1 s2.0 S0009250919306360 Main

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Chemical Engineering Science 208 (2019) 115150

Contents lists available at ScienceDirect

Chemical Engineering Science


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

An analytical solution to the fixed pivot fragmentation population


balance equation
Andreas Håkansson
Department of Food Technology, Engineering and Nutrition, LTH at Lund University, Lund, Sweden

h i g h l i g h t s

 The fixed pivot technique (FPT) is one of the most used PBE discretization methods.
 An analytical solution to the fragmentation FPT is presented.
 It agrees with previously presented analytical solutions and to numerical results.
 Computational cost is reduced by a factor 17–40 compared to the traditional method.

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

Article history: Drop breakup in high-energy emulsification is often modelled using a population balance equation (PBE)
Received 30 March 2019 framework. For cases with high emulsifier to disperse phase concentration, the PBE is dominated by the
Received in revised form 16 May 2019 fragmentation term. Since there are no analytical solutions to the continuous PBE for physically reason-
Accepted 4 August 2019
able fragmentation rate expressions, a discretization is often needed to evaluate the PBE, turning it into a
Available online 6 August 2019
set of ordinary differential equations that can be solved numerically. The fixed pivot technique (Kumar
and Ramkrishna, 1996) is one of the most often applied class discretization methods. This contribution
Keywords:
suggests an analytical solution to the fixed pivot technique fragmentation equation, that can be used
Population balance equation
Population balance model
instead of the traditional numerical approach provided that the fragmentation rate is constant over time.
Emulsification The proposed solution compares favorably to two special cases where analytical solutions for the contin-
Fragmentation uous PBE are available, and to two more realistic PBE emulsification problems (with varying fragmenta-
Break-up tion intensity and varying number of fragments formed per breakup), while offering a substantial
reduction in computational time compared to the traditional approach of solving the discretized equa-
tions numerically.
Ó 2019 Elsevier Ltd. All rights reserved.

1. Introduction and Håkansson (2016) for reviews. These investigations show that
if the disperse phase volume fraction is sufficiently low and if the
The population balance equation (PBE) has been applied exten- emulsifier concentration is sufficiently high, coalescence has a neg-
sively to describe the evolution of the particle size distribution ligible effect on the process (Henry et al., 2009; Lobo et al., 2002;
(PSD) of disperse systems (Ramkrishna, 2000). High-energy emul- Mohan and Narsimhan, 1997; Taisne et al., 1996). Under these con-
sification–such as that occurring in high-pressure homogenizers ditions the evolution of the PSD will be governed by the fragmen-
and rotor-stator mixers (Håkansson, 2019; Walstra, 2005)–is an tation terms of the PBE (Ramkrishna, 2000),
application with special relevance for several branches of chemical Z 1
@nðv ; tÞ
engineering. The emulsification process is often described as a ¼ gðv Þnðv ; tÞ þ f ðv ; v 0 Þgðv 0 Þnðv 0 ; tÞdv 0 ; ð1Þ
@t v
combination of fragmentation of large initial drops and (re-)
coalescence of insufficiently stabilized fragments (Håkansson where g(v) is the fragmentation rate and f(v,v0 ) is the fragment size
et al., 2009; Maindarkar et al., 2015; Raikar et al., 2009, 2010, distribution.
2011). The relative importance of coalescence has been the subject No general analytical solution has been presented to the contin-
of substantial experimental investigations, see Jafari et al. (2008) uous PBE. However, analytical solutions to several special cases
have been proposed, e.g., Ziff and McCrady (1985) and Ziff (1991)
for some idealized fragmentation-only problems. For more realistic
E-mail address: andreas.hakansson@food.lth.se kernels, emulsification researchers rely on numerical methods to

https://doi.org/10.1016/j.ces.2019.08.008
0009-2509/Ó 2019 Elsevier Ltd. All rights reserved.
2 A. Håkansson / Chemical Engineering Science 208 (2019) 115150

Nomenclature
g fragmentation rate, s1
Abbreviations I number of size classes, –
CFD computational fluid dynamics M0 total number of drops, –
FPT fixed pivot technique n(v,t)dv number of drops per unit volume with volume in
PBE population balance equation [v, v + dv], m3
PSD particle size distribution Nb number of fragments formed per breakup event, –
QMOM quadrature methods of moments ni,j contribution to size class i from a fragmentation in class
j in the discretized equations, –
Nj number of drops per unit volume in size class j, m3
Symbols 
%E relative difference between the numerical FPT-solution Nj average number of drops per volume, m6
and Eq. (13), % p, q parameters in the lognormal distribution, –
Akj constants in Eq. (8), – Sj source of drops to size class j, m3 s1
Cj constants in Eq. (A.4), – t time, s
d drop diameter, m v drop volume, m3
d32 sauter mean diameter, m x pivot volume, m3
f(v,v0 )dv number of drops created with volume in [v, v + dv] due
to a fragmentation of a drop of volume v0 , –

