Accepted Manuscript

Computational Fluid Dynamics Modeling and Experimental Validation of Heat

Transfer and Fluid Flow in the Recovery Boiler Superheater Region

Viljami Maakala, Mika Järvinen, Ville Vuorinen

PII: S1359-4311(18)30358-2
DOI: https://doi.org/10.1016/j.applthermaleng.2018.04.084
Reference: ATE 12081

To appear in: Applied Thermal Engineering

Received Date: 16 January 2018

Accepted Date: 16 April 2018

Please cite this article as: V. Maakala, M. Järvinen, V. Vuorinen, Computational Fluid Dynamics Modeling and
Experimental Validation of Heat Transfer and Fluid Flow in the Recovery Boiler Superheater Region, Applied
Thermal Engineering (2018), doi: https://doi.org/10.1016/j.applthermaleng.2018.04.084

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.
 Computational Fluid Dynamics Modeling and Experimental Validation of Heat Transfer
and Fluid Flow in the Recovery Boiler Superheater Region

Viljami Maakalaa,∗, Mika Järvinenb , Ville Vuorinenb

a Andritz, Helsinki, Finland
b Aalto University, Department of Mechanical Engineering, Espoo, Finland

Abstract
Development of predictive computational fluid dynamics (CFD) methods for recovery boilers would be highly beneficial for
the development of such very large-scale energy production applications. Herein, unique experimental data is compared with a
developed CFD framework demonstrating the predictive character of the present simulations. The novelty of the work consists of the
following: 1) We report two sets of previously unpublished full-scale temperature and flow field measurements from the recovery
boiler superheater region. The data from these challenging measurements is very valuable, since reported experimental data on
recovery boilers is scarce in literature. 2) We introduce a detailed, three-dimensional CFD model for the recovery boiler superheater
region. The results of the model are verified computationally and validated with the experimental data. 3) We demonstrate the
added-value of the developed CFD model with a detailed analysis of the three-dimensional flow field and heat transfer results. In
addition, we consider the implications of the three-dimensional solution for the estimation of fouling.
Keywords:
heat transfer, efficiency, measurements, computational fluid dynamics, recovery boiler

1. Introduction

Recovery boilers are an integral part of the chemical recovery

cycle of pulp mills. Such boilers are utilized to combust black
liquor, which contains the cooking chemicals along with lignin
and other organic compounds separated from the wood during
the pulping process. The two main functions of recovery boilers
are to regenerate the chemicals, such that they can be re-used
in the pulping process, and to generate high-pressure steam for
heat and power generation. Black liquor is considered a renew-
able fuel, since its combustible part originates from wood, and
thus recovery boilers are an important source of renewable en-
ergy. Figure 1 shows a modern, high-capacity recovery boiler.
The global production of chemical kraft pulp is approxi-
mately 130 million air dry tonnes per year and it is expected
to grow by approximately 1% per year (Pöyry, 2015). The pro-
duction of one tonne of air dry pulp produces one and a half
tonnes of dry black liquor solids. This means that 195 mil-
lion tonnes of dry black liquor solids are combusted in recov-
ery boilers every year. In terms of energy, this is 731 TWh per
year, assuming a typical higher heating value of 13.5 MJ/kg for
the dry black liquor solids. Most of the energy generated in the
recovery boilers is utilized for the process purposes of the pulp
mills, but especially modern units generate substantial amounts
of additional energy, which can be utilized for supplying renew- Figure 1: A modern, high-capacity recovery boiler. The boiler consists of two
able electricity to the power grid. As an example, in Finland in main sections: a furnace section and a heat transfer section, which are separated
2016, 8% of total electricity and 18% of renewable electricity by the nose. The superheater region, which is the focus of this work, is high-
lighted. High-capacity recovery boilers are physically large, even 20 m wide,
20 m deep, and 80 m high.
∗ Corresponding author. Email: [email protected]

Preprint submitted to Applied Thermal Engineering April 13, 2018

 Nomenclature

f~ body force [N] T temperature [K]

~g gravitational acceleration [m/s2 ] T gas flue gas temperature [K]
~u velocity [m/s] T int water/steam side (internal) temperature [K]
Dn, eff effective diffusivity of the species n [m2 /s] T surface surface temperature [K]
eavg average relative error [%] T 15 sticky temperature [K]
h representative cell size [mm] T 70 flow temperature [K]
h sensible enthalpy [J/kg] y+ dimensionless wall distance [-]
hn sensible enthalpy of the species n [J/kg] Yn mass fraction of the species n [-]
Jn diffusion flux of the species n [kg/(m2 s)] GCIavg average grid convergence index [%]
k turbulent kinetic energy [m2 /s2 ]
Greek Symbols
kfluid inner tube surface to water/steam heat transfer coefficient
λ thermal conductivity [W/(mK)]
[W/(m2 K)]
λdeposit deposit thermal conductivity [W/(mK)]
kgas flue gas side convective heat transfer coefficient [W/(m2 K)]
λeff effective thermal conductivity [W/(mK)]
kint water/steam side (internal) heat transfer coefficient [W/(m2 K)]
λtube tube thermal conductivity [W/(mK)]
l characteristic length [m]
µeff effective viscosity [kg/(ms)]
Nu Nusselt number [-]
ρ density [kg/m3 ]
p apparent order of the discretization method [-]
p pressure [Pa] Abbreviations
q surface heat flux [W/m2 ] CFD computational fluid dynamics
qconv convection heat flux [W/m2 ] ES equiangle skewness
qrad radiation heat flux [W/m2 ] FSR full superheater region
r refinement ratio [-] GCI grid convergence index
r tube radius [m] HHV higher heating value
Rn source of the species n from chemical reactions [kg/(m3 s)] NHV net heating value
Re Reynolds number [-] OQ orthogonal quality
sdeposit deposit thickness [m] RANS Reynolds-averaged Navier-Stokes
S rad radiation source term [W/m3 ] SH superheater
stube tube thickness [m] SSR slice superheater region

