Hassan 2012
Hassan 2012
a r t i c l e i n f o a b s t r a c t
Article history: This paper deals with gas–liquid mass transfer in an aerated stirred tank containing Newtonian or shear-
Received 17 June 2011 thinning fluids. The aim is to demonstrate that, for a given mixing system, an unique dimensionless
Received in revised form correlation gathering all the mass transfer rates (150 kl a measurements) can be obtained if and only if the
24 November 2011
variability of the rheological material parameters is correctly considered when implementing the theory
Accepted 14 December 2011
of similarity. More particularly, it is clearly illustrated that a too gross simplification in the relevant list
of the parameters characterizing the dependence of apparent viscosity with shear rates leads to pitfalls
Keywords:
when building the -space set. This is then a striking example showing that a robust predictive correlation
Dimensional analysis
Variable material property
can be established when the non-constancy of fluid physical properties ceases to be neglected.
Gas–liquid mass transfer © 2012 Elsevier B.V. All rights reserved.
Shear-thinning fluids
Aerated stirred tank
Model of Williamson–Cross
1385-8947/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.cej.2011.12.066
R. Hassan et al. / Chemical Engineering Journal 184 (2012) 42–56 43
Nomenclature
Sc Schmidt number defined according to gas phase
a specific interfacial area (m−1 ) properties (Eq. (31))
CO2 concentration in dissolved oxygen (kg/m3 ) ∗
tw dimensionless time number issued from the
CO∗ concentration in dissolved oxygen at saturation Williamson–Cross’s model (Eq. (45))
2
(kg/m3 ) Ug∗ dimensionless superficial gas velocity (Eq. (26))
d air sparger diameter (m)
D impeller diameter (m) Subscripts
D oxygen diffusion coefficient in the liquid phase g gas phase
(m2 /s) l liquid phase
G/Vl mass throughout per unit of liquid volume (kg/s/m3 )
g gravity acceleration (m/s2 )
H Henry’s constant (Pa) Another approach is to use correlations with dimensionless
Ht tank height (m) groups; contrary to dimensional correlations, they guarantee firm
kl liquid-side mass transfer coefficient (m/s) basis for process scale-up, provided that they must be established
K consistency index from the model of Ostwald–de- with respect to the theoretical context of the theory of similarity.
Waele (Eq. (14)) (Pa snost ) The pioneer work’s of Zlokarnik [7] has established the relevant
Kl overall mass transfer coefficient (m/s) list of influencing intensive parameters and proposed the following
kl a overall volumetric gas–liquid mass transfer coeffi- dependence between dimensionless numbers:
cient (s−1 ) ∗ ∗
∗ P Qg
KMO Metzner–Otto constant (Eq. (5)) (kl a) = f1 , , ∗ , Sc, Si∗
Qg Vl
m Henry’s constant ⎧ 1/3 ∗
nost flow index from the model of Ostwald–de-Waele ⎪
⎨ (kl a)∗ = kl a · g2l ,
P
=
P
· [l · (l · g)
2/3 −1
] , (2)
Qg Qg
(Eq. (14)) where ∗ 1/3
nw flow index from the model of Williamson–Cross (Eq. ⎪
⎩ ∗ = · [l · (4 · g)1/3 ]−1 , Qg = Qg · l
l Vl Vl g2
(15))
N rotational impeller speed (s−1 ) Note that, in Eq. (2), Sc is the Schmidt number and Si* is a material
P/Vl power dissipated per unit of volume (W/m3 ) dimensionless parameter which describes coalescence behaviour
Ps pressure in the system (Pa) of solutions (i.e. ionic strength, electrical charge of ions, . . .). Thanks
Qg gas flow rate (m3 /s) to this approach, Zlokarnik [7] could rigorously distinguish dif-
t time (s) ferent process relationships depending whether the system is
tw time parameter from the model of coalescent or non-coalescent.
Williamson–Cross (Eq. (15)) (s) Few years later, Judat [8] has critically examined the existing
Tl temperature of the liquid phase (K) publications on gas–liquid mass transfer (coalescing systems) in
Tt vessel diameter (m) stirred vessels. Description of experimental data with the aid of
Ug superficial gas velocity (m/s) intensive parameters has leaded this author to (±30% deviation):
V tank volume (m3 )
∗0.40
Vl liquid tank volume (m3 ) ∗ (P/Vl )
(kl a) = 9.8 × 10−5 · (3)
B−0.6 + 0.81 × 10−0.65/B
Greek letters
ε mean standard deviation (Eq. (50)) where B = (Qg /Tt2 ) · (l · g)−1/3 , the others numbers being defined
˙ shear rate (s−1 ) as in [7]. Judat [8] has then shown that a monoparametric represen-
˙ o reference shear rate (s−1 ) tation of (kl a)* versus (P/Vl )* is inadequate, and that only a -space
˙ av average shear rate defined from the Metzner–Otto representation containing both dimensionless power per unit vol-
concept (Eq. (4)) (s−1 ) ume and superficial gas velocity can satisfactorily correlate (kl a)*.
dynamic viscosity (Pa s) This author has also put forward that another -space, contain-
a apparent viscosity (Pa s) ing non intensive parameters (rotational impeller speed instead
o reference apparent viscosity parameter defined at of power per unit liquid volume) could be used to describe the
reference shear rate (Pa s) measures of (kl a)*, but this i -space is larger than the previous one.
w viscosity parameter from the model of Few authors (for example [9,10]) have conserved the dimen-
∗ 1/3
Williamson–Cross (Eq. (15)) (Pa s) sionless group (kl a) = kl a · (l /g 2 ) defined by these two pioneer
cinematic viscosity (Pa s) works for modelling absorption processes. Most of them have
density (kg/m3 ) adopted, with or without theoretical backgrounds, others defini-
surface tension (N/m) tions for making dimensionless kl a [11–17]. They include notably
a modified Sherwood number (kl a · Tt2 /D) or Stanton number
Dimensionless numbers (kl a · Vl /Qg ) [3].
Fr Froude number (Eq. (27)) When shear-thinning fluids are involved, two additional ques-
kl a* dimensionless volumetric mass transfer coefficient tions arise unfortunately in a point of view of the theory of
(Eq. (25)) similarity:
* dimensionless viscosity (Eq. (29))
i dimensionless number deduced from the theory of - How should we proceed to guarantee that the results obtained
similarity with Newtonian fluids can be extended to these non-Newtonian
* dimensionless density (Eq. (28)) liquids? The spatial distribution of liquid viscosity in the tank, due
* dimensionless surface tension (Eq. (30)) to its dependency with shear rates, constitutes a major difficulty
with regard to the choice of a representative viscosity.
