UV Photoreactor

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Environ. Sci. Technol.

2010, 44, 2056–2063

the chemicals (photochemical/chemical kinetics), and the


Simulation of UV Photoreactor for interaction of these phenomena with one another. The UV
Degradation of Chemical reactor performance model (semimechanistic approach)
consists of a system of partial differential equations governing
Contaminants: Model Development various phenomena in the UV reactor that should be solved
simultaneously. Due to the complexity of the reactor
and Evaluation geometry and interaction of many phenomena, analytical
solutions do not exist, and a numerical solution is the only
SIAMAK ELYASI* AND practical approach, using relevant techniques such as
FARIBORZ TAGHIPOUR computational fluid dynamics (CFD).
Chemical and Biological Engineering Department, University The modeling results for reactor performance should be
of British Columbia, 2360 East Mall, Vancouver, validated against experimental data (hydrodynamics, radia-
BC V6T 1Z3, Canada tion, and species concentration) to ensure their applicability
to reactor design and optimization. A number of researchers
Received August 13, 2009. Revised manuscript received have developed models of photoreactor performance (e.g.,
January 31, 2010. Accepted February 2, 2010. see refs 2-7); however, they have used conventional methods
for model evaluation. These methods compare the modeling
predictions with the experimental results for the overall
reactor performance, which is determined by comparing the
Models simulating the performance of UV reactors enhance ratio of the outlet to the inlet concentration of the reactor.
our understanding of the fundamental principles governing the This method does not reveal any discrepancies that might
operation of these units. When modeling the performance of exist between the model predictions and the experimental
UV reactors, governing equations for all related phenomena are values for velocity, radiant energy, and concentration of
derived and solved. This research presents a step-by-step species inside the photoreactors. Some other researchers
(8, 9) have compared the hydrodynamics from the reactor
methodology to setup and solve the governing equations
performance model with experimental velocity data inside
determining the performance of UV reactors and to evaluate the photoreactor obtained using particle image velocimetery
the results. A computational fluid dynamic (CFD) model (PIV). But, in these cases, the CFD model predictions have
was developed in order to simulate UV photoreactors in the been typically validated using PIV experimental data for only
Eulerian framework for chemical removal using a UV- a few strategic locations inside the photoreactors, instead of
based hydroxyl radical initiated oxidation process. Verifying throughout the entire reactor domain.
the results of the integrated CFD model, a novel method was In this research, a general methodology is presented for
developed using a modified planar laser-induced fluorescence developing an integrated CFD model of UV reactor perfor-
technique for measuring tracer concentration profiles inside mance and evaluating all parts of the model separately. The
the UV reactor. In addition, the components of the CFD results of each part of the integrated model (hydrodynamic,
modelshydrodynamics and radiationswere evaluated using radiation, species concentration) are evaluated separately
experimental profile throughout the entire reactor. This verified against experimental values in a UV photoreactor in order
procedure can be applied to the simulation and optimization to develop a more practical and reliable integrated model.
of UV photoreactors with various geometries and operating The approach can be summarized as follows:
• The simulated velocity field is compared with PIV
conditions.
measurements at every point, and velocity vector
1. Introduction profiles for the entire reactor cross-section are
presented.
UV technology, in combination with an oxidant such as • The radiation field is modeled and compared with
hydrogen peroxide, provides an effective method of water measured values, which are recorded using a photo-
and wastewater treatment. This photoinitiated oxidation diode sensor located at discrete positions throughout
process (also referred to as the UV-based advanced oxidation for the entire domain of the reactor.
process, AOP) can be applied when treating optically clear • The photoreaction rate of a chemical used as a model
water or wastewater that contains contaminants with con- contaminant for the UV-advanced oxidation process is
centrations of less than 1000 ppm (1). In UV-based AOP, measured in a bench-scale photoreactor under con-
hydroxyl radicals which are generated through the reaction trolled conditions to obtain a photoreaction kinetic
of UV radiation with hydrogen peroxide play a major role in model.
oxidizing persistent chemicals. A mathematical model simu- • An integrated model of hydrodynamics, radiation,
lating the performance of UV reactors can contribute to our and conservation of a chemical species is developed
understanding of UV technology for water treatment, allowing to determine the concentration profile throughout
us to obtain the full benefit of UV photoreactors. the reactor. The integrated model is evaluated using
The modeling of a UV photoreactor is a challenging area concentration data obtained with a newly developed
of research because of the multiphysics nature of this type and modified planar laser-induced fluorescence
of reactor. The model should allow for the effects of fluid (PLIF) method. This powerful technique reveals the
movement and mixing (hydrodynamics), the distribution of concentration profile throughout the entire domain
radiant energy (radiation field), the rate of deterioration of of the reactor and can be useful to evaluated the full
model.
* Address correspondence to either author. Tel.: (+1) 604-822- The strategy presented can be employed for any type of UV
3238. Fax: (+1) 604-822-6003. E-mail: siamak.elyasi@gmail.com (S.E.); reactor in order to obtain a reliable integrated model for
fariborz@chbe.ubc.ca. performance simulation.
2056 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 6, 2010 10.1021/es902391t  2010 American Chemical Society
Published on Web 02/12/2010
2. Theory 2.3. Species Mass Conservation. For each individual
2.1. Mass and Momentum Conservation (Hydrodynamics). chemical (species m) in the computational domain, the mass
The velocity field can be obtained by solving the equations conservation equation is
of mass and momentum conservation. The general forms of
the conservation of mass and momentum (Navier-Stokes ∂
(x F) + ∇ · (xmFu) ) -∇ · jeff + Sm (5)
equation) are ∂t m

