Fuel 158 (2015) 139–151

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

2D dynamic mesh model for deposit shape prediction in boiler banks

of recovery boilers with different tube spacing arrangements
Manuel García Pérez ⇑, Esa Vakkilainen, Timo Hyppänen
Lappeenranta University of Technology, Energy Technology, P.O. Box 20, FIN-53851 Lappeenranta, Finland

h i g h l i g h t s

 A new CFD model for deposit shape prediction is proposed.

 This model was tested for different geometries of a boiler bank in a Kraft Recovery Boiler.
 A dynamic mesh deforms the mesh, and the deposit grows according to the deposition.
 Results highlight deposition rates, shapes, and heat transfer penalties.

a r t i c l e i n f o a b s t r a c t

Article history: CFD tools are of great value in the design and operation of boilers. One particular aspect that can be mod-
Received 10 November 2014 eled by CFD is the ash deposition and plugging of heat transfer surfaces of boilers. Fouling and slagging
Received in revised form 18 March 2015 are the most typical causes of unscheduled boiler shutdowns. Consequently, appropriate predictions of
Accepted 28 April 2015
deposit geometries and rates are of considerable interest. CFD multiphase flow simulations are capable
Available online 19 May 2015
of modeling particle-laden streams and constitute a suitable tool for study of material deposition.
Fouling phenomena have a complicated and multidisciplinary nature involving thermo-fluid
Keywords:
mechanics, sticking/rebounding of particles, sintering, etc. If the deposit growth rate has been accurately
Kraft Recovery Boiler
Fume
calculated by a multiphase model appropriate for particle-laden flows, CFD dynamic mesh techniques are
Deposition able to move the interphase fluid-deposit according to the growth rate. This work develops and presents a
Dynamic mesh CFD model for prediction of deposition shapes in a classical boiler bank of a Kraft Recovery Boiler by
Black liquor combining a multiphase and a dynamic mesh model. The effect of tube transversal spacing is also
analyzed. The paper highlights the features and effects of the dynamic mesh model.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction Modeling of boilers is a topic of considerable interest within

Since their invention in 1934, recovery boilers have operated in
the pulp and paper mill industry. As a part of the Kraft pulping
process, they allow the recovery of inorganic cooking compounds
which are residuals from the fiber extraction process, for further
cyclical reutilization. As a secondary function, they also generate
steam for the mill. Since the black liquor is not a typical fuel and
there exist two main desired outputs (energy and reduction) from
the boiler, their operation, modeling and design are usually more
difficult than for other boiler applications [1]. In addition, the fact
that black liquor is a very ashy fuel makes the operation of Kraft
Recovery Boilers (KRBs) very challenging.
⇑ Corresponding author.
E-mail address: manuel.garcia.perez@lut.fi (M. García Pérez).
both academia and industry. CFD provides a powerful tool to
numerically solve the complex Navier–Stokes partial differential
equations. By using CFD it is possible to resolve the complex phe-
nomena in boilers such as fluid flow, turbulence, heat transfer,
chemical reactions and combustion, transport of mass/particles,
agglomeration, deposition, fouling, erosion, and pollutant forma-
tion [2–5].
A tool capable of modeling accurately and simultaneously all
the involved phenomena is not yet available as this is a multidisci-
plinary problem which entails really complex formulations, and
also because the domain itself (i.e., the target region to be
modeled) is of high complexity. Even with a narrow domain it
may happen that a particular problem is still hard to handle, e.g.,
an analysis of chemistry should take into account hundreds of
reactions among different phases of reactants, including phase
changes. As a consequence, CFD tools often try to model a specific

http://dx.doi.org/10.1016/j.fuel.2015.04.074
0016-2361/Ó 2015 Elsevier Ltd. All rights reserved.
140 M. García Pérez et al. / Fuel 158 (2015) 139–151