was produced by black liquor combustion in recovery boilers (2012) studied the causes for asymmetric furnace temperatures
(Statistics Finland, 2017). using CFD modeling and compared the results against valida-
In recent years, the global goals regarding the climate change tion measurements. Bergroth et al. (2010) and Engblom et al.
mitigation have focused considerable attention on energy gen- (2010a) focused on understanding the physical and chemical
eration from renewable sources. Accordingly, the traditional phenomena occuring in combustion on the char bed, which is
role of the recovery boilers as chemical recovery units is chang- located at the bottom of the furnace, and developed a CFD-
ing, and focus is shifting toward simultaneously maximizing based model for the dynamic evolution of the char bed. Fer-
the generation of renewable energy. To achieve these goals, reira et al. (2010) utilized CFD modeling to study the effects
it is important to understand the combustion and heat transfer of various combustion air injection schemes. Li et al. (2012)
processes of such boilers and to develop modeling methods for developed thermal boundary conditions for the furnace walls
optimizing their efficiency. and compared CFD results against measurements. Multiple re-
searchers have also focused on black liquor combustion. Walsh
Computational fluid dynamics (CFD) modeling can be con- and Grace (1988), Frederick (1990), Saastamoinen (1996), and
sidered to be an established research tool for ensuring and op- Verrill and Wessel (1998) have studied droplet combustion and
timizing the performance of recovery boilers, in both research published increasingly detailed CFD-applicable models. The
organizations and industry. CFD modeling of recovery boil- most detailed CFD-applicable droplet combustion model up-
ers was pioneered approximately 30 years ago when several re- to-date has been developed by Järvinen et al. (2002, 2011). Our
search groups independently developed the first comprehensive earlier work was also connected to combustion in the recov-
CFD models for recovery boilers (Grace et al., 1989; Jones, ery boiler furnace (Maakala and Miikkulainen, 2015; Maakala
1989; Walsh, 1989; Sumnicht, 1989; Karvinen et al., 1991; Sal- et al., 2017).
cudean et al., 1993; Abdullah et al., 1994). Since then, the work
has been continued and increasingly detailed recovery boiler Several researchers have published data measured from re-
CFD models have been developed (Blasiak et al., 1997; Wes- covery boilers and performed important validation work of
sel et al., 1997; Vakkilainen et al., 1998; Mueller et al., 2004; CFD models. In addition to the previously mentioned, Grace
Jukola et al., 2014). et al. (1998), Saviharju et al. (2004, 2007), Brink et al. (2010),
The main focus of the recent research has been the model- Miikkulainen et al. (2010), Vainio et al. (2010), and Engblom
ing of the furnace processes and combustion. Engblom et al. et al. (2010b) have contributed to model validation. However,
2
the amount of published measurement data and validation work changers (Prithiviraj and Andrews, 1998a,b; Hayes et al., 2008;
of recovery boiler CFD models is limited, since full-scale mea- Shi et al., 2010; Kritikos et al., 2010; Yang et al., 2014).
surements on a recovery boiler are extremely challenging and The distributed resistance method has often been utilized in
time consuming. In addition, physical scale modeling of recov- boiler simulations in a simple manner with approximated pres-
ery boilers is impractical or even impossible. sure loss coefficients and predetermined volumetric heat sink
A considerable amount of CFD research has also been done values (Le Bris et al., 2007; Dı́ez et al., 2008; Choi and Kim,
in connection to coal-fired utility boilers. Yin et al. (2002, 2009; Miikkulainen et al., 2010; Al-Abbas et al., 2012; Maakala
2003) and He et al. (2007) simulated coal-fired boilers with and Miikkulainen, 2015). Yin et al. (2002, 2003) and He et al.
a particular focus on the gas flow deviations and uneven wall (2007) modeled the first two superheaters as thin, constant-
temperatures in the upper furnace and crossover pass regions. temperature platens and the rest of the superheaters and re-
Le Bris et al. (2007), Dı́ez et al. (2008), and Choi and Kim heaters as porous media with predetermined pressure loss co-
(2009) simulated the combustion processes with a focus on the efficients. Park et al. (2010) applied the distributed resistance
modeling of NOx emissions. Park et al. (2010), Edge et al. method in a more sophisticated manner, such that the heat sink
(2011), and Schuhbauer et al. (2014) studied the combustion values were connected to a process simulation model. Drosatos
processes and the steam cycle via coupled CFD-process simu- et al. (2014, 2017) calculated the pressure loss coefficients from
lations. Nikolopoulos et al. (2011) and Al-Abbas et al. (2012) empirical correlations and solved the heat transfer using the
simulated lignite-fired boilers under air-fired and oxy-fuel con- macro heat exchanger model. Schuhbauer et al. (2014) uti-
ditions. Drosatos et al. (2016) modeled a lignite-fired boiler lized pressure loss coefficients based on design data and mod-
under full load and various partial load conditions. Drosatos eled heat transfer by considering convection and radiation sepa-
et al. (2014, 2017) presented decoupled simulations of the fur- rately. The convective coefficients were calculated from empiri-
nace and convective section in two lignite-fired boilers. cal Nusselt number correlations. The radiative coefficients were
In the present work, we mainly focus on the superheater re- calculated based on the flue gas properties and particle load. In
gion of the recovery boiler (see Figure 1). After combustion in addition, the heat sink values were coupled with a process sim-
the furnace, the hot flue gas flows to the superheater region. The ulation model. The review by Dı́ez et al. (2005) presents several
purpose of the superheater region is to heat the high-pressure heat transfer models for coal-fired boilers, many of which are
steam flowing inside the superheaters to a high temperature via applicable in the context of the distributed resistance method.
heat transfer from the hot flue gas. Since approximately 30% of The computational cost of the distributed resistance method is
the recovered energy is captured through heat transfer to the su- small and it is a reasonable simplification especially when the
perheaters, they are one of the largest and most important heat focus of the research is not a detailed solution in the heat trans-
transfer surfaces in the recovery boiler. In addition, the per- fer section.
formance of the superheaters is critical for the efficiency and However, the distributed resistance method has several dis-
availability of the whole recovery boiler power plant. advantages. Paraphrasing Zhang et al. (2009), the three main
Hence, it can be understood that the superheater region has drawbacks are: 1) many additional geometrical parameters,
been one of the focal research topics in the past. Vakkilainen such as volumetric porosities and surface permeabilities, must
et al. (1991, 1992) studied the optimal shape of the boiler nose be known, 2) distributed resistances and heat transfer coeffi-
using both physical and CFD modeling. Kawaji et al. (1995) de- cients must be provided from existing experimental correla-
veloped a model for predicting heat transfer in the superheater tions, and 3) detailed characteristics of flow and heat transfer
region. Saviharju et al. (2004) modeled the superheater region on the flue gas side are not obtained as a part of the solution. In
in two recovery boilers and reported full-scale measurements addition, as has been highlighted by Schuhbauer et al. (2014),
for model validation. Leppänen et al. (2014a) modeled the flow the model does not correctly take into account the thermal ra-
field in the superheater region and developed new deposition diation interaction between the flue gas and the porous media,
models. Pérez et al. (2016) developed a detailed CFD model since the tube surfaces are not physically present in the model.
for predicting deposition rates and shapes. Deposition has been The three-dimensional slice model, also known as the slice
the focus of other researchers as well, for example, Jokiniemi superheater region (SSR) model, is another method which has
et al. (1996), Pyykönen and Jokiniemi (2003), Wessel and Bax- been used, in particular, in recovery boiler simulations (Savi-
ter (2003), and Weber et al. (2013). Our recent, ongoing work harju et al., 2004; Leppänen et al., 2013, 2014a,b; Leppänen
has also focused on better understanding of the superheater re- and Välimäki, 2016; Maakala et al., 2015, 2016). This ap-
gion. In Maakala et al. (2015), we utilized CFD modeling, op- proach takes advantage of the repeating pattern of the super-
timization, and surrogate modeling to optimize heat transfer to heater platens and considers only a thin slice of the full three-
the superheaters. dimensional geometry. This simplification can be considered
Modeling of the heat transfer sections of boilers is challeng- valid for the central region of the heat transfer section when:
ing due to the very large dimensions and complex geometries. 1) the effect of the side walls to the flow field can be consid-
One of the standard modeling approaches is the distributed ered minor, and 2) the flow field is even and not swirling at
resistance method, also known as the porous media method, the furnace outlet. These conditions typically hold reasonably
which was originally developed for the modeling of heat ex- well in high-capacity recovery boilers, since they are physically
changers (Patankar and Spalding, 1974). In fact, this approach wide and the combustion air is injected in a symmetric fashion
has been widely utilized for simulating various types of heat ex- from the front and rear walls, and a significant distance before
3
the furnace outlet. The three-dimensional slice model is ade- values for the flow field at the nose level, when the flue gas en-
quate for a multitude of purposes but it does not fully resolve ters the superheater region. The domain outlet is located well
the complex three-dimensional flow structures, which are also after the boiler bank region to ensure that the outlet boundary
connected to other processes, such as heat transfer, deposition, condition will not affect the solution in the superheater region.
and fouling. The boiler walls, rear wall screen, and boiler bank are so called
Based on the presented literature review, we identify the fol- boiling surfaces, which are used for boiling water into steam.
lowing research gaps: 1) To the best of our knowledge, all Because of the phase change of water, temperature inside these
of the previously published recovery boiler superheater region surfaces is assumed to be constant. The superheaters 1A, 1B,
CFD models have considered only a thin slice of the full ge- 2, 3, and 4 are used for heating high-pressure steam to a high
ometry or utilized the distributed resistance approach. A full temperature. Thus, temperature of the steam flowing inside the
three-dimensional model has not been developed and full three- superheaters depends on the heat transfer rate from the flue gas.
dimensional results have not been published. 2) The amount The boundary conditions are described in more detail in the
of experimental data, especially from full-scale measurements, Subsection 2.3.
available for recovery boiler model validation is small. Con- Figure 2b shows an isometric view of the FSR model. Each
sequently, the amount of published validation studies is also heat transfer surface consists of a row of platens (in the z-
small. 3) Understanding of the three-dimensional physical phe- direction). In reality, each platen consists of a number of tightly
nomena in the superheater region is limited. Because a full spaced tubes but in this work the geometry is simplified such
three-dimensional model has not been available, the complex that the individual tubes are not considered. Because the geom-
three-dimensional character of the flow field is not known. In etry is fully symmetric in the z-direction a symmetry boundary
addition, it is not well-understood how the three-dimensional condition is used in the middle of the boiler.
flow field phenomena affect processes such as heat transfer, de- Figure 2c shows the slice superheater region (SSR) model
position, and fouling. which has been used in previous research (Saviharju et al.,
The purpose of this work is to improve the understanding 2004; Leppänen et al., 2013, 2014a,b; Leppänen and Välimäki,
of the complex three-dimensional physical phenomena in the 2016; Maakala et al., 2015, 2016). In the SSR model, only a
recovery boiler superheater region. We present and validate a single gap between two superheater platens is modeled in the
new CFD model which includes the full superheater region ge- z-direction by utilizing symmetry boundary conditions on both
ometry and many state-of-the-art sub-models for describing the sides. This is a reasonable approach for obtaining the solu-
flow field, heat transfer, and special characteristics related to tion in the central part of the boiler but no information is ob-
the recovery boiler process. Specifically, the main objectives of tained about three-dimensional flow phenomena or variation of
this work are: the fields in the z-direction. The present FSR model is thus a
significant improvement compared to the SSR model.
1. Presenting a detailed, three-dimensional CFD model for
In addition to the extended domain, the FSR model includes
the recovery boiler superheater region, called the full su-
many state-of-the-art sub-models for describing the flow field,
perheater region (FSR) model.
heat transfer, and special characteristics related to the recovery
2. Reporting two sets of previously unpublished full-scale boiler process, which have not been utilized in the previous su-
temperature and flow field measurements from the recov- perheater region models. These sub-models are described in the
ery boiler superheater region. The data from these chal- following subsections.
lenging measurements is very valuable, since the amount
of reported experimental data from recovery boilers is
small. 2.2. Governing Equations and Modeling
3. Verifying the results of the FSR model computationally The full superheater region (FSR) model solves the gov-
and validating them with the experimental data. erning equations for mass continuity, momentum, turbulence,
4. Showing the added-value of the FSR model by a detailed species, energy, and radiation. The equations are solved in the
analysis of the three-dimensional flow field and heat trans- steady-state, Reynolds-averaged form and incompressible flow
fer results and by discussing the implications of the three- is assumed.
dimensional solution for the estimation of fouling. The mass continuity equation is

∂ρu j
2. Full Superheater Region CFD Model =0 (1)
∂x j
2.1. General Description of the Model where ρ is the density and u j are the components of velocity
Figure 2a shows the domain of the full superheater region (tensor notation).
(FSR) model. The part of the domain which is marked as the The steady-state Navier-Stokes momentum equations are
superheater region is the target of the accurate solution. It ex-
tends from the nose level to the beginning of the boiler bank ∂ρu j ui ∂ ∂ui ∂u j ∂p 2 ∂ρk
= (µeff ( + )) − −
area. The domain inlet is located well below the nose level for ∂x j ∂x j ∂x j ∂xi ∂xi 3 ∂xi (2)
computational reasons, that is, for obtaining realistic, developed + ρgi + fi for i = 1, 2, 3
4
Figure 2: a) A side view of the domain of the full superheater region (FSR) model. Superheater is abbreviated as SH. b) An isometric view of the domain of the
FSR model. The boiler walls are transparent to show the construction of the heat transfer surfaces (superheaters, rear wall screen, and boiler bank). c) The domain
of the slice superheater region (SSR) model that has been used in previous research. A single gap between two superheater platens is modeled.