∂ where x, jeff, and S are mass fraction, effective diffusive flux,


(F) + ∇ · (Fu) ) 0 (1)
∂t and source/sink term of species m, respectively. The diffusive
term can be derived using Fick’s law with the molecular
∂ diffusion coefficient of the species in the medium. It was
(Fu) + ∇ · (Fuu) ) ∇p - ∇ · τ + Fg + F (2) shown in our previous work that a turbulent Schmidt number
∂t
of 0.7 is an appropriate value to be used for calculating the
effective diffusive flux (22).
where t, F, u, p, τ, g, and F are time, medium density, velocity
The source/sink term depends on the nature of the
vector, pressure, viscous stress tensor, gravitational accelera-
reaction occurring in the domain of the reactor. Adding an
tion, and external body force, respectively. An analytical
oxidant (e.g., hydrogen peroxide) to the UV-based photo-
solution of the system of partial nonlinear differential
reactor produces hydroxyl radicals, very strong oxidants that
equations is not available for complex geometries, so these
can oxidize many chemical contaminants. Three potential
must be solved numerically. The three-dimensional, time-
primary reactions occur in parallel for the degradation of a
dependent numerical solution of eq 2 (direct numerical
chemical contaminant: (1) direct oxidation of chemical
simulation (DNS)) is only applicable for a very small
contaminant with oxidant; (2) photolysis by UV radiation;
computational domain and a laminar flow regime due to the
and (3) reaction with hydroxyl radicals (23). The value of the
extensive computational resources required. For the turbu-
sink term in the species conservation equation for the reactant
lent regime, an acceptable engineering approach to solve eq
is the sum of these three reaction rates. For many chemicals,
2 is the use of statistical methods or the classical approach
the rates of reactions are available (24); however, for certain
that solves the Reynolds averaged Navier-Stokes (RANS) form
chemicals, the rates need to be measured. In addition, if a
of the equations (10). In the RANS approach, the Reynolds
chemical of interest (contaminant) exists in the presence of
stress tensors are semiempirically correlated using algebraic
other impurities (which is the case in many practical
(11), one-equation (12), two-equation (e.g., standard k-ε
applications), the rates may be affected and, hence, should
(13), RNG k-ε (14), realizable k-ε (15), or standard k-ω
be measured experimentally. The reaction rates can be
(16)), or multiple equation models of turbulence such as the
determined using a bench-scale collimated-beam UV pho-
Reynolds stress model (RSM) (17-19).
toreactor under controlled conditions (i.e., well-mixed with
2.2. Radiant Energy Conservation. The simplified form
a known radiation field). It is important to note that the
oftheradiationtransferequation(RTE)(20)istheBeer-Lambert
photoreaction rate is a function of the radiant power (fluence
Law that is applicable in many UV reactors, due to a lack of
rate) distribution in the medium. Therefore, the fluence rate
scattering (no significant concentration of particulates) and
should be simulated prior to photoreaction rate modeling.
emission (relatively low temperature) throughout the me-
In addition to the fluence rate, which appears in the source
dium. The differential form of the Beer-Lambert Law for
term of the species mass conservation equation, the velocity
steady-state conditions is
also appears in the convection term of this equation.
Therefore, the velocity field needs to be simulated. As a result,
dI(s, Ω)
+ k(s, Ω)I(s, Ω) ) 0 (3) for developing an integrated model of reactor performances,
ds which reveals the concentration profile of species, it is
essential to also develop models of hydrodynamics and
where I and k are the intensity and the absorption coefficient radiation distribution.
for a specific solid angle (direction) Ω and position vector
s. For a photoreactor, changes in intensity depend not only 3. Experimental Methods
on the medium absorption coefficient but also on the 3.1. Flow-Through Pilot-Scale Photoreactor. A flow-
refraction/reflection through/from different materials in the through pilot-scale photoreactor was studied for UV reactor
medium, such as the quartz body of the UV lamp, the quartz modeling and model evaluation under different flow
protector of the UV lamp (sleeve), and the body of the reactor. conditions. The dimensions of the reactor were selected
These refraction/reflection phenomena should be integrated to meet the criteria for planar laser-induced fluorescence
into eq 3. that were recommended by Menton and Lipp (26) in order
The photoreaction rate is a function of the radiant power to minimize the optical absorbance. The body of the reactor
obtained by integrating the intensity over the entire solid was built from glass (5 mm thickness), instead of clear
angle (4π). Considering all the above-mentioned effects, the polymeric material such as polymethacrylate, to prevent
radiant power per unit area, or fluence rate, at a point for its transparency being affected by either UV radiation or
one specific wavelength is (21) chemicals adsorption over the experiment period. At the