part of the boiler and solve only a limited set of the phenomena. necessary here. The model has been shown to work well for flows
Weber et al. [2] analyzed the state-of-the-art of CFD usage for where the volume fraction of particles is below 10% [6]. Higher vol-
ash behavior prediction. In their work, it was noted that CFD tools ume fractions would lead to a too strong coupling between the
are not widely used for boiler operation analysis and that many dilute and continuous phase motions.
results are still considered merely indicative. It was furthermore Tomeczek et al. [7] created a somewhat similar model for shape
observed that little work has been done on models that include prediction in Fluent. Even though they performed a steady state
accurately the effect of flow patterns over tube arrays. For instance, model, they showed that the mass can be computed in FLUENT
some models fail to fulfill the resolution requirements necessary with the user-defined function DEFINE_DPM_EROSION. In this
for accurate prediction of particle motion. It was also highlighted model, that routine was coded to add up the mass of each particle
that not much work has been done on models that consider the which sticks to a face.
flow over tube arrays, rather than a single tube. In addition, tran-
sient studies are often omitted, which means that the effects of
Von Kármán vortex shedding and the Coanda effect are lost. 2.2. Fume stream motion
These effects lead to a swinging fashion motion of the flow over
tube arrays [4]. A carryover-free stream is simulated. Due to inertial impaction,
The model presented here simulates and calculates fume depo- only the windward edges of the leading tubes would be prone to
sition on tubes of a boiler bank of a KRB by means of dynamic carryover deposits. This model focuses on fume deposition and
meshes. Special consideration is given to resolving the flow pattern its mechanisms.
around a tube array consisting of periodical repetitions in the Since fume is to be analyzed, special considerations must be
transversal spacing of a row of 4 tubes. In order to illustrate the taken. The motion of very small particles is significantly affected
effect of different flow patterns, different transversal tube spacings by flow drag. Therefore, inertial impaction (i.e., impaction due to
are simulated. A discrete phase model is used to simulate the ash the flow being unable to drag the particle away as it avoids the
particles in the flow. The deposition rate in the tube surfaces is windward side of a tube) is not expected to be very significant.
computed, and the deposit subdomain grows accordingly. On the other hand, thermophoresis and Brownian motion consti-
tute the major mechanisms of deposition for sub-micron particles
1.1. Study cases [1,5]. These two phenomena are considered in this model by
enabling their respective options in the software package.
Fouling is a major concern in KRB operation, especially in the
boiler banks (Fig. 1) [1,5]. This work presents CFD calculations of
fume deposit growth simulating the conditions in these heat
exchangers. A parametric analysis with 8 different values for the
transversal spacing between tube rows, st , is carried out. Each case
presents a computational domain consisting of a row of 4 tubes,
with a typical boiler bank tube diameter of D = 50 mm. The longi-
tudinal pitch remains constant at 3  D ¼ 150 mm between the cen-
ters of the tubes. Translational periodic boundary conditions on the
sides of the domain ensure an array configuration. Symmetry
boundary conditions should not be applied since the flow pattern
is not symmetrical although the geometry is. The transversal pitch
st corresponds to the width of the domain. The parametric analysis
of the transversal spacing consists of 8 different cases, varying the
spacing from 1.25 to 3.00 times the tube diameter, in steps of
0:25  D.
A discrete phase model is used to simulate the ash particles in
the flow. The deposition rate in the tube surfaces is computed,
and the dynamic mesh model deforms the mesh to make the
deposits grow according to the deposition rates. Transient consid-
erations are highlighted. Ansys FLUENT enhanced with
User-Defined Functions is used in this work.

2. Model description

2.1. Discrete phase model

A Lagrangian approach is used to track the particles. This kind of

model (also called the Discrete Phase Model) first solves the con-
tinuous phase (gas) alone and then models the discrete phase (par-
ticulate). The discrete parcels of particles are uniformly distributed
and injected through the inlet boundary condition at each
time-step. The number and mass of parcels is automatically set
to match the fume concentration (6 g/m3, as discussed in a later
section). Each parcel is tracked separately. It is possible to neglect
Fig. 1. Top: Deposit plug in a KRB shutdown between the boiler bank and the boiler
the particulate phase influence to the continuous phase, or bank screen. Deposits on the leading edge are close to plugging the whole space
consider coupling among phase calculations. Since the mass and between them. Bottom: Micrograph of fume retrieved at an electrostatic precip-
volume load of fume in gas are very low, the coupling is not itator of a KRB in eastern Finland.
 M. García Pérez et al. / Fuel 158 (2015) 139–151
2.3. Transient considerations
The combination of Kármán vortex streets and the Coanda
effect leads to a periodically unsteady flow in the wake of the
tubes. When tubes are arranged in different manners, different pat-
terns of flow are shed. Some of these patterns are prone to deposi-
tion of small-sized particles in the leeward area of the tube due to
vortice generation and swinging of the main flow direction [4].
Therefore, a transient consideration of the problem is required.
When platen-based heat exchangers (e.g. superheaters) are mod-
eled, the oscillations are not significant and thus the model may
be approximated as being steady-state flow [2,7]. Fig. 9, in the
results section, highlights this effect. The velocity field contains a
particle-trapping vortex surrounded by oscillating jets. It would
be impossible to catch such a pattern with a steady-state
simulation.
It has been suggested that a full oscillation period T should con-
tain at least 25 time-steps [6]. An empirical correlation to deter-
mine this oscillating period would be very handy. Unfortunately,
no work has been done to determine a correlation of the vortex
streets in a tube array for the Reynolds number range typical in
boiler applications. In this work a time step of 0.1 ms is chosen
and assumed as fine enough. It will be shown in Table 2,
Section 4.1, that this time step fits the criterion of 25 time steps
per oscillation.
Due to computing time reasons it is not possible to execute cal-
culations over an operationally significant time period (hours) with
such a short time step. In order to obtain statistically reliable data,
it is necessary that sufficiently long simulations be performed
before updating (moving) the dynamic mesh. The following para-
graphs describe how this issue is resolved.
The model routine proceeds as follows: First, the flow is simu-
lated with particles until vortex shedding becomes quasi-stable.
Then, the period of vortex shedding, T, is measured. Once T is
known, 10 periods of flow time are modeled. This duration of
10  T (typically 0.1 s) will be referred to here as a simulation. The
main output of this simulation is the total mass deposition rate
(in kg/s) that each tube collects over the duration of the simulation
time. This mass rate is used to calculate the total mass that the sur-
faces would collect over a long period of flow time (typically, sev-
eral minutes). This long period is referred to here as a cycle. The
duration of a cycle depends on the transversal pitch st and its
determination is explained in Section 4.1.
In other words, the total mass deposited during a period of time
of 10  T is used to compute the mass deposited during the total
cycle t cycle :
Fig. 2. Mesh for st = 1.75D. This figure focuses on the space between the 2nd and 3rd tube. Two periodic repetitions are shown for a better sketching of the tube pitches.
141
tcycle
mcycle ¼ msimulations ; ð1Þ
10  T
then, the dynamic mesh procedures modify the fouled surfaces
according to mcycle . Each mesh update accounts for the deposited
mass in that cycle only, which is reset to 0 each time. After this is
done, a new simulation is run with the modified fouled surfaces.
Ideally, the length of a simulation should be extended over
many flow oscillation periods in order to obtain better, less biased,
and more robust results. However, this penalizes the computation
time. Simulations of 10 periods were assumed to be a good
compromise between computation time and statistical reliability.
2.4. CFD model configuration
2.4.1. Two-dimensional flow
Deposition and other phenomena involved, such as turbulence,
are three-dimensional mechanisms. However, due to computing
limitations and the geometry of the domain, the model used here
assumes a 2D crossflow over tubes as other researchers have mod-
eled [2,3,8] with good results.
Therefore the results obtained by the model will be per unit of
length perpendicular to the domain, in this work denoted as L? .
2.4.2. Mesh
As discussed earlier, the diameter of the tubes is D ¼ 50 mm.
Rectangular domains with the tubes in the center line were built.
It was assumed that the inlet should start from 4  D upstream of
the leading edge of the first tube and span until 7  D beyond the
lee edge of the last tube. The height of those rectangles corre-
sponds to st (different in each simulation).
Due to dynamic mesh requirements which shall be discussed
later, triangular cells are required, and an initial deposit zone has
to exist in each tube. Therefore, the tubes were surrounded initially
with a deposit layer of 0.1 mm. The low particle Stokes number
(9:72  104 ) requires a specially fine meshing. The tubes and the
deposit-flow interfaces were meshed with 380 nodes, following
the suggestions of [3] on mesh resolutions for accurate fine particle
tracking. Triangular meshes were created in each deposit. For the
flow zone, a mesh size function was set up to control the cell
growth as a function of the distance from the tubes. A growing
factor of 1.2 was set, starting from size 0.25 mm, allowing a max-
imum cell size of 2.5 mm. The final mesh was as shown in Fig. 2.
142 M. García Pérez et al. / Fuel 158 (2015) 139–151

