Bioresource Technology 101 (2010) 70057013

Contents lists available at ScienceDirect

Bioresource Technology
journal homepage: www.elsevier.com/locate/biortech

CFD optimization of continuous stirred-tank (CSTR) reactor

for biohydrogen production
Jie Ding a,*, Xu Wang a, Xue-Fei Zhou b, Nan-Qi Ren a,**, Wan-Qian Guo a
a
b

State Key Laboratory of Urban Water Resource and Environment (HIT), Harbin Institute of Technology, Harbin 150090, China
State Key Laboratory of Pollution Control and Resource Reuse Research, Tongji University, Shanghai 200092, China

a r t i c l e

i n f o

Article history:
Received 4 February 2010
Received in revised form 23 March 2010
Accepted 30 March 2010
Available online 27 April 2010
Keywords:
Biohydrogen production
Continuous stirred-tank reactors (CSTR)
Computational uid dynamics (CFD)
Hydrodynamics
Reactor design

a b s t r a c t
There has been little work on the optimal conguration of biohydrogen production reactors. This paper
describes three-dimensional computational uid dynamics (CFD) simulations of gasliquid ow in a laboratory-scale continuous stirred-tank reactor used for biohydrogen production. To evaluate the role of
hydrodynamics in reactor design and optimize the reactor conguration, an optimized impeller design
has been constructed and validated with CFD simulations of the normal and optimized impeller over a
range of speeds and the numerical results were also validated by examination of residence time distribution. By integrating the CFD simulation with an ethanol-type fermentation process experiment, it was
shown that impellers with different type and speed generated different ow patterns, and hence offered
different efciencies for biohydrogen production. The hydrodynamic behavior of the optimized impeller
at speeds between 50 and 70 rev/min is most suited for economical biohydrogen production.
2010 Elsevier Ltd. All rights reserved.

1. Introduction
Hydrogen energy is an obvious candidate as a sustainable
replacement for fossil fuel, as it produces lower emissions in use
and is potentially environmentally benign and cleaner than fossil
fuels (Ren et al., 1997; Kapdan and Kargi, 2006; Li et al., 2008). Biohydrogen production through anaerobic fermentation of organic
substrates has been extensively researched, as the process can
use materials regarded as pollution to generate hydrogen (Ren
et al., 2006; Cai et al., 2004; Bhaskar et al., 2008). Although biohydrogen production is a complex, multiphase chemical, biological
and physical process with numerous internal interactions between
gas, liquids and solids, present research on biohydrogen production has focused primarily on the biological and chemical characteristics. A number of chemical and biological factors which
affect the efciency of hydrogen production, such as fermentation
type, anaerobic fermentation terminal products and the effect of
different substrates, have been investigated (Chang et al., 2002;
Lee et al., 2004; Nath and Das, 2004; Guo et al., 2009). Through previous research several fundamental breakthroughs have been
achieved in the understanding of biohydrogen production, including the isolation of microbial strains with high hydrogen production capability, identication of high-efciency and low-cost
* Corresponding author. Tel./fax: +86 0 451 86282193.
** Corresponding author. Tel./fax: +86 0 451 86282193.
E-mail addresses: dingjie123@hit.edu.cn (J. Ding), rnq@hit.edu.cn (N.-Q. Ren).
0960-8524/$ - see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biortech.2010.03.146

carbon sources and optimization of the microbial fermentation

process (Guo et al., 2008a,b; Cao et al., 2010).
In contrast, physical characteristics affecting the efciency of
biohydrogen production have received very little attention.
Although a variety of laboratory-scale and pilot-scale reactors with
either continuous ow or microbial growth carrier media made
from foam or plastic have been developed (Kraemer and Bagley,
2005; Logan et al., 2002; Gavala et al., 2006), most of these reactors
were designed by semi-empirical correlation. The effect of velocity
elds, distributions of shear stresses, turbulent intensity and volume fraction of multiphases that signicantly affect the composition of the microbial community, biomass activity and settling
rate of the activated sludge, have largely been ignored (Jin and
Lant, 2004; Hamzehei and Rahimzadeh, 2009; Meroney and
Colorado, 2009). Understanding the hydrodynamic phenomena involved in biohydrogen production is a necessary precursor to
industrial scale application. To optimize the reactor conguration
and therefore improve the performance of a biohydrogen production reactor, it is essential to develop and apply new methods to
enhance our understanding of reactor hydrodynamics.
Modern computational uid dynamics (CFD) software can predict uid ow, heat and mass transfer, chemical reactions and
other related phenomena by solving a set of appropriate mathematical equations, describing these processes as mass, momentum,
energy and species balances, methods, which have been widely
and successfully employed in water and wastewater treatment
systems (Wang et al., 2009; Terashima et al., 2009; Pougatch

