Sie sind auf Seite 1von 15

Chemical Engineering Communications

ISSN: 0098-6445 (Print) 1563-5201 (Online) Journal homepage: https://www.tandfonline.com/loi/gcec20

Probability density functions for bubble size


distribution in air–water systems in stirred tanks

Henrique J. O. Pinho, Dina M. R. Mateus & Sebastião S. Alves

To cite this article: Henrique J. O. Pinho, Dina M. R. Mateus & Sebastião S. Alves (2018)
Probability density functions for bubble size distribution in air–water systems in stirred tanks,
Chemical Engineering Communications, 205:8, 1105-1118, DOI: 10.1080/00986445.2018.1434159

To link to this article: https://doi.org/10.1080/00986445.2018.1434159

Published online: 09 Apr 2018.

Submit your article to this journal

Article views: 183

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=gcec20
CHEMICAL ENGINEERING COMMUNICATIONS
2018, VOL. 205, NO. 8, 1105–1118
https://doi.org/10.1080/00986445.2018.1434159

none defined

Probability density functions for bubble size distribution in air–water


systems in stirred tanks
Henrique J. O. Pinhoa,b , Dina M. R. Mateusa,c , and Sebastião S. Alvesb
a
Unidade Departamental de Engenharias, Instituto Politécnico de Tomar, Estrada da Serra, Quinta do Contador, Tomar, Portugal; bCERENA,
Centro de Recursos Naturais e Ambiente, Instituto Superior Técnico, Universidade de Lisboa, Lisboa, Portugal; cGEOBIOTEC, GeoBioSciences,
GeoTechnologies and GeoEngineering, Universidade de Aveiro, Campus Universitário de Santiago, Aveiro, Portugal

ABSTRACT KEYWORDS
Bubble size distribution (BSD) is relevant to the design of gas–liquid systems, as it determines the Nukiyama–Tanasawa
interfacial area available in heat and mass transfer processes. Although data on BSD in stirred distribution; particle size
aerated tanks are available, a systematic comparison of alternative modeling functions for these distribution; probability
density function; Rosin–
data is lacking. In this work, BSDs obtained in air–water dispersions in a stirred aerated tank with a
Rammler distribution; Sauter
Rushton turbine and BSDs available in the literature for similar systems were modeled by 14 mean diameter; two-phase
empirical probability density functions (PDFs). It was found that both the distribution of Nukiyama– flow
Tanasawa with three adjustable parameters and the Rosin–Rammler distribution with two
adjustable parameters reasonably fit original and literature BSDs. It is also concluded that it is
possible to correlate the PDF parameters with the power dissipated by the agitator in the liquid
phase, allowing the BSD to be modeled with only two parameters in a range of dissipated power
from 0.5 to 2.3 kW/m3. BSDs thus modeled provide good predictions of average bubble size.
Abbreviations: BE, bi-exponential distribution; BS, Birnbaum–Saunders distribution; BSD, bubble size
distribution; CDF, cumulative distribution function; CFD, computational fluid dynamics; F, Frechet
distribution; G, gamma distribution; G–L, gas–liquid; IG, inverse Gaussian distribution; J, Johnson
distribution; LH4, four parameter log-hyperbolic distribution; LL, log-logistic distribution; LN, log-
normal distribution; N, normal distribution; NT, two-parameter Nukiyama–Tanasawa distribution;
NT3, three-parameter Nukiyama–Tanasawa distribution; PDF, probability density function; PSD,
particle size distribution; RR, Rosin–Rammler distribution; UL3, upper limit distribution

Introduction Sauter mean diameter represented by d32 in


Equation (1), where Ni represents the number of
Particle size is relevant to several unit operations
bubbles with diameter di and M the number of
typically used in chemical, biochemical, energy,
different bubble diameters.
and environmental processes, among other fields
of engineering, particularly for the modeling and PM
Ni di3
scale-up of transport phenomena. Modeling of d32 ¼ Pi¼1
M 2
ð1Þ
mass and heat transfer involving a dispersed phase, i¼1 Ni di

for example, requires knowledge of the interphase To obtain d32, the bubble size distribution (BSD)
area, which depends upon particle size and is required. The BSD may also be required in
dispersed phase holdup. applications where a single average is not good
One problem in studying these phenomena enough, such as modeling with computational fluid
derives from the fact that, due to fragmentation dynamics (CFD) (Kálal et al., 2014; Montante et al.,
and dispersion dynamics, solid particles, liquid 2008; Petitti et al., 2010; Sajjadi et al., 2012), or
drops, and gas bubbles are generally found in a in validation of new experimental methods to
range of sizes instead of a single well-defined characterize dispersions (Busciglio et al., 2010).
dimension. This usually implies the use of an Mathematical description of raw BSD data is
appropriate average size. In gas–liquid heat and thus important and there are no studies available
mass transfer applications, this average is the on the analysis of alternative BSD models for

CONTACT Henrique J. O. Pinho hpinho@ipt.pt Unidade Departamental de Engenharias, Instituto Politécnico de Tomar, Estrada da Serra, Quinta do
Contador, 2300-313 Tomar, Portugal.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/gcec.
© 2018 Taylor & Francis
1106 H. J. O. PINHO ET AL.

