Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Contents lists available at ScienceDirect

Chemical Engineering & Processing: Process Intensification

journal homepage: www.elsevier.com/locate/cep

A hydrodynamic model for biomass gasification in a circulating fluidized T

bed riser
⁎
Mohamed I. Hassana,1, Yassir T. Makkawib,
a
European Bioenergy Research Institute (EBRI), School of Engineering and Applied Science, Aston University, Birmingham, B4 7ET, UK
b
Chemical Engineering Department, College of Engineering, American University of Sharjah, P.O. Box 26666, Sharjah, United Arab Emirates

A R T I C LE I N FO A B S T R A C T

Keywords: This study presents a three-dimensional Computational Fluid Dynamic (CFD) model and experimental mea-
Circulating fluidized bed surements of the hydrodynamics in the riser section of a Circulating Fluidized Bed (CFB) biomass gasifier
CFD modeling consisting of a binary mixture of polydisperse particles. The model is based on multi-fluid (Eulerian-Eulerian)
Hydrodynamics approach with constitutive equations adopted from the Kinetic Theory of Granular Flow (KTGF). The study first
Experiment validation
presents an assessment of the various options of the constitutive and closure equations for a binary mixture
Gasification
followed by sensitivity analysis of the model to the solution time step, cell size, turbulence and the alternative
formulations of the granular energy equation. Accordingly, a robust and reliable hydrodynamic model is re-
commended and validated using conventional pressure measurements and Positron Emission Particle Tracking
(PEPT) technique. Furthermore, the model predictions and experiments revealed evidence of the particle re-
circulation within the lower part of the riser, which is an important feature contributing to rapid mass and heat
transfer in a CFB gasifier. The present hydrodynamic model can be further developed; by incorporating ap-
propriate reactions and heat transfer equations, in order to fully predict the performance and products of a CFB
biomass gasifier.

1. Introduction waste and conversion of biomass to bio-fuels are nowadays frequently

reported in the literature [2,3]. The advances in computational power
The application of Circulating Fluidized Beds (CFB) to biomass has also been matched with the development of user-friendly com-
thermal conversion (gasification and pyrolysis) is currently receiving mercial software. Examples of some of the widely used commercial
increasing attention due to its good mixing, high thermal efficiency and software for particle-gas flow simulation are ANSYS (FLUENT and CFX),
most importantly, its excellent scale-up potentials up to around COMSOL and Barracuda. Other highly functional open-source CFD
1000 tonnes of dry biomass feed per day [1]. However, practical ex- software is also available free for academic use such as MFIX and
perimental investigation in a CFB, for testing new feedstock or para- OpenFoam. Most of these codes offer the solution of multiphase flows in
metric sensitivity analysis, is often difficult, expensive and requires Eulerian-Eulerian or Eulerian-lagrangian approaches. The latter, which
high expertise due to the complexity associated with the high-tem- is also referred to as Eulerian-DEM, has the advantage of being more
perature operation and release of particulate and toxic/highly flam- accurate by tracking each individual particle, but at the expense of high
mable gases. Computational Fluid Dynamic (CFD) models integrated computational time (CPU time). The former treats both of the solid and
with equations describing the heat transfer, transport of species and gas phases as an interpenetrating continuum based on ensemble aver-
chemical reactions, offers an alternative solution. Such models can be aging, hence relatively faster in handling large systems but less re-
solved using standard computers at a reasonable computational time to vealing at the discrete particle level. Nevertheless, the Eulerian-Eu-
reveal detailed features of the reactor such as the solid and gas dis- lerian approach remains the most popular option among academic
tribution, velocities, pressure, temperature, gas species concentrations researchers for the simulation of multiphase flow with chemical reac-
and product quality. As a result, studies using CFD models for the si- tions. Having said that, it must be noted that applying the Eulerian-
mulation and analysis of processes involving rapid thermochemical Eulerian approach to three-dimensional simulations of polydispersed
conversion, such as in catalytic reforming of gases, combustion of solid suspension requires a good understanding of the various constitutive

⁎
Corresponding author.
E-mail addresses: mihassan@jazanu.edu.sa (M.I. Hassan), ymakkawi@aus.edu (Y.T. Makkawi).
1
Current address: Chemical Engineering Department, Faculty of Engineering, Jazan University, P.O. Box 114, Jazan, Saudi Arabia.

https://doi.org/10.1016/j.cep.2018.05.012
Received 3 March 2018; Received in revised form 18 May 2018; Accepted 22 May 2018
Available online 24 May 2018
0255-2701/ © 2018 Elsevier B.V. All rights reserved.
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Nomenclature ur particle terminal velocity (m s−1)

I unit tensor (–) Greek letters

Cd particle drag coefficient (–)
Cfr Friction coefficient (–) ε volume fraction (–)
C1, C2 C3 turbulence model constants (–) β momentum exchange coefficient (kg m−3 s-1)
di diameter of solid phase i (m) γ collisional energy dissipation (kg m−1 s-3)
e particle restitution coefficient (–) θ granular temperature of solid phase i (m2 s−2)
g0 radial distribution function (–) k diffusion coefficient of granular energy (kg m−1 s−1)
Gk Gb turbulence kinetics due to velocity gradient and buoyancy, ∅′ specularity coefficient (–)
respectively (kgm−1s-2) ρ density (kg m−3)
→
g gravity (m s−2) μeff effective turbulence viscosity (kg m−1 s−1)
K solid-solid momentum exchange coefficient (kg m−3 s-1) τ shear stress tensor (kg m−1 s-2)
ke turbulence kinetic energy (m2 s−2) Δcell computational face cell size (m)
wt turbulence dissipation energy (m2 s−3)
σe σt prandtl numbers for kinetic and dissipation energies, re- Subscripts
spectively (–)
Nc Courant number (–) s, g solid and gas phases, respectively
P pressure (Pa) w wall
Re Reynolds number (–) col wall
t time (s) kin kinetic
ug⃑ , us⃑ gas and solid velocity vector (m s−1)

and closure equations in addition to careful setting of the solution Fig. 1] shows two examples of the arrangements of a DFB reactor for
procedure. biomass steam gasification. In this study, the focus is made on the si-
As noted above, one of the major attractive features of the CFB mulation of the riser, shown in the left section of the CFB arrangement
gasification technology is its high thermal efficiency; it allows for the given in Fig. 1-b. Here, the gasification is carried out in the riser and the
supply of the heat required to derive a highly endothermic thermo- reaction is entirely driven by hot circulating inert solid introduced at
chemical conversion process in a closed loop without the need of ex- the bottom of the riser from the connected combustor.
ternal heating. For example, in steam gasification (also referred to as Review of the recent literature on modeling of biomass gasification
pyrolytic gasification), this is achieved by coupling two reactors, in fluidized bed reactors show that the CFB type received less attention
creating what is usually referred to as dual fluidized bed (DFB) gasifier. compared to the bubbling bed. See Table 1 for examples of the most
In this arrangement, one reactor is used for the gasification and another recent studies. Modeling of the flow hydrodynamics in a CFB gasifier is
is used for the char combustion. In the gasifier, the biomass is brought challenging due to the existence of multi-solids which undergo complex
into contact with the fluidizing gas (steam or air/steam mixture) and a collisional interactions while dispersing in a continuum gas phase.
heat carrier solid, ideally to maintain the gasifier within the range of Furthermore, the accuracy of the Eulerian-Eulerian model for a poly-
750–950 °C [2,4]. In the combustor, the transferable heat carrier solid dispersed suspension is heavily dependent on a range of constitutive
(such as sand) is raised to a high temperature by char combustion. The and closure equations, as noted earlier. Therefore, the overall objective
use of steam in biomass gasification is particularly attractive as it en- of this study is to identify a reliable and robust hydrodynamic model,
hances hydrogen production through a water gas shift reaction [5,6]. which can be used as a platform for further development of a full

Fig. 1. Examples of dual fluidized bed reactors for biomass gasification (a) bubbling bed coupled with a riser (b) two coupled risers.