7006

J. Ding et al. / Bioresource Technology 101 (2010) 70057013

Nomenclature
Ci
DI
g
h
k
m
n
N
p
Q
pini
P
r
Re
t
ti
U

tracer concentration in the efuent at ti (g l1)

impeller diameter (m)
gravitational constant (m s2)
depth of mixture in the reactor (m)
turbulent kinetic energy (m2 s2)
tracer mass injected (g)
impeller speed (rev)
impeller speed (rev m1)
pressure(N m2)
tracer ux (l s1)
initial pressure (N m2)
power (W)
radial distance (m)
Reynolds number
time (s)
sampling time (s)
vector of velocity (m s1)

et al., 2007) eliminating the requirement for expensive post-construction eld tests. Furthermore, if performed before construction
this approach eliminates the painful realization that a system is
inefcient after installation.
This paper establishes a computational model of a continuous
stirred-tank reactor (CSTR) for biohydrogen production and the
simulation results are validated by examination of residence time
distribution (RTD). The CFD simulation is used to portray hydrodynamics behavior in the reactor, including the velocity eld, biogas
volume fraction, turbulence kinetic energy and shear strain rate.
Conguration optimization of the reactor is achieved by optimizing
the impeller design. Comparisons were then made between normal
and optimized impeller performance and the inuence of hydrodynamic behavior on biohydrogen production was predicted and
validated.
2. Methods
2.1. Reactor conguration and operating conditions
A continuous stirred-tank reactor (CSTR) with a total capacity of
17 L was operated in a continuous ow mode for biohydrogen production (Fig. 1A). Normal molasses, containing about 53% sugars,
was diluted by water to a chemical oxygen demand (COD) concentration of 3000 mg/L and used as a substrate. The inuent substrate
was pumped into the reactor continuously from the feed tank and
then mixed with anaerobic-activated sludge by impeller. The shaft
of the top-driven impeller was concentric with the axis of the reactor. The normal impeller design, with a blade angle of 45 and an
external diameter of 100 mm (Fig. 1B), was run at varying speeds
to get different ow patterns. The optimized impeller has a blade
angle of 45 and an external diameter of 120 mm (Fig. 1C). Four
bafes were equally placed around the inner tank, each with a
width of 20 mm.
The temperature was maintained at the level of 35 1 C.
NaHCO3 was added to the feed solution to maintain pH at 6.5
7.5 in the inuent and hence to keep a pH level of 5.0 in the reactor.
The hydraulic retention time (HRT) was 8 h. COD Nitrogen: Phosphorus ratio in the inuent was maintained at an average of
250:5:1 by adding synthetic fertilizer in order to supply the microorganisms with adequate nitrogen and phosphorus. The mixed liquor volatile suspended solid (MLVSS) in the reactor was
approximately 13.5 g/L. Different impeller geometries were used
to determine the hydrodynamic effect impeller design on biohy-

Vtip
V
VFa
VFb
z

e
l
q
h

s
ra
CD
SMSa

Cab
Ma

impeller tip velocity (m s1)

velocity in reference frame (m s1)
volume fraction of phase a
volume fraction of phase b
axial distance (m)
turbulent energy dissipation rate per unit mass (m2 s3)
molecular viscosity (Pa s)
density of mixture (kg m3)
dimensionless time, dened in Eq. (14)
the mean residence time (s)
volume fraction of phase a
dimensionless drag coefcients
specied mass sources
mass ow rate per unit volume from phase a to phase b
interfacial forces acting on phase a due to the presence
of phases b

drogen production, running each design at impeller speeds of 50,

70, 90, 110 and 130 rev/min.
2.2. Analytical methods
The hydrogen fermentation performance was evaluated using
key parameters, such as pH, oxidation reduction potential (ORP),
alkalinity, ethanol concentration, volatile fatty acid (VFAs) distribution and biogas production (Lin and Hung, 2008). In this work,
temperature, pH, alkalinity, biogas yield and composition were
measured or monitored daily. Analyses of viscosity, density, total
suspended- and volatile-suspended solids of mixture were carried
out once a week. These analyses were performed according to
appropriate standard methods (APHA, 1998). Biogas yield was
measured at room temperature by a wet gas meter (LMF-2, Far
Asia Co. Ltd.), while its constituents were analyzed using an on-line
industrial gas chromatography (RQD-102, Chongqing Sichuan
Instrument Co. Ltd.) or an outline gas chromatography (SC-7,
Shandong Lunan Instrument Factory). The outline gas chromatography was equipped with a thermal conductivity detector and a
stainless steel column (2 m  5 mm) lled with Porapak Q
(5080 meshes). Nitrogen was used as the carrier gas at a ow rate
of 40 ml/min. Dose of injected sample was 0.5 ml each time. Based
on the percentage of hydrogen in biogas, the hydrogen yield could
be calculated. The hydrogen yield could be then calculated based
on the percentage of hydrogen in the biogas.
2.3. Methodology for computational uid dynamics model generation
2.3.1. Computational uid dynamics model
The equations which describe the processes of momentum,
heat and mass transfer are discretized and solved iteratively for
each control volume. As a result, an approximation of the value
of each variable at specic points throughout the domain can be
obtained. The turbulence equations were also solved in conjunction with the continuity equation, the NavierStokes equation,
and the energy equation. According to Meroney and Colorado
(2009), the standard k-e model is the most adopted turbulence
closure because of its simplicity, low computational requirement
and good convergence for complex turbulent ows. We assume
that the mixture of substrate and activated sludge in the biohydrogen production reactor is homogeneous and incompressible,
and that the various components of the mixture shared the same
mean velocity, pressure and temperature elds. The biogas which