gas–liquid (G–L) contactors. In contrast, the with diameters always positive. Several PDFs, with
particle size distribution (PSD) of granulated a variable number of parameters, may be used to
solids and gas-dispersed liquids has received fulfill these criteria. However, there is a trade-off
more attention, particularly in the energy and between fitting quality and parameter determi-
combustion fields (Babinsky and Sojka, 2002; nation difficulties and stability (Paloposki, 1994).
Dunbar and Hickey, 2000; González-Tello et al., Table 1 includes a selected set of PDFs that meet
2008; Paloposki, 1994). BSD basic criteria, most of which are based on
In most of the available literature, probability two parameters for the sake of simplicity. Normal
density functions (PDFs) are used to model PDF was also included as well as some multipara-
experimentally observed BSD. The use of such an meter PDFs usually successful when applied to
empirical approach was found to be a practical PSDs. To avoid an excessive number of symbols
and successful alternative to more complex in Table 1, a and b were used preferentially as
theoretical models (Macías-García et al., 2004). symbols to identify PDF parameters, which may
The objective of this work was to identify the not correspond to common definitions. For the
best suitable PDFs to model BSD in G–L-stirred most common or the most successful PDFs, the
contactors and to evaluate the effects of operating usual nomenclature was used to avoid misunder-
conditions on the PDFs’ parameters. To achieve standing in the following article sections.
this goal, original data were obtained from experi- Table 1 includes the PDFs’ definition and the
mental work conducted in air–water dispersions corresponding cumulative distribution function
in a lab stirred tank reactor and representative (CDF). All PDFs used are normalized density
literature data on gas–liquid systems were gathered functions according to Equation (2):
and analyzed.
Zþ1
pðxÞdx ¼ 1 ð2Þ
Probability density functions for bubble 1

size distributions The CDF is defined accordingly Equation (3):


The extensive work of Paloposki (1994) identifies Zx
and characterizes 16 examples of PDFs potentially P ð xÞ ¼ pð yÞdy ð3Þ
applicable to model PSDs. In his work, the author
1
applies six PDFs to experimental drop size distribu-
tions on air-dispersed liquids: the log-normal, upper In Table 1, erf(t) represents the error function,
limit, Rosin–Rammler and Nukiyama–Tanasawa Γ(t) represents the gamma function, and Γ(t,z) is
distributions, and the four- and three-parameter the upper incomplete gamma function. K1(t) repre-
forms of the log-hyperbolic distribution. sents the modified Bessel function of the third kind
There is no similar systematic work on BSD data of order 1.
in aerated stirred tanks. However, a few researchers Among the distributions described by Paloposki
attempted to fit PDFs to experimental BSDs: Zhang (1994), a few were not tested in the present
et al. (1989) observed that bubbles’ diameters in work: the square-root normal distribution, the
air–water system approach normal distribution; three-parameter Weibull distribution (the two-
Barigou and Greaves (1992) concluded that the parameter form of which is the Rosin–Rammler
Weibull distribution may fit the BSD of coalescing distribution), the Li–Tankin distribution, the
air–water systems, but noncoalescing systems BSD Griffith distribution, and different forms of the
are better fitted by exponential-type distributions; hyperbolic distribution. The decision not to use
Martin (1995) argues that bimodal BSDs in both these distributions resulted from the observation
coalescent and noncoalescent air–water systems that these are minor variants of other distributions,
may be modeled by a set of superimposed log- as confirmed by preliminary fitting tests. On the
normal distributions. other hand, two-parameter distributions not referred
Bubble size distributions are typically non- by Paloposki, such as the Birnbaum–Saunders and
normal, positively skewed, and positive only and the Johnson distributions, were used in this work.
CHEMICAL ENGINEERING COMMUNICATIONS 1107

Table 1. Selected probability density functions (PDFs). PDFs with more than two parameters are identified with a number in the
abbreviation. It is also represented the cumulative distribution functions (CDFs).
Name Abb. Par. PDF ¼ p(x) CDF ¼ P(x)
Bi-exponential BE a,b ab � ax � 1 � xb
� xa

e e bx a 1 e bð1 e Þ
b a a b
Birnbaum–Saunders "� � � �3=2 # ( " rffiffiffi rffiffiffi!#)
BS a,b 1 b 1=2 b 1 x b 1 1 x b
pffiffiffiffiffi þ e½ þð
2a2 b x
2Þ�
1 þ erf pffiffi
2ab 2p x x 2 a 2 b x
Frechet a
F a,b aba x 1 a e ðbÞ
x

a
eðx=bÞ
Gamma G a,b a 1 Cða; x=bÞ
x
1
ex=b ba CðaÞ CðaÞ
Inverse Gaussian rffiffiffiffiffiffiffiffiffi � � ( "rffiffiffiffi #) ( "rffiffiffiffi #)
IG a,b b bðx aÞ2 1 b� x� 1 b� x�
e 2xa2 1 erf 1 þ eð2b=aÞ 1 erf 1þ
2px 3 2 2x a 2 2x a
Johnson � � ��
J a,b b 1 2 1 a þ b ln x
pffiffiffiffiffi e 2ðaþb ln xÞ 1 þ erf pffiffi
x 2p 2 2
Log-hyperbolic pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi �pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ffi � pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Z �p ffiffi2ffiffiffiffiffiffiffiffiffiffiffiffiffi2ffi �
LH4 a,b,δ,µ a2 b2 2 a2 b 2 x
� pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi� e d þðx lÞ þbðx lÞ
� pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi� e d þðy lÞ þbðy lÞ
dy
2
2adK1 d a 2 b 2adK1 d a2 b2 0
Log-logistic
LL a,b axa 1 xa
ðb þ a
x a Þ2 ba xa
Log-normal � � � � ��
LN μ,σ 1 ðln x lÞ2 1 ln x l
pffiffiffiffiffi e 2r2 1 þ erf pffiffi
xr 2p 2 r 2
Normal � � � � ��
N μ,σ 1 ðx lÞ2 1 x l
pffiffiffiffiffi e 2r2 1 þ erf pffiffi
r 2p 2 r 2
Nukiyama–Tanasawa 3
� �
NT b,q qbq 2 bxq C q3 ; bx 2
� �x e ��
1
C q3 C q3
Nukiyama–Tanasawa
pþ1
� �
NT3 b,p,q qb q C pþ1 p
� � xp e bx q q ; bx
1 � �
C pþ1
q C pþ1
q
Rosin–Rammler
RR m,n n � x �n 1 ðx=mÞn 1 e ðx=mÞn
e
m m � � � � � ���
Upper limit UL3 a,δ,xmax dxmax 2 2
1 ax
pffiffiffi e d ½lnðxmax xÞ�
ax
1 þ erf d ln
x ðxmax x Þ p 2 xmax x