( )
bottom of the reactor, a quartz sleeve was attached to hold
n m

∫ ∑∫
a large UV lamp (low pressure, 200 W, arc length 1.07 m,
∏T
4π 1
G) n
I0(Ω) exp[- ki(s, Ω) ds] i dΩ with more than 90% of its emission at 255 ( 5 nm, from
0 overLi
( ∑
i)1
Li)2
i)1 i)1
Emperor Aquatics). A hole at the end of the sleeve with air
suction at the other end, with flow rate of 0.0013 ( 0.003
(4) m3/s and inlet/outlet temperatures of 22 ( 1/25 ( 1 °C,
maintained an ambient skin temperature on the surface
where Li, I0, ki, and Ti are the path length of a ray through of the sleeve and maintained isothermal conditions. Figure
the ith medium, intensity at source, absorption coefficient S1 (of the Supporting Information) shows the dimensions
of medium i, and fraction of the ray transmitted from one and configuration of the pilot-scale reactor.
medium to another, respectively. The fluence rate is the key 3.2. Material. A fluorescent chemical, Rhodamine WT,
parameter in the photoreaction rate correlation. abbreviated as RhWT, was selected as the chemical candidate

VOL. 44, NO. 6, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2057


for photoreactor modeling and model evaluation. RhWT, with fields from the three sections were stitched together to
the CAS registry number 37299-86-8 and the index name represent the velocity profile inside the reactor (0.25 m length
Xanthylium, 9-(2,4-dicarboxyphenyl)-3,6-bis(diethlamino)-, of the reactor).
chloride, disodium salt in a 20 wt % aqueous solution was 3.5. Radiant Energy Measurement. A radiometer (IL1700,
supplied by Turner Designs Inc. Potassium iodide, potassium from International Light Technologies) and a solar blind UV
iodate, and borax were used to calibrate the UV radiant energy sensor (SIC01M-C, from Roithner Laser Technik) were used
passing through a collimated-beam bench-scale photoreactor to measure the irradiance rate at different points inside the
(used for kinetic measurements). The bench-scale photo- reactor (where no liquid was run through the reactor, which
reactor contains a low pressure UV lamp with more than was filled with air). The top side of the reactor was removed,
90% of its emission at 255 ( 5 nm from Emperor Aquatics. and the sensor was inserted into the flow pathway of the
Hydrogen peroxide was used as an oxidant and was measured reactor. The midplane inside the reactor was scanned,
using potassium iodide, sodium hydroxide, and ammonium changing the position of the sensor inside the reactor (see
molybdate tetrahydrate. Ammonium molybdate tetrahydrate Figure S3 of the Supporting Information).
was supplied by Acros Organics, and all of the other chemicals 3.6. Photoreaction Rate Measurement. The photore-
were supplied by Fisher Scientific. action rate of RhWT (as a model contaminant) and hydrogen
3.3. Concentration Measurement. The absorbance of
peroxide (as an oxidant) in the presence of UV radiant energy
different weight percentages of RhWT (10 ppb to 10 ppm) at
was measured in a customized UV collimated-beam pho-
different hydrogen peroxide concentrations was measured
toreactor under controlled conditions. The reactor consisted
at 555.5 nm (maximum absorbance) using a Cary 100 UV-vis
of: two midsize UV lamps on an aluminum reflector below
spectrophotometer to obtain its extinction coefficient (21).
a beam collimator; a double-jacketed reactor of 250 mL, with
The concentration of hydrogen peroxide in the solution was
measured using the method presented by Klassen et al. (25). a round quartz window (3 mm thickness) at the bottom; and
Hydrogen peroxide can oxidize iodide solutions under a variable speed stirrer at the top. The photoreactor sample
alkaline pH conditions and generates an I3- ion. The port (bottom) was connected to the flow-through cuvette of
absorbance of the solution at 352 nm was used to back- a UV-vis spectrometer (Cary 100) via the circulating pump.
calculate the concentration of hydrogen peroxide. Prior to measuring the photoreaction rate of the RhWT, the
3.4. Velocity Measurement. Particle image velocimetry actinometry solution of iodide/iodate and the online spec-
was applied to measure the velocity field. The flow-through trometer were used to accurately determine the UV radiant
photoreactor was installed in a piping network consisting of power passing through the collimated-beam photoreactor
pumps, piping, instrumentation, a storage tank, and a particle (29). The concentration of RhWT solution was 126 ( 1 ppb,
image velocimetry (PIV) setup. Figure S2 (of the Supporting and the hydrogen peroxide concentration was 10.14 ( 0.05
Information) shows a schematic diagram of the PIV setup. ppm. These were measured as a function of absorbed
The same setup was used to measure the concentration radiation to find the photoreaction rate (see Figure S4 of the
profile by planar induced-laser fluorescence (PLIF). Supporting Information) (21).
The PIV setup employed was the “FlowMap 2D” system 3.7. Concentration Profile of RhWT Measurement
from Dantec Dynamics which consists of software for (PLIF). Fluorescent chemicals (e.g., RhWT) upon absorption
capturing and processing images; a 12-bit digital camera of light at a given wavelength (e.g., 532 nm) can generally
(“HiSense MKII” from Hamamatsu Photonic K.K.) equipped emit light at higher wavelengths (e.g., 580 nm). The intensity
with a narrow-band filter at 532 nm; and a laser pulse source of the emitted light depends directly on the concentration
at 532 nm (“Laser Solo III-15 Hz” from NewWave Research of the fluorescent chemical. This behavior allows the
Company). PIV measurement criteria (27, 28) were taken measurement of the fluorescent chemical concentration
into consideration in order to minimize errors during the profile using the same setup as for PIV. The technique is
experiments. Seeding the fine particles (polyamide with referred to as planar laser-induced fluorescence (PLIF).
diameters of 10-20 µm) causes the reflection of light Replacing the narrow-band filter of the camera in the PIV
produced by the pulsed laser source. During the very short setup with a high-pass filter (>550 nm) can convert the PIV
lifetime of the pulse (5-10 ns), the displacement of the setup to a PLIF setup. Due to the nature of pulsed lasers, the
particle is almost zero, and it can be concluded that the beam energy is not spatially uniform. In addition, the laser
scattered light represents the location of the particles energy varies over time in consecutive pulses ((7% as
captured by the digital camera. After a short period of time, measured in our experiments). To increase the accuracy of
a second image is captured using a second laser pulse.
the measurements by considering these variations, the laser
Comparison of the two images (cross correlation) using a
beam was split into two parts using a beam splitter. The
proper length scale reveals the length and direction of the
energy of one beam was measured using another 12-bit digital
velocity vector in the plane of the laser sheet. To reduce the
camera with a narrow-band (532 nm) filter from a reflective
noise, a Gaussian window function with a coefficient of 1
surface as the reference. The other beam excited the
was used. Because window functions do not use the
information near the edges of an interrogation area, a 25% fluorescent solution through the reactor and this was
overlap (4 pixels) of the interrogation window of 16-by-16 captured using the second camera with a high-pass filter
pixels was considered in the calculations. In addition, to (>550 nm). This technique provides a way of accounting for
broaden the narrow correlation peaks and remove the effect the pulse-to-pulse variation in laser energy and, hence,
of the neighboring points in the correlation plane, a low- increases the accuracy of the concentration measurements.
pass Gaussian filter with a coefficient of 1.5 was applied to For the PLIF calibration process, different solutions of
the frequency domain of the Fourier transform calculation RhWT (10-130 ppb) with a constant hydrogen peroxide
(27). concentration (9.65 ( 0.56 ppm) were pumped through the
Using the aforementioned criteria, a 0.25 m length of the reactor (while the UV lamp was off), and 400 images were
reactor (from reactor inlet) was studied using PIV. This zone captured for each concentration using both cameras. The
was divided into three sections. For each section, 400 double value of each pixel on the resulting image, after subtracting
images were captured at two different mass flow rates of the dark base (no laser source) was divided by the total value
0.005 ( 0.002 and 0.014 ( 0.002 kg/s. The values are reported of reflected light (captured by the second camera) to allow
with 95% confidence intervals. The 400 images were pro- normalization based on the reference beam. Finally, the
cessed to reveal the average velocity field. Finally, the velocity normalized value for each pixel was plotted in relation to the