J. Ding et al. / Bioresource Technology 101 (2010) 70057013

7007

Fig. 1. (A) Schematic diagram of the CSTR biohydrogen production system; schematic diagram of the impeller (B) normal impeller; (C) optimized impeller; (D) mesh layout
for reactor geometry; and (E) location of planes and lines for evaluation.

was produced in process of fermentation was introduced into the

reactor as additional uid-mass source from twelve points that
were distributed over the reacting area of the reactor. It was also
assumed that biogas bubbles were distributed in the mixture as
spherical particles with a mean diameter of 1 mm. From the above
assumptions, there is one gas phase (biogas) and one liquid phase
(mixture) in the reactor, so the simulation can adopt multiphase
ow models. Two distinct multiphase ow models are available
in ANSYS CFX: an EulerianEulerian multiphase model and a
Lagrangian Particle Tracking multiphase model. In the rst approach, the particles are modeled with a certain mass to impose
the momentum exchange at the point locations of the particles.
In the second approach, particles are tracked through the ow in
a Lagrangian way. In this case, an EulerianEulerian multiphase
model has been used to describe the ow behavior of each phase,
and the gas and liquid phases are treated as different continua,
interpenetrating and interacting everywhere in the computational
domain. The mixture acts on each biogas particle inuencing its
path. The biogas particles in turn affect the turbulence quantities
of the mixture. For each phase, the full set of conservation equations was solved and each phase had a different velocity eld.
The mechanisms of the interaction of the phases were the ow
resistance modeled by momentum transfer, the phase changes
modeled by mass transfer and the heat conduction modeled by
energy transfer (Murthy et al., 2007). Two different sub-models,
the homogeneous model and the inhomogeneous model are avail-

able for EulerianEulerian multiphase ow. The inhomogeneous

particle model used in this simulation assumes that the mixture
is continuous (phase a) and the biogas is dispersed (phase b).
The momentum equation of the multiphase system is given
below:

@
r a qa U a r  ra qa U a  U a
@t
r a rpa r  r a la rU a rU a T

Np
X
Cab U b  Cba U a Ma

b1

where Ma describes the interfacial forces acting on phase a due to

the presence of phase b.
The following general form is used to model interphase drag
forces
d

Ma cab U b  U a

2
d

where coefcient cab was computed applying the dimensionless

drag coefcient

CD 1

q U a  U b 2 A
2 a

cab

CD
Aab qa jU b  U a j
8

3
4

7008

J. Ding et al. / Bioresource Technology 101 (2010) 70057013

Continuity equations and conservation equations were described as follows:

Np
X
@
Cab
r a qa r  ra qa U a SMSa
@t
b1
!
 X
NP
X
X 1 @ q
1
a
r  ra qa U a
SMSa
Cab
@t
a qa
a qa
b1

5
6

2.3.2. Computational domain and boundary conditions

The multiple reference frame (MRF) approach is one of the most
commonly used numerical methods to model stirred reactors. This
method is adequate for simulating the ow and mixing generated
by the impellers and benets from a low total computational time
(Dakshinamoorthy et al., 2006; Pramparo et al., 2008). The reactor
was divided into a rotating domain (rotating reference frame) and
a stationary domain (stationary reference frame). The boundary of
the rotating domain comprising the impeller and part of the impeller shaft positioned from z = 60 mm to z = 80 mm with r = 70 mm.
A MRF with Frozen Rotor interaction scheme was used in the simulations to bond the stationary and the rotating domains. The mixture and biogas were used as the domain uid, and the uid
properties and other key parameters such as density and dynamic
viscosity, were specied according to measurement results. The
multiphase ow was regarded as buoyant and was implemented
by a density difference model. The additional buoyancy force was
calculated by considering the difference in the densities between
phases. The buoyancy reference density was set as that of the less
dense uid (the biogas).The gravity vector is aligned with the axis
of rotation.
Before leaving the tank by the outlet pipe, the mixture overows continuously through the bafe at the top of the reactor
where a free surface was formed. The height of this surface was
used to calculate initial value of volume fraction and the static
pressure

