Vous êtes sur la page 1sur 10

Chemical Engineering and Processing 49 (2010) 1249–1258

Contents lists available at ScienceDirect

Chemical Engineering and Processing:


Process Intensification
journal homepage: www.elsevier.com/locate/cep

CFD simulation and optimization of effective parameters for biomass production


in a horizontal tubular loop bioreactor
S.M. Mousavi, S.A. Shojaosadati ∗ , J. Golestani, F. Yazdian
Biotechnology Group, Chemical Engineering Department, Tarbiat Modares University, P.O. Box 14115-143, Tehran, Iran

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

Article history: The focus of the current study was to perform an experimental investigation and computational fluid
Received 4 April 2009 dynamic (CFD) simulation of flow hydrodynamics in a forced-liquid horizontal tubular loop bioreactor for
Received in revised form 31 August 2010 the production of biomass. The simulations were performed using the FLUENT commercial CFD package,
Accepted 19 September 2010
a segregated unsteady solver and a two-phase Eulerian multiphase model. To validate the simulation
Available online 25 September 2010
results, several experiments were performed in a pilot bioreactor. In addition, the design of experiments
methodology using a Taguchi orthogonal array (OA) was applied to evaluate the influence of four factors
Keywords:
on the hydrodynamic behavior of the bioreactor. The effective parameters considered for optimization
CFD simulation
Optimization
were air inlet velocity, liquid inlet velocity, bubble diameter, and viscosity. An L9 OA was used to conduct
Gas hold up the Taguchi experiments to study the significance of these parameters and the possible effects of any
Biomass production two-factor interactions. The optimum conditions and most significant process parameters affecting the
Forced-liquid horizontal tubular loop hydrodynamic behavior were determined using an analysis of variance model. The results showed that
bioreactor the liquid inlet velocity had the most influence on the air volume fraction in the bioreactor. A subsequent
Taguchi method confirmatory test demonstrated that the results were within the confidence interval.
© 2010 Elsevier B.V. All rights reserved.

1. Introduction heat transfer, mass transfer [7] and liquid mixing [8]. Thus, under-
standing gas holdup is essential for a reliable description of a loop
Bacteria that can utilize methane as a sole source of carbon and bioreactor [9].
energy were first recognized in 1906 [1]. Recent studies have indi- Computational fluid dynamics (CFD) is a powerful technique for
cated the potential for methane-oxidizing bacteria to be used as a the acquisition of a complete picture of the transient behavior of
source of protein as either a food supplement or fodder [2]. Despite fluids under given conditions [8,10–13], and it is less expensive
its low solubility, flammability, and lack of purity in natural sources, than a high quality experimental facility [14]. Due to the significant
methane is a good candidate for biomass production due to its enhancement of CFD models and increases in calculation speeds,
non-toxicity, selectivity, and volatility [3]. CFD has been used with increasing frequency to simulate hydrody-
Loop reactors, which are characterized by a definitely directed namics in complex flows [15]. Because multiphase mixtures occur
circulation flow that can be driven in fluid or fluidized systems very frequently in processing industries, modeling methods need
[4,5], have shown an acceptable performance in the production to be extended to deal with these multiphase flows [16]. Due to the
of biomass from natural gas [4] because of their unique hydro- inherent complexity of two-phase flows from both a physical and
dynamic characteristics. They are especially appropriate for fluid numerical point of view, “general”-use CFD codes are non-existent.
systems requiring highly coalesced dispersion properties. On the The main reasons for this absence include the physical complexity
other hand, their simple construction and operation result in low of two-phase flow phenomena, complex interactions between the
investment and operational costs. To be used in large scale produc- gas and the liquid phases [17] and the complexity of the governing
tion plants, however, much study is still required to optimize their equation numerics [18].
design and operation [4,5]. Taguchi design of experiment methods are extensively used
Gas holdup is an important hydrodynamic characteristic of mul- to optimize the critical parameters of any process [19]. They
tiphase systems [6] not only for its effect on the liquid circulation largely minimize the number of experiments required to achieve an
rate, but also for its effect on gas residence time, oxygen transfer, optimal set of performance characteristics for a process by incorpo-
rating readily available fractional factorial matrices or orthogonal
arrays (OA) [20]. This approach is a combination of mathematical
∗ Corresponding author. Tel.: +98 21 82883341; fax: +98 21 82884931. and statistical techniques used in empirical studies. In the Taguchi
E-mail address: shoja sa@modares.ac.ir (S.A. Shojaosadati). approach, OA and analysis of variance are used as the analytical

0255-2701/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2010.09.013
1250 S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258

Table 1
Bioreactor dimensions and operational parameters.

Description Value

Bioreactor diameter (m) 0.03


Working volume (l) 3
Separator diameter (m) 0.11
Separator height (m) 0.19
Bioreactor height (m) 0.4
Bioreactor length (m) 2.2
Static mixer 2 Komax, 0.1 m length, 0 inclination angle
Number of sparer holes 6

presented in Table 1. Medium temperature was controlled at 30 ◦ C


(Fig. 1).