2058 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 6, 2010


FIGURE 1. PIV results at mid cross-section of the reactor for two different mass flow rates: 0.005 (A) and 0.014 (B) kg/s. The results
are shown for the first half of the reactor, in which hydrodynamics is most important, due to the flow being far from fully developed.
Because of the light scattering and reflection, unrealistic velocity vectors were generated from some parts of the reactor; these have
been removed (blank areas).

concentration of RhWT to determine a linear correlation for the inlet average Reynolds number. As a result, in case A, the
that pixel. This resulted in a two-dimensional (2D) calibration flow regime is laminar throughout the entire reactor length;
map. in case B, however, the area close to the bottom of the reactor
A stock solution of hydrogen peroxide (10.30 ( 0.05 ppm) (near the sleeve) is in the transition regime, whereas the area
and RhWT (126 ( 1 ppb) at 21.4 ( 0.3 °C was fed into the far from the sleeve is laminar. The aforementioned observa-
reactor while the UV lamp was stabilized. Two different flow tions were modeled using different viscous models (laminar,
rates of 0.006 ( 0.002 and 0.015 ( 0.002 kg/s were tested. For transitional,andturbulent)withdifferentnumericalapproaches.
each operating condition, the inlet, middle, and outlet of the The CFD results from laminar and standard k-ω tran-
reactor were studied, and 400 images were captured for each sitional flow simulations using the first-order discretization
section. The average of the normalized images, after sub- method yielded the best fit of the PIV data for cases A and
tracting the glowing background image, was used to calculate B, respectively (Figure 2). Overall, the flow patterns were
the RhWT concentration profiles inside the flow-through UV well-predicted by the model.
photoreactor. This procedure was applied for three zones In order to show the deviation of the simulation results
through the entire length of the reactor. from the experimental measurements (PIV), the velocity
vectors of the simulation results were subtracted from the
4. CFD Model Setup PIV measurement for each corresponding point for the two
The details of the CFD model set up are provided in the different cases (see Figure S6 of the Supporting Information).
Supporting Information. Considering the scale of the images, the error corresponding
to the PIV measurements, and the instrumental errors, there
5. Results and Discussion is a good correspondence for both cases, although with better
5.1. Evaluation of the Hydrodynamic Model. 5.1.1. Velocity agreement for case B. In the inlet zone, none of the models
Field Measurement (PIV). The results of the PIV measurement could predict the high circulation flow accurately. In addition,
for the two flow rates for the entrance region of the reactor there is a narrow band below the contour of the turning
are shown in Figure 1. For each image, the length of the points for which both models predicted a smaller velocity
arrows is proportional to the velocity magnitude. The value field than was found experimentally. The overall degree of
of R2 (the square of the correlation coefficient between two agreement can be evaluated using the coefficient of deter-
observed data values) for the average of 400 and 350 images mination (R2) between the PIV velocity vectors and those of
for each of the two flow rates was close to 1, suggesting that simulation, which is expressed as
400 images were statistically sufficient for averaging.
The flow pattern within the UV reactor begins at the inlet
R2 ) 1 -
∑ (Vx - Vx PIV SIM)
2
+ (VyPIV - VySIM)2
jet and passes over the hemisphere of the sleeve in the inlet
section where it diverts to the main section of the reactor. ∑ (Vx - Vx
PIV PIV,Ave)
2
+ (VyPIV - VyPIV,Ave)2
Due to the high gradient of momentum in different directions (6)
of the jet stream, a back-circulation of the fluid at the inlet
zone is produced causing recycle of fluid near the top of the where Vx and Vy are the x and y components of the velocity
reactor. Higher inlet velocities (with higher gradients) result vector, respectively, and indices PIV, SIM, and Ave represent
in greater vorticity and flow rate of the recycled flow and, the PIV measurements, simulation results, and the average
consequently, reduce the path height of the main fluid at the of the measured values (PIV) throughout the entire domain,
bottom of the reactor (close to sleeve)(see Figure S5 of the respectively. The R2 for cases A and B are 0.86 and 0.92,
Supporting Information). respectively (a perfect match would result in R2 ) 1.0). Better
5.1.2. Simulation of Hydrodynamics. The average Rey- statistical agreement was found between the simulation
nolds numbers corresponding to the flow rates over the entire results and experimental values for case B. It is expected that
cross-section at the round inlet of the photoreactor (Figure the simulation and experimental concentration profiles,
S1 of the Supporting Information) are 530 ( 230 and 1480 which are highly affected by the velocity field, will show better
( 280 for cases A and B, respectively. Due to variations in agreement in case B, as well.
the velocity at each cross-section, the local Reynolds number 5.2. Evaluation of the Radiation Model. Considering the
differs on any vertical line (from top to bottom) within the UV lamp as a line or volume source (I0 modeling in eq 4),
reactor, but the maximum Reynolds number cannot exceed the UV radiant emission from the source can be modeled