where µeff is the effective viscosity, p is the pressure, k is the tur- Even though species sources from chemical reactions are not
bulent kinetic energy, gi is the gravitational acceleration in the included in this work, the species transport equations are in-
xi -direction, and fi are the other body forces in the xi -direction. cluded herein since they are considered to be a feature of the
The momentum equations are closed using the k-ω SST tur- model. Therefore, additional species and chemical reactions
bulence model (Menter, 1994). A pressure-based solver is used can be readily included in future work. The additional compu-
and the pressure-velocity coupling is implemented with the seg- tational cost associated with solving the scalar species transport
regated SIMPLE scheme (Patankar and Spalding, 1972). equations is considered minor.
The transport equations for the N species are
∂ρu j Yn ∂ ∂Yn Table 1: The heat capacity functions used for the species. The functions are
= (ρDn,eff )+Rn for n = 1, 2, ..., N species (3) piecewise-polynomial with the form c p (T ) = α0 + α1 T + α2 T 2 + α3 T 3 + α4 T 4 .
∂x j ∂x j ∂x j
Species α0 α1 α2 α3 α4
where Yn is the mass fraction of the species n. Dn,eff is the
effective species diffusivity and Rn is the species source from Range: 300 ≤ T < 1000 [K]
chemical reactions. The transport equations are solved for the H2 O 1.563e+3 1.604e+0 −2.933e−3 3.216e−6 −1.157e−9
gaseous water (H2 O), oxygen (O2 ), and carbon dioxide (CO2 ). O2 8.348e+2 2.930e−1 −1.496e−4 3.414e−7 −2.278e−10
The mass fraction of nitrogen (N2 ) is found by subtracting the CO2 4.299e+2 1.874e+0 −1.966e−3 1.297e−6 −4.000e−10
mass fractions of the other species from one. The heat capacity, N2 9.790e+2 4.180e−1 −1.176e−3 1.674e−6 −7.256e−10
thermal conductivity, and viscosity of the individual species are Range: 1000 ≤ T < 5000 [K]
functions of temperature (see Table 1). Mass-weighted species H2 O 1.233e+3 1.411e+0 −4.029e−4 5.543e−8 −2.950e−12
O2 9.608e+2 1.594e−1 −3.271e−5 4.613e−9 −2.953e−13
mixing laws are used for obtaining the local flue gas proper-
CO2 8.414e+2 5.932e−1 −2.415e−4 4.523e−8 −3.153e−12
ties. The density of the flue gas is assumed to be a function
N2 8.686e+2 4.416e−1 −1.687e−4 2.997e−8 −2.004e−12
of temperature by the ideal gas law. In the present work, com-
bustion processes are assumed to be completed when the gas
enters the superheater region, thus species sources from chem-
ical reactions are not included. This assumption has been con- The particle phase in the recovery boiler superheater region
firmed in multiple measurement campaigns (see, e.g., Saviharju can be considered to consist of carryover (combusted black
et al. (2004) for experimental data from four recovery boilers). liquor droplets) and of fume (condensed sodium and potassium
5
salts). In this work, the average carryover concentration was
measured to be 1.1 g/Nm3 , which is in line with values reported
in Vakkilainen (2005). The fume concentration is approximated
to be 15.6 g/Nm3 , according to the work of Mikkanen et al.
(1999). Based on these values, the total volume fraction of the
particles is approximately 1.4 × 10−6 and the particle loading
ratio is approximately 1.3 × 10−2 (see Elghobashi (1994) and
Di Giacinto et al. (1982) for details). Therefore, the effect of
the particles on the flow field is considered minor and it is not
included in the model. Since particle-based erosion is typically
minor in recovery boilers (Vakkilainen, 2005), a one-way cou-
pled solution of the particle tracks is not included in this work.
The energy equation is

∂ρu j h ∂ ∂T X
N
= (λeff − hn Jn, j ) + S rad (4)
∂x j ∂x j ∂x j n=1

where h is the sensible enthalpy, λeff is the effective thermal

conductivity, T is the temperature, hn is the sensible enthalpy
of the species n, Jn, j is the diffusion flux of the species n in the
x j -direction, and S rad is the radiation source term. The term
S rad is solved from the radiative transfer equation using the dis-
crete ordinates method (Raithby and Chui, 1990). The non-gray
weighted sum of gray gases method with five wavelength bands
is used for the local flue gas radiation properties (Dorigon et al.,
2013). Figure 3: The final computational grid from the side. The cell sizes are shown
The effect of the recovery boiler fume on radiation is taken in the coordinate directions. The detailed view shows the mesh between the
superheater platens. For visualization purposes, the resolution shown between
into account by assuming that in the superheater region the the platens (z-direction) is only one-fourth (25 mm) of the final grid resolution
fume consists mainly of sodium sulfate (Na2 SO4 ) particles with (6.25 mm). The refinement region covers the superheater region (see Figure 2a),
a diameter of 1 µm (McKeough and Janka, 2001). As was men- which is the area of interest in this work.
tioned before, the fume particle concentration is approximated
to be 15.6 g/Nm3 in the flue gas (Mikkanen et al., 1999). The
contribution of the fume on the scattering coefficient is calcu-
lated with a model based on the work by Wessel et al. (2000). and because no detailed information about the inlet profiles is
The effect of the fume on the absorption coefficient is neglected available, the inlet boundary conditions are set using constant
because it is considered to be minor compared to the effect of profiles. In accordance with this approach, the inlet of the com-
the flue gas itself (Wessel et al., 2000). putational domain is located well below the nose level, for two
The equations are solved using a commercial CFD solver and reasons: 1) for obtaining a realistic, developed flow field for
the sub-models are programmed as user-defined functions. The the flue gas at the nose level and 2) for reducing the sensitivity
computational grids used are block-structured and comprised of the solution to the uncertainties in the selection of the inlet
entirely of hexahedral cells (Figure 3). Local mesh refinements profiles. We consider the present approach acceptable for the
are handled via the standard 2:1 cell splitting approach. The purposes of this work but acknowledge that more detailed in-
ICEM CFD program is used for grid generation. Details of the formation regarding the inlet profiles could have increased the
numerical grids are reported in Subsection 4.1 together with the accuracy of the solution.
results of the model verification. The values of the boundary conditions are mainly based on
the experimental data obtained from the measurement cam-
2.3. Boundary Conditions paign, which include the boiler operating parameters (e.g., fuel
As was mentioned previously, the goal of the present work is flow and combustion air flow) and the measured values (e.g.,
an accurate solution in the superheater region (see Figure 2a). fuel composition, flue gas composition, and steam flow and
It is assumed that the profiles at the nose level, when the flue temperature before and after each superheater). Two sets of
gas enters the superheater region, are mainly defined by the measurements, A and B, were taken on the boiler on different
geometry and are not any more significantly affected by the days. Accordingly, the boundary conditions corresponding to
combustion processes and air injection schemes in the furnace. each set are somewhat different.
This typically holds reasonably well in modern recovery boil- Table 2 summarizes the values of the boundary conditions
ers, since the combustion air is injected in a symmetric, non- at the model inlet. The mass flow boundary condition is uti-
swirling fashion from the front and rear walls, and a significant lized for the momentum equations. The mass flow rate at the
distance before the furnace outlet. According to these reasons, inlet was obtained from a balance calculation using the mea-
6
sured fuel and air mass flows. The calculated value was con-
firmed with a measured value collected from the boiler control
system. The inlet velocity given in the table was calculated
from the mass flow rate. The inlet temperature was calculated
using CFD such that the measured temperature was approxi-
mately obtained at the measurement location closest to the inlet
(point 5 in Figure 5). The turbulence intensity at the inlet was
estimated to be 15%, which corresponds to a relatively high tur-
bulence level, as indicated by previous simulations (Maakala
and Miikkulainen, 2015). In addition, preliminary CFD sim-
ulations indicated that the value of the turbulence intensity at
the nose level is not very sensitive to the precise value of the
turbulence intensity at the domain inlet. The flue gas composi-
tion was obtained in a similar fashion as the mass flow rate, that
is, calculated from a balance and confirmed with the measured
values.

Table 2: The values of the boundary conditions at the model inlet in the mea-
surement sets A and B. The length scale of the Reynolds number (Re) at the
inlet is the hydraulic diameter of the furnace cross section. The Reynolds num- Figure 4: a) A schematic of the boundary between the flue gas side and the
ber in the superheater region is given for reference, where the length scale is water/steam side. The T int and kint , which are given as boundary conditions, are
the spacing between the superheater platens. circled. The red line shows the temperature profile and the dashed green line
indicates the average deposit thickness. b) The water/steam side temperature
Set A Set B Source T int on the heat transfer surfaces. On the boiling surfaces, the T int is constant.
Velocity 4.6 m/s 4.7 m/s Balance, measurements On the superheater surfaces, the T int is interpolated in the x-direction using their
steam inlet and outlet temperatures, and is constant in the y- and z-directions.
Temperature 1 340 K 1 410 K Measurements
Turbulence intensity 15% 15% Estimation, previous work
Re, inlet 284 000 266 000 -
Re, superheater region 15 000 14 000 - is the tube thermal conductivity, sdeposit is the deposit thickness,
Flue gas composition and λdeposit is the deposit thermal conductivity.
Carbon dioxide (CO2 ) 19 wt% 19 wt% Balance, measurements The boundary conditions on the heat transfer surfaces are
Gaseous water (H2 O) 15 wt% 14 wt% Balance, measurements summarized in Table 3 and visualized in Figure 4b. The tem-
Oxygen (O2 ) 3 wt% 4 wt% Balance, measurements peratures, pressures, and steam flow rates before and after each
Nitrogen (N2 ) 63 wt% 64 wt% Balance, measurements superheater were collected from the boiler control system dur-
ing the measurement campaign. Therefore, the average heat
transfer rates to the superheaters could be calculated. The mea-
The thermal boundary conditions are set using the standard sured temperature values were directly utilized to set the T int
convective boundary condition, where the water/steam side boundary conditions. The average heat transfer rates were used
temperature T int (internal temperature) and the water/steam side for setting the kint boundary conditions, such that the kint values
heat transfer coefficient kint (internal heat transfer coefficient) were adjusted during the CFD simulations until the calculated
are given as boundary conditions. The surface heat flux q and average heat transfer rates were obtained. Similar approaches
the surface temperature T surface are calculated as a part of the have been utilized in previous research (Saviharju et al., 2004;
CFD solution from the following system of equations Leppänen et al., 2014a; Schuhbauer et al., 2014). Table 3
also shows the estimated sdeposit values, which have been cal-
q = kgas (T surface − T gas ) + qrad culated from the kint values using the Equation (6). The value
(5)
= kint (T int − T surface ) of λdeposit = 1.5 W/(mK) has been used in the estimations, fol-
lowing Schuhbauer et al. (2014) and Leppänen et al. (2014a).
where kgas is the convective heat transfer coefficient on the flue The sdeposit values were not directly used in the simulations.
gas side, T gas is the flue gas temperature, and qrad is the radiation The present approach can be understood by considering the
heat flux. The schematic of the boundary in Figure 4a clarifies challenges associated with direct, a priori estimation of the kint
how the variables are defined. from the Equation (6). In the present work, the kfluid , stube ,
The significance of the water/steam side heat transfer coeffi- and λtube are well known or can be readily estimated. How-
cient kint is seen from the equation (see also Figure 4a) ever, the sdeposit and λdeposit are very hard to estimate and, based
1 stube sdeposit
! on the measured values, they constitute over 95% of the to-
kint = 1/ + + (6) tal thermal resistance. In the literature, typical reported values
kfluid λtube λdeposit
for the sdeposit are between 5–60 mm (Adams, 1997; Li et al.,
where kfluid is the heat transfer coefficient from the inner surface 2013; Leppänen et al., 2014a) and for the λdeposit between 0.1–
of the tube to the water/steam, stube is the tube thickness, λtube 2.5 W/(mK) (Rezaei et al., 2000; Baxter et al., 2001; Zbogar
7
Table 3: The boundary conditions on the heat transfer surfaces in the measurement sets A and B. The T int values were measured and the kint values were calculated
from the measured values such that the average heat transfer rates observed during the measurement campaign were satisfied. The simulations converged well to
the tabulated kint values. The sdeposit values (not directly used in the simulations) were estimated from the kint values using the Equation (6).
Set A Set A Set A Set B Set B Set B
T int k sdeposit T int k sdeposit
h int i h int i
W W
[K] m2 K
[mm] [K] m2 K
[mm]