44 R. Hassan et al. / Chemical Engineering Journal 184 (2012) 42–56
Table 1 The sparger was located 30 mm from the bottom of the tank, in
Dimensions of the experimental setup.
the axis of the impeller. The gas flow rate (Qg ) was regulated by
Tank size Tt = 0.212 m using a manometer (Samson® 47 08-1155) and measured with
Ht = 0.316 m a volumetric flow meter (Brooks® R2-25-C) with an accuracy of
V = 10 L
0.05 L/min. Ranged from 0.33 to 3.33 L/min, these values remained
Hl = 0.212 m
Vl = 7.4 L
quite narrow when compared to the available literature, they were
Baffles wb = Tt /10 (width, in m) initially imposed by the application underlying this work (namely
bb = Tt /50 (distance from walls, in m) the Autothermal Thermophilic Aerobic Digestion of sludge, see
Impeller (six-concave-blade D = 0.4·Tt (impeller diameter, in m) [19]). In terms of gas–liquid regime, it can be noticed (visual obser-
turbine) Ds = 3·Tt /4 (disk diameter, in m)
vations) that the operating conditions under test (N, Qg , fluids)
C = Tt /4 (clearance from the bottom, in m)
b = D/5 (blade height, in m) leaded to a complete dispersion regime, meaning thus that bub-
w = 0.5 D/5 (blade width, in m) bles were almost uniformly distributed throughout the tank. One
l = D/4 (blade length, in m) exception was for N = 200 rpm where the loading regime took place
Sparger d = D (sparger diameter, in m)
(presence of bubbles only in the upper part of the tank).
- How should we proceed to guarantee that the results obtained at 2.2. Methods of overall volumetric gas–liquid mass transfer
lab-scale will be also scalable at industrial scale? coefficient measurement
Until now, most of works encountered in the literature did not 2.2.1. Standard dynamic method
take care about these questions: they simply consist in replacing For implementing the standard dynamic method, the liquid
the Newtonian viscosity by an apparent viscosity defined from the phase was deoxygenated by flushing with nitrogen. Then, after
Ostwald–de-Waele’s model in which an average shear rate is con- replacing nitrogen by air, the variation in dissolved oxygen con-
sidered according to the well-known concept of Metzner–Otto: centrations with time was measured until reaching the saturation.
For that, two probes (InPro-6050, Mettler-Toledo® ) and an acqui-
˙ av = KMO · N (4)
sition card were implemented, as well as the LabView® software
where the constant KMO depends on the agitation system. Such for data acquisition. The positions of the probes are represented in
choice of apparent viscosity is in many cases questionable, in par- Fig. 1: they are located vertically at 4 cm and 18 cm above the bot-
ticular when (i) the flow regime is not laminar (the use of Eq. (4) tom of the vessel, and horizontally at 1.5 cm from walls. Assuming
becomes then quite haphazard) and (ii) the rheological behaviour a well mixed liquid phase, the mass balance in dissolved oxygen
of fluids cannot be described in the whole range of shear rates by concentration is given by
the Ostwald–de-Waele’s model.
The present paper aims at rigorously answering these two lat- dCO2
= Kl · a · (CO∗ − CO2 ) (5)
ter questions, starting from the theoretical background underlying dt 2
For a given operating condition (N and Qg ), the mean kl a in the tank during an aeration time taeration , with the small quantity of Na2 SO3
was calculated by averaging the values measured by the two probes introduced:
and for the three runs (N = 3), such as:
1
Na2 SO3 + O2 → Na2 SO4 (11)
1 |kl atop + kl abottom | 2
kl a =< kl a >= · (9)
N 2 The mass of Na2 SO3 to initially introduce (mt ) must be chosen
N
carefully, as it should enable to keep a zero oxygen concentration
The axial homogeneity (h) of volumetric gas–liquid mass transfer during the aeration time (taeration ) while avoiding an excessive use
coefficients was also evaluated by using the criterion h defined as of Na2 SO3 . Indeed, it is important to guaranty that the coalescing
follows [24]: properties of the liquid phase were not affected by the presence
of Na2 SO3 . A good compromise was to maintain an initial concen-
1 |kl atop − kl abottom |
h= · (10) tration below 0.5 g/L (i.e. mt < 3.5 g with Vl = 7.4 L) [25]. Note that
N < kl a > to minimize mt , it was also possible to play on the aeration time
N
(taeration ), which was here typically ranged from 1.5 min to 9 min.
where N was the number of experiments (N = 60 for each liquid The optimization of both mt and taeration was made easier by the
phase). fact that the orders of magnitude of kL a were known thanks to the
measurements issued from the dynamic method.
2.2.2. Chemical method When such conditions are respected, Painmanakul et al. [27]
When applying the latter dynamic method in viscous fluids, the have shown that the overall volumetric gas–liquid mass transfer
impact of the probe dynamics and of the liquid film in front of the coefficient can be deduced from:
membrane on the probe can no more be ignored, as possibly bias- (1/2)(MO2 /MNa2 SO3 ) · (mt − mr )
ing the measurements of kl a [25]. In addition, in such fluids, the kl a = (12)
taeration · Vl · CO∗
gas hold-up structure is known to be very different from the one 2
observed in water and other low-viscosity liquids: many tiny bub- where mt is the total mass of Na2 SO3 initially introduced, mr is
bles appear to accumulate during aeration and circulate with the the mass of Na2 SO3 remaining in the tank after an aeration period
liquid while large bubbles are also observed (bimodal bubble popu- taeration , and CO∗ is the concentration in dissolved oxygen at satura-
lation). These tiny bubbles can actively contribute to mass transfer, 2
tion. For glycerine solutions, CO∗ was by default considered equal to
depending whether they are in equilibrium with the level of dis- 2
solved solute in the liquid phase (high residence times) or not [26]. 8.8 mg L−1 as in deionised water at 20 ◦ C. An identical assumption
For these reasons, it has been chosen to implement a second was made for CMC and xanthan gum solutions in agreement with
method for kl a measurement. It will then enable to test the valid- the data reported by [28] who showed that, in the range of con-
ity and accuracy of the dynamic method in the viscous fluids centrations here involved, no major variation of CO∗ occurs when
2
involved. This alternative method was the chemical method devel- compared to water.
oped by [27], based on a mass balance on sodium sulphite (Na2 SO3 ) At last, for each condition, three samples (10 mL) were taken in
concentrations during a given aeration time. Nitrogen was firstly the tank, immediately mixed with 10 mL of standard iodine reagent
injected into the liquid phase in order to eliminate the dissolved at 0.12 equiv./L. The titration of these samples with a sodium thio-
oxygen present in the tank. When the concentration of dissolved sulphate solution (0.05 equiv./L) and a starch indicator (iodometry
oxygen reached nearly zero, an adequate amount of Na2 SO3 was titration) gave access to the concentration of Na2 SO3 remaining in
introduced; air was then introduced in the tank and will react, the tank after an aeration period taeration , and thus to mr .