Materials and methods Filtered air was introduced through a 0.088-m-


diameter ring distributor with 14, 0.6-mm-
Experimental setup
diameter holes, placed 0.103 m below the turbine.
This work A rotameter measured the airflow rate. The
Bubble size distributions were obtained by photo- operative conditions combined three agitation
graphing air–water gas dispersions in a stirred tank speeds of 300, 400, and 500 rpm (respectively,
reactor. The experiments were performed with tap 5.00, 6.67, and 8.33 revolutions/s) with three gas
water (surface tension of 71.8 mN/m) for various flow rates of 0.265, 0.529, and 0.968 vvm (vessel
combinations of agitator speeds and air flow volumes/minute; corresponding to superficial gas
rates. The lab-scale reactor consisted of a 0.232 m velocities of 1.03, 2.03, and 3.74 � 10−3 m/s,
diameter, flat-bottomed, and fully baffled glass respectively).
vessel with steel bottom, described in Figure 1. Photographs of gas dispersions were obtained
The liquid level was kept equal to the tank diameter using a digital camera (Pentax, Model *istD), with
resulting in a working volume of 9.8 L. Agitation a Pentax SMCDA 18–55 mm f/3.5–5.6 AL lens.
was provided by a 0.090-m-diameter Rushton The focal plane was set at 20 mm from the tank
turbine placed 0.118 m above the tank bottom. wall. A small Perspex box full of water attached
1108 H. J. O. PINHO ET AL.

Figure 1. Details of the experimental setup (not to scale). Dimensions (mm): B ¼ 28; C ¼ 118; D ¼ 90; Dd ¼ 60; De ¼ 17; Ds ¼ 28;
h ¼ 32; H ¼ 232; L ¼ 20; T ¼ 232; W ¼ 30; x1 ¼ 2; x2 ¼ 3.

to the tank wall eliminated the curvature effect. the tank volume (m3), g is the gravitational acceler-
The images have 49 mm height and 77 mm width ation (m/s2), and W the width of the impeller
and half height point 70 mm below the liquid level. blades (m). The estimated power ranges from 371
The photographs were analyzed using the Scion to 2281 W/m3.
Image for Windows software, version 4.0.3.2
P ¼ P0 qND3 D5 ð5Þ
(Scion, USA). Equivalent diameter db for ellipsoid
bubbles was calculated through Equation (4): � � ! 0:2
0:25
PG QG ND2 D4
�13 ¼ 0:10 2 ð6Þ
2 P ND V gWV 3
db ¼ d M dm ð4Þ

where dM is the length of the major axis and dm the


length of the minor axis. Approximately 1000 Data from the literature
bubbles were measured for each of the nine combi- Bubble size distributions from three published
nations of agitator speed and gas flow rate used. works were analyzed for comparison with the
The power transferred to the liquid was present work, covering a range of stirred tank sizes,
estimated from a corrected power number in operating conditions, and bubble sizing methods.
accordance with Brauer’s work (1985), to include Samaras et al. (2014) obtained bubble size
the effect of tank and stirrer geometric details, distributions in a 5-L stirred tank reactor, a fully
and with Hughmark’s (1980) model, to account baffled Plexiglas cylindrical tank of 0.190 m diam-
for the gas holdup effect on power. The power eter with a curved bottom. Agitation was provided
number (P0) calculated from Brauer’s (1985) set by a 0.063-m-diameter Rushton turbine, placed
of equations depends on the Reynolds number 0.063 m above the tank bottom. The liquid level
but is near 8 for the three agitations speeds used was kept equal to the tank diameter. A nozzle with
in the present work. The power dissipated by the 2 mm internal diameter placed at the bottom’s
impeller was then calculated using Equation (5) center was used to introduce the gas. The article
and corrected to gassed conditions with Hughmark’s provides graphical results for the arithmetic and
correlation [Equation (6)], where ρ is the specific Sauter mean bubble diameters and experimental
weight of the liquid phase (kg/m3), ND is the impel- PDFs, for experiments with 4.3 mM sodium chlor-
ler rotational speed (/s), D is the diameter of the ide aqueous solutions, which the authors claim to
impeller (m), QG is the gas flow rate (m3/s), V is be similar to tap water (conductivity 0.5 mS/cm,
CHEMICAL ENGINEERING COMMUNICATIONS 1109