149
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Table 1
Summary of some recent studies on modeling of biomass gasification in fluidized bed (FB) reactors.
Modeling approach and software Gasifier type Main subject of the study Authors

Model: 3D Eulerian-Lagrangian Bubbling FB - Model-set-up Ku et al. [7]

Software: OpenFOAM - Reactor hydrodynamics and temperature
- Effect of steam to biomass ratio
- Product gas composition
Model: 3D Eulerian-Eulerian Bubbling FB - Model sensitivity analysis Gel et al. [8]
Software: MFIX - Effect of grid resolution
- Effect of particle size and steam to biomass ratio
- Product gas composition
Model: 2D Eulerian-Eulerian Bubbling FB - Hydrodynamics and turbulence Anil et al. [9]
Software: ANSYS FLUENT - Equivalence and steam to biomass ratios
- Effect of steam and air temperatures
Model: 3D MP- approach.PIC* Bubbling FB - Model set-up Yan et al. [10]
Software: OpenFOAM - Hydrodynamic predictions
- Product gas composition
Model: 3D MP-PIC* Bubbling FB in a Dual FB system - Steam gasification Kraft et al. [11]
Software: Barracuda - Hydrodynamics and temperature distribution
- Product gas composition
Model: 2D Eulerian-Eulerian Internal circulating FB - Hydrodynamics and reactor temperature Juhui et al. [2]
Software: ANSYS FLUENT - Granular temperature
- Product gas composition
Model: 3D MP-PIC* Bubbling FB in a Dual FB system - Hydrodynamics Liu et al. [12]
Software: Barracuda - Solid circulation rate Liu et al. [13]
- The impact of gasifier temperature
- Effect of steam to biomass ratio
- Effect of air supply to the combustor
- Product gas composition.

*MP-PIC: Multi-Phase Particle-In-Cell (MP-PIC) is a Lagrangian-based.

predictive tool of biomass gasification in a CFB riser.

The model is solved using ANSYS FLUENT commercial CFD code.
This code has the advantage of providing a range of built-in functions
and options that allow for the assessment of the various constitutive
equation and solution procedures, as well as providing UDF access to
the code solver. The hydrodynamic predictions are validated with ex-
perimental data obtained in a cold flow CFB riser using pressure mea-
surements and positron emission particle tracking (PEPT).

2. Experiments

Experiments in a cold flow CFB (isothermal and non-reactive) have

been carried out for two purposes. First, to collect data for the valida-
tion and assessment of the hydrodynamic model and its constitutive
equations. Second, to visually observe the flow characteristics and es-
tablish a rough estimate of the range of operating condition for future
simulation of biomass gasification in a CFB riser. The set-up shown in
Fig. 2 mainly comprises a relatively short riser with aspect ratio
(height/diameter) = 32.6, downer section, two cyclones, solid re-
ceiving tank and a data acquisition system. The dimensions of the riser
were carefully selected to allow covering the range of flow regimes
commonly observed in a biomass gasifier (i.e. dense bottom and rapid
dilute suspension at the top). The riser is made of glass and fitted with
eight pressure probes connected to piezoresistive pressure sensors/
transmitters. The downer side was equipped with an inlet for auxiliary
air and a diversion valve to allow measuring the total solid circulation
rate. Two rotameters were used to measure the airflow required to
fluidize the bed and maintain the solid circulation. In order to resemble
the particulate in a biomass gasifier, a mixture of sand and wood par-
ticles (Geldart group B particles) is used as the bed material. In some
cases, narrow-sized glass bead mixtures were used instead of sand, ei-
ther as a mixture with wood or as a single solid phase. Tables 2 gives
the details of the equipment, material and range of operating condition
considered in the experiments.

Fig. 2. Schematic representation of the cold flow CFB. All dimensions are in
centimeter.

150
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Table 2 Center. Prior to fluidization experiment, a single tracer particle was

Summary of the experimental unit and range of operating conditions. activated with Fluorine-18 and placed inside the CFB solid receiving
Experimental unit tank. The activated particle was then tracked during circulation using
Gamma photons detector plates located around the lower part of the
Riser Diameter = 5 cm; height = 163 cm; material: glass CFB riser, as shown in Fig. 3. The experiments were conducted using a
Solid receiving tank Cylindrical shape, 1000 m3 volume; material: steel
bed material consisting of sand–wood mixture. Two sets of data were
Air flow meter Two rotameters, combined flow of 1900 l/min (max)
Pressure transmitter 0-70 mbar (model CTEM70070GY4)
produced; one with the system tracking an activated wood particle and
Cyclones Main: 47 cm overall height and 12 cm top diameter another with an activated sand particle.
Secondary: 22.5 cm overall height and 6 cm top The recorded data were analyzed using MATLAB software to obtain
diameter transient and time-averaged information on the particle velocity and
Operating conditions
location within the PEPT detection zone. Because the radioactive par-
Particle sizes and Sand: d = 200−700 μm; ρs = 2500 kg / m3
density ticle was circulating between the riser and downer side, there was few
Wood: d = 500−1000 μm; ρs = 585 kg / m3
minutes delay between each set of detections, as will be demonstrated
Glass beads: dp = 360 − 750 μm; ρs = 2500 kg / m3
later. The total duration time used for the collection of the data was
Fluidization velocity 2−5 m/s
Solid circulation rate 40 g/s max (20 kg/m2s) around 2 h, beyond this time, the activity of the particles decays ra-
Fluidization medium Air at ambient condition pidly. All the measurements were taken at one fixed level within the
lower part of the CFB riser due to difficulty in moving the sensors to
higher levels (see Fig. 3b).
2.1. Pressure measurements
3. Hydrodynamic model equations
The axial pressure profile in the riser was measured using eight
pressure sensors/transmitters (Range: 1–70 mbar; model: The mathematical model used to simulate the CFB riser hydro-
CTEM70070GY4) connected to probes inserted at various heights above dynamics is described below. The multiphase flow mixture in the riser
the gas distributor (21 cm, 33 cm, 45 cm, 76 cm, 107 cm and 138 cm), is assumed to consist of a polydispersed binary solid mixture and air as
as shown in Fig. 2. The lower probes were distributed at short distances the fluidizing agent.
in order to capture the steep variations in pressure in this region. The
tips of the probes were covered by a fine metal mesh to prevent the
3.1. Main model equations
particles from entering the probes. A data acquisition system was used
to record the pressure at the rate of 1 Hz. The pressure at each point was
The main transient model equations include continuity, momentum
calculated using localized data averaged over 5–10 min of continuous
and granular energy in addition to turbulence closure equations. The
and steady circulation.
Continuity equations are given by:

2.2. Particle tracking ∂ (εg ρg )

+ ∇ (εg ρg ug⃑ ) = 0
∂t (1)
Positron Emission Particle Tracking (PEPT) is a non-invasive tech-
∂ (ε si ρsi )
nique that allows tracking the motion of a single radioactively labeled + ∇ (ε si ρsi ⇀
u si ) = 0 i = 1 or 2
tracer particle within a multiphase flow. The principle of the PEPT ∂t (2)
technique and its application in studying particle dynamics can be 2
found in the open literature [14–17]. The facility used in this study was ∑ εsi + εg = 1
located at the University of Birmingham, Nuclear Physics Research i=1 (3)

Fig. 3. A schematic description of the PEPT detectors surrounding the lower part of the CFB riser (all dimensions are in cm) and a Photo of the PEPT facility used
(Birmingham University, UK).

151
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

where ε is the volume fraction, ρ is the density, u ⃑ is the velocity vector. measurements. Gidaspow [30] derived the following drag law by
The subscripts g and s stand for the air and solid phases, respectively, combining Ergun [31] correlatn for packed bed with Wen and Yu [32]
while i = 1 or 2 refers to any of the two solid phases. Note that the right correlation for a dilute flow as follows:
side of Eqs. (1) and (2) is zero because no mass transfer takes place
3 εsi εg ρg |usi − ug | −2.65
(isothermal and non-reactive system). βsi = Cd εg if εg > 0.8
The momentum equations are given by: 4 dsi (8)

∂ (εg ρg ug⃑ ) 2 εsi (1 − εg ) μg εsi ρg |usi − ug |

+ ∇ (εg ρg ug⃑ ug⃑ ) = −εg ∇Pg + ∇τg− ∑ βg − si ( ug⃑ − usj⃑ ) + εg ρg→
g βsi = 150
εg dsi2
+ 1.75
dsi
if εg ≤ 0.8
∂t (9)
j=1

(4) where CD is the single particle drag coefficient given by:

∂ (εsi ρsi ⇀
usi ) 24
+ ∇ (εsi ρsi ⇀
usi ⇀
usi ) = εsi ∇Pg − ∇Psi + ∇τsi + βg − si (⇀
ug − ⇀u si ) CD = [1 + 0.15(εg Re )0.687] if εsi Re ≤ 1000
∂t εg Re (10)
+ Kij (⇀ usj ) + εsi ρ →
usi − ⇀ sig i and j = 1 or 2 CD = 0.44 if εsi Re ≥ 1000 (11)
(5)
Syamlal et al. [33] reported another formula described to be ap-
where β and K are the solid-gas and solid-solid momentum exchange plicable to a wide range of solid concentration as follows:
→
coefficient, respectively. P , τ and g are the solid pressure, solid shear
3εsi εg ρg Re ⎞
stress tensor and gravity constant, respectively. βsi = CD ⎛⎜ ⎟ |usi − u g | i = 1 or 2
The energy equation, derived from the principles of the Kinetic 4ur2, si dsi u
⎝ r , si ⎠ (12)
Theory of Granular Flow (KTGF) [18–20] is used to calculate the where ur , si and CD are the terminal velocity and single particle gas drag
granular temperature, an important parameter for calculating the par- coefficients, respectively (detailed in Table 4). In the rest of this study,
ticles interaction and the energy lost during. This is given by the fol- Eq. (12) will be used as default unless otherwise stated.
lowing pseudo-thermal energy balance equation: [18–20] Compared to the gas-solid drag, the momentum exchange between
the solid-solid in polydispersed mixtures is less understood and often
(εsi ρsi θsi + ∇ . (εsi ρsi ⇀
usi θsi ) ⎤ = (−Psi I + τsi ): ∇⇀
2 ∂
3⎡
usi + ∇ . (ksi ∇θsi ) − γsi
⎣ ∂t ⎦ neglected in the modeling of dilute suspension. This force, which
− 3βsi θsi araises from the velocity difference between the interacting solids, be-
i = 1 or 2 comes important when the particles are significantly different in size
(6) and/or density, such as in the case of sand and biomass in gasification
reactors. It is also crucially important in modeling segregation and
where θ , k and γ represent the granular temperature, diffusion coeffi-
mixing in dense bubbling fluidized beds [34]. In this study, the solid-
cient and collisional dissipation of energy. The last term on the right
solid momentum exchange coefficient is incorporated in the model by
side represents the energy exchange between the solid phase i and the
using the equation proposed by Syamlal [35] as follows:
fluid. Eq. (6) can be reduced to an algebraic expression by neglecting
the convection and diffusion terms to give the following equation: [21]
Kij =
3(1 + eij ) ( π
2
+
π2
C
8 fr , ij ) ρ ρ ε ε (d
si sj si sj si +dsj )
2g
0, ij
(usi − usj ) i and j
0 = (−Psi I + τsi ): ∇⇀
usi − γsi − 3βsi θsi i = 1 or 2 (7) 2π (ρsi dsi3 + ρsj dsj3 )
The validity of this simplified equation for the case under con- = 1 or 2 (13)
sideration will be discussed in the results section.
where Cfr,ls is the coefficient of friction between the two solid phases
(assumed zero in this study due to the dilute nature of the flow), g0 is
3.2. Constitutive and closure equations the radial distribution function (discussed in details in Section 3.2.2)
and e is the particle restitution coefficient (given in Table 3). The
Constitutive equations for the Eulerian two-phase flow model are subscripts i and j represent the two solid phases given that i ≠ j.
widely reviewed and validated for the case of a monodispersed dense-
intermediate suspension [22–29]. In particular, the solid-gas mo- 3.2.2. Radial distribution function
mentum exchange equation (gas drag low) for bubbling fluidized beds The KTGF is based on the assumption that particles collision obeys
received the most attention [26–28]. However, less attention has been Chapman-Enskog theory for molecules interactions in dense gases [36],
given to the case of polydispersed dilute suspension, such as in CFB but with the granules exhibiting loss of momentum due to inelasticity.
systems. In this section, the focus is made on the constitutive and clo- The probability of collisions are introduced in the Eulerian-Eulerian
sure equation describing the solid-gas and solid-solid momentum ex- model through the radial distribution function, g0 , which appears in a
change, radial distribution function, solid pressure and turbulence. The number of the KTGF constitutive equations such as the solid pressure,
remaining constitutive and closure equations, which are used as de- the solid-solid momentum exchange and the solid shear viscosity. The
fault, are given in Table 4 as part of the final recommended constitutive
and closure equations. Table 3
Values of the fixed parameters used in the solution of the hydrodynamic model.
3.2.1. Solid-gas and solid-solid momentum exchange coefficients
Parameter Value
Multiphase flow in a CFB riser generally falls under the category of
rapid particle motion. However, the bottom of the riser exhibits a re- Particle-particle restitution coefficient, es and ew sand: 0.8
latively dense flow structure that gradually decreases to form a rapid (–)* glass beads: 0.9
dilute flow toward the exit. It is also recognized that the major heat and wood: 0.7
Specularity coefficient, ∅′ (–)* 0.5
mass transfer in a CFB gasifier take place in the bottom part of the riser.
Maximum solid fraction, εs − max (–) 0.63 (for all solid phases)
Therefore, it is important to apply a solid-gas drag equation capable of
Turbulence model constants, C1, C2 and C3 (–) 1.44, 1.92, 0 (respectively)
capturing such variations in the flow regime. Here, the two most widely Coefficient of friction, Cfr (–) 0
used models, reported to satisfy this requirement, will be used and their
capabilities in correctly predicting dilute binary mixture hydro- *Vales are based on the wide reported literature on fluidization of glass beads,
dynamics will be assessed by comparing with experimental sand and wood particles.

152
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

2
Table 4 1 3 εsj
The recommended constative and closure equations. g0, i = + dsi ∑ i and j = 1 or 2

Gas-solid momentum exchange coefficient [31]

( 1−
εs, all
εs, max ) 2 j=1
dsj
(17)
3εsi εg ρg Re
βsi = CD ⎛ si ⎞ |usi − ug | i = 1 or 2 A similar formula for multi-solids was derived by Lebowitz [44] and
4ur2, si dsi ⎝ ur , si ⎠
used by Syamlal [33] for a binary mixture as follows:
where
2
CD = ⎛0.63 +
4.8 ⎞ 1 3dsi dsj εsk
⎝
⎜
Resi / ur , si
⎟