2.3. Experimental procedures

The overall gas hold up ε was determined by the volume expan-


Fig. 1. Schematic diagram of the forced-liquid horizontal tubular loop bioreactor sion method [4,22]. The mixing time tm and circulation time tc were
(HTLB) used for the experiments: (1) natural gas capsule; (2) air capsule; (3) gas determined using the tracer response technique, which is based on
pump; (4) liquid pump; (5) gas–liquid separator; (6) condenser; (7) helium capsule; the observation that a decaying sinusoidal type of response can be
(8) dissolved methane detector. Magnified views of the (A) static mixer and (B)
detected downstream of the injection point after a pulse of tracer
sparger.
is injected into the flow [23]. In this study, this procedure, based on
the work of Blenke (1979), used a dye (Brilliant Blue G,  = 595 nm)
as the tracer. In the loop reactor, the tracer (0.5 ml) was injected
tools. OA can minimize the number of experimental replicates, into the gas–liquid separating section, and the optical density of
and ANOVA can estimate the effect of a factor on the characteris- the samples was measured by their absorbance at 595 nm (A595 )
tic properties. A conventional statistical experimental design uses using a UV–visible spectrophotometer (Varian, Australia). The opti-
measured values of the characteristic properties to can determine cal density changes were recorded until the response of the pulse
the optimal conditions, whereas the Taguchi method identifies the was completely damped. All of the experiments were performed
experimental condition having the least variability to determine in triplicate. The circulation time was regarded the time difference
the optimal condition. As a result, a Taguchi experiment can be between two adjacent peaks (not taking the first peak into account),
accomplished within a limited time frame, at a reduced cost, and and the mixing time was regarded as the time required until the
with results comparable to those of a full factorial experiment. response oscillation was damped in all subsequent measurements
The specific objectives of this investigation are to simulate flow within 10% of the final response [23]. The residence time tr was cal-
behavior in a bioreactor and apply Taguchi OA experimental design culated according to the following equation [4] using the inlet air
to optimize the influence of various process parameters involved volume divided by air flow rate as follows:
in biomass production in a forced-liquid HTLB.
Vg
tr = ·
(1)
2. Materials and methods Vg

2.1. Microorganism and growth medium


2.4. Analytical methods
The microorganism (Methylomonas spp.) used in this work was
isolated from samples obtained from an oil field in Iran. The bac- 2.4.1. Bubble size determination
teria were cultured on methane salt broth, a carbonless salt broth Bubble size was measured in the horizontal part of the loop
medium [21]. reactor using a photographic technique with a high speed camera
(Canon Power shot S3 IS). The experimental data was processed
2.2. Bioreactor characteristics using Sigma plot software to evaluate the bubble diameter and
determine the distribution. The correction to actual size was based
A schematic drawing of the single-walled forced-liquid HTLB on the scale attached to the horizontal tube. For ellipsoidal bub-
with a dissolved methane detector is shown in Fig. 1. The experi- bles, the major and minor axes of the bubble were measured. The
ments were carried out in a bioreactor with a diameter, height and Sauter mean bubble diameter (d32 or dp ) was calculated using the
length of 3, 40, and 220 cm, respectively. The air and natural gas following equation [24]:
were distributed by a perforated tube through a gas pump in dif-  3
ferent zones within horizontal section, and the liquid medium was nd
d32 = dp =  i i2 (2)
circulated by a liquid pump. The horizontal section contained two ni di
Komax static mixers, which are used to generate a swirling motion
(vortex) as the liquid passes through. Vorticity is a mathematical
concept used in fluid dynamics related to the amount of circula- 2.4.2. Viscosity determination
tion or rotation in a fluid. The vorticity at a point is a vector that The viscosity of the medium, which is composed of distilled
is defined as the curl of the velocity. Following measurement with water and aqueous carboxymethyl cellulose solutions of 0.01%
a gas flow meter, the gas was injected through annular holes in wt/v, 0.02% wt/v and 0.05% wt/v, was measured using an Ostwald-
front of the static mixer. The geometric details of the bioreactor are type viscometer.
S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258 1251

3. Mathematical formulations and


dq2
3.1. Governing equations p = (10)
18q
The governing equations were solved sequentially. The non- The drag coefficient depended on the flow regime and the liquid
linear governing equations were linearized to produce a system of properties. For rigid spheres, the drag coefficient is usually approx-
equations for the dependent variables in every computational cell. imated by the standard drag curve [25]:
The resulting linear system was then solved to yield an updated 
flow-field solution. A point implicit (Gauss–Seidel) linear equation 24(1 + 0.15 Re0.687 )/Re if Re < 1000
CD = (11)
solver was used in conjunction with an algebraic multigrid method 0.44 if Re ≥ 1000
to solve the resulting scalar system of equations for the dependent Because of internal gas circulation and deformation, however,
variable in each cell. bubbles do not necessarily behave as rigid spheres; their drag coef-
The continuity equation for phase q is: ficient can therefore differ from that predicted by the standard drag
∂ curve. According to Clift et al., the drag on bubbles in pure water
(˛q q ) + ∇ · (˛q q vq ) = 0 (3)
∂t is less than the drag predicted by the standard drag curve. On the
other hand, in contaminated systems, the surfactants tend to col-
The description of the multiphase flow as an interpenetrating
lect at the rear of the bubble, reducing the slip along the surface
continuum incorporated the concept of phasic volume fraction,
of the bubble and causing the bubbles to behave more like rigid
denoted by ˛q . The volume fractions represent the space occupied
particles [26]. As tap water is used in most experiments reported
by each phase, and the laws of conservation of mass and momen-
in the literature, the standard drag curve equation can be used as
tum are satisfied by each phase individually. The derivation of the
an estimate for the drag coefficient of a spherical bubble [27].
conservation equations can be done by calculating the ensemble
The relative Reynolds number (Re) for the primary phase (liquid)
average of the local instantaneous balance for each of the phases or
q and secondary phase p (air) can be obtained from the following
by using the mixture theory approach.
equation:
The volume fraction of each phase was calculated from a conti-
nuity equation: q |vp − vq |dp
Re = (12)
∂ q
(˛q ) + ∇ · (˛q vq ) = 0 (4)
∂t Nearly all definitions of f include a drag coefficient that is based
The momentum balance for phase q yields: on the relative Reynolds number (Re), but this drag function differs
among the exchange-coefficient models. For all these situations,

