Sei sulla pagina 1di 13

International Journal of Multiphase Flow 85 (2016) 223235

Contents lists available at ScienceDirect

International Journal of Multiphase Flow


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

On the implementation of moment transport equations in OpenFOAM:


Boundedness and realizability
Antonio Buffo a,b,, Marco Vanni a, Daniele L. Marchisio a
a
b

Dipartimento di Scienza Applicata e Tecnologia, Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Torino, Italy
Department of Biotechnology and Chemical Technology, Aalto University, Kemistintie 1, 02150 Espoo, Finland

a r t i c l e

i n f o

Article history:
Received 15 September 2015
Revised 21 June 2016
Accepted 22 June 2016
Available online 25 June 2016
Keywords:
Moment boundedness
Moment realizability
Generalized population balance equation
(GPBE)
Quadrature-based moment methods
(QBMM)
OpenFOAM

a b s t r a c t
In this contribution some crucial numerical aspects concerning the implementation of quadrature-based
moment methods (QBMM) into the open-source Computational Fluid Dynamics (CFD) code OpenFOAM
are discussed. As well-known QBMM are based on the simple idea of solving a kinetic master equation,
not in terms of the underlying number density function (NDF), but in terms of the moments of the NDF
itself, via moment transport equations. These numerical aspects are in fact very often overlooked, resulting in implementations that do not satisfy the properties of boundedness and realizability for the moments of the NDF. Boundedness is an important property (i.e., moments of the NDF have to be bounded
between some minimal and maximum values), that in turn depends on the initial and boundary conditions. Boundedness can be guaranteed by using a consistent approach with respect to the constraints imposed on the transport variables, such as dispersed phase volume fraction. Realizability is instead related
to the existence of an underlying NDF that corresponds to a specic moment set. It is well-known that
time and spatial discretization schemes can corrupt a moment set unless they are specically designed
to preserve realizability. One popular choice is to use ad-hoc pseudo high-order schemes to ensure realizable moment set is obtained. In this work, moment transport equations are implemented in the CFD
code OpenFOAM using a similar form proposed by Weller (2002) for preserving the boundedness of the
volume fraction, together with the numerical schemes of Vikas et al. (2011) for ensuring moment realizability. The effectiveness of such implementation is illustrated on two simple examples, taken from our
work on the simulation of uid-uid multiphase systems.
2016 Elsevier Ltd. All rights reserved.

1. Introduction
The generalized population balance equation (GPBE) describes
the evolution of multiphase particulate polydisperse systems
(Marchisio and Fox, 2013; Ramkrishna, 20 0 0) that nd applications in many different elds, from physics and chemistry to engineering. Many different solution methods for the GPBE were proposed in the literature in the latest years (for an overview of
these approaches, the reader may refer to the work of Marchisio
and Fox, 2013). Among them, quadrature-based moment methods (QBMM) gained huge popularity in the last decade due to
their low computational demand. Many methods belong to this
category: the Quadrature Method of Moments (QMOM) (McGraw,
1997), the Direct Quadrature Method of Moments (DQMOM)

Corresponding author at: Department of Biotechnology and Chemical Technology, Aalto University, Kemistintie 1, 02150 Espoo, Finland.
E-mail address: antonio.buffo@aalto. (A. Buffo).

http://dx.doi.org/10.1016/j.ijmultiphaseow.2016.06.017
0301-9322/ 2016 Elsevier Ltd. All rights reserved.

(Marchisio and Fox, 2005), the Conditional Quadrature Method


of Moments (CQMOM) (Yuan and Fox, 2011), and the Extended
Quadrature Method of Moments (EQMOM) (Yuan et al., 2012).
These methods are based on the idea of solving the GPBE by tracking the time and spatial evolution of some integral properties of
the underlying number density function (NDF), which in turn denes the state of the multiphase polydisperse system. These integral properties are known as moments of the NDF, and for the
solution of many theoretical and practical problems only a small
set of lower-order moments are necessary. Due to their characteristics, QBMM are particularly suitable for solving spatially inhomogeneous problems, where the dispersed population of particles,
bubbles or droplets, constituting the polydisperse multiphase system moves into the physical space. Hence the integration of these
methods into existing commercial, academic in-house and opensource computational uid dynamics (CFD) codes is the subject of
many recent works (Buffo et al., 2013a; 2012; 2013b; Marchisio
et al., 2003; Pea-Monferrer et al., 2016; Silva et al., 2008; Silva
and Lage, 2011; Zucca et al., 2007).

224

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

However, the way in which the moment transport equations are


implemented and numerically solved in CFD codes is an aspect often overlooked. The moments of the NDF are linked to each other
by complex mathematical relationships, that express the existence
of an underlying NDF. If the numerical schemes (for time and space
integration) used for the solution of the moment transport equations do not preserve these relationships, the corresponding moment set, after numerical integration, can become unrealizable, as
described in detail by Wright Jr. (2007) who rst identied this
moment corruption problem. To guarantee the realizability of the
moment set specic numerical methods must be used (Marchisio
and Fox, 2013). Vikas et al. (2011) proposed particular realizability preserving high-order discretization schemes for the moment
transport equations. Nevertheless, these numerical methods do not
guarantee that the moments of the NDF lay between physically
meaningful minimum and maximum values specic of the particular problem analyzed during the calculations. These constraints
may depend on to the boundary and/or initial conditions and can
be respected by seeking numerical procedures that keep the solution bounded. On this particular aspect, it is important to analyze the approaches adopted for other bounded variables. The simplest example refers to a passive scalar bounded between a minimum and a maximum value: in this respect many discretization
schemes are suitable to ensure this property, from the standard
rst-order upwind to the so-called bounded Total Variation Diminishing (TVD) schemes (Ferreira et al., 2012; Herrmann et al., 2006;
Song et al., 20 0 0; Waterson and Deconinck, 20 07; Zhu, 1992). Another important example is that of the volume fraction of the dispersed phase which is directly related to a moment of the NDF.
Clearly, the volume fraction has to be bounded between zero and
one by the numerical procedure. Rusche (2002) pointed out that
the existing bounded discretization schemes may not guarantee
boundedness of volume fraction owing to the strong link between
volume fraction and multiphase velocity coupling. In this respect,
Weller (2002) proposed an ecient boundedness preserving implementation of the volume fraction equation, currently available
inside the CFD code OpenFOAM.
Also the moments of the NDF are physical quantities that have
to be bounded between a certain range determined by the uid
dynamics of the system. Similar to the disperse phase, other moments of NDF are advected with their own velocity eld but with
the difference that the minimum and maximum values of the moments are not known a-priori. Therefore, to ensure both boundedness and realizability of the moments of the NDF from the
numerical point of view, it is necessary to adopt specic measures in terms of implementation and numerical solution of the
moment transport equations inside CFD codes. Of course, these
have to be implemented consistently with the structure of the
CFD code utilized, and for this reason some of the solutions
here proposed are not universal, but depend on the adopted CFD
code.
In this work, which is the follow-up of a previous work (Buffo
et al., 2014; 2013a), a specic implementation, that guarantees
boundedness, together with numerical schemes that preserve realizability, is discussed and tested, for the rst time, for the opensource CFD code OpenFOAM. This implementation is based on the
extension of the aforementioned method of Weller (2002) to the
moment transport equations, together with the use of the realizable discretization schemes proposed by Vikas et al. (2011). The
manuscript is structured as follows. First the governing equations
are presented, showing the main features of QBMM. Then the definition of the desired properties of moment boundedness and realizability are introduced and the numerical methods for ensuring
them in a CFD code implementation are explained. Eventually two
test cases are introduced and in the last section the obtained results are discussed.

