You are on page 1of 10

Experimental and numerical investigation of hydrodynamics in raceway

reactors used for algaculture


Rainier Hreiz
a,b,c,
, Bruno Sialve
d
, Jrme Morchain
a,b,c,
, Renaud Escudi
d
, Jean-Philippe Steyer
d
,
Pascal Guiraud
a,b,c
a
Universit de Toulouse; INSA, UPS, INP; LISBP, 135 Avenue de Rangueil, F-31077 Toulouse, France
b
INRA, UMR792 Ingnierie des Systmes Biologiques et des Procds, F-31400 Toulouse, France
c
CNRS, UMR5504, F-31400 Toulouse, France
d
INRA, UR0050, LBE, Avenue des Etangs, F-11100 Narbonne, France
h i g h l i g h t s
The hydrodynamics in a raceway reactor is studied experimentally and numerically.
Effects of the paddlewheel speed and geometry are investigated.
Simulations successfully predict the ow rate and the impeller power consumption.
Mixing rates are related to the ow low-frequency unsteadiness.
Even at moderate intensities, wind has a crucial impact on the hydrodynamics.
a r t i c l e i n f o
Article history:
Received 28 October 2013
Received in revised form 24 February 2014
Accepted 9 March 2014
Available online 19 March 2014
Keywords:
Raceway
Mixing
Tracer pulse injection
CFD
Sliding-mesh
Pulsed Ultrasonic Doppler Velocimetry
a b s t r a c t
Raceways are nowadays the most used large-scale reactors for microalgae culture. This paper focuses on
the hydrodynamics in such a reactor, and emphasizes on the effects of the paddlewheel geometry (two
impeller congurations are tested). The global hydrodynamics behavior of the raceway (ow velocity,
mixing time) is characterized experimentally by the tracer pulse injection method, and local velocity
measurements are acquired by Pulsed Ultrasonic Doppler Velocimetry. Finally, the ow is modeled by
using the sliding-mesh CFD technique, a method overcoming many limitations of the simulation
approaches used in the literature. CFD simulations successfully estimate the ow rate in the reactor
and the power consumption of the paddlewheel. It is shown that the mixing efciency in the raceway
reactor is closely related to the low-frequency ow unsteadiness that arises from the periodic motion
of the blades. Concerning local velocities, CFD and experimental data are in good agreement at some posi-
tions in the reactor, but a signicant disagreement is observed at some other locations. Taking into
account the wind presence in the simulations reduces the discrepancy between the experimental and
numerical results, showing that, even with a moderate intensity, wind has an important effect on the
hydrodynamics in the reactor.
2014 Elsevier B.V. All rights reserved.
1. Introduction
Nowadays, growing concerns about global warming and energy
shortage motivate the society to seek renewable and carbon neu-
tral fuels as alternatives to fossil fuels. Eukaryotic microalgae and
cyanobacteria (to which we will refer collectively by microalgae)
are a potential source of various biofuels [1,2] by virtue of their fast
growth rate, and their oil yields, substantially higher then terres-
trial oil crops, even exceeding 80% of the dry biomass weight under
certain stress conditions [3]. Cultivation of microalgae in wastewa-
ter treatment reactors is even interesting from economical and
synergic perspectives [4,5]. Today cultured microalgae are used
for pharmaceuticals, nutraceuticals, cosmetics, aquaculture and
production of high value molecules (pigments, stable isotope
biochemicals. . .) [6,7].
Microalgae culture systems can be broadly classied into closed
systems where the culture is enclosed into a translucent vessel,
http://dx.doi.org/10.1016/j.cej.2014.03.027
1385-8947/ 2014 Elsevier B.V. All rights reserved.

Corresponding authors at: INSA, LISBP (Laboratoire dIngnierie des Systmes


Biologiques et des Procds), 135 Avenue de Rangueil, F-31077 Toulouse, France.
Tel.: +33 (0)561559774.
E-mail addresses: hreiz.rainier@gmail.com (R. Hreiz), jerome.morchain@
insa-toulouse.fr (J. Morchain).
Chemical Engineering Journal 250 (2014) 230239
Contents lists available at ScienceDirect
Chemical Engineering Journal
j our nal homepage: www. el sevi er . com/ l ocat e/ cej
and open-air systems where the culture medium is directly
exposed to the environment. Open-air cultures are grown outdoors
and utilize sunlight. This is also the case for some closed systems,
while others (referred as photobioreactors) are operated
indoors with articial illumination. Photobioreactors allow higher
productivities [1] since the physicochemical environment of the
culture is well controlled, light intensity and cycles optimized,
and risks of invasion by competing microorganisms minimized.
On the other hand, open-air systems are susceptible to location-
related variables as weather conditions, and the daily and seasonal
variations in light levels and temperature. However, most commer-
cial systems used today are outdoor open-air reactors [7,8]. In fact,
photobioreactors are expensive to purchase, maintain and operate,
difcult to scale up, and suffer from several drawbacks as over-
heating, bio-fouling and toxic oxygen accumulation, while open-
air systems are simpler to design, cheap and easy to build and
operate. Among open culture bioreactors, raceway ponds are by
far the most widely used nowadays [9,10].