solve the PBE. The two dominant types of numerical PBE methods result and device hydrodynamics (Becker et al., 2014; Janssen and
are the ‘moment methods’ and the ‘class methods0 . Moment based Hoogland, 2014), the computational cost of using the FPT becomes
methods such as the QMOM (quadrature method of moments) considerable. In general, each size class in the discretization intro-
(Marchisio et al., 2003; McGraw, 1997) translates the PBE to a duces an additional transport equation to the CFD. A reduction in
set of discrete equations for moments of the distributions, solve the computational cost of solving the discretized equations could,
for these and attempt to reconstruct the PSD from the moments. therefore, substantially decrease computation cost of the PBE-CFD.
These methods often have a relatively low computational cost Some attempts have been made to give an analytical solution to
but might suffer from some drawbacks when difficulties arise with the discretized equations. Reid (1965) presented a general solution
the reconstruction. The class methods constitutes the other domi- to the similar case of batch grinding. Vankova et al. (2007)
nant group of methods to discretize and solve the PBE numerically. sketched a method for back-calculating size classes as a part of a
The drop size (diameter or volume) range is divided into a finite suggestion for an inverse method to determine fragmentation
number of classes and the PBE is translated into a set of ordinary rates, but did not present an explicit solution. This contribution
differential equations for the number of drops in each size class, uses a similar technique to derive an explicit analytical solution
Nj. Several such methods have been suggested, see Ramkrishna that can be used for any time independent fragmentation rate
(2000) and Kumar et al. (2009) for reviews. The fixed pivot tech- and fragment size distribution to solve the FPT fragmentation
nique (FPT) (Kumar and Ramkrishna, 1996) is one of the most equations.
widely used class methods, due to its simplicity and its suitability The rest of the text is organized as follows. Section 2 gives the
for use with a general discretization of the size range. Whereas no derivation for the analytical solution. Section 3 tests the solution
discretization (on any finite sized grid) will provide an exact repre- by comparing it to specific analytical solutions to the continuous
sentation of the continuous PBE, the FPT captures two of the PBE and to traditional numerical methods for solving the FPT equa-
moments of the continuous problem exactly. The FPT proposes tions. Section 4 provides a short discussion and Section 5 concludes
the following discretization of Eq. (1): the study.

dNj XI
¼ g j  Nj þ n  g i  Ni ;
i¼j j;i 2. Theory
dt ð2Þ
g j ¼ gðxj Þ:
First assume that the fragmentation rate and the fragment size
In Eq. (2), xj is the representative volume of class j, the ‘pivot0 , distribution are independent of time. This is a reasonable assump-
positioned at the center of each size class, i.e., if vj and vj+1 are the tion when using the type of semi-spatially resolved PBE modelling
limits of size class j, then xj = (vj + vj+1)/2. The distribution matrix, that is common in emulsification studies, e.g., dividing the device
nj,i, describes the contribution to class j following a fragmentation into a series of slices or stages, each of which is assumed to have
event in class i. Note that nj,j is not necessarily 0 in the FPT, since a uniform level of turbulence, see Håkansson et al. (2009) and
part of a fragment can be assigned to the same class as the mother Maindarkar et al. (2015) for a detailed discussion and examples.
drop to achieve the desired conservation of moments. The time-independent fragmentation rate can also be a reasonable
When choosing to keep the number of drops and the total mass assumption when using the PBE coupled to a steady-state CFD sim-
of disperse phase (zeroth and first moment) equal to that of the ulation, either when the flow is in the laminar regime or when a
continuous formulation, nj,i is given by (Kumar and Ramkrishna, Boussinesq closure is used under turbulent conditions. When the
1996) condition of time-independent fragmentation is met, the number
Z Z
xjþ1
xjþ1  v xj
v  xj1 of drops in the largest size class, NI, can be obtained directly by
nj;i ¼ f ðv ; xi Þdv þ f ðv ; xi Þdv : ð3Þ integrating the discretized PBE (Eq. (2)) for j = I:
xj xjþ1  xj xj1 xj  xj1
dNI
As previously noted, the FPT has several advantages. However, ¼ g I  NI  ð1  nI;I Þ; ð4Þ
dt
with the increasing interest in coupling the PBE to computational
fluid dynamics (CFD) to investigate the link between emulsification resulting in,
A. Håkansson / Chemical Engineering Science 208 (2019) 115150 3

NI ðt Þ ¼ NI ð0Þ  expðg I  ð1  nI;I Þ  t Þ: ð5Þ Note that each Al,j only depends on the Ak,i from larger size
classes (i.e., it only depends on indices k = j + 1. . .I). Moreover, the
Once we have an explicit expression for NI(t), the number of
value for the largest size class is given directly by the initial condi-
drops in the second largest size class, NI1, can be obtained by
tion AI,I = NI(0) according to Eq. (8). Thus, all Al,j values can be cal-
inserting Eq. (5) into the FPT equation (Eq. (2)) for j = I  1:
culated explicitly from Eqs. (8) and (12) by starting from the
dNI1 largest size class, j = I, and continuing stepwise down to the small-
¼ g I1  NI1  ð1  nI1;I1 Þ þ g I  NI ð0Þ  nI1;I
dt est size class, j = 1. A MATLAB implementation of the scheme has
 expðg I  ð1  nI;I Þ  t Þ: ð6Þ been supplied in the supporting material to further clarify the
procedure.
After integration and algebraic manipulation, this results in the In summary, a general solution to the fragmentation FPT equa-
solution: tions is provided by:
 0
 0
NI1 ðtÞ ¼ NI1 ð0Þ  A  expðg I1  ð1  nI1;I1 Þ  tÞ þ A
X
I