Walls and roof 598 58 25 597 58 25

Rear wall screen 598 140 10 597 83 17
Boiler bank 598 183 8 597 170 8
Superheater 1A 598–616 111 13 598–615 58 25
Superheater 1B 616–644 52 28 615–642 35 42
Superheater 2 638–690 43 34 639–689 30 48
Superheater 3 682–753 53 27 686–751 33 45
Superheater 4 736–788 48 30 741–788 28 53

et al., 2005). Depending on the time after the boiler startup,

Table 4: The main operating parameters of the boiler. The values are reported
ash composition, sootblowing, and the location of the particu- according to the conventions given in Vakkilainen (2005), where additional de-
lar heat transfer surface, the sdeposit can vary from values close tails can be found. The following abbreviations are used: ds (dry solids), tds/d
to zero (clean tubes) to values that completely plug the flue gas (tonnes of dry solids per day), HHV (higher heating value), and NHV (net heat-
ing value).
passages. On the other hand, the λdeposit is affected by the par-
ticle size distribution, porosity, and sintering degree of the de- Parameter Value
posit. Therefore, it is evident that: 1) the heat transfer coef- Boiler type Kraft recovery boiler (subcritical)
ficient kint is highly sensitive, in particular, to the sdeposit and Fuel type Softwood black liquor
λdeposit and 2) the a priori estimation of the sdeposit and λdeposit Black liquor capacity, tds/d 2 400
is very challenging due to the complex nature of the recovery Black liquor HHV, MJ/(kgds) 13.0
boiler process. Black liquor dry solids content, wt% 72
The approach of calculating the kint through the measured av- Thermal capacity, MW 360
erage heat transfer rates is further justified in the present work, Air ratio (λ) 1.13
since the goal is the validation of the CFD model, submodels, Main steam flow, kg/s 92
and the flow and temperature field solutions in typical operat- Main steam temperature, K 788
ing conditions of the recovery boiler. The essential feature, and Main steam pressure, bar 110
an advantage, of this method is that the measured values (tem- Boiler efficiency (NHV), % 87
peratures and heat transfer rates) from the water/steam side can Reduction efficiency, % 96
be utilized as boundary conditions. Therefore, the surface tem-
perature and heat flux profiles are calculated as a part of the
CFD simulation together with the three-dimensional flow and
temperature field solutions. pulp mill. Hence, the number of measurement locations and re-
peated measurements normally have to be limited in full-scale
3. Measurements measurement campaigns. Furthermore, the locations available
for measurements are limited by the number of openings exist-
The measurement campaign was conducted on a modern, ing in the boiler walls. On the other hand, the flow and tem-
medium-capacity recovery boiler. The main operating param- perature fields are always time-dependent even in steady boiler
eters of the boiler are shown in Table 4. In this work, we re- operation, and thus each pointwise recording should be mea-
port two sets of measurements, called sets A and B, which were sured multiple times to obtain a reliable time-averaged value.
taken at the boiler on different days. In the measurement campaign, the flow and temperature
Full-scale measurements on a recovery boiler are time con- fields were measured from the openings in the boiler side walls
suming and expensive because of the large number of personnel (Figure 5). In the x- and y-directions, the number of measure-
required for the measurements, sample taking, and operating ment locations was limited by the number of openings avail-
the boiler in a steady fashion. The high-temperature and dusty able. In the z-direction, all of the measurements were carried
conditions inside the boiler are especially challenging. Addi- out at the same depth of approximately 1.7 m. This was mainly
tional time is required every time the boiler operation mode or because of the limited physical dimensions of the probes, but
load is changed because it takes several hours for the recovery also because of the previously mentioned time constraints. Due
boiler operation to reach a steady state after every change. The to the challenging conditions and limited time available for the
total time available for measurements is often limited by the measurements, it was possible to repeat each measurement only
8
from one to three times at each measurement location. The the convergence criteria for reaching a steady state had to be
photos taken during the measurement campaign illustrate the somewhat relaxed around the corner vortex region. The results
challenging conditions in the recovery boiler superheater region indicated that grid convergence was achieved in the x- and y-
(Figure 5). directions with a resolution of 100 mm. The GCI error esti-
The flue gas temperatures were measured using a suction py- mates were 1.9% in temperature, 8.9% in velocity magnitude,
rometer (Figure 6a). The suction pyrometer draws the flue gas and 3◦ in velocity direction.
inside the ceramic sleeve, where a thermocouple for measuring Next, grid convergence is assessed by further grid refinement
the temperature is located. The purpose of the ceramic sleeve in the z-direction. This is important because a fine grid resolu-
is to shield the thermocouple from direct thermal radiation heat tion in the z-direction is essential for accurately capturing the
transfer, which would introduce bias to the measurements. flow structures inside the relatively small gaps between the su-
The velocity measurements were done using a special two- perheater platens. Table 5 summarizes the computational grids
chamber Pitot probe (Figure 6b). The Pitot probe is air cooled that are used in the verification.
and designed for high-dust conditions. The probe is rotated
manually around its axis in the measurement location to find
Table 5: The computational grids used in the verification. The cell sizes are
the orientation which shows the largest pressure difference be- given in the refinement region (see Figure 3). The representative cell sizes
tween the two chambers. The flow direction is obtained from and refinement ratios were calculated according to Celik et al. (2008). The
the orientation and the velocity magnitude can be calculated first cell y+ values and the errors in platen heat transfer were obtained by two-
from the pressure difference. Because of the nonstandard shape dimensional simulations of flow between two superheater platens. The errors
in platen heat transfer are relative to a dense grid solution (64 cells between
of the two-chamber Pitot probe, the velocity calculation for- platens). The values of equiangle skewness and orthogonal quality range from
mulas have been calibrated with measurements done using a 0 (degenerate) to 1 (excellent).
standard Pitot probe.
The construction of the Pitot probe (Figure 6b) effectively fil- Grid 1 Grid 2 Grid 3
ters out the velocity vector component parallel to the shaft of the Total cells 4.8 M 9.0 M 17.4 M
probe, i.e., z-direction. Because of this, we are assuming that Cell x-size 100.00 mm 100.00 mm 100.00 mm
the probe measures the xy-velocity direction and xy-velocity Cell y-size 100.00 mm 100.00 mm 100.00 mm
magnitude. The effect of this assumption was later confirmed Cell z-size 25.00 mm 12.50 mm 6.25 mm
by the modeled results to be minor because the z-component of Representative cell size, h 60.09 mm 47.70 mm 37.86 mm
the velocity was relatively small in the measurement locations. Refinement ratio, r - 1.26 1.26
In addition to the temperature and flow field measurements, Cells between superheater platens 8 16 32
samples of black liquor, auxiliary fuel, ash, carryover, smelt, First cell y+ between platens 118.8 65.9 44.5
and circulation water were collected during the measurement Error in platen heat transfer 1.7% 1.5% 0.5%
campaign. The flue gas composition was measured and the Equiangle skewness, mean / min 0.85 / 0.32 0.84 / 0.31 0.84 / 0.31
boiler operation data was collected from the control system. Orthogonal quality, mean / min 0.93 / 0.49 0.92 / 0.48 0.92 / 0.48

4. Results
The grid resolution in the z-direction was further verified by
4.1. Model Verification two-dimensional simulations of flow between two superheater
The accuracy of the computational solution is assessed with platens (see Table 5). The first cell y+ value is in the appropri-
a grid convergence study. The uncertainty due to discretization ate range of 30 < y+ < 200 with all of the grids. The error in
is reported using the standard grid convergence index (GCI) platen heat transfer is also reasonably low with all of the grids,
method (Celik et al., 2008). between 1.7%–0.5%, with the resolution of 32 cells providing
Because of the large size of the modeled domain it was not the lowest error value of 0.5%. The velocity and temperature
possible to study grid convergence with the available comput- profiles between the platens are adequately captured when the
ers by refining the grid simultaneously in all three coordinate resolution is between 16–32 cells. Thus, based on the two-
directions. Thus, grid convergence was studied by first refining dimensional simulations, we consider that the flow field and
the grid in the x- and y-directions and then in the z-direction. heat transfer between the superheater platens can be accurately
The results concerning the grid refinement in the x- and y- described with the resolutions of the grids 2 and 3.
directions were reported in our previous work (Maakala et al., Next, the grid convergence is studied using the full three-
2016). The most significant observation was that the steady- dimensional model. Table 5 shows the equiangle skewness (ES)
state RANS solution does not achieve perfect iterative conver- and orthogonal quality (OQ) values for the three-dimensional
gence, which is typical in such high Reynolds number flows grids. The mean values of both ES and OQ are high and their
in complex geometries (see Grace (1995) for details). Iterative minimum values are well above the typically accepted mini-
convergence was satisfactory in the majority of the superheater mum values of 0.05 (ES) and 0.01 (OQ). Therefore, according
region, except in the vortex region in the corner of the front to these measures, the numerical grids are of high quality, espe-
wall and the roof (near the points 1, 2, and 3, see Figure 5), cially considering the present relatively complex geometry.
where minor transient behavior remained in the solution. Thus, Figure 7 shows the solved temperature profiles along the
9
Figure 5: The measurement points and lines which are used for profiles when comparing modeled results to the measurements. The line 1 is drawn from between
the points 1 and 2 because they are not exactly aligned in x-direction. I) View from the opening in point 1. Lightly fouled platens of the superheater 2 can be seen.
II) A cooled carryover probe that has been held inside the boiler for approximately five minutes. White fume deposition can be seen on the surface. III) View from
the opening in point 7. Localized deposition can be seen in the central region of the superheater 4. IV) View from the opening in point 11. Deposition can be seen
on the platens of the superheater 4.

the grid 3 are 0.6% in temperature, 3.8% in velocity magnitude,

and 2◦ in velocity direction. We consider the error estimates
reasonably low and the computational solution reliable. The re-
sults in the rest of this study are reported using the finest grid 3
offering the lowest GCI error level.

