Graphical Abstract

Wurster Fluidized Bed Coating: Coarse-Graining Technique within CFD-

DEM in Conjunction with Heat and Mass Transfer

Philip Kjaer Jepsen, Luis Martin De Juan, Srdjan Sasic

Highlights
Wurster Fluidized Bed Coating: Coarse-Graining Technique within CFD-
DEM in Conjunction with Heat and Mass Transfer

Philip Kjaer Jepsen, Luis Martin De Juan, Srdjan Sasic

• Logarithmic simulation speed-up in CFD-DEM via coarse-graining (parti-

cle scaling).

• Modelling of solid, gas, and liquid phases in the Wurster fluidized bed coat-
ing process.

• Evaluation of granular flow rheology, with analysis of inter-phase momen-

tum, heat, and mass transfer.

• Experimental validation of CFD-DEM predictions with PEPT data.

• Contrasting CFD-DEM simulations and the effects of progressive coarse-

graining.
 Wurster Fluidized Bed Coating: Coarse-Graining
Technique within CFD-DEM in Conjunction with Heat
and Mass Transfer
Philip Kjaer Jepsena , Luis Martin De Juanc , Srdjan Sasic1
a
Department One, Address One, City One, 00000, State One, Country One
b
Department Two, Address Two, City Two, 22222, State Two, Country Two
c
Department Two, Address Two, City Two, 22222, State Two, Country Two

Abstract

In this study, we assess predictive modelling in Wurster fluidised bed coaters us-
ing Computational Fluid Dynamics-Discrete Element Method (CFD-DEM) with
a coarse-graining technique. We conducted simulations of the full three-phase
system for the original and three coarse-grained systems, analysing velocity dis-
tributions, macroscopic solid stresses, coating uniformity, and phase temperatures.
By aggregating 8, 27, and 64 original particles, we reduced the particle count from
5.1 million to 80,000, achieving a logarithmic increase in simulation speed while
maintaining fidelity to the original CFD-DEM system. Our findings indicate that
coarser grains, representing up to 64 original particles, provide a practical sim-
plification for CFD-DEM simulations in Wurster coaters, thereby broadening its
applicability to larger systems currently limited by computational capabilities.
Keywords: CFD-DEM, Three-phase system, Particle scaling, Coarse-graining,
Wurster fluidized bed coater, spray coating
PACS: 0000, 1111
2000 MSC: 0000, 1111

Preprint submitted to Powder Technology January 17, 2024

Nomenclature H enthalpy, J
h heat transfer coefficient, W/(m2 K)
Abbreviations hc mass transfer coefficient, m/s
CT cycle time I Inertial number
MCC microcrystalline cellulose Ii moment of inertia, kg/m2
PCM particle centroid method kn Hertz stiffness coefficient, Pa m
PEPT positron emission particle track- l branch vector, m
ing m mass, kg
RT residence time Mv vapour molar mass, g/mol
Greek Symbols Nu Nusselt number
α voidage p pressure, Pa
ωi angular velocity, 1/s Qi heat transfer, J
τ shear stress tensor, N/m 2 R radius, m
λ thermal conductivity, m2/s Ra universal gas constant, J/(K mol)
µ fluid viscosity, Pa s Re Reynolds number
µ s particle friction coefficient S h Sherwood number
ϕ relative humidity T temperature, K
ψ scaling factor t time, s
ρ density, kg/m 3 V volume, m3
ε coefficient of restitution Yv vapour mass fraction
Latin Symbols Subscript
′ effective
n̂i j normal unit vector
t̂ i j tangential unit vector cg, o coarse grain, original particle
Fi force, N i particle index
g gravity, m/s 2 n, t normal, tangential component
ui particle velocity, m/s o original particle
v gas velocity, m/s s solid
xi particle position, m v vapour
A area, m 2

Cd drag force coefficient δ relative displacement, m

c p specific heat capacity, J/(kg K) γ̇ s solids shear strain tensor
D diffusion coefficient, m /s 2 ηn viscous damping coefficient
d diameter, m γ damping ratio
E Young’s modulus, Pa ν Poisson ratio

2
1. Introduction

Fluidized bed spray coating of particles, pellets or granules is ubiquitous in indus-

tries such as pharmaceuticals, chemicals, and food production, serving distinct
functional objectives such as modified release, masking of sensory attributes, and
extension of shelf life, depending on the specific application [1]. A type of flu-
idized bed is the Wurster coater, known for its efficiency in producing high-quality
films with minimal defects [2, 3]. Within the array of fluidised bed coaters, the
Wurster coater is distinguished for its proficiency in fabricating high-quality films
with minimal defects [2, 3]. This proficiency is attributable to its unique air-
flow and geometry, which correlate to produce systematic, cyclic trajectories of
particles [4, 5]. In this apparatus, particles are conveyed from the bed to the
Wurster tube via atomizing and fluidizing airflows for spraying. Subsequently,
they advance to the drying region for solvent evaporation and film layer forma-
tion, traversing the three regions illustrated in Figure 1. To capitalize the benefits

Figure 1: Schematic outlining the characteristic regions in the Wurster coater system: 1) the bed
region, 2) tube region, and 3) drying region.

of this technology, design, optimisation, and risk mitigation are crucial, address-

3
ing challenges like film degradation and particle agglomeration [6]. Advanced
monitoring methods in laboratory-scale Wurster coaters have been used to study
process variables and geometric effects [2, 7, 8]. However, challenges in particle
enumeration and spatial resolution, along with the elevated costs and complexity
for equivalent industrial-scale data accuracy, limit their use [9, 10].

In the preceding decade, advancements in computational capabilities have ex-

panded the utility of predictive modelling for multiphase gas-solid systems, com-
mon in industrial applications [5, 11, 12]. Two main numerical frameworks are
Euler-Euler [13, 14], which conceptualises both the solid and fluid phase as in-
terpenetrating continua, and Euler-Lagrange, known as CFD-DEM, where parti-
cles are tracked individually as discrete elements in a Lagrangian reference sys-
tem [15, 16]. In the Euler-Euler framework, the solid phase is modelled via the
kinetic theory of gases, assuming instantaneous collisions and uncorrelated ve-
locities (molecular chaos) [17]. Conversely, CFD-DEM simulates particle inter-
actions as deformable mechanical elements, capable of accommodating multiple
sustained contacts [18].

Selecting an appropriate numerical framework necessitates comprehension of the

intrinsic model assumptions and their pertinence across distinct granular flow
regimes. The observed behaviour varies significantly within these regimes, char-
acterised by the complex ability of particles to flow collectively in manners remi-
niscent of solids, liquids, or gases, dependent on the extent of packing and agita-
tion [19]. In very dense regions, where particles undergo slow shear and frictional
stresses arise from prolonged, multiple contacts, a quasi-static flow regime is dis-
cernible [20, 21]. The inertial regime, characterised by increased particle inertia

4
(strong agitation), leads to erratic motion predominantly governed by collisional
stresses. However, the extent of frictional interactions in this regime is depen-
dent on solid concentration: in a dilute inertial regime, collisional rapid flow ex-
hibits gas-like characteristics, whereas, in a dense inertial regime, it transitions to
a liquid-like flow where frictional contacts are influential [22–24].