A raceway is a pond consisting of long channels arranged in a
racetrack closed loop, in which the algae growth medium is cycled
continuously around the central dividing walls by means of a slow-
moving paddlewheel (Figs. 1 and 2). The pond is kept shallow, typ-
ically 1540 cm, to ensure an adequate penetration of sunlight in
the liquid.
Liquid in the raceway is mechanically circulated at about
0.150.4 m/s, which ensures mixing of the culture medium thanks
to the turbulence of the ow. Although various devices have been
proposed to circulate and mix the culture medium [9,11,12], pad-
dlewheels remain by far the mostly used. In fact, they match well
the pumping requirements of the ponds as they are high ow rate,
low head devices [13]. Moreover, they are mechanically simple and
require little maintenance, and their gentle mixing minimizes
damage to occulated or fragile algae. To minimize backow, the
paddlewheel may sit in an invagination in the pond oor, keeping
a small clearance between the blades tip and the bottom of the
channel: the paddlewheel will then behave more as a positive
displacement pump [13].
Mixing the culture medium is necessary to avoid gradients in
nutrient concentration, temperature and pH. It also prevents mic-
roalgae sedimentation, increases liquid-to-cell mass transfer rates,
and enhances algal photosynthetic efciency. In fact, mixing con-
tinuously moves the algal cells between the well-lit zone (close
to the free interface) and the dark zone (bottom of the pond), what
reduces photolimitation and photoinhibition [1]. In the raceway
straight sections, there are no features that disturb the ow, and
therefore, despite turbulent dispersion, mixing is relatively poor
as shown experimentally by Mendoza et al. [10]. However, near
the paddlewheel and in the bends, dispersion rates are important
Fig. 1. Main dimensions of the raceway reactor used in this study. (a) Top view and positions of the PUDV measurements planes, (b) side view, and (c) side view of the
numerical domain.
R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239 231
[10,11]. Indeed, secondary ows of Prandtls rst kind develop
in the bends, what enhances macro-mixing.
Thanks to mixing, raceways productivity can be up to ten-fold
higher than that of unmixed ponds [6,14]. However, mixing is of
great signicance in terms of input energy costs. Chisti [15] esti-
mated that 28% of the global energy input to the algal oil produc-
tion process is for microalgae cultivation (so mainly for liquid
circulation). According to Neenan et al. [16], mixing accounts for
more than 69% of the utilities cost. Hence, reducing head losses
in the raceway is of primary importance in order to make the
industry competitive, particularly if the culture is dedicated to bio-
fuel production. Thus, one of the greatest challenges in algaculture
is to adequately mix the culture medium with minimum power
consumption.
The raceway design must address in particular the problem of
dead zones (separated ow) that develop near the middle wall
downstream of bends, since they increase energy dissipation and
reduce the pond capacity. Moreover, the resulting non-uniform
velocity eld leads to uneven cells residence time in the reactor,
which is harmful to the pond productivity, especially if the race-
way operates in continuous mode. Different strategies have been
proposed to minimize the dead zones extent, among which install-
ing ow deectors, rectiers or guide vanes, and modifying the
island or the bends design [13,17,18]. Reducing the areas of stagna-
tion straightens the ow, what lowers head losses in the reactor
[10].
In the context of improving the hydrodynamics in bioreactors,
CFD (Computational Fluid Dynamics) is a powerful low-cost tool
that has already proven its efciency with respect to photobioreac-
tors design (see [19] for a review on the topic). In fact, CFD models
the spatio-temporal hydrodynamics variations, and hence, it
allows characterizing/estimating various key variables that are
practically inaccessible to experimental measurements, as local
shear stresses (that above a certain limit lead to shear-induced
injury of algal cells) or cells individual history. For example, Pern-
erNochta and Posten [20] simulated microalgae trajectories in a
tubular photobioreactor to obtain light intensity uctuations as
experienced by individual cells. Pruvost et al. [21] designed an
innovative reactor with CFD support. Despite these successes, it
is only recently that CFD has been applied to raceway ponds, with
the works of Hadiyanto et al. [17] and Liffman et al. [18]. These
works focused on the evaluation of different raceway geometries
in terms of ow uniformity and head losses in the reactor (how-
ever, none of the numerical results was confronted to experimental
data). For simplication purposes, in both papers, the free surface
was considered at (with a slip boundary condition), and the
paddlewheel was not included in the numerical domain. Liffman
et al. [18] modeled the propulsive thrust effect of the paddlewheel
by adding a body force term in the Reynolds equations. This
approach had already been used to simulate the ow in aquacul-
ture ponds [22,23]. A different modeling strategy was used by
Hadiyanto et al. [17]. The section of the channel surrounding the
paddlewheel was removed from the numerical domain, and an
inlet (downstream of the wheel position) and an outlet (upstream
of the wheel position) boundary conditions were used to generate
the ow. These two modeling approaches are interesting, since they
permit low time-consuming calculations, allowing therefore a rapid
evaluation of the performance of various raceway designs. However,
both approaches have some major drawbacks, as they cannot capture
several aspects of the hydrodynamics in the raceway:
They cannot take into account the paddlewheel geometry, a
parameter that however has a crucial effect on the mixing
performances in the reactor. Thus, both methods only consider
the effects of the pond geometry and the liquid ow rate, and
cannot be used to study the impact of the impeller or to
improve its design.