h 0:26m

VF b stepz  h=1

VF a 1  VF b
pini q  g  h  z  VF a

9
10

The inlet where the inuent substrate was pumped into the
reactor was modeled with a mass ow rate inlet boundary condition. The turbulence boundary conditions at the inlet were given
through the low turbulence intensity (1%). The outlet of the mixture was set as a static pressure outlet boundary condition where
the atmospheric pressure was specied, whereas the outlet of
the biogas at the top of the reactor was set as the opening boundary condition. All other solid surfaces including impeller blades,
shaft, bafe and reactor walls were dened by wall boundary conditions with free slip for the biogas and no slip for the mixture.
Uniform temperature distributions were assumed.
2.3.3. Numerical solution
In this work, full computational geometry was used for CFD calculations to capture more accurate results from transient characteristics. The geometry and the unstructured grid of the reactor
used in the experiments were generated by ANSYS ICEM with a
set of user-specied mesh characteristics, which enables the nest
and coarsest grid to be set up in each coordinate direction, with the
gradient of the mesh being rened near solid boundaries (Fig. 1D).
A tetra meshing algorithm was used to ll the volume with tetrahedral elements and to generate a surface mesh on the object surfaces. The predictions of turbulent quantities are usually quite
sensitive to the number of grid nodes used in the solution domain,

so it is very important to use an adequate number of computational cells while the governing equations over the solution domain are solved numerically. The pressure was selected to
conduct the mesh test. The simulation results vary little with grid
density so truncation errors in the numerical simulation can be neglected. An analysis independent of grid was performed to eliminate errors in simulation accuracy, numerical stability,
convergence and computational step related to grid coarseness.
The grid independent analysis was done with three different cell
numbers. When the optimum cell number was used, the difference
in pressure drop was below 5%, which means the most positive
outcome. The optimized mesh for the reactor is of 9,433,203 volume elements and 1,618,028 nodes.
To minimize calculation time, the simulations were divided into
two parts. First, a steady state simulation of the complete turbulent
ow eld for multiphase in the reactor was carried out. The second
step was the transient simulation of the RTD with the rst step results as initial value input le. For convenient comparison with the
RTD experiment results, a tracer transport method was adopted
which is generated by a step boundary condition at the inlet, at
time zero of the transient run (Moullec et al., 2008). The response
at the outlet boundary condition is monitored.
A commercial computational uid dynamics (CFD) code,
namely ANSYS CFX, was employed to explain the reactor hydrodynamic behavior by numerical methods. The simulations were carried out on a PC equipped with an Intel Xeon 2.4 GHz processor
and 8 GB RAM. Although the upwind scheme is very robust, it does
introduce diffusive discretization errors. A high resolution scheme
which is both accurate and bounded was used along with automatic timescale control to achieve steady state conditions. In the
steady state simulations stage, the convergence criterion of
1.E4 for the root mean square (RMS) residual target was reached
within 400 iterations. The total simulation time for each case was
around 50 h. For a transient simulation with time steps of 5 min, a
convergence criterion of 1.E4 for RMS residual target was reached
within 600 iterations.
2.4. Trace experiment
Experimental measurements of the ow eld are necessary for
calibrating and validating the simulation models. Further renements to the model would be required to improve further the
agreement of the simulation results with experimental values. Considering the complex and small scale nature of the structure under
examination, intrusive measurement techniques, such as impedance probes, optical ber probes, ultrasound probes and hot lm
anemometry cannot be applied to investigate ow behavior in a
laboratory-scale biohydrogen production reactor. The presence of
the probe will inevitably affect the ow within the boundary layer,
which is in turn likely to disturb the ow pattern under consideration. Particle image velocimetry (PIV) (Krepper et al., 2008;
Darmana et al., 2005), laser doppler anemometry (LDA) (Kulkarni
et al., 2007) and tomography (Vesselinov et al., 2008) are widely
used non-intrusive techniques to measure ow elds in transparent
uids. However, measuring opaque multiphase ows in a biohydrogen production reactor is challenging, since there is no straightforward application of the type widely used for transparent
single-phase measurement techniques. In this work, RTD was carried out to compare experimental with simulation results. Although
experimental validation by RTD only is not sufcient for a deep
investigation of the ow eld, RTD is a fundamental parameter in
reactor design which can give information on how long the
substrate has been in the reactor for anaerobic fermentation. A
two-point detection method was used, allowing the measurement
of the concentration evolution at both the inlet and outlet of the
reactor. A small quantity of lithium chloride (LiCl) was used as

