.0, 1.22, 0.09, 1.44 and 1.92, respectively.

826

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

The velocitypressure coupling has been eected through the SIMPLEC algorithm (Van
Doormal and Raithby, 1984) which provides the basis for updating pressure and correcting velocity components for continuity.
The ow equations were solved using the general purpose code, CFDS-FLOW3D (1994)
(currently known as CFX-F3D).
3. Combustion model methods
3.1. Mixed-is-burnt combustion model
Combustion of the gas was modelled using the `Mixed-is-burnt' model which assumes instantaneous combustion as soon as the fuel and oxidant mix with each other. This is normally a
good approximation because gas phase combustion is usually rapid. The eect of intermittency in
oxygen and fuel concentrations caused by turbulence is accounted for using a double delta
function pdf technique.
A conserved scalar in the form of a mixture fraction is dened and transport equations for the
mixture fraction and its variance are solved to determine the concentrations of reactants and
products.
Conservative transport equations for the mean F of the mixture fraction f, and its variance G
are given in the form:



lT l
r  qUF r 
rF 0;
11

rT rL



lT l
e
r  qUG r 
rG Cg1 lT rF 2 Cg2 q G:

12
rT rL
k
Here rL and rT are the equivalent Prandtl numbers and Cg1 and Cg2 are model constants. The
mixture fraction f is dened as
f

v vO
;
vFu vO

13

where
mO
14
i
Here i is the stoichiometric ratio and m is the mass fraction with subscripts Fu and O denoting
fuel and oxidant, respectively. When v 0, mixture fraction f attains its stoichiometric value FST
given by
v mFu

1
:
15
1i
The mixed-is-burnt model is based on the assumption that fuel and oxidant cannot coexist instantaneously. The instantaneous mass fractions of fuel, oxidant and products are dened as
follows.
When f > FST , the mixture consists only of fuel and products such that:
FST

f FST
; mO 0; mp 1 mFu
1 FST
and when f < FST , the mixture consists only of oxidant and products such that:
mFu

16

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

827

f
; mp 1 m O ;
17
FST
where the subscript p refers to products.
The PDF of the mixture fraction needed to determine the mean properties and variation of
species mass fraction and temperature is assumed here to be a double delta density function of the
form
mFu 0;

mO 1

pf Adf F Bdf F ;

18

where F F a and F F a with the constants A, B and a are determined from and G. The
mean mass fractions are then dened as:



f FST
; 0 pf df ;
1 FST

19



f
max 1
; 0 pf df ;
FST

20

Z1
mF

max
0

Z1
mO
0

mp 1 mF mO :

21

Radiation heat transfer within the boiler was modelled using the discrete transfer method (Shah,
1979) and assuming a gray gas model. The gas absorption coecient for the furnace gases was set
to be 0.25, following Mackenzie (1990) and Smith et al. (1989). The number of rays to be tracked
from each node was set to be 6.
In setting up the gaseous combustion, full representation of all the constituents of plant o-gas,
which as one would expect dier distinctly to those of conventional gas fuels, could signicantly
increase the computational size of the problem. Hence the focus was on accurate representation of
the reacting species and simplication of the inert constituents of the plant o-gas mixture. For
example, N2 is replaced with CO2 since neither take part in the reactions and hence there should
be no eect on the stoichiometry. The composition and density of the gas have been modied
accordingly.
3.2. Eddy break-up combustion model
Investigations of reaction rates by Malcolm McIntosh and Denise Goch (1995) show that
combustion critically depends on the mixture temperature emphasising the necessity of an accurate temperature prediction. A more accurate temperature prediction would be obtained using
an Eddy break-up combustion model which included some account of both mixing kinetics and
nite rate chemical kinetics for gas phase combustion. Such a model may also be suitable for
investigating combustion stability and ignition.
In this paper we use a simple Eddy break-up combustion model based on that of Bakke and
Hjertager (1987) in which the turbulence mixing time is compared with the chemical induction
time at each computational cell of the model domain and a decision made on the ignition/extinction of the fuel-air mixture available at that cell. If the gaseous mixture ignites, combustion is
assumed to take place in that cell at the mixing rate. The ratio of a turbulent and a chemical
kinetic scale is dened as
se
D
;
22
sch