They require experimental measurements to be able to repro-
duce given experimental conditions: the ow rate for the
Liffman et al. [18] approach, and the paddlewheel mechanical
power for the Hadiyanto et al. [17] method (so the additional
term in the Reynolds equations can be calculated).
The Liffman et al. [18] approach fails to reproduce streamlines
straightening immediately downstream of the paddlewheel,
while the Hadiyanto et al. [17] method imposes a uniform ow
at the inlet plane located downstream of the paddlewheel.
In these two studies, the steady-state equations were solved,
and thus, the ow unsteadiness related to the periodic motion
of the blades could not be accounted for. As it will be shown
later, these low-frequency waves have a crucial impact on the
mixing rate in the reactor.
The Liffman et al. [18] approach does not model reliably the
paddlewheel effect. According to their method, the ow rate
in the pond will be determined by the viscous dissipation (that
must balance the paddlewheel power), while in reality, as the
paddlewheel behaves as a positive displacement pump, the ow
rate is only slightly affected by the viscous dissipation (viscous
dissipation however determines the energy demand. The ow
rate depends mainly on the geometry of the pond and the
paddlewheel). In other words, this method imposes the
paddlewheel power consumption, while in reality, it is a result
of the paddlewheel speed set by the operator.
Fig. 2. Geometry of the paddlewheel: schematic representation and CFD modeling.
232 R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239
This paper tries to overcome these limitations through the use
of a more general modeling approach. First, the hydrodynamics
in a semi-industrial sized raceway pond is investigated experimen-
tally. Two paddlewheels geometries are tested, which is novel to
the authors knowledge. The tracer pulse injection method is used
to determine the global hydrodynamics characteristics in the pond
(i.e. the residence time and the degree of mixing) for different
operating conditions. Results demonstrate that the low-frequency
unsteadiness arising from the periodic blades movement has a
major impact on the mixing process in the reactor. Local velocity
measurements are then acquired at different locations in the pond,
using the pulsed Doppler ultrasonic technique. Detailed local
velocity measurements in a raceway have never been reported so
far in the literature (although Chiaramonti et al. [11] conducted
microacoustic Doppler velocimeter measurements, they only
reported qualitative results in their paper). Then, the ow in the
raceway is predicted through a numerical simulation. The paddle-
wheel rotation, the moving free surface as well as wind effects are
explicitly taken into account. This is achieved through the combi-
nation of sliding-mesh and VOF (Volume of Fluid) techniques.
Numerical results under various operating conditions are then con-
fronted to experimental measurements. Power consumption, cir-
culation time and mean velocities are accurately predicted
through the simulation; local velocity proles are more difcult
to reproduce numerically. Taking into account the wind effect
reduces this discrepancy between the experimental and numerical
results at some measurements planes.
2. Materials and methods
2.1. Experimental set-up
2.1.1. The raceway pond
The raceway investigated in this study is a semi-industrial scale
reactor operated outdoors, and located at Narbonne, in the south of
France. The region climate is characterized by high sunlight inten-
sities and powerful winds. The raceway main dimensions are
shown in Fig. 1. The algal culture depth was set to 40 cm during
all the experiments. The pond is plastic lined to reduce head losses
and eliminate water loss by percolation. The ow is circulated by a
1.75 m wide metallic paddlewheel (Fig. 2), with an inner and outer
diameter of 0.33 m and 0.65 m respectively. The impeller is com-
posed of three sections including 5 blades each: this arrangement
allows an angular offset between the paddles of the different sec-
tions. Two paddlewheel congurations were investigated in this
study, with 22.5 (case illustrated in Fig. 2) and without any angu-
lar offset between the different sections. These two arrangements
are referred as non-aligned blades and aligned blades congu-
rations respectively. The paddlewheel is connected to a three-
phase motor through a variable-speed drive (which controls the
rotational speed of the impeller). Its shaft is located 0.84 m above
the pond oor; therefore, the minimum gap between the blades
tip and the pond bottom is 19 cm, which is quite important.
2.1.2. Global hydrodynamics characterization
The tracer experiments are used to determine the global hydro-
dynamic behavior of the reactor, and more precisely, the impact of
both the blades conguration and the paddlewheel speed on the
circulation rate and mixing time. A similar method is used in the
recently published paper of Mendoza et al. [10]. A pulse input of
5 L of salt water (5.5 mol L
1
of NaCl) is added upstreamof the pad-
dlewheel and used as a tracer. The response to the pulse is mea-
sured by a conductivity probe (WTW Multi 3410). Mixing time is
dened as the time required to reach about 95% of the nal stable
value of conductivity. Stimulusresponse experiments were
performed for both paddlewheel congurations, and for three rota-
tional speeds: 3.5, 8 and 12.5 rpm.
Tracer experiments are analyzed using the model of Voncken
et al. [24] (see Fig. 3). The model is suitable for loop reactor and
describes the variation of the tracer concentration at the injection
point assuming a plug ow model with dispersion. The equation
for the scalar transport in a non-dimensional form is:
@C
@h

@C
@s

1
Pe
@
2
C
@s
2
0
where C is the tracer concentration, s the curvilinear abscissa, t the
time, h = t/T
c
the reduced time, with T
c
the circulation time dened
as the ratio between the average length of one turn, L, and the aver-
age velocity U, D the axial dispersion coefcient and Pe the Peclet
Number dened as Pe = U.L/D.