J. Ding et al. / Bioresource Technology 101 (2010) 70057013

tracer and injected by syringe into the inlet tube, simulating a pulse
with minimal disturbance to the ow inside the reactor.
3. Results and discussion
3.1. Model validation
In the RTD experiment, comparison of the obtained inlet and
outlet curves allows an estimation of the mean residence time. Previous results from steady state simulations have been taken as initial conditions for transient RTD simulations at the same impeller
speeds. The results of the experiment and simulation are interpreted in dimensionless and normalized form. h is the dimensionless time and s is the mean residence time. Eh is normalized RTD
function

11

P
tC
s Pi i
Ci
Eh

Q
Ch
m

12
13

As far as the CFD model verication is concerned, Fig. 2A and B

illustrates a comparison between the experimentally measured
and the simulated data of the RTD showing a good level of agreement between the measured and predicted values. The variance
of the RTD curve with different impeller at N = 90 rev/min. The
simulation and experimental results are compared showing that
the relative error between the measured and simulated data was
within 20% indicating that the model provides a good overall
description of reactor behavior.
3.2. Hydrodynamics evaluation
Five steady state simulations, at stirrer speeds from 50 rev/min
to 130 rev/min at intervals of 20 rev/min, were conducted for two

7009

impeller designs. The biogas volume fraction, velocity eld, kinetic

energy from turbulence and shear strain rates were evaluated in
certain lines and planes passing through the reactor. The location
of these lines and planes are shown in Fig. 1E.
Fig. 3 AD shows the mixture velocity in Reference Frame generated by different impellers at various rotation rates. Using this
variable instead of mixture velocity results in the velocity vectors
appears to be continuous at the interface between the rotating
and stationary domains. Velocity variables that do not include a
frame specication always use the local reference frame. Fig. 3A
D shows the velocity vector over a cross-sectional plane (plane 1)
which was placed at the lengthways mid section of the reactor.
The mixture ow agitated by the impeller travels upward in the radial direction and then splits into two streams near the wall of the
tank. One stream above the impeller creates an upward vortex area
near the surface of mixture; another below the impeller creates a
large downward vortex area towards the bottom of the reactor.
The mixture ows up in the inner part of the reactor tank and then
returns downwards at the periphery, diverted by the separating
bafe (see Fig. 1A for a diagram of the inner tank and separating
bafe). The normal impeller (Fig. 3A and B) generates a more powerful vortex area near the bottom which suspends more sedimentary activated sludge than the optimized impeller, which (Fig. 3C
and D) brings a stronger vortex higher up the reactor, which is directly available for mixing of fermentation substrates and anaerobic-activated sludge in the top area of the reactor. Vortices are also
indicating areas of locally higher residence times.
The predicted distributions of turbulent kinetic energy (k) in the
bottom region of the reactor (Fig. 3E and F) show that the values of
k in the impeller region were higher than those in the bulk ow region due to large spatial velocity gradients. The energy associated
with the normal impeller ow is higher than that of the optimized
impeller ow in the bottom region of the reactor, and so the turbulence intensity for the normal impeller is also relatively high in this
region. Moreover, the increased speed of optimized impeller does
not have much inuence on the turbulence kinetic energy in the
bottom of reactor. These simulation results show reasonable

Fig. 2. RTD curves with different impeller at N = 90 rev/min (A) normal impeller; (B) optimized impeller); velocity prole of the line (C) normal impeller; and (D) optimized
impeller).

7010

J. Ding et al. / Bioresource Technology 101 (2010) 70057013

Fig. 3. Velocity vectors of plane 1 (A) normal impeller with 50 rev/min; (B) normal impeller with 130 rev/min; (C) optimized impeller with 50 rev/min; (D) optimized
impeller with 130 rev/min; turbulence kinetic energy contour of reactor bottom region (E) normal impeller; and (F) optimized impeller.

correlation with experimental observation of the sedimentary

sludge in the bottom of the reactor. Different ow patterns in the

top area of the reactor (Fig. 4A and B) show the velocity contours
of plane 3 (a cross section in the top of the inner tank) with respect

J. Ding et al. / Bioresource Technology 101 (2010) 70057013

7011

Fig. 4. Velocity contour of plane 3 (A) normal impeller with different speed; (B) optimized impeller with different speed; shear strain rate contour of plane 2 (C) normal
impeller; and (D) optimized impeller.