2. GPBE and QBMM


In this section we summarize the essential theoretical features
concerning the solution of the GPBE with QBMM. Let us consider a
generic particulate multiphase ow, with a continuous uid phase
and a disperse phase, that might be composed of solid particles,
droplets or bubbles. These disperse elements can be thought of
as a population evolving chaotically in space and time and being
characterized by different properties, such as size, chemical composition, velocity, temperature, etc. These are usually referred as
the internal coordinates of the population, which are different from
the external coordinates, namely the space coordinates and time.
This population can be represented from the mathematical point
of view by a smooth and differentiable function, the Number Density Function (NDF), obtained by ensemble-averaging and dened
so that the following quantity:

n ( , u; x, t )d du dx
represents the expected number of elements of the disperse
phase in the innitesimal volume dx, around the physical point x,
with the internal coordinates ranging between and + d , where
= (1 , 2 , . . . , M ) is the vector that contains M relevant properties of the system and with velocity within an innitesimal interval around u. In many implementations of QBMM in CFD codes,
instead of considering n the following marginal NDF, obtained by
integrating out particle velocity, is considered:

n ( ; x, t ) =



n ( , u; x, t )du,

(1)

representing the expected number density of particles with internal coordinate vector within and + d (regardless of their velocity). The continuity statement for this function corresponds to
the following GPBE (Marchisio and Fox, 2013) (by omitting the dependencies on internal coordinates, physical space and time in the
notation):

+
( Un ) = H
( n )
t x

(2)

where H( ; x, t) represents the source term due to collisional


events and the last term accounts for the drift in internal coordinate space (with drift velocity vector ( ; x, t )). The real-space
particle velocity U( ; x, t) is dened as:


un ( , u; x, t )du
U ( ; x, t ) =   
,
n ( , u; x, t )du

(3)

and represents the velocity of the particles with internal coordinates vector equal to .
Eq. (2) can be used to derive transport equation for the moments of the NDF, dened as follows (Marchisio and Fox, 2013):

Mk ( x , t ) =




n( ; x, t )1k1 2k2 . . . MkM d =




n( ; x, t ) d , (4)
k

where  represents the phase-space generated by all the possible values of the internal coordinate considered, while the vector
k = (k1 , k2 , . . . , kM ) contains the order of the moments with respect to each internal coordinate. The nal transport equation for
the moment becomes (Buffo and Marchisio, 2014):

Mk
+ (Uk Mk ) = Hk + Gk ,
t

(5)

where Hk and Gk are, respectively, the discontinuous and continuous event source terms, usually grouped together into a single
source term for the moment of order k: Sk = Hk + Gk , and where

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

Uk is the velocity of the moment of order k:

Uk =

U ( )n ( ) d
=

k
n ( ) d
k

U ( )n ( ) d
.
Mk
k

(6)

As well-known Eq. (5) is not closed and closure is achieved by


using a quadrature approximation, that in the case of CQMOM results in the following approximation for the moment of order k:

Mk ( x , t ) =

N1 
N2


1 =1 2 =1

NM


w1 w1 ;2 . . . w1 ;2 ;...;M

M =1

(1;1 ) 1 (2;1 ,2 ) 2 . . . (M;1 ,2 ;...;M ) M ,


k

(7)

where w1 , w1 ;2 , etc. are the N1 N2 ... NM weights and


1;1 , 2;1 ,2 , etc. are the N1 N2 ... NM nodes of the quadrature approximation. Quadrature weights and nodes can be calculated from the moments of the NDF through different inversion
algorithms (Marchisio and Fox, 2013). In the case of only one internal coordinate (M = 1) QMOM is used, while in case of more
than one internal coordinate (M > 1) multivariate methods must
be used, such as for example CQMOM.
It is useful to remind here that the quantity U( ), appearing in
Eq. (6), is usually calculated within the CFD code independently
from the moment transport equation. In this work we assume that
all the moments of the NDF move with one single velocity, Ud , that
characterizes the disperse phase, as the extension to other more
complicated situations is straightforward. When QBMM are coupled with the EulerianEulerian two-uid model (TFM) (Hill, 1998;
Ishii and Mishima, 1984; Kataoka, 1986), Ud is calculated by solving the following set of equations:

i i
+ (i i Ui ) = i
t
i i Ui
+ (i i Ui Ui )
t
= (i i ) i p + i i g + i Ui + i

(8)

(9)

for the continuous phase i = c and for the disperse phase i = d,


where i is the volume fraction of the generic ith phase, i is its
density, i is the tensor accounting for both viscous and turbulent
stresses, p is the pressure eld shared between all the phases, g is
the gravity,  i and i are respectively the mass and momentum
exchange term between phase i and all the other phases present
in the system. It is useful to mention that both equations are
derived through ensemble-averaging (Drew and Passman, 1998),
since they refer to turbulent systems. Therefore the quantities i
and Ui should be intended as ensemble-averaged. As pointed out
in our previous works (Buffo and Marchisio, 2014; Buffo et al.,
2013a), both governing equations can be also derived from the
general framework of GPBE.
3. Moment boundedness and realizability
As it is well-known, the numerical solution of partial differential equations (such as the Finite-Volume, FV, method typical of
CFD codes) may introduce errors, that may sum together, eventually leading to wrong predictions (Ferziger and Peric, 2001):
namely the discretization and the iteration errors. The discretization error, representing the difference between the exact solution
of the conservation equations and the exact solution of the linearized algebraic system of the equations derived from the discretization, can be usually reduced by decreasing the grid size or
by using high-order time and space discretization methods. The iteration error, intrinsic to the numerical procedure, since it is the
difference between the iterative and the exact solution of the linearized equations, can be limited by setting proper tolerances or

225

adopting particular measures in the implementation of the equations. Also with respect to the numerical solution of the moment
transport equations in CFD codes, it is evident that these two
types of error cannot be avoided, affecting moment boundedness
and realizability, but only be limited by using proper numerical
procedures.
3.1. Moment boundedness
As mentioned the moments of the NDF represent physical
quantities and thus assume values bounded between minimum
and maximum values. These are not simple scalar quantities, advected with one single velocity shared with the continuous phase,
but representing the polydispersity of a multiphase system require
special attention. Moreover, the moments must respect complex
relationships for realizability constraints and the maximum and
minimum values are not known a priori. Boundedness can be locally violated due to iteration errors, but using a proper numerical implementation this issue can be solved. It should be noticed,
in fact, that the moments are transported in the physical space
with the velocity of the disperse phase Ud according to Eq. (5) and
obey a transport equation similar to that of the volume fraction
of the disperse phase d , as clearly evident by observing Eq. (8).
By denition d is bounded between zero and one and as pointed
out by Hill (1998) and Rusche (2002) the numerical solution of its
transport equation may violate this physical constraint, even when
using bounded discretization schemes. Moreover, the fact that the
volume fraction of the disperse phase is related to the moments of
the NDF suggests that the numerical procedure adopted to avoid
iteration errors and to ensure boundedness for d can be applied
also to the moment transport equation.
Let us describe rst the problem related to the volume fraction.
The governing equations for d and for his complement to one,
c , follow from Eq. (8) and can be written in the following nonconservative form:



d

d
+ (d Ud ) = d d
+ Ud d ,
t
d d t


c
c c c
+ (c Uc ) =

+ Uc c ,
t
c c t

(10)