The efficacy of CFD-DEM in diverse fluidized bed systems, including Wurster

coaters, is widely recognized [4, 25, 26]. While the computational demands of
tracking individual particles predominantly confine its use to laboratory-scale sys-
tems, improvements in parallel computing and DEM algorithms have broadened
its scope, facilitating simulations in fluidized bed systems comprising up to 25
million particles [12, 27]. However, industrial processes with billions of particles
remain prohibitively expensive. As a countermeasure, simplification strategies
encompassing particle scaling—identified as coarse-graining techniques—have
been extensively utilized. In practice, this entails the aggregating of original par-
ticles into coarser grains while preserving the physical behaviour of the original
particle system [28]. Following the classification by Washino [29], two principal
coarse-graining methods are identified: 1) direct force scaling, which proportion-
ally adjusts forces on each coarse grain to their original counterparts, conforming
to similarity laws [30, 31]; and 2) parameter scaling, wherein the adjustment of
contact forces is explicitly dictated by the selected contact model and the scaling
laws elucidated through dimensional analysis [32–35].

Progress in developing and validating scaling techniques for Wurster coater sys-
tems remains constrained, with only a few foundational studies in which direct
force scaling has been used [36–38]. This paper sets itself apart from earlier stud-

5
ies in two critical respects: firstly, whereas previous studies primarily employed
scaling for computational efficiency, this research presents a comparative exami-
nation of both the unscaled and scaled systems; secondly, in contrast to the earlier
studies with focus on direct force scaling, this study delves into the application of
parameter scaling.

2. Methodology

2.1. Continuous-phase modelling

Commencing with the volume-averaged equations for the conservation of mass

and momentum in the gas phase:

∂
αρ + ∇ · αρv = 0 (1)
∂t
∂
αρv + ∇ · αρvv = −α∇p + ∇ · ατ + αρg + Sm (2)
∂t

In the equations, α, ρ, v, and Sm denote the voidage, density, velocity, and in-
terphase momentum exchange of the gas, respectively. The shear stress tensor is
defined by τ = µ′ (∇v + ∇vT − 23 ∇ · vI), where I is the identity matrix and µ′ is
the effective viscosity, determined from the realisable k-ε turbulence model. The
realisable version extends the standard model by introducing corrections to mit-
igate its tendency to overestimate turbulent viscosity and excessive dissipation,
culminating in improved jet spreading [39].

The presence of humidity in the gas phase necessitates concurrently solving the
energy conservation equation for the gas with an additional mass conservation

6
equation for water vapour:

∂
αρc p T + ∇ · αρvc p T = ∇ · (αλ∇T + ρD′v ∇Yv Hv ) + S h (3)
∂t
∂
αρYv + ∇ · αρvYv = ∇ · αρD′v ∇Yv + S v (4)
∂t

where c p , T, λ, D′v , Yv , and Hv represent the specific heat capacity, gas temper-
ature, thermal conductivity, effective mass diffusivity, vapour mass fraction, and
the water vapour enthalpy, respectively. The source term, S h , denotes the trans-
fer of convective heat from the particles to the gas, while S v represents the mass
flux of vapour from wet particles undergoing evaporation. For a detailed mesh
sensitivity analysis, refer to Appendix A.

2.2. Inter-phase coupling

The fluids conservation equations 1-3 are coupled with the solid phase in each
computational cell Vc , by the presence of P particles via voidage and source terms
Sm , S h , S v . Voidage is computed via the particle centroid method (PCM), as
αc = 1− i=1 Vi /Vc , based on particles with their centroids located in the cell. Nev-
PP

ertheless, PCM’s ability to conserve quantities is compromised in highly dense

systems or when Vi ∼ Vc [40]. This limitation is critical in scenarios that require
a fine mesh for accurate resolution of fluid phase turbulence. To address this, we
employ a diffusion-based coupling method when αc < 0.4, implementing a diffu-
sion equation of the form α̇c = −D∇2 αc across all domain cells. This equation,
initially computed using PCM cell values, is iteratively solved while enforcing
a no-flux boundary condition until αc ≥ 0.4 [41, 42]. For details regarding the
numerical implementation, refer to Appendix B.

The momentum exchange term, S m , is evaluated from the hydrodynamic forces

7
F f,i exerted on each particle, as follows:
P P
1 X 1 X 1 
Sm = − F f,i = Vi ∇p − ρA′i Cd (v − ui )|v − ui | (5)
Vc i=1 Vc i=1 2

In the equation, Vi , A′i , and ui denote the volume, projected area, and velocity of
the ith particle, respectively. The Gidaspow drag model is chosen to calculate the
drag coefficient Cd = f (α, Rei ) as justified by the comparative analysis by Li et
al. [15], which examined different drag models within the Wurster coater system.
The second source term, S h , found in Equation 3, encapsulates the convective
inter-phase heat exchange between the solids and the gas, expressed as:
P P
1 X 1 X
Sh = − Q f,i = hAi (T i − T ) (6)
Vc i=1 Vc i=1

In this formulation, Q f,i denotes the convective heat flux from the ith particle,
with Ai and T i serving as the surface area and temperature, respectively, of the
particle’s surface, while h represents the convective heat transfer coefficient. The
third source term, S v , is the sum of water evaporated from the set of particles
present within the computational cell, evaluated as:
P P
1 X hc Ai Mv ϕpv pv,i
" #
1 X
Sv = ṁv,i = − (7)
Vc i=1 Vc i=1 Ra T Ti

In the provided expression, ṁv,i signifies the rate of evaporation from the ith parti-
cle. Meanwhile, hc , Mv , Ra , and ϕ stand for the mass transfer coefficient, vapour
molar mass, gas constant, and localized relative humidity, respectively. In Equa-
tion 7, the vapour concentration difference at the particle surface relative to the
bulk phase is evaluated using an ideal gas law. Furthermore, saturation at the par-
ticle surface is assumed, enabling the estimation of its partial pressure pv using an
Antoine equation. The heat and mass transfer coefficients in Equations 6 and 7 are

8
determined using the Sherwood and Nusselt numbers, defined as S h = hc di /Dv
and Nu = hdi /λ respectively, through Gunn’s correlation in packed and fluidized
beds [43].

2.3. Discrete-phase modelling

In the Lagrangian framework, the translational and angular velocities of the ith
particle, denoted as ui and ωi and possessing a mass mi and moment of inertia Ii ,
are ascertained through the following governing equations:
J
dui X
mi = mi g + F f,i + (Fn,i j + Ft,i j ) (8)
dt j=1
J
dωi X
Ii = Ri n̂i j × Ft,i j (9)
dt j=1

where Fn,i j and Ft,i j denote the normal and tangential contact forces, respectively,
emerging from interactions with J other particles or boundaries, conforming to
a soft-sphere formulation, indicated by the overlap, which for two identical par-
ticles, occurs when δn = di − (xi − x j ) · n̂i j > 0, with the normal unit vector
n̂i j = (xi − x j )/|xi − x j | [16]. As seen from equation 9, the rotational motion is
produced by collisions, as the torques generated are proportionally related to the
tangential contact force. The normal force is defined per Hertz theory, augmented
by Tsuji et al.’s viscous modifications while for the tangential component, the
Mindlin-Deresiewicz model is applied, as detailed below [44–46]:

Fn,i j = −kn δn n̂i j − ηn vn n̂i j (10)

s
  6m′ µ s |Fn,i j | 1/4
Ft,i j = −µ s |Fn,i j | 1 − ξ3/2 t̂ i j − γt ξ vt t̂ i j (11)
δt,max

9
 √
In equation 10, the stiffness and damping coefficients are defined as kn = 43 E ′ R′ δn
√
and ηn = γn 5m′ kn , respectively with γn being the damping ratio empirically as-
certained using Schwager and Pöschel’s method [47]. The variables vn and vt , are
the normal and tangential components of the relative contact velocity and t̂ i j the
tangential unit vector aligned with the latter. The tangential component, given by
equation 11, describes both elastic and frictional attributes through sticking and
sliding, conditioned by a maximum tangential deformation related to the normal
overlap, as described by δt,max = (0.5−νi )/(1−νi )µ s δn with µ s denoting the friction
coefficient. A specified factor, ξ = 1 − min(|δt |, δt,max )/δt,max , triggers sliding when
δt > δt,max and causes the tangential force component, given by Equation 11, to
p
conform to Coulomb’s law. The term γt = ln ε/ ln2 ε + π2 acts as the tangential
damping ratio, akin to γn .