828

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

where D is a Damkohler number. Ignition is assumed to take place when D P Die , where Die
(ignitionextinction) is an empirical constant which was set as used in Hjertager (1982). The
turbulent scale is given by
k
se ;
23
e
while the chemical induction time is dened as
 
TA
qmF a qmO2 b ;
24
sch Ach exp
T
where: Ach is a rate constant, TA an activation temperature, and a and b are the exponents for the
fuel and oxygen density, respectively.
These parameters were set for plant-o-gas based on the reaction kinetics derived from the
investigations of Malcolm McIntosh and Denise Goch (1995) using Chemkin (1980).
Since plant o-gas is a mixture of various gases, the reaction kinetics data specied for the
model based on any one of the main reactants should reasonably reect combustion characteristics of the mixture under typical operating conditions. The plant o-gas has combustible constituents, H2 and CO. Based on earlier observations (Malcolm McIntosh and Denise Goch, 1995),
the H2 component of the plant gas appears to burn very quickly under the typical temperatures
that prevail in the combustion zone. The combustion of the other combustant CO is slower and
depends rather heavily on temperature. Two dierent approaches were considered to address this
issue. In the rst approach plant gas is assumed to have the reaction kinetics of CO while retaining
its actual heating value. Since in reality the H2 component will burn much faster, it is reasonable
to assume that the heat energy released by the combustion of H2 is sucient to locally increase the
temperature of the CO component of the gaseous mixture for it to burn as well. In other words
the reaction kinetics of the plant gas can be assumed to be that of H2 for the combustion
modelling. This constitutes the second approach of the Eddy break-up modelling. It turns out that
both approaches resulted in very similar solutions.
A quick check on the ammability limits (Lefebvre, 1983) of the limiting CO, reveals a weak
limit of 12.5% of fuel by volume when compared with the available air. In the present case the
bulk of the combustion takes place when the plant gas comes in contact with the primary air.
Calculation of the fuel concentration in terms of % by volume as compared to the available air
gives a value of 14.5% which is above the weak limit of the ammability as noted above. However
it should be noted that the ammability limits have a weak dependence on pressure and a relatively strong dependence on the temperature. It is also aected by the presence of inert gases.
In this amelet modelling approach, coal combustion was not modelled simultaneously with
the gas combustion. Instead, the spatial distribution of heat energy released by char combustion
in the previous models was stored and the same distribution was set in this model releasing an
equivalent amount of heat energy. Though this approach limits the accuracy of temperature
predictions, it allows a clear comparison of dierent gas combustion models. Primary concentrate
particles are tracked as a post processing exercise based on the already converged results.
3.3. Particle tracking method
In this gasparticle system, where the mass ow rate of the gas phase is signicantly higher
than that of the particulate phase and particleparticle collision rate is therefore small, there is no
advantage in treating the particulate phase as continuum. Indeed the representation of the particles phase using the Lagrangian technique is superior since it allows more accurate specication
of dierent characteristics of the particles phase. Hence the Lagrangian technique, in which

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

829

representative particles with pre dened initial starting locations, velocities, directions and
compositions are tracked in the computational domain, is employed to model the particle phase.
These particles interact with the gas phase through source terms in the mass, momentum and
energy equations of the continuum gas phase, a method known as the Particle- Source-In Cell
(PSI-Cell) technique (Crowe et al., 1977).
Particles are assumed to be spherical in shape and particleparticle interactions are neglected.
When particle trajectories hit a wall they are assumed to bounce inelastically with a coecient of
restitution of 0.8.
Particle-momentum conservation: Particle velocities, trajectories, composition and other properties are determined by integrating the momentum, mass and energy conservation equations of
the particulate phase. The rate of change of velocity of the particle is derived from Newton's
second law of motion
du
F;
25
dt
where F is the force on the particle which includes the drag, buoyancy, and m its mass. The drag
force, force exerted on the particle by the continuous phase, is given by
m