⎠
g0, ij =
εg
+ 2 ∑ dsk
i and j = 1 or 2
εg (dsi + dsj ) k=1 (18)
ur , si = 0.5(εg4.14−0.06Resi + (0.06Rsi )2 + 0.12Resi (1.6εg1.28−εg4.14 ) + εg8.18 )
Note: This model is applicable when εg ≤ 0.85 , otherwise, the term 1.6εs1.28 should which reduces to the following formula for a selected solid phase i :
be 2εs2.65 . 2
1 3d εsk
Solid-solid momentum exchange coefficient [33] g0, i = + si2 ∑ i = 1 or 2
π π2
εg 2εg k=1
dsk (19)
3(1 + eij ) ⎜⎛ + C ⎟⎞ ρ ρ εsi εsj (dsi +dsj )2g0, ij
⎝2 8 f , ij ⎠ si sj
Kij = 3 + ρ d3 ) (usi − usj ) i and j = 1 or 2 where εg is the void fraction given by:
2π (ρsi dsi sj sj
where Cfr,ls = 0 (no friction) 2
Granular Viscosity [31] εg = 1−∑ εsj
μsi = μsi, col + μsi, kin i = 1 or 2 j (20)
θsi 1/2
4
μsi, col = εsi dsi ρsi g0i (1 + esi )
5 ( )
π It is worth noting that, in most of the currently available radial
μsi, kin =
εsi dsi ρsi θsi π 2
⎡1 + 5 (1 + esi )(3esi−1) εsi g0i⎤ distribution functions, including the ones above, it is assumed that the
⎣
6(1 − εsi ) ⎦
function is independent of the particle density. However, there is an
Radial distribution function [38,21]
g0i dsi + g0j dsj argument that such assumption may not be accurate, especially if the
g0, ij = i and j = 1 or 2
dsi + dji densities are considerably different. Wang et al. [38] derived a modified
εs, all 1/3 −1 εsj radial distribution function including the particle densities using data
g0i = ⎡1−⎛ ⎤ 1 2
⎞ + dsi ∑ j = 1
⎢ ⎝ εs, max ⎠
⎣
⎥
⎦
2 dsj from discrete hard-sphere model simulation. The modified function
where εs, all = ∑ j εsj
2 produced accurate results, however, within the low range of particle
Solids pressure [28,19] volume fraction (< 0.2), the effect of density was clearly negligible. It
2 (dsi + dsj ) 3 was also shown that the computed results from Syamlal’s formula (Eq.
Psi = εsi ρsi θsi + 2εsi ρsi θsi ∑ j = 1 ⎡ ⎤ (1 + es, ij ) g0, ij εs, ij i = 1 or 2
⎣ 2dsi ⎦ (8)) are in good match with the simulation.
Energy equation
0 = (−Psi I + τsi): ∇usi − γsi − 3βg − si θsi i = 1 or 2 3.2.3. Solid pressure
Energy dissipation [19] The solid pressure is a measure of the momentum transfer due to the
12(1 − e 2) g0i
γsi = ρsi εsi2 θsi2 3 i = 1 or 2 streaming motion of the particles. It is an important parameter involved
di π
in the solution of the solid phase momentum and kinetic energy
equations (Eqs. (5) and (6), respectively). The widely used solid pres-
majority of literature discussing the radial distribution function has sure equation for the simple case of a monodispersed suspension was
been related to dense collisional flow, such as in bubbling fluidized beds derived from the principle of the KTGF and is given by the following
[37–40]. The extent of its influence to dilute suspension, such as in a equation: [19]
CFB, is less understood and is certainly worth investigating. Ps = εs ρs θs + 2ρs (1 + es ) εs2 g0 θs (21)
The most widely used radial distribution function for a mono-
dispersed suspension is the one originally proposed by Bagnold [41] where the first term in the right side represents the kinetic contribution
and later applied by Lun et al. [19] and Savage [39] as follows: arising from the momentum transfer due to the solid movement across a
shear layer and the second term represents the collisional contribution
1/3 −1
⎡ ε ⎤ arising from the direct solid-solid contacts. For a dilute suspension, it is
g0 = ⎢1−⎛⎜ s ⎞⎟ ⎥
ε suggested that the collisional term could be neglected as its contribu-
⎣ ⎝ max ⎠
s ,
⎦ (14)
tion to the total solid pressure becomes small [33]. However, for the
where εs, max is the maximum allowable solid volume fraction, or solid concentration within the range of < 0.2, calculations have shown
packing limit. For the case of a binary solid mixture, the radial dis- that the kinetic and collisional contributions are equally important
tribution is given by: [29], hence, both terms are retained in this study to ensure applicability
of the solid pressure equation to the range of solid concentration
g0i dsi + g0j dsj commonly observed in CFB risers.
g0, ij = i and j = 1 or 2
dsi + dji (15) For the case of a binary mixture, Gidaspow [30] modified Eq. (21)
to take care of the existence of two different particles by adding up the
where g0, i is for a selected solid phase i taking into account the effect of collisions. Accordingly, the solid pressure for a selected solid phase i in
its own class in addition to the effect of the other particle. This is given a binary mixture is given by:
by a modified Bagnold function as follows: [21].
2
(dsi + dsj ) ⎤3
⎡ εs, all ⎞
1/3 −1
⎤ 1
2
εsj Psi = εsi ρsi θsi + 2εsi ρsi θsi ∑ ⎡⎢ ⎥
(1 + es, ij ) g0, ij εs, ij i = 1 or 2
g0i = ⎢1 − ⎜⎛ + dsi ∑ i and j = 1 or 2 j=1 ⎣ 2dsi ⎦
εs, max ⎠
⎟
⎥ 2 dsj
⎣ ⎝ ⎦ j=1 (16) (22)
The maximum allowable solid volume fraction in a binary mixture, where θsi is the granular temperature of the solid phase i .
εs, max , is calculated by using the values of maximum packing for each
solid phase [19,21,42]. The total solid volume fraction, εs, all , is obtained 3.2.4. Turbulence
by the summation of the volume fraction of the two solid phases. The CFB hydrodynamics is expected to be affected by the swirl and
Iddir and Arastoopour [43] proposed an alternative formula of g0 rotation, depending on the magnitude of the mean flow Reynolds
which reduces to the following simplified form: number. In this study, the dispersed RNG k-ε model has been used to

153
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

assess the model sensitivity to such effects. Compared to other optional an average Courant number of 0.15, a cell face size of 1.0 cm and a gas
turbulence models available in FLUENT, the dispersed RNG k-ε model is velocity of 2 m/s, then the recommended time step would be
described to be more accurate and reliable for a wider class of flows Δt = 0.00075 s. Using this value as a rough guide, the choice has been
[21]. The model involves the solution of two equations calculating the made here to start the simulation with a step size of 0.0005 s. This was
turbulence kinetic energy, ket , and its dissipation, ωt , as follows: then increased to 0.001 s for the rest of the simulation once the first few
seconds converges. While this proved successful in avoiding computa-
∂
(ρ ket ) + ∇ (ρg ket ug ) = ∇ (σk μeff ∇ket ) + Gk + Gb−ρg ωt −YM tional difficulties and stiffness of the solution at the start, it also allowed
∂t g (23)
a significant reduction in the computational time without jeopardizing
∂ ω ω2 the solution accuracy.
(ρg ωt ) + ∇ (ρg ωt ug ) = ∇ (σω μeff ∇ωt ) + C1 t (Gk + C3 Gb)−C2 ρg t In specifying the boundary conditions, the bottom of the riser and
∂t ket ket
solid inlet were set at a fixed air velocity and fixed solid mass flow rate,
(24)
respectively. An auxiliary air, specified at a fixed velocity, was set at the
where Gk and Gb represent the generation of turbulence kinetic energy solid feeding to create positive pressure, hence, ensuring smooth solid
due to the velocity gradients and buoyancy, respectively. YM represents flow and preventing it from backflow. The boundary at the top of the
the contribution of the fluctuating dilatation incompressible turbulence riser was set at atmospheric pressure. At the wall boundary, the gas
to the overall dissipation rate. C1, C2 , and C3 are constants (values given phase was set at zero velocity (no-slip condition) while the solid phase
in Table 3). σk and σω are the turbulence Prandtl numbers for ket and ωt , is satisfying the widely used boundary conditions: [48]
respectively. The details of calculating the effective turbulence visc-
osity, μeff , and other parameters of the RNG k-ε model can be found in usi 3 πφ′goi ρsi εsi |usi | θsi
n. τsi + =0
ANSYS FLUENT documentarian [21]. The model sensitivity to the tur- |usi | 6εs − max (26)
bulence and comparison with the laminar flow assumption will be
3 πρsi goi εsi θsi
discussed in the result section. ksi
∂θsi
= ⎡φ′ |usi |2 − 3θsi (1 − eiw
2 ⎤
)
∂y 6εs − max μsi ⎣ 2 ⎦ (27)
4. Simulation geometry, meshing and solution procedure
where eiw is the particle-wall restitution coefficient for solid phase i and
φ′ is the specularity coefficient. The values of these parameters, as well
The simulation geometry and domain is shown in Fig. 4. In defining
as the values of the other parameters appearing in the model con-
the mesh size, the rule of thumb for a monodispersed solid mixture
stitutive equations, are given in Table 3.
suggests a minimum cell face size of around 10 times the particle dia-
Initially, the simulation domain was set to atmospheric pressure and
meter. Applying this to a binary mixture, then the face cell size would
the gas and solid phases were introduced at the specified flow condi-
be 10 times the larger size as an upper limit and 10 times the smaller
tions. For post-processing and statistical data analysis, the simulation
particle as a lower limit. This is used as guide for initial set of the mesh
results were sampled every 10 time-steps, but autosaved every 20 time-
size, however, because the computational time and solution accuracy
steps to avoid unnecessary data storage. The steady state flow condition
are both critically dependent on the number of cells, sensitivity analysis
was achieved after around 10 s of real-time simulation (corresponds to
deemed necessary in order to optimize the solution procedure and
around 24 h of CPU time). The time-averaged steady state data was
avoid the risk of losing relevant flow structures of any of the two par-
ticles in the mixture. ANSYS Workbench software was used to generate
the final simulation geometry and meshes based on the specified
minimum and maximum cell face sizes. The tetrahedral meshing
function, available as one of the options in the software, was used to
generate an unstructured element of different sizes and shapes de-
pending on the element position.
In setting the solution time step, it is important to avoid instability,
solution stiffness, and unrealistic long computational time, especially
for a complex reactive system, such as the case considered here. It is
generally understood that the stiffness and instability in solving the
partial differential equations (PDEs) can be avoided by adopting a
suitable time step and discretization scheme. The main model PDEs
given in Section 3.1 were discretized using the first order upwind im-
plicit method. To control the update of computed variables at each
iteration, the under-relaxation factors for the pressure and momentum
were set to 0.2 and 0.5, respectively. In setting the step size, the fol-
lowing simple rule of thumb was used:
Δt
Nc = ug i = 1 or 2
Δcell (25)

where Nc , Δcell , Δt and ug are the Courant number, cell face size and the
vertical component of the gas velocity, respectively. For a mono-
dispersed mixture, Cornelissen et al. [45] recommended a Courant
number within the range of 0.03–0.3 while Gobin et al. [46] re-
commended the maximum value of 0.3. For a polydispersed mixture,
Coroneo et al. [47] suggested that a Courant number within the range
of 0.028–0.15 always gives an error below 10%. However, it has to be
noted that generalization of such values should be treated with caution
since the Courant number strongly depends on the particle sizes used. Fig. 4. (a) Schematic representation of the CFB and the simulation domain (b)
Nevertheless, if assuming the range given by Corone et al. [47] is the 3D geometry of the simulation domain generated by ANSYS Workbench soft-
most appropriate (they used particles of Geldart B group) and assuming ware. All the dimensions are in centimetres.

