Introduction

Groundwater is a key resource of potable
water. Soluble ferrous iron is probably the most common contaminant of
groundwater: It causes unpleasant taste in mouth and anesthetic red/brown
stains on clothes and sanitary facilities; it promotes the build-up of deposits
that can constrict pipeline and favors the development of iron bacteria that
cause corrosion of equipment and make the taste and odor issues worse.
Consequently, guidelines for soluble iron are between 0.1 and 0.2 mg/L. In
practice, current iron(II) concentrations in groundwater lie about 0.7 mg/L,
but values about 80 mg/L have already been reported.

Iron removal is a conventional operation of
water treatment that can be conducted using either a physicochemical or a
biological process, each of them with several variants. The common
physicochemical pathway consists in an oxidation-precipitation process,
followed by either filtration or decantation: Soluble iron(II) is oxidized into
insoluble iron(III) hydroxides, especially ferric hydroxide Fe(OH)₃.
Among the variants, aeration is usually applied for oxidizing ferrous iron in
water exhibiting iron(II) concentrations higher than 5 mg/L because 1 mg
dissolved oxygen is needed to oxidize 7 mg iron. In this case, the biological
process may also need aeration and it has been already shown that biological
and physicochemical mechanisms usually proceed in parallel. As a result, an
accurate description of the kinetics of physicochemical iron(II) removal is
required for both pathways. A key issue is that this kinetics is highly
sensitive to pH, alkalinity and ionic strength; in addition, an autocatalytic
effect of ferric hydroxide has also been highlighted. Consequently, iron
removal can sometimes be faster than expected, which results in oversized
designs, but opposite trends have also been reported.

In the last decade,
airlift reactors have been shown to be versatile tools for iron(II) removal in
groundwater because
they combine good mixing, mass transfer and pH control properties. Now, the
objective is, therefore, to develop a multi-scale modeling approach able to
take hydrodynamics, mass transfer and pH change into account, so that robust
scale-up can be achieved.   

Materials and methods

A multi-scale modeling strategy was
developed using simultaneously 1D, 2D and 3D modeling and simulation tools that
accounts for local and global hydrodynamic, mass transfer and reaction
features. The experimental data from El Azher et al. (2005, 2008) was
used to calibrate the 1D model. In these works, experiments were carried out in
a split-rectangular internal airlift reactor of 63L, consisting of a square
column (0.20m×0.20m) of 2m height, divided equally into a riser and a downcomer
section by a baffle. Available data describes the respective influence of
superficial gas velocity from 0.0085 to 0.068 m/s and of water properties on
hydrodynamics, mass transfer and reaction using real drinking water spiked with
various amounts of soluble ferrous iron and insoluble iron hydroxide. Water
properties were varied as a function of pH and of the concentrations of soluble
iron(II) and insoluble iron(III) hydroxide. Constant pH was maintained using CO₂
injection coupled with air supply. Due to the presence of black solid particles
of ferric hydroxide that induced turbidity and color, local hydrodynamics and
bubble size could not be measured and hydrodynamic data only consisted of
average gas hold-up in both sections, and overall liquid recirculation
velocity. For mass transfer, only the average k_La on the reactor was
determined.

The modeling strategy consisted, first, of
a 1D model adapted from the approach developed by Talvy et al. (2005).
This involves mass and momentum equations and is based on the drift flux model both in the riser and
in the downcomer, extended by the Higbie penetration model to include mass
transfer and by the autocatalytic kinetics of iron removal to account for the
chemical reaction. Matlab® was used to solve the resulting set of ordinary
differential equations and for estimating the three adjustable variables that
are, namely, the distribution parameter of the drift flux model in the riser,
the average bubble size derived from the interphase slip velocity and the singular
pressure drop coefficient in the airlift reactor. Then, 2D and 3D modeling was
applied to simulate the local hydrodynamics, mass transfer and kinetics of iron
removal in the airlift reactor using a commercial computational fluid dynamics
(CFD) software, Ansys Fluent®. A two-fluid
model
based on the Reynolds-averaged Navier?Stokes equations in
two-phase flows was retained. Mass transfer and chemical reaction kinetics were
implemented using user-defined functions.

