Accepted Manuscript

Modeling of hydrodynamics and mixing in a submerged membrane bioreactor

Zaineb Trad, Christophe Vial, Jean-Pierre Fontaine, Christian Larroche
PII:
DOI:
Reference:

S1385-8947(15)00609-9
http://dx.doi.org/10.1016/j.cej.2015.04.119
CEJ 13608

To appear in:

Chemical Engineering Journal

Please cite this article as: Z. Trad, C. Vial, J-P. Fontaine, C. Larroche, Modeling of hydrodynamics and mixing in
a submerged membrane bioreactor, Chemical Engineering Journal (2015), doi: http://dx.doi.org/10.1016/j.cej.
2015.04.119

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Modeling of hydrodynamics and mixing in a submerged membrane

bioreactor
Zaineb TRADa,b,c, Christophe VIALb,c*, Jean-Pierre FONTAINEb,c, Christian LARROCHEb,c
a

Clermont Universit, Universit Blaise Pascal, LABEX IMobS3, BP 10448, F-63000

F-63171 Clermont-Ferrand, France
b
Clermont Universit, Universit Blaise Pascal, Institut Pascal, BP 20206,
F-63174 Aubire cedex, France
c
CNRS, UMR 6602, IP, F-63171 Aubire, France
Corresponding author: Zaineb.TRAD@univ-bpclermont.fr
Tel: +33(0)473405266 - Fax: +33(0)473407829

ABSTRACT
This work deals with the optimization of mixing in an anaerobic submerged membrane bioreactor (AnMBR)
devoted to the production of biohydrogen as a 2nd generation biofuel using the dark fermentation process from
lignocellulosic waste. The AnMBR consists of an unbaffled mechanically-stirred tank equipped with a two-stage
radial impeller which is coupled to an external hollow fibre membrane module placed in a forced circulation
loop with permeate suction. In the tank, mixing conditions must enable the suspension of solid waste,
homogenize local concentrations, and enhance hydrogen desorption simultaneously. For optimizing reactor
design, mixing was investigated both experimentally and numerically. The computational strategy consisted in
the combination of 1D and 3D methodologies including single-phase and two-fluid CFD models. The
experimental approach encompassed RTD and mixing time measurements, the analysis of vortex formation and
the description of solid suspension using straw as waste. The results showed that a laminar flow pattern prevailed
in the recirculation loop and the membrane unit, while turbulent flow was observed in the stirred tank in which
mixing time was far smaller than residence time in single-phase flow. A compromise was defined using CFD,
which prevented vortex formation and promoted homogeneous suspension, by optimizing impeller position and
rotation speed. The positions of the inlet and the outlet of the forced circulation loop in the tank were deduced
from CFD simulations and validated by experimental data. Finally, this work confirms the potential interest of
AnMBR for biohydrogen production and the applicability of CFD for the optimization of abiotic factors.
Keywords: anaerobic membrane bioreactor, biohydrogen, CFD, hydrodynamics.

1. Introduction
Biohydrogen (BioH2) is a 2nd and a 3rd generation biofuel. It constitutes undoubtedly a promising
renewable and sustainable alternative to fossil fuels and combines the advantages to face with the issue
of mobility of the future and to reduce greenhouse gas emissions [1]. Second generation BioH2 can be
obtained through anaerobic digestion of agricultural and food waste in the liquid phase, provided
methanogenesis is prevented. This technology has been the subject of intensive research in the recent
years, as this process can use solid waste as a raw material to produce BioH2 using either pure [2] or
mixed cultures [3]. Mixed cultures present the advantage to avoid aseptic conditions, but with a lower
yield. However, the production of BioH2 is necessarily coupled with that of volatile fatty acids (VFA),
while soluble hydrogen, H2 partial pressure and VFA are all able to inhibit BioH2 production; this
means that VFA must be removed continuously. As a result, BioH2 production relies not only on
1

complex biological mechanisms, but also on many physical processes because it involves a multiphase
flow with suspended solid waste and biogas (CO2 and H2) bubbles in a continuous liquid phase. This
implies that liquid-solid and gas-liquid mass transfer with gas desorption kinetics may compete with
those of biological processes. Up to now, although numerous research breakthroughs have been
reported on the efficiency of BioH2 production by evaluating the chemical and the biological factors
[3-5], the abiotic factors have received only minor consideration, despite their key influence on the
economic competitiveness of BioH2 production and downstream purification technology [6].
In practice, bioreactor design constitutes a key issue of BioH2 production by dark fermentation [7].
Recently, various designs have been compared, such as anaerobic sequencing batch and fluidized bed
reactors, or continuously stirred bioreactors. The applicability of membrane bioreactors has also been
analyzed in several review papers for methane and BioH2 production through anaerobic fermentation
[7-9]. The advantage of anaerobic membrane bioreactors (AnMBRs) is to achieve a very high solid
retention time, while the hydraulic retention time can be controlled independently. AnMBRs are
divided in two main classes. In the first one, the membrane is located inside the bioreactor and a
negative pressure serves to suck permeate through the submerged membrane. In the second one, the
membrane is placed in an external loop connected to the bioreactor in which the driving force for
permeate recovery is obtained from the recirculation pump. The first internal design allows to
minimize power input, but requires a larger surface area to maintain a high permeate flux, whereas an
external membrane module is easier to implement, to clean and to replace, as fouling remains the key
issue of membrane technology. As a result, hollow fibre modules of polymeric membranes are usually
implemented with stirred tanks because this design minimizes fouling and enhances H2 desorption, so
as to avoid inhibiting the catalytic activity of hydrogenase enzymes by dissolved H2 [10]. However,
despite the recent advances, BioH2 production in AnMBRs is still disregarded in comparison to
methane production [7].
For promoting the development of AnMBRs devoted to BioH2 production, mixing appears as a key
parameter. First, it generates a spatial maldistribution of the suspended solid phase and the
inhomogeneity of the concentrations in the liquid phase, which affects the biochemical reaction rates.
Similarly, it enhances gas-liquid mass transfer and prevents the inhibition of anaerobic digestion by
soluble BioH2. Local hydrodynamics close to the membrane may also reduce fouling, leading to
higher productivity and preventing the inhibition of BioH2 production by soluble VFA. All these
processes require highly turbulent flow conditions, but a compromise is necessary, first to limit power
input, but mainly because the biological mechanisms of anaerobic fermentation usually require gentle
mixing conditions. Even though mixing has been studied for a long time in the process industries as a
function of reactor geometry, feeding strategy or fluid properties [11-14], AnMBRs combine
biochemical reactions and filtration in three-phase flow, which means that reliable data on mixing and
hydrodynamics are difficult to obtain. Computational Fluid Dynamics (CFD) could, therefore, be
2