(11)

and together with the two conditions d + c = 1 and d + c = 0,


Eqs. (10) and (11) guarantee that the total mass of the system is
properly conserved. It is important to observe also that, in the case
of no mass transfer and when both phases are incompressible, the
source terms of both equations are equal to zero. The discretization scheme adopted may play a role for preserving the boundedness of d and c between 0 and 1 also in the numerical solution of Eqs. (10) and (11). In general, for scalar quantities, it is
known that the rst-order upwind scheme is capable of guaranteeing boundedness, while standard higher-order schemes may produce unbounded solutions, when the grid adopted is too coarse
(Ferziger and Peric, 2001). In the recent years, many high-order
discretization schemes were proposed specically for quantities
that are bounded between a dened range (Ferreira et al., 2012;
Herrmann et al., 2006; Song et al., 20 0 0; Waterson and Deconinck, 2007; Zhu, 1992). However, in case of iteration errors affecting the solution of the velocity elds, the value of d may be unbounded, regardless the particular discretization scheme adopted
for the volume fraction equation. In this case, a different implementation strategy of Eq. (10) can solve this issue (Rusche, 2002)
as described in the following.
In the TFM usually implemented in CFD codes only Eq. (10) is
solved, then the volume fraction of the continuous phase c is
calculated as 1 d . The total mass conservation is enforced by

226

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

summing the Eqs. (10) and (11) leading to the following


expression:



d c d d
(Uvm ) =
+

+ Ud d
d c d t


c c

+ Uc c
c t

(12)

where Uvm = d Ud + c Uc is the mean volumetric velocity, whose


eld is divergence free, if both phases are incompressible and there
is no mass transfer. The Poisson equation for calculating the pressure is derived from Eq. (12), ensuring that the interpolation of
Uvm at cell faces satises the mixture continuity equation exactly
(from the numerical point of view). However, as pointed out by
Rusche (2002), Ud might not be divergence free. This fact, together
with the iteration errors intrinsic to the numerical procedure, may
lead to local violation of the boundedness constraint for the disperse volume fraction when Eq. (37) is solved.
Different approaches proposed to overcome this issue were
tested by Rusche (2002), demonstrating that an ecient way
to ensure boundedness for the volume fraction, is the one proposed by Weller (2002). This method starts from the denition
of the mean volumetric velocity Uvm and the relative velocity
Ur = Uc Ud , so that is is possible to write the velocity of the disperse phase in the following way:

Ud = Uvm c Ur = Uvm (1 d )Ur

(13)

and by substituting this expression into Eq. (10) the following expression is obtained:

d
+ (d Uvm ) (d (1 d )Ur )
t



d
= d d
+ Ud d .
d d t

d
,
d + c

Mk
+ (Uvm Mk ) (Mk (1 d )Ur ) = Sk
t

(16)

where the second term is bounded when Uvm is divergence free (in
case of incompressible phase with no mass transfer) and the third
term goes to zero in the limit of d tending to unity, namely when
the population of disperse elements is becoming the continuous
phase (the so-called phase inversion) and therefore the moments
are no longer dened, leading to an unphysical solution.

(14)
3.2. Moment realizability

In case of no mass transfer (i.e., d = 0) and when the disperse


phase is incompressible, it can be shown that the implementation reported in Eq. (14) is capable of preserving the boundedness
of the disperse phase volume fraction: in fact, the second term
gives bounded solution if the Uvm eld is divergence free and the
third term guarantees that 0 d 1 since it goes to zero at
both limits. Rusche (2002) remarked that this numerical procedure
may still not guarantee the boundedness of the volume fraction if
standard high-order discretization schemes are adopted, especially
in the case of coarse grids; however this inconvenience can be
avoided by using bounded discretization schemes (Passalacqua and
Fox, 2011). Moreover, another known limitation of this approach is
the non-linearity of Eq. (14) with respect to d : boundedness is
guaranteed only if the equation is solved fully implicitly, requiring
iterations to reach the convergence. But, as pointed out by Rusche
(2002), the iteration procedure may not be always convergent, especially when the time step is not suciently small: but still this
aspect does not represent a problem in transient simulations, because small time steps are usually required for stability issues. In
case of steady-state simulations, instead, this approach is still not
sucient to guarantee bounded elds of d and the additional solution of the transport equation for the continuous phase volume
fraction c is also required. After each iteration the boundedness
of the volume fraction is locally enforced by means of the following equation:

d =

work, we are interested in transient simulations and the steadystate approach is reported just for the sake of completeness, but is
not implemented.
It is important to remark that boundedness of d is guaranteed by the implementation shown in Eq. (14) for transient simulations only if the two phases are both incompressible and there
is no mass transfer. Otherwise, when one of the two conditions
are not veried, boundedness is not necessarily guaranteed by this
specic numerical procedure, since mass is no more a conserved
quantity. In these cases, particular attention should be paid when
the equations are solved, by using an appropriate time step (or
under-relaxation factors in case of steady state simulations); however, the non-linear term present in Eq. (14) should in any case
reduce the eventuality of an unbounded solution.
As previously mentioned, the transport equation of the generic
moment of the NDF is similar to the disperse phase volume fraction equation, and therefore it is clear that a similar approach for
preserving boundedness can be applied also to the moment transport equation. In fact, if the moment transport equation is implemented as written in Eq. (5), similar issues to those appearing for
d may occur, since the velocity Ud is used to transport the moments in the physical space. By substituting the disperse phase velocity Ud dened in Eq. (13), the following expression similar to
Eq. (14) can be written as follows:

(15)

where the volume fraction d is now bounded between 0 and


1. In this case, the total mass conservation (i.e., d + c = 1) is
ensured only when nal convergence is reached. However, in this

As previously mentioned, the realizability of the moment set, is


related to the existence of an underlying NDF, that corresponds to
that specic moment set. Realizability impacts the stability of the
simulation, as in QBMM an inversion algorithm is adopted to reconstruct the NDF; when an unrealizable moment set is fed to the
inversion algorithm an unrealizable reconstructed NDF is obtained
leading to unwanted numerical instabilities.
The generation of unrealizable or corrupted moment sets may
primarily arise when standard high-order time and spatial discretization schemes are used for transporting the moments of the
NDF (Marchisio and Fox, 2013; Wright Jr., 2007). For example spatial high-order interpolation schemes for moment values at the
face between two neighboring cells may turn a realizable set into
an unrealizable one, and this corrupted set may propagate rapidly
in the computational domain through the advective term. In recent works, different strategies were proposed to correct a corrupted moment set (McGraw, 2012; Wright Jr., 2007), but these are
only capable to restore the set, not to prevent and permanently
solve the corruption problem. Moreover, these corrections were
proposed for a monovariate NDF, namely when only one internal coordinate is considered, because in this case the realizability
condition can be veried a-priori through the HankelHadamard
determinants (Shohat and Tamarkin, 1943). For multivariate NDF,
namely with multiple internal coordinates, the mathematical theory for realizability is not available, and it is not possible to apriori establish if a moment set is realizable before applying the
inversion algorithm. However, it is important to remark that if a
valid set is transported properly in the physical space, the resulting moment set preserves the realizability property (Marchisio and
Fox, 2013; Vikas et al., 2011).