surface tension 71.7 mN/m). Agitation speeds sizes were organized into discrete classes (bins)
ranged from 500 to 800 rpm, with gas flow rates with identical width. For the literature data, the
of 7.05 and 12.9 � 10−3 m/s. Bubble size was original bins were used, except when a minimum
obtained by photography. The average sample size number of five bubbles were not guaranteed, in
was 500 bubbles. The authors did not provide which case bins were modified or grouped to
power data. Using a power number of 5 and ensure statistical significance. The best set of
Hughmark’s correlation (1980), gassed power was parameters for each combination of operative
estimated to range from 289 to 1282 W/m3. conditions and for each of the 14 tested PDFs
Martin’s (1995) experimental setup consisted was the one which minimized the chi-square stat-
of a fully baffled Perspex cylindrical stirred tank istics [Equation (7)], where (Nhyp,i) and (Nexp,i)
reactor with 138 L liquid capacity and 0.56 m are the theoretical and experimental numbers of
diameter. A Rushton turbine with 0.168 m in bubbles within each bin and M represents the
diameter was placed 0.14 m above the tank bottom. number of bins.
Oil-free air was introduced through a 0.10-m- �2
XM
Nhyp;i Nexp;i
diameter ring with 16 2.5-mm-diameter holes 2
v0 ¼ ð7Þ
placed at the center of the tank below the impeller. i¼1
Nhyp;i
The liquid media consisted of deionized water
The theoretical number of bubbles within each
(surface tension 71.79 mN/m). Agitation speeds
bin was calculated for each PDF using the hypo-
ranged from 200 to 500 rpm, with gas flow rates
thetical parameter values and the corresponding
between 3.74 and 18.7 � 10−3 m/s. Bubble size
CDF (P(x)) by Equation (8), where N is the total
was obtained by a video technique. The average
number of bubbles, xi is the upper end, and xi−1
sample size was 500 bubbles. The author obtained
the lower end of the ith bin.
the gassed power experimentally by measuring
the torque in the agitator shaft. Gassed power Nhyp;i ¼ N ½Pðxi Þ Pðxi 1 Þ� ð8Þ
ranges from 167 to 2415 W/m3, estimated from
graphical data given by the author. All calculations were performed with Microsoft
Barigou (1987) made an extensive work in a Excel® 2016 (Microsoft, USA), using the solver
800-L stirred tank consisted in a cylindrical tank functionality to minimize the value of the chi-
made of Perspex, with 1.0-m-diameter, fully square statistics [Equation (7)]. Test computations
baffled, and a 0.33-m-diameter Rushton turbine. of the chi-square statistics were performed with
The turbine was placed 0.25 m above the tank IBM’s SPSS® package version 21 (IBM, USA),
bottom. Filtered air was introduced through a which validated the Excel results.
19-mm-diameter nozzle placed at the center of The chi-square statistics in Equation (7) may be
the tank and below the impeller. The liquid media used to test if a sample of data came from a popu-
consisted in softened tap water (surface tension lation with a specific distribution (Snedecor and
70.99 mN/m). Agitation speeds range from 100 Cochran, 1989). A sample of bubbles comes from
to 300 rpm, with gas flow rates of 2.09 and a population with a specific distribution if
5.58 � 10−3 m/s. Bubble size was obtained by the Equation (9) holds, where the right-hand side is
capillary suction probe technique. The average the chi-square critical value with DF degrees of
sample size was 500 bubbles. The author obtained freedom and significance level a, DF being the dif-
the gassed power experimentally by measuring ference between the number of data bins, M, minus
the torque in the agitator shaft. Gassed power one, and the number of the adjusted parameters, k
ranges from 106 to 2026 W/m3, as calculated from (DF ¼M − 1 − k). In general, the significance level
a correlation obtained by the author. is chosen at 5% (a ¼ 0.05). When the chi-square
statistics does not exceed the critical value, the true
hypothesis that the data follow a specified distri-
Fitting procedure and criteria
bution cannot be rejected at the stated statistical
Parameter optimization for PDF fitting of bubble significance level (Paloposki, 1994).
size distributions followed the procedures of
Paloposki (1994). Experimentally measured bubble v20 < v21 a;DF ð9Þ
1110 H. J. O. PINHO ET AL.

p-Values for the chi-square test, which represent diameter for each bin (di ¼ (xi þxi−1)/2), and Ni
the probability that the critical value of the chi- either Nhyp,i or Nexp,i.
statistics exceeds the value obtained in the fitting To evaluate the bin processing effects, d32,pop
procedure, were used to compare different PDFs was also estimated using the complete bubble
taking into account the different numbers of population in the present work, replacing the total
parameters. Higher p-values mean greater agree- number of bubbles, N, for the number of bins, M,
ment between experimental and fitted results. and the real measured individual bubble diameters
Coefficient of determination (R-squared, or for the mean bin diameter, di, in Equation (1). This
simply R2) is widely used to measure the goodness calculation was not possible to perform for the
of fit for linear models. Despite being deemed a literature data.
poor evaluator of the goodness of fit of nonlinear
models (Spiess and Neumeyer, 2010), R-squared
was used here due to its universality and simplicity Results and discussion
of reading in comparing the goodness of fit. How-
PDF assessment
ever, it was not the main criterion to distinguish
the fittings. Data from this work
Table 2 contains the experimental conditions of
Sauter mean diameter calculations the present work, the number of bubbles measured
for each set of experimental conditions, and the
Sauter mean diameter (d32) is the most important
calculated arithmetic (d10) and Sauter mean
parameter obtained from a BSD, since it is directly
diameters.
used in the calculation of heat and mass transfer
Figure 2 presents the average goodness of fit
area. Agreement between d32 directly calculated
criteria evaluated for the 14 PDFs tested. The
from experimental BSD and calculated from BSD
values presented represent the average of the nine
modeled through PDF is therefore an alternative
experimental data sets. The vertical intervals in
valuable criterion for goodness of PDF fitting.
the figure represent the spread of the chi-square
Equation (10) is a measure of the Sauter mean
statistics for the experimental data sets for each
diameter relative deviation, d32,dev
fitted PDF.
K �
� �
1X d32;hyp;j d32;exp;j � Since the number of bins was equal to nine for
d32;dev ¼ ð10Þ all data sets, the critical chi-square value is 12.6
K j¼1 d32;exp;j
for the two-parameter PDFs, 11.1 for the three-
where d32,hyp,j represents the Sauter mean diameter parameter PDFs (NT3 and UL3), and 9.5 for the
calculated with the number of bubbles within each four-parameter LH4 PDF. p-Values may be directly
bin, Nhyp,i, estimated using a PDF, while d32,exp,j comparable because they account for the number
represents the Sauter mean diameter calculated of adjustable parameters. Average p-values showed
with the number of experimental bin-grouped bub- significance for the NT3, RR, and UL3 PDFs
bles, Nexp,i. K represents the number of evaluated (higher than 0.05).
data sets. d32,hyp,j and d32,exp,j were calculated by For individual data sets (Figure 3), NT3 PDF
Equation (1), where di represents the mean shows statistical significance for eight of the nine