helpful to better elucidate the complex interplay between the reactor geometry and the operating
conditions of bioreactors. Since the 2000s, CFD has become a powerful and versatile tool for
simulating turbulent non-reactive and reactive turbulent multiphase flows in mechanically stirred tank
[15], as in bioreactors [16]. However, the applications to BioH2 production remain scarce [17] and also
for AnMBR devoted to methanation process [18], as it implies CPU- and memory-intensive
computations for 3D flows, especially for rotating impellers in multiphase flows.
Finally, the objective of the present work is to assess the applicability of an innovative AnMBR
design for BioH2 production by analysing and enhancing the effect of some abiotic factors, in
particular mixing conditions. Only non-reacting flow conditions will be considered. This AnMBR has
been designed atypically to combine the advantages of internal and external membrane bioreactors:
The bioreactor consists of an unbaffled mechanically-stirred tank equipped with a two-stage
impeller, while the membrane is placed externally on a forced circulation loop. The role of the
bottom impeller is to achieve solid suspension, while that of the top impeller is to maintain flow
homogeneity and enhance BioH2 desorption.
Even though it is equipped with an external membrane module, the AnMBR is driven as a
submerged membrane bioreactor, as the driving force for permeate recovery comes from the
suction exerted by an external withdrawal pump.
A joint experimental/numerical simulation strategy will be applied, in which experiments (such as
mixing time measurements and Residence Time Distribution, RTD) will lead to 1D modelling and
CFD to 3D predictions with a special focus on the numerical issues. In detail, the objective is to
experimentally analyze the mixing performance of the AnMBR, including distributive mixing
(concentration homogenization) and dispersive mixing (solid dispersion and vortex formation), to
assess that CFD is able to predict experimental data, and finally to optimize the bioreactor design and
operating conditions, including impeller design and the positions of the inlet and the outlet of the
recirculation loop.
2. Materials and Methods
2.1. Reactor design and operating conditions
Experiments were carried out in a laboratory-scale AnMBR (Figure 1). This consists of a 5-L
mechanically-stirred tank associated with an external hollow fibre microfiltration (MF) membrane
module that operates in the tangential outside-in crossflow mode. This module is placed on an
external recirculation loop in which the flow rate is imposed using a peristaltic circulation pump (up to
400 mL/min). The minimum residence time in the tank is 12.5 min. The permeate is recovered using a
withdrawal peristaltic pump that sucks the liquid phase, causing a negative gauge pressure inside the
fibres. In the MF module, the driving force of tangential filtration is the transmembrane pressure that
remains small in the present case, as in submerged membrane bioreactors. As a result, the tank and the
circulation loop can be operated under atmospheric pressure, which avoids inhibition by the partial
3

pressure of BioH2. The variable-flow permeate pump has a maximum flow rate far lower than the
circulation pump, about 15 mL/min, and is reversible, so as to enable backwashing of the membrane
module. This is equipped with 142 free polymeric fibres in PVDF with a 0.2 m cut-off diameter
placed in a 32.5 cm length housing. This housing is cylindrical and the module packing configuration
corresponds to a fibre bundle in which CO2 gas injection at the bottom can also be used to prevent
membrane fouling. The objective is to operate in the fed-batch mode (cyclic removal of permeate
compensated by the addition of an equivalent volume of culture medium). As a result, the membrane
will maintain the solid substrate and the microorganisms in the AnMBR, while soluble molecular
compounds, in particular VFA, will be recovered with the permeate.
The stirred tank consists of a round-bottom vessel stirred using a two-stage impeller placed in the
center of the tank (see Figure 2 and Table 1). An unbaffled tank was preferred to avoid the presence of
stationary solids close to baffles, with the drawback of vortex formation at high rotation speed. The
internal diameter of the vessel, D, is equal to 170 mm for a liquid height of HL of 270 mm. The bottom
impeller is a four-blade disk Rushton turbine (i.e. a radial impeller) placed at a distance h from the
bottom of the tank. The second impeller is a three-blade, 45 pitched blade turbine placed at a distance
C above the Rushton turbine. C can be varied between 90 and 110 mm, while the impeller clearance
off the tank bottom ranges between 46 and 69 mm in order to enhance the suspension of the solid
substrate. Contrary to the Rushton turbine, the pitched blade turbine produces a flow discharge both
axially and radially, nearly in equal proportion due to the 45 angle. To avoid the inhibition of BioH2
production, the rotation speed of the impeller remains low, but high enough to ensure that the solid
phase is fully suspended. In this work, the rotation speed has been varied between 30 and 500 rpm.
The long-term objective is to carry out anaerobic cultures using a mesophilic microbial consortium
at 35C and under controlled pH conditions to prevent pH decrease due to the production of VFA with
lignocellulosic waste as the solid substrate, but preliminary tests have been driven using straw. As
straw is present in the form of large tabular particles of about 2 mm length, it does not really modify
the viscosity of the fluid. As a result, the fluid remained Newtonian, with a viscosity between 1.3 and
2.0 mPa.s at 20C, as measured using an AR-G2 rheometer (TA Instruments, USA) equipped with a
double-Couette geometry.
2.2. Analysis of macromixing
Macromixing was studied in the stirred and the recirculation loop using a conductivity tracer
technique based on pulse injection. A conductivity meter (CDM210, Radiometer Analytical, France)
with a separate probe was used and the linearized signal of the meter was recorded using an ADLINK
usb-1901 data device at a frequency of 1 Hz. Mixing was analyzed using experiments in the different
sections of the AnMBR, i.e. both on the stirred tank and the membrane module. Due to their difference
in volume, 5 L for the stirred tank and 0.5 L for the membrane module, the residence time in the latter
was 10 times lower. As a result, only mixing and circulation times tm and tc could be measured in the
4

stirred tank that was treated as a batch system, as the residence time in the tank was expected to be far
higher than tm; this will be verified experimentally. Conversely, RTD was measured on the membrane
module with and without permeate extraction. In the tank, the mixing time tm is defined as the time
period to achieve the desired level of homogeneity which can be expressed in terms of the degree of
mixing Y:
 = 

 

 

(1)

where   is the tracer concentration at time t, while  and  are the initial and final average

concentration, respectively. The desired level of mixing was Y=0.95 and the pulse injection of a very
small volume (10 mL) of a salt solution (usually 1.0 M NaCl) was used. Three configurations were
tested (Figure 3).
In the present paper, mixing experiments have been used to assess CFD simulations, as the
conventional flow visualization techniques (such as laser Doppler, particle image or hot-film
velocimetry) are unable to deal with the solid suspensions encountered in anaerobic bioreactors.
However, it is also worthy of note that mixing properties not only describe the suspension of organic
waste and the degree of homogeneity of local concentrations, but also enhance hydrogen desorption or
may disturb the biological processes. In addition, solid suspension was also investigated using flow
visualization as a function of rotation speed, of the amount of straw (from 5 to 10 g/L) and of impeller
position (Table 1) using two black and white cameras (USB 3.0, 4 Mpx, 60 fps, from IDS GmbH,
Germany) in the shadowgraphy mode. Quantitative data on the segregation of straw in the tank could
be extracted and compared to CFD simulations. Combined to data on distributive mixing, these may
provide useful information to set the position of the inlet of the membrane module in the tank, first for
preventing the aspiration of suspended solids in the loop, but also if possible, where VFA
concentrations are maximal to avoid local inhibition of biological processes. The same cameras were
also used in this work to analyze vortex formation (gas/liquid interface shape). As an unbaffled tank
was preferred to avoid solid segregation, the vortex could modify the hydrodynamics of the tank and
favor the recirculation of BioH2 bubbles, which should be avoided.
2.3. Modeling tools
The computational strategy consists of the combination of 1D and 3D CFD methodologies [16],
using a comparison between experimental data on RTD and mixing time based on a conductivity
tracer technique with the simulations. For 1D modeling, the RTD free software package from CISP
St Petersburg Ltd. (Russia) was used to describe the RTD response curves for the membrane module.
For 3D modeling, a commercial CFD software package was used, Phoenics, from CHAM Ltd. (UK),
because flow in unbaffled stirred tanks has been shown to be highly complex [19]. This solves the
continuity, the Navier-Stokes for a turbulence model equation by a finite-volume approach. The