The normalized concentration at the injection point E(h) = C(h)/
C
1
is given by:
Eh
1
2

Pe
ph
r
X
p
j0
exp
Pe
4h
j h
2

Pe and T
c
were identied through a least square method for various
experimental conditions.
2.1.3. Local velocity measurements
The Pulsed Ultrasonic Doppler Velocimetry (PUDV) technique is
used to conduct velocity measurements in the raceway pond.
These experimental measurements will serve to validate our CFD
modeling approach. To the authors knowledge, no detailed local
velocity measurements in raceway reactors have been reported
so far in the literature.
PUDV is a slightly intrusive technique that can perform velocity
measurements in opaque liquids, what makes it particularly suit-
able for algal reactors. A transducer emits a short ultrasonic burst
periodically, then receives the echoes scattered by seed particles
contained in the owing liquid (e.g. microalgae in our experi-
ments). PUDV only measures the velocity component in the direc-
tion of the ultrasonic beam axis. For given operating parameters of
the instrument, there is a maximum measurable velocity, and a
maximum distance from the transducer at which measurements
can be performed. These maximum depth and velocity are inver-
sely proportional, what is the main drawback of the technique.
Details on the PUDV working principles can be found in [25].
Fig. 3. Typical system response to the tracer pulse injection (case of 10 rpm
paddlewheel speed, non-aligned blades conguration). The rounded symbols
correspond to the experimental data, and the continuous line to the Voncken
et al. [24] model.
R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239 233
The PUDV instrument DOP2000 (Signal Processing) was used
in the present investigation. The piezoelectric element of the trans-
ducer is 10 mm diameter, and emits 1 MHz ultrasonic waves.
Experiments were realized during a day when the wind speed
was relatively low. Only the experimental case with the aligned
blades paddlewheel conguration at a rotational speed of
12.5 rpm has been investigated. The PUDV probe was inclined
about 45 from the horizontal (as shown in Fig. 1b) to prevent
ultrasonic reections from the bottom wall from disturbing the
measurements. The probe was oriented in the opposite direction
than the mean ow to minimize the disturbance of the ow in
the measurements zone. To ensure that the transducer remains
within the liquid phase (40 cm height), the probe was placed
35 cm above the pond bottom, apart in sections F, H, L and P where
it was pushed down by additional 5 cm (in these planes, the liquid
height is low near the middle wall because of centrifugal effects).
In fact, due to its very low acoustic impedance, an air layer
between the transducers face and the liquid would virtually reect
all the ultrasonic waves, preventing its penetration into the culture
medium. Measurements were acquired in sections A, B, C, D, F, H, I,
J, L and P (Fig. 1a), at several transverse positions d (0.9 m, 1.3 m,
1.7 m and 2.1 m).
2.2. CFD modeling
Numerical modeling of the ow in the raceway pond is carried
out using the commercial CFD package ANSYS 13. The computa-
tional domain geometry is constructed using Design Modeler
(Fig. 2). The sliding-mesh technique is used to account for the pad-
dlewheel rotation. Unlike the approaches used in the literature,
this method actually represents the movement of the entire pad-
dlewheel. Since the paddlewheel is not entirely submerged, the
computational domain was extended up to 2 m from the ground
level in order to fully encompass the impeller. Therefore, it consists
in the raceway itself and a volume of air above the water such that
the entire paddlewheel is included inside (Fig. 1c). The sliding-
mesh technique requires dividing the domain in two zones: a
cylindrical zone encompassing the paddlewheel (rotating-mesh
zone), and the rest of the computational domain (immobile zone).
The two zones slide past each other at a cylindrical interface. The
sliding-mesh method does not require a prior knowledge of any
of the ow characteristics: the reactor geometry and the impellers
rotation speed are the only needed inputs. Regarding the modeling
of the paddlewheel itself, neither the paddlewheel shaft nor the
plates connecting the paddles were included in the numerical
geometry. This simplication results in the geometry presented
in (Fig. 2): only the blades, modeled as zero-thickness walls (given
that their thickness is about 2 mm only) are considered. Although
they seem to oat in the domain, the movement of the blades is
prescribed by the rotation of the cylindrical zone. As the cylindrical
zone is rotating, the blades actually turn around a virtual axis (not
represented) corresponding to the shaft of the real paddlewheel.
The main drawback of the sliding-mesh method is that it is
computationally expensive, since the blades position is updated
at each time step (thus, transient calculations are performed).
Therefore, to estimate the mesh size achieving grid-independent
results, preliminary tests were conducted using the Hadiyanto
et al. [17] method (i.e. the paddlewheel section is removed from
the computational domain, and the ow is circulated using inlet
and outlet boundary conditions). Based on these results, the
computational domain was nally discretized using a multi-bloc
structured mesh of 310,000 cells (mainly hexahedral non-uniform
elements). Grid rening is adopted close to the pond walls to allow
the y + values to be in the correct range (30300, since wall
functions are used). A ne mesh is used in high gradients zones
(e.g. bends intrados, zone around the paddlewheel), while larger
cells are used in the gas phase domain and in the straight areas
of the pond.