46 R. Hassan et al. / Chemical Engineering Journal 184 (2012) 42–56
viscosity (Pa.s)
1
- the surface-aeration: it corresponds to the mass transfer occurring
at the free surface which importance depends strongly on the
0.1
surface motion. It also includes the aeration associated with the
bubbles entrained from the surface into the liquid bulk;
0.01
- the volume-aeration: it is induced by the bubbles generated at the
sparger directly inside the liquid bulk.
0.001
This can be expressed such as: 0.1 1 10 100 1000 10000
shear rate (s-1)
kl a|t = kl a|surf + kl a|vol (13)
(b) 100
Some measurements are made with mechanical agitation and with-
Exp.
out bubbling at the sparger. They enable to get an idea of the relative 10
importance of each contribution. Ostwald
Williamson
viscosity (Pa.s)
1
2.3. Fluids
0.1
The application underlying this study dealt with investigations
on aeration performances in an Autothermal Thermophilic Aero-
0.01
bic Digestion (ATAD) process for treating sludge issued from waste
water treatment plant [19]. As sludge was a very complex mate-
rial, it was decided in a first step to work with model fluids instead 0.001
of sludge. Their formulation was chosen so as to obtain rheologi- 0.1 1 10 100 1000 10000
cal behaviours as close to sludge as possible, in particular in terms shear rate (s-1)
of shear-thinning properties. For these reasons, and with regard 100
to literature, aqueous solutions of carboxymethylcellulose (CMC)
(c)
Exp.
and xanthan gum were selected. The use of two types of fluids
10 Ostwald
offered the advantage to cover a wider range of combinations of
viscosity (Pa.s)
flow and consistency indexes. To ensure their stability in time (dur- Williamson
1
ing several days), NaCl was added at 0.1% (w/v) to the solutions
of xanthan gum solutions [29] and NaHCO3 (0.1 mol/L) + Na2 CO3 ,
10H2 O (0.1 mol/L) to the solutions of CMC [30]. Various concentra-
tions of CMC and xanthan gum were tested, and finally converged 0.1
towards the following ones: 4 and 6 g/L for CMC, and 1 and 2 g/L for 0.01
xanthan gum. In addition to these non-Newtonian fluids, deionised
water and two aqueous solutions of glycerine (50% and 70%, v/v) 0.001
were also chosen as Newtonian fluids. 0.1 1 10 100 1000 10000
The rheology of these fluids was measured, at 20 ◦ C, by a shear rate (s-1)
rotational stress-controlled rheometrer (MCR500, PAAR Physica® )
(d) 100
equipped a cone-plate device (50 mm in diameter, 3 degree in cone
angle). Their density and surface tension were determined using Exp.
a densimeter ERTCO® and a tensiometer involving the Wilhelmy 10 Ostwald
plate method (3S GBX® ). The physical and rheological proper- Williamson
viscosity (Pa.s)
Table 2
Physical and rheological properties of the fluids.
the range of shear rate investigated, no major elastic property was describing the physical absorption process according to the two-
highlighted. film theory:
Based on these findings, some rheological models were chosen
G G
to mathematically describe the variation of apparent viscosity (a ) = kl · a · C or kl · a = (16)
Vl Vl · C
with shear rates ranging from 0.1 to 3000 s−1 . Among the large
variability available in the literature, two models were selected: where G/Vl is the mass throughput per unit volume of liquid and
C is a characteristic concentration difference. Eq. (16) implicitly
- the Ostwald–de-Waele’s model assumes that (i) the intensity of gas–liquid contacting is so high
that a quasi-uniform system is produced, (ii) the gas-phase mass
a = K · ˙ nost −1 (14) coefficient kg is negligible when compared to kl (low soluble gases),
(iii) the absorption rate at the interface is extremely fast, resulting in
where K and nost are respectively the consistency and flow indexes an equilibrium concentration of the dissolved gas at the interface C*
respectively. (e.g. C = C* − C). Hence, the establishment of the list of parameters
- the Williamson–Cross’s model influencing the main parameter kl a should respect the following
w rules [7]: (i) kl a must be independent of all geometrical parameters
a = (15)
˙ 1−nw
1 + (tw · ) (i.e. diameters of stirrer and tank, etc.), (ii) kl a must be independent
of the material parameters of gas phase, and (iii) kl a is an intensive
where w is a parameter describing a pseudo-Newtonian quantity because of its volume-related formulation.
behaviour for the smallest shear rates, tw is a time parameter As previously mentioned the number of the parameters influ-
characterizing the transition between the “pseudo-Newtonian” encing kl a, even performed in Newtonian liquids, is large and can
and purely shear-thinning behaviours, and nw is the consistency be decomposed according to:
index.
• The geometric parameters (see legend in Table 1), characterizing
The values of K, nost , w , tw , nw are reported in Table 2 for - the tank: Tt , Hl , curvature radius and angle (for tank’s bottom),
each fluid; they have been obtained by multi-parameter optimiza- ...
tions using the software Auto2fit® . In terms of consistency index, - the impeller: D, Ds , C, w, b, l, . . .
the most shear-thinning fluid appeared to be the solution of xan- - the sparger: d, number, diameter and shape of holes, . . .
than gum at 2 g/L. Based on the latter parameters, the apparent • The material parameters:
viscosities predicted by Eqs. (14) and (15) were compared to the l , l , g , g , , D, C*
experimental ones in Fig. 2. For all the fluids, a better agree- • The process parameters:
ment with experiments was obtained with the Williamson–Cross’s
model, insofar as it enables to describe the most faithfully possible Qg
g, N, Ug = , Tl , Ps , . . .
the shape of the rheograms over the whole range of shear rates. This · Tt2 /4
demonstrates that the fluids under test were not shear-thinning on
the whole range of shear rates investigated, three parameters being
required to well describe their behaviour.