Reynolds-averaged form of the continuity and the Navier-Stokes equations for a turbulent flow
 =0

(RANS) can be written as follows for a single phase incompressible fluid:





 +  =  +  +  +     

(2)
(3)

where  is the mean Reynolds-averaged velocity vector,  is the fluid density,  is the pressure, g is
the acceleration of gravity and  stands for turbulent velocity fluctuations. The viscous stress tensor is
 =   +  

related to the mean velocity gradients determined by the following equation:

where  is the viscosity of the fluid. Additional equations are required for model closure, in particular
(4)

to estimate the turbulent stress tensor    . In this work, the k- model is used with the transport
equation for k and






expressed as:

! + (! = "

#$

%&
#$

!' + ( - 

  + (  = " ' + *+ ( - -+ 

%)

+.
,

(5)
(6)

in which G is the production rate of turbulent kinetic energy, /, , /+ , *+ and -+ are the constants of
the k- model, and   is the turbulent viscosity defined as:
  =0

,.
+

(7)

Many variants of the k- model are available in the literature, starting from the standard k- model,

among which the RNG k- model is the most common. In practice, the various k- models can be
divided in two groups: the conventional (or high-Re) models and the low-Re models that better
account for the near-wall and low-velocity regions, but are computationally more intensive. In this
study, the standard k- model has been compared to the RNG k- model that corrects Equation (7) by
model that introduces an additional time scale !( in Equation (6) [20]. For each of them, the low-Re

reducing the overestimation of turbulence production rate in small vortices, and to the Chen-Kim

and high-Re models have also been compared. For mixing, closure also requires a tracer transport
equation as:
2


+ 3  = 4 3  + 53

(8)

where 3 is the concentration of the tracer or any (scalar) compound x, 4 is the turbulent diffusion
coefficient and 53 is a source term that is equal to 0 when no chemical reaction or interfacial mass

transfer involving x have to be considered. 4 is usually deduced from   using the assumption that the

turbulent Schmidt number   4  is close to 0.7. In single-phase flows, Equations (2), (3), (5), (6)
and (8) have been solved on a 6-core 12-thread PC computer using the parallel Phoenics solver,
together with the following boundary conditions:
no-slip at the walls of the tank;
the gas liquid-interface was defined as a solid wall with slip allowed, which will be validated using
VOF simulations below a rotation speed of 200 rpm.
6

In the literature, several CFD strategies have been developed for modeling rotating flows driven by
impellers. The most common way is the Multiple Reference Frame (MRF) approach, which is a steady
approximation of rotating flows [21]. Stationary zones are defined far from the impellers and moving
zones around them in which a rotating velocity is imposed. In fact, neither the grid nor the impeller
move, but a relative velocity formulation is superimposed on the rotation speed. The first alternative
consists in using a sliding mesh in the moving zones that follows the impeller. This implies transient
simulations that are far more CPU-intensive than MRF steady simulations. However, both MRF and
sliding mesh approaches usually require unstructured grids in the moving zones due to the complex
geometry of the impeller. Unlike these two common methods, Phoenics proposes a third way,
denoted MOFOR (Moving Frame of Reference) and defined as follows [20]:
A structured Cartesian grid is defined using the cut-cell technique in which the cells crossed by
solid wall receive a particular numerical treatment;
The impeller moves in the stationary structured grid without affecting the grid, which requires
transient simulations.
This approach has been used in this work. It presents the advantage to avoid complex unstructured
grids, but transient simulations are needed. The CFD geometry and a typical example of the structured
grids used in the simulations (between 300,000 and 800,000 cells to investigate the effect of grid size)
are presented in Figure 4. The influence of the discretization scheme of the advective term has also
been analyzed (from the first-order upwind to the second/third-order MinMOD and MUSCL
schemes), as it may play a more important role on the simulations in the cut-cell technique. The time
step was varied between 0.01 s and 0.001 s, depending on the rotation speed; relative residuals lower
than 10-3 were required at each time step on all variables. However, these single-phase simulations are
not able to detect the onset of vortex formation and to predict the dispersive mixing of straw. This is
the reason why multiphase CFD has also been applied, using the VOF (Volume of Fluid) method and
the Algebraic Slip Model (ASM). These are detailed below.
2.3.1. VOF method
The VOF method can describe the location of two immiscible fluids (or phases) by solving a single
set of momentum equations and tracking the volume fraction of each fluid for segregated fluid flows,
such as gas-liquid flows with a free interface. So, it is able to capture the topology of the free-surface
in the unbaffled stirred tank. The VOF method coupled to the k- turbulence model with a hybrid
linear scheme implemented in the Phoenics CFD software has been chosen in this work. This was
used to estimate the influence of the rotation speed and of the position of the impeller on the
deformation of the interface and to analyze its effect on the hydrodynamics of the liquid phase
compared to a rigid flat interface.
phase (i.e. the gas phase 67 ), but a single-fluid approach is used for velocity  because the two phases

Numerically, the VOF method has to simulate the transport of the volume fraction of the secondary

coexist only at the interface. As a result, the momentum equations are similar to those of single-phase
68 + 67 = 1  = 68 8 + 67 7

 = 6 8 8 + 6 7  7

flows (Equation 3), but with average material properties:

(9)

Similarly, the continuity equation (Equation 2) is solved together with the transport equation of the
volume fraction:
:;


+ 67  = 0

(10)

In this work, simulations using the VOF method have been conducted without the transport of tracer
and compared with visualization experiments of vortex formation. The same geometry and grid
described above have been used, but with a gas phase above the liquid phase and a no-slip boundary
wall condition at the top of the tank. The initial height of the liquid phase was always HL=27 cm, with
H-HL=9 cm of gas above (Table 1). The results obtained on single-phase simulations on the effect of
turbulence model and discretization scheme were applied to reduce the number of simulations using
the VOF method. However, the time step had to be reduced in comparison to single-phase simulations
to below 10-3 s.

2.3.2. Algebraic Slip Model (ASM)

The modeling and the simulation of the dispersive mixing of a solid phase in a mechanically stirred
tank under turbulent flow is a challenge in its own because it is always computationally intensive and
this is particularly true when the MOFOR and the cut-cell approaches are used. For this reason, ASM
has been selected, which represents an acceptable compromise between physical modeling and
computational requirements. The algebraic slip approximation assumes that the liquid-solid dispersion
is a mixture in which the phases move at different velocities, but with a local mechanical equilibrium
between the phases, which means that the relative velocity between the phases can be expressed
without solving the momentum equation of the dispersed phase. Contrary to the VOF method, the
phases can be interpenetrating. As a consequence, a mixture velocity must be defined in addition to the
 = 6= = =

mixture properties similar to those of Equation (9) between straw and the liquid phase, as follows:
Turbulence is shared by the two phases. The slip velocity > between the phases is deduced from the
(11)

balance between body forces ?@,@ (i.e. buoyancy, centrifugal and Coriolis forces) and drag (i.e.
friction forces). Finally, one gets:
|> |- = " ' F
D
C

GH G
GH

I JK ?@,@

(12)

in which JK and K are the diameter and the density of the spherical solid particles, respectively. In
Equation (12), L is the drag coefficient that is calculated using a correlation similar to that of Schiller
and Nauman:

L = M

24

, YZ [\] 10C
`
0.44, YZ [\] > 10C

*P.*RST
.UVW
STX

L is a function of the relative Reynolds number [\] , which is defined as

[\] =

Gab cH
#P#$

(13)

(14)

under turbulent flow conditions.