2.4.3. Dynamic mesh, requirements 2.5. Mass scattering and spreading

As stated earlier, a dynamic mesh model is used to modify the
deposit layer in each tube, according to the computed mass, by
moving the deposit-flow interface. In order to avoid overlapping
and extremely skewed elements in the process, the so-called
smoothing and remeshing methods are necessary [6].
Spring-based smoothing is used in this work. It treats the grid as
a network of interconnected springs. Hence, a moving zone would
pushes the nodes away instead of absorbing them. Therefore the
displacement is shared and absorbed by many cells within some
distance away from the moving deposit surface. Thus, nodes can
re-locate and accommodate in order to preserve minimum mesh
quality; especially close to the boundary. The damping of the net-
work is controlled by the spring constant factor: a value close to
zero means almost no damping, therefore, a displacement of a
given zone will affect cells far away; on the other hand, a value
close to one will mean that the movement is completely absorbed
by the closest cells. A spring constant factor of 0 was found to
behave best for this model.
The accumulated displacement of the interfaces after several
cycles will be too large, even when using spring smoothing meth-
ods, for the mesh to absorb the grid deformation. Eventually, some
cells will become very coarse and stretched in the deposit, and
some other cells will become very small in the fluid zone. To avoid
such a behavior, a local-face remeshing algorithm is used to col-
lapse clusters of cells whose sizes have fallen below a threshold
into bigger ones; and reversely, to split large cells into smaller
ones. This also helps to reduce the skewness of cells.
Unfortunately, the algorithm is limited to triangular (tetrahedral
in 3D) cells only and hence, quadrilateral cells are not possible
here. Consequently, pure triangular paved schemes have been used
to mesh the model. The minimum cell size allowed was 0.4 mm,
whereas the maximum cell size threshold was set to 3 mm.
2.4.4. Dynamic mesh node update
Whereas the mass deposition occurs in the faces of the inter-
phases deposit-fluid of the model, the growth of the deposit is dri-
ven by displacing its nodes. To determine these displacements,
mass balances can be rearranged (deposited mass divided by
deposit density yields the expansion of the surface, which is
directly related to the displacement of the points).
However, it has been stated [9] that direct individual-face based
mass balances (each face displacement covers a volume according
to the mass the face itself has collected) lead to model instabilities
due to numerical cancellation. To solve this issue, an alternative
formulation suggests that the nodes be moved in a direction deter-
mined by averaging the area vectors of its two adjacent faces. A
deposit mass balance comprising the two adjacent faces will set
the adequate magnitude of this displacement.
With this solution, the volume expansion generated by a given
node displacement ~ r does not match each mass collection of its
neighboring faces exactly, but the sum of both. Nonetheless, this
solution does not entail an important error, smooths the displace-
ments, and is numerically stable. The final expression for this new
procedure to obtain ~r which is surrounded by the faces i and j reads
as [9]:
As noted in Section 2.3, the flow is simulated and the results are
extrapolated to several minutes of actual flow time. Even after 10
full periods of simulation time, the collected mass as deposited is
still biased and needs a filtering procedure in order to simulate a
more realistic distribution.
Due to random turbulent effect on particle motion and to a
non-infinite number of particles tracked, a given face might not
collect any particle at all during a simulation, whilst the neighbor-
ing faces have collected several ones. Fig. 3 shows an example of
the mass collected by the second tube for the case st ¼ 1:75  D
after its first simulation and extrapolation. Fig. 4 sketches the def-
inition of the angle h used in the mass distribution figures. Each
point represents the mass collected by each face. The distribution
is highly discontinuous and it takes only discrete values due to
the impaction of integer numbers of particles.
In addition to being unrealistic, excessively scattered distribu-
tions lead to mesh instabilities and to eventual model crashes, as will
be further discussed in the results section. To counteract this prob-
lem, the model includes a filtering algorithm which smooths these
distributions. This algorithm spreads the mass m collected by a given
face j among the four neighboring faces and that face itself, i.e.,
among the faces j  2; . . . ; j þ 2. The mass is distributed symmetri-
cally but not equally among these faces. The closer a face is to the
face j, the more mass share it will receive. Respective weights of
1
=9 ; 2 9 , and 3 =9 (the last one for the face j) were selected for the mass
share among the faces. Fig. 5 presents this idea graphically.
Implementing this algorithm in a distribution such as the one
presented in Fig. 3 leads to a smoother and more realistic distribu-
tion, in which there would not be faces without deposition, and at
the same time high peaks will share mass among their neighboring
faces. Successive iterations of the algorithm will lead to smoother
distributions. If applied an infinite number of times, the resulting
distribution would be uniform and equal to the average mass
Fig. 3. Mass distribution as collected for the first tube of the case st ¼ 1:75  D.