Note that, in the present study, all the experiments were con-
ducted at room temperature and atmospheric pressure, inducing
3. Dimensional analysis for Newtonian and non-Newtonian thus that temperature and absolute pressure will not be listed. The
Fluids geometry of the tank was also unchanged, as well as the type and
position of both sparger and impeller. As a consequence, the list
3.1. Generation of i -sets governing aeration process for of individual physical quantities could be reduced: Table 3 shows
Newtonian fluids the dimensional matrix obtained with the reduced list of relevant
parameters.
In the present stirred tank, bubbles were directly generating It is voluntarily chosen to list the non-intensive parameters N
inside the liquid bulk by means of a gas sparger. Surface-aeration and D instead of power per unit of liquid volume (P/Vl ). The main
was of course present, but its contribution remained minor when motivation is that P/Vl is an intermediary variable which is not
compared to volume-aeration (see Section 4). Consequently, the always available (in particular at industrial scale), and thus using
overall volumetric gas–liquid mass transfer coefficient, kl a, can be such intensive variable would restrict the field of applications of
considered as the tractable quantity which is significantly influ- the final dimensional correlation which will be established.
enced by aeration conditions: it will be thus taken as target variable. In Table 3, the columns are assigned to the individual phys-
Remind that such choice is based on the following relationship ical quantities and the rows to the exponents appearing when
48 R. Hassan et al. / Chemical Engineering Journal 184 (2012) 42–56
Table 3
Dimensional matrix of the influencing parameters (Newtonian fluids).
g g g kl a Ug N D l l D
Mass, M (kg) 1 1 0 0 0 0 0 1 1 1 0
Length, L (m) −3 −1 1 0 1 0 1 −3 −1 0 2
Time, T (s) 0 −1 −2 −1 −1 −1 0 0 −1 −2 −1
Table 4
Unit core matrix obtained by linear transformations of dimensional matrix (Newtonian fluids).
g g g kl a Ug N D l l D
1
M + T + 2A 1 0 0 3
− 13 1
3
− 23 1 0 − 13 −1
3M + L + T + A 0 1 0 − 13 1
3
− 13 2
3
0 1 4
3
1
A = − 13 × (3M + L + 2T ) 0 0 1 2
3
1
3
2
3
− 13 0 0 1
3
0
R. Hassan et al. / Chemical Engineering Journal 184 (2012) 42–56 49
Such formulation of dimensionless numbers offers the advantage to material function s(p) which is apparent viscosity a (),
˙ the
enable the impact of all influencing parameters to be studied sepa- Pawlowski’s work [34] can be summed up, as follows:
rately, or in others words each dimensionless number is defined for
a single influencing variable. This is not the case in the literature • Firstly, all the dimensional parameters defined in the relevant list
where, for example, most of the authors used the aeration num- with fluids having constant properties (Newtonian case) should
ber, Na = Flg = Qg /(N · D3 ), in which the effect of gas flow rate is not be conserved, except for the variable physical properties in ques-
decoupled from the one of rotational impeller speed. tion (here apparent viscosity a ()).˙
At this state, the dependence of Eq. (32) is all that can be con- • Secondly, the Newtonian viscosity should be replaced by a refer-
tributed by the theory of similarity. The mathematical expression ence apparent viscosity, o , calculated at a reference shear rate,
for the function f, namely for the process relationship, has to be ˙ o . It is important to point out that any value for the reference
determined experimentally. shear rate can be chosen.
• Thirdly, the reference shear rate should be added in the list of
3.2. Extension of the theory of similarity to the cases of variable relevant parameters. At this stage, we can point out that some
material properties exceptions to this rule exist. Indeed, it has been demonstrated
that adding the reference point is not necessary when the mate-
When using the dimensional analysis to model system answers, rial function s(p), here a (),
˙ can be described by the following
it is generally assumed that the material properties remain unal- family of curves:
tered in the course of the process. However, the invariability of
s(p) = (A + B · p)c or s(p) = exp(A + B · p) (33)
material properties cannot be assumed when non-Newtonian flu-
ids are involved. Indeed, at the least one of the material properties, where A, B and C are three independent constants.
the apparent viscosity, can no longer be considered as a constant • Finally, a set of additional dimensionless numbers should be
inside the whole volume of the aerated stirred tank, insofar as the added in the relevant (dimensional) list to take into account the
dependency of this latter with the shear rates () ˙ generates a spa- dependency of material function. These dimensionless parame-
tial distribution of the liquid viscosity. The underlying question ters, called rheol , correspond to all the dimensionless ratios i
addressed to the dimensional analysis is now: how must the space which appear in the expression of the function u, except for the
of dimensionless numbers, i , be built in presence of such variable ratio /
˙ ˙ 0 :
material property?
In the case of materials with constant properties, no special pre-
1
da
caution should be made to guarantee that a process relationship {rheol } = {i } such as u = (˙ − ˙ 0 ) ·
correlating a set of dimensionless ratios is also applicable to another a (˙ 0 ) d˙ =
˙ ˙ o
material. This is not true for materials with variables properties, as ˙
demonstrated by [34]. In this case, we should first ensure as prior- =g ; {i } (34)
ity that a certain similarity exists for materials in order to extend ˙ 0
the range of validity of the process relationship to other materi- When integrating the previous guidelines, the list of the influ-
als. Despite that, the theory of similarity has little changed since encing parameters established for Newtonian fluids becomes for
its beginning, and the dimensional modelling involving materials non-Newtonian fluids:
with variable physical properties remains treated, in most of the
papers, as the case with constant material properties. The authors {kl a, g , g , g, Ug , N, D, l , , D, a (˙ o ), ˙ o , rheol } (35)
ignore then the fact that the spatio-temporal variability of mate- The next step is to find the type of material function describing the
rial properties influences the course of the process! One exception rheological behaviour of viscous fluids having shear-thinning prop-
is the modelling of the transformation processes where a mate- erties. As presented in Section 2, two models are well adapted for
rial having a temperature-dependence in viscosity is submitted to the investigated fluids: the Ostwald–de-Waele’s model (Eq. (14))
heat transfer condition. Most of attempts made to take into account and the Williamson–Cross’s model (Eq. (15)). As these models are
the variability of product properties in the reactor have consisted able to describe the variation of viscosity with shear rate for all the
in adding a new ratio raised to a certain exponent to character- aqueous solution of CMC and xanthan gum, each of them constitute
ize the system response. This ratio is defined by the ratio between one possible material function.