 expðg I  ð1  nI;I Þ  tÞ; ð7aÞ Nj ðtÞ ¼ Ak;j expðg k  ð1  nk;k Þ  tÞ ð13aÞ


k¼j

nI1;I  g I  NI ð0Þ 8
A0 ¼ : ð7bÞ PI
> 1
g I  ð1  nI;I Þ  g I1  ð1  nI1;I1 Þ < gk ð1nk;k Þgj ð1nj;j Þ i¼jþ1 nj;i  g i  Ak;i
> k ¼ j þ 1; :::I
Ak;j ¼ N ð0Þ  PI k¼j ð13bÞ
This backward-calculation approach can be continued (at the >
> j i¼jþ1 Ai;j
expense of increasingly complicated algebra) to obtain an analyti- :
0 k<j
cal solution for each size class, see Vankova et al. (2007) for a sim-
ilar attempt for the special case of n(j,j) = 0, j = 1,. . .I). However, the Note that Eq. (13) also includes the exact analytical solution to
solution can also be formulated explicitly and more compact, using the discrete fragmentation PBE (i.e., as compared to Eq. (2) which is
a method similar to what Reid (1965) used for batch grinding. a discretization of a continuous PBE), this corresponds to the
From inspection of Eqs. (5) and (7), it is reasonable to assume that expression presented by Reid (1965) for batch grinding when n
a solution to Eq. (2) can be found on the form: (k,k) = 0 for all k = i. . .I.
The same procedure as outlined above can also be extended to
X
I
some other cases. One such case of special interest is when there is
Nj ðtÞ ¼ Ak;j expðg k  ð1  nk;k Þ  tÞ; ð8Þ
a net influx term to each size class. This could arise either when the
k¼j
PBE describes a case of simultaneous fragmentation and nucleation
with time-independent constants Ak,j. To fulfill the initial condition or when the fragmentation-only PBE is coupled to steady-state
on the number of drops in each size class, Nj(t = 0) = Nj(0), such a CFD. Since material can then be transported between the CFD mesh
solution must fulfill, points, this introduces a source term, Sj, to each of the FPT equa-
X
I tions–the explicit formulation of which will depend on the dis-
Nj ð0Þ ¼ Ak;j ; ð9Þ cretization scheme used for the convection-diffusion term of the
k¼j CFD. A solution for the case of a set of FPT equations with a source
term, following the same principles as described above, is pre-
which can be formulated as a requirement on Aj,j:
( sented in Appendix A.
PI
Nj ð0Þ  m¼jþ1 Am;j j ¼ i:::I  1
Aj;j ¼ : ð10Þ
Nj ð0Þ j¼I 3. Results

Inserting Eq. (8) in the FPT equations (Eq. (2)) for a general size 3.1. Comparison to analytical solutions
class j, results in:
PI 3.1.1. Test case I: g(v) = v
m¼j  g m  ð1  nm;m Þ  Am;j expðg m  ð1  nm;m Þ  tÞ ¼
PI PI PI For a fragmentation rate proportional to the drop volume (g(v)
¼ i¼j nj;i  g i k¼j Ak;i expðg k  ð1  nk;k Þ  tÞ  g j m¼j Am;j expðg m  ð1  nm;m Þ  tÞ:
= v), a uniform fragment size distribution with an average of two
ð11Þ
fragments formed per breakup ( f(v,v0 ) = 2/v0 ) and a monodisperse
Note that both sides of the equality sign in Eq. (11) consists of initial drop size distribution located in the largest size class vI = 1
sums of exponential factors of the form expðg l  ð1  nl;l Þ  tÞ, (n(v,0) = d(v  vI)), Ziff and McCrady (1985) have presented an ana-
where l = j. . .I. The equality holds for each t if the factors multiply- lytical solution to the dimensionless continuous PBE with fragmen-
ing each exponential factor are equal. Identifying all factors on tation only:
each side of the equality multiplying a general exponential term  
nðv ; tÞ ¼ expðv I  tÞ  dðv  v I Þ þ 2t þ t2  ðv I  v Þ expðv  tÞ;
expðg l  ð1  nl;l Þ  tÞ, for each l = j. . .I, results in:
ð14Þ
X
I
g l  Al;j  ð1  nl;l Þ ¼ nj;i  g i  Al;j  g j  Al;j ð12aÞ where d(x) is the Dirac distribution.
i¼j
Fig. 1A displays the total number of drops as a function of time
Eq. (12a) can be used to find an explicit expression for Al,j , by (line), obtained from integration of the analytical solution in Eq.
first extracting the i = j term from the summation, (14):
Z 1
X
I
g l  Al;j  ð1  nl;l Þ ¼ nj;i  g i  Al;j  g j  Al;j  ð1  nj;j Þ; ð12bÞ M0 ðtÞ ¼ nðv ; tÞdv : ð15Þ
0
i¼jþ1
By discretizing the size range with a geometric grid, vi+1/vi = 2,
and then rearranging:
with 30 size classes, a fixed pivot discretization can be applied to
1 XI the problem. The total number of drops resulting from the tradi-
Al;j ¼ nj;i  g i  Al;i : ð12cÞ tional numeric solution (circles) and the analytical solution
g l  ð1  nl;l Þ  g j  ð1  nj;j Þ i¼jþ1
(crosses) to the FPT equations have been inserted in Fig. 1A. (All
4 A. Håkansson / Chemical Engineering Science 208 (2019) 115150