VOL. 44, NO. 6, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2059


FIGURE 2. Simulated velocity profile (CFD results) at mid cross-section of the reactor for two different flow rates: 0.005 (A) and 0.014
(B) kg/s.

FIGURE 3. Irradiance rate at different distances indicated by ], 0, ×, ∆, and * for distances 1, 10, 20, 30, and 50 mm from the
surface of the lamp, respectively. Experimental values for the radiation power (symbols) vs the 1D model-predicted values
(continuous lines) and the 3D model, cylindrical emitter with 7 mm diameter, values (dashed lines).
using measured boundary conditions. It is important to note the presence of UV is the summation of the three primary
that the UV lamp does not behave as a uniform emission rates of direct oxidation, direct photolysis, and reaction with
source because the moving air (used in this experiment as hydroxyl radicals (generated from UV photo oxidation of
a coolant in the gap of the sleeve and lamp to maintain hydrogen peroxide). The rate of direct oxidation and direct
ambient temperature on the outer sleeve surface) changes photolysis were found to be insignificant. The total photo-
the temperature gradient along the side of the lamp plasma initiated oxidation rate of RhWT and hydrogen peroxide after
dramatically. This causes a nonuniform emission that is lower receiving H J/m2 UV fluence was measured to be (21)
closer to the lamp tip, where air enters the gap. As a result,
radiant power should be measured close to the surface of
the lamp/sleeve to calculate the intensity of the UV lamp at
the source (boundary conditions). In addition, for a more
ln ( CRhWT
130 )) (-4.757 ( 0.092) × 10-3H (7)

realistic model, reflection from the body of the reactor should


where CRhWT is the concentration of RhWT after receiving H
be integrated into the model. A comparison of the experi-
(J/m2) fluence. This correlation was developed for the
mental and modeling results for the irradiance rate inside
concentration of RhWT of less than 130 ppb (initial con-
the air-filled reactor (absorption coefficient is equal to zero)
centration of RhWT) and hydrogen peroxide at 10 ppm.
is shown in Figure 3.
Considering the volume of the bench-scale collimated-beam
Modeling the radiation (considering refraction and re-
photoreactor (in which the reaction rate was measured) and
flection from the reactor walls), with the radiant source as
the agitation effect of the impeller, the photoreactor is a well-
a volumetric or linear emitter, provided a reasonable
mixed batch reactor. Using the chain rule, the rate of reaction
agreement with the experimental results. However, the results
for such reactor from eq 7 can be written as (21)
from the linear source model were more satisfactory over
the various distances from the lamp. As a result, the line
source (1D) model was selected for modeling the radiation dCRhWT
) (-4.757 ( 0.092) × 10-3GabsCRhWT (8)
field inside the reactor, taking into account the absorption dt
coefficient of RhWT and hydrogen peroxide (4.48 m-1), the
reflection/refraction from/through the sleeve and the lamp where Gabs is the absorbed fluence rate.
quartz body, as well as the reflection from the side walls of Considering the temperature of reaction rate tests during
the reactor (21). The simulated fluence rate profiles inside the experimental measurements (21 ( 1 °C), eq 8 is valid
the photoreactor are shown in Figure S7 (of the Supporting only for ambient temperature. Passing cooling air over the
Information). In the areas about 3 cm above the lamp sleeve UV lamp in the PLIF measurements maintained the tem-
surface, the fluence rate is six times less than in the area perature of reaction zones at 21 ( 1 °C. During all tests, the
close to the surface of the sleeve. skin temperature of the sleeve while the UV lamp was
5.3. Kinetic Model Determination. The total rate of operated did not exceed 24 °C, and hence, the isothermal
reaction of the rhodamine WT with hydrogen peroxide in assumption using eq 8 is valid.