Table 2. Experimental conditions and observed representative bubble diameters.


Agitator speed Gas flow rate Estimated Number Minimum Maximum
Data set (rps) (rpm) (�10−3 m/s) (vvm) power (W/m3) of bubbles diameter (mm) diameter (mm) d10 (mm) d32 (mm)
1 5.00 (300) 1.03 (0.265) 513 926 0.10 5.01 1.31 2.54
2 2.04 (0.529) 432 940 0.05 4.67 1.41 2.59
3 3.74 (0.968) 371 1015 0.08 5.58 1.71 3.00
4 6.67 (400) 1.03 (0.265) 1189 1095 0.14 5.09 1.32 2.23
5 2.04 (0.529) 1001 1191 0.09 4.98 1.43 2.31
6 3.74 (0.968) 861 997 0.10 5.29 1.51 2.58
7 8.33 (500) 1.03 (0.265) 2281 1129 0.13 5.23 1.30 2.05
8 2.04 (0.529) 1920 1264 0.13 4.83 1.32 2.25
9 3.74 (0.968) 1651 1135 0.10 5.99 1.32 2.32
CHEMICAL ENGINEERING COMMUNICATIONS 1111

Figure 2 Goodness of fit indicators. Bars represent average reduced chi-square for tested PDFs fitted to experimental data. Vertical
lines represent minimum and maximum values of reduced chi-square. Circles represent average p-value and squares represent
average R-squared.

data sets, RR PDF shows significance for four data have average relative deviations of 3.3, 3.7, 3.8,
sets, LH4 PDF for three data sets, and UL3 PDF for and 4.1%, respectively. The N PDF was surprisingly
one data set. NT PDF and G PDF also show good at predicting d32 after data fitting. d32 calcu-
significance for one data set. lated with the NT3 PDF was very close to the
The R-squared was higher for the same four experimental results for all tested operating
PDFs (Figure 2). The G, BE, and NT PDFs also conditions (Figure 5). The other PDFs showed high
showed an average R-squared greater than 0.95. relative d32 deviations for some operating con-
Figure 4 shows the average d32 deviation with ditions. RR shows higher deviations for lower input
d32 calculated from experimental data after bin power conditions with a maximum deviation of
grouping. Five PDFs showed an average relative 6.9% for data set 7. N PDF does not show a regular
deviation lower than 5%. NT3 PDF has an average trend on the d32 relative deviation. LH4 PDF shows
relative deviation of 0.4%, and a maximum devi- very high deviations for data sets 1 and 4, two of
ation of 0.8%. The N, RR, LH4, and UL3 PDFs the three lower power input conditions. UL3 PDF

Figure 3. Individual data set p-values for three-parameter Nukiyama–Tanasawa (NT3), Rosin–Rammler (RR), four-parameter
log-hyperbolic (LH4), and three-parameter upper limit (UL3) PDFs. Data set operating conditions according to Table 2.
1112 H. J. O. PINHO ET AL.

Figure 4. Average relative deviation between d32 calculated from PDF fitting results and d32 calculated from experimental
bin-grouped data. Intervals represent minimum and maximum values. Maximum values higher than 40% are not showed.

ranks the fifth best average relative deviation but volume used by Samaras et al., agitation speed
with low dispersion from that value. and superficial gas velocity ranges exceed that of
In general, the PDFs with the best fitting to the present work. With the exception of one data
experimental data were the NT3, RR, LH4, and set, however, the gassed power was within the
UL3, in that order. The N PDF was revealed to range of this work. Samaras et al. data were orga-
be an acceptable alternative to estimate the d32 nized in nine bins for direct comparison with data
after data fitting. RR and N have the advantage from this work.
of requiring only two fitting parameters. Figure 6 presents the results from the fitting
of the 14 tested PDFs. The critical values of the
Literature data chi-square statistics are the same as referred
Six data sets were extracted from Samaras et al. for the present work data. Although all fittings
(2014) work. The BSDs were estimated from the show chi-square values higher than the critical
author’s figures presenting bubble size probability ones, the UL3 and NT3 PDFs have the
against bubble diameter. Due to the smaller tank best behavior, followed by the NT and G PDFs.