In this work, simulations using ASM have been compared with visualization experiments of straw
suspension vs. rotation speed and impeller position. The volume-average diameter of the particles
based was estimated about 0.7 mm using image analysis. As the density of straw was difficult to
estimate (dry straw flotates, but settles after a few hours in water), wet straw was treated as a pseudophase. The initial condition is that this pseudo-phase occupies the round-bottom region of the tank
when straw content is 10 g/L and the solid-liquid mixture exhibits a density close to water in this
region (about 1010 kg/m3). The geometry, grid and boundary conditions described above for singlephase simulations have been used. As for the VOF method, the time step ranged between 10-4 and 10-3
s. These results obtained with single-phase simulations on the effect of turbulence model and
discretization scheme were applied.
3. Results and discussion
3.1. Single-phase mixing experiments
Conductivity tracer experiments will be discussed first for clearances h=69 mm and C=110 mm,
respectively. Figure 5a illustrates the evolution of the normalized salt concentration for a salt injection
close to the free surface using the configuration of Figure 3.1. In the figure, rotation speed was 30 rpm
and tm was about 1288 s. tm was not affected by the flow rate in the recirculation loop because tm was

far lower than the minimum space time  of the fluid in the tank (Table 3). Figure 5a shows also that tc
was lower than 20 s for 30 rpm and it was smaller than 10 s for higher agitation speed (Table 3).

Above 100 rpm, tc was too small to be accurately measured. Roughly, the dimensionless mixing time
(tm multiplied by rotation speed) was between 60-70 and tc was about tm/10. Even though the Reynolds
number varied between 1,000 and 20,000 for both impellers (i.e. from the transitional to the fully
developed turbulent flow conditions), all the response curves presented a similar profile. This is in
agreement with the gradual transition between the flow regimes, usually observed in stirred tanks. As
a conclusion, the stirred tank of the AnMBR appears to be perfectly mixed in less than 100 s above 50
rpm and in less than 30 s above 100 rpm. Above 150 rpm, tm is also lower than the smallest space time
of the fluid in the MF module () in Table 3, which implies that the position of both the inlet and the
outlet of the recirculation loop play a weaker role on the respective composition of the fluid in the loop
and the tank. Finally, these results were also only slightly influenced by h and C in the range defined
in Table 1.

In the membrane module, RTD curves E() were obtained as a preliminary result. It appeared that
when the when the module was connected to the tank, tracer experiments using the configuration
described in Figure 3.2 were very close to E(). Figure 5b presents a typical response curve vs.
normalized time =t/ obtained without the extraction of permeate. This curve exhibits an atypical
shape, with a peak that emerges at half space time in the module, i.e. about =0.5 (Table 3), followed
by a long tail. Actually, the curves always presented the same shape that nearly fitted the theoretical
curve E() under laminar flow conditions for an empty pipe in Figure 5b. Moreover Re, defined as in a
pipe, was lower than 1000 at the highest flow rate in the loop. Even though the height of the
experimental peak is smaller than expected theoretically, it must be reminded that the shape of the
peak is always affected by the non-ideality of the injection and by the axial dispersion at the inlet. In
addition, the presence of the fibres also affects the local flow in the module. Conversely, the
experimental data highlighted that the effect of rotation speed in the stirred tank was negligible on by
the peak time, as shown in Table 3, and also on the shape of the response curves. When the permeate
was sucked at maximum flow rate of the permeate pump, as in Figure 3.3, the response curves were
always close to the E() curves obtained without permeate recovery, which shows that the permeate
flow rate was low in comparison to the recirculation flow rate (Table 3). For validating these results, a
simple 1D model was established, first combining a laminar flow reactor with a perfectly mixed tank
in the loop. The simulation for a tracer injection with the configuration of Figure 3.2 is reported in
Figure 5b and shows that only the end of the tail, when >1 is slightly affected. If a non-ideal tank is
assumed, described for example the loop by a plug flow reactor followed by two tanks-in-series,
similar results are obtained in Figure 5b. Finally, this confirms that the MF module can always be
modelled as a simple pipe under laminar flow conditions.
As a conclusion, mixing in the AnMBR can be described roughly by a perfectly mixed tank with a
space time and a dimensionless mixing time about 60-70, connected to a recirculation loop that could
be assimilated to a tubular laminar flow region of space time /10 when rotation speed was higher than
150 rpm. Theoretically, this behavior could be biased, as it was not possible to place the probe close to
the bottom turbine in the tank; in addition, a vortex was observed at high rotation speed; consequently,
these two points have been addressed specifically in the next section. This behavior is a direct
consequence of the use of a submerged membrane system in a recirculation loop. The advantage is that
the reactor can be driven under atmospheric pressure, which prevents hydrogen inhibition; the
drawback is that filtration is carried out at low transmembrane pressure and thus requires large
membrane surface area, i.e. a large housing; this implies a low velocity in the MF module because the
recirculation flow rate is limited to avoid the accumulation of solid at the inlet or inside the MF
module. Laminar flow conditions could be considered as a drawback in terms of fouling, but as
mentioned above, the main objective is to avoid the presence of straw in the membrane module, which
implies a small inlet tube and a low velocity in the loop. This differs from the common strategy
against fouling when the membrane is submerged in the tank, but this approach seems justified by
10