FD 18pd 2 qCD jVR jVR :

26

Here d is the particle diameter, q the density of the continuous phase. VR the relative velocity of
between the particle and the gas phase. The drag coecient CD is dened as
24
1 0:15Re0:687 :
Re
Particle-energy conservation: The energy conservation of the particle is eected using
CD

27

dTp
Qc Qr Qb ;
28
dt
where mp and Tp refers to the mass and temperature of the particle and Cp is the specic heat of
the particle. Qc ; Qr and Qb represent the heat transfer due to convection, radiation and combustion, respectively.
Based on the assumption that the particles are spherical in shape, convective heat transfer for
the particle is dened as
mp Cp

Qc pdkg NuTg Tp ;

29

where kg and Tg denote the thermal conductivity and the temperature of the gas phase. Nu is the
Nusselt number given by

1=3
CP
0:5
l
:
30
Nu 2 0:6Re
kg
The radiative heat transfer is determined using the relation
Qr ep pd 4 I rTp4 ;

31

where ep is the emissivity of the coal particle, I the radiative ux at the current computational cell
and r the StefanBoltzmann constant. Qb , the heat released due to char oxidation is based on the
char combustion model described later.
The eect of turbulence in the gas phase on particle motion was modelled using a stochastic
method (Gosman and Ioannides, 1981).
The source terms of the Eulerian gas phase due to the particles are determined by computing
changes in particle properties as they traverse each computational cell. For each representative

830

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

particle, k, the number of particles proceeding along the representative trajectory per unit time, nk ,
is known from the boundary condition at the injector exit. Then the exchange rate between the
phases, for any generic property /, is given by the following (Crowe et al., 1977)
Sfi;j

N
X
k1

nk mk fk in mk fk out i;j ;

32

where i, j designate the computational cell, N is the total number of representative particles, and
mk the mass of the representative particle.
3.4. Char combustion
In the char combustion model of Field et al. (1967), the oxidation rate of the char is calculated
on the assumption that the reaction is under mixed rate control, with both the rate of diusion of
oxygen to the surface of the char particle and the chemical reaction kinetics accounted for. In the
Field model the rate of diusion of oxygen is given by kd Pg Ps , where Pg is the partial pressure
of oxygen in the gases far from the particle boundary layer and Ps is the oxygen partial pressure at
the particle surface. The diusion coecient kd is given by


2:53  107 Tp Tg
PA
kd
0:75 ;
33
Rp
2
p
where Rp is the particle radius, Tp the particle temperature, Tg the far-eld gas temperature, P the
local pressure and PA the atmospheric pressure. The char oxidation rate per unit area of the
particle surface is given by kc Ps . The chemical rate coecient kc is given by


Ec
;
34
kc Ac exp
RTp
where the parameter Ac is the Arrehenius rate constant for char oxidation, Ec the activation
energy for char oxidation and R the gas constant. Based on experimentally determined data of a
typical bituminous coal (Smith et al., 1989), char reaction kinetic parameters are specied as
Ac 162:0 kg=m2 atm and Ec 93:4 MJ=kmol. Then the overall rate coecient is
P
:
35
PA
Tracking of the primary concentrate particles was done as a part of the post-processing after the
temperature and ow elds were calculated.
kd1 kc1 Pg 4qR2p