the viscosities at bulk temperature and at wall temperature. How-
ever, this kind of enlargement of the set of dimensionless ratios 3.2.1. Case No. 1: Ostwald–de-Waele’s model.
is theoretically valid only for few limited cases, that is to say only When the material function corresponds to the Ostwald–de-
if the material function (here viscosity versus temperature) sat- Waele’s model, the function u can be expressed as:
isfies specific criteria [34]. In other words, such method leads to ˙
biased predictions and thus, cannot be generalised when handling u= − 1 (nost − 1) (36)
other fluids. Despite this fact, this ratio remains systematically ˙ 0
used, whatever the products investigated, in most of the studies Consequently,
since [35].
{rheol } = {nost } (37)
The theoretically consistent way of proceeding has been
introduced by Pawlowski in 1971 [34] and remembered by [32,36]. The Ostwald–de-Waele’s model, defining as a = K (Eq. · ˙ nost −1
The method consists in introducing some additional parameters in (14)), verifies Eq. (33), implying thus that the reference shear rate,
the relevance list so as to take into account the variation in the ˙ o , can be removed from the relevance list. As a consequence, for
flow domain of the physical property, noted s (for example viscos- purely viscous fluids having shear-thinning properties, the addi-
ity), as a function of a parameter noted p (for example temperature tional parameters is restricted to nost and the Newtonian viscosity
or shear rate); s(p) is called the material function (for example (T) is replaced by a (˙ o ). Eq. (32) established for Newtonian fluids
or ()).
˙ This implies that the i -space governing the process will becomes then:
be extended in comparison to fluids having constant properties.
∗ a (o )
The guidelines to introduce the right number of additional dimen- kl a = g Ug∗ , Fr, ∗ , ∗ , Sc, ∗ = , nost (38)
g
sional parameters are detailed in [34]. Hence, when dealing with a
50 R. Hassan et al. / Chemical Engineering Journal 184 (2012) 42–56
Table 5
Dimensional results for kl a (expressed in s−1 ): Newtonian and non-Newtonian fluids.
When degenerated to the Newtonian case, the material function dimensionless numbers defining aeration process is enlarged by
associated with the Ostwald–de-Waele’s model leads to nost = 1 and two dimensionless numbers:
* = a /g where a = l is the Newtonian viscosity.
1
1/3
∗ g
9 = nw and 10 = tw = · (45)
3.2.2. Case No. 2: Williamson–Cross’s model tw g2
When the material function corresponds to the
Eq. (32) established for Newtonian fluids becomes then:
Williamson–Cross’s model, the function u can be expressed
as:
a (1/tw )
˙ − ˙ (n − 1) · (t · ˙ )(−nw ) kl a∗ = h Ug∗ , Fr, ∗ , ∗ , Sc, ∗ = ∗
, nw , tw (46)
0 w w 0 g
= · (39)
˙ 0 1 + (tw · ˙ 0 )1−nw
Note that the Williamson–Cross’s model leads to a Newtonian
Consequently, behaviour for particular values of nw and tw : namely nw = 1 and
{rheol } = {nw , tw · ˙ 0 } (40) tw = 1. In this case, l = w /2.
Water
[µ * = 54, σ * = 73385] 10-4
Gly50
[µ * = 589, σ * = 45066]
10-5
Gly70
[µ * = 1886, σ * = 50704]
(k l a )*mod
10-6
on Fr and Ug∗ , as already encountered in the literature (see review of encountered in the literature. Based on that, the effect of surface
1/3
[3]). Thus, Fig. 4 offers the advantage to easily appreciate the neg- tension, ∗ = /(g3 · g4 · g) , has been added in Eq. (51) (Case
ative effect of * on aeration performances: whatever (Ug∗ · Fr)2/3 , No. 3): the constant is in return increased and a negative exponent
an increase of * from 54 (water) to 1886 (glycerine 70%) leads to appears for * (−0.245). This latter logically confirms the positive
a drastic reduction of (kl a)* (more than 85%). impact of decreasing surface tension on aeration: when surface ten-
When imposing the exponent 2/3 to (Ug∗ · Fr), the best fit- sion is reduced (for example when adding surfactant or alcohol in
ting between experimental data and Eq. (49) without taking into the liquid phase), smaller bubble sizes are generated, leading to a
account * leads to: rise in interfacial area and thus in overall volumetric mass trans-
fer coefficient. In the literature [7], this usually results in positive
kl a∗ = 0.1420 · (Fr · Ug∗ )2/3 · (∗ )−0.591 (51) exponents for the Weber number.
The associated mean standard deviation is equal to 17.1%. To test At last, at a given temperature, under a given pressure, for the
the robustness of such correlation (in particular of the exponents geometry of the aerated stirred tank under test (in particular for
found), several cases are now tested when implementing the multi- Hl /Tt = 1, D/Tt = 0.4, Ds /Tt = 0.75 and for the other geometrical ratios
parameter optimization: characteristics of the system), the dimensional analysis states that
the process relationship for Newtonian fluids is expressed by:
- Case No. 1: all the exponents in Eq. (49) are kept free and * is 2/3 −0.591 −0.245
kl a∗ = 0.2097 · (Fr · Ug∗ ) · (∗ ) · ( ∗ )
neglected; 0.096 < Fr < 2.4, 0.0029 < Ug∗ < 0.029, 54 < ∗ < 1886 (52)
- Case No. 2: the exponents of Fr and of Ug∗ are imposed identical valid for
847 < ∗ < 1015, 50704 < ∗ < 73385, Sc = 7850
without any specified value, and * is neglected;
- Case No. 3: the exponents of Fr and Ug∗ are imposed equal to 2/3
Fig. 5 compares the experimental data with the dimensionless
(as in Eq. (51)), but now the effect of * is taken into account.