Fig. 1. (A) Total number of drops, M0, as a function of time,



t, comparing the Ziff and McCrady (1985) solution (line), the numerical solution to the FPT equations (circles) and
Eq. (13). (B) The average number of drops per size class, N j , at three points in time–102 (light gray, dotted), 104 (dark grey, dashed) and 106 (black, solid)– compared between
the three solution methods. Test case I: g(v) = v, f(v,v ) = 2/v0 , n(v,0) = d(v  vI).
0

numerical integrations of the FPT were done with the ode15s sol- Fig. 2A compares the evolution of the total number of drops and
ver in MATLAB 2017a (MathWorks, Natick, MA), using a non- Fig. 2B compares the average number of drops in the different size
negativity constraint and the numerical differentiation formula classes between the Ziff and McCrady solution to the continuous
scheme (Shampine and Reichelt, 1997)). As expected, the Ziff and equations, the traditional numerical solution to the FPT equations
McCrady solution corresponds closely to the numerical solution and Eq. (13). As in the previous case, the analytical and numerical
of the FPT equations (cf. Kumar and Ramkrishna, 1996). Moreover, solutions to the FPT equations correspond closely to each other and
both show excellent agreement with the analytical solution in Eq. to the analytical solution of the continuous PBE.
(13). Note that this holds despite the long integration time, result-
ing in a factor 106 increase in the total number of drops. 3.2. Comparison to numerical results
Fig. 1B plots the average number of drops in each size class,

Z v jþ1 Analytical solutions are only available for a limited number of
1
Nj ¼ nðv ; tÞdv ; ð16Þ PBE problems, and these differ substantially from the kernels sug-
v jþ1  v j v j
gested in drop breakup and emulsification literature (Coulaloglou
comparing the Ziff and McCrady (1985) solution, the traditional and Tavlarides, 1977; Lasheras et al. 2002; Liao and Lucas, 2009;
numerical solution of the FPT equations and Eq. (13) for three points Lou and Svendsen 1996). Therefore, Eq. (13) was subjected to fur-
in (dimensionless) time (t = 102, 104 and 106). A slight deviation can ther testing by comparing solutions with a broader range of condi-
be found between the analytical solution and the fixed pivot results tions to the results obtained by solving the FPT equations
(as previously reported by Kumar and Ramkrishna, 1996, p. 1321). numerically.
However, the analytical and numerical solutions to the FPT equa-
tions are indistinguishable. 3.2.1. Test case III: Turbulent breakup with varying number of
fragments
3.1.2. Test case II: g(v) = v2 Consider a pre-emulsion with a PSD that can be described by a
As a second test case, consider the same problem as in Sec- lognormal distribution with parameters p = 8.8 and q = 0.8
tion 3.1.1 but with a fragmentation rate proportional to the square
ðlogv  pÞ
2
1
of drop volume (g(v) = v2). Ziff and McCrady (1985) arrived at the nðv ; 0Þ ¼ pffiffiffiffiffiffiffi exp ; ð18Þ
following analytical solution to the continuous problem: v  q 2p 2q2

nðv ; tÞ ¼ expðt  v 2 Þ  dðv  v I Þ þ 2  t  v I  hðv I  v Þ: ð17Þ this corresponds to a mean diameter of 207 mm and standard devi-
ation of 0.04 mm, see illustration in Fig. 3. The fragmentation rate is

Fig. 2. (A) Total number of drops, M0, as a function of time, t, comparing the Ziff and McCrady (1985) solution (line), the numerical solution to the FPT equations (circles) and
Eq. (13). (B) The average number of drops per size class at three points in time–102 (light gray, dotted), 104 (dark grey, dashed) and 106 (black, solid)– compared between the
three solution methods. Test case II: g(v) = v2, f(v,v0 ) = 2/v0 , n(v,0) = d(v  vI).
A. Håkansson / Chemical Engineering Science 208 (2019) 115150 5

Fig. 3. The discrete volumetric PSD (xjNj) at three points in time–0 ms (black, solid),
Fig. 5. The discrete volumetric PSD (xjNj) at three points in time–0 ms (black, solid),
910 ms (dark grey, dashed) and 1400 ms (light grey, dotted)–comparing the
910 ms (dark grey, dashed) and 1400 ms (light grey, dotted)–comparing the
traditional numerical solution to the FPT equations (circles) and the analytical
traditional numerical solution to the FPT equations (circles) and the analytical
solution in Eq. (13) (crosses). Test case III (see Section 3.2.1).
solution in Eq. (13) (crosses). Test case IV (see Section 3.2.2).

assumed to be given by the model proposed by Coulaloglou and


ber of fragments, the numerical and analytical solution to the FPT
Tavlarides (1977), and each fragmentation event is assumed to give
agree closely, the maximum relative difference is less than 0.05%
rise to Nb equally sized fragments,
(Fig. 4C).
 