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

Very recently a class of high-order numerical schemes was introduced (Vikas et al., 2011), based on the kinetic nite-volume
scheme, that guarantees the realizability of a set of moments. This
class of discretization schemes is based on the idea of calculating the moment uxes by interpolating at the faces of the cells
the values of quadrature weights and nodes, representing the reconstructed NDF, instead of the moments themselves: in this way,
it is possible to build discretization schemes that always prevent
the rise of the moment corruption problem. In particular, it is has
been demonstrated that the moment set is always realizable when
the quadrature nodes are interpolated by using the rst-order upwind and the quadrature weights are interpolated with high-order
schemes, providing that the time integration scheme is explicit
and the time step used follows some specic constraints. Readers interested in the details can refer to the work of Vikas et al.
(2011) and Marchisio and Fox (2013). Moreover, it is also possible
to prove that it is mathematically equivalent to use the rst-order
upwind method for both weights and nodes or to directly interpolate the moment themselves with the rst-order upwind schemes
(Desjardins et al., 2008); for this reason, in our previous works on
both commercial and open-source CFD codes, the use of rst-order
upwind for the moments equation was recommended (Buffo and
Marchisio, 2014; Buffo et al., 2013a; 2013b).
For simplicity, we limit the following discussion to a spatial
one-dimensional case, therefore Eq. (5) becomes:

Mk

+
(U M ) = Sk ,
t
x d k

(17)

where the velocity Ud = Ud (x, t ), as well as Mk = Mk (x, t ) and


Sk = Sk (x, t ). By dening the entire moment set with M =
[M0,0,... ; M1,0,... ; . . . ; M0,1,... ; M1,1,... ; . . . ]T and the moment ux with
F (Ud , M ) = [Ud M0,0,... ; Ud M1,0,... ; . . . ; Ud M0,1,... ; Ud M1,1,... ; . . . ]T , Eq.
(17) can be written as follows:

M F (Ud , M )
+
=S
t
x

(18)

with S = [S0,0,... ; S1,0,... ; . . . ; S0,1,... ; S1,1,... ; . . . ]T . By using a uniform


grid with spacing x, it is possible to evaluate the volume-average
moment set (in the nite-volume sense) at time t = n t, in the
following way:

Mni =

1
x

xi + x/2
xi x/2

M(x, n t )dx,

(19)

where xi represents the cell center. By using a rst-order forward


Euler time-integration scheme, the updated moment set in time
can be written as follows:

Mni +1

Mni

[G (Mni+ 1 ;l , Mni+ 1 ;r ) G (Mni 1 ;l , Mni 1 ;r )]


2
2
2
2
x

+ t S ,

(20)

where the subscript i 1/2 represents the face between the grid
cells i 1 and i (and i + 1/2 between i + 1 and i). The subscript l
indicates the limit approaching the face from the left (and r the
right). The denition of the numerical ux G is the following:

G (Ml , Mr ) = Ud+;l Ml + Ud;r Mr ,

(21)

where
=
+ |Ud;l | ) and
=
|Ud;r | ). It is important to notice that the estimation of the moment sets Ml and
Mr , located to the left and to the right of the cell interfaces
i 1/2 and i + 1/2, is required. As pointed out in the work of
Vikas et al. (2011), moment set reconstruction at faces Ml can be
performed in two ways: through direct interpolation of the moment set Mi or through the interpolation of quadrature weights
and nodes. As previously mentioned, the two choices are equivalent when the rst-order upwind scheme is used. When highorder methods are adopted, the rst option may lead to an unrealizable moment set, while only with the second option realizability is ensured (Marchisio and Fox, 2013; Vikas et al., 2011).
Ud+;l

1
2 (Ud;l

Ud;r

1
2 (Ud;r

227

It was proven that if the rst-order upwind scheme is used for


the quadrature nodes and any high-order scheme is used for the
quadrature weights, and if the time step t respects the following
condition (Vikas et al., 2011):

t min
i,

wm
x
i,

wm
U m+ wm
U m
i1/2, ,r d i1/2,r
i+1/2, ,l d i+1/2,l

(22)

realizability for the moment set is ensured.


By combining the scheme that guarantees boundedness, reported in Eq. (16):

M F 1 (Uvm , M ) F 2 (Ur , M )
+
+
=S
t
x
x

(23)

where Ur = (1 d )Ur , with the scheme that guarantees the realizability of the moment set, reported in Eq. (20), the following
overall scheme is obtained:

Mni +1 = Mni

t
[G (Mni+ 1 ;l , Mni+ 1 ;r ) G 1 (Mni 1 ;l , Mni 1 ;r )]
2
2
2
2
x 1

t
[G (Mni+ 1 ;l , Mni+ 1 ;r ) G 2 (Mni 1 ;l , Mni 1 ;r )]
2
2
2
2
x 2
+ t S
(24)

where
+

G 1 (Ml , Mr ) = Uvm
Ml + Uvm
;r Mr ,
;l

(25)

G 2 (Ml , Mr ) = Ur+
Ml + Ur
;r Mr .
;l

(26)

This slight modication of the scheme reported in Eq. (20) is capable of keeping bounded the value of the moments even when
using realizable high-order discretization schemes.
It is important to remark that the solution of the algebraic system of equation reported in Eq. (24) is explicit in time; moreover,
the uxes must be calculated in a consistent way, by using the values at time n for the solution at n + 1. This is particularly important when the moment equations are implemented in a CFD code,
since it tells us which is the optimal order for the solution of the
different equations in the TFM. In the next section, the details of
the implementation in the CFD code OpenFOAM are discussed.

3.3. Implementation in OpenFOAM


OpenFOAM is a FV library written in C++, composed of different routines and solvers for CFD calculations (Weller et al., 1998).
For details on its structure, the reader may refer the freely available user guides (OpenFOAM Foundation Ltd., 2015a; 2015b). In
this work, the TFM solver compressibleTwoPhaseEulerFoam
of the OpenFOAM version 2.2.x is modied with our own QBMM
implementation for the description of uid-uid or particle-uid
systems. This solver is based on the work of Weller (2002) for the
solution of the volume fraction and the PIMPLE algorithm (combination of SIMPLE and PISO) for the solution of the pressurevelocity coupling. This procedure allows under-relaxation within
the time step, enhancing the convergence and robustness of the algorithm (for a detailed description, see Passalacqua and Fox, 2011).
A schematic representation of the solver procedure is reported in
Fig. 1. The equations solved in this code are the following:

d
+ (d Uvm ) (d (1 d )Ur )
t



d
= d d
+ Ud d ,
d d t

(27)

228

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

The stress tensors d and c , expressing both viscous and turbulent


stresses, are formulated according to Newton and Boussinesq approximations. The standard k
model for the continuous phase
is used as turbulence model:



c
t,c
+ (c Uc ) c
= c (G
),
t
c

(34)





c

2
t,c
+ (c
Uc ) c

= c C
,1 G C
,2
.
t
c

(35)
The model constants are those of the standard
model: C =
0.09, = 1.0,
= 1.3, C
,1 = 1.44, and C
,2 = 1.92. The term G
is the turbulence production rate dened as: G = 2 tc,c (S : uc ),
where the strain rate tensor is in turn dened as S = 12 ( Uc +
( Uc )T ).
As depicted in Fig. 1 the optimal is to solve the moment transport equations before the solution of the volume fraction equation, d , so that the realizability criterion on the time step can
be enforced. The extension of the scheme reported in Eq. (24) to
three-dimensional cases is straightforward: at the beginning of every time step, the moment set in every cell of the domain is inverted to nd the quadrature weights and nodes, then the weights
and nodes values are reconstructed at the faces and nally the
value of the moments at the faces are calculated from the reconstructed weights and nodes, in order to calculate the numerical
uxes reported in Eqs. (25) and (26). The realizability criterion for
the three-dimensional case leads to the following expression:

t < min
i,

Fig. 1. Simplied owchart of the two-uid algorithm implemented in OpenFOAM


(compressibleTwoPhaseEulerFoam).



d c d d
(Uvm ) =
+

+ Ud d
d c d t


c c

+ Uc c ,
c t

(28)

d d Ud
+ (d d Ud Ud )
t
= (d d ) d p + d d g + d Ud + d ,

(29)

c c Uc
+ (c c Uc Uc )
t
= (c c ) c p + c c g + c Uc + c ,

(30)

Mk
+ (Uvm Mk ) (Mk (1 d )Ur ) = Sk ,
t

(31)

where the following conditions for the mass and momentum


transfer rates are also enforced:

d + c = 0,

(32)

d Ud + c Uc + d + c = 0.