4. Gridding issues
4.1. Complex geometry
The geometry of waste heat boiler is complex, for example it has multiple ducts (see Fig. 1) and
a multitude of air jets at dierent levels. The model is dened using a number of blocks and a nonorthogonal body-tted mesh. These blocks are three-dimensional arrays of computational cells
and can be joined with each other. Denition of the complex waste heat boiler model using such a
block structured mesh resulted in a constrained mesh which in turn lead to a slower convergence.
A stable numerical solution was achieved using the standard techniques like initial under relaxation of the momentum and the turbulence quantities and deferred corrections on the turbulence quantities. It was also found advantageous to use algebraic multi grid (AMG) solver for

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

831

the pressure correction equation, as the geometry has a complex shape and multiple blocks. The
non-orthogonality of the grids in the circular cross sections was minimised by a smoother grid
which matched the inner nodes with the curvature of the boundary edges using elliptic smoothing
technique. Solution stability was also found to be sensitive to the initial values of the turbulence
quantities.
4.2. Mesh sensitivity
In solving the numerical problem, the focus was on accurate prediction of multiple air jets and
their mixing with the plant o-gas without unduly increasing the computational size to prohibitively high levels. Initially a grid of size 40  8 was used, with a minimum of 40 cells needed in the
rst direction to adequately represent all the air jets. With the predictions showing air jets to be
highly diusive, the grid was made ner to 40  16. This has helped in reducing the level of
numerical diusion to acceptable levels. Since a further renement of mesh to 40  32 did not
generate any signicant changes to the solution, a nal mesh size of 40  16 was used for the
model.
5. Simulation of example waste heat boiler
5.1. Model description
The example waste heat boiler which is modelled here is illustrated in Fig. 1. The boiler consists
of radiant and convective passes to be constructed onto the existing after-burner and the air
required for combustion was to be injected through sets of slots in the radiant pass in three stages.
At each level air enters through nine injection ports, ve on one side of the boiler and the rest on
the opposite side. The plant o-gas carries char particles and dust particles known as primary
concentrate particles which are not chemically reactive. The plant o-gas is composed of combustible components such as CO and H2 , some moisture and a signicant percentage of inert
gases like N2 and CO2 .
Computation begins half-way up the former after-burner with the assumption of a uniform
ow eld and particle loading across this section, and extends down the radiative pass and up to
the entrance of the convective pass. Air injection ports are represented in the model by rectangular slots of equivalent cross sectional area. The resistance oered to the ow by the convective
banks was simulated by restricting the ow at the outlet.
The constructed model had 25 blocks of computational cells, subdivided into over 70,000
computational cells.
5.2. Isothermal model
After setting up the geometry, the ow was rst modelled under isothermal conditions to
qualitatively assess the major characteristics of the ow eld. For this purpose the plant o-gas
was assumed to have the properties of air at 800C.
Fig. 1 shows the overall geometry of the parts of the boiler and duct-work modelled, with
shaded contours of vertical velocity indicating the ow pattern. In addition, ow vectors are
overlaid to indicate the velocity direction and magnitude. Negative values indicate the ow going
downwards (dark shaded), while the positive values represent ow going upward (light shaded).
The approximate locations of the air jets are also indicated in the gure. Upward ow along the
front wall above the primary air jet indicates the presence of an established recirculating ow

832

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

pattern extending up into the hood. Another recirculation between the secondary and tertiary air
jets is also evident from the gure. The recirculation becomes weaker as the distance from the wall
increases (i.e. as the distance from the viewer increases). Below the level of the tertiary air jets, the
ow velocities are smaller and there is little recirculation.
To sum up, the isothermal results suggested that good mixing was likely, particularly between
the primary and the secondary air jets, and also indicated the typical recirculation ow pattern
that could be expected.
5.3. Mixed-is-burnt combustion model
The necessity to set up a radiation grid for the complex geometry with many convex surfaces of
after-burner region was obviated by the realisation that combustion is restricted to the boiler and
will not have any signicant upstream eect in the after-burner region. Thus it was decided to
exclude the after-burner from the combustion calculations. However to account for upstream ow
eects, the inlet boundary conditions dened at the entry to the hot-tie-in duct were set based on
the isothermal ow proles.
5.3.1. Gas phase
The predicted temperature distribution for the combustion model is given in Fig. 2, where
shaded contours indicate the temperature (given in K in the key). The velocity vectors indicate the
presence of a dominant recirculation ow pattern above the primary air jets. However the recirculation between the secondary and the tertiary air jets as observed in the isothermal predictions has almost disappeared with most of the ow tending to go down. This is mainly due to the
combustion taking place between the primary and secondary air jets which creates expansion of