a~i þ a~j mi þ mj 1
~
r¼   ; ð2Þ
j a~i þ a~j j j a~i j þ j a~j j qdep

where the subscripts i; j refer to the two different faces adjacent to

the node that is being moved; and for a given face k, a~k would
denote the area vector of that face (pointing outside the tube) and
mk would be the total deposited mass in the face. Finally, qdep stands
for the density of the deposit. Fig. 4. Referencing of the angle coordinate h for tubes.
 M. García Pérez et al. / Fuel 158 (2015) 139–151 143

Table 1
Main flow parameters.

Magnitude Symbol Value Units

Temperature T1 569 °C
Pressure p 99.706 kPa
Specific heat cp 1.258 kJ/kg K
Thermal conductivity k 0.0555 W/m °C
Viscosity l 3:38  105 kg/m s

Fig. 5. Sketch of the spreading algorithm when a particle of mass m hits the face j.
collected in the tube. In this work, the algorithm is applied a total
of 10 times for each simulation prior to the dynamic mesh rou-
tines. Fig. 6 shows the effect of this filtering method when applied
successively to the same initial mass distribution.
The parameters of the algorithm (four neighboring faces, ten
iterations and the mass weights) were tuned to obtain reasonably
realistic shapes based on visual observations of industrial deposits.
It is not trivial to obtain real data and deposit densities based on
visual experience. Additional work is required to study the algo-
rithm further and test different parameter possibilities.
selected for the flow. Other selected parameters of the incoming
gas flow are shown in Table 1.
The flow is considered as an ideal gas with a molecular weight
of 28.13 g/mol. The temperature of the tube wall surface is set to
305 °C. The mass flow of fume must be calculated before the parti-
cle injection. A typical value of fume concentration of 6 g/m3 has
been selected corresponding to actual measurements in particulate
flow [11]. For a 2D model, the mass flow per unit of perpendicular
length is:
mfume
¼ st  u1  ½F; ð3Þ
L?
where [F] denotes the fume concentration.

3. Parameter set-up 3.2. Properties of particles

3.1. Main flow conditions Although particles typically follow a specific diameter distribu-
tion, a very high peak appears at dp ¼ 0:7 lm, whereas other diam-
Typical flow features of a classic boiler bank in a KRB are simu- eters seem not to have a significant weight [11,12]. Thus, for sake
lated in this work. Based on [10], an incoming speed of 11.6 m/s is of simplicity, only that constant value of particle diameter was

Fig. 6. Resulting mass distributions after (in reading order) 1, 2, 5 and 10 iterations of the spreading algorithm done on the distribution shown in Fig. 3.
144 M. García Pérez et al. / Fuel 158 (2015) 139–151