(33)

wi,
.
(wi, Ud i )

(36)

where the subscript i refers to the ith computational cell and to


the th of the N1 N2 NM quadrature nodes.
The sequence of the operations performed by the moment equations module implemented in the OpenFOAM solver
compressibleTwoPhaseEulerFoam is the following:
1. Weights and nodes of the quadrature are evaluated at the cell
centers from the moments through the inversion algorithm.
2. Boundary conditions for weights and nodes are applied.
3. Velocities are evaluated at the faces through interpolation according to their discretization schemes, since OpenFOAM uses
a collocated grid arrangement (Rusche, 2002).
4. The time step t is evaluated, according to the realizability
condition. If the t found is less than the one used in the TFM,
it is overwritten on the previous one.
5. Weights are interpolated at the faces by using a higher-order
upwind scheme.
6. Nodes are interpolated at the faces by using a rst-order upwind scheme.
7. Fluxes of the moments are evaluated with the interpolated
weights, nodes and velocities.
8. Source terms of the moments are calculated in every cell, according to the value of weights and nodes found in step 1.
9. Solution of the dened linear system of the equations.
4. Numerical details and test cases
In this work, three different geometries were considered. In order to show the advantages of the realizable discretization scheme,
some preliminary simulations were performed for an idealized
spatially mono-dimensional (1D) system, where the disperse phase
ows in one direction and the continuous phase is perfectly mixed.
Subsequently, two realistic three-dimensional systems were considered with the aim of testing the implementation of moment

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

transport equations for the preservation of moment boundedness


and realizability. The physical system considered in all the simulations is composed of water (continuous phase) and air (disperse
phase), considering the size of the bubbles L as the only internal
coordinate, resulting in the use of a monovariate (M = 1) quadrature approximation (QMOM). This latter choice represents a simplication in terms of modeling hierarchy, since bubbles may be characterized by many other properties such as bubble composition,
temperature, momentum, etc. (Buffo and Marchisio, 2014), however it is suciently complex to show the features of the proposed
methodology for the advection of the moments in physical space.
The advection of mixed-order moments of a multivariate NDF can
be in fact treated in the same way. The system is isothermal and
both phases are considered incompressible, without mass exchange
between them and only buoyancy and drag forces are assumed to
be important, neglecting all the other forces in the evaluation of
the momentum transfer rates.
It is worth pointing out that when bubble size L is chosen as
the only internal coordinate of the NDF, the moments of the distribution are dened as:

Mk ( x , t ) =

n(L; x, t )Lk dL.

(37)

Low-order moments have a clear physical meaning: M0 represents


the total number of bubbles per unit volume, M1 represents the
total length of bubbles per unit volume, M2 is proportional to the
total interfacial area between bubbles and liquid and M3 is proportional to the gas volume fraction d . This latter relationship is
particularly interesting, in fact it shows that if the volume fraction is bounded between zero and one, because of the following
relationship:

d = kV M3

(38)

where kV is the volumetric shape factor, equal to /6 for spheres,


M3 has to be bounded between zero and d /kV = 6/ 1.90986.
As a consequence, the initial condition for all the moments was
xed to zero in all the domain, except for M3 whose value was
equal to 6/ 1.90986 above the liquid free-surface and zero under the liquid level to be consistent with the actual situation of the
system at the beginning of the simulation. The source terms of the
moment transport equation due to coalescence and breakage are
calculated as follows (Buffo and Marchisio, 2014):

Sk =

N1

N
N
k/3 k k 


1 
ai, j wi w j L3i + L3j
Li L j +
bi wi Pki 1 ,
2
i=1 j=1

i=1

(39)
where ai, j = a(Li , L j ) and bi = b(Li ) are the coalescence and breakage kernels, estimated according to models based on the homogeneous isotropic turbulence theory (Buffo and Marchisio, 2014;
Buffo et al., 2013a). The term Pki is the moment of order k of the
 +
daughter distribution function:, Pki = 0 P (L|Li )Lk dL, that for some
daughter distribution functions has an analytical solution (Buffo
et al., 2013a). The N quadrature weights wi and N quadrature nodes
Li are calculated from the rst 2N moments of the NDF through the
Wheeler inversion algorithm (Buffo and Marchisio, 2014). Moreover, it is important to mention that coalescence and breakage are
neglected when the disperse phase is very diluted (d < 1 106 )
or when it becomes continuous ( > 0.8), since both are very unlikely to occur under these conditions.
In the mono-dimensional test case, the method was veried
by implementing and solving the numerical scheme reported in
Eq. (20) for a quadrature approximation with two nodes (N1 = 2),
transporting therefore four moments. For this case, advection in
the physical space is the only process considered, as the source
term is assumed null. The velocity with which the moments are

229

advected was assumed to be constant and the initial condition for


the moments was equal to zero in all the domain, corresponding
to a system in which an advancing front is moving in one direction without dispersion: the so-called Riemann problem (LeVeque,
2002). Time integration was performed by using a rst-order forward Euler scheme; different grid sizes and different discretization schemes were used to assess the accuracy of the realizable
schemes for the moments, compared to other standard discretization methods.
The rst realistic three-dimensional test case corresponds to a
partially aerated rectangular bubble column of Diaz et al. (2008),
with the inlet located in the middle of the bottom. The grid is
constituted by a limited number of rectangular non-uniform cells
(32 width 11 depth 70 height) in order to enhance the effects of numerical diffusion and hence see the advantage of using higher-order schemes. However, it should noted that this is the
same grid used in our previous work (Buffo et al., 2013a), where
the model validation through comparison with the experimental
data of Diaz et al. (2008) was performed, showing that the grid
size is anyway enough to catch the main uid dynamics features of
the column (i.e., the characteristic oscillating motion of the bubble
plume). A sketch of the numerical grid is reported in Fig. 2. The
simulated gas supercial velocity corresponds to 5 mm/s, which is
a realistic velocity for a bubble column working in the homogeneous regime. All the simulations performed are transient with a
time step equal to 0.001 s, that can be reduced during the simulation to ensure the realizability condition expressed in Eq. (36).
At the initial time, the interface between the gas and liquid is located at two third of the column. For this case, bubble coalescence
and breakage were considered: details on the sub-models used, on
the discretization schemes for all the other variables and on the
boundary conditions for this system can be found in our previous
works (Buffo and Marchisio, 2014; Buffo et al., 2013a).
The following test cases are carried out: rst the moments are
transported by using Eq. (5), namely with an implementation that
does not ensure boundedness of the moments with the rst-order
upwind as a discretization scheme, then by using Eq. (16), namely
ensuring boundedness with the rst-order upwind discretization
scheme and nally by using the numerical approach reported in
Eq. (24) with different discretization schemes for the weights and
nodes of the quadrature approximation (i.e., second-order upwind
for the quadrature weights and rst-order upwind for the quadrature nodes).
The last geometry here considered is an aerated stirred tank reactor experimentally investigated by Laakkonen et al. (2006) and
modeled in our previous works (Buffo et al., 2012; Petitti et al.,
2010). This system is a 0.194 m3 tank with four baes (with a particular conguration with upper extensions), agitated by a standard
six-blade Rushton turbine and with a circular metal porous sparger
of diameter equal to 3.3 cm and pores with an average diameter of
15 mm, located at 10.5 cm under the impeller. A schematic representation of the geometry is reported in Fig. 2. The motion of the
Rushton turbine is modeled by using the multiple reference frame
(MFR) approach (Wechsler et al., 1999). The computational grid is
composed of approximately 230,0 0 0 hexahedral cells: only half of
the stirred tank is considered due to geometrical symmetry and
system periodicity. The operating condition considered consists of
an air ow rate of 0.093 volumes of gas per volume of reactor per
minute (vvm) and a stirring rate of 250 rpm, leading to a global
hold-up of about 1.5%. Such operating condition may lead to the
formation of gas cavities behind the impeller blades, regions where
the local gas volume fraction may signicantly differ from the
rest of the vessel. For further details on boundary conditions, the
reader is referred to previous works (Buffo et al., 2012; Petitti et al.,
2010). For this case the following simulations are performed: rst
the moments are transported by using Eq. (5) without ensuring