(˛q q vq ) + ∇ · (˛q q vq vq ) = −˛ ∇ p + ∇ · ¯¯ q + ˛q q g Kpq should tend to zero whenever the primary phase is not present
∂t within the domain. To enforce this requirement, the drag function

n f is always multiplied by the volume fraction of the primary phase
+  pq ) + ˛q q (Fq )
(R (5) q.
p=1 In comparison with single-phase flows, the number of terms to
be modeled in the momentum equations in multiphase flows is
The q phase stress–strain tensor (¯¯ q ) was defined as follows: large, which complicates the modeling of turbulence in multiphase
 2
 simulations.
q ∇ · vq Ī¯
T
¯¯ = ˛q q (vq + vq ) + ˛q q − (6) In the present study, standard a k–ε turbulence model was used.
3
The equations k and ε that describe the model are as follows
The interphase force, R  pq , depended on the friction, pressure, [28]:
cohesion, and other effects and was subject to the conditions that ∂
 
t,m
R  qp , R
 pq = −R  qq = 0 and that (m k) + ∇ · (m vm k) = ∇ · ∇ k + Gk,m − m ε (13)
∂t k
n
n
and
 pq = Kpq (vp − vq )
R (7)

 
t,m
p=1 p=1 (m ε) + ∇ · (m vm ε) = ∇ · ∇ ε + C1ε Gk,m − C2ε m ε (14)
∂t ε
where Kpq (Kqp ) is the interphase momentum exchange coefficient.
where the mixture density and velocity, m and vm , are computed
In the simulations of the multiphase flow, the lift force can be
from:
considered for the bubbles in the secondary phase. This force is
important if the bubble diameter is very large. The software pro- 
N

gram assumed that the bubble diameters were smaller than the m = ˛i i (15)
distance between them, so the lift force was insignificant compared i=1
with the other forces, such as drag force. Therefore, there was no and
reason to include this extra term. N
The exchange coefficient for these types of bubbly, gas–liquid ˛i i v
vm = i=1 (16)
mixtures can be written in the following general form: N
˛
i=1 i i
˛q ˛p p f
Kpq = (8) The turbulent viscosity t,m is computed from:
p
k2
where the drag function f was defined differently for the each of t,m = m C (17)
ε
the exchange-coefficient models (as described below) and the par-
ticulate relaxation time  p was defined as: The production of turbulent kinetic energy Gk,m is computed
from:
CD Re
f = (9) T
Gk,m = t,m (∇ vm + (∇ vm ) ) : ∇ vm (18)
24
1252 S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258

Fig. 2. (a) Actual image of the laboratory set up. (b) Schematic diagram of the simulated bioreactor: (1) static mixer; (2) air sparger; (3) liquid inlet; (4) mixture outlet.

3.2. Species transport equations were negligible, (b) the structure of the static mixer was simplified,
and (c) the four jet holes were considered to be four active holes
To solve conservation equations for chemical species, software in the spargers. The experimental set up and a schematic of the
predicts the local mass fraction of each species, Yi , through the simplified bioreactor are shown in Fig. 2.
solution of a convection-diffusion equation for the ith species. This The phase-coupled simple (PC-SIMPLE) algorithm, which
conservation equation takes the following general form: extends the SIMPLE algorithm to multiphase flows, was applied
to determine the pressure–velocity coupling in the simulation. The