Table 6: The values calculated for the discretization error on the grid 3 using
the GCI method. The values are shown for temperature (T ), velocity magni-
tude (~u mag), and velocity direction (~u dir). The GCIavg is the estimate for the
average discretization error.
T ~u mag ~u dir

Apparent order of the method, p 5.03 5.29 4.84

Average relative error, eavg 1.1% 7.2% 3◦
Average grid convergence index, GCIavg 0.6% 3.8% 2◦

Figure 6: a) The suction pyrometer used for the temperature measurements. b)

The two-chamber Pitot probe used for the velocity measurements.
4.2. Model Validation
The accuracy of the model is next validated with the exper-
imental data. During the measurement set A, the boiler was
lines 1–6, which cover the whole superheater region and the operating at 87% heat load of the full capacity. The velocity
majority of the measurement locations. As can be seen from measurements were repeated from one to three times at each
Figure 7, effect of the grid refinement is minor in the solved measurement location while the temperatures were measured
profiles. Apparent differences can be seen only along the lines 1 only once. During the measurement set B, the boiler was op-
and 2, which are close to the corner vortex region. The er- erating at 84% heat load of the full capacity. Only temperature
ror bounds calculated for the dense grid solution using the GCI was measured at each location but the measurements were re-
method are small in the majority of the superheater region. peated twice.
Table 6 summarizes the values calculated for the discretiza- The confidence intervals for the measured values were cal-
tion error using the GCI method. The GCI error estimates on culated using the t-distribution with a 95% confidence level.
10
 measure and to model. The difference at the point 19 could also
be explained by a corner vortex, which might exists in reality
but is not captured by the model. Another possible explanation
is localized fouling, which is not included in the model but is
quite common in practice in the boiler bank region. At most of
the measurement locations, the relative difference is clearly be-
low 15% (colored green). Thus, we consider the overall match
between the modeled results and the measurements to be rela-
tively good.
In velocity direction (Figure 8), the average absolute differ-
ence between the modeled and measured values is 15◦ . The
largest differences (above 30◦ , colored red) are observed at the
points 4 (51◦ ) and 1 (37◦ ). However, the modeled values are
clearly inside the confidence intervals of the measurements at
both locations. The point 1 is also in the challenging corner
vortex region. We consider that the solved velocity direction
values correspond well to the measurements.
Figure 9 shows the solved temperature field. The solved and
measured values of temperature are shown at each measurement
location. The average absolute difference between the modeled
and measured values is 30 K or 3%. The largest differences
(above 5%, colored red) are seen at the points 20 (8%), 2 (7%),
and 6 (5%). The difference at the point 20 can be possibly ex-
plained by localized fouling in the boiler bank region. The dif-
ference at the point 2 can be explained by the corner vortex re-
gion. Some of the differences can also result from uncertainties
in the measurements. Since the temperatures were measured
only once at each location, it is possible that all of the mea-
sured values are not representative of time-averaged boiler op-
eration. Despite the various uncertainties, we consider that the
Figure 7: The temperature profiles along the lines 1–6 (see Figure 5) solved
using the grids 1–3. The GCI error bounds are shown in gray. The profiles measured and modeled values of temperature correspond well
of velocity magnitude and velocity direction show similar results and are thus to each other.
omitted for brevity. Figure 10 shows the solved and measured values of velocity
magnitude (a), velocity direction (b), and temperature (c) along
the lines 1–6. The velocity magnitude profiles correspond very
well to the measurements on the lines 2, 4, 5, and 6. There are
Even though the measured mean values are considered rep-
only relatively small differences on the lines 1 and 3. The ve-
resentative, the large confidence intervals calculated with the
locity direction profiles are inside the confidence intervals of the
rather strict 95% confidence level demonstrate the experimental
measurements at each of the points. The solved values are also
uncertainty associated with these challenging measurements.
close to the measured mean values. The confidence intervals are
Therefore, a perfect correspondence between the measured and
wide since the direction measurements were challenging due to
modeled values cannot be expected.
the fluctuating flow field in the recovery boiler. In the tempera-
ture results, there are apparent differences between the modeled
4.2.1. Validation Against the Measurement Set A and measured values only at some individual points, most no-
Figure 8 presents the solved velocity field by streamlines. tably, on the lines 1 and 3. On average, the solved profiles of
The solved and measured values of velocity magnitude and velocity magnitude, direction, and temperature correspond well
direction are shown at each measurement location. The aver- to the measured values. It is significant that the solved profiles
age absolute difference between the modeled and measured ve- clearly follow the measured profiles.
locity magnitudes is 0.9 m/s or 20%. The largest differences
(above 30%, colored red) are observed at the points 11 (61%), 4.2.2. Validation Against the Measurement Set B
12 (58%), 1 (43%), 19 (39%), and 7 (34%). We do not consider Figure 11 shows the simulated temperature field while Fig-
the differences at the points 7, 11, and 12 to be very significant ure 12 shows the solved profiles along the lines 1–6. The sim-
because of the wide confidence intervals of the measurements ulated and measured values are shown at each measurement lo-
at these locations. In addition, at these locations the modeled cation. In this set, the measured mean values calculated with
values are inside, or close to, the confidence intervals of the two measurements are more reliable than the values from sin-
measurements. The difference at the point 1 can be explained gle measurements in the set A. However, the confidence inter-
by the corner vortex region, which is very challenging both to vals, calculated with the 95% confidence level, are very large
11
Figure 8: The solved velocity field shown by streamlines on the xy-surface corresponding to the measurement locations. The solved (not in parentheses) and
measured (in parentheses) velocity magnitude and velocity direction values are shown at each measurement location. The measurement locations have been colored
by the relative difference of the modeled and measured values. At each location, the upper color corresponds to velocity magnitude and the lower color to velocity
direction. The measured values are from the set A.

because of only two repeated measurements. The average con- ure 13, which shows the modeled values in comparison to the
fidence interval is ±311 K and the average absolute difference measured values in both sets, A and B. The overall match be-
between the repeated measurements is 49 K. tween the modeled and measured values is considered good.
The average absolute difference between the modeled and
measured temperature values is 35 K or 4% (see Figure 11). 4.3. Analysis of the Three-dimensional Simulation Results
The largest differences (above 5%, colored red) are at the The added-value of the present CFD model is next illustrated
points 3 (12%), 18 (10%), 14 (7%), and 12 (6%). Again, the with a detailed analysis of the three-dimensional simulation re-
difference at the point 3 can be explained by the corner vortex sults. The focus is on new results which have not been reported
region, where both the numerical solution and measurements in previous modeling or experimental studies along with find-
are challenging. It is unclear whether the other differences re- ings that the simplified models, which were discussed in the
sult from experimental or modeling error since the measure- literature review, cannot adequately capture. The results in this
ment uncertainties are still reasonably large. Regardless, the section correspond to the CFD simulation of the measurement
differences are small (below 3%, colored green) in most of the set A.
measurement locations.
The solved temperature profiles correspond very well to the 4.3.1. Vortex Structures and Overall Character of Heat Trans-
measurements on the lines 1, 3, and 4 (Figure 12). On the other fer
lines, there are apparent differences only at individual points. The three-dimensional character of the flow field is demon-
The profiles are also inside the confidence intervals of the mea- strated in Figure 14a, which shows the vortex structures iden-
surements and they follow the shapes outlined by the measured tified via the standard Q-criterion method. The vortex in the
mean values. We consider that the measured and modeled val- corner of the front wall and the roof is clearly visible (I). In
ues correspond well to one another also in this measurement addition, there is a symmetric vortex pair in the front cavity re-
set. gion (II) and complex vortex structures below the boiler nose
The results of the model validation are summarized in Fig- (III). When the flow enters between the superheater platens, the
12
Figure 9: The solved temperature field shown by contours on the xy-surface corresponding to the measurement locations. The solved (not in parentheses) and
measured (in parentheses) values are shown at each measurement location. The measurement locations have been colored by the relative difference of the modeled
and measured values. The point 5 is the inlet temperature reference location. The measured values are from the set A.

large-scale vortex structures are effectively filtered out and thus most likely a major portion of the radiation heat flux is due
small-scale vortices are seen throughout this region. to close range radiation from the flue gas flowing through
Figures 14b and 14c show the simulated heat flux solution the superheaters. This consideration is also supported by the
on the walls, boiling surfaces, and superheaters. The vortex two-dimensional simulations of flow between two superheater
structures shown in Figure 14a affect the heat flux solution, platens (see Subsection 4.1), where the contribution of the ra-
particularly on the boiler walls. This indicates that the three- diation heat transfer was significant. The ratio of the radiation
dimensional flow is closely connected to the heat transfer. The heat flux to the total heat flux is highest for the superheater 2
heat flux to the furnace walls is clearly greater in the central (0.70) and superheater 3 (0.66), which are directly exposed to
region of the walls than close to the corners (IV). In the super- thermal radiation from regions below the nose.
heater region, the heat flux to the furnace walls reduces sharply Figure 16 shows the ratio of the radiation heat flux to the total
when the flow enters between the superheater platens (V). The heat flux in detail on the boiler walls (a) and on the superheaters
heat flux is highest to the superheaters which are directly ex- and boiling surfaces (b). The ratio varies locally from the values
posed to thermal radiation heat transfer from the furnace, that on the boiler walls which are close to 1.0 (I) to the values on the
is, from below the nose (VI). However, there is substantial lo- superheater 4 which are close to 0.0 (II). The local variations in
cal variation in the heat flux to the superheaters and the boiling the gas and surface temperatures and the shadowing effects of
surfaces because of the effect of the flow field (VII). These re- the geometry affect these values significantly.
sults indicate that the heat transfer solution is strongly affected
Since the share of the radiation heat transfer is large in the
by the geometry, flow field, and radiation heat transfer effects.
whole superheater region, its accurate modeling is considered
highly important for the overall accuracy of the heat transfer
4.3.2. Radiation and Convection Heat Transfer solution. For an accurate solution of the radiation heat transfer,
The average ratios of the radiation heat flux to the total the full three-dimensional representation of the superheaters is
heat flux on the superheater surfaces are shown in Figure 15a. considered important. This is due to the global character of the
The ratio is substantial for all of the superheaters, between radiation heat transfer process and the shadowing effects of the
0.45–0.70. Considering the layout of the superheater region, geometry.
13
Figure 10: The solved profiles on the lines 1–6 (see Figure 5) compared to the measured values. The error bounds are estimated for the solved profiles using the GCI
method. The horizontal lines show the 95% confidence intervals. The inlet temperature reference location (point 5) is indicated with a blue square. The measured
values are from the set A.