230

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

Fig. 2. Graphical representation of the three-dimensional grids used in this work. Left: rectangular section partially aerated bubble column. Right: aerated stirred tank with
a porous sparger.

Table 1
L1 -error of different schemes for the 1D case.
Num. scheme

Num. of cells

L1 -error

10
25
50
100
200

0.0472
0.0272
0.0195
0.0136
0.0097

Standard
second-order
upwind

10
25
50
100
200

0.0342
0.0196
0.0129
0.0088
0.0069

Realizable
second-order
upwind

10
25
50
100
200

0.0328
0.0146
0.0096
0.0058
0.0038

First-order
upwind

Fig. 3. Solution of the mono-dimensional Riemann problem for a computational


grid composed by 100 cells. Solid line: analytical solution; dash-dot line: rst-order
upwind; dotted line: second-order upwind; dashed line: realizable second-order
upwind.

boundedness with the rst-order upwind as a discretization


scheme for the moments and then by using Eq. (16) ensuring
boundedness with the rst-order upwind discretization scheme.
The results obtained for these different cases are discussed in
the next section.
5. Results and discussion
Let us start the discussion with the simplied spatially monodimensional Riemann problem. The results obtained by using
different discretization schemes are shown in Fig. 3. As evident
from the picture, the analytical solution of the problem (solid line)
is a steep gradient, smoothed out in the numerical solutions by
all the numerical schemes adopted here, due to the unavoidable
numerical diffusion. As expected the solution obtained with the
rst-order upwind scheme (dash-dot line) shows the largest deviation from the analytical solution without producing overshoots.
The standard second-order upwind scheme (dotted line), instead,
produces overshoots after the steep gradient, since the front is
moving from left to right of the domain: the values of the moments are negative in this region and the inversion of the moment
set would lead to non-realizable quadrature weights and nodes. It
should be noticed that it was possible to show the solution for this
case just because any other physical process other than advection
was neglected, and hence there was no need to use an inversion
algorithm: otherwise, the use of a corrupted moment set, resulting
in an unrealizable reconstruction for the NDF, would make the

simulation highly unstable. Eventually, the realizable second-order


upwind scheme (dashed line) is less diffusive than the rst-order
scheme and results in a solution closer to the analytical solution, without unphysical overshoots and guaranteeing that the
moment set is always realizable in the entirety of the domain. A
quantitative analysis of the error for the mono-dimensional case
is reported in Table 1, where it is evident that the error steadily
decreases with the increase of the number of cells used in the
computational grid and is smaller for the realizable second-order
upwind scheme.
Let us now consider the results from the rst three-dimensional
test case: the rectangular bubble column. In Fig. 4, the instantaneous contour plots for the gas volume fraction and for the moment of order three with respect to bubble size are reported at
t = 2 s for the case in which the moments are transported by using the procedure described in Eq. (5) that does not ensure boundedness of the moments and with the rst-order upwind spatial
discretization scheme for the moments. As depicted in the gure,
the gas phase is fed from the bottom of the column and leaves
from the top, after the phase inversion located in correspondence
of the liquid free-surface. The simulation time is not enough to see
the characteristic oscillating behavior of the bubble plume (Peger
et al., 1999), which is visible just after about 10 seconds. In Fig. 4,
some cells of the domain located above the liquid free-surface
show higher values for M3 with respect to the maximum allowable value, hence violating the boundedness constraint. It is worth
reminding that the moments of the bubble distribution are not

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

231

Fig. 4. Instantaneous contour plots of d (left) and M3 (right) in the central plane of the rectangular bubble column at t = 2 s for the case in which the moment transport
equations are implemented by using Eq. (5).

Fig. 5. Instantaneous contour plots of d (left) and M3 (right) in the central plane of the rectangular bubble column at t = 2 s for the case in which the moment transport
equations are implemented by using the preserving boundedness procedure reported in Eq. (16).

dened when the disperse phase becomes continuous (namely


when d 1) because the bubbles are no longer dispersed in that
region. In the particular case considered in this work, the area in
which the moment boundedness is not preserved is not particularly important, since it is outside of the region of interest. However, there are situations in which this is not true: it is always
possible that the disperse phase tends to accumulate in particular
zones of the domain (usually characterized by low vorticity). Since
the equations of the moments are solved in all the computational
domain and not just when the disperse phase is actually disperse,
the boundedness and the realizability of the moment sets must be
preserved for the stability of the simulation. Therefore, this situation is not acceptable.
In Fig. 5, the instantaneous contour plots for the gas volume
fraction and for the moment of order three with respect to bubble

size are reported at t = 2 s. In this case the moments are transported by using Eq. (16), with the numerical implementation that
ensures boundedness of the moment values by using the rstorder upwind as discretization scheme. In this case, it is possible to
see that the two contour plots are similar. As expected M3 is now
bounded between values corresponding to d = 0 and d = 1 in
all the computational domain. The differences in the two contour
plots are in this case due to the different discretization schemes
used for the volume fraction (blended upwind scheme with van
Leer ux-limiter) and M3 (rst-order upwind); these are particularly evident in the zone just above the upper gas-liquid interface,
where the solution of d is less diffusive than M3 , since a lower order scheme is used for the moments on a coarse grid. Further simulations on rened meshes (not reported here) showed that this
difference can be signicantly reduced, as expected.

232

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

Fig. 6. Proles of normalized M3 at the central line for different heights of the bubble column. Dashed line: unbounded numerical procedure of Eq. (5) and rst-order
upwind discretization scheme for the moment transport equation. Solid line: bounded numerical procedure of Eq. (16) and rst-order upwind discretization scheme for the
moment transport equation. Left: time t = 2 s. Right: time t = 90 s.

Fig. 7. Instantaneous contour plots of the impeller region of the investigated aerated stirred tank, calculated with an unbounded rst order upwind scheme for the moments
of the NDF (Eq. (5)). Left: disperse phase volume fraction, d . Right: Third-order moment of the NDF with respect to bubble size, M3 .