Fig. 2. Thermal and ow predictions.

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

833

the mixture and drives it downwards. The temperature distribution also indicates that most of the
combustion takes place in this zone as a result of the intensive mixing of the fuel and the oxidant.
The lower temperature near the tertiary jets indicates less combustion in that zone. Lower done in
the boiler, the mixture cools as it leaves the radiant pass because of the heat losses to the walls.
The gas temperature in the lower part of the radiant pass is between 550C and 1100C (dark
shaded in the gure).
Fig. 3 depicts the temperature prole in a plane half way between the front and the back walls.
It is worth noting that the Mixed-is-burnt combustion model predicted two horse shoe shaped hot
spots closer to the walls between the primary and the tertiary air jets. Dark spots present in the
main combustion zone indicate the array of air jets.
Prediction of fuel mass fraction (Fig. 4) showed that almost all of the fuel is consumed in a
zone around the primary and secondary air jets. At the outlet, air occupies roughly 23% of the overall
mass ow. Prediction of combustion products of plantgas (Fig. 5) indicated clearly the entrainment of air and combustion products up into the hood by the recirculation above the primary air jets.
5.3.2. Char particles
Char combustion was modelled by tracking 100 representative char particles of three dierent
sizes ranging from 15 to 70 lm in diameter. In the absence of a measured size distribution, equal
numbers of the particles are tracked in the three size fractions. Particles are distributed uniformly
across the inlet, an assumption which is reasonable based on the particle tracks generated in the
isothermal predictions. Fig. 6 presents sample char tracks for 50 lm particles with shade intensity
representing particle mass (including ash). The ash content is 12%, so for a 50 lm particle the ash
mass is 1:6  1011 kg. Analysis of the particle track results le shows that char burnout is close to
100%, the residual mass in Fig. 6 being ash. A small fraction of the char particles are caught in the
recirculation above the primary jets and carried upward towards the hood. As particles pass

Fig. 3. Thermal prediction in a plane half way between front and back walls.

834

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

Fig. 4. Fuel mass fraction Mixed-is-burnt model.

Fig. 5. Combustion products mass fraction Mixed-is-burnt model.

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

835

Fig. 6. Sample tracks of char particles.

through the primary air jet plane, many are momentarily caught in one or other of the air jets and
move horizontally before continuing their descent. The `wiggly' nature of the tracks is a result of
turbulent eddies continually disturbing the paths one way or the other. Below the tertiary jets,
turbulence is less and the tracks are straighter.
Only a few of the particles of size 50 lm in diameter land in the hopper of the radiant pass:
hardly surprising given the small residual mass after burnout. A greater fraction of char particles
larger than 50 lm would fall into the hopper, but since a detailed particle size distribution is not
available, quantitative evaluation is not possible. Of all the char particles tracked, only 510% fall
into the hoppers.
The maximum temperature of any char particle when it hits the wall does not appear to exceed
its inlet temperature of 800C. As would be expected, there are few particles which hit the upper
part of the back wall after the ow is turned downward by the hood. They should not be sticky
since their temperature is still around 800C. There are also regions near the bottom of the radiative and convective passes where impact of particles occurs as a result of the changes in ow
direction. Here the particles are cooled suciently for sticking not to be a problem.
5.3.3. Primary concentrate particles
Three dierent sizes of primary concentrate particles ranging from 15 to 70 lm in diameter
have been tracked. Based on the results of particle tracking done under isothermal conditions, the
prole of particle concentration was taken to be uniform across the duct at the upstream inlet.
Fig. 7 shows sample tracks of primary concentrate particles with their temperature (in K)
indicated by varying shade intensity. Dark shading represents particle temperature of 850C and