Figure 5. Relative deviation for d32 calculated from PDF fitting results and d32 calculated from experimental bin-grouped data.
Three-parameter Nukiyama–Tanasawa (NT3), Rosin–Rammler (RR), Normal (N), four-parameter log-hyperbolic (LH4), and three-para-
meter upper limit (UL3) PDFs.
CHEMICAL ENGINEERING COMMUNICATIONS 1113

Figure 6. Goodness of fit indicators for Samaras et al. (2014) data sets. Bars represent average reduced chi-square for tested PDFs
fitted to experimental data. Vertical lines represent minimum and maximum values of reduced chi-square. Triangles represent average
d32 deviation and squares represent average R-squared.

For these data sets, the RR PDF only obtains the which to d32 is only about 0.5%. As all the tested
11th position but gives the best estimation of the PDFs are unimodal, these very small bubbles were
d32, showing a d32 average deviation of 0.7%. neglected in the calculations.
From Martin’s data (1995), eight data sets were Bubble size distribution data were organized in
extracted. The operating conditions for three data nine bins to allow a direct comparison with the
sets are slightly below the ranges covered in the present work’s results. Maximum bubble sizes
present work. Gassed power of four data sets is interpreted from the author’s work was 6 mm.
outside the present work’s range of values. The average d32 deviation between the bin-grouped
Bubble size distributions were estimated from diameters and the d32 given by the author
the author’s figures presenting bubble number fre- was 2.0%.
quency against bubble diameter. Martin’s data are Figure 7 presents the results from the fitting of
bimodal, with a high frequency peak very close to all the tested PDFs except for LH4 PDF, which
zero diameters, the maximum contribution of does not converge for two data sets. The critical

Figure 7. Goodness of fit indicators for Martin’s (1995) data sets. Bars represent average reduced chi-square for tested PDFs fitted to
experimental data. Vertical lines represent minimum and maximum values of reduced chi-square. Triangles represent average d32
deviation and squares represent average R-squared.
1114 H. J. O. PINHO ET AL.

values of the chi-square statistics are the same as Table 3. Nukiyama–Tanasawa and Rosin–Rammler parameter
those for the present work’s data, as already correlations with the input gassed power.
mentioned. All fittings present chi-square values PDF parameter correlations R2 F-value

1:48 (12)
b ¼ 6:43 � 10 6 PVG 0.876 0.016
higher than the critical ones, and higher than the �
p ¼ 9:43 � 10 15 PVG
4:29 (13) 0.852 <0.001
values obtained with Samaras et al.’s (2014) data. �
q ¼ 14:8 PVG
0:279 (14) 0.838 0.038
The lower chi-square values were obtained for � 0:0892 (15)
m ¼ 2:89 PVG 0.669 0.024
the NT3, N, RR, and UL3 PDFs by that order. �0:171 (16) 0.713 0.004
n ¼ 0:475 PVG
d32 deviations follow the same order: 0.6% for the
NT3 and N PDFs, and 14 and 20%, respectively,
for the RR and UL3 PDFs. There is not a single PDF that accurately
From Barigou’s data (1987), three data sets were describes all the literature data analyzed. The LN
extracted. BSDs were estimated from the author’s PDF, sometimes mentioned as appropriate for
figures presenting bubble number frequency BSD description, is not among the best ranked
against bubble diameter. Maximum bubble size PDFs.
was 5.65 mm. The NT3 and RR PDFs, which best described the
The average d32 deviation between the present authors’ original data, showed to be a
bin-grouped diameters and the d32 given by the reasonable descriptor of BSD in the case of NT3
author was 1.8%. Figure 8 presents the results from PDF, and a reasonable estimator of d32 in the case
fitting the 14 tested PDFs to the experimental data. of RR PDF.
The critical value of the chi-square statistics is the
same as the value found for the present work’s
data. The lower chi-square value was obtained Correlation of PDF parameters with
for LH4 PDF, but all values were higher than the experimental conditions
critical ones except for the G and UL3 fittings to Data from this work
one data set. All fitted PDFs give very low d32 The dependency of the adjusted PDF parameters
deviations, less than 2% except for N, F, and BE, on the operative conditions was tested for the four
the PDFs with the worst fittings. RR PDF does PDFs which best fitted the original experimental
not rank well, but the fitting quality indicators data and for the normal PDF. The latter was stud-
are similar to the best ranked PDF. ied because of its simplicity and universality and its

Figure 8. Goodness of fit indicators for Barigou’s (1987) data sets. Bars represent average reduced chi-square for tested PDFs
fitted to experimental data. Vertical lines represent minimum and maximum values of reduced chi-square. Triangles represent average
d32 deviation and squares represent average R-squared.
CHEMICAL ENGINEERING COMMUNICATIONS 1115

PDF parameter, ϕk and ψk are the corresponding


correlation parameters.
� �wk
PG
ck ¼ /k ð11Þ
V
Equation (11) is used since it has been shown
that in aerated stirred tanks, bubble size correlates
well with PG/V for both coalescing and noncoales-
cing systems, both at impeller discharge and in the
bulk of the tank (Alves et al., 2002).