The thermal energy balance for solids, as expressed in Equation 12, is predicated
on the assumptions of a small Biot number and large Péclet numbers, rendering
internal gradients and conductive heat transfer through contacts negligible [48].
dT i
mi c p,i = Q f,i + ṁv,i H f g,i (12)
dt
in the equation, c p,i denotes the specific heat capacity of the solids, and H f g,i the
latent heat of evaporation at the particle temperature. For the liquid injections, a
static spray zone is established above the nozzle with liquid spray being continu-
ously administered at each time step through particle enumeration and uniformly
distributing the liquid, ensuring an even allocation of liquid volume to each parti-
cle [37, 49].

10
2.4. Coarse-graining technique

In our study, we apply a coarse-graining technique to formulate scaling laws as

dimensionless groups Πk , derived via dimensional analysis of the specific con-
tact models, following the established methodologies referenced in [32–35]. The
foundation of this derivation lies in the foremost dimensionless group Π1 = Rcg /Ro ,
which signifies geometric similarity through the relationship between the radii of
coarse-grained and original particles, denoted as Rcg and Ro , respectively. This
group is subsequently recognized as the scaling factor ψ, which serves to quan-
tify the level of coarse-graining by the number of original particles combined into
each coarse-grained unit, as represented by ψ3 . The selection of ψ must therefore
take into consideration the geometric constraints of the enclosing system, ensur-
ing that the particle size exerts minimal impact on the overall flow behaviour of
the bulk material, thus facilitating optimal implementation.

To elucidate the method of identifying dimensionless groups and deriving scaling

laws, our analysis commences with a generalized version of the Hertzian spring-
dashpot model, expressed as follows:

m′ an = kn δn + ηn vn (13)

In this equation, an represents the relative acceleration arising from the rela-
tive displacement during particle contact. For the formulation of dimensionless
groups, we introduce the following set of dimensionless variables for mass, ac-
celeration, overlap, and displacement velocity: m̂ = m′ /(ρi R′3 ), â = an R′ /(E ′ /ρi ),
δ̂ = δn /R′ , and v̂ = vn / E ′ /ρi . Given that the scaling of relative displacement
p

δn ∝ ψ, its dimensionless form, δ̂, remains constant. Substituting these dimension-

less variables into Equation 13 leads to the dimensionless form of the equation:

11
 ηn
! !
kn
m̂â = ′ ′ δ̂ + ′2 √ ′ v̂ (14)
RE R ρE
| {z } | {z i }
Π2 Π3

The dimensionless groups, Π2 and Π3 , represent the scaling laws which, per the
dynamic similarity principles, are required to remain invariant. By maintaining
constant values for particle density and Young’s modulus, Equation 14 delineates
the precise interrelation between the original and scaled systems in terms of these
dimensionless parameters.
kn,cg kn,o ηn,cg ηn,o
= , = 2 (15)
Rcg Ro R2cg Ro
If the coefficient of restitution is held constant, the energy dissipated in binary
collisions within the scaled system is consistent with ψ3 collisions in the original
particle system. Furthermore, due to the size-independent packing of spheres, the
voidage remains invariant under scaling [28].

As per the definitions of the stiffness and damping coefficients, we observe: kn,cg ∝
ψkn,o and ηn,cg ∝ ψ2 ηn,o . Hence, Π2 and Π3 remain unaffected by the scaling. Al-
though, the relative scaling of the normal force will adhere to Fn,cg /Fn,o ∝ ψ2 , and
an examination of the tangential component reveals analogous scaling behaviour,
provided that the solids friction coefficient, µ s , remains constant. Equation 15 is
in agreement with Chialvo et al.’s findings [50], indicating invariant stresses in
very dense regimes, exemplified by p ∝ kn /Ri . This is substantiated by analysing
the particle stress tensor, as outlined in Equation 16, which is independent of ψ
and thus unaffected by the scaling approach [33, 34].
1 X 
σi = − mi (ui − ū) ⊗ (ui − ū) + Fc,i j ⊗ lc,i j (16)
Vi j

12
In Equation 16, ū is the local mean velocity of the particles, while lc,i j represents
the branch vector situated between the centres of mass of the colliding particles.

With scaling laws established, including constants ρi , E ′ , ε, and µ s , the focus shifts
to the remaining terms in Equations 8 and 12. According to similarity principles,
Fi ∝ ψ3 and Qi ∝ ψ3 are requisite. Given Vi ∝ ψ3 and Ai ∝ ψ2 , adjustments are
necessary only in the drag force, convective, and evaporative heat transfer terms.
The involved coefficients Cd , h, and hc are determined via empirical correlations,
which are theoretically unfounded in scaled analogies. This issue is addressed by
employing the original particle radius in these computations, ensuring consistency
in Reynolds, Nusselt, and Sherwood numbers under similar conditions. Conse-
quently, a ψ factor results in similarity in each of these terms. Furthermore, the
DEM time step, defined as 20% of the Rayleigh time, increases linearly: ∆ti ∝ Ri ,
thereby enhancing stability and permitting larger steps.

2.5. Assessment of rheology in granular flows

In rheological studies of granular flows, determining specific flow regimes and
their transitional phases involves contrasting the macroscopic shear deformation
q
timescale, 1/γ̇ s , with the particles’ inertial timescale, ρi di2 /p s [17, 22, 23]. This
comparison yields a dimensionless parameter, the inertial number, as given in
Equation 17, which assesses whether the granular pressure p s is sufficient for con-
solidating the particles or if the magnitude of shear strain rate, γ̇ s , is sufficiently
high to cause dilation.
di γ̇ s
I= p (17)
p s /ρi
In simulations and experiments in various geometries, three different flow regimes
are observed based on the relationship between the inertial number and the effec-

13
tive friction coefficient µ′s = |τ s |/p s , which quantifies the macroscopic solid shear
stress [24]. The quasi-static flow regime is characterised by its stress-rate indepen-
dence, as indicated by a constant µ′s , which has been identified at I ≲ 10−3 − 10−2 .
With an increased rate and inertia becoming influential, a rate-dependent pattern
emerges, seen by a monotonic increase in µ′s with I, defining the dense inertial
flow regime. The transition into collisional flow, consistent with kinetic theory
postulations, is observed at I ≳ 10−1 − 100 , where µ′s exhibits a plateau [14, 50].
Further details on deriving macroscopic granular continuum-equivalent properties
from discrete particle data are presented in Appendix C.