v0
f ðv ; v 0 Þ ¼ N b  d v : ð19Þ
Nb
3.2.2. Test case IV: turbulent breakup with varying intensity
A size discretization with vi+1/vi = 2 and 30 size classes, ranging Lou and Svendsen (1996) suggested a rate kernel free from
from d1 = 0.25 mm to dI = 200 mm, is introduced to obtain a set of empirical fitting parameters that has attracted much attention
discrete FPT equations describing the problem. Fig. 3 shows the and been used as the basis for many later developments
discrete volumetric PSD (xjNj) at three time-points (0 ms, 910 ms (Andersson and Andersson, 2006; Becker et al., 2014; Han et al.,
and 1400 ms) for a dissipation rate of turbulent kinetic energy equal 2011; Solsvik et al., 2016). As a fourth test case, the same lognor-
to 107 m2/s3, comparable to that found in high-energy emulsifica- mal initial size distribution as used in Section 3.2.1 was combined
tion devices (Håkansson et al., 2011; Mortensen et al., 2018), with this fragmentation rate and a binary uniform fragment size
assuming binary breakup (Nb = 2). As seen in the figure, Eq. (13) distribution (f(v,v0 ) = 2/v0 ). The drop volume domain was dis-
corresponds closely to the traditional numerical fixed pivot cretized with vi+1/vi = 1.5, I = 30 and dI = 200 mm. The average num-
solution. ber of drops per size class is showed in Fig. 5 for a dissipation rate
Fig. 4A compares the total number of drops as a function of time of turbulent kinetic energy of 107 m2/s3, comparing the numerical
and Fig. 4B compares the evolution of the Sauter mean diameter, (circles) and analytical solution (Eq. (13)) (crosses) of the FPT frag-
d32, as a function of processing time (t = 0. . .10 ms) for the same mentation equation. Good agreement can be seen in the figure.
problem. As expected, the mean diameter decreases (and M0 Fig. 6 compares the total number of drops (Fig. 6A) and the Sau-
increases) initially, until it approaches a steady-state PSD. Fig. 4 ter mean diameter (Fig. 6B) as a function of time, for three different
displays these results for three different assumptions on the num- dissipation rates of turbulent kinetic energy (106, 107 and 108 m2/
ber of fragments formed per breakup event, Nb = 2, 4 and 8. A larger s3). As seen in the figure, the numerical and analytical solutions to
number of fragments results in a smaller mean drop diameter (and the FPT agrees for this case as well. The relative difference in the
larger number of drops, Fig. 4A). However, regardless of the num- mean diameter is less than 0.015% (Fig. 6C).

Fig. 4. Total number of drops, M0, (A) and Sauter mean diameter, d32, (B) as a function of time, t; comparing the numerical solution to the FPT equations (circles) and Eq. (13)
(crosses). (C) Relative differences in d32 between numerical FPT solution and Eq. (13). Results are displayed for three different number of fragments formed per breakup event:
2 (black), 4 (dark grey) and 8 (light gray. Test case III (see Section 3.2.1).
6 A. Håkansson / Chemical Engineering Science 208 (2019) 115150

Fig. 6. Total number of drops, M0, (A) and Sauter mean diameter, d32, (B) as a function of time, t; comparing the numerical solution to the FPT equations (circles) and Eq. (13)
(crosses). (C) Relative differences in d32 between numerical FPT solution and Eq. (13). Results are displayed for three different dissipation rates of turbulent kinetic energy:
106 m2/s3 (light gray), 107 m2/s3 (dark grey) and 108 m2/s3 (black). Test case IV (see Section 3.2.2).