The Fluent code is used to resolve the Reynolds Averaged
NavierStokes equations in transient mode (also called Unsteady
or U-RANS simulation). Turbulence is modeled through the high-
Reynolds realizable ke turbulence model. The free interface is
tracked using the Volume of Fluid (VOF) technique (via the implicit
compressive scheme). The objective is to be able to reect changes
in liquid height near the paddlewheel as well as wind effect rather
than predicting the free surface shape and wave motion accurately.
The convective terms are discretized using the QUICK scheme, and
the diffusion terms are central-differenced. Simulations were per-
formed for both aligned and non-aligned blades congurations, for
a paddlewheel speed of 12.5 rpm. Advancement in time is done
through a second-order implicit scheme with a 0.009 s time step,
until the permanent ow regime is reached. From that point
onward, instantaneous values are stored and further averaged to
produce mean values to compare with experimental data.
3. Results and discussion
3.1. Tracer experiments
Fig. 3 shows a comparison between experimental data (non-
aligned blades conguration, paddlewheel speed of 10 rpm) and
the Voncken et al. [24] model prediction. A good agreement was
observed between model and tracer concentration for all the
experimental conditions tested. By using the calibration parame-
ters of the model, it is possible to extract the Peclet number (Pe)
and the circulation time (T
c
) for each conditions. The liquid bulk
velocity and the dispersion coefcient versus the paddlewheel
velocity are plotted in Figs. 4 and 5 respectively, for both paddle-
wheel congurations. Note that 2 experiments were run for each
experimental condition (Table 1): discrepancy between their
results certainly includes wind effects.
Fig. 4 shows a linear relation between the ow rate and the pad-
dlewheel speed, indicating that the paddlewheel behaves like a
positive displacement pump. It is however possible that this linear
behavior slightly changes for higher liquid ow rates, as the liquid
height at the intake of the paddlewheel decreases with increasing
Fig. 4. The ow bulk velocity versus the paddlewheel speed.
234 R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239
viscous dissipation. As they sweep the same liquid volume (gray
area in Fig. 1b), for the same speed, both paddlewheel congura-
tions generate similar liquid ow rates.
Fig. 5 and Table 2 show that the mixing rate is enhanced when
the liquid ow rate increases, as expected. However, a very
interesting point is revealed by these data: the aligned blades
conguration is quite superior to the non-aligned blades arrange-
ment in terms of mixing performances (this superiority gets more
pronounced as the liquid ow rate increases), and this for a slightly
higher power demand only (Table 2). The reason behind this
observation has to be elucidated thanks to CFD simulations. A more
efcient paddlewheel allows a great reduction in operating costs as
mixing can be performed with a lower energy input. Note that the
non-aligned blades conguration is recommended in the literature,
in order to reduce pulsations on the drive train and in the water
ow [13].
3.2. CFD results and comparison with experimental data
3.2.1. Aligned-blades conguration: experimental versus numerical
results
Fig. 6 shows CFD predictions of the mean velocity magnitude at
0.3 m above the ground level, for the case where the blades are
aligned and the paddlewheel speed is 12.5 rpm. The ow acceler-
ates near the intrados of the bends, and decelerates near the extra-
back. In accordance with preliminary visualizations, dead zones
develop downstream of the bends what results in a non-uniform
ow in a cross-section. Flow streamlines are greatly straightened
after the ow passage through the paddlewheel, and the velocities
get more uniform. This effect cannot be simulated by any of the
modeling approaches used in the literature. Fig. 7 shows the prole
of |V
Z
|
avg
along the raceway as calculated by CFD. s represents the
curvilinear abscissa (nil at measurements section A), |V
Z
|
avg
the
absolute value of the mean vertical velocity, averaged over a verti-
cal section. |V
Z
|
avg
is strongly correlated with the vertical mixing in
the reactor, which ensures that the microalgae are intermittently
exposed to light. According to CFD results, the vertical velocity
component is one or two order of magnitude smaller than the bulk
liquid velocity. Vertical mixing is more pronounced near the pad-
dlewheel due to the movement generated by the blades passage,
and in the bends, due to secondary ows. These results are in
agreement with the experimental measurements of Mendoza
et al. [10].
Figs. 8 and 9 show a comparison between experimental and
numerical results, along the raceway at 0.18 m above the ground
level, and at different measurements sections, respectively. The
Fig. 5. The dispersion coefcient versus the paddlewheel speed.
Table 1
Experimental conditions and identied parameters for tracer experiments.
Experimental conditions Identied parameters
Paddlewheel geometry Paddlewheel speed (rpm) T
c
(s) D (m
2
s
1
) U (m s
1
) Pe
Aligned blades 3.5 324 0.072 0.091 37
3.5 321 0.043 0.092 63
Non-aligned blades 3.5 349 0.047 0.084 53
3.5 354 0.045 0.083 55
Aligned blades 8 143 0.121 0.207 51
8 144 0.143 0.205 42
Non-aligned blades 8 140 0.076 0.210 82
8 140 0.095 0.211 65
Aligned blades 12.5 93 0.180 0.319 52
12.5 89 0.178 0.333 55
Non-aligned blades 12.5 88 0.121 0.337 82
12.5 88 0.117 0.334 84
Table 2
Experimental conditions and the corresponding mixing time and electrical power consumption.