14
2.6. Simulation setups

2.6.1. Experimental validation of numerical framework

In the current study, a Rocky-Fluent coupled solver is deployed within a paral-
lel architecture, executing the CFD on 26 CPU cores and the DEM on a single
NVIDIA Tesla V100 32GB GPU. For the validation of the CFD-DEM frame-
work, data from Positron Emission Particle Tracking (PEPT) experiments con-
ducted by Li et al. in a STREA-1TM Wurster coater are utilised. In their ex-
periments, selected microcrystalline cellulose (MCC) pellets were tagged with
Fluorine-18 to monitor their trajectories over an extended period. This provided
data on the particle velocity, residence time (RT) per pass, and cycle time (CT)
distribution [2]. The material properties of the MCC pellets utilised in the sim-
ulations are presented in Table 1. Details on the system specific and dimensions

Table 1: Physical and mechanical properties of the simulated MCC pellets.

Physical Property Numerical Value

Particle diameter 1749 µm
Particle density 1420 kg/m3
Young’s modulus of particles 1 MPa
Poisson’s ratio of particles 0.30
Poisson’s ratio of walls 0.33
Particle-particle friction coefficient 0.53
Particle-wall friction coefficient 0.20
Particle-particle restitution coefficient 0.83
Particle-wall restitution coefficient 0.80

are provided in [2], and simulation-specific parameters are found in the subse-
quent paper [3], by Li et al., where they employed CFD-DEM to yield results
comparable to the experiment, using the MultiFlow solver [3, 15]. Furthermore,
Böhlinger et al. corroborated their CFD-DEM framework using the XPS/AVL-

15
Fire solver [9], offering dual reference points for a comparative assessment of our
CFD-DEM framework with the current solver. Consequently, we will juxtapose
our simulated outcomes with both the experimental data reported in [2] and the
simulated results of the aforementioned CFD-DEM studies [3, 9, 15].

2.6.2. System adaptation and coarse-graining design

Building on the system settings outlined in the previous section, this section incor-
porates three levels of coarse-graining (ψ = 2, 3, 4), in addition to the original par-
ticle system (ψ = 1), achieved by grouping sets of 8, 27, and 64 original particles
into coarse grains. For an accurate evaluation, we aim to minimize the influence
of the coarse grain size on bulk properties. Accordingly, the original particle size
is determined by ensuring the diameter of the coarsest grains (ψ = 4) does not
exceed one-tenth of the partition gap. This is the narrowest gap in the geometry,
defined as the distance between the tube and the distributor plate. Following this
criterion, the diameter of the coarsest grains is 1500 µm, corresponding to original
particles of 375 µm diameter.

Contrary to the Geldart D categorized particles used for validation, these particles
fall under the Geldart type B category, necessitating corresponding adjustments
in the airflow. To minimize excessive particle collisions at the outlet boundary,
the velocity of the atomizing gas is reduced by 5 m/s, and the fluidizing gas flow
is adjusted per Ergun’s equation to preserve a consistent pressure drop across the
bed. The updated gas-phase inlet conditions, including air humidity, are presented
in Table 2, alongside the modelling parameters for the spray zone and the desig-
nated coarse-graining sequence. Heat losses to the environment, due to an am-
bient temperature of 20 ◦ C, are quantified with a wall heat transfer coefficient

16
Table 2: Simulation parameters and their change during coarse-graining, CFD boundary condi-
tions, and spray zone modelling parameters.

Framework Numerical value Units

DEM
Scaling factor 1, 2, 3, 4 -
Particle diameter 375, 750, 1125, 1500 µm
Particle count 5.1, 0.64, 0.19, 0.08 M
Time step 7.7, 15.5, 22.3, 30.1 µs
CFD
Time step 1 × 10−4 s
Fluidizing gas inlet:
Velocity inner annulus 2.01 m/s
Velocity outer annulus 0.41 m/s
◦
Temperature 50 C
Relative humidity 6 %
Atomizing gas inlet:
Velocity 45 m/s
◦
Temperature 20 C
Relative humidity 6 %
Spray
Liquid spray rate 10 g/min
Water density 998 kg/m3

of 30 W/(m2 K), following the approach outlined in [51] for a Glatt GPCG-2 6′′
Wurster coater geometry. The specific heat capacity of the particles is taken as
200 J/(kg K), as reported for MCC pellets in [5].

17
3. Results and discussion

3.1. Experimental validation of CFD-DEM framework

Figure 2 compares the time-averaged vertical particle velocities at a height of 90
mm above the distributor, encompassing both simulation and experimental results.
The alignment between the simulated and the experimental particle velocities is
evident. The negative velocity at the interior tube wall quantifies the prevalence
of particle circulations within the tube [2].
2.5
Wurster tube wall

Experiment PEPT, [2]

Vertical particle velocity (m/s)

2
Simulation (MultiFlow), [3]
1.5 Simulation (XPS/AVL-Fire), [9]
Simulation (current work)
1
0.5
0
-0.5
-1
-1.5
-2
0 10 20 30 40 50 60 70 80
Radial distance from center (mm)

Figure 2: Vertical particle velocity profile, 90 mm above the distributor plate, including the exper-
imental values from PEPT, simulated results in the current study, and the two reference papers.

Figure 3 presents the average particle RTs per cycle in each region. The RTs
within the tube, as simulated in the two referenced papers, are underestimated,
whereas the current study overestimates them, albeit with the smallest absolute
error. In the bed region, our simulated RTs closely align with the experimental
data, avoiding the underestimation evident in the other simulations.
Figure 4 shows the CTDs derived from recorded particle cycles shorter than 25 s.
A consistent trend across the simulations is evident, where cycles shorter than 5 s

18
 2 7
Experiment PEPT, [2]
6.5

Residence time in the bed region (s)

1.75 Simulation (MultiFlow), [3]
Simulation (XPS/AVL-Fire), [9] 6
1.5 Simulation (current work)
Residence time (s)
5.5
1.25 5
1 4.5

0.75 4
3.5
0.5
3
0.25 2.5
0 2
Drying region Tube region Bed region

Figure 3: Residence times, per cycle, within the distinct zones of the Wurster coater from the
PEPT experiment [2] and simulations from current and reference papers [3, 9].

are overestimated, while longer cycles are underestimated. In our study, the RTs
16
Experiment PEPT, [2]
Percentage of cycles (%)

14 Simulation (MultiFlow), [3]

12 Simulation (XPS/AVL-Fire), [9]
Simulation (current work)
10

0
0 4 8 12 16 20 24
Cycle time (s)

Figure 4: Particle CT distribution, confined to cycles shorter than 25 s, derived from experiment
[2], simulation outcomes of the current study, and reference papers [3, 9].

suggest a heavier tail in the CTD compared to other simulations. However, Li et

al. present CTDs that contradict this expectation, possibly due to their observation

19
of stable CTDs in simulations exceeding 25 s. In contrast, this was not observed
in our simulations, where the CTD showed a continual drive towards longer cycle
times from t = 25 s to t = 30 s. Nevertheless, the results obtained with our CFD-
DEM framework already show strong concordance with the experimental values,
thus eliminating the necessity for extended simulations or further analysis.

20
3.2. Coarse-graining with Heat and Mass Transfer Integration

3.2.1. Dynamics of Particle Motion and Momentum

To assess the effects of progressive coarse-graining on momentum and motion,
relative to the unscaled system, simulations spanning 20 s were employed. Dur-
ing these simulations, particle data from the designated regions are captured at
0.05-second intervals, yielding a total of 400 samples.