preliminary results with straw highlighting that fouling increased with the circulation speed in the
loop.
3.2. Experimental analysis of vortex formation and solid suspension
The experimental estimation of a vortex size and shape is an uneasy task [22]. Figure 6 shows that
the gas-liquid interface was fairly flat when rotation speed was 100 rpm, while the vortex appeared
clearly at 200 rpm and reached the three-blade impeller above 400 rpm. The onset of vortex formation
was observed around 150 rpm. When the impeller was closer to the liquid surface to enhance mass
transfer, the vortex appeared at a lower rotation speed. For 100 rpm, this corresponds to a critical
Froude number based on the diameter of the pitched blade turbine (Table 1) of about 0.05, which is a
classical value in the literature for turbines [23]. When the vortex depth reached the level of the
turbine, the radial discharge of this impeller promoted the formation of bubbles, which was an
undesired behavior. Figure 6 illustrates these trends for h=69 mm and C=110 mm. This figure also
shows that the vortex presents an atypical shape at 200 rpm: it is asymmetrical in the picture and two
air fingers can be seen in Figure 6b. In fact, using the two cameras placed perpendicularly, three
fingers appear clearly, resulting from the depression behind the pitched blades. Consequently, the
vortex depth and the finger length increased with increasing the rotation speed (Table 4) because of
the large centrifugal force which drains the water towards the wall of the tank and of the pressure
difference between the front and back faces of the pitched blades that sucks the gas behind the blades.
The vortex seemed reduced in the presence of a dispersed solid phase. It is difficult to conclude
whether this is due to the straw itself, or to a slight raise in viscosity due to the soluble compounds
released by straw in water, or by fines formed when straw had been grinded. The experimental results
showed that straw suspension started when rotation speed was between 150 rpm and 200 rpm,
depending on the position of the bottom Rushton turbine (Table 4), as seen in Figure 7a with only
small particles suspended at 100 rpm. Conversely, the suspension appeared to be nearly homogeneous
at 200 rpm (Figure 7c). In between, an axial segregation emerged with two interfaces and three regions
in Figure 7b: a stationary solid region at the bottom, a region of suspended solids in the middle and at
the top, as In Figure 7a, only small particles are present. All these trends seemed nearly unaffected by
the initial amount of dry straw between 5 and 10 g/L, with a suspension velocity increasing with this
amount. This minimum impeller speed for solid suspension was difficult to predict using conventional
methods [24], first because of the uncertainty on straw properties and on the shape of the straw
particles. The correlations of [24] based on the characteristic length of the bottom impeller always
overestimated the minimum suspension velocity. Other possible reasons are, therefore, that suspension
is favored by a small clearance C and by a round-bottom tank, or by the circulation between the
bottom and the top impellers highlighted in the previous single-phase mixing experiments.
As a conclusion, the experimental data confirms that the selected impeller geometry achieves the
desired objectives: suspension of the solid phase and good distributive mixing in the tank even at low
11

rotation speed. This presents a key advantage: an homogeneous suspension can be obtained without
imposing a forced circulation in the loop, while a partial suspension of solids is possible by decreasing
rotation speed when filtration operates, using an inlet of the forced circulation loop in the top of the
tank. This could avoid the accumulation of solids close to this inlet and inside the circulation loop.
However, the results also point out that the range of rotation speed for which solids are completely
suspended and no vortex is observed is narrow. As a result, the operation range between partial
suspension and complete suspension without and with vortex is tiny and requires an accurate
prediction of the multiphase flow inside the AnMBR for scale-up purpose. This emphasizes the need
for advanced modelling and simulation tools.
3.3. Simulation of mixing in single-phase flows
As reported above, vortex generation can be avoided and this is delayed at rotation higher speed
when straw is added. Thus, the deformation of the gas-liquid interface can be initially neglected in the
single-phase flow simulations carried out to assess the physical and numerical CFD modelling
approach and to simulate mixing experiments in the tank.
Preliminary simulations focused on the mesh size influence. As already mentioned, the MOFOR
approach allows the use of a structured mesh with finer grids near the two impellers. However,
unphysical trends were observed at 100 rpm with a 300,000 cells grid, while grid independency was
achieved when the number of cells was 600,000 or higher, as shown in Figure 8a. This value of
600,000 cells has been retained for further simulations. The numerical differences on velocity profiles
between the standard, RNG and Chen-Kim k- models remained slight over 100 s real time; they were
more visible between the low- and high-Re variants of the k- models on k and values: maximum
value in Table 2 showed that low-Re k- models predicted nearly a laminar flow, which contradicts
mixing time experiments. For high-Re k- models, the results on were very close in Table 2 and the
high-Re Chen-Kim k- model was selected for further simulations because it exhibited much smaller
computation time. Furthermore, larger discrepancies between simulations for the standard, RNG and
Chen-Kim k- models were only related to the first or second-order discretization scheme of the
convective term (Table 2). This probably results from the cut-cell technique when moving boundaries
in a fixed structures grid are involved, which requires higher accuracy. The second-order MinMod
scheme was selected because it had the advantage to reduce computation time by 7% with the k-
Chen-Kim model and by a factor 2 for the standard and the RNG k- turbulence models (Table 2).
This result was strongly linked to the interplay between the turbulence model and the discretization
scheme, as the computational time was identical for the simulations with the MUSCL and MinMod
schemes when laminar flow conditions were assumed.
Using MOFOR, the simulations were able to describe the development of the radial discharge of
each turbine and also the time necessary to reach the pseudo-transient trends deduced through a
stationary approach, such as MRF. For example, Figure 8b shows that far from the two turbines, about
12

35 s are necessary at 100 rpm to achieve a cyclic behavior in terms of axial velocity. Under transient
conditions (Figure 8c), the radial discharge of the two turbines was shown to impinge on the vessel
wall. However, beyond this transient period, the simulations always predicted an interaction between
the two turbines (Figure 8d), in agreement with the mixing experiments. Consequently, only the
discharge of the pitched bladed turbine impinged on the wall because the discharge flow of the Ruston
turbine was deviated by the interaction between the two impellers, although it is commonly admitted
that the interaction between the circulation loops of impellers is negligible when C is higher than 1.5
times the impeller diameter, which was the case if the average diameter of the two impellers was
accounted for. This could also result from the two following reasons:
Due to the small clearance off the tank bottom, the flow near the Rushton turbine was distorted in
comparison to the classical situation and the fluid moved upward near the blades (Figure 8d);
The relative diameter of the pitched blade turbine Dp1/T was close to 1/2, higher than the
conventional 1/3 value.
The local velocity was decomposed into its axial, radial and tangential components for both the
pitched blade and the Rushton turbines. In the planes of the impellers, the local velocity is mainly
tangential in Figure 9, as expected. The radial component is negligible, even for the pitched blade
turbine; its order of magnitude is close or even lower than that of the axial velocity component. As a
result, Figure 9 confirms Figure 8c and shows that the fluid moves upwards close to the Rushton
turbine, between the impellers and close to the pitched blade impeller, in particular between the blades
of the pitched blade turbine, as highlighted by Figure 9. Conversely, the fluid moves downwards close
to the wall, which confirms the development of a recirculation loop between the bottom and top region
of the tank. In addition, the flow pattern did not strongly vary as a function of rotation speed in the
range studied, which means that the key features of the flow varied nearly proportionally with the
rotation speed. It is also worthy of note that the axial velocity values remained low, except in the
regions close to the impellers. In particular, there was a region between the impellers in which the
velocity close to the wall was minimum in the reactor. This corresponds to the region of Figure 7 in
which the straw particles moved nearly horizontally, which highlights that the flow in the solid
suspension is probably close to that of the single-phase flow. Although the flow was obviously
turbulent on the basis of Re and mixing experiments, kinematic turbulent viscosity remained low; for
example, it was never higher than 20/ at 100 rpm and, actually, it did not exceed 10/ except in
the region of the paddles (Figure 10). This was not strongly modified when the clearances off bottom
and between the impellers were varied as defined in Table 1, with only slight differences: Figure 10
demonstrates that the change of turbulent viscosity values remained negligible, but that the size and
shape of the recirculation loops were modified. In practice, the comparison of configurations C, D and
E on this figure showed that lower h values at constant clearance between impellers did not really
improved fluid circulation below the bottom impeller. Similarly, the comparison of configurations A,
B and C demonstrated that increasing the distance between the top impeller and the fluid surface
13