154
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

calculated after excluding the first 5 s. computational time in the case of algebraic solution potion. This sug-
gests that the kinetic energy dissipate locally, hence the contributions of
5. Results and discussion convection and diffusion kinetic energies may be omitted in modeling a
CFB without critically losing the main hydrodynamic features of the
5.1. Model sensitivity analysis flow. The solution of the model for a binary solid mixture using the full
energy equation (Eq. (6)) has been found to be very sensitive to the
Before going into the model validity, it is important first to establish solution time-step, such that convergence can only be achieved at a
the conditions required to achieve solution independent of grid size and very small time step, and in many cases the solution fails even at a very
time step. It is also important to assess the various option of constitutive small time step of 10−6 s.
equations as well as to identify the best solution procedure for eco- Fig. 8 shows the predicted cross-sectional average granular tem-
nomic computational time and reliable results. perature and the corresponding solid concentration profiles for the two
dispersed solid phases along the riser height. The larger particle of
5.1.1. Cell size and time step 755 μm is at higher concentration and higher granular temperature
Fig. 5 shows the predicted solid concentration profile computed than the smaller particle of 360 μm. This suggests that the simplified
using three different solution time steps. The results produced using a energy equation is capable of predicting physically sound trends since,
time step of 0.1 s clearly deviate from the rest, particularly near the wall within the dilute regime, the granular temperature is proportional to
region. This is not surprising since the relatively dense wall layer is a the solid concentration [49] (θ ∝ εs2/3 ). Quantitatively, the values of the
major source of turbulence due to the expected large velocity and granular temperature obtained here are close to the data reported for
pressure gradients. This essentially requires refined time step to resolve. Geldart Group B particles in a CFB riser [49], hence giving more con-
The default time step of 0.001 s, which was decided earlier based on the fidence on the adopted solution of the energy equation.
Courant number, appear to produce results closely matching that pro-
duced at a time step of 0.01 s. Accordingly, all the results shown in this
study were produced using a time step of 0.001 s, unless otherwise 5.1.3. Turbulence
stated. Fig. 9a shows the predicted pressure profile across the riser height
Several different cell sizes have been investigated to show the model with and without incorporating the dispersed RNG k-ε turbulence
sensitivity and confirm the appropriate cell or element numbers for the model. The result clearly shows the model to be very sensitive to the
current simulation. This is demonstrated in terms of the velocity pro- turbulence. This is expected since the turbulence transfer among the
files of the wood particles at a selected operating condition as shown in phases in a binary mixture plays a more dominant role especially when
Fig. 6. The results were produced with varying the number of cells there are rapidly strained flows and swirl effects. Fig. 9b shows the
through changing the upper and lower face size limits. With the ex- variation of the solid velocity at the solid entrance region. This due to
ception of the finer cell size (total cells number of 69,433), all of the the entrance effect associated with the introduction of a high solid flux
other settings appear to show similar profiles with negligible numerical from one side of the reactor assisted with a secondary airflow to create
differences. The observed deviation at the finer cell is due to the fact positive pressure at the entrance point. This appears to cause high
that the specified lower face limit, which is 0.5, falls below the downfall velocity and non-uniform velocity distribution, which, in turn,
minimum required size (i.e. 10 times the larger particle size in the causes significant turbulence that, propagates towards the top parts of
binary mixture), therefore, the model failed to correctly resolve the the reactor. The work of Hartge et al. [51] on modeling of turbulence in
flow structure relevant to this particle size. Accordingly, in order to CFB (dilute suspension) and that of Wang et al. [52] on bubbling bed
satisfy a solution independent of the cell size for a polydisperse mixture (dense suspension), have both shown that the dispersed RNG k-ε model
the recommended rule of thumb is: produces improved predictions, hence confirming its applicability to
wider flow regimes. In the rest of this study, the turbulence model will
cell face size ≥ 10 ds, max . (28)
be used as default unless otherwise stated.
where ds, max is the diameter of the larger particle in the mixture.

5.1.2. Solution of the energy equation

The granular temperature is an important parameter required for
the description of the solid phase stresses. In the context of Eulerian-
Eulerian modeling, the granular temperature represents the solid phase
kinetic energy fluctuation due to the random motion of particles. As
discussed in Section 3.1, the model can be solved by using the full
partial differential equation (PDE) of the kinetic energy (Eq. (6)) or by
using its simplified algebraic version (Eq. (7)), which neglects the
convection and diffusion terms. In terms of the solution stability and
computation time, the experience has shown that the algebraic equa-
tion produces a stable solution at a reasonable computational time step.
This is simply because the energy equation, in this case, is solved in a
single step (i.e. non-iterative solution). On the other hand, the full ki-
netic energy for a binary mixture is stiff and the solution requires set-
ting a relatively small time step to avoid instability and complicated
divergence problem, which, in turn, results in increasing the compu-
tational time. In modeling of a CFB reactor with multi-solid phases and
added complexity of reactions and heat transfer, it is important to seek Fig. 5. The solid volume fraction of glass beads at 30 cm above the bottom of
methods to minimize the computational time. the riser demonstrating the model sensitivity to the solution time step.
To start with, Fig. 7 compares the predicted average granular tem- Simulation conditions: a Binary mixture of particles sizes ds1 = 755 μm and
perature profile along the riser height for a monodispersed system ds2 = 360 μm and equal densities of ρs = 2500 kg/m3, mixed at the ratio of
(single solid phase). Here, it is clear that both options produce similar 70:30 wt% respectively and fluidized at the air velocity of 4.6 m/s and a solid
results with negligible numerical differences, although with much less circulation rate of 27 g/s.

155
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Fig. 8. Predicted average granular temperature and the corresponding solid

concentration profiles for a polydispersed binary mixture using the simplified
Fig. 6. Vertical velocity profile of the wood particles along the riser height
algebraic solution of the energy equation. Simulation conditions: a Binary
demonstrating the model sensitivity to the cell size. Simulation condition: a
mixture of particles sizes ds = 755 μm and ds = 360 μm of equal densities of
Binary mixture of sand (ds1 = 200 μm, ρs1 = 2600) and wood (ds2 = 500 μm,
ρs = 2500 kg/m3, mixed at the ratio of 70:30 wt% respectively, and fluidized at
ρs2 = 585 kgm−3), mixed at the ratio of 70:30 wt%, respectively, and fluidized
the air velocity of 4.6 m/s and a solid circulation rate of 27 g/s.
at the air velocity of 4.6 m/s and a solid circulation rate of 27 g/s.

at the solid volume fraction 0.2, while Syamlal’s model has the
advantage of being continuous over a wide range of solid
concentration. Moreover, Almuttahar [27] suggested that Syamlal
model is of better predictive capability for the range of operating
conditions used in CFB risers. In the rest of this study, the solid-gas
momentum exchange coefficient of Syamlal (Eq. (13)) will be used as
default unless otherwise stated. The validity of this choice for a binary
mixture will be demonstrated in Section 5.3 when comparing the model
predictions with the experimental measurements.