4. Discussion (Lasheras et al., 2002; Liao and Lucas, 2009). However, it should be
noted that this limitation also applies for analytical solutions to the
Whereas the continuous fragmentation-only PBE does not gen- continuous PBE such as the ones presented by Ziff and McCrady
erally have an explicit analytical solution, the discrete form is read- (1985) and Ziff (1991).
ily solved by Eq. (13). This allows for a two-step approach for Moreover, the methodology as presented in Eq. (13) assumes
solving the PBE problem for emulsification processes with negligi- that the process is entirely governed by fragmentation. A general-
ble coalescence, using the methodology outlined in Section 3: ization to problems including coalescence can, unfortunately, not
Firstly, a class discretization is used to convert the continuous be obtained using the same procedure since the solution method-
PBE (an integro-partial differential equation) into a finite number ology relies heavily on the fact that an upper size class can be iden-
of ordinary differential equations (ODEs). Secondly, the ODEs are tified where the evolution of the number of drops does not depend
solved by Eq. (13). This can be compared to the traditional on that of any other size class, allowing for the back-calculation
approach where the second step is to solve the ODEs numerically. indicated in Section 2, and the back-calculation of the constants
The test cases in Section 3 illustrates that the two methods provide Ai,j in Eq. (13).
nearly identical results.
The analytical solution to the FPT equations suggested here has 5. Conclusions
the advantage of substantially reducing the computational time
compared to solving the FPT numerically. For the test cases An explicit analytical solution to the fragmentation-only fixed
reported in Section 3, the computational time is reduced by a factor pivot technique equation with general size dependent fragmenta-
17–40 compared to the numerical approach. Solving the ODEs tion kernels was presented. The solution compares favorably, both
resulting from the fixed pivot technique is not time demanding– to test cases where analytical solutions are available to the contin-
none of the numerical solutions in Section 3 required more than uous PBE, and to the results from the traditional numerical solution
0.5 s on a compact laptop (dual core 2.9 GHz CPU and 8 GB of to the fixed pivot discretization. This suggest an alternative
RAM). However, when used in combination with a CFD simulation method for simulating the evolution of the drop size distribution
(which is of increasing interest in emulsification research), this in emulsification systems where coalescence is negligible: by first
reduction in computational cost can be highly significant. Due to discretizing the continuous population balance equation using the
the high computational cost of class methods, CFD-PBE simulations fixed pivot technique (or alternatively, one of the other class dis-
often use moments based techniques (e.g. QMOM) instead. With cretization methods found in literature), and then using the analyt-
the analytical solution presented here, the FPT (even when using ical solution to the obtained set of ordinary differential equations.
a large number of size classes) can become an economically feasi- As seen from the test-cases, this results in a substantial reduction
ble complement to the moment based methods for these types of in computational cost.
investigations.
The focus of this contribution has been on cases when the dis- Declaration of Competing Interest
cretization of the PBE into a set of discrete ODEs is done using
the FPT. However, there are several other promising class based The authors declare that they have no known competing finan-
techniques (Kumar et al., 2009; Ramkrishna 2000). These differ cial interests or personal relationships that could have appeared
in how they do the discretization and also in terms of efficiency to influence the work reported in this paper.
and accuracy. Consequently, they will also result in a different dis-
tribution matrix, ni,k, and/or different summation indices compared Acknowledgments
to the FPT. However, they will still result in a set of ODEs compa-
rable to Eq. (2), and can, therefore, be solved using a similar This study was financed by the Swedish Research council (VR),
methodology to that presented in Section 2, for cases where there grant number 2018-03820.
is no coalescence/aggregation present.
Some limitations to the proposed method should also be men-
Appendix A. The solution with a source term
tioned. The solution assumes that the fragmentation kernels (g
and f) are independent of time (but allows for any form of volume
Assume that Sj is the source term to size class j. A source could
dependence). This makes the methodology unsuitable for coupling
arise as a net influx from adjacent cells in a coupled PBE-CFD setup,
it to setups with transient flow, since most theoretically suggested
or due to nucleation. The corresponding discretized equations (cf.
kernels depend on the dissipation rate of turbulent kinetic energy
Eq. (2)) are:
A. Håkansson / Chemical Engineering Science 208 (2019) 115150 7

dNj XI In summary, an analytical solution to Eq. (A.1) can be obtained


¼ g j  Nj þ Sj þ nj;i  g i  Ni : ðA:1Þ from:
dt i¼j
X
I
Under the assumption that the source term is independent of Nj ðtÞ ¼ C j þ Ak;j expðg k  ð1  nk;k Þ  tÞ; ðA:11aÞ
time, the solution to Eq. (A.1) can be obtained using a method sim- k¼j
ilar to that outlined in Section 2. The evolution of the number of
8  P 
drops in the largest size class, NI, can be solved directly by integra- < 1
Sj þ Ii¼jþ1 nj;i  g i  C i j ¼ 1:::I  1
g j ð1nj;j Þ
tion of Eq. (A.1): Cj ¼ ðA:11bÞ
: SI
j¼I
g I ð1nI;I Þ
dNI
¼ g I  NI  ð1  nI;I Þ þ Sj ; ðA:2Þ
dt 8 PI
> 1
resulting in: < gl ð1nl;l Þgj ð1nj;j Þ i¼jþ1 nj;i  g i  Al;i
> l¼j

  Al;j ¼ Nj ð0Þ  C j  PI l ¼ j þ 1:::I ðA:11cÞ


>
> m¼jþ1 Am;j
SI SI :
NI ðtÞ ¼ þ NI ð0Þ  expðg I  ð1  nI;I Þ  tÞ: 0 l<j
g I  ð1 þ nI;I Þ g I  ð1 þ nI;I Þ
ðA:3Þ As expected, Eq. (A.11) simplifies to Eq. (13) when Sj = 0 for all
j = 1, . . . , I.
This suggests that a more general solution can be found on the
form:
Appendix B. Supplementary material
X
I
Nj ðtÞ ¼ C j þ Ak;j expðg k  ð1  nk;k Þ  tÞ ðA:4Þ Supplementary data to this article can be found online at
k¼j
https://doi.org/10.1016/j.ces.2019.08.008.
with constants Ak,j and Cj. In order to fulfill the initial condition,
Nj(t = 0) = Nj(0), these constants must fulfill: References