The convection heat transfer is studied by calculating the av- correlations.

erage Nusselt numbers on the superheater surfaces. The Nusselt
number is defined as The Nusselt numbers are calculated from the Equation (7) by
kgas l using the simulated kgas values, following a similar approach as
Nu = (7)
λ in Jang and Yang (1998). The values are compared to a well-
where kgas is the convective heat transfer coefficient of the flue established experimental correlation for tube bundles in cross
gas, l is the characteristic length, and λ is the thermal conduc- flow (Gnielinski, 1975), which was recently utilized for the su-
tivity of the flue gas. The characteristic length is defined as the perheaters and reheaters of a coal-fired boiler by Schuhbauer
streamed length of a single tube, l = πr, where r is the tube et al. (2014). The experimental correlation is considered for
radius, since this is the typical convention used in experimental two conditions: 1) cross flow perpendicular to the tubes and 2)
14
Figure 11: The solved temperature field shown by contours on the xy-surface corresponding to the measurement locations. The solved (not in parentheses) and
measured (in parentheses) temperature values are shown at each measurement location. The measurement locations have been colored by the relative difference of
the modeled and measured values. The point 5 is the inlet temperature reference location. The measured values are from the set B.

cross flow 35◦ oblique to the tubes. Additional details can be 4.3.3. Heat Flux Profiles in the Width Direction
found in the mentioned publications and in VDI-Gesellschaft
(1993). The total heat flux and thermal radiation heat flux profiles of
each superheater are shown in the width direction of the boiler
Figure 15b shows the average Nusselt numbers on the super-
(z-direction) in Figure 17. The simulated average heat flux val-
heater surfaces. The profiles from the CFD simulation and the
ues match the corresponding values observed during the mea-
experimental correlations show a similar trend. Compared to
surement campaign, since the average heat transfer rates were
the Nusselt numbers from the CFD solution, the values from
used for defining the boundary conditions (see Subsection 2.3).
the experimental correlation for perpendicular flow (Exp 1) are
However, the results show that the heat flux profiles differ sig-
on average 18.4% higher and the values from the experimen-
nificantly from the average values. The heat flux is substantially
tal correlation for oblique flow (Exp 2) are on average 13.9%
higher to the platens in the central region than to the platens
lower. It is considered that the match between the experimental
close to the side walls (I). The largest differences are noted at
correlations and the CFD solution is acceptable, since neither
the superheater 4, where the heat flux toward the platens in the
of the assumptions for the flow direction holds exactly in the
central region is over 90% higher compared to the platens near
present geometry. The local variation in the flow direction over
the side walls.
the superheater tubes can be significant (see Figure 8).
The results indicate that with the experimental correlations In Figure 17, there is a clear depression in the heat flux pro-
an average accuracy of roughly 10–20% can be achieved. How- files in the central region of the superheaters 3, 4, 1B, and 1A
ever, applying the correlations in the present geometry is chal- (II). The CFD solution indicates that this phenomenon could
lenging, since it is difficult to estimate the direction of the flow result from a small, lower temperature vortex region which oc-
field a priori or to simulate it accurately without including the curs in the central part of the boiler when the flow entering the
geometry of the superheaters in the model. Therefore, the full superheater region separates from the boiler nose. The vortex
three-dimensional representation of the superheaters is consid- region is not visible in Figure 14a since the view is obstructed
ered important also for an accurate solution of the convective by the superheater platens and the large number of small-scale
heat transfer. vortex structures.
15
 Figure 13: The modeled values in comparison to the measured values. Inside
the dashed error bounds, the match between the modeled and measured values
is considered to be good (colored green in Figures 8, 9, and 11). The values
from the set A are indicated with blue circles and the values from the set B are
indicated with red squares. Confidence intervals are not shown.

range is utilized to illustrate the implications of the three-

Figure 12: The solved profiles of temperature along the lines 1–6 compared to dimensional CFD solution for the estimation of fouling. The
the measured values. The average GCI error estimate (Table 6) is used for the
error bounds of the solved profiles since only the dense grid solution is available potential deposits are assumed to consist of 50 wt% carryover
for this measurement set. The estimated error bounds are so small that they are particles (combusted black liquor droplets) and 50 wt% fume
barely visible. The horizontal lines show the 95% confidence intervals. The particles (condensed sodium and potassium salts). The chemi-
inlet temperature reference location (point 5) is indicated with a blue square.
cal compositions of the carryover and fume were obtained from
The measured values are from the set B.
the samples taken during the measurement campaign. Further-
more, it is assumed that the potential deposits are carried with
the flue gas flow and that they are at the same temperature as the
4.3.4. Implications for the Estimation of Fouling flue gas. The locations where the fouling potential of the ash is
Recovery boiler ash is considered sticky (i.e., fouling) if it is highest are then identified as the regions where the flue gas tem-
between 15% and 70% in the liquid phase (Isaak et al., 1986; perature is in the sticky temperature range. The transport of the
Tran et al., 2002). Below 15% liquid content, the particles are ash particles to the heat transfer surfaces is not modeled. This
too dry and therefore not sticky. Above 70% liquid content, the simple approach is considered only as an indicative worst-case
particles can be sticky, but the deposits stop growing since addi- estimate for the purposes of the present discussion, and it is not
tional material transported to the surfaces becomes molten and intended to be a detailed model for fouling (for detailed models,
flows off. The sticky temperature T 15 (15% liquid content) and see Weber et al. (2013)).
flow temperature T 70 (70% liquid content) are typically used to Figure 18 shows the potential regions of fouling on the fur-
characterize the sticky temperature range of the ash, T 15 –T 70 . nace walls, superheater 2, and superheater 4. The identified
This range can be obtained via thermodynamic calculations if regions are clearly connected to the three-dimensional vortex
the chemical composition of the ash is known (Backman et al., structures of the flow field (compare Figures 14a and 18a). On
1987). In general, the melting behavior of the ash can vary sig- the furnace walls, the largest regions are in the central part of
nificantly depending of the pulp mill and process conditions. It the front wall (I) and the corners of the rear wall (II). Signifi-
is especially sensitive to the chloride and potassium content of cant sticky areas are also in the corner region of the front wall
the ash. The sticky temperature range is an established concept and the roof (III). These numerical observations can be linked
and it has been utilized in a number of previous studies, such to practical observations, which typically indicate deposition
as, Mueller et al. (2003, 2005). in these regions. On the superheater 2, the regions containing
Next, a simple approach based on the sticky temperature sticky ash are mainly in the top part, top corners, and near the
16
Figure 14: a) Three-dimensional vortex structures identified via the standard Q-criterion method with Q = 0.01 and colored by the magnitude of vorticity. b) Heat
flux toward the boiler walls. c) Heat flux toward the superheaters and boiling surfaces.

Figure 15: a) The ratio of the radiation heat flux to the total heat flux on the su-
perheater surfaces. The average values (Mean) and standard deviations (Stdev)
are shown. b) The average Nusselt numbers on the superheater surfaces. The
values are shown from the present simulation (CFD) and from an experimental
correlation (Gnielinski, 1975), with cross flow conditions perpendicular to the
tubes (Exp 1) and 35◦ oblique to the tubes (Exp 2).

side walls. On the superheater 4, the regions of sticky ash are

almost in the opposite locations, that is, mainly in the central
part. This numerical result is also supported by practical ob- Figure 16: The ratio of the radiation heat flux to the total heat flux on the a)
boiler walls and b) superheaters and boiling surfaces.
servations. Deposition was observed visually during the mea-
surement campaign in the central part of the superheater 4 (see
Figure 5).
The results indicate that the fouling behavior of the ash can 5. Conclusions
be considerably non-intuitive, since it is closely connected to
the three-dimensional flow and temperature field solutions. The We presented a new model for the recovery boiler super-
potential locations of fouling can also vary notably between dif- heater region, called the full superheater region (FSR) model,
ferent superheaters. The results indicate that for an accurate so- and reported two sets, A and B, of previously unpublished full-
lution of fouling in the superheater region, a detailed fouling scale temperature and flow field measurements from the super-
model should be coupled with a three-dimensional CFD model. heater region. The accuracy of the computational solution was
17
 structures and their consequent impact on the heat trans-
fer, b) the detailed heat flux solutions toward the furnace
walls, boiling surfaces, and superheaters, c) the detailed
analysis of the radiation and convection heat transfer, and
d) the quantitative variation of the heat flux profiles of the
superheaters in the width direction.
5. The results indicated that the fouling behavior of the ash is
closely connected to the three-dimensional flow and tem-
perature field solutions. It was considered that for an ac-
curate solution of fouling a detailed fouling model should
be coupled with a three-dimensional CFD model.
The results of this work and the developed CFD model are
valuable for optimizing the efficiency of recovery boilers. Due
to the large size of the global pulp industry, the amount of
black liquor combusted in recovery boilers every year is very
large. Therefore, the potential in optimizing the recovery boil-
ers for the generation of renewable energy is tremendous. In
the present work, the developed CFD framework was utilized
in connection to recovery boilers but it could also be applica-
ble to other types of large-scale energy production applications,
such as biomass-fired industrial boilers or utility boilers.

Acknowledgments

The authors wish to thank Ilkka Välipakka, Lars-Gunnar

Magnusson, Niko Metsämuuronen, and Arto Paunonen who
were responsible for the measurement campaign. The sup-
Figure 17: The total heat flux and radiation heat flux profiles of each superheater
port obtained from the mill personnel during the campaign is
in the width direction of the boiler (z-direction, see Figure 5 for the locations also acknowledged. In addition, we gratefully acknowledge
of the superheaters). The heat fluxes have been calculated for each platen as Mohammad Hadi Bordbar for implementing and providing the
area-weighted averages. The positive platen numbers correspond to the right code for the non-gray weighted sum of gray gases method and
side of the boiler and the negative platen numbers to the left side of the boiler.
The average total heat flux is also shown.
for his valuable advice with radiation modeling.

verified with a grid convergence study by utilizing the stan-

dard grid convergence index (GCI) method. The modeled re-
sults were validated against the measurement sets A and B. The
added-value of the CFD model was illustrated with a detailed
analysis of the three-dimensional simulation results. Based on
the results, we make the following conclusions:

1. The error estimates calculated with the GCI method were

considered low and the computational solution reliable.
2. The overall match between the modeled and measured val-
ues was considered good and the model validation suc-
cessful. Due to the experimental uncertainty associated
with these challenging measurements, a perfect correspon-
dence could not be expected.
3. The present measurement data is considered extremely
valuable, since reported experimental data on recovery
boilers is scarce in literature. Further comprehensive mea-
surement campaigns are necessary for the development
and validation of the recovery boiler CFD models.
4. To the best of our knowledge, the following results were
reported for the first time: a) the three-dimensional vortex
18
Figure 18: The identified regions of potential fouling on the a) furnace walls, b) superheater 2 (SH 2), and c) superheater 4 (SH 4). The regions where the temperature
of the ash is in the sticky range are shown in yellow (T 15 –T 70 ). The regions where the temperature of the ash is outside the sticky range are shown in blue (below
T 15 or above T 70 ). The figures b and c show the yz-surfaces in the middle of the superheaters in the x-direction (the surfaces are also marked in the figure a). In the
figures b and c, the superheater platens are visible as vertical black lines.