impaired the circulation close the surface, while the circulation between the two impellers was reduced
when their clearance was increased. This highlights that a compromise must be found between the
possible objectives that encompasses mass transfer at the top, solid suspension at the bottom and
concentration homogenization. It must be stressed that the trends of Figure 10 are representative
because vortex is never observed at 100 rpm, but that they may be modified at higher speed when
vortex depth becomes significant.
To confirm these trends, simulations involving the mixing of a passive tracer were carried out. For
the prediction of mixing from CFD, it was necessary to achieve, first, established flow conditions. For
example, for a rotation speed of 100 rpm, Figure 8a had shown that a reproducible cyclic behavior was
achieved after about 35 s, which implies that the injection of the passive tracer was simulated when
t>40 s. The comparison showed higher mixing times in the simulations than in the experiments; thus,
CFD predicted effectively the global circulation observed in the mixing experiments and the influence
of the impeller position, but with a slightly overestimated dimensionless mixing time (about 80) in
comparison with the experiments (about 70). This may be partially due to the fact that in the
experiments, injection in regions of low velocity modifies the local flow field and probably leads to an
underestimation of the experimental mixing time, which was confirmed by visualization experiments.
Finally, Figure 11 also highlights that once the tracer is in the region of the top impeller, dispersion is
driven first by the tangential and the radial discharge of the impeller, followed by the downward
circulation of the fluid in the wall region. Surprisingly, the region in which homogeneity seemed to be
the most difficult to achieve was just above the bottom impeller because it develops a low velocity
region just above the disk of the turbine, as illustrated by Figure 8c.
3.4. Two-phase flow CFD simulations
When the vortex depth becomes significant, simulations in single-phase flows may be biased. This
can be circumvented by the VOF method. Figure 12 shows the predicted profiles of the air-water
interface in the stirred tank reactor for two different rotation speeds using the VOF method. The
simulations perfectly predict that the interface is flat at 100 rpm: velocity profiles are similar for single
phase and VOF simulations (Figure 12). A vortex with gas fingers due to depression emerges behind
the paddles appears at 200 rpm. A comparison of Figure 12 with Figures 6.a and 6.b confirms,
visually, the good agreement, as already seen in the literature [20, 21]. A quantitative comparison is
available in Table 4 and shows that it is possible to predict the shape of the gas-liquid interface in
mixing simulations. Conversely, a key limitation is that it is not possible to combine a VOF method to
a two-fluid Euler-Euler approach, i.e. to simulate at the same time vortex formation and solid
dispersion with Phoenics. Experimentally, the vortex seemed to be reduced when straw was added.
Using this assumption, the Algebraic Slip Model was used to describe the suspension of straw in
the tank. As shown in Figure 13, the solid phase stands at the bottom of the tank when rotation speed
is 100 rpm or below, the suspension is partial when the impellers are driven at 150 rpm and it becomes
14

more homogenous when rotation speed reaches 200 rpm. This fairly agrees with the experimental
results of Figure 7, even though the simulations predict a low solid concentration at the top of the tank
and in the region close to the impeller. A quantitative comparison is reported in Table 4. Slight
differences remain, however, between experiments and simulations, mainly coming from the
heterogeneity of the size and shape distribution of straw particles that are not accounted for by the
modelling methodology. As a rule of thumb, when the initial concentration of straw increases,
suspension is delayed to higher speed; this mainly comes from the volume occupied by the straw
before stirring. Finally, a key result of this work is that CFD is able to roughly predict the onset of
solid suspension, which can be used for scale-up purpose, even though the description of the solid
phase should still be improved.
As a conclusion, these results coupled to the single-phase simulations show that the position and
the shape of the bottom impeller can still be modified for better mixing performance, but that the
present design has an advantage: solid suspension can be controlled at low rotation speed due to the
low relative density of straw. As a result, this design opens the way to the definition of a cyclic
method for recovering permeate that derive both from experiments and simulations:
The tank can be driven at about 200 rpm because this speed prevents vortex in the presence of
straw, while it maintains a circulation of the liquid and the solid phases in the whole reactor;
When permeate has to be recovered, the rotation speed may be slowed down to 100 rpm with an
inlet point of the circulation loop placed close to the top impeller, as this speed ensures good
distributive mixing conditions, while the presence of solids is prevented at the top of the tank, and
thus in the recirculation loop, because straw has settled. This will avoid fouling and help maintain
the effectiveness of the membrane. In this case, the position of the outlet point of the MF module is
less critical.
4. Conclusions
In this work, an AnMBR has been designed using an unbaffled stirred tank coupled to an external
MF module, even though the bioreactor behaves as a submerged membrane device. In the MF module,
low velocity and laminar flow prevail to avoid the presence of solid in the loop. The tank is under
turbulent flow conditions and good distributive mixing is achieved using a two-stage impeller, which
ensures that the mixing time in the tank is shorter than the space time in the circulation loop when
rotation speed is higher than 150 rpm. This is confirmed both by experiments and CFD simulations,
and explained by a macro-circulation between the bottom and the top of the tank. These simulations
have been conducted using the MOFOR approach that relies on the cut-cell technique, which had not
been used for simulating bioreactors up to now, as far as the authors know. These are able to predict
the transient flow and the mixing time, even though it is computationally intensive, coupled to the k-
Chen-Kim turbulence model for closure and to the MinMod discretization scheme for the convective
terms, as it is a good compromise in terms of computation time and accuracy. The CFD simulations
15