(Yi ) + ∇ · (vYi ) = ∇ · j + Ri + Si (19) velocities were solved coupled by phases, but in a segregated fash-
∂t
ion. The block algebraic multigrid scheme used by the coupled
here Ri is the net rate of production of species i by chemical reac- solver was used to solve a vector equation formed by the velocity
tion and Si is the rate of creation by addition from the dispersed components of all phases simultaneously. Then, a pressure correc-
phase plus any user-defined sources. An equation of this form will tion equation was built based on total volume continuity rather
be solved for N − 1 species where N is the total number of fluid phase than mass continuity. The pressure and velocities were then cor-
chemical species present in the system. Since the mass fraction of rected to satisfy the continuity constraint. The volume fractions
the species must sum to unity, the Nth mass fraction is determined were obtained from the phase continuity equations.
as one minus the sum of the N − 1 solved mass fractions. To min- To satisfy these conditions, the sum of all volume fractions
imize numerical error, the Nth species should be selected as that should be equal to one.
species with the overall largest mass fraction, such as N2 when the For the continuous phase (liquid phase), the turbulent contribu-
oxidizer is air. tion to the stress tensor was evaluated by the k–ε model described
by Sokolichin and Eigenberger (1999) [29] using the following
3.2.1. Mass diffusion standard single-phase parameters: C = 0.09, C1ε = 1.44, C2ε = 1.92,
In Eq. (19), Ji is the diffusion flux of species i, which arises due to k = 1 and ε = 1.3.
concentration gradients. By default, FLUENT uses the dilute approx- The flow boundary conditions applied to each phase set the inlet
imation, under which the diffusion flux can be written as: gas velocity to 10.6 m s−1 , the inlet liquid velocity to 0.24 m s−1 , the
gas phase fraction to 1 and the bubble (air) diameter to 15 mm
Ji = −Di,m ∇ Yi (20) for all simulations except one, which investigated the effect of air
bubble diameter on flow hydrodynamics. The initial mass fraction
Here Di,m is the diffusion coefficient for species i in the mixture.
of the biomass in the system was set to 0.1 for the simulations.
A constant bubble diameter was used as another assumption and
4. Numerical implementation simplification in the simulations. The constant air bubble diameter
was obtained from the average bubble diameter that was measured
4.1. Simulation characteristics in the laboratory. To investigate the effect of air bubble diameter
on flow hydrodynamics, three different sizes (6, 9 and 12 mm, but
In the present work, a commercial grid-generation tool, GAMBIT not 15 mm) were applied. The gravitational acceleration vector (g)
2.2 (Fluent Inc., USA) was used to create the geometry and gener- was [0,0,−9.8] m s−2 , and the time step size was 0.01.
ate the grids. The use of an adequate number of computational cells The discretization scheme for each governing equation involved
while numerically solving the governing equations over the solu- the following procedure: PC-SIMPLE for the pressure–velocity cou-
tion domain is very important. To divide the geometry into discrete pling and “first order upwind” for the momentum, volume fraction,
control volumes, more than 4.7 × 105 3D tetrahedral computa- turbulence kinetic energy and turbulence dissipation rate. The
tional cells and 112,793 nodes were used. Because the bioreactor under-relaxation factors that determine how much control each of
geometry for this simulation was so complicated, the following the equations has in the final solution were set to 0.5 for the pres-
simplifications were assumed in the simulated geometry: (a) the sure and volume fraction, 0.8 for the turbulence kinetic energy and
effects of the liquid pump and the separator part of the bioreactor turbulence dissipation rate, and 0.6 for all species. The gas holdup,

Table 2
Comparison between three grid resolutions.

Case number Number of nodes Grid size Air volume fraction Air velocity Time (s)

1 89235 361696 0.070267 0.18 200


2 112793 470428 0.08398 0.19 200
3 137395 591377 0.086932 0.196 200
S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258 1253

Table 3
Studied experimental control factors and their levels.

Factor Level 1 Level 2 Level 3

(A) Air inlet velocity (m/s) 10.62 21.23 31.85


(B) Liquid inlet velocity (m/s) 0.24 0.35 0.47
(C) Bubble diameter (mm) 6 9 12
(D) Viscosity (×10−3 Pa s) 1.085 1.243 1.301

Table 4
The basic Taguchi L9 (34 ) orthogonal array.

Simulation number A B C D

1 1 1 1 1
2 1 2 2 2
3 1 3 3 3
4 2 1 2 3
5 2 2 3 1
6 2 3 1 2
7 3 1 3 2
8 3 2 1 3
9 3 3 2 1

Fig. 3. Size distribution of bubble diameter.


cases. A comparison of the data generated at 200 s for these three
grid sizes is given in Table 2. Based on the accuracy of resulting data,
measured as the phase volume fraction, was recorded after every which relied on the air volume fraction and air velocity, case 2 with
five time steps. 470,428 cells was selected as the optimum mesh size. Although
Using mentioned values for the under-relaxation factors, a rea- the final volume fractions and velocities for cases 2 and 3 were
sonable rate of convergence was achieved. The convergence was very similar, case 2 was selected for the simulations because of the
considered to be achieved when the conservation equations of smaller number of meshes.
mass and momentum were satisfied, which was considered to
have occurred when the normalized residuals became smaller 5. Taguchi optimization methodology
than 5 × 10−5 . The normalization factors used for the mass and
momentum were the maximum residual values after the first few 5.1. Orthogonal array
iterations.
The Qualitek-4 (version 4.82.0) software package was used for
4.2. Confirmation of grid independence the Taguchi design and subsequent analyses, including evaluations
of the influence of individual factors and the interaction between
The results are grid independent. To select the optimized num- the factor pairs as well as the determination of the optimum con-
ber of grids, a grid independence check was performed. The three ditions for the bioreactor performance. The appropriate OA for the
mesh sizes that were examined included a total of 361,696, 470,428 experiment was determined by the software. To obtain the best
and 591,377 cells, respectively. The data were recorded at 200 s, performance of the bioreactor, the program selected an L9 (34 )
which was the point at which the system stabilized for all three orthogonal array with four control factors and three levels to eval-

Fig. 4. Temporal evaluation of the air volume fraction in the bioreactor after (a) 0.5 s, (b) 1.0 s, (c) 1.5 s, and (d) 4.0 s.
1254 S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258

Table 5
Interactions between factor pairs.