Figure 9. Parity plots for PDF parameters from gassed power Nukiyama–Tanasawa and Rosin–Rammler PDFs
correlations vs. Parameters directly from experimental data. (a) Table 3 presents the resulting expressions after ϕk
RR: ▪ parameter m; ● parameter n; (b) NT3: Δ parameter b; and ψk optimization of the three-parameter
� parameter p; & parameter q. Dashed line represents identity.
Nukiyama–Tanasawa and Rosin–Rammler PDFs.
Probability density function parameter regres-
accuracy in predicting d32. A simple correlation sions are significant for both correlations: b, p, q
with gassed power (PG) per liquid volume (V) for NT3 PDF and m, n for RR PDF (all with
was tested [Equation (11)]. γk represents each F-value <0.05, which represents the Fisher test

Figure 10. Comparison of fitted NT3 PDF to the experimental data sets, identified inside brackets from 1 to 9. X-axis is the average
bubble diameter for each bin (mm) and Y-axis is the number frequency (/mm). ▪ Experimental data; — NT3 PDF directly from
experimental data; ···· NT3 PDF with parameters calculated from correlation.
1116 H. J. O. PINHO ET AL.

p-value). The corresponding correlation para- RR PDFs. High power input to the dispersion
meters are also significant (ϕk and ψk always with causes a pronounced number frequency peak at
P < 0.05, for both correlations). Figure 9 presents lower bubble diameters and a decrease in number
parity plots for both correlations: parameters frequencies at higher bubble diameter sizes,
correlated with gassed power against the original decreasing the BSDs tails. Both the NT3 and RR
parameters fitted directly to the experimental data. PDFs correctly follow these BSD profiles.
Figures 10 and 11 present the comparisons Three-parameter Nukiyama–Tanasawa distri-
between the experimental BSDs for the nine bution PDF with parameters calculated from the
experimental data sets and the fitted three- power correlation [Equations (13)–(15) of
parameter Nukiyama–Tanasawa and Rosin– Table 3)] has the biggest deviation from the experi-
Rammler PDFs, respectively. The figures also mental data on the data set 7, which corresponds to
include the BSDs calculated using the same the higher input power. RR PDF, with only two
distributions but with PDF parameters computed parameters [from the power correlations of
from the correlation with gassed power. The Equations (16) and 17) of Table 3], presents a more
experimental data show a pronounced effect of stable description of power effects.
the operative conditions on the BSD shape, which No expressions for the normal, log-hyperbolic,
can be satisfactory modeled by both the NT3 and and upper limit PDFs parameters are presented,

Figure 11. Comparison of fitted RR PDF to the experimental data sets, identified inside brackets from 1 to 9. X-axis is the average
bubble diameter for each bin (mm) and Y-axis is the number frequency (/mm). ▪ Experimental data; — RR PDF directly from
experimental data; ···· RR PDF with parameters calculated from correlation.
CHEMICAL ENGINEERING COMMUNICATIONS 1117

since they do not correlate significantly well as a Conclusion


function of PG/V.
Bubble size distributions in aerated stirred tanks
have been successfully modeled by PDFs. However,
Literature data
the literature is lacking in:
Three-parameter Nukiyama–Tanasawa and Rosin–
i. A systematic comparison of alternative
Rammler PDFs’ parameters correlated in the
modeling functions and
previous section [Equations (12)–(16) in Table 3]
ii. correlations between modeling function
were used to estimate the BSDs for Samaras et al.
parameters and operating conditions.
(2014), Martin (1995), and Barigou (1987), from
To systematically evaluate the modeling of
experimental or calculated gassed power in these
bubble size distribution in air-sparged vessels
authors’ experiments. The resulting BSDs were
agitated by Rushton turbines, 14 different PDFs
then used to calculate d32 using the procedure
were studied and fitted to bubble size distributions,
described in the “Sauter mean diameter calcula-
both original and retrieved from the literature. The
tions” section.
results obtained indicate that:
Figure 12 compares d32 thus calculated with the
i. Bubble size distributions in air-sparged
d32 obtained after bin grouping of experimental data.
vessels agitated by Rushton turbines may be
For Samaras’ data, d32 deviation was 5.6 and 6.3%
reasonably modeled by the three-parameter
for the NT3 and RR PDFs, it was 4.8 and 7.5% for
Nukiyama–Tanasawa PDF, and acceptably by
Martin’s data and 20.9 and 24.5% for Barigou’s data,
the two-parameter Rosin–Rammler PDF;
respectively. Globally, considering the data from all
ii. the parameters of these two PDFs can be
authors, including the present work, d32 deviation
correlated with the experimental power of
was 7.5 � 3.2% for the NT3 PDF (P ¼ 0.33;
the impeller; and
r ¼ 0.525) and 8.4 � 3.5% for the RR PDF (P ¼ 0.53;
iii. both PDFs, with parameters correlated with
r ¼ 0.491). Results show that the BSDs in aerated
power data, provide good predictions of
agitated tanks may be reasonably represented by
bubble size distribution and Sauter mean
two-parameter correlations on dissipated power,
diameter.
resulting in good approximations for the Sauter
The correlations obtained may be useful in
mean diameter. It should be recognized that the data
improving the application of CFD to air–water
were retrieved from the literature and used in this
systems in stirred tanks, and set an example
work for comparison results from a smaller number
for similar correlations in other multiphase
of measured bubbles, which may significantly
systems.
contribute to the poorer fittings.

Nomenclature
a upper limit distribution parameter
b Nukiyama–Tanasawa distribution parameter
D impeller diameter, m
d10 arithmetic mean diameter, m
d32 Sauter mean diameter, m
d32,dev Sauter mean diameter relative deviation
d32,exp,j Sauter mean diameter calculated with
the experimental number of bin-grouped
bubbles, m
d32,hyp,j Sauter mean diameter calculated with
the theoretical number of bin-grouped
bubbles, m
Figure 12. Parity plot of d32 calculated with correlated
parameters against bin-grouped experimental data d32, for all d32,pop Sauter mean diameter calculated with the
analyzed data sets: ● Present author’s data; Δ Samaras et al. whole bubble population, m
(2014) data; � Martin’s (1995) data; & Barigou’s (1987) db bubble equivalent diameter, m
data. (a) Three-parameter Nukiyama–Tanasawa PDF; (b) Rosin– di bubble diameter, m
Rammler PDF. DF degrees of freedom
1118 H. J. O. PINHO ET AL.