5.1.4.2. Solid-solid momentum exchange coefficient. Fig. 11 shows a

comparison between the predicted pressure profile with and without
solid-solid momentum exchange for a binary mixture. The results
confirm the critical sensitivity of the solution to this parameter,
particularly at the bottom of the riser where the solid concentration
is high and the velocity difference between the two solid phases is
significant. This interesting, since there is a general understanding that
the momentum exchange between the solid phases in a polydispersed
dilute suspension, such as in CFBs, is negligible, especially if the solid
phases are of limited size and density differences. In a biomass gasifier,
Fig. 7. Comparision of the predicted cross-sectional average granular tem- the physical differences between the solid phases (e.g. wood particles
perature using two different solutions methods of the kinetic energy equation. and sand) is high, therefore, the incorporation of the solid-solid
Simulation conditions: single solid phase (ds = 200 μm, ρs = 2600) fluidized at
momentum exchange in the hydrodynamic model becomes essential.
the air velocity of 4.7 m/s and solid circulation rate of 36.0 g/s.
It is worth noting that Yin and Sundaresan [53,54] and Holloway et al.
[55], using Lattice-Boltzmann simulations, reported similar conclusion,
5.1.4. Constitutive equations though in different ranges of particle sizes, mixing ratios and Reynolds
5.1.4.1. Solid-gas momentum exchange coefficient (gas drag). Fig. 10 number than the ones considered here.
shows the predicted pressure profiles along the riser height using the
two solid-gas momentum exchange equations proposed by Gidaspow
[30] and Syamlal et al. [33], as given in Section 3.2.1. The simulations 5.1.4.3. Radial distribution function and solid pressure. The radial
distribution function appears in various constitutive equations, as
were carried out using single solid phases, each of different particle size
and flow conditions. The profiles on the left side of Fig. 10 are for a noted earlier, among which, is the solid pressure (Eq. (22)). It is of
interest here to quantify and assess the degree of the model sensitivity
small particle size at dilute circulation, while the ones in the right are
for a relatively larger particle size and denser circulation. It is clear that and deviations when applying any of the three radial distribution
both models produce comparable results at the dilute condition. In the functions given earlier in Section 3.2.2. This is less understood and
denser condition, the differences become well pronounced. This rarely discussed in the literature for a binary solid mixture of
suggests that these drag models behave differently with increasing considerable physical differences. Fig. 12 shows the variations of the
solid concentration. This discrepancy is understood to arise from the radial distribution function and the solid pressure for a binary solid
fact that Gidaspow’s model switches between two different drag models mixture as a function of the total solid concentration. The values
calculated by Syamlal [33] and the modified Bagnold [21] models are

156
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Fig. 9. (a) Comparision of the predicted pressure profile along the riser height demonstrating the effect of turbulence model. (b) contours of the sand velocity vectors
at the solid entrance region. Simulation conditions: a Binary mixture of particle sizes ds1 = 400 μm and ds2 = 200 μm of equal densities of ρs = 2500 kg/m3, mixed at
the ratio of 70:30 wt% respectively and fluidized at the air velocity of 3.1 m/s and solid circulation rate of 23.0 g/s.

clearly very close, however, both models behave differently at the high solid concentration this is dominated by collisions. Quantitatively, it is
concentration near the maximum packing. On the other hand, the clear that the difference in the computed solid pressure is insignificant,
modified Bagnold [21] and Iddir and Arastoopour [43] models both especially within the range of intermediate solid concentration.
appear to correctly increase exponentially towards infinity as the solid Therefore, it is concluded that the radial distribution functions of say-
concertation approached maximum packing. mala [33] and the modified Bagnold [21] function can both be used in
In a dilute suspension, the impact of the solid pressure on the overall the simulation of the CFB riser with negligible numerical differences. In
flow hydrodynamic is limited (i.e. the pressure term appearing in the the rest of this study, the saymala [33] function will be used as the
solid momentum equation, Eq. (5), will be factored by a small number). default, unless otherwise specified.
However, it is of interest to here to assess the sensitivity of the solid
pressure and the overall hydrodynamic prediction to the employed
radial distribution function. Fig. 13 shows the calculated solid pressure 5.2. Recommended constitutive and closure equations
as a function of the solid concentration using two different radial dis-
tribution functions. The curves trend is in good agreement with the data Following the above analysis and based on experience in modeling
produced experimentally by Gidaspow et al. [49] and Campbell and gas-solid flows, the recommended constitutive and closure equations
Wang [56]. Within the range of low solid concentration relevant to CFB, for modeling the hydrodynamics of a binary solid mixture in a CFB riser
the solid pressure is dominated by kinetic interactions, while in the high using the Eulerian-Eulerian KTFG model are summarized in Table 4.
This model will now be validated against experiment measurement in

Fig. 10. Comparision of the pressure profiles along the riser height at two different solid circulation rates for a single solid phase predicted by using two different gas-
solid momentum exchange models.

157
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Fig. 13. Comparison of the computed solid pressure using two different radial
distribution functions. Data calculated for a binary solid mixture consisting of
particles sizes of ds1 =755 ⎧m and ds2 =400 ⎧m, both of density = 2600 kg/
Fig. 11. Comparision of the predicted pressure profile along the riser height m3; particle-particle restitution coefficient = 0.9.
demonstrating the model sensitivity to the solid-solid momentum exchange.
Simulation condition: Binary mixture of sand (ds1 = 200 μm, ρs1 = 2600) and
wood (ds2 = 500 μm, ρs2 = 585 kgm−3), at the ratio of 85:15 wt%, respectively, 5.3.1. Pressure profile
and fluidized at the air velocity of 3.3 m/s and a solid circulation rate of 35 g/s. The model validation was first considered by comparing the pre-
dicted axial pressure profile with the experimental measurement, as
shown in Fig. 14. The profiles show classic feature of CFB risers where
the pressure gradually decreasing towards the exit. The calculate mean
deviation from the experimental measurements is 1.2% for the curve to
the left side of Fig. 14, compared to 4.5% deviation for the curve to the
right. The improved accuracy in the first case suggests that the model
has better predictive capabilities at the uniform flow of limited velocity
and pressure variations. Nevertheless, the deviations in both cases re-
main considerably small, and therefore, the model predictive cap-
abilities are deemed good enough to reveal the main features of the
flow structure in the CFB riser, as further demonstrated here.

5.3.2. PEPT measurements and particle dynamic behavior

As described earlier, the PEPT technique applied here is based on
tracing a single particle at a time within a given time and space domain.
This allows calculating the discrete particle position and velocity.
Fig. 15 shows an example of the PEPT detections of a radioactive wood
particle in a binary mixture of wood and sand during circulation be-
tween the riser and downer sides of the CFB system. The detections
Fig. 12. Comparison of the value of radial distribution functions computed show the particle axial position within the lower part of CFB riser. In
using different alternative functions. Data calculated for a binary solid mixture; Fig. 15-a it is clear that there is around 3–6 min delay between each
particles sizes of ds1 = 755 μm and ds2 = 400 μm, both of density = 2600 kg/ detection as the particle circulation completes a full cycle. As a result,
m3; particle-particle restitution coefficient = 0.9. there was around 20–40 set of detections in each 2 h run. Most inter-
esting, Fig. 15-b indicates that the particle dynamic movement in this
Section 5.3. zone is more complex than it is thought. The particle appears to travel
upward and downward and sink in the bottom denser region of the riser
as indicated by the gap in signal detection. The corresponding particle
5.3. Model validation velocity is shown in Fig. 16. Here it is shown that the particle accel-
erates and descend with the velocity ranging between 3 m/s and −2 m/
This section shows the results of model validation by comparing the s. This was not an isolated case, but a rather consistent behavior in most
experimental measurements with the hydrodynamic predictions ob- of the recorded PEPT data, especially when operating at low fluidiza-
tained by solving the main model with the recommendations given tion velocity.
earlier in Table 4. In addition, this section includes comments on some Fig. 17 shows the predicted time evolution of the particle con-
of the interesting hydrodynamic features of the CFB riser considered. It centration within the lower part of the riser. Here, the prediction shows
should be noted that the validation is limited to the range of solid the gradual increase in the solid concentration below the feeding point
circulation rate, particle size, concentration and velocities commonly as the time progresses. This is well resembling the particle dynamics at
used in a fast CFB gasifier. Caution must be exercised if the model is to the bottom of the riser as detected by the PEPT measurement and
be used for conditions beyond these regimes. hence, lending support to the model validity and capability in pre-
dicting the experimentally observed behavior. It is believed that the
particle circulation in the bottom of the riser relates to the gas swirling

158
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Fig. 14. Comparison of the experimental measurement with the model prediction of the axial pressure distribution in the CFB riser for binary solid mixtures at two
different flow conditions. The inset shows the time series data of the pressure measured at 139 cm.

and/or increased solid exchange between the dense-wall layer and the time-space averaged discrete particle velocity within the same space,
core in this region. A similar bottom bed behavior has been previously determined by the PEPT. Clearly, there is good agreement between the
reported by Grace et al. [2] when experimentally studying the wall-core measured and predicted velocity. It also observed that the sand particle
particle exchange in a fast fluidized bed reactor. Similarly, Chan et al. velocity is low and is less sensitive to the changes in the air velocity
[57], also observed descend of particles near the solid feeding point in a compared to the wood. This is due to the fact that the sand is of higher
CFB riser by using particle tracking technique. In biomass gasification, density and is more likely to sink before accelerating towards the top.
such behavior is desirable to enhance the heat exchange between the
heat carrier (sand) and the biomass, and hence positively contributes to
the rapid thermal decomposition (drying and 5.3.3. Granular temperature
devolatilization) of the fresh biomass feed at the lower part of the Fig. 19 shows the predicted granular temperatures for a binary solid
riser. mixture as a function of the solid volume fraction. This is obtained by
Fig. 18 shows a comparison between the measured and predicted recording the predicted granular temperature of each solid phase at a
solid phase velocity at the bottom of the riser. It is recognized that the central line extending from the bottom to the top of the riser, thus
Eulerian-Eulerian model provides the velocity of the dispersed solid covering a wide range of solid concentration.
phase while the PEPT data provides the discrete particle velocity. Here, The predicted values of granular temperature, as well as the curve
the comparison is made between the time-space averaged axial velocity trend, is in agreement with the published data for the particle size range
of the solid phase, determined by the Eulerian-Eulerian model, and the considered. For example, the extensive work by Gidaspow and co-
workers [18,20,49,50] on fluidized bed reactors has shown the granular

Fig. 15. Example of axial dynamic movement of a radioactive wood particle detected by the PEPT system. Experiment carried out using a mixture of 97% sand (ds =
700 μm, ρs = 2500 kg/m3) and 3% wood (ds = 1000 μm, ρs = 585 kg/m3) fluidized at 4.7 m/s and solid circulation rate of 27 g/s.