Fig. 8. Instantaneous contour plots of the impeller region of the investigated aerated stirred tank, calculated with the proposed preserving boundedness approach for the
moments of the NDF (Eq. (16)). Left: disperse phase volume fraction, d . Right: Third-order moment of the NDF with respect to bubble size, M3 .

Fig. 6 reports the horizontal proles for the moment of order three with respect to the bubble size plotted on the central line of the column at different heights and different times.
The proles are normalized to the maximum value of M3 , namely
d /kV = 6/ 1.90986, in such a way that the points where the
normalized M3 is larger than unity highlight the locations where
boundedness is violated. As seen from the gure, when the moments transport equations are implemented using Eq. (5) the normalized M3 is larger than one at the height of the column H = 0.46
m, where the liquid free surface is located. When the moment
transport equation is instead implemented by using Eq. (16), the

values of M3 lay between the proper range also close to the free
surface.
Figs. 7 and 8 report a similar comparison between different
implementation approaches of the moment transport equation
for the aerated stirred tank case. As previously mentioned, under certain operating conditions the formation of gas cavities behind the blades of the impeller can be observed, namely zones
in which the gas may accumulate due to the local pressure gradients. Fig. 7 clearly shows how an unbounded scheme may lead
to numerical problems that affects the correct evaluation of the
moments evolution, jeopardizing the stability of the simulation. In

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

233

Fig. 9. Prole of normalized M3 on a line belonging to the 45 degree plane between two baes, at the height of H = 0.22 m (corresponding approximately to the height of
the impeller). Dashed line: unbounded numerical procedure of Eq. (5) and rst-order upwind discretization scheme for the moment transport equation. Solid line: bounded
numerical procedure of Eq. (16) and rst-order upwind discretization scheme for the moment transport equation. Left: time t = 1 s. Right: time t = 2.4 s.

Fig. 10. Instantaneous contour plots of d (left) and M3 (central and right) in the central plane of the rectangular bubble column at t = 30 s for the case in which the
moment transport equations are implemented by using the preserving boundedness scheme reported in Eq. (24). The gure at the center is obtain when rst-order upwind
is used, instead at right the realizable second-order upwind discretization scheme for the moment equations.

this case, in fact, it is not possible to proceed further than 2.4 s


since in few cells of the domain innite values for the moments
are found, interrupting the simulation. As clearly shown, the zone
where the disperse phase tends to accumulate is in this case inside the region of interest, and therefore it is very important to
preserve the boundedness of the moments. By using the proposed
bounded implementation approach for the same case, as depicted
in Fig. 8, not only the obtained results are meaningful, but the simulation is also numerically stable for its entire duration.
Fig. 9 shows the prole of the normalized M3 on one radial line
located on the 45 degree plane between the two baes at the
height H = 0.22 m (slightly above the impeller) at two different
times. As previously pointed out, in this case the values of M3 are
several order of magnitude higher than one if the moment transport equations are solved by using the unbounded procedure: this
fact causes the interruption of the simulation few time steps after
2.4 s due to the very large and unphysical values of the moments.
Instead, with the use of the bounded procedure, the moments are
properly calculated and the radial proles are reasonable.
In Fig. 10, the comparison between the rst-order upwind and
a realizable second-order upwind discretization schemes in terms
of the moment of order three with respect to bubble size, M3 ,
is reported for t = 30 s for the rectangular bubble column case.
In this case, the oscillating behavior of the bubble plume can be
visualized since the simulated time is large enough to see the
lateral motion of the central plume. As reported in the previous
sections, in this case the moment uxes are calculated by using
the face-interpolated values of weights and nodes of the quadrature: the former by using second-order upwind and the latter

by using the rst-order upwind, building a realizable high-order


scheme.
It is clear that the numerical diffusion plays an important role
in this test case: the solution obtained with the rst-order is
more diffusive than the realizable quasi second-order discretization
scheme. This behavior was here enhanced by the use of a coarse
computational grid; however, the use of high-order discretization
schemes is always advisable because it is able to improve the quality of the solution without increasing the computational time. In
fact, as it is clearly shown in Fig. 10, the contour plot of M3 is
similar to the prole of the gas volume fraction (with which the
proportionality holds) obtained in turn by means of a high-order
discretization scheme (i.e., blended upwind scheme with van Leer
ux-limiter).
The only difference is in the upper part of the column, where
the gaseous phase becomes continuous: since the moments are
inverted in all the computational domain for calculating the moment uxes, the values must be realizable in every cell of the domain and thus far from the continuous phase limit ( d 1). This
is achieved by imposing as initial condition the value of M3 = 0
also for this region of the domain, and considering coalescence and
breakage only for a certain range of disperse phase volume fraction
values (as previously mentioned, 1 106 < d < 0.8). The value of
the moments where the gas is not disperse is in fact not meaningful from a physical point of view, since the bubbles are no more
disperse and therefore the NDF is no more dened. It should be
noticed that, when the rst-order upwind is used, the moment advection term is calculated without inverting the moment set, and
therefore the value of the moment of order three with respect to

234

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235

Fig. 11. Proles of normalized M3 at the central line for different heights of the bubble column. Solid line: bounded procedure of Eq. (16) with rst-order upwind discretization scheme. Dashed line: bounded procedure of Eq. (16) with realizable second order discretization scheme. Left: time t = 30 s. Right: time t = 75 s.

bubble size is exactly: M3 = d /kV = 6/ , in the upper zone of the


column, without compromising the stability of the simulation.
The differences between rst-order upwind and realizable
second-order upwind scheme are evident also in Fig. 11, where
the proles of the normalized M3 on the central line at different
heights of the bubble column are reported. As it is possible to note,
the solution obtained with the high-order discretization scheme
shows sharper proles than the solution obtained with the rstorder scheme, due to the reduced numerical diffusion. Moreover,
it may be seen also that the values of the normalized M3 in the
upper part of the column are always lower than one in both cases,
since the boundedness is preserved by the adopted numerical
procedure.
6. Conclusions
In this work, the problem of moment boundedness and realizability in QBMM is considered. These aspects are closely connected
to often overlooked numerical details concerning the implementation of QBMM into CFD codes. Since these constraints have to be
satised in a consistent way, a novel implementation is proposed,
based on the work of Weller (2002) for the numerical solution
of the disperse phase volume fraction. This methodology is here
presented in conjunction with realizable higher-order spatial discretization schemes (Vikas et al., 2011) and is implemented for the
rst time in the open-source CFD code OpenFOAM. Three different
geometries are investigated as test cases for different numerical
implementations of the moment transport equation. The analysis
of the results shows that the proposed numerical approach is capable of preserving the desired properties of the moment set in a
stable and ecient way. When a realizable higher-order discretization scheme is used for the moments, the numerical diffusion of
the solution is signicantly reduced, allowing to use a coarser grid
and therefore decreasing the computational costs, without resulting in moment corruption or stability issues.
References
Buffo, A., Marchisio, D., Vanni, M., 2014. On the implementation of moment transport equations in OpenFOAM to preserve conservation, boundedness and realizability. In: Proceedings of the 13th International Conference on Multiphase
Flows in Industrial Plants, Sestri Levante, Italy.
Buffo, A., Marchisio, D.L., 2014. Modeling and simulation of turbulent polydisperse
gas-liquid systems via the generalized population balance equation. Rev. Chem.
Eng. 30, 73126.
Buffo, A., Marchisio, D.L., Vanni, M., Renze, P., 2013a. Simulation of polydisperse
multiphase systems using population balances and example application to bubbly ows. Chem. Eng. Res. Des. 91, 18591875.