are not only able to predict mixing time, but also the formation of a vortex using the VOF method and
the suspension of straw using ASM, which gives access to the necessary information for scale-up
purpose. While using a large top impeller is an advantage for liquid circulation, it induces the
formation of a vortex that is delayed by the presence of solids, which prevents the use of high rotation
speeds. The bottom impeller is designed to promote solid suspension. Despite its high clearance and
its small diameter, this gives access to a precise control of the suspension of straw and to the definition
of optimum rotation speeds when batch cultures are operated and permeate is recovered, respectively,
with an inlet point of the MF recirculation loop close to the top impeller. The perspectives of this work
are, now, to validate the Eulerian simulations using Particle Image Velocimetry and to implement a
simplified biokinetic model of BioH2 production by dark fermentation in the CFD code using the
transport equations of additional scalar variables, first with a simple substrate, such as glucose, for
which the metabolic pathways have already been described in the literature.
5. Acknowledgements
LABEX IMobS3 Innovative Mobility: Smart and Sustainable Solutions, the French National Centre for
Scientific Research (CNRS), Auvergne Regional Council and the European Funds of Regional
Development (ERDF/FEDER) are gratefully acknowledged.
6. References
[1] A. Pandey, C. Larroche, Biofuels: Alternative Feedstocks and Conversion Processes, Academic Press,
Oxford, 2011.
[2] P. Abdeshahian, N.K. Nasser Al-Shorgani, B., N.K.M. Salih, H. Shukor, A. Kadier, A.A. Hamid, M.S. Kalil,
The production of biohydrogen by a novel strain Clostridium sp. YM1 in dark fermentation process, Int. J.
Hydrogen Energy, 39 (2014) 1252412531.
[3] H. Han, H., L. Wei, , B. Liu, H. Yang, J. Shen, Optimization of biohydrogen production from soybean straw
using anaerobic mixed bacteria, Int. J. Hydrogen Energy, 37 (2012) 1320013208.
[4] W.Q. Guo, N.Q. Ren, X.J. Wang, W.S. Xiang, J. Ding, Y. You, B.F. Liu, Optimization of culture conditions
for hydrogen production by Ethanoligenens harbinense B49 using response surface methodology, Bioresour.
Technol. 100 (2009) 11921196.
[5] E. Bittencourt Sydney, C. Larroche, A.C. Novak, R. Nouaille, S.J. Sarma, S.K. Brar, L.A. Junior Letti, V.T.
Soccol, C.R. Soccol, Economic process to produce biohydrogen and volatile fatty acids by a mixed culture
using vinasse from sugarcane ethanol industry as nutrient source, Bioresour. Technol. 159 (2014) 380386.
[6] P. Bakonyi, N. Nemestthy, K. Blafi-Bak, Biohydrogen purification by membrane: an overview on the
operational conditions affecting the performance of the non-porous, polymeric and ionic liquid based gas
separation membranes, Int. J. Hydrogen Energy 38 (2013) 96739687.
[7] P. Bakonyi, N. Nemestthy, V. Simon, K. Blafi-Bak, Fermentative hydrogen production in anaerobic
membrane reactors: A review, Bioresour. Technol. 156 (2014) 357363.
[8] H. Lin, W. Peng, M. Zhang, J. Chen, H. Hong, Y. Zhang, A review on anaerobic membrane bioreactors:
applicants, membrane, fouling and future perspectives, Desalination 314 (2013) 169188.
[9] H. Ozgun, R.K. Dereli, M.E. Rsahin, C. Kinaci, H. Spanjers, J.B. Van Lier, A review of anaerobic membrane
bioreactors for municipal treatment: integration operations, limitations and expectations, Sep. Purif. Technol.
118 (2013) 89104.
[10] J. Wang, W. Wan, Factors influencing fermentative hydrogen production: a review, Int. J. Hydrogen
Energy, 34 (2009) 799811.
[11] S.A. Martnez-Delgadillo, J. Ramrez-Muoz, H.R. Mollinedo, V. Mendoza-Escamilla, C. Gutirrez-Torres,
J. Jimnez-Bernal, Determination of the spatial distribution of the turbulent intensity and velocity field in an
electrochemical reactor by CFD, Int. J. Electrochem. Sci. 8 (2012) 274298.
[12] Y. Han, J.-J. Wang, X.-P. Gu, L.-F. Feng, Numerical simulation on micromixing of viscous fluids in a
stirred-tank reactor, Chem. Eng. Sci. 74 (2012) 917.

16

[13] K. Yang, G.W. Chu, L. Saho, Y. Xiang, L.L. Zhang, J.F. Chen, Micromixing efficiency of viscous media in
micro-channel reactor, Chin. J. Chem. Eng. 17 (2009) 546551.
[14] J. Bertrand, J.-P. Couderc, H. Angelino, Power consumption, pumping capacity and turbulence intensity in
baffled stirred tanks: Comparison between several turbines, Chem. Eng. Sci. 35 (1980) 21572163.
[15] H. Singh, D. Fletcher, J.J. Nijdam, An assessment of different turbulence models for predicting flow in a
baffled tank stirred with a Rushton turbine, Chem. Eng. Sci. 66 (2011) 59765988.
[16] Ch. Vial, Y. Stiriba, Characterization of bioreactors using computational fluid dynamics, in: C.R. Soccol, A.
Pandey, C. Larroche (Eds.), Fermentation Processes Engineering in the Food Industry, CRC Press, Boca
Raton, 2013, pp. 121164.
[17] X. Wang, J. Ding, N.-Q. Ren, B.-F. Liu, W.Q. Guo, CFD simulation of an expanded granular sludge bed
(EGSB) reactor for biohydrogen production, Int. J. Hydrogen Energy, 34 (2009) 96869695.
[18] Y. Wang, T.D. Waite, G.L. Leslie, Computational fluid dynamics (CFD) analysis of membrane reactors:
modelling of membrane bioreactors for municipal wastewater treatment, in: A. Basile (Ed.), Handbook of
Membrane Reactors, Woodhead Publishing, Cambridge, 2013, pp. 532568.
[19] T. Mahmud, J.N. Haque, K.J. Roberts, D. Rhodes, D. Wilkinson, Measurements and modelling of freesurface turbulent flows induced by a magnetic stirrer in an unbaffled stirred tank reactor, Chem. Eng. Sci. 64
(2009) 41974209.
[20] CHAM, The Phoenics Encyclopedia, http://www.cham.co.uk/ChmSupport/encindex.php, 2014.
[21] V.V. Ranade, Computational Flow Modeling for Chemical Reactor Engineering, Academic Press, London,
2002.
[22] J.-P. Torr, D.F. Fletcher, T. Lasuye, C. Xuereb, An experimental and computational study of the vortex
shape in a partially baffled agitated vessel, Chem. Eng. Sci. 62 (2007) 19151926.
[23] E.S. Gaddis, Heat transfer and power consumption in stirred vessels, in: VDI Heat Atlas (2nd ed.), Springer,
Heidelberg, 2010, pp. 14511484.
[24] A.W. Nienow, The suspension for solid particles, in: N. Harnby, M.F. Edwards, A. Nienow (Eds.), Mixing
in the Process Industries (2nd ed.), Butterworth-Heinemann, 1997, Oxford, pp. 364392.
List of symbols
C
CD
3
C0

0 , *+ , -+
dp
Dp
Dp1
Dr1
Dr
Ds
Dt
E()
?@,@
g
G
H
H(t)
HL
h
hbo
hbr
hp
d
!
Lbr
P
Re

Clearance between the two impellers (m)

Drag force coefficient
Concentration of component x at time t (mol/L)
Initial concentration (mol/L)
Final tracer concentration (mol/L)
Coefficients of the k- model
Particle diameter (m)
Pitched blade impeller diameter (m)
Characteristic dimension of the pitched blade impeller (m)
Characteristic dimension of the Rushton impeller (m)
Rushton impeller diameter (m)
Shaft diameter (m)
Turbulent diffusivity (m2/s2)
Normalized RTD
Body force between two phases (N)
Acceleration of gravity (m2/s)
Volumetric production rate of turbulent kinetic energy (W/m3)
Tank height (m)
Heaviside function
Liquid height in the tank (m)
Impeller clearance off the tank bottom to impeller (m)
Height of the round-bottom section of the tank (m)
Height of the blade of the Rushton turbine (m)
Blade length of the pitched blade impeller (m)
Unit tensor
Turbulent kinetic energy (m2/s2)
Blade length of the Rushton impeller (m)
Pressure (Pa)
Reynolds number

17

Rer
53
T
t
tc
tm

>

W
Y

Relative Reynolds number

Source term of species x
Tank diameter (m)
Time (s)
Circulation time (s)
Mixing time (s)
Reynolds-average mean velocity (m/s)
Slip velocity between the two phases (m/s)
Turbulent velocity fluctuation (m/s)
Width of the impeller blades (m)
Degree of mixing

Greek letters
6@



e

K
/, , /+


f

Volume fraction of phase

Turbulent energy dissipation rate (W/kg)
Molecular viscosity (Pa.s)
Turbulent viscosity (Pa.s)
Dimensionless time
Fluid density (kg.m-3)
Solid density (kg.m-3)
Coefficients of the k- model
Space time (s)
Viscous stress tensor (Pa)
Angle of turbine impeller