Interacting factor pairs Interaction severity index (%)

A–C 61.62
A–D 43.6
B–C 29.65
B–D 13.37
A–B 4.06
C–D 2.32

performed. Thus, running a confirmation experiment once the opti-


mum condition has been determined is recommended.

6. Results and discussion

The bubble size distribution in gas–liquid reactors influences


Fig. 5. Comparison of the simulated and experimental data for the air volume frac- gas holdup, residence time distribution, and the gas–liquid inter-
tion using different running times.
face area for mass transfer. The distribution of bubble sizes that
were photographically measured in the horizontal zone could
uate the optimal values. Thus, nine experiments involving different approximate the normal distribution for the combinations of gas
combinations of the factors should be conducted to study the main and liquid flow rates that were used. Increasing the gas flow
effects and interactions. Four parameters, the air inlet velocity (A), rate under a constant liquid flow rate can positively or nega-
the liquid inlet velocity (B), the bubble diameter (C), and the viscos- tively influence the median bubble size, depending on the value
ity (D), were selected because they were considered to be typically of the liquid flow rate used. The Sauter mean bubble size gener-
influential based on experimental experience. The control factors ally declines as the liquid flow rate increases. This trend occurs
with the three levels are presented in Table 3. The structure of the because at any fixed aeration rate, an increase in the liquid flow
Taguchi’s L9 design, including the selected parameters and their rate increases turbulence, which in turn, enhances the turbulence-
levels, is given in Table 4. associated bubble breakup. Furthermore, a forced flow of liquid
reduces bubble–bubble encounters that occur as a consequence
5.2. Confirmation test of the radial movement of bubbles, which reduced bubble coa-
lescence. In the present study, the forced circulation of liquid
The final step in the Taguchi optimization method is the con- prevented the formation of a heterogeneous flow regime over the
firmation test, which represents an important stage. To verify the entire range of operation. The distribution of bubble diameters in
conclusions drawn from the experiments and verify the applica- the forced-liquid loop bioreactor are shown in Fig. 3.
bility and feasibility of the optimal parameters, a confirmation test An examination of Figs. 4 and 5 revealed the manner in which
should be performed. Because the OA represents only a small frac- the air moved along the bioreactor. The temporal evaluation of the
tion of all the possibilities, the optimum run may not be necessarily average air volume fraction at four different times is shown in Fig. 4.
have been one of the many experiments that had been previously The gas phase holdup was obtained by averaging all the values in

Fig. 6. Effect of various parameters on the air volume fraction.


S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258 1255

Table 6
ANOVA analysis.

Factor Degrees of freedom (DOF) Sum of squares (S) Variance (V) F-Ratio (F) Pure sum (S’) Percent (%)

(A) Air inlet velocity (m/s) 2 0.001 0 – 0.001 21.765


(B) Liquid inlet velocity (m/s) 2 0.005 0.002 – 0.005 72.091
(C) Bubble diameter (mm) 2 0 0 – 0 4.814
(D) Viscosity (×10−3 Pa s) 2 0 0 – 0 0.053
Other/error 0

Total 8 0.007 100

the whole bioreactor. A comparison between the simulated and


experimental values of the air volume fraction at different running
times is shown in Fig. 5. The simulated and experimental results
that were obtained are in good agreement.
The average values of air volume fraction for all factors at each
level are plotted in Fig. 6. The highest value of performance for each
factor helped determine the best level. The optimum condition was
obtained using air inlet velocity level 3, liquid inlet velocity level 1,
bubble diameter level 1, and viscosity level 3.
Understanding the interaction between two factors resulted
in a better insight into the overall process. Given that any factor
may interact with any of the other factors, there is the possibil-
ity of generating a large number of interactions. The estimated
interaction, also known as the severity index (SI), of the differ-
ent factors under study can help determine the influence of two
individual factors at various levels. The SI was 100% for a 90◦
angle between the lines but was 0% for parallel lines (Table 5).
When the factors listed in Table 3 were analyzed, air inlet veloc-
ity and bubble diameter had the maximum interaction with an SI
of 61.62%. The other two interactions that produced a significant
SI were air inlet velocity–viscosity and liquid inlet velocity–bubble
diameter.
The ANOVA for these experiments are given in Table 6. The
degree of freedom (DOF) measures the amount of information that
can be uniquely determined from a given set of data, and the vari-
ance describes the distribution of the data around the data mean.
Because the simulation data is only representative of part of all pos-
sible data, DOF was used in the variance calculation instead of the
number of observations. The F-ratio, which is the ratio of the vari-
ance due to the effect of a factor and variance due to the error, is
used to measure the significance of the factor under investigation
with respect to the variance of all the factors included in the error
term. The pure sum of the squares is the sum of the squares minus
the DOF times the error variance. In this study, the DOF for each
factor was 2 and the total DOF was 8, resulting in an error term
with a DOF of 0. As a result, the variance for the error term (Ve ),
obtained by calculating the error sum of the squares divided by the
error degrees of freedom, could not be calculated and thus, the F-
ratio could not be calculated. To eliminate the zero DOF from the
error term, a pooled ANOVA was applied. Pooling is the process of
disregarding an individual factor’s contribution and subsequently
adjusting the contribution of the other factors. If the calculated F-
ratio for a parameter is less than the F-ratio tabulated at a stated
confidence level (95% confidence level), the effect of the parameter
is insignificant and the parameter is pooled [30]. If the parameter
is pooled, the sum of the squares due to the parameter is added to
the error sum of the squares, and other ANOVA terms are modified.
The values of F-ratio that were calculated after pooling for viscosity
are given in Table 7. From the calculated F-ratios, the parameters
and the interactions considered in the experimental design can be
inferred to be statistically significant at the 95% confidence level. Fig. 7. Comparison of the simulated and experimental data under each Taguchi run
The percentage contribution of each factor on the air volume condition for the following variables: (a) air volume fraction; (b) mixing time; (c)
fraction, which was calculated by the ratio of the variance for each residence time. The statistics for the measurement of air volume fraction, mixing
time and residence time are shown using 99%, 95% and 98% confidence error bars,
factor to the total variance, is shown in Table 6. The contribution
respectively.
of the liquid inlet velocity was the greatest (72.091%) with the
1256 S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258