considered here. Typically fume has a 0.95 mass fraction of sodium
sulphate [1]. Therefore, the physical properties of the particle (den-
sity, thermal conductivity and specific heat) have been taken as
those of sodium sulphate.
Stickiness of particles is a very complicated phenomenon.
Ideally, functions that calculate the stickiness efficiency of particles
(ratio of particles that stick in a surface over the total amount of
particles that hit the surface) should be included in the model.
Such a function would depend on many parameters (particle veloc-
ity, molten fraction, impaction angle, surface properties).
Temperature is of a major importance: particles are not pure sub-
stances, therefore they are very sticky within a temperature win-
dow at which the molten mass fraction spans from 15% to 70% [1].
According to [13] the fume is around 10% molten at the model
flow inlet temperature. In the absence of a reliable stickiness effi-
ciency correlation or measured value, a 50% of particles are
assumed to stick when hitting faces [1,5].
3.3. Deposits
The major feature of the deposit affecting the model is its shape,
the motion updating of which has already been described in
Section 2.4.4.
The bulk density of the deposit is computed as qdep ¼   qfume ,
where  is the deposit porosity. The fume density qfume ¼
2600 kg/m3 was obtained by mass-averaging the density of the dif-
ferent fume components [1]. The porosity is not uniform nor con-
stant over time since it eventually reduces as a consequence of
sintering [14]. Real-time volume shrinkage and hardening of the
deposit will be the subject of a future study; they fall out of the
scope of the present one. For this work, a constant uniform
value of  = 0.5 is assumed.
3.4. Turbulence
The Reynolds number based on the tube diameter and upstream
velocity is 6867, falling in the turbulent regime. Therefore, fully
developed turbulent flow is assured since velocities are typically
higher than that value when the flow surrounds the tubes
(due to transversal area constriction). Since the accuracy is critical
in near-wall regions, an SST k–x model has been selected with
standard input.
For the turbulent boundary conditions, a turbulent intensity of
7% and a viscosity ratio of 10 were set.
3.5. Flow-particle interaction
Due to the low mass and volume fraction of particles in the
stream, a weak coupling between the discrete and dilute phases
is expected. In a transient solution, the model tracks particles indi-
vidually. The sources of the dilute phase motions is affected by the
particles only once at the beginning of each time step. There is no
particle-flow coupling recalculation within a time-step iteration.
Stoke’s drag law is typically suggested for the tracking of very
small particles. A Cunningham correction factor of 5.49 has been
computed following the instructions and formulas in [6]. The
model will not account for buoyancy effects.
4. Results
4.1. Length of cycles
The flow was simulated along 10 motion swings each time.
Hence, these simulations consist actually of a short sampling time
(typically 0.15 s). Then, the collected mass in this short period is

Table 2
Cycle duration determination.

st /D 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00

Flow oscillation period (ms) 11.42 11.19 13.33 13.95 14.58 14.69 15.21 15.92
Simulation flow time (s) 0.114 0.112 0.133 0.140 0.146 0.147 0.152 0.159
Average deposition (g/m\) 0.12 0.09 0.10 0.10 0.07 0.06 0.03 0.04
Fouling layer growth (nm) 223 167 186 186 130 112 56 75
Time to reach 0.5 mm (min) 4.3 5.6 6.0 6.2 9.3 11.0 22.7 17.8
Cycle duration (min) 4 6 6 6 10 12 20 20
Extrapolating factor 2102 3217 2701 2581 4115 4901 7890 7538

Fig. 7. Negative volume formation, second tube, case st =D = 1.25.

 M. García Pérez et al. / Fuel 158 (2015) 139–151 145

extrapolated to a whole cycle of a certain duration (from 4 to
20 min, depending on st ). This means multiplying the collected
mass by an extrapolating factor, obtained by dividing the selected
cycle duration by the time the flow was actually simulated (10
oscillations). The duration of a cycle was determined after the first
simulation (for each st case) took place. It was adjusted to be an
integer number of minutes such that, upon the dynamic mesh
update, the deposit would experience a moderate growth
( 0:5 mm); not allowing for cycles longer than 20 min in any case.
The tube which presented the largest deposition was selected for
this observation. These operations and the selected cycle durations
are summed up in Table 2. Consequently, the cycle duration could
not be determined a priori as the results of the first simulation
were required. A per-cycle fouling growth of 0.5 mm was chosen
as a compromise between computational costs (if the growth is
too small) and early dynamic mesh instabilities crashing (if the
growth is too large).
4.2. Model challenges
In spite of applying mass scattering and spreading as described
in Section 2.5, major model-crashing problems arise eventually.
The dynamic mesh updating is not able to absorb too large accu-
mulated displacements, leading to negative-volume cells and
cracks. Fig. 7 highlights the formation of negative volumes. Cells
of the flow domain (in red) have entered (overlapped) the deposit
region (orange), and vice versa. So that the cells are still connected
to the interphase, some cells have had to turn over. This failure is
detected when the CFD package computes the volume (which
corresponds to the cell area in a 2D mesh) of the cell. This

Fig. 8. Cracks appearing at the surface of the deposit. Further iterations on the mesh according to Eq. 2 would eventually lead to the collapse of the interfaces and to an
incorrect numerical solving of the Navier–Stokes equations locally.

Fig. 9. Instantaneous capture of flow field pattern. The view is between the 1st and 2nd tube, st =D = 1.75. The surfaces are already fouled after 54 min (9 cycles of 6 min).
146 M. García Pérez et al. / Fuel 158 (2015) 139–151

Fig. 10. Mass density distribution for each tube (from top to bottom) for values of st =D of 1.50 (left), 2.25 (middle) and 3.00 (right). For a given plot, a line is plotted for each
cycle, showing the temporary evolution for each case. Each line represents the final distribution after each cycle, therefore they are separated by several-minute intervals, as
shown in Table 2. Note that the final deposited mass does not correspond to the same fouling duration among columns; each column has its own fouling time.

computation is done as a product of the vectors which connect the In addition to negative volumes, crack formation is another
nodes. Since these nodes are always accounted in the same order challenge of this model. Cracks tend to appear when neighboring
and the cell has turned over, the computed volume is negative. faces experience very different mass depositions. Therefore, these
The simulation cannot continue anymore beyond that point. faces, which would be expected to move towards the same
 M. García Pérez et al. / Fuel 158 (2015) 139–151 147