In Figure 5, the unfiltered particle velocity probability distribution functions are

shown for each region, based on all recorded particle values in each sample. In
the tube region, coarse-graining induces a shift towards marginally lower veloc-
ities, a trend that is consequently mirrored in the drying region. However, the
shift in the drying region is more marked, attributable to two factors: Firstly, the
elevated velocities originating from the tube which, secondly, cause particles to
ascend higher before descending, thus enhancing the downward velocities due to
increased falling heights. In the bed region, nearly identical results are observed,
except for the coarsest grains, which exhibit slightly lower velocities.

1.8 =1 1.2 =1 7 =1
Velocity distribution (s/m)

Velocity distribution (s/m)

Velocity distribution (s/m)

=2 =2 =2
1.5 1 =3
=3 =3 5.6
=4 =4 =4
1.2 0.8
4.2
Velocity distribution (s/m)

7
0.9 0.6 5.6
4.2
2.8
0.6 0.4 2.8
1.4

0.3 0.2 1.4 0

0 0.1 0.2 0.3 0.4 0.5
Particle velocity (m/s)

0 0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Particle velocity (m/s) Particle velocity (m/s) Particle velocity (m/s)

(a) Tube region (b) Drying region (c) Bed region

Figure 5: Probability distributions of particle velocities in the tube (a), drying (b), and bed (c)
regions in the original particle system (ψ = 1) and the coarse-grained systems (ψ = 2, 3, 4).

21
The time-volume averaged inertial number and the corresponding effective fric-
tion coefficients in each distinct region are depicted in Figure 6. The bed region
conforms to a dense inertial regime, as characterised by the inertial number. This
is corroborated by the mean relative magnitude of collisional to frictional stress,
approximately 0.65 in the original system, which progressively decreases to ap-
proximately 0.57 with the coarsest grains. Conversely, within the tube and drying
regions, the frictional component of the solid stresses becomes negligible, indi-
cating a collisional flow regime. This is further supported by the inertial numbers
observed in these regions and the saturation in µ′s , attributable to the large shear
stress relative to the confining pressure. As ψ increases, this saturation in µ′s takes
place at lower values, concurrently exhibiting increased variability in the inertial
number. Figure 7 presents the radial profiles of the solids volume fraction in a

1.6 A =1
Effective friction coefficient

A =2
1.4 A =3
A =4
1.2

0.8
Bed region Tube region Drying region
0.6

0.4

0.2

10-2 10-1 100

Inertial number

Figure 6: The time-volume averaged values of effective friction coefficients and inertial numbers,
in each region for the original (ψ = 1) and scaled systems (ψ = 2, 3, 4), including error bars.

cross-section situated 15 mm above the distributor plate. Near the interior tube

22
wall, the original particle system exhibits increased solid concentrations, leading
to elevated local pressure (flow resistance), which in turn results in steeper gradi-
ents, particularly in terms of radial pressure and velocity. Despite the jet flow, the
velocity profile remains largely unaltered, but an increment in air mass flow to-
wards the centre is observed. This is attributed to the fluidizing gas moving closer
to the jet, bringing particles from the bed region with it. However, in comparison
45

Wurster tube
40
=1
Solids volume fraction (%)

=2
35
=3
30 =4
25
20
15
10
5
0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
Radial distance from center (mm)

Figure 7: Solids volume fraction profiles radially in a 10 mm cross-section, located 15 mm above

the distributor plate (lower section of the tube, adjacent to the lower tube entry), segmented into 2
mm radial increments, showcased for the unscaled (ψ =1) and scaled (ψ = 2, 3, 4) systems.

to the coarse-grained systems, there is a slight reduction in the mass flow through
the tube due to axial gas flow along the tube walls. While these differences are mi-
nor and local, they are discernible in the solid concentration on the tube’s exterior,
where this axial flow causes more rigorous fluidization. This becomes notable in
the coarsest system, where the solids start to concentrate radially outward in the
bed, as a consequence of this.

23
3.3. Thermal and Mass Transfer Dynamics

To ascertain the accuracy of integrating heat and mass transfer phenomena in the
coarse-grained system, a dynamic two-stage analysis was conducted. This in-
volved 20-second simulations comparing the system with the highest degree of
coarse-graining (ψ = 4) against the original system (ψ = 1). Initiating the simula-
tions from a packed bed configuration with a uniform particle temperature of 0 ◦ C,
the first stage, lasting 10 seconds, involve convective heating of particles with the
hot fluidizing air. The subsequent stage defines a spray cooling phase, where pure
water is injected at a constant rate.

The regionally averaged gas and solid phase temperature profiles over 20 s, en-
compassing the convective and spray cooling stages consecutively, are depicted
in Figure 8. The temperatures closely align, with the original particle system re-
maining marginally warmer. The most pronounced difference is observed during
42 42 42
Average temperature (°C)

Average temperature (°C)

Average temperature (°C)

36 36 36

30 30 30

24 24 24

18 Solid phase 18 Solid phase 18 Solid phase

=1 =1 =1
12 =4 12 =4 12 =4
Gas phase Gas phase Gas phase
6 =1 6 =1 6 =1
=4 =4 =4
0 0 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Time (s) Time (s) Time (s)

(a) Tube region (b) Drying region (c) Bed region

Figure 8: Transient profiles of the gas and solid phases’ mean temperatures in each region over 20
s for the original particle system (ψ = 1) and the systems with the coarsest grains (ψ = 4).

the bed’s heating-up stage, wherein the gas temperatures of the system diverge by
2.6 ◦ C after ten seconds. In this initial stage, a gradual convergence over time is

24
noted, yet it does not reach a plateau before the onset of the second stage, indicat-
ing a trend towards converging steady-state temperatures. The temperature differ-
ences in the system illustrate effective phase mixing, despite the coarse-grained
system’s delayed stabilization in the tube. This delay correlates with the dispari-
ties noted in the bed regions, attributable to the more humid air traversing the tube
in the coarse-grained system. In the ensuing spray cooling stage, the difference
between the systems quickly diminishes, culminating in near-identical conditions
after 20 s, with a maximum temperature variation below 0.5 ◦ C.

Following the spray cooling phase, the moisture distribution across particles in-
trinsically governs the effectiveness of the coating. Consequently, Figure 9 de-
picts the distribution functions of the particle coating concentration in each region.
Each region in both systems exhibits a log-normal coating distribution, consistent
with experimental observations [52].
0.4 0.6 0.8
=1 =1 =1
Coating distribution (kg/g)

Coating distribution (kg/g)

Coating distribution (kg/g)

=4 =4 =4
0.3 0.6
0.4

0.2 0.4

0.2
0.1 0.2

0 0 0
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Moisture content (g/kg) Moisture content (g/kg) Moisture content (g/kg)

(a) Tube region (b) Drying region (c) Bed region

Figure 9: Moisture probability distribution function in each region after 20 s for the original
particle system (ψ = 1) and the systems with the coarsest grains (ψ = 4). Values are presented per
solid, measured in grams of liquid per kilogram of solid.

25
3.3.1. Computational efficacy
For the simulations of the original CFD-DEM system (ψ = 1), one second of the
simulation was generated for every 12 hours of real-time processing. In contrast,
the corresponding coarse-grained system with ψ = 4, achieved a ten-fold increase,
rendering 10 s of simulation in the same period, thereby exhibiting a logarithmic
reduction in simulation time.