Buffo, A., Vanni, M., Marchisio, D.L., 2012. Multidimensional population balance
model for the simulation of turbulent gasliquid systems in stirred tank reactors. Chem. Eng. Sci. 70, 3144.
Buffo, A., Vanni, M., Marchisio, D.L., Fox, R.O., 2013b. Multivariate quadrature-based
moments methods for turbulent polydisperse gasliquid systems. Int. J. Multiph.
Flow 50, 4157.
Desjardins, O., Fox, R.O., Villedieu, P., 2008. A quadrature-based moment method for
dilute uid-particle ows. J. Comput. Phys. 227, 25142539.
Diaz, M.E., Iranzo, A., Cuadra, D., Barbero, R., Montes, F.J., Galan, M.A., 2008. Numerical simulation of the gasliquid ow in a laboratory scale bubble column
inuence of bubble size distribution and non-drag forces. Chem. Eng. J. 139,
363379.
Drew, D., Passman, S., 1998. Theory of Multicomponent Fluids. Springer, New York,
USA.
Ferreira, V., de Queiroz, R., Lima, G., Cuenca, R., Oishi, C., Azevedo, J., McKee, S.,
2012. A bounded upwinding scheme for computing convection-dominated
transport problems. Comput. Fluids 57, 208224.
Ferziger, J.H., Peric, M., 2001. Computational Methods for Fluid Dynamics, third ed.
Springer, Berlin, Germany.
Herrmann, M., Blanquart, G., Raman, V., 2006. Flux corrected nite volume scheme
for preserving scalar boundedness in reacting Large-Eddy simulations. AIAA J.
44, 28792886.
Hill, D.P., 1998. The Computer Simulation of Dispersed Two-phase Flows. University
of London, Imperial College of Science, Technology and Medicine, London, UK
(Ph. d. thesis).
Ishii, M., Mishima, K., 1984. Two-uid model and hydrodynamic constitutive relations. Nucl. Eng. Des. 82, 107126.
Kataoka, I., 1986. Local instant formulation of two-phase ow. Int. J. Multiph. Flow
12, 745758.
Laakkonen, M., Alopaeus, V., Aittamaa, J., 2006. Validation of bubble breakage, coalescence and mass transfer models for gas-liquid dispersion in agitated vessel.
Chem. Eng. Sci. 61, 218228.
LeVeque, R.J., 2002. Finite Volume Methods for Hyperbolic Problems, rst ed. Cambridge University Press, Cambridge, UK.
Marchisio, D.L., Fox, R.O., 2005. Solution of population balance equations using the
direct quadrature method of moments. J. Aerosol Sci. 36, 4373.
Marchisio, D.L., Fox, R.O., 2013. Computational Models for Polydisperse Particulate
and Multiphase Systems. In: Cambridge Series in Chemical Engineering. Cambridge University Press, Cambridge, UK.
Marchisio, D.L., Vigil, R.D., Fox, R.O., 2003. Quadrature method of moments for aggregation-breakage processes. J. Colloid Interface Sci. 258, 322334.
McGraw, R., 1997. Description of aerosol dynamics by the quadrature method of
moments. Aerosol Sci. Technol. 27, 255265.
McGraw, R., 2012. Correcting transport errors during advection of aerosol and
cloud moment sequences in Eulerian models. In: Druyan, L. (Ed.), Climate Models. InTech, pp. 297310. Available from: http://www.intechopen.com/books/
climate-models.
OpenFOAM Foundation Ltd., 2015a. OpenFOAM the open source CFD toolbox
Programmerss guide.
OpenFOAM Foundation Ltd., 2015b. OpenFOAM the open source CFD toolbox
Users guide.
Passalacqua, A., Fox, R.O., 2011. Implementation of an iterative solution procedure
for multi-uid gas-particle ow models on unstructured grids. Powder Technol.
213, 174187.
Petitti, M., Nasuti, A., Marchisio, D.L., Vanni, M., Baldi, G., Mancini, N., Podenzani, F., 2010. Bubble size distribution modeling in stirred gas-liquid reactors
with QMOM augmented by a new correction algorithm. AIChE J. 56, 3653.
Pea-Monferrer, C., Passalacqua, A., Chiva, S., Muoz-Cobo, J., 2016. CFD modelling
and validation of upward bubbly ow in an adiabatic vertical pipe using the
quadrature method of moments. Nucl. Eng. Des. 301, 320332.

A. Buffo et al. / International Journal of Multiphase Flow 85 (2016) 223235


Peger, D., Gomes, S., Gilbert, N., Wagner, H.G., 1999. Hydrodynamic simulations of
laboratory scale bubble columns fundamental studies of the EulerianEulerian
modelling approach. Chem. Eng. Sci. 54, 50915099.
Ramkrishna, D., 20 0 0. Population Balances: Theory and Applications to Particulate
Systems in Engineering, rst ed. Academic Press, San Diego, USA.
Rusche, H., 2002. Computational Fluid Dynamics of Dispersed Two-Phase Flows at
High Phase Fractions. University of London, Imperial College of Science, Technology and Medicine, London, UK (Ph. D. thesis).
Shohat, J.A., Tamarkin, J.D., 1943. The Problem of Moments. American Mathematical
Society, New York, USA.
Silva, L.F.L.R., Damian, R.B., Lage, P.L.C., 2008. Implementation and analysis of numerical solution of the population balance equation in CFD packages. Comput.
Chem. Eng. 32, 29332945.
Silva, L.F.L.R., Lage, P.L.C., 2011. Development and implementation of a polydispersed
multiphase ow model in OpenFOAM. Comput. Chem. Eng. 35, 26532666.
Song, B., Liu, G., Lam, K., Amano, R., 20 0 0. On a higher-order bounded discretization
scheme. Int. J. Numer. Methods Fluids 32, 881897.
Vikas, V., Wang, Z.J., Passalacqua, A., Fox, R.O., 2011. Realizable high-order nite-volume schemes for quadrature-based moment methods. J. Comput. Phys. 230,
53285352.
Waterson, N., Deconinck, H., 2007. Design principles for bounded higher-order convection schemes a unied approach. J. Comput. Phys. 224, 182207.

235

Wechsler, K., Breuer, M., Durst, F., 1999. Steady and unsteady computations of turbulent ows induced by a 4/45 pitched-blade impeller. J. Fluids Eng. Trans. ASME
121, 318329.
Weller, H.G., 2002. Derivation, Modelling and Solution of the Conditionally Averaged
Two-phase Flow Equations. Technical Report. Nabla Ltd.. TR/HGW/02
Weller, H.G., Tabor, G., Jasak, H., Fureby, C., 1998. A tensorial approach to computational continuum mechanics using object-oriented techniques. Comput. Phys.
12, 620631.
Wright Jr., D.L., 2007. Numerical advection of moments of the particle size distribution in eulerian models. J. Aerosol Sci. 38, 352369.
Yuan, C., Fox, R.O., 2011. Conditional quadrature method of moments for kinetic
equations. J. Comput. Phys. 230, 82168246.
Yuan, C., Laurent, F., Fox, R.O., 2012. An extended quadrature method of moments
for population balance equations. J. Aerosol Sci. 51, 123.
Zhu, J., 1992. On the higher-order bounded discretization schemes for nite volume
computations of incompressible ows. Comput. Methods Appl. Mech. Eng. 98,
345360.
Zucca, A., Marchisio, D.L., Vanni, M., Barresi, A.A., 2007. Validation of bivariate DQMOM for nanoparticle processes simulation. AIChE J. 53, 918931.

Potrebbero piacerti anche