836

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

Fig. 7. Sample tracks of primary concentrate particles.

below. From a temperature of 800C at the inlet, the particles heat up to around 1275C in the
combustion zone before nally cooling to 300C or so at the outlet. The particles are hotter than
900C for only a short part of their passage through the boiler, namely in the combustion zone
near the primary and secondary air jets. In the combustion zone the walls are water walls at
around 300C and this may reduce the tendency of the particles to stick to the walls.
The maximum temperature when the particles hit the walls is predicted to be just a few degrees
higher than the inlet value. This means there is little likelihood of the primary concentrate particles sticking to the wall.
The impact speed of primary concentrate particles hitting walls in the hood and combustion
zones was recorded during the trajectories calculation and analysed to estimate the likelihood of
erosion, since the rate of erosion depends strongly on velocity of impact. In this case most impacts
are at speeds lower than 6 m/s, though there are localised regions of higher speed collisions. One
region near the primary air jet plane is subject to impacts of 10 m/s and higher. Some particles
drop into the hopper of the radiant pass, while most of them are carried away into the convective
pass.
5.3.4. Comments on combustion model and predictions
As already noted, the combustion model used (Mixed-is-burnt) assumes instantaneous combustion of fuel as soon as it comes in contact with oxidant. This is usually an excellent approximation for gaseous combustion. In the present case, however, the plant o-gas has a low
caloric value, and as a result of the quenching eect of the diluent components, the fuel will burn
at a lower temperature and hence at a slower rate than conventional higher caloric value fuels.

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

837

For this reason, the actual maximum temperature may be somewhat lower than predicted by the
model which is based on instantaneous combustion.
5.4. Eddy break-up model
5.4.1. Gas phase
Discussions are on the predictions using the second approach, i.e., reaction kinetics of the gas
based on corresponding values of H2 . Fig. 8 depicts the overall thermal and ow eld. The
combustion is mainly conned to the primary air jet plane and below (light shaded), which is
dierent from the predictions using Mixed-is-burnt combustion model where the combustion zone
extends up to the boiler hood. This happens despite a recirculation above the primary air jets
similar to that in the Mixed-is-burnt model.
Fig. 9 shows some of the hot spots that occur closer to the wall opposite to the dividing wall in
a plane perpendicular to the one shown in Fig. 6. In this plane there is a clearly marked combustion zone dened by the primary and tertiary air jets, of nearly uniform temperature of around
1100C. Again when compared to the predictions of the earlier model the extent of hot spots near
the wall is signicantly less using the Eddy break-up model. This may lessen the probability of
primary concentrate particles sticking to the walls. The model predicts a maximum temperature of
1315C which is 40C below that predicted with Mixed-is-burnt model. The outlet temperature is
similar to the earlier predictions.The fuel mass fraction prole as shown in Fig. 10 emphasises the
earlier observation that the combustion is mainly conned to the primary air jet plane and below.

Fig. 8. Thermal and ow eld predictions Eddy break-up model.

838

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

Fig. 9. Temperature prole in a plane half way between front and back walls.