4. Conclusions

Simulations of the Wurster coating process were executed in the original CFD-
DEM formulation and with an integrated coarse-graining technique, wherein up
to 64 original particles were represented by larger grains, effectively reducing
the particle count from 5.1 million to 79 thousand. While culminating in a log-
arithmic increase in simulation speed, results closely congruent with those from
the original CFD-DEM simulation were achieved. Initial experimental validation
corroborated high accuracy in particle velocities, RT, and CT distributions.

The efficacy of the coarse-grained system in simulating macroscopic granular flow

properties akin to those of the original particles was apparent. The most notable
difference in the coarse-grained system, compared to the original particles, was
observed in the tube, where grain size marginally influenced the granular flow. In-
corporating spray modelling into the framework facilitated the assessment of heat
and mass transfer between MCC pellets and humid air, as well as their transient
temperature profiles under different conditions. This produced coating distribu-
tions that were in excellent agreement.

The marked reduction in computational demands achieved via coarse-graining

26
highlights its capability to overcome CFD-DEM limitations while retaining reso-
lution, paving the way for industrial-scale Wurster coater system simulations.

27
Appendix A. Mesh Analysis

Mesh sensitivity was evaluated by analysing steady-state gas phase solutions.

Convergence was established by observing flow properties like gas velocity, dissi-
pation rate, and kinetic turbulence, with steady-state solutions deemed invariable
and acceptable at a residual tolerance below 10−4 . Radial velocity profiles of the
gas flow at 15 and 165 mm above the distributor plate are depicted in Figure A.10,
illustrating the effect of progressively refining the mesh from 50 to 105 thousand
cells. Analysis concluded that a mesh with 60 thousand cells sufficiently captures
the gas dynamics.

(a) 15 mm above the distributor plate. (b) 165 mm above the distributor plate.

Figure A.10: Radial gas velocity profiles, at steady-state, 15 and 165 mm above the distributor
plate, at four levels of mesh refinements.

Appendix B. Implementation of diffusion-based particle redistribution method

The redistribution method can be implemented by iteratively solving a discretized

form of the diffusion equation to redistribute excessive solid fractions from high-
concentration cells to neighbouring cells. The transferred quantity is then calcu-

28
lated based on the difference in the solid fraction between the neighbouring cells,
and the iteration continues until the specified maximum solid volume fraction or
the maximum number of iterations is reached. An explicit form of Fick’s second
law of diffusion is obtained by assuming a homogeneous allocation of the com-
putational cells. Then, the discretized diffusion equation that computes the new
concentration αc after diffusion from an initial concentration αc0 , provided by the
PCM, to N neighbouring cells can be expressed as follows:
ΩX
αc = αc0 + (αn − αc,0 ) (B.1)
Vc n∈N

The constant Ω is proportional to the diffusion constant and determines the itera-
tion rate of the solid volume fraction. A larger of Ω leads to faster iteration, but
can compromise stability. In a non-uniform Eulerian grid, the value of Ω only
needs to be the same within the sets of cells that exchange solid volume fractions.
Thus, an optimal value of Ω is determined per cell as Ω = Vc /2N, which if sub-
stituted into B.1 gives equation B.2 below, that represents the maximum rate at
which the cell can undergo a stable exchange.
1 X
αc = αc0 + (αn − αc0 ) (B.2)
2N n∈N

Therefore, for any set of exchanging cells, if a neighbouring cell has a lower
value of Ωn , this value must be adopted for summation over all these cells, thus
restricting the overall diffusion rate for that set. By implementing this knowledge,
equation B.1 can be expressed as follows:
1 X h i
αc = αc0 + min Ωn , Ω (αn − αc0 ) (B.3)
Vc n∈N

The presented equation is applied to every cell within the Eulerian domain and
solved iteratively to obtain the updated value of αc in each cell during the map-

29
ping process of solid volume fraction between the Lagrangian and Eulerian frame-
works. Additionally, the diffusion of solid volume fraction is employed to facil-
itate the mapping of energy and momentum exchange, with additional weighting
based on the fluid volume present in the current cell at a specific time [40, 42].
The application of the method to other quantities is analogous. For smoothing of
a given force component Fc0 , the new force is obtained by:
1 X h i Fn Fc0 
F c = F c0 + min ρn Ωn , ρΩ − (B.4)
Vc n∈N ρn ρ

seen from Equation B.4, the force components are weighted by the specific mass
of gas occupying the cells. Conversely, for heat transfer:
1 X h i Q n Q c0 
Q c = Q c0 + min ρn c p,n Ωn , ρc p Ω − (B.5)
Vc n∈N ρn c p,n ρc p

Appendix C. Macroscopic Stress and Strain Computations

To study the collective granular material behaviour at a macroscopic scale, the

transition from discrete particle stresses to a continuum-based framework neces-
sitates considering a control volume V s ≫ Vi . The macroscopic stress tensor of the
granular assemblies, σ s , is calculated as the volume-weighted sum of the stresses
of the individual particles, given by σ s = i σi Vi /V. The granular pressure, p s ,
P

is derived from the isotropic part of this tensor as p s = tr(σ s )/3. The magnitude
of shear stresses, |τ s |, is obtained from the second invariant of the deviatoric part
of σ s , and their ratio defines the effective or bulk friction coefficient µ′s = |τ s |/p s .
The shear strain rate magnitude, γ̇ s , is determined from the deviatoric part of the

s , as γ̇ s = 2||D s ||.
macroscopic granular strain rate tensor, Ddev dev

30
References

[1] K.-S. Seo, R. Bajracharya, S. H. Lee, H.-K. Han, Pharmaceutical applica-

tion of tablet film coating, Pharmaceutics 12 (9) (2020). doi:10.3390/
pharmaceutics12090853.
URL https://www.mdpi.com/1999-4923/12/9/853

[2] L. Li, A. Rasmuson, A. Ingram, M. Johansson, J. Remmelgas, C. von Cor-

swant, S. Folestad, Pept study of particle cycle and residence time dis-
tributions in a wurster fluid bed, AIChE Journal 61 (3) (2015) 756–768.
doi:10.1002/aic.14692.

[3] L. Li, J. Remmelgas, B. G. van Wachem, C. von Corswant, M. Johansson,

S. Folestad, A. Rasmuson, Residence time distributions of different size par-
ticles in the spray zone of a wurster fluid bed studied using dem-cfd, Powder
Technology 280 (2015) 124–134. doi:10.1016/j.powtec.2015.04.031.

[4] L. Fries, S. Antonyuk, S. Heinrich, S. Palzer, Dem–cfd modeling of a flu-

idized bed spray granulator, Chemical Engineering Science 66 (11) (2011)
2340–2355. doi:https://doi.org/10.1016/j.ces.2011.02.038.
URL https://www.sciencedirect.com/science/article/pii/
S0009250911001333

[5] M. Trogrlić, S. Madlmeir, T. Forgber, S. Salar-Behzadi, A. Sarkar, P. Liu,

L. Contreras, A. Carmody, A. Kape, J. Khinast, D. Jajčević, Numerical
and experimental validation of a detailed non-isothermal cfd-dem model
of a pilot-scale wurster coater, Powder Technology 391 (2021) 97–113.
doi:https://doi.org/10.1016/j.powtec.2021.05.100.