159
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

Fig. 16. Example of axial velocity of a radioactive wood particle detected by the PEPT system. Experiment carried out using a binary mixture of 97% sand (ds =
700 μm, ρs = 2500 kg/m3) and 3% wood (ds = 1000 μm, ρs = 585 kg/m3) fluidized at 4.7 m/s and solid circulation rate of 27 g/s.

Fig. 17. Contours of the evolution of solid concentration in the lower part of the
riser. Simulation condition: binary solid mixture of 97% sand (ds = 700 μm, ρs
= 2500 kg/m3) and 3% wood (ds = 1000 μm, ρs = 585 kg/m3) fluidized at
4.7 m/s and solid circulation rate of 27 g/s. The color bar represents the range Fig. 19. Predicted granular temperatures for a binary mixture using a simpli-
of the solid volume fraction. fied algebraic solution of the energy equation. Simulation condition: a Binary
mixture of glass beads consisting of 70 wt% of ds1 = 755 μm and 30 wt% of
ds2 = 360 μm) fluidized at the velocity of 4.6 m/s and a circulation rate of 27 g/
s.

which is commonly classified as kinetic flow regime, the effect of in-

creasing solid concentration on the granular temperature is similar to
the effect of compression on gas temperature, i.e. giving the propor-
tional relation θs ∝ εs2/3 . This is in good agreement with the data re-
ported by Gidaspow [30] for dilute flow. On the other hand, the de-
crease in the granular temperature at increasing solid concentration
beyond ∼0.1 is attributed to the decrease in mean free path due to high
solid collisions, this gives the inversely proportional relation θs ∝ εs−2 , as
reported by Gidapsow [30] for collissional flow. At the other extreme
end near to maximum packing (not covered in this study), the granular
temperature correctly approach zero.
In summary, there is evidence of good agreement between the
prediction and the reported data on granular temperature. Hence, the
Fig. 18. Comparison between the predicted and measured solid velocity at the solution of the granulat energy equation using the simplified algenraic
entrance region of the CFB for the case of a binary solid mixture of 97% sand (ds alternative offers a reliable approach for the simulation of a poly-
= 700 μm, ρs = 2500 kg/m3) and 3% wood (ds = 1000 μm, ρs = 585 kg/m3). dispered mixture in a CFB gasifier.

temperature of particle class Geldart B to fall within the range of

6. Conclusions
0.0001–0.5 m2/s2. It is interesting to note the increase in granular
temperature as the solid concentration increases towards ∼0.1, which
This study presented a multi-fluid (Eulerian-Eulerian) model of a
then decreases as the concentration approaches the maximum packing
binary mixture of polydisperse particles in a CFB riser. The study was
condition. Within the range of low solid volume fraction below ∼0.1,
mainly focused on identifying the appropriate constitutive/closure