Plant o-gas (light shaded) is fully burnt (dark shaded) in the main combustion zone. Most of the
fuel is burnt between the primary and the secondary air jets. The unused oxygen mass fraction at
the outlet is essentially the same as predicted by the Mixed-is-burnt model.
5.4.2. Primary concentrate particles
Particles receive heat from the hot gases and rise in temperature up to a maximum of 1240C in
the combustion zone. In this run none of the particles hit the walls when they are hotter than
815C. It is worth emphasising that the predictions are based on the assumption of uniform
distribution of the particles at inlet and deviation from this assumption may alter the particle
tracks. Most impacts of primary concentrate particles hitting walls in the hood and combustion
zones are at speeds lower than 6 m/s and occur at the back wall above the primary air jet.
6. Conclusions
A model has been developed of a waste heat recovery boiler utilising typical plant o-gas
consisting of both gaseous and particulate combustibles. The model allows the calculation of
temperatures of gas and particles within the boiler and hence the likelihood of deposition onto the
boiler walls. The model was applied to a typical waste heat boiler geometry, and a typical o-gas
composition including a mixture of combustibles (char) and non-combustible particulates. Mixing
in the burner region, char burnout and char particle temperature were analysed using the model.

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

839

Fig. 10. Fuel mass fraction Eddy break-up model.

Combustion stability was also studied using a simple Eddy break-up model which accounts for
combustion kinetics and the results compared with a Mixed-is-burnt model.

Acknowledgements
The authors wish to acknowledge the support for this work by the Cooperative Research
Centre For New Technologies From Low Rank Coal which is funded in part by the Cooperative
Research Centres Program of the Commonwealth of Australia. The authors would also like to
thank Dr. Alan Manzoori and Mr. Malcolm McIntosh for their useful comments during the
course of the work, and Denise Goch for carrying out the Chemkin (1980) calculations.
References
Bakke, J.R., Hjertager, B.H., 1987. The eect of explosion venting in empty vessels. Int. J. Numer. Meth. Engrg. 24,
129140.
CFDS-FLOW3D., 1994. User Guide, Release 3.3., AEA Industrial Technology, Harwell Laboratory, Oxfordshire,
UK.
CHEMKIN: Kee, R.J., Miller, J.A., Jeerson, T.H., 1980. CHEMKIN: A General Purpose Problem Independent,
Transportable FORTRAN Chemicals Kinetics Code Package, Sandia Laboratories, Report SAND 80-8003.
Crowe, C.T., Sharma, M.P., Stock, D.E., 1977. The particle-source-in cell (PSI-CELL) model for gas-droplet ows.
Trans. ASME, J. Fluids Engrg. 99, 325.

840

M. Manickam et al. / Appl. Math. Modelling 22 (1998) 823840

Field, M.A., Gill, D.W., Morgan, B.B., Hawksley, P.G.W., 1967. The Combustion of Pulverised Coal, British Coal
Utilisation Research Association, Leatherhead, Surrey, p. 175.
Gosman, A.D., Ioannides, E., 1981. Aspects of Computer Simulation of Liquid-Fueled Combustors, AIAA Paper No.
81-0323.
Hjertager, B.H., 1982. Numerical Simulation of Turbulent Flame and Pressure Development in Gas Explosions. FuelAir Explosions, SM Study No. 16, University of Waterloo Press, Ontario, Canada, p. 407.
Lefebvre, A.H., 1983. Gas Turbine Combustion. Hemisphere Publishing Corporation.
Mackenzie H.J., 1990. The Numerical Modelling of the Interaction of Burner Jets in Brown Coal Fired Boilers, Ph.D.
Thesis, Monash University.
Malcolm McIntosh and Denise Goch., 1995. Private communication.
Van Doormal, J.P., Raitoby, G.D., 1984. Enhancements of the SIMPLEC method for Predicting Incompressible Fluid
Flows, Numerical Heat Transfer 7, 147163.
Smith I.W., Wall T.F., Baker J.W., Holcombe D., Harris D.A., Juniper L.A., Truelove J.S., 1989. The Combustion
Behaviour of Low-Volatile Australian Coals. NERDDC Project 1110, End of Grant Report, November 1989.
Shah, N.G., 1979. New Method of Computation of Radiation Heat transfers in Combustion Chambers, Ph.D. Thesis,
University of London.

Potrebbero piacerti anche