to impeller speed. It can be seen that with an increase in the impeller speed, the velocity of the mixture at the top of the reactor has
increased. In the case of the normal impeller, the stagnation region
is obvious when impeller speed is under 70 rev/min. Uniformity of
velocity distribution is achieved as the impeller speed increases. In
contrast, optimized impellers can generate better velocity distributions in the top area of the reactor at lower impeller speeds.
Fig. 2C and D shows the velocity prole of line A (see Fig. 1E for
Line A). The samples which compose line A are plotted along the
inner tank wall of the reactor, with a non-dimensional distance
from the wall of the samples of 0.3. Velocity distributions are represented by ve speeds with a normal impeller (Fig. 2C) and ve
speeds with the optimized impeller (Fig. 2D). The velocities are
normalized using the impeller tip velocity (Vtip). The axial distance
is the non-dimensional distance from the bottom of the inner tank.
As can be seen, the normal velocity prole gives a maximum value
in the bottom of the inner tank and declines to minimum value
near the top, whereas optimized velocity prole gives a maximum
value above the impeller and not much decrease in the top area.
The inner tank is the primary area for the process of anaerobic fer-

mentation, and a well-proportioned velocity value means effective

use of the volume of the reactor, therefore, an optimized impeller
offers best efciency gains.
The magnitude of the uid hydraulic forces, often expressed as
shear stress or shear rate, is another important parameter for bioreactor design and operation. Areas of high shear strain rate or
shear stress are also typically areas where the highest mixing occurs, but the associated high shear stress can cause damage to
microorganism cells and sludge ocs, and should be avoided (Cao
and Alaerts, 1995; Luo and Muthanna, 2008). Proper understanding of the hydrodynamic shear stress is therefore crucial for the
successful design of a biohydrogen production reactor. Fig. 4C
and D shows the shear strain rate contour of plane 2, which is a
cross section taken just above the impeller (see Fig. 1E for plane
2). Sudden changes with the optimized impeller speed in the shear
strain rate can be clearly observed in Fig. 4D. The shear strain rate
changes change smoothly with variation in the normal impeller
speed (Fig. 4C).
Another important hydrodynamic characteristic which inuences the process of biohydrogen production is the biogas volume

7012

J. Ding et al. / Bioresource Technology 101 (2010) 70057013

fraction in the reactor. The biogas is mainly composed of CO2 and

H2 with the percentage of H2 ranging from 35% to 45% and results
from anaerobic fermentation. As it is a product of fermentation
metabolism, any biogas holdup in the mixture will restrain the
reaction. Previous study indicates that a great amount of CO2 and
VFAs (such as acetic acid, propionic acid, buritic acid, etc.) inhibit
the metabolic activities of anaerobic bacteria. Since H2 is a byproduct of fermentation, it is reasonable that H2 production is reduced
when the metabolic pathways for VFAs and CO2 are inhibited. Contours of the biogas volume fraction in the reactor are shown in
Fig. 5. This picture indicates that the biogas volume fraction increased in proportion to the rotational speed of the normal or optimized impeller. The optimized impeller generates a higher biogas
volume fraction than the normal impeller at the same speed. The
buoyancy of the biogas drives the gas ow upwards. With both
impellers, the biogas volume fraction was reached a higher value
near the surface and the separating bafe where biogas is separated from mixture.
3.3. Effect on biohydrogen production

Fig. 6. Average biogas yield and hydrogen yield by normal impeller and optimized
impeller with different speed.

Fig. 6 shows the average biogas yield and hydrogen yield by

normal impeller and optimized impeller with different speeds
from 50 to 130 rev/min. In a reactor congured with the normal
impeller, the average biogas yield increased rapidly from 11.8 L/d
to 26.1 L/d when the impeller speed increased from 50 to 70 rev/
min, and reached a peak value 29.2 L/d when the impeller speed
was 90 rev/min. A maximum hydrogen production rate was obtained when impeller speed was 110 rev/min. Both average biogas
and hydrogen yield decreased when the impeller speed increased
to 130 rev/min. Comparatively, the reactor congured with an
optimized impeller obtained a high biogas yield (24.3 L/d) at a lower impeller speed and needed less startup time to reach a steady
state. The peak of biogas yield was reached at an impeller speed
of 70 rev/min, from then onward the biogas yield decreased
acutely along with the increase in impeller speed. From these re-

sults it is clear that the impeller type and speed affect the process
of biohydrogen production. The hydrodynamic behavior of a normal impeller in the speed range between 90 rev/min and
110 rev/min is more suitable for biohydrogen production. By integrating with the previous results of simulation, a qualitative relationship between hydrodynamics and biohydrogen production
can be obtained. Although the uniformity of velocity distribution
in the reactor is improved, along with the increase of impeller
speed, the average yield of hydrogen does not continually increase.
The optimized impeller can better generate a velocity distribution
in the reactor at a lower impeller speed, so higher average hydrogen yield and less startup time are needed compared to a normal
impeller at an impeller speed of 50 rev/min. Shear strain stress
and biogas volume fraction also increase with the impeller speed

Fig. 5. Biogas volume fraction contour of plane 1 (A) normal impeller; and (B) optimized impeller.