2060 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 6, 2010


FIGURE 4. Concentration profile of rhodamine WT (ppb) in the UV reactor with the laser energy at a steady-state condition for two
different mass flow rates: 0.006 (A) and 0.015 (B) kg/s.

FIGURE 5. Concentration profile of rhodamine WT in the UV photoreactor calculated by the integrated model for two different mass
flow rates: 0.006 (A) and 0.015 (B) kg/s. Concentrations less than 80 ppb are shown as white.

5.4. Evaluation of the Integrated Model. The concen- of RhWT was very low. In other words, higher concentrations
tration profile of RhWT is the result of the interaction of of RhWT are expected in this zone, in contrast to what was
hydrodynamics, fluence rate, and reaction rate. Therefore, observed experimentally (Figure 4). This can be explained
measuring the concentration profile throughout the entire by considering the flow pattern in the reactor (Figure 1),
reactor is the best indicator for the evaluation of the which plays a major role in controlling the concentration.
photoreactor model performance. As far as the authors are Low velocity (flow) of fluid at the reactor inlet zone increases
aware, no experimental measurement of the concentration the residence time of chemicals and consequently enhances
profile in a photoreactor has been reported in the open the conversion consequently. In addition, the recycled flow
literature. Figure 4 shows the measured concentration of the of low concentration fluid from downstream to upstream
RhWT profile through the midcross-section of the photo- dilutes the concentration of the inlet stream. As a result,
reactor. The concentration profile through the entire length lower concentrations levels of RhWT are expected at the inlet
of the reactor was studied by investigating three sections of zone. In the area closer to the lamp (sleeve) where the fluence
the reactor from inlet to outlet separately. The three images rate is at its maximum level, a minimum concentration of
(averaged from 400 images for each zone) were combined RhWT should be observed. These features are clearly
and stitched together. Due to the technical limitations in demonstrated in Figure 4. Error in concentration can occur
keeping the setup perfectly consistent while studying different where the concentration gradient is high, with higher
zones, it was not possible to obtain a perfect match in the concentration at top is higher (due to a high absorption of
areas where two adjacent images overlapped. The values laser light at the reactor exit) than at the bottom. Therefore,
close to the boundaries of the reactor walls were masked the measurements in the outlet sections are expected to have
because they did not represent the true concentrations (as a relatively high degree of uncertainty for all cases and this
a consequence of reflections from the white glue joints in explains the similar concentration profiles found for both
these regions of the reactor). For all images, a horizontal cases in Figure 4, at the end of the reactor.
median filter of rank 40 was applied to remove the shadowing The integrated model of reactor performance simulated
effects of the laser sheet in the images. the concentration profile in the UV reactor by solving the
Radiation measurement in the inlet zone (-5 to 1 cm governing equations of mass, momentum, radiant energy,
from the tip of the lamp) showed a very low level of radiant and species conservation. The results are shown in Figure
energy (almost zero). This implies that the rate of conversion 5. The main features of the concentration profile through

FIGURE 6. Concentration difference (CPLIF - CSimulation) in parts per billion for two different mass flow rates: 0.006 (A) and 0.015 (B) kg/
s. Concentration differences higher than 30 ppb are shown as white.