Table 7
Pooled ANOVA analyses.

Factor Degrees of freedom (DOF) Sum of squares (S) Variance (V) F-Ratio (F) Pure sum (S’) Percent (%)

(A) Air inlet velocity (m/s) 2 0.001 0 403.361 0.001 24.323


(B) Liquid inlet velocity (m/s) 2 0.005 0.002 1336.014 0.005 80.704
(C) Bubble diameter (mm) 2 0 0 89.222 0 5.333
(D) Viscosity (×10−3 Pa s) (2) (0) – Pooled – –
Other/error 2 −0.001 −0.001 −10.36

Total 8 0.007 100

Fig. 8. Temporal evaluation of air volume fraction in the bioreactor for (a) Taguchi run #3 and (b) Taguchi run #7. (c) The optimum conditions proposed by the Taguchi
method.

air inlet velocity and bubble diameter being 21.765% and 4.814%,
respectively.
The confirmation test indicated that the gas hold up using the
new design experiments was 0.145, with a 95% confidence interval
of 0.143 ± 0.002. The mean calculated from confirmation test was
0.145, which is within the confidence interval. The 95% confidence
interval (CI) of the confirmation experiment was calculated using
the following equation:
 
1 1
CI = F˛ (1, fe )Ve + (21)
neff R

where Ve is the error variance, F˛ (1,fe ) is the F-ratio at the confi-


dence level of (1 − ˛) against a DOF between 1 and the error degree
of freedom, fe ˛ and
N
neff = (22)
1 + [total DOF associated in the estimate of mean]

where N is the total number of results and R is the sample size


used for the confirmation experiments. The experimental results
confirmed the validity of the applied technique for optimizing the
bioreactor performance. Thus, the gas hold up can be significantly
increased using the proposed statistical technique.
A comparison between the simulation and experimental results,
including gas hold up, mixing time and residence time in the biore-
actor, that were obtained by the L9 array is shown in Fig. 7. It is
clear that the results of simulated runs and experimental data are
in good agreement, and the simulation data can verify.
The temporal evaluations of the air volume fraction in the biore-
actor for Taguchi runs #3 and #7 are shown in Fig. 8(a) and (b),
Fig. 9. Temporal evaluation of the methane mass fraction in the bioreactor after (a)
respectively. Runs #3 and #7 represent the most insignificant and 0.5 s and (b) 1.0 s.
significant combinations, respectively, of the factors that affect
S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258 1257

Fig. 10. Temporal evaluation of the biomass mass fraction in the bioreactor after (a) 0.5 s, (b) 1.0 s, (c) 2.0 s and (d) 2.0 s. Note that part (c) used Taguchi run #7, whereas part
(d) used the optimum conditions proposed by the Taguchi method.