J. Ding et al. / Bioresource Technology 101 (2010) 70057013

counteracting biohydrogen production, so the average yield of

hydrogen decreases at higher impeller speed. Shear strain stress
brought by optimized impeller increases intensively along with
the increasing stirring speed leading to low biohydrogen production results. The optimized impeller is not as cost-effective as a
normal impeller for hydrogen production at higher speeds. Our results clearly indicate that the hydrodynamics behavior of a biohydrogen production reactor can affect the chemicalbiological
reactions occurred within and conrm CFD-based software as a
powerful tool for the prediction of ow patterns in the reactor.
4. Conclusions
In this paper, three-dimensional CFD simulations of a gasliquid
two-phase ow in a laboratory-scale CSTR for biohydrogen production have been performed. It has been shown that impeller type
and speed signicantly affects ow patterns, and thus offer different optimal efciencies for biohydrogen production. By integrating
the results of simulations with process experimental observations,
it is clearly shown that an optimized impeller can generate better
velocity distribution in the reactor with lower impeller speed, so
higher average hydrogen yield and less startup time are needed
than a normal impeller when the impeller speed is in the range
from 50 to 70 rev/min.
Acknowledgements
The authors would like to thank the National Natural Science
Fund of China (No. 30870037), the National Natural Science Key
Fund of China (No. 50638020), the Project 50821002 (National
Creative Research Groups) supported by National Nature Science
Foundation of China, the State Key Laboratory of Urban Water
Resource and Environment, the Harbin Institute of Technology
(2008QN02), the Fundamental Research Funds for the Central
Universities (Grant No. HIT.NSRIF.2009115), and the Development
Program for Outstanding Young Teachers in Harbin Institute of
Technology (HITQNJS.2009.046) for their support of this study.
References
American Public Health Association (APHA), 1998. Standards methods for the
examination of water and wastewater, 20th ed., Washington DC, America.
Bhaskar, Y.V., Mohan, S.V., Sarma, P.N., 2008. Effect of substrate loading rate of
chemical wastewater on fermentative biohydrogen production in biolm
congured sequencing batch reactor. Bioresource Technology 99 (15), 69416948.
Cao, G.L., Ren, N.Q., Wang, A.J., Guo, W.Q., Yao, J., Feng, Y.J., Zhao, Q.L., 2010.
Statistical optimization of culture condition for enhanced hydrogen production
by thermoanaerobacterium thermosaccharolyticum W16. Bioresource
Technology 101 (6), 20532058.
Cai, M.L., Liu, J.X., Wei, Y.S., 2004. Enhanced biohydrogen production from sewage
sludge with alkaline pretreatment. Environmental Science and Technology 38
(11), 31953202.
Cao, Y.S., Alaerts, G.J., 1995. Inuence of reactor type and shear stress on aerobic
biolm morphology, population and kinetics. Water Research 29 (1), 107118.
Chang, J.S., Lee, K.S., Lin, P.J., 2002. Biohydrogen production with xed-bed
bioreactors. International Journal of Hydrogen Energy 27(1112), PII S03603199(0302)00130-00131.
Dakshinamoorthy, D., Khopkar, A.R., Louvar, J.F., Ranade, V.V., 2006. CFD simulation
of shortstopping runaway reactions in vessels agitated with impellers and jets.
Journal of Loss Prevention in the Process Industries 19 (6), 570581.
Darmana, D., Deen, N.G., Kuipers, J.A.M., 2005. Detailed modeling of hydrodynamics,
mass transfer and chemical reactions in a bubble column using a discrete
bubble model. Chemical Engineering Science 60 (12), 33833404.
Gavala, H.N., Skiadas, L.V., Ahring, B.K., 2006. Biological hydrogen production in
suspended and attached growth anaerobic reactor systems. International
Journal of Hydrogen Energy 31 (9), 11641175.