VOL. 44, NO. 6, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2061


the reactor is well-predicted by the model. The difference Gabs ) absorbed fluence rate (W/m2)
between the modeling and experimental results of RhWT g ) gravitational acceleration (m/s2)
concentration for the two flow rates (cases A and B) are shown H ) local absorbed fluence (dose) by microorganisms or
in Figure 6. The reliability of the simulation results (Figure chemicals (J/m2)
6) should be compared with the uncertainty associated with I ) average radiant intensity of entire spectrum (W/sr)
the calibration map for calculating the concentration con- I0 ) average radiant intensity of entire spectrum at source
tours in the PLIF procedure (see Figure S8 of the Supporting
of radiation (W/sr)
Information).
jeff ) effective molecular diffusive rate (kg/s · m3)
Overall, there is close agreement between the modeling
and experimental results throughout the entire cross-section k ) absorption coefficient of the medium for the whole
of the reactor. The relatively high degree of disagreement at spectrum (1/m)
the lower end of the reactors is likely due to the difficulties ki ) absorption coefficient of medium, ith (1/m)
in PLIF measurements as a result of light reflection in these Li ) pathway length of ray through medium, ith (m)
regions. The differences between the model predictions and p ) pressure (Pa)
the experimental results are in the range of measurement R2 ) coefficient of determination
uncertainty (see Sources of Uncertainty and Errors in the Sm ) source term of species m (kg/s · m3)
Supporting Information). The agreement is better for case s ) position vector (m)
B, likely due to a better prediction of the velocity profile. For t ) time (s)
both cases, in a small zone at the end of the reactor, adjacent u ) velocity vector (m/s)
to the UV lamp, the difference in concentration shows a Vx ) x component of velocity vector (m/s)
notable deviation. In that zone, the significant vertical
Vy ) y component of velocity vector (m/s)
concentration gradient causes an error in the PLIF measure-
xm ) mass fraction of species m
ment. Overall, the disagreement between the results can be
attributed to the uncertainty associated with experiments, F ) density (kg/m3)
such as instrumental error and error of the PLIF method, as τ ) viscous stress tensor (Pa)
well as the errors associated with numerical methods and Ω ) solid angle (steradians)
model parameters.
The PLIF method presented here cannot be used directly
for industrial photoreactors; considering it requires the body Supporting Information Available
of the reactor to be manufactured from optically accessible
CFD model setup (30), source of uncertainty and errors,
(transparent) material. In addition, for reactors with relatively
and Figures S1-S8. (Figure S1) The photoreactor used for
large diameter, a scanning method should be employed to
first find the concentration close to the reactor wall and then PLIF measurements (all internal dimensions are expressed
continue the scanning up to the center of the reactor, taking in cm). (Figure S2) Schematic view of pilot-scale photo-
into account any absorbance correction. However, this reactor, which consists of: product reservoir (1), feed
technique could be directly used for small pilot scale reservoir (2), stirrer (3), centrifugal pump (4), flow/
transparent reactors for better understanding of the reactor pressure/temperature meter (5), photoreactor (6), laser
behavior or for evaluating the general-purposed integrated source (7), digital camera (8), PIV/PLIF control unit (9),
models of reactor performance. data acquisition system (10), and online spectrophotometer
To summarize, conventional methods for evaluating (11). (Figure S3) Reflector glasses over the UV lamp in the
photoreactor models typically rely exclusively on concentra- quartz sleeve. The apparatus consists of two window glasses
tion measurements at the reactor inlet and outlet. These (1), a UV photodiode extension arm (2), a UV sensor (3),
methods cannot show the discrepancies between the model a UV lamp (4), a quartz sleeve (5), and the position of the
predictions and real values inside the photoreactor, nor can
hole in the sleeve (6). (Figure S4) Schematic diagram of
they reveal the causes of the deviation. The methods utilized
bench-scale collimated-beam UV photoreactor: parabolic
in this research can be applied to evaluate photoreactor
models, as well as models of similar systems. This approach reflector (1), UV lamp (2), collimator (3), double jacket
evaluates the accountability of each component of the rector (4), stirrer (5), and thermometer (6). (Figure S5)
integrated model as well as the results of the integrated model Velocity (x-component) profile and contour of turning
for the entire computational domain. As a result, the points at different sections of the reactor for two different
discrepancy for each component is revealed, and the model flow rates. Vertical lines indicate different xpositions in
can be applied to a wide range of operating conditions. This centimeters. (Figure S6) Velocity difference (vectors), VPIV-
approach reduces uncertainty in the integrated model setup VSIM, and x-components of velocity reactor. Vertical lines
and provides a solution for individual phenomena. Conse- indicate x-position, and curves show zero axial velocity
quently, it decreases the bias of the final integrated model difference (VxPIV-VxSIM). The color bar (linear scale)
solution. indicates the velocity difference (vectors) from 0.015 (red)
Considering the uncertainties in the measurements of and 0.03 (red) to 0 m/s (blue) for cases A and B, respectively.
velocity, fluence rate, and concentration, overall, the favorable
The arrows show the flow directions. The velocity mea-
agreement between the experimental data and the simulated
surements near the surfaces are removed (white regions)
results for each governing equation (momentum, mass, and
radiant energy), and the integrated system of equations because of the high uncertainty in the measurement due
(species mass conservation) verifies the reliability of the to the reflections. (Figure S7) Fluence rate (W/m2) profile
presented methodology. at the mid-cross-section of the photoreactor. (Figure S8)
Uncertainty in the RhWT concentration profile measure-
Appendix A ment (ppb) with a 95% confidence interval in the UV reactor
for two different mass flow rates: 0.006 (A) and 0.015 (B)
Nomenclature kg/s. The color bar (linear scale) indicates the concentra-
CRhWT ) concentration of rhodamine WT (ppb) tion uncertainty from (30 (red) to 0 ppb (blue), and the
F ) volumetric external forces (N/m3) arrows show the flow directions. This material is available
G ) fluence rate (W/m2) free of charge via the Internet at http://pubs.acs.org.