Experimental conditions Experimental results
Paddlewheel geometry Paddlewheel speed (rpm) Mixing time (s) Electrical power consumption (W)
Aligned blades 3.5 2000 95
Non-aligned blades 3.5 >2300 94
Aligned blades 8 800 137
Non-aligned blades 3.5 1500 128
Aligned blades 12.5 500 216
Non-aligned blades 12.5 1000 201
R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239 235
coordinate u represents the position along the ultrasonic beam axis
(which is inclined 45 from the horizontal. The reference value,
u = 0, corresponds to the ground level.), V
u
the mean velocity com-
ponent along the ultrasonic beam axis (positive velocities repre-
sent velocities pointing in the same direction as the mean ow,
negative ones represent reverse ow). Each graph or prole is ref-
erenced by a value d representing the transverse distance from the
middle of the central divider to the measurements location (see
Fig. 1). This point to point comparison against experimental data
is a severe test for the simulation and it must be outlined here that
the experimental positioning of the probe can never be as stable
and accurate as the numerical one. Despite the mismatch between
experimental and numerical results at some measurements planes,
one observes that owing to the U-RANS simulation, the general fea-
tures of the ow (Fig. 8), as well as local proles are predicted with
a fairly acceptable accuracy (Fig. 9). The simulation underestimates
the width of the dead zones (see for example the results at section
H in Fig. 9), and as the raceway is an endless loop, this error prop-
agates and alters the numerical results in the whole domain. The
discrepancy between numerical and experimental results can arise
from an inaccurate calculation of the wall shear stresses (by the
classical law of the wall) or from the used turbulence model, but
it certainly includes the wind effects as it will be discussed later.
CFD predicts a ow rate of 360 kg/s, i.e. a liquid bulk velocity of
about 0.47 m/s. Surprisingly, CFD results are in good agreement
with local velocity measurements, but the calculated ow mean
velocity is quite different from that measured by the tracer exper-
iments, i.e. 0.32 m/s. This indicates an important difference
between the tracer and the PUDV results. To better understand
the reason behind this disagreement between the experimental
data, a new CFD simulation was run, this time using a paddlewheel
external radius of 0.585 m instead of 0.65 m (10% lower). The goal
is to study the sensibility of the liquid ow rate relative to
uncertainties on the paddlewheel dimensions or on the liquid
height in the raceway. The ow rate generated by the smaller
paddlewheel is about 263 kg/s, i.e. a circulation velocity of
0.34 m/s. This great variation in the circulation velocity is due to
the low submersion of the blades, which make the ow rate very
sensitive to the liquid height. In fact, the liquid volume swept by
the blades is equal to the paddlewheel width, times the gray area
shown in Fig. 1b. The blade-sweeping area is given by:
Area R
2
cos
1
R h
R

R h

2Rh h
2
q
with R the paddlewheel radius, and h the maximum height that
blades submerge in the liquid.
If the paddlewheel radius is reduced from 0.65 m to 0.585 m,
the blade-sweeping area decreases from 0.139 m
2
to about
0.076 m
2
, what explains this important variation in the generated
ow rate. Thus, we think that the liquid height was actually lower
than 0.4 m during the tracer experiments. Another possibility is
that the shaft vertical position has changed between the two series
of experiments, as the paddlewheel has been taken off then reas-
sembled after the tracer experiments.
CFD shows a non-linear relationship between the blade-sweep-
ing area and the ow rate in the raceway. This effect is due to the
low submersion of the paddles which leaves an important gap
between the blades tip and the pond bottom. In fact, the global
paddlewheel action deviates from the positive displacement pump
behavior when this gap increases, since a greater part of the liquid
ow rate ows due to its inertia from below the impeller (i.e.
without being directly pushed by the paddlewheel).
3.2.2. Aligned versus non-aligned-blades conguration: numerical
results
The numerical simulation with the non-aligned blades congu-
ration predicts a mass ow rate of 350 kg/s, which is substantially
similar to the 360 kg/s generated by the aligned blades congura-
tion. This result is in agreement with experiments, which showed
that the ow rate is independent of the paddlewheel conguration
(Fig. 4). Concerning energy demand, the calculated power con-
sumption (torque exerted on the paddlewheel times the impeller
rotational speed) for the aligned and non-aligned congurations
are respectively 229 W and 198 W, what is in an excellent agree-
ment with experimental measurements (Table 2): simulations sat-
isfyingly predict a slightly higher power demand for the aligned
blades conguration. Thus, the sliding mesh technique is a power-
ful tool that successfully takes into account the effect of the pad-
dlewheel geometry on the ow rate and the power consumption.
Fig. 6. Colored contour plot of the mean velocity magnitude at 0.3 m above the ground level (numerical results): aligned-blades conguration, paddlewheel speed at
12.5 rpm.
Fig. 7. Prole of |V
Z
|
avg
along the raceway (numerical results).
236 R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239
Fig. 8. Numerical versus experimental velocities at 0.18 m above the ground level. The star symbols represent experimental measurements and the rounded symbols with a
continuous line represent numerical results.