31
 URL https://www.sciencedirect.com/science/article/pii/
S0032591021005179

[6] C. Frey, Chapter 8 - fluid bed coating-based microencapsulation, in:

R. Sobel (Ed.), Microencapsulation in the Food Industry (Second Edi-
tion), second edition Edition, Academic Press, 2023, pp. 83–115.
doi:https://doi.org/10.1016/B978-0-12-821683-5.00024-8.
URL https://www.sciencedirect.com/science/article/pii/
B9780128216835000248

[7] M. Foroughi-Dahr, R. Sotudeh-Gharebagh, N. Mostoufi, Devel-

opment of a pat tool for monitoring the wurster coater perfor-
mance, International Journal of Pharmaceutics 561 (2019) 171–186.
doi:https://doi.org/10.1016/j.ijpharm.2019.02.023.
URL https://www.sciencedirect.com/science/article/pii/
S0378517319301462

[8] Y. Song, T. Zhou, R. Bai, M. Zhang, H. Yang, Effect of op-

erational and geometric parameters on the hydrodynamics of a
wurster coater: A cfd-dem study, Particuology 85 (2024) 62–76.
doi:https://doi.org/10.1016/j.partic.2023.03.019.
URL https://www.sciencedirect.com/science/article/pii/
S167420012300086X

[9] P. Böhling, J. G. Khinast, D. Jajcevic, C. Davies, A. Carmody, P. Doshi,

M. T. Am Ende, A. Sarkar, Computational fluid dynamics-discrete element
method modeling of an industrial-scale wurster coater, Journal of Pharma-

32
 ceutical Sciences 108 (1) (2019) 538–550. doi:10.1016/j.xphs.2018.
10.016.

[10] D. Jones, E. Godek, Chapter 35 - development, optimization, and scale-up

of process parameters: Wurster coating, in: Y. Qiu, Y. Chen, G. G. Zhang,
L. Yu, R. V. Mantri (Eds.), Developing Solid Oral Dosage Forms (Second
Edition), second edition Edition, Academic Press, Boston, 2017, pp. 997–
1014. doi:10.1016/B978-0-12-802447-8.00035-2.

[11] A. Zhu, Q. Chang, J. Xu, W. Ge, A dynamic load balancing algo-

rithm for cfd–dem simulation with cpu–gpu heterogeneous com-
puting, Powder Technology 428 (2023) 118782. doi:https:
//doi.org/10.1016/j.powtec.2023.118782.
URL https://www.sciencedirect.com/science/article/pii/
S0032591023005661

[12] D. Jajcevic, E. Siegmann, C. Radeke, J. G. Khinast, Large-scale cfd–dem

simulations of fluidized granular systems, Chemical Engineering Science 98
(2013) 298–310. doi:https://doi.org/10.1016/j.ces.2013.05.014.

[13] R. Šibanc, S. Srčič, R. Dreu, Numerical simulation of two-phase

flow in a wurster coating chamber and comparison with experi-
mental results, Chemical Engineering Science 99 (2013) 225–237.
doi:https://doi.org/10.1016/j.ces.2013.05.057.
URL https://www.sciencedirect.com/science/article/pii/
S0009250913003965

[14] S. Schneiderbauer, A. Aigner, S. Pirker, A comprehensive frictional-

33
 kinetic model for gas–particle flows: Analysis of fluidized and mov-
ing bed regimes, Chemical Engineering Science 80 (2012) 279–292.
doi:https://doi.org/10.1016/j.ces.2012.06.041.
URL https://www.sciencedirect.com/science/article/pii/
S0009250912003971

[15] L. Li, J. Remmelgas, B. G. van Wachem, C. von Corswant, S. Folestad,

M. Johansson, A. Rasmuson, Effect of drag models on residence time dis-
tributions of particles in a wurster fluidized bed: a dem-cfd study, KONA
Powder and Particle Journal 33 (2016) 264–277. doi:10.14356/kona.
2016008.

[16] P. A. Cundall, O. D. L. Strack, A discrete numerical model for granular

assemblies, Géotechnique 29 (1) (1979) 47–65. doi:10.1680/geot.1979.
29.1.47.

[17] S. Chialvo, S. Sundaresan, A modified kinetic theory for frictional granular

flows in dense and dilute regimes, Physics of Fluids 25 (7) (2013) 070603.
doi:10.1063/1.4812804.
URL https://doi.org/10.1063/1.4812804

[18] N. Deen, M. Van Sint Annaland, M. Van der Hoef, J. Kuipers, Re-
view of discrete particle modeling of fluidized beds, Chemical En-
gineering Science 62 (1) (2007) 28–44, fluidized Bed Applications.
doi:https://doi.org/10.1016/j.ces.2006.08.014.
URL https://www.sciencedirect.com/science/article/pii/
S0009250906004830

34
[19] H. M. Jaeger, S. R. Nagel, R. P. Behringer, Granular solids, liquids, and
gases, Rev. Mod. Phys. 68 (1996) 1259–1273. doi:10.1103/RevModPhys.
68.1259.
URL https://link.aps.org/doi/10.1103/RevModPhys.68.1259

[20] C. S. Campbell, Granular material flows – an overview, Powder Technology

162 (3) (2006) 208–229. doi:https://doi.org/10.1016/j.powtec.
2005.12.008.
URL https://www.sciencedirect.com/science/article/pii/
S0032591005005486

[21] O. Pouliquen, C. Cassar, P. Jop, Y. Forterre, M. Nicolas, Flow of dense

granular material: towards simple constitutive laws, Journal of Statisti-
cal Mechanics: Theory and Experiment 2006 (07) (2006) P07020. doi:
10.1088/1742-5468/2006/07/P07020.
URL https://dx.doi.org/10.1088/1742-5468/2006/07/P07020

[22] F. Y. . P. O. Jop, P., A constitutive law for dense granular flows., Nature 441
(2006) 727–730. doi:https://doi.org/10.1038/nature04801.

[23] F. da Cruz, S. Emam, M. Prochnow, J.-N. Roux, F. m. c. Chevoir, Rheo-

physics of dense granular materials: Discrete simulation of plane shear
flows, Phys. Rev. E 72 (2005) 021309. doi:10.1103/PhysRevE.72.
021309.
URL https://link.aps.org/doi/10.1103/PhysRevE.72.021309

[24] G. MiDi, On dense granular flows, Eur. Phys. J. E (2004) 341–365doi:

https://doi.org/10.1140/epje/i2003-10153-0.

35
[25] M. J. A. de Munck, E. A. J. F. Peters, J. A. M. Kuipers, Cfd-dem
fluidized bed drying study using a coarse-graining technique, Indus-
trial & Engineering Chemistry Research 62 (48) (2023) 20911–20920.
arXiv:https://doi.org/10.1021/acs.iecr.3c02960, doi:10.1021/
acs.iecr.3c02960.
URL https://doi.org/10.1021/acs.iecr.3c02960

[26] J. Hilton, D. Ying, P. Cleary, Modelling spray coating using a combined

cfd–dem and spherical harmonic formulation, Chemical Engineering
Science 99 (2013) 141–160. doi:https://doi.org/10.1016/j.ces.
2013.05.051.
URL https://www.sciencedirect.com/science/article/pii/
S0009250913003825

[27] P. Liu, C. M. Hrenya, Challenges of dem: I. competing bottlenecks in

parallelization of gas–solid flows, Powder Technology 264 (2014) 620–626.
doi:https://doi.org/10.1016/j.powtec.2014.04.095.
URL https://www.sciencedirect.com/science/article/pii/
S0032591014005385

[28] A. Di Renzo, E. S. Napolitano, F. P. Di Maio, Coarse-grain dem modelling in

fluidized bed simulation: A review, Processes 9 (2) (2021). doi:10.3390/
pr9020279.