X
I X
I Andersson, R., Andersson, B., 2006. Modeling the breakup of fluid particles in
Nj ð0Þ ¼ C j þ Ak;j ¼ C j þ Aj;j þ Ak;j ; ðA:5Þ turbulent flows. AIChE J. 52 (6), 2031–2038. https://doi.org/10.1002/aic.10832.
k¼j k¼jþ1 Becker, P.J., Puel, F., Jakobsen, H.A., Sheibat-Othman, N., 2014. Development of and
improved breakage kernel for high dispersed viscosity phase emulsification.
which can be formulated as a constraint on Aj,j: Chem. Eng. Sci. 109, 326–338. https://doi.org/10.1016/j.ces.2014.02.008.
Coulaloglou, C.A., Tavlarides, L.L., 1977. Description of interaction processes in
X
I agitated liquid-liquid dispersions. Chem. Eng. Sci. 32, 1289–1297. https://doi.
Aj;j ¼ Nj ð0Þ  C j  Am;j : ðA:6Þ org/10.1016/0009-2509(77)85023-9.
Han, L., Luo, H., Liu, Y., 2011. A theoretical model for droplet breakup in turbulent
m¼jþ1
dispersions. Chem. Eng. Sci. 66, 766–776. https://doi.org/10.1016/j.
ces.2010.11.041.
Expressions for the remaining constants are obtained by insert-
Henry, J.V.L., Fryer, P.J., Frith, W.J., Norton, I.T., 2009. Emulsification mechanism and
ing Eq. (A.4) in the discretized equations (Eq. (A.1)): storage instabilities of hydrocarbon in-water sub-micron emulsions stabilized
PI with Tweens (20 and 80), Brij 96v and sucrose monoesters. J. Colloid Interface
 g m ð1  nm;m ÞAm;j expðg m ð1  nm;m ÞtÞ ¼
m¼j Sci. 338 (1), 201–206. https://doi.org/10.1016/j.jcis.2009.05.077.
P Håkansson, A., Trägårdh, C., Bergenståhl, B., 2009. Dynamic simulation of emulsion
¼ g j C j  g j Im¼j Am;j expðg m ð1  nm;m ÞtÞ ðA:7Þ formation in a high pressure homogenizer. Chem. Eng. Sci. 64, 2915–2925.
P h P i https://doi.org/10.1016/j.ces.2009.03.034.
þ Ii¼j nj;i  g i C i þ Ik¼j Ak;i expðg k ð1  nk;k ÞtÞ Håkansson, A., Fuchs, L., Innings, F., Revstedt, J., Trägårdh, C., Bergenståhl, B., 2011.
High resolution experimental measurement of turbulent flow field in a high
Each side of the equality sign in Eq (A.7) is a sum of I-j exponen- pressure homogenizer model its implication on turbulent drop fragmentation.
Chem. Eng. Sci. 66, 1790–1801. https://doi.org/10.1016/j.ces.2011.01.026.
tial terms and one time-independent term. If both the constant Håkansson, A., 2016. Experimental methods for measuring coalescence during
terms on each side of the equality sign and the factors multiplying emulsification – a critical review. J. Food Eng. 178, 47–59. https://doi.org/
the exponentials are equal, then Eq. (A.4) is a solution to Eq. (A.1). 10.1016/j.jfoodeng.2016.01.006.
Håkansson, A., 2019. Emulsion formation by homogenization: current
For the constant terms: understanding and future perspectives. Ann. Rev. Food Sci. Technol. 10, 239–
258. https://doi.org/10.1146/annurev-food-032818-121501.
X
I
Jafari, S.M., Assadpoor, E., He, Y., Bhandari, B., 2008. Re-coalescence of emulsion
0 ¼ g j  C j þ nj;i  g i  C i droplets during high-energy emulsification. Food Hydrocoll. 22, 1191–1202.
i¼j https://doi.org/10.1016/j.foodhyd.2007.09.006.
X
I Janssen, J.J.M., Hoogland, H., 2014. Modelling strategies for emulsification in
¼ g j  C j þ nj;j  g j  C j þ nj;i  g i  C i : ðA:8Þ industrial practice. Can. J. Chem. Eng. 92 (2), 198–202. https://doi.org/10.1002/
cjce.21942.
i¼jþ1
Kumar, J., Warnecke, G., Peglow, M., Heinrich, S., 2009. Comparison of numerical
methods for solving population balance equations incorporating aggregation
Eq. (A.8) can be reformulated to give an expression for the C-
and breakage. Powder Technol. 189, 218–229. https://doi.org/10.1016/j.
constant: powtec.2008.04.014.
! Kumar, S., Ramkrishna, D., 1996. On the solution of Population Balance Equations by
1 XI
Discretization - I. A Fixed Pivot Technique. Chem. Eng. Sci. 51, 1311–1332.
Cj ¼ Sj þ nj;i  g i  C i : ðA:9Þ https://doi.org/10.1016/0009-2509(96)88489-2.
g j  ð1  nj;j Þ i¼jþ1 Lasheras, J.C., Eastwood, C., Martínez-Bazán, C., Montañés, J.L., 2002. A review of
statistical models for the break-up of an immiscible fluid immersed into a fully
For a general exponential term exp(-gl(1-nl,l) t), the equality in developed turbulent flow. Int. J. Multiph. Flow 28, 247–278. https://doi.org/
Eq. (A.7) results in (cf. Eq. (12)): 10.1016/S0301-9322(01)00046-5.
Liao, Y., Lucas, D., 2009. A literature review of theoretical models for drop and
1 XI bubble breakup in turbulent dispersions. Chem. Eng. Sci. 64, 3389–3406.
Al;j ¼ nj;i  g i  Al;i : ðA:10Þ https://doi.org/10.1016/j.ces.2009.04.026.
g l  ð1  nl;l Þ  g j  ð1  nj;j Þ i¼jþ1 Lobo, L., Svereika, A., Nair, M., 2002. Coalescence during emulsification. 1. Method
development. J. Colloid Interface Sci. 253, 409–418. https://doi.org/10.1006/
Just as in Section 2, the constants Cj, and Al,j can always be jcis.2002.8560.
Luo, H., Svendsen, H.F., 1996. Theoretical model for drop and bubble breakup in
explicitly calculated by starting from the largest size class (j = I) turbulent dispersions. AIChE J. 42, 1225–1233. https://doi.org/10.1002/
and continuing step-wise down to the smallest size class (j = 1). aic.690420505.
8 A. Håkansson / Chemical Engineering Science 208 (2019) 115150