Fig. 9. Numerical versus experimental velocities at different measurements planes. Black rounded symbols, red stars, green crosses and blue diamonds represent
experimental measurements at d = 0.9 m, d = 1.3 m, d = 1.7 m and d = 2.1 m respectively. The continuous black line, the dashed red line, the green dashed dotted line and the
dotted blue line represent CFD results at d = 0.9 m, d = 1.3 m, d = 1.7 m and d = 2.1 m respectively. (For interpretation of the references to colour in this gure legend, the
reader is referred to the web version of this article.)
R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239 237
Concerning the ow eld, according to CFD results, the paddle-
wheel conguration has only a slight inuence on the mean ow
hydrodynamics: the mean velocity eld is nearly the same for both
paddlewheel congurations (comparison not shown here). How-
ever, this is not at all the case of the low-frequency unsteady
effects. As can be seen from Fig. 10, the torque exerted on the pad-
dlewheel exhibit much higher amplitude variations when the
aligned blades arrangement is used (the dominant frequency is
the blades passage frequency: 1.04 Hz). The same effect is
observed concerning the ow rate (however, the oscillations
amplitude decreases while going downstream of the paddlewheel).
In fact, the aligned blades conguration results in a non-smooth
ow generation conditions: the paddles of the three sections of
the paddlewheel advance simultaneously, pushing forward a large
amount of water all at once. It is this effect that enhances the
mixing rate in the reactor, by increasing both macro (convection)
and micro mixing (turbulence).
3.2.3. Wind effect
Wind effect can be a potential cause of the discrepancy between
numerical and experimental results in the raceway reactor. The
experimental facility operates outdoors, thus experimental veloc-
ity measurements are inuenced by both wind direction and inten-
sity. Thereby, an additional simulation accounting for the wind
presence was run. Wind enters the computational domain through
an inlet boundary condition. Its intensity was taken as 20 km/h,
and its direction from left to right on Fig. 1a. These are only rough
estimations of the wind characteristics when PUDV measurements
were conducted as no precise measurements are available.
Compared to the previous simulation (Fig. 9), taking into
account the wind presence considerably improved the numerical
results at some measurement sections (Fig. 11). CFD results were
however altered at some other measurement planes. It must be
kept in mind that wind intensity and direction continuously vary
with time, which is very difcult to reproduce numerically. Thus,
the goal of this last simulation is not to completely reconcile exper-
imental and numerical results, but to demonstrate that even at a
moderate intensity, wind has a considerable effect on the hydrody-
namics in the reactor. It can be expected that its effect would be
more pronounced if the raceway reactor is longer, if the circulation
velocity is decreased, or obviously, if the wind intensity is higher.
This last case is very difcult to simulate with the sliding mesh
and VOF techniques. In fact, for high wind intensities, the free inter-
face gets wavy, and the airliquid momentum transfer gets closely
related to the interfacial roughness (ripples and low-amplitude
Fig. 10. Temporal variation of the torque exerted on the paddlewheel (permanent
regime), for both blades arrangements (numerical results).
Fig. 11. Numerical versus experimental velocities at different measurements planes: case when the wind effect is accounted for. Black rounded symbols, red stars, green
crosses and blue diamonds represent experimental measurements at d = 0.9 m, d = 1.3 m, d = 1.7 m and d = 2.1 m respectively. The continuous black line, the dashed red line,
the green dashed dotted line and the dotted blue line represent CFD results at d = 0.9 m, d = 1.3 m, d = 1.7 m and d = 2.1 m respectively. (For interpretation of the references to
colour in this gure legend, the reader is referred to the web version of this article.)
238 R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239
waves). Simulating such small deformations requires a prohibitive
calculation time.
The results of this last simulation emphasize the need for future
experiments on a pilot scale raceway operating indoors, where the
wind characteristics can be controlled. Understanding the wind
impact on the hydrodynamics in the raceway is crucial to choose
an adequate orientation of the pond with regard to the prevailing
winds, so as to take advantage of the wind to improve the mixing
efciency in the reactor (and so decrease the energy consumption).
4. Conclusion
The hydrodynamics in the raceway is critical to obtain high
microalgae productivity, and CFD is a powerful tool of assistance
in the reactor design. This paper focused on a numerical and exper-
imental investigation of the hydrodynamics in such a pond. Two
paddlewheel congurations have been tested in the present study.
Results demonstrate that the blades arrangement do not inu-
ence the generated ow rate, but however, has a crucial impact
on the mixing rate. Contrary to the CFD approaches previously pre-
sented in the literature, the sliding mesh technique used here cap-
tures the low-frequency motion induced by the paddlewheel. It is
found that the mixing process is strongly correlated to the ow
unsteadiness resulting from the periodic passage of the blades.
The numerical simulations calculated accurately the ow rate and
the pumping force in the pond, for both paddlewheel congura-
tions. Numerical results were in good agreement with velocity
measurements at some locations in the reactor, but a signicant
discrepancy existed at some other locations. Taking into account
the wind effect in the simulations improved the concordance
between experimental and numerical results at some measurement
planes. The result demonstrated that even at a moderate intensity,
wind has an important effect on the hydrodynamics of a raceway.
Acknowledgment
The authors are grateful to OSEO FUI Salinalgue for its nancal
support.