[29] K. Washino, E. L. Chan, T. Kaji, Y. Matsuno, T. Tanaka, On large

scale cfd–dem simulation for gas–liquid–solid three-phase flows,
Particuology 59 (2021) 2–15, towards Modelling Solids and Mul-
tiphase Flow for Large Scale Industrial Systems and Applications.

36
 doi:https://doi.org/10.1016/j.partic.2020.05.006.
URL https://www.sciencedirect.com/science/article/pii/
S1674200120300754

[30] M. Sakai, S. Koshizuka, Large-scale discrete element modeling in pneumatic

conveying, Chemical Engineering Science 64 (3) (2009) 533–539. doi:
https://doi.org/10.1016/j.ces.2008.10.003.

[31] M. Sakai, H. Takahashi, C. C. Pain, J.-P. Latham, J. Xiang, Study on a large-

scale discrete element model for fine particles in a fluidized bed, Advanced
Powder Technology 23 (5) (2012) 673–681. doi:https://doi.org/10.
1016/j.apt.2011.08.006.

[32] D. S. Nasato, C. Goniva, S. Pirker, C. Kloss, Coarse graining for large-scale

dem simulations of particle flow – an investigation on contact and cohesion
models, Procedia Engineering 102 (2015) 1484–1490. doi:https://doi.
org/10.1016/j.proeng.2015.01.282.

[33] C. Bierwisch, T. Kraft, H. Riedel, M. Moseler, Three-dimensional discrete

element models for the granular statics and dynamics of powders in cavity
filling, Journal of the Mechanics and Physics of Solids 57 (1) (2009) 10–31.
doi:https://doi.org/10.1016/j.jmps.2008.10.006.

[34] S. Radl, C. A. Radeke, J. G. Khinast, S. Sundaresan, Parcel-based approach

for the simulation of gas-particle flows, 2011.

[35] D. Queteschiner, T. Lichtenegger, S. Pirker, S. Schneiderbauer, Multi-level

coarse-grain model of the dem, Powder Technology 338 (2018) 614–624.
doi:https://doi.org/10.1016/j.powtec.2018.07.033.

37
[36] S. Heinrich, M. Dosta, S. Antonyuk, Multiscale Analysis of a Coating Pro-
cess in a Wurster Fluidized Bed Apparatus, Vol. 46, 2015, pp. 83–135.
doi:10.1016/bs.ache.2015.10.012.

[37] H. Che, H. Wang, L. Xu, R. Ge, Investigation of gas-solid heat

and mass transfer in a wurster coater using a scaled cfd-dem
model, Powder Technology 406 (2022) 117598. doi:https:
//doi.org/10.1016/j.powtec.2022.117598.
URL https://www.sciencedirect.com/science/article/pii/
S0032591022004922

[38] H. Che, D. Liu, W. Tian, S. Gao, J. Sun, L. Xu, Cfd-dem study of gas-solid
flow regimes in a wurster type fluidized bed with experimental validation
by electrical capacitance tomography, Chemical Engineering Journal 389
(2020) 124280. doi:https://doi.org/10.1016/j.cej.2020.124280.
URL https://www.sciencedirect.com/science/article/pii/
S1385894720302710

[39] T.-H. Shih, W. W. Liou, A. Shabbir, Z. Yang, J. Zhu, A new k-

eddy viscosity model for high reynolds number turbulent flows, Comput-
ers Fluids 24 (3) (1995) 227–238. doi:https://doi.org/10.1016/
0045-7930(94)00032-T.

[40] R. Sun, H. Xiao, Diffusion-based coarse graining in hybrid contin-

uum–discrete solvers: Theoretical formulation and a priori tests, Interna-
tional Journal of Multiphase Flow 77 (2015) 142–157. doi:10.1016/j.
ijmultiphaseflow.2015.08.014.

38
[41] J. Zhang, T. Li, H. Ström, B. Wang, T. Løvås, A novel coupling method
for unresolved cfd-dem modeling, International Journal of Heat and Mass
Transfer 203 (2023) 123817. doi:10.1016/j.ijheatmasstransfer.
2022.123817.

[42] R. Sun, H. Xiao, Diffusion-based coarse graining in hybrid contin-

uum–discrete solvers: Applications in cfd–dem, International Jour-
nal of Multiphase Flow 72 (2015) 233–247. doi:10.1016/j.
ijmultiphaseflow.2015.02.014.

[43] D. Gunn, Transfer of heat or mass to particles in fixed and fluidised beds,
International Journal of Heat and Mass Transfer 21 (4) (1978) 467–476.
doi:https://doi.org/10.1016/0017-9310(78)90080-7.

[44] H. Hertz, Ueber die berührung fester elastischer körper. 1882 (92) (1882)
156–171. doi:doi:10.1515/crll.1882.92.156.

[45] Y. Tsuji, T. Tanaka, T. Ishida, Lagrangian numerical simulation of plug flow

of cohesionless particles in a horizontal pipe, Powder Technology 71 (3)
(1992) 239–250. doi:10.1016/0032-5910(92)88030-L.

[46] M. Emam, L. Zhou, W. Shi, H. Chen, L. Bai, R. Agarwal, Theories and

applications of cfd–dem coupling approach for granular flow: A review,
Archives of Computational Methods in Engineering 28 (04 2021). doi:
10.1007/s11831-021-09568-9.

[47] T. Schwager, T. Pöschel, Coefficient of restitution and linear dash-

pot model revisited, Granular Matter 9 (02 2007). doi:10.1007/
s10035-007-0065-z.

39
[48] M. de Munck, E. Peters, J. Kuipers, Fluidized bed gas-solid heat transfer us-
ing a cfd-dem coarse-graining technique, Chemical Engineering Science 280
(2023) 119048. doi:https://doi.org/10.1016/j.ces.2023.119048.
URL https://www.sciencedirect.com/science/article/pii/
S0009250923006048

[49] M. Heine, S. Antonyuk, L. Fries, G. Niederreiter, S. Heinrich, S. Palzer,

Modeling of the spray zone for particle wetting in a fluidized bed,
Chemie Ingenieur Technik 85 (3) (2013) 280–289. arXiv:https:
//onlinelibrary.wiley.com/doi/pdf/10.1002/cite.201200148,
doi:https://doi.org/10.1002/cite.201200148.
URL https://onlinelibrary.wiley.com/doi/abs/10.1002/cite.
201200148

[50] S. Chialvo, J. Sun, S. Sundaresan, Bridging the rheology of granular flows in

three regimes, Phys. Rev. E 85 (2012) 021305. doi:10.1103/PhysRevE.
85.021305.
URL https://link.aps.org/doi/10.1103/PhysRevE.85.021305

[51] S. Madlmeir, T. Forgber, M. Trogrlic, D. Jajcevic, A. Kape, L. Contr-

eras, A. Carmody, P. Liu, C. Davies, A. Sarkar, J. Khinast, Quantifying
the coating yield by modeling heat and mass transfer in a wurster flu-
idized bed coater, Chemical Engineering Science 252 (2022) 117505.
doi:https://doi.org/10.1016/j.ces.2022.117505.
URL https://www.sciencedirect.com/science/article/pii/
S0009250922000896

[52] H. R. Norouzi, Simulation of pellet coating in wurster coaters, Inter-

40
national Journal of Pharmaceutics 590 (2020) 119931. doi:https:
//doi.org/10.1016/j.ijpharm.2020.119931.
URL https://www.sciencedirect.com/science/article/pii/
S0378517320309169

41