Subscripts
Y, g
G
L

Principal direction
Gas
Liquid

Abbreviations
AnMBR
ASM
BIOH2
CFD
MF
MOFOR
MRF
RANS
RTD
VFA

Anaerobic Membrane Bioreactor

Algebraic Slip Model
Biohydrogen
Computational Fluid Dynamics
Microfiltration
Moving Frame of Reference
Multiple Reference Frame
Reynolds-average Navier-Stokes equations
Residence Time Distribution
Volatile Fatty Acids

List of figures
Fig. 1. Experimental set-up: AnMBR with an external hollow fibre membrane module. 1. Tank (5 litres), 2.
Control unit, 3. Impeller, 4. Peristaltic pump, 5. Hollow fibre membrane module, 6. Permeate withdrawal
pump; Pinlet, Poutlet, Pdiff are pressure sensors.
Fig. 2. Stirred tank geometry: picture of the 5-L stirred tank equipped with a two-stage impeller (left); Schematic
diagram of the stirred tank reactor (right).
Fig. 3. Experimental set-up for mixing time analysis in the stirred tank (1), for mixing analysis in the membrane
module without permeate extraction (2), and for mixing analysis in the membrane module with permeate
extraction (3).
Fig. 4. (a) CFD geometry of the tank; (b) front face view of the structured mesh.
Fig. 5. (a) Normalized concentration vs. time in the batch stirred tank (C=110 mm, h=69 mm, 30 rpm); (b)
Normalized response curves (E) vs. normalized time with a circulation flow rate of 300 mL/min compared
to the1D model (plug flow reator + one or two mixed tanks).

Fig. 6. Pictures describing vortex formation for a rotation speed of: (a) 100 rpm, (b) 200 rpm and (c)
400 rpm, respectively (h=69 mm, C=110 mm).

18

Fig. 7. Pictures illustrating the evolution of solid suspension (10 g/L) with the rotation speed of the impeller: (A)
100 rpm, (B) 150 rpm, (C) 200 rpm.
Fig. 8. Example of CFD data for rotation speed 100 rpm, C=110 mm and h=69 mm: (a) Velocity profile for
checking grid independency; (b) Evolution of the local tangential velocity vs. time in the reactor close to the
wall at the plane of the top impeller; (c) velocity field in the middle plane (t=3.4 s); (d) velocity field in the
middle plane (t=40 s).
Fig. 9. Velocity field along the pitched blade impeller turbine and the Rushton turbine at 100 rpm using the
Chen-Kim k- model in single phase flow using a no-slip condition at the gas-liquid interface (h=69 mm,
C=110 mm).
Fig. 10. Evolution of turbulent kinematic viscosity with impeller position and clearance (100 rpm t=40 s). A.
h=6.9 cm, C=9 cm; B. h=5.9 cm, C=10 cm; C. h=6.9 cm, C=11 cm; D. h=4.9 cm, C= 11 cm; E. h=5.9 cm,
C=11 cm (the kinematic viscosity of the liquid phase is 10-6 m/s).
Fig. 11. Example of tracer dispersion based on a Heaviside injection at t=15 s after injection for rotation speed
100 rpm (normalized data: C1=100% without tracer and C1=0% at injection concentration, h=69 mm, C=110
mm).
Fig. 12. Plot of the phase indictor function describing the vortex depth predicted by the VOF method for a
rotation speed of: (a) 100 rpm and (b) 200 rpm, respectively (h=69 mm, C=110 mm).
Fig. 13. Spatial distribution of the solid phase using simulations based on ASM as a function of rotation speed
using 10 g/L straw (h=69 mm, C=110 mm).
List of tables
Table 1. Main geometric features of the stirred tank and impellers.
Table 2. Influence of turbulence modelling and discretization scheme on the computation time and the
maximum turbulent viscosity value for 100 rpm and simulations on 2.4 s real time (h=69 mm, C=110 mm).
Table 3. Typical results on mixing and circulation times in the tank and value of the peak in RTD
measurements as a function of rotation speed in the tank, circulation and flow rate and permeate flow rate. Space
times () values in the tank and the loop are added for comparison purpose.
Table 4. Comparison of multiphase experimental and CFD data: vortex depth (with VOF simulations)

and height of settled straw (with ASM simulations).

19

Geometry of stirred tank and turbine impeller

Stirred tank reactor
Tank diameter
T=170 mm

Pitched blade
impeller
3 blades
=45

Rushton impeller
4 blades
=90

Tank height
H=360 mm

Diameter
Dp =82 mm
Dp/T=0.48

Diameter
Dr=56 mm
Dr/ T=0.33

Liquid height
HL =270 mm

Dp1=56 mm

Dr1=40 mm

Tank bottom height

hbo=85 mm

Height
hp= 25 mm

Height
hbr= 12 mm

Clearance off bottom

46 mm < h <69 mm
Shaft

Clearance between impellers

90 mm < C < 110 mm
Ds = 12mm

20

Upwind (1st order)

Minmod (2ndorder)
Muscl (2nd order)

k- - Low Re
8336 s
8146 s
14210 s

k-
2326 s
2951 s
6092 s

Upwind (1st order)

Minmod (2ndorder)
Muscl (2nd order)

k- - Low Re
0.01
0.01
0.3

k-
0.9
0.7
0.7

Computation time
Chen-Kim
RNG
1906 s
2098 s
2598 s
2526 s
2781 s
5591 s
Maximum (m2/s 3)
Chen-Kim
RNG
0.01
0.9
0.7
0.7
0.4
0.4

Chen-Kim - Low Re
7615 s
Chen-Kim - Low Re
0.01

21

Tank

hi (s)

hj (s)

0
0
0
0

100
100
100
100
150
150
150
150

Circulation
flow rate
(mL/min)

Impeller
speed
(rpm)

Permeate
Flow rate
(mL/min)

0
100
200
300

50
50
50
50

0
0
0
0

0
100
200
300

100
100
100
100

100
200
250
300
0
100
200
300

Loop
(s)

peak

(s)

6 1
7 1
72
6 1

3000
1500
1000

0.51
0.48
0.50

300
150
100

423
403
423
413

51
52
5 1
5 1

3000
1500
1000

0.52
0.49
0.53

300
150
100

15.3
15.2
15.5
15.1

3000
2000
1500
1000

0.51
0.47
0.48
0.52

300
200
150
100

0
0
0
0

272
252
242
251

<3
<3
<3
<3

3000
1500
1000

0.52
0.49
0.47
0.49

300
150
100

905
914
895
903

22

Speed (rpm)
CFD
Experiments

Vortex depth (cm)

100
150
200
Flat
1.1
3.0
Flat
1.2
3.1

400
divergence
8.7

Height (cm) of settled straw (10 g/L)

100
150
200
Speed (rpm)
8.3
1.6
0.4
CFD
8.15
1.2
0.6
Experiments

23

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

HIGHLIGHTS

Mixing conditions were investigated in an anaerobic submerged membrane bioreactor

Experimental and modeling tools were used, including computational fluid dynamics
Mixing time, vortex formation and solid suspension data were compared to CFD
Optimum rotation speed and mixer design agree between experiments and simulations
Anaerobic submerged membrane bioreactors are promising for biohydrogen production