Fig. 11. Aspect of final deposit shapes and mesh. From top to bottom, st =D ¼ 1:50, 2.25 and 3.00. The longitudinal spacing between the tubes has been reduced to save space.
The flow comes from left to right. The fouling times are, from top to bottom, 1:54, 3:10, and 5:20 [h:mm]. The deposits are coloured by temperature in °C. The tubes are
oriented as in Fig. 4. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

direction, may start to move and grow facing each other. This is
highlighted in Fig. 8. When this happens, even though the model
is still able to simulate and continue the calculations; the flow
region inside the cracks will not be predicted accurately, entailing
convergence problems. These zones become singular regions
where the CFD package is not able to solve the Navier–Stokes equa-
tions properly.
Due to a strong Coanda effect, the high-velocity streams leaving
the tight spacing between the tube rows deviate towards the oppo-
site side of the following tube, leading to a heavy and irregular
deposition in angles close to h  3p=4. Fig. 9 illustrates this speci-
fic moment. The figure shows that a strong jet impacts the second
tube around the previously mentioned angle. The stream becomes
faster and stronger as st reduces, because of the constriction of area
for the gas to flow. These regions are prone to the discussed mesh
problems.
It may be possible to utilize further filtering and smoothing of
the deposition surface to avoid the above-mentioned problems.
Other smoothing techniques different from the one presented here
may be an interesting scope of future research.
For each different st , simulations were carried out until the first
negative-volume cell appeared. This event constitutes a model fail-
ure which makes it impossible to continue further. Typically, each
case crashed (i.e., presented a negative-volume cell) after around
20 cycles, or mesh iterations. Since the length of each cycle is dif-
ferent for each st , the final collected mass or the final deposit shape
for a given case does not correspond to the same fouling time as
another case with another st . This has to be taken into account
when interpreting the results and figures presented in the next
sections.
4.3. Mass distributions
The deposition for each tube in each cycle was calculated. The
density of deposition (mass of deposit per unit of tube area) as a
function of the tube angle, h for each cycle, tube and st is presented.
Fig. 10 consists of an array of graphs where these distributions are
shown. Each row corresponds to a tube (from top to bottom, 1st tube
to 4th tube). Columns, from left to right, stand for st =D = 1.50, 2.25
and 3.00 respectively. Only these cases of st are shown here, for
the sake of space. As mentioned earlier, the final value of total col-
lected mass does not correspond to the same amount of fouling time
for different cases of st , since each one has its own cycle duration and
its own number of total calculated cycles.
An increased density of deposition towards h  0 and h  p
can be noted. This increase is due to particle trapping in vortexes
in the longitudinal gaps between the tubes as shown in Fig. 9.
Fume particles get trapped in these vortexes and thermophoresis
has enough time to act on them and push them towards the sur-
faces. However, in the transversal gaps between the tubes, the
main gas stream moves too fast, dragging the fume and not letting
it travel towards the tube. As st increases, the speed reached in this
area is smaller since there is a smaller area constriction for the flow
to travel through it. Hence, the effect of having large peaks around
148 M. García Pérez et al. / Fuel 158 (2015) 139–151
Fig. 12. Comparison of the initial (in blue) versus the last (in orange) deposition rate distributions after 4 h and 20 min of fouling. Case st =D ¼ 2:75. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)
leading and rear edges of the tube and low deposition rates around
the sides (h   p2 ) is expected to decrease with st , as can be seen in
the plots of Fig. 10.
4.4. Fouling shapes
Similarly to the collected mass deposition, the final shapes of
the deposits are shown in Fig. 11 for the same st values used in
Fig. 10. The appearance of these figures matches with what intu-
itively one would expect to see of a fouled tube. The surface rough-
ness and non symmetrical shape of the deposit are due to random
turbulent fluctuations that the CFD package simulates on the fume
particle trajectories, leading to a somewhat random roughness in
the tubes as happens in real boilers.
5. Discussion
5.1. Effect of the deposit on successive deposition rates
The observed deposition rate was found to be dependent with
the time, since for a given tube, the deposition rate over a clean
tube (i.e., the mass deposition distribution collected in the first
cycle) does not match the last distribution (fouled tube). Fig. 12
illustrates this issue. This implies that the deposit may affect the
flow and/or the deposition mechanisms, since it is the only entity
varying between different cycles.
It is observed that the deposition rates generally (not particu-
larly for each location of a given tube, but in most areas over the
circumference and the overall value) decrease with the time
(Fig. 12). This is also mentioned in the next section. The coating
of the tube leads to weaker thermophoretical forces, explaining
this effect. A second effect which can be observed is a more irreg-
ular profile in the case of the fouled tube when compared to the
clean surface. These irregularities may be explained by the gener-
ation of peaks in the surface: the initial deposition entails some-
what smooth irregularities corresponding to a somewhat
non-uniform fouled shape after the first mesh update. The areas
that point out of the deposit tend, at least slightly, to stagnate
the flow around them; thus they are able to capture more particles
and favoring increasing deposition on them. This would result in
an increasingly pointed deposit after a sufficiently long time,
explaining the growing irregularities over time and the rapid for-
mation of peaks. The aggregated effect of the irregularities can be
seen in Figs. 7 and 10, also in Section 4.2.
5.2. Per-tube mass deposition comparison
The mass density deposited in the tubes was integrated along
the angle and plotted for comparison. Fig. 13 represents the total
mass deposited at different times, comparing all the tubes, rear
and front parts.
The general trend that can be observed from Figs. 11 and 13 is
that the deposition rates generally decrease with the parameter st .
The fluid reaches high velocities at low st and drags the fume
towards the wind side of the following tube. Interestingly, the front
of the first tube is the surface that has received the smallest
amount of mass. Most of the particles avoid the first tube, just
being dragged by a smooth flow pattern. It should be remarked
that this only happens in the first tube: it is the only one where
the front deposition is smaller than the lee one. Afterwards, they
 M. García Pérez et al. / Fuel 158 (2015) 139–151 149

Fig. 13. Total mass deposited in tubes, per tube, per st . Each graph corresponds to a given time, which is specified. Cases of st which have crashed before specified time are
marked with a black bold X.