Marchisio, D.L., Virgil, R.D., Fox, R.O., 2003. Quadrature method of moments for Raikar, N.B., Bhatia, S.R., Malone, M.F., McClements, D.J., Henson, M.A., 2011.
aggregation-breakage processes. J. Colloid Interface Sci. 258, 322–334. https:// Predicting the effect of the homogenization pressure on emulsion drop-size
doi.org/10.1016/S0021-9797(02)00054-1. distributions. Ind. Eng. Chem. Res. 50, 6089–6100. https://doi.org/10.1021/
Maindarkar, S.N., Hoogland, H., Hansen, M.A., 2015. Predicting the combined effects ie101818h.
of oil and surfactant concentrations on the drop size distributions of Ramkrishna, D., 2000. Population Balances-Theory and Applications to Particulate
homogenized emulsions. Colloids Surf. A Physicochem. Eng. Asp. 467, 18–30. Systems in Engineering. Academic Press, San Diego.
https://doi.org/10.1016/j.colsurfa.2014.11.032. Reid, K.J., 1965. A solution to the batch grinding equation. Chem. Eng. Sci. 20, 953–
McGraw, Robert, 1997. Description of aerosol dynamics by the quadrature method 963. https://doi.org/10.1016/0009-2509(65)80093-8.
of moments. Aerosol Sci. Technol. 27 (2), 255–265 http:// Shampine, L.F., Reichelt, M.W., 1997. The MATLAB ODE suite. SIAM J. Sci. Comput.
www.tandfonline.com/doi/abs/10.1080/02786829708965471. https://doi.org/ 18, 1–22. https://doi.org/10.1137/S1064827594276424.
10.1080/02786829708965471. Solsvik, J., Skjervold, V.T., Han, L., Lou, H., Jakobsen, H.A., 2016. A theoretical study
Mohan, S., Narsimhan, G., 1997. Coalescence of protein-stabilized emulsions in a on drop breakup modeling in turbulent flows: the inertial subrange versus the
high-pressure homogenizer. J. Colloid Interface Sci. 192, 1–15. https://doi.org/ entire spectrum of isotropic turbulence. Chem. Eng. Sci. 149, 249–265. https://
10.1006/jcis.1997.5012. doi.org/10.1016/j.ces.2016.04.037.
Mortensen, H.H., Innings, F., Håkansson, A., 2018. Local levels of dissipation rate of Taisne, L., Walstra, P., Cabane, B., 1996. Transfer of oil between emulsion droplets. J.
turbulent kinetic energy in a rotor-stator mixer with different stator slot Colloid Interface Sci. 184, 378–390. https://doi.org/10.1006/jcis.1996.0632.
widths-an experimental investigation. Chem. Eng. Res. Des. 130, 52–62. https:// Walstra, P., 2005. Emulsions. In: Lyklema, J. (Ed.), Fundamentals of Interface and
doi.org/10.1016/j.cherd.2017.12.006. Colloid Science. Elsevier, Amsterdam, pp. 8.1–8.94.
Raikar, N.B., Bhatia, S.R., Malone, M.F., Henson, M.A., 2009. Experimental studies and Vankova, N., Tcholakova, S., Denkov, N.D., Vulchev, V.D., Danner, T., 2007.
population balance equation models for breakage prediction of emulsion drop Emulsification in turbulent flow 2. Breakage rate constants. J. Colloid Interface
size distributions. Chem. Eng. Sci. 64, 2433–2447. https://doi.org/10.1016/j. Sci. 313, 612–629. https://doi.org/10.1016/j.jcis.2007.04.064.
ces.2009.01.062. Ziff, R.M., 1991. New solutions to the fragmentation equation. J. Phys. A: Math. Gen.
Raikar, N.B., Bhatia, S.R., Malone, M.F., McClements, D.J., Almeida-Rivera, C., Bongers, 24, 2821–2828. https://doi.org/10.1088/0305-4470/24/12/020.
P., Henson, M.A., 2010. Prediction of emulsion drop size distributions with Ziff, R.M., McCrady, E.D., 1985. The kinetics of cluster fragmentation and
population balance equation models of multiple drop breakage. Colloids Surf. A: depolymerisation. J. Phys. A: Math. Gen. 18, 3027–3037. https://doi.org/
Physiochem. Eng. Asp. 361, 96–108. https://doi.org/10.1016/ 10.1088/0305-4470/18/15/026.
j.colsurfa.2010.03.020.

You might also like