dm length of bubble minor axis, m Barigou, M., and Greaves, M. (1992). Bubble-size distribu-
dM length of bubble major axis, m tions in a mechanically agitated gas–liquid contactor,
g gravitational acceleration, m/s2 Chem. Eng. Sci., 47, 2009–2025.
k number of adjustable parameters Brauer, H. (1985). Stirred vessel reactors, in H. Brauer ed.,
K number of evaluated data sets Biotechnology, vol. 2, Fundamentals of Biochemical Engin-
m Rosin–Rammler distribution parameter eering, VCH, Weinheim.
M number of bins Busciglio, A., Grisafi, F., Scargiali, F., and Brucato, A. (2010).
n Rosin–Rammler distribution parameter On the measurement of bubble size distribution in
N total number of bubbles gas–liquid contactors via light sheet and image analysis,
ND impeller rotational speed, /s Chem. Eng. Sci., 65, 2558–2568.
Nexp,i number of experimental bin grouped bubbles Dunbar, C., and Hickey, A. (2000). Evaluation of probability
Nhyp,i theoretical number of bin grouped bubbles density functions to approximate particle size distributions
Ni number of bubbles with diameter di of representative pharmaceutical aerosols, J. Aerosol Sci.,
p Nukiyama–Tanasawa distribution parameter 31, 813–831.
P0 power number González-Tello, P., Camacho, F., Vicaria, J., and González, P.
P power dissipated by the impeller, W (2008). A modified Nukiyama–Tanasawa distribution func-
PG power dissipated by the impeller in the tion and a Rosin–Rammler model for the particle-size-
aerated vessel, W distribution analysis, Powder Technol., 186, 278–281.
q Nukiyama–Tanasawa distribution parameter Hughmark, G. A. (1980). Power requirements and interfacial
QG gas flow rate, m3/s area in gas–liquid turbine agitated systems, Ind. Eng. Chem.
V vessel volume, m3 Process Des. Dev., 19, 641–646.
W width of the impeller blades, m Kálal, Z., Jahoda, M., and Fort, I. (2014). Modelling of the
xi upper bin limit, m bubble size distribution in an aerated stirred tank: Theor-
xi−1 low bin limit, m etical and numerical comparison of different breakup
xmax upper limit distribution parameter models, Chem. Process Eng., 35, 331–348.
a significance level, probability distribution Macías-García, A., Cuerda-Correa, E., and Díaz-Díez, M. (2004).
function parameter Application of the Rosin–Rammler and Gates–Gaudin–
b probability distribution function parameter Schuhmann models to the particle size distribution analysis
γk generic representation of correlated of agglomerated cork, Mater. Charact., 52, 159–164.
parameter Martin, T. (1995). Gas dispersion with radial and hydrofoil
δ probability distribution function parameter impellers in fluids with different coalescence characteristics,
μ probability distribution function parameter Ph.D. thesis, Birmingham University, United Kingdom.
ρ specific weight of the liquid phase, kg/m3 Montante, G., Horn, D., and Paglianti, A. (2008). Gas-liquid
σ probability distribution function parameter flow and bubble size distribution in stirred tanks, Chem.
ϕk correlation parameter Eng. Sci., 63, 2107–2118.
ψk correlation parameter Paloposki, T. (1994). Drop size distributions in liquid sprays,
v20 chi-square statistics Acta Polytech. Scand. Mech. Eng. Ser., 114, 1–209.
v21 a chi-square critical value Petitti, M., Nasuti, A., Marchisio, D., Vanni, M., Baldi, G.,
Mancini, N., and Podenzani, F. (2010). Bubble size distribution
modeling in stirred gas–liquid reactors with QMOM augmen-
ted by a new correction algorithm, AIChE J., 56, 36–53.
ORCID Sajjadi, B., Raman, A., Ibrahim, S., and Shah, R. (2012).
Review on gas–liquid mixing analysis in multiscale stirred
Henrique J. O. Pinho http://orcid.org/0000-0002-1344-
vessel using CDF, Rev. Chem. Eng., 28, 171–189.
6517
Samaras, K., Kostoglou, M., Karapantsios, T. D., and Mavros,
Dina M. R. Mateus http://orcid.org/0000-0001-8374-3384
P. (2014). Effect of adding glycerol and Tween 80 on gas
holdup and bubble size distribution in an aerated stirred
References tank, Colloids Surf. A, 441, 815–824.
Snedecor, G. W., and Cochran, W. G. (1989). Statistical Methods,
Alves, S. S., Maia, C. I., Vasconcelos, J. M. T., and Serralheiro, 8th Ed., Iowa State University Press. Iowa City, USA.
A. J. (2002). Bubble size in aerated stirred tanks, Chem. Spiess, A.-N., and Neumeyer, N. (2010). An evaluation of R2
Eng. J., 89, 109–117. as an inadequate measure for nonlinear models in
Babinsky, E., and Sojka, P. (2002). Modeling drop size distri- pharmacological and biochemical research: A Monte Carlo
butions, Prog. Energy Combust. Sci., 28, 303–329. approach, BMC Pharmacol., 10, 1–11.
Barigou, M. (1987). Bubble size, gas holdup and interfacial area Zhang, Z., Dai, G., and Chen, M. (1989). Study on bubble size
distributions in mechanically agitated gas–liquid reactors, distribution in an aerated stirred vessel, Huagong Xuebao
Ph.D. thesis, The University of Bath, United Kingdom. (Chin. Ed.), 40, 183–189.

Das könnte Ihnen auch gefallen