overall volumetric gas–liquid mass transfer coefficients predicted
by Eq. (52). A good agreement is observed, as the mean standard
In Table 6 are collected, for each case, the constant and expo-
deviation remains smaller than 17%. It can been observed that some
nents of Eq. (49) deduced from the multi-parameter optimization.
data corresponding to the smallest kl a* (obtained for low speeds in
In a general point of view, no major difference appears between the
glycerine) tends to move away from the straight lines representing
different cases: the exponent of * remains close to −0.59 (devia-
(kl a∗ )exp = ±20% · (kl a∗ )pred . This should be linked to the fact that,
tion < 1.3%), and the exponents of Fr and Ug∗ do not vary significantly
in such conditions, the gas–liquid regime corresponds to a loading
whether they are imposed identical or not (deviation < 6.5%). These
regime, and not to a complete dispersion regime characterizing all
findings confirm that the orders of magnitude of the exponents
the other operating conditions. The mechanical agitation is not then
found in Eq. (51) are relevant, and thus, that this correlation is
sufficient to disperse uniformly the bubbles in the whole volume of
mathematically robust. Note that, even if Fr and Ug∗ have the same
tank, in particular in the lower part of the tank (below the impeller).
exponent, the contribution of N is two times higher than the one
Hence, the relative contributions of the mechanisms controlling kl a
of superficial gas velocity, as the Froude number is expressed in
(mechanical agitation against sparger aeration) deviates from the
squared rotational impeller speed (Eq. (27)); this is coherent with
ones acting when a complete dispersion regime takes place. When
these observations made on the dimensional results (Section 4.1).
such changes in regime occurs, it becomes then difficult to define
The cases No. 1 and No. 2 leads to a slight improvement (smaller
an unique and accurate process relationship able to describe the
than 1%) of the mean standard deviation (ε); however, the expo-
entire range of operating conditions. Some deviations can be also
nents of Eq. (51) will be thereafter conserved, insofar as they lead to
observed for the highest vales of kl a; they can be here associated
the most simple formulation, and also as the value of 2/3 is already
with an increasing contribution of the surface aeration (mass trans-
fer occurring at the free liquid surface) when rotational impeller
Table 6 speed rises. Indeed, for the highest N, a significant vortex appears
Dimensionless modelling for Newtonian fluids using Eq. (49) (in bold the case in the centre of the tank and entrains many bubbles. In such condi-
retained).
tions, the mechanism (or regime) of aeration is deviated to a pure
Case ˛ a b c d ε (%) volume-aeration, and thus, the choice of kl a (and not kl a ) for
t surf
No. 1 0.02125 0.747 0.66 −0.605 – 16.1 tracting volume-aeration state is less representative.
No. 2 0.01535 0.683 0.683 −0.592 – 16.8 To conclude, it should be kept in mind that the validity of Eq.
No. 3 0.2097 2/3 2/3 −0.591 −0.245 16.8
(52) can be at present guaranteed only in the range of dimensionless
R. Hassan et al. / Chemical Engineering Journal 184 (2012) 42–56 53
Table 7
Dimensionless numbers characteristics for liquid properties.
* * nost ∗ost nw ∗
tw ∗w
numbers above mentioned, at given temperature and pressure, and In Fig. 6, (kl a)* is plotted as a function of (Ug∗ · Fr)2/3 for all the
in the geometry defined in Section 2. fluids. Such figure is particularly important as the effects of * and
nost can be visualized separately. Indeed, at a given *, an decrease
4.3. Dimensionless results for non-Newtonian fluids in nost clearly induces an increase in (kl a)*, or in other words,
the shear-thinning character of fluids favours the aeration perfor-
4.3.1. When considering the model of Ostwald–de-Waele mances, probably due to the spatial heterogeneity of viscosity. This
As any value can be considered (see Section 3.2), the refer- is illustrated when comparing either (i) the glycerine at 70% and
ence shear rate, ˙ o , has been arbitrary chosen equal to 120 s−1 . The the aqueous solutions of CMC at 4 g/L which have almost the same
dimensionless numbers describing the rheological properties asso- dimensionless apparent viscosity (1886 against 1866), but differ-
ciated with the Ostwald–de-Waele’s model (Table 2) and such ˙ o ent flow indexes (1 against 0.642), or (ii) the glycerine at 50% and
are collected in Table 7 (fourth and fifth columns). It can be logi- the aqueous solutions of xanthan gum at 1 g/L which have almost
cally observed that the flow index (nost ) decreases when increasing the same dimensionless apparent viscosity (589 against 538), but
the concentration, the smallest value being obtained for the most different flow indexes (1 against 0.543). The impact of nost seems
concentrated solution of xanthan gum. Dimensionless apparent however less pronounced than the one of *.
viscosity (∗ost ) are higher for aqueous solutions of CMC than the Fig. 7 compares the dimensionless overall volumetric mass
ones of xanthan gum. transfer coefficients measured in presence of non-Newtonian fluids
Basing on the reasons mentioned in the case of Newtonian fluids, with the ones predicted using the process relationship estab-
the possible changes in * and in Sc will be neglected in Eq. (38) lished for Newtonian fluids (Eq. (52)). The points related to each
previously established when the model of Ostwald–de-Waele is non-Newtonian fluid are regrouped along individual straight lines
considered. Thus, Eq. (38) becomes: which are parallel to each others and to the curves previously
obtained for Newtonian fluids. This illustrates that the introduc-
a (o ) tion of nost into the dimensionless modelling (Eq. (54)) could be
kl a∗ = g Ug∗ , Fr, ∗ , ∗ = , nost (53)
g a mean for differentiating the non-Newtonian fluids from each
others and from the Newtonian ones. Nevertheless, this group of
The simplest monomial form is here also looked for the g -relation four straight lines (one for each fluid) is not classified according
(no mechanistic information available): to the values of flow index, in particular the associated nost are
not decreasing when deviating from the Newtonian curve (nost = 1).
kl a∗ = ˛ · (Fr)a · (Ug∗ )b · (∗ )c · ( ∗ )d · (nost )e (54) Indeed, the straight line the closest from the Newtonian curve
corresponds to nost = 0.543 (XG 1 g/L), followed by nost = 0.373 (XG
where ˛ , a , b , c , d and e are respectively the constant and
2 g/L), then nost = 0.642 (CMC 4 g/L), until reaching the most distant
the exponents to which the dimensionless Froude, superficial gas
line, nost = 0.527 (CMC 6 g/L). This finding tends to show that gather-
velocity, viscosity, surface tension and flow index numbers are
ing all the data related to non-Newtonian fluids over the Newtonian
raised (Auto2fit® ). However, it is important to keep in mind that
ones will be difficult by adding only the flow index nost in the
the modelling for Newtonian fluids is a degraded case of the one
relevant list. In others words, this would mean that the material
for non-Newtonian fluids. As a consequence, the exponents of Fr,
function associated with the model of Ostwald–de-Waele could be
Ug∗ , * and * (namely a , b , c , d ) are already defined, and will be
insufficient and/or unsuitable for describing completely the effect
thus taken equal respectively to 2/3, 2/3, −0.591 and −0.245 as in
of the shear-thinning properties on (kl a)*.
the Newtonian case (Table 6, Case No. 3).