Fig. 14. Temperature field, in °C, for the tube array (bottom) and detail of the third tube (top). Note the significant temperature drop due to tube coating. st =D = 1.25.

accelerate in the transversal pitches and enter a vortex-dominated impaction is not significant for the first tube, but that the condi-
flow pattern inside the longitudinal pitch between two tubes. The tions of the oscillating flow patterns between tubes do project
combination of these two phenomena causes high depositions on the particles fast to the following tube, making it collect more
the front side of the following tube. It can be deduced that inertial particles.
150 M. García Pérez et al. / Fuel 158 (2015) 139–151
Also, it can be noticed that each tube experiences a larger depo-
sition in the wind side than the preceding one. At first sight, this is
intuitively contradictory with the idea that a larger temperature
gradient would lead to stronger thermophoretic strengths; since
the temperature difference between the tubes and flow would
decrease due to the gas cooling down. However, according to
Talbot et al. [15], this thermophoretic force is also inversely pro-
portional to the static pressure, which decreases tube after tube
due the flow motion itself. This explains why each tube suffers
from a larger deposition that the preceding one. In addition, this
also explains why this effect becomes weaker as st increases, since
a larger st entails smaller pressure drops over the fluid domain.
5.3. Heat transfer penalty and predicted sootblowing rates
Being porous and an insulating material, the thermal conductiv-
ity of the deposit is very low. As a consequence, the deposit growth
coats the tubes and entails a penalty on heat transfer. For this
work, a conductivity of 0.08 W/m K was chosen. This is a low value
but still within the typical limits for fume deposits [16]. Fig. 14
shows the temperature field for the case st /D = 1.25. The tempera-
ture gradient is practically concentrated within the deposit, and
therefore, the gas does not cool down as it did at the initial condi-
tions (with clean tubes).
It is possible to evaluate the performance by computing the
total temperature drop within the domain (T inlet  T outlet ). This tem-
perature difference will reduce as the deposits grows. Hence, the
relative heat performance is defined here as
T inlet  T outlet ðtÞ
gðtÞ ¼ ; ð4Þ
T inlet  T outlet ð0Þ
where T denotes the averaging of the temperature, first on space (to
obtain the average temperature of the profile) and then in time (to
obtain the mean value out of an expected time-dependent oscillat-
ing profile). T inlet is constant and uniform as stated in Section 3.1
(569 °C).
These results are summed up in Fig. 15. Generally, it is observed
that the efficiency drops drastically at the beginning of the fouling
process. This penalty on heat transfer seems to be more problem-
atic at smaller st , since these arrangements experienced the largest
depositions.
Fig. 15. Heat transfer performance evolution for all the cases of st =D (presented in
the legend).
The marginal penalty in the performance reduces with time.
Coating becomes less effective as the deposit grows since the mar-
ginal increase in heat transfer resistance becomes less significant
over time. Moreover, the thermophoresis becomes less important
as the temperature of the deposit surfaces increases and
approaches the flow temperature. The deposit absorbs a more sig-
nificant share of the temperature gradient each time (see upper
part of Fig. 13). Thus, the thermophoretic force is expected to
weaken.
The value of the heat transfer performance is directly related to
the fouling rate. Typically, sootblowing is recommended when the
performance drops to 0.8 or 0.7 [1,5]. Looking at Fig. 15, the model
is capable of predicting effectively how often sootblowing is
needed, after comparing the predicted values with boiler operation
experience [5,10].
In addition, the weak coupling effects that fume concentration
exerts on the flow field suggests that these results would behave
very close-to-proportionally with the fume concentration [F].
Under this assumption, a similar simulation carried out with a dif-
ferent fume concentration would be expected to show a propor-
tional growth rate and hence, inversely proportional sootblowing
rates. For instance, if upon inspection of Fig. 15 we deduce that
sootblowing will happen hourly, a reduction of a 20% in [F] would
extend the sootblowing rates by around 12 min.
Unfortunately, the same proportional scaling cannot be said
about the thermal conductivity of the deposit. To illustrate this,
the first tube of the case st =D ¼ 2:25 could be examined in detail.
The average temperature of the deposit external surface is
521.8 °C after 4 h of fouling. The flow temperature is 569 °C and
the inner tube temperature is 305 °C (as stated in Section 3.1).
There are two heat transfer mechanisms: convection from gas to
deposit, and conduction through deposit to tube surface. Each ther-
mal resistance will be proportional to its temperature drop. With
the temperatures earlier described, this means that the conduction
resistance represents about the 82% of the total. The conductive
resistance is significantly larger than the convective resistance,
but the former still cannot be neglected. Moreover, if these calcu-
lations had been done on an earlier cycle (i.e., after 1 or 2 h of foul-