References
[1] Y. Chisti, Biodiesel from microalgae, Biotechnol. Adv. 25 (2007) 294306.
[2] R.J. Craggs, S. Heubeck, T.J. Lundquist, J.R. Benemann, Algae biofuel from
wastewater treatment high rate algal ponds, Water Sci. Technol. 63 (2011)
660665.
[3] F.B. Metting, Biodiversity and application of microalgae, J. Ind. Microbiol. 17
(1996) 477489.
[4] J.B.K. Park, R.J. Craggs, A.N. Shilton, Wastewater treatment high rate algal
ponds for biofuel production, Bioresour. Technol. 102 (2011) 3542.
[5] L. Christenson, R. Sims, Production and harvesting of microalgae for
wastewater treatment, biofuels, and bioproducts, Biotechnol. Adv. 29 (2011)
686702.
[6] J.R. Benemann, W.J. Oswald, Systems and economic analysis of microalgae
ponds for conversion of CO
2
to biomass, Final Report, Subcontract XK 4-04136-
06, Pittsburgh Energy Technology Center Grant No. DE-FG22-93PC93204,
1996.
[7] P. Spolaore, C. Joannis-Cassan, E. Duran, A. Isambert, Commercial applications
of microalgae, J. Biosci. Bioeng. 101 (2006) 8796.
[8] M.A. Borowitzka, Commercial production of microalgae: ponds, tanks, tubes
and fermenters, J. Biotechnol. 70 (1999) 313321.
[9] M.A. Borowitzka, N.R. Moheimani, Algae for Biofuels and Energy, Springer,
2013.
[10] J.L. Mendoza, M.R. Granados, I. De Godos, F.G. Acin, E. Molina, C. Banks, S.
Heaven, Fluid-dynamic characterization of real-scale raceway reactors for
microalgae production, Biomass Bioenergy 54 (2013) 267275.
[11] D. Chiaramonti, M. Prussi, D. Casini, M.R. Tredici, L. Rodol, N. Bassi, G.C.
Zittelli, P. Bondioli, Review of energy balance in raceway ponds for microalgae
cultivation: Re-thinking a traditional system is possible, Appl. Energy 102
(2013) 101111.
[12] B. Ketheesan, N. Nirmalakhandan, Development of a new airlift-driven
raceway reactor for algal cultivation, Appl. Energy 88 (2011) 33703376.
[13] J.C. Weissman, J.P. Goebel, Design and analysis of microalgal open pond
systems for the purpose of producing fuels, Subcontract report for the U.S.
department of energy, 1987.
[14] A. Darzins, P.T. Pienkos, L. Edye, Current status and potential for algal biofuels
production, Report for IEA Bioenergy Task 39, 2010.
[15] Y. Chisti, Response to Reijnders: do biofuels from microalgae beat biofuels
from terrestrial plants? (letters response), Trends Biotechnol. (2008) 351
352.
[16] B. Neenan, D. Feinberg, A. Hill, R. McIntosh, K. Terry, Fuels from Micro-algae:
Technology Status, Potential and Research Requirements, SERI/SP-231-2550,
Solar Energy Research Institute, Golden, Colo., 1986.
[17] H. Hadiyanto, S. Elmore, T. Van Gerven, A. Stankiewicz, Hydrodynamic
evaluations in high rate algae pond (HRAP) design, Chem. Eng. J. 217 (2013)
231239.
[18] K. Liffman, D.A. Paterson, P. Liovic, P. Bandopadhayay, Comparing the energy
efciency of different high rate algal raceway pond designs using
computational uid dynamics, Chem. Eng. Res. Des. 91 (2013) 221226.
[19] J.P. Bitog, I.-B. Lee, C.-G. Lee, K.-S. Kim, H.-S. Hwang, S.-W. Hong, I.-H. Seo, K.-S.
Kwon, E. Mostafa, Application of computational uid dynamics for modeling
and designing photobioreactors for microalgae production: a review, Comput.
Electron. Agricult. 76 (2011) 131147.
[20] I. Perner-Nochta, C. Posten, Simulations of light intensity variation in photo-
bioreactors, J. Biotechnol. 131 (2007) 276285.
[21] J. Pruvost, L. Pottier, J. Legrand, Numerical investigation of hydrodynamic and
mixing conditions in a torus photobioreactor, Chem. Eng. Sci. 61 (2006) 4476
4489.
[22] Y.H. Kang, M.O. Lee, S.D. Choi, Y.-S. Sin, 2-D hydrodynamic model simulating
paddlewheel-driven circulation in rectangular shrimp culture ponds,
Aquaculture 231 (2004) 163179.
[23] E.L. Peterson, J.A. Harris, L.C. Wadhwa, CFD modelling pond dynamic
processes, Aquacultural Eng. 23 (2000) 6193.
[24] R.M. Voncken, G.D. Holmes, H.W. Den Hartog, Fluid ow in turbine-stirred,
Bafed Tanks-II, Chem. Eng. Sci. 19 (1964) 209213.
[25] M. Messer, Pulsed ultrasonic Doppler velocimetry for measurement of velocity
proles in small channels and capillaries, Masters thesis, The Georgia Institute
of Technology, 2005.
R. Hreiz et al. / Chemical Engineering Journal 250 (2014) 230239 239

You might also like