-5 -4
10-6 10 10 10-3
10-3
(k l a )*exp
Water
Gly50 n1.0E-04 -4
ost =0.543 10
Gly70
CMC4 n ost =0.373
CMC6
n ost =0.642
XG1 10-5
XG2 n ost =0.527
(k l a )* mod
10-6
-6 -5 -4
10-6 10-5 10-4 10-3 10 10 10 10-
10-3 10-
(k l a )*exp (k l a *)exp
Water
Water
Gly50
Gly50
10-4 Gly70 10-4
Gly70
CMC4 CMC4
CMC6 CMC6
XG1 10-5 XG1 10
-5
XG2
XG2
(k l a )* mod
(k l a *)mod
10-6 -6
10
Fig. 8. Modelling using the Ostwald–de-Waele’s model: comparison between the
Fig. 9. Modelling using the Williamson’s model: comparison between the experi-
experimental (kl a)* and the values predicted by Eq. (55). The dotted lines correspond
mental (kl a)* and the values predicted by Eq. (57). The dotted lines correspond to a
to a deviation of ±20%.
deviation of ±20%.
(Case No. 5) in which * is not taken into account, and the second tively) are determined by the multi-parameter optimization. The
one (Case No. 6) in which * is taken into account with a non- following process relationship is then obtained:
imposed exponent. The results are collected in Table 8 (Case No. 4 kl a∗ = 0.02109 · (Fr · Ug∗ )
2/3
· (∗ )
−0.591
· ( ∗ )
−0.245
· (nw )
−2.399 ∗
· (tw )
−0.168
cases (deviations below 2.4% and 3.2% respectively). Note also that,
The present paper dealt with a consistent dimensionless analy-
when the exponent of * is kept free in the optimization, the value
sis of gas–liquid mass transfer in an aerated stirred tank containing
found (d) is close to the one already obtained for the Newtonian
purely viscous fluids with shear-thinning fluids. More particularly,
case. These tests clearly validate Eq. (57) in a mathematical point
this work showed how to proceed:
of view.
In addition, it can be observed that the exponent of tw ∗ , and
∗
thus the impact of tw on kl a*, is negative. This is illustrated - to construct a complete list of relevant parameters able to build
in Fig. 10 where kl a* is plotted (Ug∗ · Fr)2/3 · ∗ −0.591 · nw −2.399 · an unique -space which keeps unchanged for both Newtonian
∗ −0.245 (according to the exponents found in Eq. (57) or Case and non-Newtonian fluids.
No. 7 in Table 9). The points related to each non-Newtonian fluid - and consequently to elaborate, without pitfalls, a set of
are regrouped along individual straight lines which deviate more dimensionless numbers characterizing all the factors governing
or less from the curves previously obtained for Newtonian flu- absorption rate coefficients (kl a) in a stirred tank where shear-
ids. These straight lines are remarkably classified according to thinning fluids are involved.
decreasing tw ∗ as far as going far from the Newtonian points (t ∗ = 1).
w
Thus, at given dimensionless apparent viscosity and flow index, the This theoretical approach was supported by a set of kl a mea-
smallest is the time parameter of the Williamson–Cross’s model surements in a tank stirred by a six-concave-blade disk turbine
(tw ) the highest is the overall volumetric mass transfer coefficient and aerated by a ring sparger, under different operating condi-
(kl a). tions (rotational impeller speed, gas flow rate) and for various fluids
All the findings demonstrate that the rigorous extension of the (water, glycerine 50% and 70%, solutions of CMC at 4 and 6 g/L,
theory of similarity to the case of variable material properties (e.g., solutions of xanthan gum at 1 and 2 g/L). These measures were val-
dependence of viscosity with shear rate) and the choice of an appro- idated notably by means of two methods (physical and chemical).
priate material function (here issued from the Williamson–Cross’s At last, a suitable dimensionless correlation could be obtained for
model) have made possible an accurate dimensionless modelling describing all the variations of kl a if and only if all the rheological
of the impact of the shear-thinning character of fluids on the aer- properties were correctly taken into account (e.g. using the three
ation performances in a stirred tank. Such approach also enables parameters involved in the model of Williamson–Cross) instead of
to understand how each parameter, either operating parameter neglecting some of them (namely, by simple fitting with the model
of Ostwald–de-Waele. It was expressed by:
References [20] W.K. Lewis, W.G. Whitman, Principles of gas absorption, Ind. Eng. Chem. 16
(1974) 1215–1220.
[1] J.B. Joshi, A.B. Pandit, M.M. Sharma, Mechanically agitated gas–liquid reactors, [21] J. Ingham, I.J. Dunn, E. Heinzle, J.E. Prenosil, Chemical Engineering Dynamics.
Chem. Eng. Sci. 37 (6) (1982) 813–844. Modeling with PC Simulation, VCH Publishers Inc., New York, 1994, ISBN 3-
[2] E.L. Paul, V.A. Atiemo-Obeng, S.M. Kresta, Handbook of Industrial Mixing: Sci- 527-28577-6.
ence and Practise, J. Wiley & sons, 2004. [22] C.O. Vandu, K. Koop, R. Krishna, Volumetric mass transfer coefficient in a slurry
[3] F. Garcia-Ochoa, E. Gomez, Bioreactor scale-up and oxygen transfer rate in bubble column operating in the heterogeneous flow regime, Chem. Eng. Sci. 59
microbial processes: an overview, Biotechnol. Adv. 27 (2009) 153–176. (2004) 5417–5423.