gas hold up. The air volume fraction under the optimum con- The biomass production from natural gas is an aerobic process, and
ditions proposed by the Taguchi method is shown in Fig. 8(c). because methane was the only carbon source, the concentration of
Under the optimum conditions, gas hold up reached its maximum methane decreased as the microorganisms grew (Fig. 9(a) and (b)).
value. In addition, a temporal evaluation of the biomass mass frac-
To investigate the performance of the forced-liquid HTLB with tion in the forced-liquid HTLB confirmed that biomass production
respect to biomass production as well as the hydrodynamic and increased the most at those points at which methane was injected
mass transfer parameters, five experiments were designed. In these (Fig. 10). A comparison of the data presented in Figs. 9 and 10
experiments, the gas and liquid velocities were set at the values revealed that methane concentration decreased as the biomass
determined under the optimum condition, and the gas stream was concentration increased (occurring mainly at 1.0 s in both figures).
variable, with different mixtures of oxygen and methane. A gas Finally, the highest biomass production occurred under the optimal
mixture of 50 vol% methane and 50 vol% oxygen resulted in the conditions suggested by the Taguchi method (Fig. 10(d)).
best biomass production (maximum optical density: 1.2; dry cell
weight; 2.45 g/l and doubling time: 104 min) [31]. The stoichiomet- 7. Conclusion
ric reaction for methane fermentation by Methylomonas is given in
the following equation: This study successfully employed a commercial package of FLU-
ENT code to solve the three-dimensional flow in a forced-liquid
2.9CH4 + 3.85O2 +0.64NH3 → C1.9 H4.6 N0.64 O1.25
HTLB and an integrated Taguchi OA to optimize the gas hold up
+ CO2 + 4.46H2 O (23) for the production of biomass. An L9 orthogonal array was imple-
mented to optimize the factors that affect the flow hydrodynamics
The combination of the hydrodynamics and the chemical species in the loop bioreactor, and air inlet velocity, liquid inlet velocity,
transport equation simulated the biomass production from natu- bubble diameter, and viscosity were chosen as the main param-
ral gas. A temporal evaluation of the methane mass fraction in the eters. By combining the hydrodynamics and the chemical species
bioreactor under optimum conditions during biomass production transport equation, the biomass production from natural gas was
after 0.5 s is illustrated in Fig. 9(a), and a sequential assessment of simulated in the HTLB. The results demonstrate the ability of CFD
the methane mass fraction in the HTLB after 1 s is shown in Fig. 9(b). to provide new insights on the biological phenomena that occur
in the gas–liquid reactors. The simulation results confirm that a
Eulerian formulation is a successful approach to predict the hydro-
Table 8 dynamics of the bioreactor because it provides good engineering
Optimum conditions and performance. descriptions and can be used reliably to predict the flow and holdup
patterns in such systems. Based on the optimization, the liquid inlet
Factor Level description Level Contribution
velocity and viscosity were the parameters that had the most sig-
(A) Air inlet velocity (m/s) 31.85 3 0.017 nificant and insignificant effects on flow behavior in the system,
(B) Liquid inlet velocity (m/s) 0.24 1 0.035
respectively. Under the optimal conditions (Table 8), the simulated
(C) Bubble diameter (mm) 6 1 0.007
(D) Viscosity (×10−3 Pa s) 1.301 3 0 and experimental data were in good agreement with the predicted
data analyzed by the Taguchi robust design method. Our data indi-
Contribution from all factors (total) 0.59 cated that an increase in biomass production occurred under the
Current grand average of performance or mean 0.84 optimal conditions suggested by the Taguchi design. Thus, the
Expected result at optimum condition 0.143
Taguchi optimization technique is a powerful tool to solve indus-
1258 S.M. Mousavi et al. / Chemical Engineering and Processing 49 (2010) 1249–1258

trial problems that can be easily implemented for many industrial References
applications.
[1] B.T. Sheehan, M.J. Johnson, Production of bacterial cells from methane, Appl.
Microbiol. 21 (1971) 511–515.
Appendix A. Nomenclature [2] B. Volesky, J.E. Zajic, Batch production of protein from ethane and
ethane–methane mixtures, Appl. Microbiol. 21 (1971) 612–614.
[3] L. Jorgensen, Method and means for the production of a microorganism cell
mass, EP Pat. 0306466A2, 1989.
Notations [4] H. Blenke, Loop reactors, Adv. Biochem. Eng. 13 (1979) 121–214.
[5] H. Eriksen, Method of fermentation, WO Pat. 016460A1, 2003.
C1ε constant [6] F. Wang, Z. Mao, Y. Wang, C. Yang, Measurement of phase holdups in
C2ε constant liquid–liquid–solid three phase stirred tanks and CFD simulation, Chem. Eng.
CD drag coefficient Sci. 61 (2006) 7535–7550.
[7] T. Zhang, J. Wang, Z. Luo, Y. Jin, Multiphase flow characteristics of a novel
CI confidence interval
internal-loop airlift reactor, Chem. Eng. J. 109 (2005) 115–122.
C constant [8] C. Freitas, M. Fialova, J. Zahradnik, J.A. Teixeira, Hydrodynamic of a three-phase
Di,m diffusion coefficient for species i in the mixture external-loop airlift bioreactor, Chem. Eng. Sci. 55 (2000) 4961–4972.
d32 sauter mean bubble diameter (mm) [9] M. Blazej, M. Kisa, J. Markos, Scale influence on the hydrodynamics of an internal
loop airlift reactor, Chem. Eng. Process. 43 (2004) 1519–1527.
dp diameter of the bubbles of phase p (mm) [10] M.H. Al-Rashed, A.G. Jones, CFD modeling of gas–liquid reactive precipitation,
f drag function Chem. Eng. Sci. 54 (1999) 4779–4784.
F˛ F-ratio at a confidence level of (1 − ˛) against DOF [11] R. Andersson, B. Andersson, F. Chopard, T. Noren, Development of a multi-scale
simulation method for design of novel multi phase reactors, Chem. Eng. Sci. 59
Fq external body force (N) (2004) 4911–4917.
g gravity acceleration (m/s2 ) [12] B. Farizoglu, B. Keskinler, Influence of draft tube cross-sectional geometry on
Ji diffusion flux of species i KL a and ␧ in jet loop bioreactors (JLB), Chem. Eng. J. 133 (2007) 293–299.
[13] S.M. Mousavi, A. Jafari, S. Yaghmaei, M. Vossoughi, I. Turunen, Experiments and
Gk,m production of turbulence kinetic energy CFD simulation of ferrous biooxidation in a bubble column bioreactor, Comp.
k turbulent kinetic energy (m2 /s2 ) Chem. Eng., in press.
kpq (kqp ) interphase momentum exchange coefficient (kg/s) [14] C. Gentric, D. Mignon, J. Bousque, P.A. Tanguy, Comparison of mixing in two
industrial gas–liquid reactors using CFD simulations, Chem. Eng. Sci. 60 (2005)
ni number of bubbles with diameter di 2253–2272.
N total number of results [15] L. Raynal, I. Harter, Studies of gas–liquid flow through reactors internals using
P pressure (Pa) VOF simulations, Chem. Eng. Sci. 56 (2001) 6385–6391.
[16] G.L. Lane, M.P. Schwarz, G.M. Evans, Predicting gas–liquid flow in a mechani-
R sample size for confirmation experiments
cally stirred tank, Appl. Math. Model. 26 (2002) 223–235.
Re relative Reynolds number [17] E. Olmos, C. Gentric, C. Vial, G. Wild, N. Midoux, Numerical simulation of mul-
Ri net rate of production of species i by chemical reaction tiphase flow in bubble column, influence of bubble coalescence and break-up,
R pq interaction force between phases (N) Chem. Eng. Sci. 56 (2001) 6359–6365.
[18] S. Ghorai, K.D.P. Nigam, CFD modeling of flow profiles and interfacial phenom-
Si rate of creation by addition from the dispersed phase plus ena in two-phase flow in pipes, Chem. Eng. Process. 45 (2006) 55–65.
any user-defined sources [19] P.J. Ross, Taguchi Techniques for Quality Engineering, 5th ed., McGraw-Hill,
tc circulation time (s) New York, 1996.
[20] N. Belavendran, Quality by Design, 1st ed., Prentice-Hall International, London,
tm mixing time (s) 1995.
tr residence time (s) [21] F. Yazdian, S. Hajizadeh, S.A. Shojaosadati, R. Khalilzadeh, M. Jahanshahi, M.