References Kakaras, E., 2014. Decoupled CFD simulation of furnace and heat exchang-
ers in a lignite utility boiler. Fuel 117, 633–648.
Abdullah, Z., Salcudean, M., Nowak, P., Xiao, Z., Savage, M. C., Markovic, C., Drosatos, P., Nikolopoulos, N., Agraniotis, M., Kakaras, E., 2016. Numerical
Uloth, V. C., Thorn, P. H., 1994. Initial validation of a mathematical model investigation of firing concepts for a flexible greek lignite-fired power plant.
of cold flow in a recovery boiler. TAPPI Journal 77 (5), 149–157. Fuel Processing Technology 142, 370–395.
Adams, T. N. (Ed.), 1997. Kraft Recovery Boilers. TAPPI Press. Drosatos, P., Nikolopoulos, N., Nikolopoulos, A., Papapavlou, C., Grammelis,
Al-Abbas, A. H., Naser, J., Dodds, D., 2012. CFD modelling of air-fired and P., Kakaras, E., 2017. Numerical examination of an operationally flexible
oxy-fuel combustion in a large-scale furnace at Loy Yang A brown coal lignite-fired boiler including its convective section using as supporting fuel
power station. Fuel 102, 646–665. pre-dried lignite. Fuel Processing Technology 166, 237–257.
Backman, R., Hupa, M., Uppstu, E., 1987. Fouling and corrosion mechanisms Edge, P. J., Heggs, P. J., Pourkashanian, M., Williams, A., 2011. An integrated
in the recovery boiler superheater area. TAPPI Journal 70 (6), 123–127. computational fluid dynamics-process model of natural circulation steam
Baxter, L. L., Lind, T., Kauppinen, E., Robinson, A., 2001. Thermal proper- generation in a coal-fired power plant. Computers & Chemical Engineering
ties of recovery boiler deposits. Proceedings of the International Chemical 35 (12), 2618–2631.
Recovery Conference 3, 133–138. Elghobashi, S., 1994. On predicting particle-laden turbulent flows. Applied Sci-
Bergroth, N., Engblom, M., Mueller, C., Hupa, M., 2010. CFD-based modeling entific Research 52 (4), 309–329.
of kraft char beds - part 1: Char bed burning model. TAPPI Journal 9 (2), Engblom, M., Bergroth, N., Mueller, C., Jones, A., Brink, A., Hupa, M., 2010a.
6–13. CFD-based modeling of kraft char beds - part 2: A study on the effects of
Blasiak, W., Tao, L., Vaclavinek, J., Lidegran, P., 1997. Modeling of kraft re- droplet size and bed shape on bed processes. TAPPI Journal 9 (2), 15–20.
covery boilers. Energy Conversion and Management 38 (10), 995–1005. Engblom, M., Brink, A., Rönnqvist, A., Mueller, C., Jones, A., Hupa, M.,
Brink, A., Lauren, T., Hupa, M., Koschack, R., Mueller, C., 2010. In-furnace 2010b. Recovery boiler char bed dynamics - measurements and modeling.
temperature and heat flux mapping in a kraft recovery boiler. TAPPI Journal Proceedings of the International Chemical Recovery Conference, 119–133.
9 (9), 7–11. Engblom, M., Miikkulainen, P., Brink, A., Hupa, M., 2012. CFD-modeling for
Celik, I. B., Ghia, U., Roache, P. J., Freitas, C. J., Coleman, H., Raad, P. E., more precise operation of the kraft recovery boiler. TAPPI Journal 11 (11),
2008. Procedure for estimation and reporting of uncertainty due to dis- 19–27.
cretization in CFD applications. Journal of Fluids Engineering 130 (7). Ferreira, D. J. O., Cardoso, M., Park, S. W., 2010. Gas flow analysis in a kraft
Choi, C. R., Kim, C. N., 2009. Numerical investigation on the flow, com- recovery boiler. Fuel Processing Technology 91 (7), 789–798.
bustion and NOx emission characteristics in a 500 MWe tangentially fired Frederick, W. J., 1990. Combustion processes in black liquor recovery: Anal-
pulverized-coal boiler. Fuel 88 (9), 1720–1731. ysis and interpretation of combustion rate data and an engineering design
Di Giacinto, M., Sabetta, F., Piva, R., 1982. Two-way coupling effects in dilute model. DOE Report DOE/CE/40637-T8.
gas-particle flows. Journal of Fluids Engineering 104 (3), 304–311. Gnielinski, V., 1975. Berechnung mittlerer wärme- und
Dı́ez, L. I., Cortés, C., Campo, A., 2005. Modelling of pulverized coal boilers: stoffübergangskoeffizienten an laminar und turbulent überströmten
review and validation of on-line simulation techniques. Applied Thermal einzelkörpern mit hilfe einer einheitlichen gleichung. Forschung im
Engineering 25 (10), 1516–1533. Ingenieurwesen 41 (5), 145–153.
Dı́ez, L. I., Cortés, C., Pallarés, J., 2008. Numerical investigation of NOx emis- Grace, T. M., 1995. A critical review of computer modeling of kraft recovery
sions from a tangentially-fired utility boiler under conventional and overfire boilers. TAPPI Journal 79 (7), 182–190.
air operation. Fuel 87 (7), 1259–1269. Grace, T. M., Lien, S., Schmidl, W., Tse, D., Abdullah, Z., Salcudean, M.,
Dorigon, L. J., Duciak, G., Brittes, R., Cassol, F., Galarça, M., França, F. H. R., 1998. Validation of CFD-based recovery furnace models. Proceedings of
2013. WSGG correlations based on HITEMP2010 for computation of ther- the International Chemical Recovery Conference 1, 271–281.
mal radiation in non-isothermal, non-homogeneous H2 O/CO2 mixtures. In- Grace, T. M., Walsh, A. R., Jones, A. K., Sumnicht, D. W., Farrington, T. E.,
ternational Journal of Heat and Mass Transfer 64, 863–873. 1989. A three-dimensional mathematical model of the kraft recovery fur-
Drosatos, P., Nikolopoulos, N., Agraniotis, M., Itskos, G., Grammelis, P., nace. Proceedings of the International Chemical Recovery Conference.