7013

Guo, L., Li, X.M., Bo, X., Yang, Q., Zeng, G.M., Liao, D.X., Liu, J.J., 2008a. Impacts of
sterilization, microwave and ultrasonication pretreatment on hydrogen
producing using waste sludge. Bioresource Technology 99 (9),
36513658.
Guo, W.Q., Ren, N.Q., Wang, X.J., Xiang, W.S., Ding, J., You, Y., Liu, B.F., 2009.
Optimization of culture conditions for hydrogen production by ethanoligenens
harbinense B49 using response surface methodology. Bioresource Technology
100 (3), 11921196.
Guo, W.Q., Ren, N.Q., Chen, Z.B., Liu, B.F., Wang, X.J., Xiang, W.S., Ding, J., 2008b.
Simultaneous biohydrogen production and starch wastewater treatment in
an acidogenic expanded granular sludge bed reactor by mixed culture for
long-term operation. International Journal of Hydrogen Energy 33 (24),
73977404.
Hamzehei, M., Rahimzadeh, T., 2009. Experimental and numerical study of
hydrodynamics with heat transfer in a gassolid uidized-bed reactor at
different particle sizes. Industrial and Engineering Chemistry Research 48 (6),
31773186.
Jin, B., Lant, P., 2004. Flow regime, hydrodynamics, oc size distribution and sludge
properties in activated sludge bubble column, air-lift and aerated stirred
reactors. Chemical Engineering Science 59 (12), 23792388.
Kapdan, I.K., Kargi, F., 2006. Biohydrogen production from waste materials. Enzyme
and Microbial Technology 38 (5), 569582.
Kraemer, J.T., Bagley, D.M., 2005. Continuous fermentative hydrogen production
using a two-phase reactor system with recycle. Environmental Science and
Technology 39 (10), 38193825.
Krepper, E., Glovera, G.C., Grahna, A., Weissa, F.P., Altb, S., Hampelb, R., Kstnerb, W.,
Kratzschb, A., Seeligerb, A., 2008. Numerical and experimental investigations for
insulation particle transport phenomena in water ow. Annals of Nuclear
Energy 35 (8), 15641579.
Kulkarni, A.A., Ekambara, K., Joshi, J.B., 2007. On the development of ow pattern in
a bubble column reactor: experiments and CFD. Chemical Engineering Science
62 (4), 10491072.
Lee, K.S., Lo, Y.S., Lo, Y.C., Lin, P.J., Chang, J.S., 2004. Operation strategies for
biohydrogen production with a high-rate anaerobic granular sludge bed
bioreactor. Enzyme and Microbial Technology 35 (67), 605612.
Li, J.Z., Ren, N.Q., Li, B.K., Qin, Z., He, J.G., 2008. Anaerobic biohydrogen production
from monosaccharides by a mixed microbial community culture. Bioresource
Technology 99 (14), 65286537.
Lin, C.Y., Hung, W.C., 2008. Enhancement of fermentative hydrogen/ethanol
production from cellulose using mixed anaerobic cultures. International
Journal of Hydrogen Energy 33 (14), 36603667.
Logan, B.E., Oh, S.E., Kim, I.S., Van Ginkel, S., 2002. Biological hydrogen production
measured in batch anaerobic respirometers. Environmental Science and
Technology 36 (11), 25302535.
Luo, H.P., Muthanna, H.A.D., 2008. Local characteristics of hydrodynamics in draft
tube airlift bioreactor. Chemical Engineering Science 63 (11), 30573068.
Meroney, R.N., Colorado, P.E., 2009. CFD simulation of mechanical draft tube mixing
in anaerobic digester tanks. Water Research 43 (4), 10401050.
Moullec, Y.L., Potier, O., Gentric, C., Leclerc, J.P., 2008. Flow eld and residence time
distribution simulation of a cross-ow gasliquid wastewater treatment reactor
using CFD. Chemical Engineering Science 63 (9), 24362449.
Murthy, B.N., Ghadge, R.S., Joshi, J.B., 2007. CFD simulations of gasliquidsolid
stirred reactor: Prediction of critical impeller speed for solid suspension.
Chemical Engineering Science 62 (24), 71847195.
Nath, K., Das, D., 2004. Improvement of fermentative hydrogen production: various
approaches. Applied Microbiology and Biotechnology 65 (5), 520529.
Pougatch, K., Salcudean, M., Gartshore, I., Pagoria, P., 2007. Computational
modelling of large aerated lagoon hydraulics. Water Research 41 (10), 2109
2116.
Pramparo, L., Pruvost, J., Stber, F., Font, J., Fortuny, A., Fabregat, A., Legentilhomme,
P., Legrand, J., Bengoa, C., 2008. Mixing and hydrodynamics investigation using
CFD in a square-sectioned torus reactor in batch and continuous regimes.
Chemical Engineering Journal 137 (2), 386395.
Ren, N.Q., Li, J.Z., Li, B.K., Wang, Y., Liu, S.R., 2006. Biohydrogen production from
molasses by anaerobic fermentation with a pilot-scale bioreactor system.
International Journal of Hydrogen Energy 31 (15), 21472157.
Ren, N.Q., Wang, B.Z., Huang, J.C., 1997. Ethanol-type fermentation from
carbohydrate in high rate acidogenic reactor. Biotechnology and
Bioengineering 54 (5), 428433.
Terashima, M., Goel, R., Komatsu, K., Yasui, H., Takahashi, H., Li, Y.Y., Noike, T., 2009.
CFD simulation of mixing in anaerobic digesters. Bioresource Technology 100
(7), 22282233.
Vesselinov, H.H., Stephan, B., Uwe, H., Holger, K., Gnther, H., Wilfried, S., 2008. A
study on the two-phase ow in a stirred tank reactor agitated by a gas-inducing
turbine. Chemical Engineering Research and Design 86 (1), 7581.
Wang, X., Ding, J., Ren, N.Q., Liu, B.F., Guo, W.Q., 2009. CFD simulation of an
expanded granular sludge bed (EGSB) reactor for biohydrogen production.
International Journal of Hydrogen Energy 34 (24), 96869695.