→ vm mixture velocity (m/s) Nosrati, Production of single cell protein from natural gas: parameter opti-
mization and RNA evaluation, Iranian J. Biotech. 3 (2005) 235–242.
vq velocity of phase q (m/s) [22] A. Fadavi, Y. Chisti, Gas holdup and mixing characteristic of a novel forced
Ve error of variance circulation loop reactor, Chem. Eng. J. 131 (2007) 105–111.
Vg inlet air volume (l) [23] M. Papagianni, M. Matty, B. Kristiansen, Citric acid production and morphology
· of Aspergillus niger as functions of the mixing intensity in a stirred tank and a
Vg air flow rate (l/s) tubular loop bioreactor, J. Biochem. Eng. 2 (1998) 197–205.
Yi local mass fraction of species i [24] P. Wongsuchoto, T. Charinpanitkul, P. Pavasant, Bubble size distribution and
gas–liquid mass transfer in airlift contactors, Chem. Eng. J. 92 (2003) 81–90.
[25] R. Clift, J.R. Grace, M.E. Weber, Bubbles Drops and Particles, Academic Press,
Greek letters New York, 1978.
[26] T.R. Auton, The dynamics of bubbles, drops and particles in motion in liquids,
˛q volume fraction of phase q Ph.D. Thesis, University of Cambridge, Cambridge, U.K., 1983.
ε turbulent dissipation rate (m2 /s3 ) [27] E. Delnoij, F.A. Lammers, J.A.M. Kuipers, W.P.M. van Swaaij, Dynamic simulation
q bulk viscosity of phase q (Pa s) of dispersed gas–liquid two-phase flow using a discrete bubble model, Chem.
Eng. Sci. 52 (9) (1997) 1429–1458.
t,m turbulent viscosity (Pa s)
[28] Fluent 6.2 Users Guide, Fluent, Inc., Lebanon, 2005.
q shear viscosity of phase q (Pa s) [29] A. Sokolichin, G. Eigenberger, Applicability of the standard turbulence model
m mixture density (kg/m3 ) to the dynamic simulation of bubble columns. Part I. Detailed numerical simu-
q physical density of phase q (kg/m3 ) lations, Chem. Eng. Sci. 54 (1999) 2273–2284.
[30] R.K. Roy, A Primer on Taguchi Method, Van Nostrand Reinhold, New York, 1990.
 ˆq effective density of phase q (kg/m3 ) [31] F. Yazdian, S.A. Shojaosadati, M. Nosrati, M.R. Mehrnia, E. Vasheghani-Farahani,
k constant Study of geometry and operational conditions on mixing time, gas hold up, mass
ε constant transfer, flow regime and biomass production from natural gas in a horizontal
tubular loop bioreactor, Chem. Eng. Sci. 64 (2009) 540–547.
p particulate relaxation time (s)
¯¯ p q phase stress–strain tensor (Pa)

Vous aimerez peut-être aussi