2062 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 6, 2010


Literature Cited (16) Wilcox, D. C. Turbulence Modeling for CFD; DCW Industries,
Inc.: La Canada, 1998.
(1) Kidman, R. B.; Tsuji, K. S. Preliminary Costs Comparison of (17) Launder, B. E. Second-Moment Closure: Present. . and Future.
Advanced Oxidation Processes; Los Alamos National Laboratory Int. J. Heat Fluid Flow 1989, 10, 282–300.
Unofficial Copy (LA-12221-MS). SJB 1-21-91, 1992.
(2) Gunten, U. V. Ozonation of drinking water: Part I. Oxidation (18) Gibson, M. M.; Launder, B. E. Ground Effects on Pressure
Kinetics and Product Formation. Water Res. 2003, 37, 1443– fluctuations in the Atmospheric Boundary Layer. J. Fluid Mech.
1467. 1978, 86, 491–511.
(3) Castrillo, S. R. V.; Lasa, H. I. Performance Evaluation of (19) Launder, B. E.; Reece, G. J.; Rodi, W. Progress in the Development
Photocatalytic Reactors for Air Purification Using Computational of a Reynolds-Stress Turbulence Closure. J. Fluid Mech. 1975,
Fluid Dynamics (CFD). Ind. Eng. Chem. Res. 2007, 46 (18), 5867– 68, 537–566.
5880. (20) Romero, R. L.; Alfano, O. M.; Cassano, A. E. Cylindrical
(4) Munoz, A.; Craik, S.; Kresta, S. Computational Fluid Dynamics Photocatalytic Reactors. Radiation Absorption and Scattering
for Predicting Performance of Ultraviolet Disinfection Sensitivity Effects Produced by Suspended Fine Particles in an Annular
to Particle Tracking Inputs. J. Environ. Eng. Sci. 2007, 6, 285– Space. Ind. Eng. Chem. Res. 1997, 36, 3094–3109.
301. (21) Elyasi, S. Development of UV Photoreactor Models for Water
(5) Ducoste, J. J.; Liu, D.; Linden, K. Alternative Approaches to Treatment. Ph.D. thesis, University of British Columbia, 2009.
Modeling Fluence Distribution and Microbial Inactivation in
(22) Elyasi, S.; Taghipour, F. Simulation of UV Photoreactor for Water
Ultraviolet Reactors: Lagrangian versus Eulerian. J. Environ. Eng.
Disinfection in Eulerian Framework. Chem. Eng. Sci. 2005, 61
2005, 131 (10), 1393–1403.
(14), 4741–4749.
(6) Kamimura, M.; Furukawa, S.; Hirotsuji, J. Development of a
(23) Oppenlander, T. Photochemical Purification of Water and Air;
Simulator for Ozone/UV Reactor Based on CFD Analysis. Water
Wiley-VCH: Weinheim, 2003.
Sci. Technol. 2002, 46 (11-12), 13–19.
(7) Janex, M. L.; Savoye, P.; Do-Quang, Z.; Blatchley, E.; Laine, J. M. (24) National Institue of Standard and Technology, Chemical Kinetics
Impact of Water Quality and Reactor Hydrodynamics on Waste- Database. http://kinetics.nist.gov/kinetics/index.jsp (accessed
water Disinfection by UV, Use of CFD Modeling for Performance May 2008).
Optimization. Water Sci. Technol. 1998, 38 (6), 71–78. (25) Klassen, N. V.; Marchington, D.; McGowan, C. E. H2O2 Deter-
(8) Sozzi, A.; Taghipour, F. The Importance of Hydrodynamics in mination by I3- Method and by KMnO4 Titration. Anal. Chem.
UV Advanced Oxidation Reactors. Water Sci. Technol. 2007, 55 1984, 66, 2921–2925.
(12), 53–58. (26) Melton, L. A.; Lipp, C. W. Criteria for Quantitative PLIF
(9) Liu, D.; Wu, C.; Linden, K.; Ducoste, J. Numerical Simulation Experiments Using High-Power Lasers. Exp. Fluid 2003, 35, 310–
of UV Disinfection Reactors: Evaluation of Alternative Turbu- 316.
lence Models. Appl. Math. Model. 2007, 31, 1753–1769. (27) FlowMap Manual. Particle Image Velocimetry Instrumentation,
(10) Ranade, V. V. Computational Flow Modeling for Chemical Reactor Installation & User’s guide; Dantec Dynamics: Ramsey, USA,
Engineering, 1st ed.; Academic Press: London, 2002. 2004.
(11) Prandtl, L. Z. Report on investigation of developed turbulence.
(Bericht über Untersuchungen zur Ausgebildeten Turbulenz (28) Willert, C. E.; Gharib, M. Digital Particle Image Velocimetry.
(German). Angew. Math. Mech. 1925, 5, 136. Exp. Fluids 1991, 10, 181–193.
(12) Spalart, P.; Allmaras, S. A One-Equation Turbulence Model for (29) Rahn, R. O.; Stefan, M. I.; Bolton, J. R.; Goren, E. Quantum Yield
Aerodynamic Flows; American Institute of Aeronautics and of the Iodide-Iodate Chemical Actinometer: Dependence on
Astronautics. Technical Report AIAA-92-0439, 1992. Wavelength and Concentration. Photochem. Photobiol. 2003,
(13) Launder, B. E.; Spalding, D. B. Lectures in Mathematical Models 78, 146–152.
of Turbulence; Academic Press: London, 1972. (30) Shilton, A. Studies into the Hydraulics of Waste Stabilization
(14) Yakhot, V.; Orszag, S. A. Renormalization Group Analysis of Ponds. Doctor of Philosophy, Massey University, New Zealand,
Turbulence, I. Basic Theory. J. Sci. Comput. 1986, 1, 1. 2001.
(15) Shih, T. H.; Liou, W. W.; Shabbir, A.; Yang, Z.; Zhu, J. A New k-ε
Eddy-Viscosity Model for High Reynolds Number Turbulent
Flows - Model Development and Validation. Comput. Fluids
1995, 24, 227–238. ES902391T

VOL. 44, NO. 6, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2063

You might also like