160
M.I. Hassan, Y.T. Makkawi Chemical Engineering & Processing: Process Intensification 129 (2018) 148–161

equations and solution procedure with the ultimate goal of developing a Engineering Foundation, New York, 1991, pp. 75–82.
reliable tool for the simulation of biomass gasification in a CFB system. [19] C.K.K. Lun, S.B. Savage, D.J. Jeffrey, N. Chepurniy, Kinetic theories for granular
flow: inelastic particles in Couette flow and slightly inelastic particles in a general
The model was solved using ANSYS FLUENT Computational Fluid flow field, J. Fluid Mech. 140 (1984) 223–256.
Dynamic (CFD) software. [20] J. Ding, D. Gidaspow, A bubbling fluidization model using kinetic theory of gran-
ular flow, AIChE J. 36 (1990) 523–538.
The model sensitivity to the various options of constitutive equa- [21] ANSYS FLUENT 17.0 Documentation: Theory Guide, (2016).
tions and alternative method for the solution of the kinetic energy [22] L. Mazzei, Eulerian Modelling and Computational Fluid Dynamics Simulation of
equation has been assessed. The model was found to be highly sensitive Mono and Polydisperse Fluidized Suspensions, PhD thesis University College,
London, UK, 2008.
to the effect of solid-solid momentum exchange (drag) but less sensitive [23] H. Enwald, E. Peirano, A.-E. Almstedt, Eulerian two-phase flow theory applied to
to the various options of radial distribution functions investigated. The fluidization, Int. J. Multiphase Flow 22 (1996) 21–66.
[24] I. Hulme, E. Clavelle, L. van der Lee, A. Kantzas, CFD modeling and validation of
inclusion of a turbulence model proved to be crucially impotent for
bubble properties for a bubbling fluidized bed, Ind. Eng. Chem. Res. 44 (2005)
accurate predictions and resemblance of the experimentally observed 4254–4266.
solid recirculation and swirl near the solid entrance at the bottom of the [25] S. Schneiderbauer, S. Pirker, A coarse-grained two-fluid model for gas-solid flui-
dized beds, J. Comput. Multiph. Flows 6 (2014) 29–47.
reactor. The two different solution methods of the energy equation, i.e. [26] C. Lohaa, H. Chattopadhyay, P.K. Chatterjee, Assessment of drag models in simu-
full PDE and simplified algebraic equation, both produced similar re- lating bubbling fluidized bed hydrodynamics, Chem. Eng. Sci. 75 (2012) 400–407.
sults with the latter being much easier to converge with the added [27] A. Almuttahar, F. Taghipour, Computational fluid dynamics of high density circu-
lating fluidized bed riser: study of modeling parameters, Powder Technol. 185
advantage of reduced computational time. Accordingly, a complete (2008) 11–23.
hydrodynamic model has been recommended and validated against [28] Y.T. Makkawi, P.C. Wright, The voidage function and effective drag force for flui-
dized beds, Chem. Eng. Sci. Chem. Eng. Sci. 58 (2003) 2035–2051.
experimental data. In summary, the model proved to be robust and [29] Y.T. Makkawi, R. Ocone, Modelling of particle stress at the dilute-intermediate-
accurate in predicting the hydrodynamic features of polydispersed dense flow regimes: a review, KONA Powder Sci. Technol. 23 (2005) 49–63.
particles (binary mixture) in a CFB riser. [30] D. Gidaspow, Multiphase Flow and Fluidization, Academic Press, Boston, MA, 1994.
[31] S. Ergun, Fluid flow through packed columns, Chem. Eng. Prog. 48 (1952) 89.
[32] C.-Y. Wen, Y.H. Yu, Mechanics of fluidization, Chem. Eng. Prog. Symp. Ser. 62
Acknowledgment (1966) 100–111.
[33] M. Syamlal, W. Rogers, T.J. O`Brien, MFIX Documentation Theory Guide. Other
Information: PBD: Dec 1993, (1993).
The authors would like to thank the staff at the University of [34] O. Owoyemi, L. Mazzei, P. Lettieri, CFD modeling of binary-fluidized suspensions
Birmingham Nuclear Physics Research Group, with special thanks to and investigation of role of particle–particle drag on mixing and segregation, AlChE
J. 53 (2007) 1924–1940.
Prof David Parker, for the support and provision of access to their [35] M. Syamlal, The Particle–particle Drag Term in a Multiparticle Model of
Positron Emission Particle Tracking (PEPT) system. Fluidization, (1987) DOE/MC/21353-2373, NTIS/DE87006500.
[36] S. Chapman, T.G. Cowling, The Mathematical Theory of Non-uniform Gases: An
Account of the Kinetic Theory of Viscosity, Thermal Conduction and Diffusion in
References Gases, 3rd ed, Cambridge Mathematical Library, 1970.
[37] B.B. Halvorsen, An Experimental and Computational Study of Bubble Behaviour in
[1] M. Worley, J. Yale, Biomass Gasification Technology Assessment, Consolidated Fluidized Beds, PhD thesis The Norwegian University of Science and Technology
Report, Harris Group Inc., Atlanta, Georgia, 2012. (NTNU), 2005.
[2] C. Juhui, Weijie Yin, Shuai Wang, Guangbin Yu, Jiuru Li, Ting Hu, Feng Lin, [38] S. Wang, G. Liu, H. Lu, B. Yinghua, J. Ding, Y. Zhao, Prediction of radial distribution
Modelling of coal/biomass co-gasification in internal circulating fluidized bed using function of particles in a gas-solid fluidized bed using discrete hard-sphere model,
kinetic theory of granular mixture, Energy Convers. Manage. 148 (2017) 506–516. Ind. Eng. Chem. Res. 48 (2009) 1343–1352.
[3] M. Bashir, X. Yu, M. Hassan, Y. Makkawi, Modelling and performance analysis of [39] S.B. Savage, Streaming motions in a bed of vibrationlly fluidized dry granular
biomass fast pyrolysis in a solar-thermal reactor, ACS Sustain. Chem. Eng. 5 (2017) material, J. Fluid Mech. 194 (1988) 457.
3795–3807. [40] B. Van Wachem, J.C. Schouten, C.M. van den Bleek, Krishna, Comparative analysis
[4] X.T. Lia, J.R. Gracea, C.J. Lima, A.P. Watkinsona, H.P. Chenb, J.R. Kim, Biomass of CFD models of dense gas–solid systems, AlChE J. 47 (2001) 1035–1051.
gasification in a circulating fluidized bed, Biomass Bioenergy 26 (2004) 171–193. [41] R.A. Bagnold, Experiments on a gravity free dispersion of large solid spheres in a
[5] P. Parthasarathy, K.S. Narayanan, Hydrogen production from steam gasification of Newtonian fluid under shear, Proc. R. Soc. Lond. A225 (1954) 49.
biomass: influence of process parameters on hydrogen yield- a review, Renew. [42] R. Fedors, R.F. Landell, An empirical method of estimating the void fraction in
Energy 66 (2014) 570–579. mixtures of uniform particles of different size, Powder Technol. 23 (1979) 225–231.
[6] L. Wei, S. Xu, L. Zhang, C. Liu, H. Zhu, S. Liu, Steam gasification of biomass for [43] H. Iddir, H. Arastoopour, Modeling of multi-type particle flow using the kinetic
hydrogen-rich gas in a free-fall reactor, Int. J. Hydrogen Energy 32 (2007) 24–31. theory approach, AlChE J. 51 (2005) 1620–1632.
[7] X. Ku, Tian Li, Terese Lovas, CFD–DEM simulation of biomass gasification with [44] J.L. Lebowitz, Exact solution of generalized percus-yevick equation for a mixture of
steam in a fluidized bed reactor, Chem. Eng. Sci. 122 (2015) 270–283. Hard spheres, Phys. Rev. A 133 (1964) 895.
[8] A. Gel, M. Shahnam, J. Musser, A.K. Subramaniyan, Jean-Francois Dietiker, [45] J.T. Cornelissen, F. Taghipour, R. Escudié, N. Ellis, J.R. Grace, CFD modelling of a
Nonintrusive uncertainty quantification of computational fluid dynamics simula- liquid-solid fluidized bed, Chem. Eng. Sci. 62 (2007) 6334–6348.
tions of a bench-scale fluidized-bed gasifier, Ind. Eng. Chem. Res. 55 (2016) [46] A. Gobin, H. Neau, O. Simonin, J.-R. Llinas, V. Reiling, J.-L.C. Selo, Fluid dynamic
12477–12490. numerical simulation of a gas phase polymerization reactor, Int. J. Numer. Methods
[9] M. Anil, S. Rupesh, C. Muraleedharan, P. Arun, Performance evaluation of fluidised Fluids 43 (2003) 1199–1220.
bed biomass gasifier using CFD, Energy Procedia 90 (2016) 154–162. [47] M. Coroneo, L. Mazzei, P. Lettieri, A. Paglianti, G. Montante, CFD prediction of
[10] L. Yan, C.J. Lim, G. Yue, B. He, J.R. Grace, Simulation of biomass-steam gasification segregating fluidized bidisperse mixtures of particles differing in size and density in
in fluidized bed reactors: model setup, comparisons and preliminary predictions, gas-solid fluidized beds, Chem. Eng. Sci. 66 (2011) 2317–2327.
Bioresour. Technol. 221 (2016) 625–635. [48] P.C. Johnson, R. Jackson, Frictional–collisional constitutive relations for granular
[11] S. Kraft, F. Kirnbauer, H. Hofbauer, CPFD simulations of an industrial-sized dual materials, with application to plane shearing, J. Fluid Mech. 176 (1987) 67–93.
fluidized bed steam gasification system of biomass with 8 MW fuel input, Appl. [49] D. Gidaspow, J. Jung, R.K. Singh, Hydrodynamics of fluidization using kinetic
Energy 190 (2017) 408–420. theory: an emerging paradigm 2002 flour-daniel lecture, Powder Technol. 148
[12] H. Liu, R.J. Cattolica, R. Seiser, CFD studies on biomass gasification in a pilot-scale (2004) 123–141.
dual fluidized-bed system, Int. J. Hydrogen Energy 41 (2016) 11974–11989. [50] M. Tartan, D. Gidaspow, Measurement of granular temperature and stresses in ri-
[13] H. Liu, R.J. Cattolica, R. Seiser, Operating parameter effects on the solids circulation sers, AIChE 1760 (2004).
rate in the CFD simulation of a dual fluidized-bed gasification system, Chem. Eng. [51] E.U. Hartge, L. Ratschow, R. Wischnewski, J. Werther, CFD-simulation of a circu-
Sci. 169 (2017) 235–245. lating fluidized bed riser, Particuology 7 (2009) 283–296.
[14] D.J. Parker, R.N. Forster, P. Fowles, P.S. Takhar, Positron emission particle tracking [52] Q. Wang, J. Lu, W. Yin, H. Yang, L. Wei, Numerical study of gas-solid flow in a coal
using the new Birmingham positron camera, Nucl. Instrum. Methods A477 (2002) beneficiation fluidized bed using kinetic theory of granular flow, Fuel Process.
540–545. Technol. 111 (2013) 29–41.
[15] D.J. Parker, C.J. Broadbent, P. Fowles, M.R. Hawkesworth, P. McNei, Positron [53] X. Yin, S. Sundaresan, Drag law for bidisperse gas−solid suspensions containing
emission particle tracking- a technique for studying flow within engineering equally sized spheres, Ind. Eng. Chem. Res. 48 (2008) 227–241.
equipment, Nucl. Instrum. Methods Phys. Res. Sect. A 326 (1993) 592–607. [54] X. Yin, S. Sundaresan, Fluid-particle drag in low-reynolds-number polydisperse gas-
[16] J.A. Seville, Single particle view of fluidization, 13th International Conference on solid suspensions, AlChE J. 55 (2009) 1352–1368.
Fluidization - New Paradigm in Fluidization Engineering, (2010). [55] W. Holloway, X. Yin, S. Sundaresan, Fluid-particle drag in inertial polydisperse
[17] J. Seville, A. Ingram, X. Fan, D.J. Parker, Positron emission imaging in chemical gas–solid suspensions, AlChE J. 56 (8) (2010) 1194–2004.
engineering, in: J. Li (Ed.), Advances in Chemical Engineering, 37 Elsevier, [56] C.S. Campbell, D.G. Wang, A particle pressure transducer suitable for use in gas-
Amsterdam, 2009. fluidized beds, Meas. Sci. Technol 1 (1990) 1275–1279.
[18] D. Gidaspow, R. Bezburuah, J. Ding, Hydrodynamics of circulating fluidized beds, [57] C. Chan, J.P.K. Seville, J. Baeyens, The solids flow in the riser of a CFB viewed by
kinetic theory approach, in: O.E. Potter, D.J. Nicklin (Eds.), Fluidization VII, positron emission particle tracking (PEPT), Proceedings of the 13th International
Proceedings of the 7th Engineering Foundation Conference on Fluidization, Conference on Fluidization- New Paradigm in Fluidization Engineering, (2010).

161