Results and discussion

First, the simulations from the 1D model
were shown to fit very accurately experimental data for hydrodynamics and oxygen
mass transfer by adjusting the distribution parameter about 1.5, the singular
pressure drop coefficient at 10.8 and the interphase slip velocity at 0.24 m/s,
which corresponded to an average bubble diameter about 4-5 mm if an ellipsoidal shape was assumed (see the figure, below). In particular, this approach was
able to predict that bubble recirculation in the dowcomer started when
superficial gas velocity was higher than 0.02 cm/s, when the overall liquid
recirculation velocity reached a plateau value. The average bubble diameter was
also in agreement with experimental k_La data. Contrary to Talvy et
al. (2007), oxygen depletion in the gas phase remained insignificant both
in the riser and the dowcomer. Consequently, the airlift reactor was closer to
the assumption of perfect mixing. This may result from the larger average
diameter, especially at low superficial gas velocity, which results from the
single-orifice nozzle used in this work for gas dispersion: The consequences
are a higher driving force for liquid recirculation and a reduction of bubble
stagnation in the downcomer, which favors the renewal of the bubbles in this
section.

Then, 2D numerical
simulations were conducted using the average bubble diameter deduced from 1D
simulations. The objective was to define the closure laws of the two-fluid
model using inexpensive calculations able to account for hydrodynamics, mass
transfer and reactions. A comparison with experimental data and 1D simulations
showed that a dispersed-phase k-epsilon turbulence model should be
preferred to predict accurately global hydrodynamics. The results showed that
turbulent dispersion force and bubble-induced turbulence formulation were of
primary importance, which stems from the single-orifice nozzle used for gas
dispersion. Conversely, the comparison between standard and RNG k-epsilon
turbulence modeling showed only minor discrepancies, the effect of lift forces
was negligible in comparison to dispersion forces. Finally, 2D simulations
predicted correctly the gas hold-up in the riser, but over-predicted the gas
hold-up in the downcomer, the overall liquid recirculation velocity and mass
transfer. Actually, 2D simulations are unable to describe accurately the gas
dispersion due to the single-orifice nozzle, which leads to a velocity profile
in the riser that peaks close the internal baffle. However, they are able to
provide rapidly and effectively the influence of modeling assumptions that have
been applied, then, to full 3D CFD simulations. These have been shown to better
predict global hydrodynamic parameters, including gas hold-up in the two
sections,
overall liquid circulation velocity, and overall transfer of oxygen. The local
velocity and gas hold-up profiles in the riser strongly differed from 2D
simulations because the bubble plume could oscillate tangentially to the
baffle. In addition, the distribution parameter of the drift flux was shown to
be close to 1.5 in the riser and to 1.1 in the downcomer, which assesses the
modeling assumptions on hydrodynamics in 1D simulations. Examples of local gas
hold-up distribution and of oxygen dispersion are shown in the figure below.

For iron(II)
removal, reaction was slow and controlled by chemical kinetics when pH was
lower than 7.5 and initial iron content lower than 20 mg/L and the assumption
of perfectly mixed reactor was confirmed by 3D CFD data. Conversely, when pH
was higher than 7.5 or when ferric hydroxide was added to enhance the
autocatalytic effect, the reaction proceeded in the intermediate regime,
leading to local heterogeneities
of dissolved oxygen concentration and reaction rate that could not be
accounted by 1D simulations, especially between the riser and the
downcomer.    

Conclusions

Iron(II) removal by aeration in a
split-rectangular airlift reactor was simulated using a multi-scale modeling
strategy that consisted in deducing the parameters that could not be measured
using 1D simulations, in using these parameters in 2D CFD simulations so as to
define the closure laws of the two-fluid model, and finally in confronting 1D
and 3D simulations to experimental data. Contrary to expectations, using a
single-orifice nozzle for gas dispersion did not restrict the accuracy of 1D
simulations for describing hydrodynamics and mass transfer. On the contrary,
this reduced the discrepancies between the riser and the downcomer when
reaction was slow. But this multi-scale strategy also presents the advantage to
highlight the development of local heterogeneities when reaction kinetics
becomes as fast as mass transfer and, in this case, to overcome the limitations
of 1D models for scale-up purpose.

References

El
Azher, N., B.Gourich, C.Vial, M.Belhaj Soulami, A.Bouzidi, M.Ziyad (2005).
Biochem. Eng. J. 23, 161?167.

El Azher, N., B.Gourich, C.Vial, M.Belhaj
Soulami, M.Ziyad (2008). Chem. Eng. Process. 47, 1877?1886.

Talvy, S., A.Cockx, A.Liné (2005). Chem.
Eng. Sci. 60, 5991?6003.

Talvy, S., A.Cockx, A.Liné (2007). AIChE J. 53,
316?326.