19
Hayes, A. M., Khan, J. A., Shaaban, A. H., Spearing, I. G., 2008. The thermal Menter, F. R., 1994. Two-equation eddy-viscosity turbulence models for engi-
modeling of a matrix heat exchanger using a porous medium and the thermal neering applications. AIAA Journal 32 (8), 1598–1605.
non-equilibrium model. International Journal of Thermal Sciences 47 (10), Miikkulainen, P., Pakarinen, L., Metsämuuronen, N., 2010. Challenges in vali-
1306–1315. dating recovery boiler furnace models in practice. Proceedings of the Inter-
He, B., Zhu, L., Wang, J., Liu, S., Liu, B., Cui, Y., Wang, L., Wei, G., 2007. national Chemical Recovery Conference 1, 60–72.
Computational fluid dynamics based retrofits to reheater panel overheating Mikkanen, P., Kauppinen, E. I., Pyykönen, J., Jokiniemi, J. K., Aurela, M.,
of no. 3 boiler of Dagang power plant. Computers & Fluids 36 (2), 435–444. Vakkilainen, E. K., Janka, K., 1999. Alkali salt ash formation in four finnish
Isaak, P., Tran, H. N., Barham, D., Reeve, D. W., 1986. Stickiness of fireside industrial recovery boilers. Energy & Fuels 13 (4), 778–795.
deposits in kraft recovery units. Journal of Pulp and Paper Science 12 (3), Mueller, C., Eklund, K., Forssen, M., Hupa, M., 2004. Influence of liquor-to-
84–88. liquor differences on recovery furnace processes - a CFD study. Proceedings
Jang, J.-Y., Yang, J.-Y., 1998. Experimental and 3-D numerical analysis of of the International Chemical Recovery Conference 2, 979–997.
the thermal-hydraulic characteristics of elliptic finned-tube heat exchangers. Mueller, C., Selenius, M., Theis, M., Skrifvars, B.-J., Backman, R., Hupa, M.,
Heat Transfer Engineering 19 (4), 55–67. Tran, H., 2005. Deposition behaviour of molten alkali-rich fly ashes - devel-
Jokiniemi, J. K., Pyykönen, J., Mikkanen, P., Kauppinen, E. I., 1996. Model- opment of a submodel for CFD applications. Proceedings of the Combustion
ing fume formation and deposition in kraft recovery boilers. TAPPI Journal Institute 30 (2), 2991–2998.
79 (7), 171–181. Mueller, C., Skrifvars, B.-J., Backman, R., Hupa, M., 2003. Ash deposition
Jones, A. K., 1989. A model of the kraft recovery furnace. Ph.D. thesis, The prediction in biomass fired fluidised bed boilers - combination of CFD and
Institute of Paper Chemistry. advanced fuel analysis. Progress in Computational Fluid Dynamics 3 (2-4),
Järvinen, M., Mueller, C., Hupa, M., Fogelholm, C.-J., 2011. A CFD-applicable 112–120.
discrete combustion model for thermally large particles. Progress in Com- Nikolopoulos, N., Nikolopoulos, A., Karampinis, E., Grammelis, P., Kakaras,
putational Fluid Dynamics 11 (6), 373–387. E., 2011. Numerical investigation of the oxy-fuel combustion in large scale
Järvinen, M., Zevenhoven, R., Vakkilainen, E. K., 2002. Auto-gasification of a boilers adopting the ECO-scrub technology. Fuel 90 (1), 198–214.
biofuel. Combustion and Flame 131 (4), 357–370. Park, H. Y., Faulkner, M., Turrell, M. D., Stopford, P. J., Kang, D. S., 2010.
Jukola, P., Kyttälä, J., McKeough, P., 2014. Predicting sodium release in recov- Coupled fluid dynamics and whole plant simulation of coal combustion in a
ery boilers in conjunction with CFD furnace modelling. Journal of Science tangentially-fired boiler. Fuel 89 (8), 2001–2010.
and Technology for Forest Products and Processes 4 (4), 48–55. Patankar, S. V., Spalding, D. B., 1972. A calculation procedure for heat, mass
Karvinen, R., Hyöty, P., Siiskonen, P., 1991. The effect of dry solids content on and momentum transfer in three-dimensional parabolic flows. International
recovery boiler furnace behavior. TAPPI Journal 74 (12), 171–177. Journal of Heat and Mass Transfer 15 (10), 1787–1806.
Kawaji, M., Shen, X., Tran, H., Esaki, S., Dees, C., 1995. Prediction of Patankar, S. V., Spalding, D. B., 1974. A calculation procedure for the transient
heat transfer in the kraft recovery boiler superheater region. TAPPI Journal and steady-state behavior of shell-and-tube heat exchangers. Heat Exchang-
78 (10), 214–221. ers: Design and Theory Source Book, 155–174.
Kritikos, K., Albanakis, C., Missirlis, D., Vlahostergios, Z., Goulas, A., Storm, Pérez, M. G., Vakkilainen, E. K., Hyppänen, T., 2016. Fouling growth model-
P., 2010. Investigation of the thermal efficiency of a staggered elliptic-tube ing of kraft recovery boiler fume ash deposits with dynamic meshes and a
heat exchanger for aeroengine applications. Applied Thermal Engineering mechanistic sticking approach. Fuel 185, 872–885.
30 (2-3), 134–142. Prithiviraj, M., Andrews, M. J., 1998a. Three dimensional numerical simulation
Le Bris, T., Cadavid, F., Caillat, S., Pietrzyk, S., Blondin, J., Baudoin, B., 2007. of shell-and-tube heat exchangers. part I: Foundation and fluid mechanics.
Coal combustion modelling of large power plant, for NOx abatement. Fuel Numerical Heat Transfer, Part A Applications 33 (8), 799–816.
86 (14), 2213–2220. Prithiviraj, M., Andrews, M. J., 1998b. Three-dimensional numerical simula-
Leppänen, A., Tran, H., Taipale, R., Välimäki, E., Oksanen, A., 2014a. Nu- tion of shell-and-tube heat exchangers. part II: Heat transfer. Numerical Heat
merical modeling of fine particle and deposit formation in a recovery boiler. Transfer, Part A Applications 33 (8), 817–828.
Fuel 129, 45–53. Pöyry, 2015. World Fibre Outlook up to 2030. [Electronic publication].
Leppänen, A., Tran, H., Välimäki, E., Oksanen, A., 2014b. Modelling fume de- Pyykönen, J., Jokiniemi, J., 2003. Modelling alkali chloride superheater depo-
posit growth in recovery boilers: Effect of flue gas and deposit temperature. sition and its implications. Fuel Processing Technology 80 (3), 225–262.
Journal of Science and Technology for Forest Products and Processes 4 (1), Raithby, G. D., Chui, E. H., 1990. A finite-volume method for predicting a
50–57. radiant heat transfer in enclosures with participating media. Journal of Heat
Leppänen, A., Välimäki, E., 2016. Improving recovery boiler availability Transfer 112 (2), 415–423.
through understanding fume behavior. TAPPI Journal 15 (3), 187–193. Rezaei, H. R., Gupta, R. P., Bryant, G. W., Hart, J. T., Liu, G. S., Bailey, C. W.,
Leppänen, A., Välimäki, E., Oksanen, A., Tran, H., 2013. CFD-modeling of Wall, T. F., Miyamae, S., Makino, K., Endo, Y., 2000. Thermal conductivity
fume formation in kraft recovery boilers. TAPPI Journal 12 (3), 25–32. of coal ash and slags and models used. Fuel 79 (13), 1697–1710.
Li, B., Brink, A., Hupa, M., 2013. CFD investigation of slagging on a super- Saastamoinen, J. J., 1996. Modelling of drying, devolatilization and swelling of
heater tube in a kraft recovery boiler. Fuel Processing Technology 105, 149– black liquor droplets. AIChE Symposium Series 92 311, 74.
153. Salcudean, M., Nowak, P., Abdullah, Z., 1993. Cold flow computational model
Li, B., Engblom, M., Lindberg, D., Brink, A., Hupa, M., Koschack, R., Mueller, of a recovery boiler. Journal of Pulp and Paper Science 19 (5), 186–194.
C., 2012. Numerical investigation of kraft recovery furnace wall tempera- Saviharju, K., Pakarinen, L., Kyttälä, J., Jukola, P., Viherkanto, K., Näkki, I.,
ture. Journal of Science and Technology for Forest Products and Processes Hämäläinen, M., 2007. Three dimensional char bed imaging for numerical
2 (5), 41–48. simulation feedback. Proceedings of the International Chemical Recovery
Maakala, V., Järvinen, M., Vuorinen, V., 2015. Improving heat transfer to re- Conference, 469–472.
covery boiler superheaters using optimization and computational fluid dy- Saviharju, K., Pakarinen, L., Wag, K., Välipakka, I., 2004. Numerical model-
namics. Proceedings of the TAPPI Peers Conference, 1415–1428. ing feedback in recovery boilers. Proceedings of the International Chemical
Maakala, V., Järvinen, M., Vuorinen, V., 2016. Experimental validation of a Recovery Conference, 247–262.
recovery boiler superheater region CFD model. Proceedings of the TAPPI Schuhbauer, C., Angerer, M., Spliethoff, H., Kluger, F., Tschaffon, H., 2014.
Peers Conference, 921–944. Coupled simulation of a tangentially hard coal fired 700 ◦ C boiler. Fuel 122,
Maakala, V., Järvinen, M., Vuorinen, V., 2017. Mixing of high momentum flux 149–163.
jets with a confined crossflow: Computational analysis and applications to Shi, Y.-L., Ji, J.-J., Zhang, C.-L., 2010. Semiporous media approach for numer-
recovery boiler air systems. Journal of Science and Technology for Forest ical simulation of flow through large-scale sparse tubular heat exchangers.
Products and Processes [In Press]. HVAC&R Research 16 (5), 617–628.
Maakala, V., Miikkulainen, P., 2015. Dimensioning a recovery boiler furnace Statistics Finland, 2017. Official Statistics of Finland: Production of Electricity
using mathematical optimization. TAPPI Journal 14 (2), 119–129. and Heat. [Electronic publication].
McKeough, P., Janka, K., 2001. Sulphur behaviour in the recovery boiler fur- Sumnicht, D. W., 1989. A computer model of a kraft char bed. Ph.D. thesis,
nace: theory and measurements. Proceedings of the International Chemical The Institute of Paper Chemistry.
Recovery Conference, 231–237. Tran, H. N., Mao, X., Kuhn, D. C. S., Backman, R., Hupa, M., 2002. The sticky

20
 temperature of recovery boiler fireside deposits. Pulp and Paper Canada
103 (9), 29–33.
Vainio, E., Brink, A., Demartini, N., Hupa, M., Vesala, H., Tormonen, K.,
Kajolinna, T., 2010. In-furnace measurement of sulphur and nitrogen species
in a recovery boiler. Journal of Pulp and Paper Science 36 (3-4), 135–142.
Vakkilainen, E. K., 2005. Kraft Recovery Boilers - Principles and Practice.
Suomen Soodakattilayhdistys.
Vakkilainen, E. K., Adams, T., Horton, R. R., 1992. The effect of recovery
furnace bullnose designs on upper furnace flow and temperature profiles.
Proceedings of the International Chemical Recovery Conference.
Vakkilainen, E. K., Hautamaa, J., Nikkanen, S., Anttonen, T., 1991. Flows in
the upper region of recovery boilers. Proceedings of the Forest Products
Symposium, 125–134.
Vakkilainen, E. K., Kjäldman, L., Taivassalo, V., Kilpinen, P., Norström, T.,
1998. High solids firing in an operating recovery boiler - comparison of
CFD predictions to practical observations in the furnace. Proceedings of the
International Chemical Recovery Conference 1, 245–256.
VDI-Gesellschaft, 1993. VDI Heat Atlas. VDI Verlag.
Verrill, C. L., Wessel, R. A., 1998. Detailed black liquor drop combustion
model for predicting fume in kraft recovery boilers. TAPPI Journal 81 (9),
139–148.
Walsh, A. R., 1989. A computer model for in-flight black liquor combustion in
a kraft recovery furnace. Ph.D. thesis, The Institute of Paper Chemistry.
Walsh, A. R., Grace, T. M., 1988. TRAC: A computer model to analyze the
trajectory and combustion behavior of black liquor droplets. Journal of Pulp
and Paper Science 15 (3), 84–89.
Weber, R., Mancini, M., Schaffel-Mancini, N., Kupka, T., 2013. On predict-
ing the ash behaviour using computational fluid dynamics. Fuel Processing
Technology 105, 113–128.
Wessel, R. A., Baxter, L. L., 2003. Comprehensive model of alkali-salt deposi-
tion in recovery boilers. TAPPI Journal 2 (2), 19–24.
Wessel, R. A., Denison, M. K., Samretvanich, A., 2000. The effect of fume on
radiative heat transfer in kraft recovery boilers. TAPPI Journal 83 (7), 1–11.
Wessel, R. A., Parker, K. L., Verrill, C. L., 1997. Three-dimensional kraft re-
covery furnace model: Implementation and results of improved black-liquor
combustion models. TAPPI Journal 80 (10), 207–220.
Yang, J., Ma, L., Bock, J., Jacobi, A. M., Liu, W., 2014. A comparison of four
numerical modeling approaches for enhanced shell-and-tube heat exchang-
ers with experimental validation. Applied Thermal Engineering 65 (1-2),
369–383.
Yin, C., Caillat, S., Harion, J.-L., Baudoin, B., Perez, E., 2002. Investigation
of the flow, combustion, heat-transfer and emissions from a 609 MW utility
tangentially fired pulverized-coal boiler. Fuel 81 (8), 997–1006.
Yin, C., Rosendahl, L., Condra, T. J., 2003. Further study of the gas temperature
deviation in large-scale tangentially coal-fired boilers. Fuel 82 (9), 1127–
1137.
Zbogar, A., Frandsen, F. J., Jensen, P. A., Glarborg, P., 2005. Heat transfer in
ash deposits: A modelling tool-box. Progress in Energy and Combustion
Science 31 (5-6), 371–421.
Zhang, J.-F., He, Y.-L., Tao, W.-Q., 2009. 3D numerical simulation on shell-
and-tube heat exchangers with middle-overlapped helical baffles and contin-
uous baffles - part I: Numerical model and results of whole heat exchanger
with middle-overlapped helical baffles. International Journal of Heat and
Mass Transfer 52 (23-24), 5371–5380.

21