[4] M. Martin, F.J. Montes, M.A. Galan, Bubbling process in stirred tank reac- [23] J.K. Bewtra, W.R. Nicholas, L.B. Polkowski, Effect of temperature on oxygen
tors. II. Agitator effect on the mass transfer rates, Chem. Eng. Sci. 63 (2008) transfer in water, Water Res. 4 (2) (1970) 115.
3223–3234. [24] F. F. Cabaret, L. Fradette, P.A. Tanguy, Gas–liquid mass transfer in unbaffled
[5] M. Martin, F.J. Montes, M.A. Galan, Mass transfer rates from bubbles in stirred dual-impeller mixers, Chem. Eng. Sci. 63 (2008) 1636–1647.
tanks operating with viscous fluids, Chem. Eng. Sci. 65 (2010) 3814–3824. [25] V. Linek, V. Vacek, P. Benes, A critical review and experimental verification of
[6] K. Van’t Riet, Review of measuring methods and results in non-viscous the correct use of the dynamic method for the determination of oxygen transfer
gas–liquid mass transfer in stirred vessels, Ind. Eng. Chem. Process Des. Dev. in aerated agitated vessels to water, electrolyte solutions and viscous liquids,
18 (3) (1979) 357–364. Chem. Eng. J. 34 (1987) 11–34.
[7] M. Zlokarnik, Sorption characteristics for gas–liquid contacting in mixing ves- [26] A.S. Khare, K. Niranjan, Impeller-agitated aerobic reactor: the influence of tiny
sels, Adv. Biochem. Eng. 8 (1979) 133–157. bubbles on gas hold-up and mass transfer in highly viscous liquids, Chem. Eng.
[8] H. Judat, Gas/liquid mass transfer in stirred vessels - a critical review, Ger. Chem. Sci. 50 (7) (1995) 1091–1105.
Eng. 5 (1982) 357–363. [27] P. Painmanakul, K. Loubière, G. Hébrard, M. Mietton-Peuchot, M. Roustan, Effect
[9] V. Schlüter, W.D. Deckwer, Gas–liquid mass transfer in stirred vessels, Chem. of surfactants on liquid-side mass transfer coefficients, Chem. Eng. Sci. 60
Eng. Sci. 47 (9–11) (1992) 2357–2362. (2005) 6480–6491.
[10] M. Roustan, A. Liné, Rôle du brassage dans les procédés biologiques d’épuration, [28] L.-K. Ju, C.S. Ho, The structure-breaking effect on oxygen diffusion coefficients in
Tribune de l’Eau 49 (1996) 109–115. electrolyte and polyelectrolyte solutions, Can. J. Chem. Eng. 67 (1989) 471–476.
[11] J.F. Perez, O.C. Sandall, Gas absorption by non-Newtonian fluids in agitated [29] T. Espinosa-Solares, E. Brito-De La Fuente, A. Tecante, P.A. Tanguy, Gas dis-
vessels, AIChE J. 20 (1974) 770. persion in rheologically-evolving model fluids by hybrid dual mixing systems,
[12] H. Yagi, F. Yoshida, Gas absorption by Newtonian and non-Newtonian flu- Chem. Eng. Technol. 25 (7) (2002) 723–727.
ids in sparged agitated vessel, Ind. Eng. Chem. Process Des. Dev. 14 (1975) [30] M. Hilal, Etude de l’écoulement d’un fluide non Newtonien à travers les réac-
488–493. teurs à lit fixe: chute de pression, dispersion axiale et transfert de matière,
[13] M. Nishikawa, M. Nakamura, H. Yagi, K. Hashimoto, Gas absorption in aerated Thèse de doctorat, Université of Nantes N 91 NANT 2049, 1991.
mixing vessels, J. Chem. Eng. Jpn. 14 (3) (1981) 219–226. [31] M. Mori, J. Isaac, I. Seyssiecq, N. Roche, Effect of measuring geometries and
[14] M. Nishikawa, M. Nakamura, K. Hashimoto, Gas absorption in aerated mix- of exocellular polymeric substances on the rheological behaviour of sewage
ing vessels with non-Newtonian liquid, J. Chem. Eng. Jpn. 14 (3) (1981) sludge, Chem. Eng. Res. Des. 86 (2008) 554–559.
227–232. [32] M. Zlokarnik, Scale-up in Chemical Engineering Second, Completely Revised
[15] E. Costa, A. Lucas, J. Aguado, J.A. Avila, Transferencia de materia en tanques agi- and Extended edition, Wiley-VCH Verlag GmbH & Co., 2006.
tados: burbujeo de gases en líquidos newtonianos y no newtonianos. I. Turbinas [33] T. Szirtes, Applied Dimensional Analysis and Modelling, McGraw-Hill, New
de 6 paletas y difusor plano, An. Quim. 78 (1982) 387–392. York, 1998.
[16] R.S. Albal, Y.T. Shah, A. Schumpe, N.L. Carr, Mass transfer in multiphase agitated [34] J. Pawlowski, Die ähnlichkeitstheorie in des physikalissch-technischen
contactors, Chem. Eng. J. 27 (1983) 61–80. Forschung- Grundlagen und Anwendungen, Springer Verlag,
[17] F. Garcia-Ochoa, E. Gomez, Mass transfer coefficient in stirred tank reactors for Berlin/Heidelberg/New York, 1971.
xanthan gum solution, Biochem. Eng. J. 1 (1998) 1–10. [35] E.N. Seider, G.E. Tate, Heat transfer and pressure drop of liquids in tubes, Ind.
[18] C. Xuereb, M. Poux, J. Bertrand, Agitation et mélange: Aspects fondamentaux Eng. Chem. 28 (1936) 1429.
et applications industrielles, L’Usine Nouvelle Dunod, 2006. [36] M. Zlokarnik, Scale-up of processes using material systems with variable phys-
[19] R. Hassan, J. Loubière, J. Legrand, Gas–liquid mass transfer in a stirred tank for ical properties, Chem. Biochem. Eng. Q 15 (2) (2001) 43–47.
non-Newtonian fluids. Application to the autothermal thermophilic digestion [37] R. Hassan, Etude expérimentale et modélisation du transfert de matière gaz-
of sludge, in: 9th International Congress of Chemical and Process Engineering liquide en cuve agitée en présence de fluides non-Newtoniens simulant des
(CHISA), Prague, Czech Republic, 2010. boues de station d’épuration, Thesis, University of Nantes, 2011.