ing), or if the thermal conductivity were larger (it should be kept in
mind that a relatively low value of thermal conductivity was
selected for this study), the share of the conductive resistance rel-
ative to the total resistance would be even smaller. Therefore, it
can be concluded that the energy performance cannot be scaled
in terms of thermal conductivity.
6. Conclusions
In this work, a CFD model for prediction of deposition of ash in
the tube banks of a Kraft Recovery Boiler was presented. Ansys
FLUENT was used with user-defined functions to calculate the
deposited mass. A Lagrangian approach was used to track the ash
particles. A single particle diameter and one Reynolds number
was used. The transversal spacing of the bundle was varied. The
flow analysis is two-dimensional and transient, and captures vor-
tex shedding in the wake of the tubes. Periodic boundary condi-
tions were used in the transversal direction. A lack of relevant
empirical data did not allow for a proper model validation espe-
cially with respect to the mass spreading. Hence, experimental
research is clearly required.
An innovative feature of the study is the dynamic mesh adap-
tion to the shape of the interface between the gas phase and the
ash deposited on a tube surface.
Difficulties with the dynamic mesh generation were illustrated
and discussed. Sample results of the mass deposition profile were
presented for some cases. The final fouling shapes were shown
and the heat performance penalty discussed.
 M. García Pérez et al. / Fuel 158 (2015) 139–151
The model showed that the tube arrangement has a major influ-
ence on the deposition, due to vortices formation on the lee side of
the tubes. The model analyzed the penalty on heat transfer capabil-
ity due to deposit growth. It was shown that larger transversal
spacings seem to show smaller depositions, and therefore smaller
penalties on heat transfer performance.
This model could be further enhanced with the inclusion of
improved particle stickiness correlations and modeling of the sin-
tering of the deposit. Further study of mass spreading–smoothing
algorithms is also required.
The usage of CFD tools for deposition phenomena is still at an
early stage, and results can only be taken as orientative (i.e., obser-
vation of trends). However, the capabilities of this sort of tools
seem to be only potentially limited by computing resources, which
are expanding constantly. Additional research is necessary in order
to expand the state-of-the-art and produce CFD models with reli-
able results.
Acknowledgments
This research has been funded by the Doctoral Program in
Energy Efficiency and Systems (EES), Finland.
The authors would like to thank the anonymous reviewers, who
have helped to improve this manuscript with valuable comments
and suggestions.
References
[1] Adams T, Frederick J, Grace T, Hupa M, Iisa K, Jones A, et al. Kraft recovery
boilers. Atlanta: Tappi Press; 1997.
151
[2] Weber R, Mancini M, Schaffel-Mancini N, Kupka T. On predicting the ash
behaviour using computational fluid dynamics. Fuel Process Technol 2013;
105:113–28.
[3] Weber R, Schaffel-Mancini N, Mancini M, Kupka T. Fly ash deposition
modelling: requirements for accurate predictions of particle impaction on
tubes using RANS-based computational fluid dynamics. Fuel 2013;108:
586–96.
[4] Ishigai S. Steam power engineering: thermal and hydraulic design
principles. Cambridge University Press; 1999.
[5] Vakkilainen E. Kraft recovery boilers: principles and practice. Helsinki
University of Technology, Energy Engineering and Environmental Protection;
2007.
[6] ANSYS Inc, Fluent 14.0 users manual, 2011.
[7] Tomeczek J, Wacławiak K. Two-dimensional modelling of deposits formation
on platen superheaters in pulverized coal boilers. Fuel 2009;88(8):1466–71.
[8] Pyykönen J, Jokiniemi J. Modelling alkali chloride superheater deposition and
its implications. Fuel Process Technol 2003;80(3):225–62.
[9] García Pérez M, Vakkilainen E, Hyppänen T. CFD for deposit formation in kraft
recovery boilers. In: Proceedings of the 2014 international chemical recovery
conference, suomen soodakattilayhdistys, The Finnish Recovery Boiler
Committee, and TAPPI, Tampere (Finland); Juna 8–13, 2014.
[10] Vakkilainen EK. Offdesign operation of kraft recovery boilers. Lappeenranta
University of Technology; 1993.
[11] Jokiniemi JK, Pyykönen J, Mikkanen P, Kauppinen EI. Modeling fume formation
and deposition in kraft recovery boilers. Tappi J 1996;79(7):171–81.
[12] Mikkanen P, Kauppinen EI, Pyykönen J, Jokiniemi JK, Aurela M, Vakkilainen EK,
et al. Alkali salt ash formation in four finnish industrial recovery boilers.
Energy Fuels 1999;13(4):778–95.
[13] Vakkilainen E. Predicting ash properties in recovery boilers. In: Proceedings of
the 2010 international chemical recovery conference. Williamsburg Lodge,
Williamsburg, Virginia (USA); March 29 – April 1, 2010.
[14] Frederick W, Vakkilainen E, Tran H, Lien S. The conditions for boiler bank
plugging by submicrometer sodium salt (fume) particles in kraft recovery
boilers. Energy Fuels 2004;18(3):795–803.
[15] Talbot L, Cheng R, Schefer R, Willis D. Thermophoresis of particles in a heated
boundary layer. J Fluid Mechan 1980;101(04):737–58.
[16] Baxter LL, Gale T, Sinquefield S, Sclippa G. Influence of ash deposit chemistry
and structure on physical and transport properties. In: Developments in
thermochemical biomass conversion. Springer; 1997. p. 1247–62.