Chemical Engineering Journal 249 (2014) 121–129

Contents lists available at ScienceDirect

Chemical Engineering Journal

journal homepage: www.elsevier.com/locate/cej

Fixed bed photocatalytic reactor for formaldehyde degradation:

Experimental and modeling study
A. Cloteaux a,b,c, F. Gérardin c,⇑, D. Thomas a,b, N. Midoux a,b, J.-C. André a,b
a
Université de Lorraine, LRGP, UMR 7274, 1 Rue Grandville, BP 20451, 54001 Nancy, France
b
CNRS, LRGP, UMR 7274, 1 Rue Grandville, BP 20451, 54001 Nancy, France
c
Institut National de Recherche et de Sécurité, Rue du Morvan CS 60027, 54519 Vandœuvre Cedex, France

h i g h l i g h t s g r a p h i c a l a b s t r a c t

 Treatment of formaldehyde-
containing solution with TiO2 coated
Raschig rings.
 Residence time distribution of a fixed
bed photocatalytic reactor.
 Modeling based on hydraulics,
chemical reactions and mass transfer
mechanisms.
 Kinetic constants of photocatalytic
degradation of formaldehyde.

a r t i c l e i n f o a b s t r a c t

Article history: Formaldehyde is toxic to humans and is classed as a category 1 carcinogen. Methods have been developed
Received 19 September 2013 to degrade this compound, but for industrial application, relevant mathematical models are required.
Received in revised form 17 March 2014 This study models a fixed bed photocatalytic reactor designed to degrade formaldehyde constituted of
Accepted 19 March 2014
TiO2-coated Raschig rings illuminated by UV-A lamps. Initially, the reactor’s hydraulic behavior was
Available online 27 March 2014
described based on an experimental residence time distribution. This model takes into account hydrau-
lics, light distribution, chemical kinetics and mass transfer in the reactor. The dispersion model satisfac-
Keywords:
torily represented the reactor’s hydraulic behavior. This model, combined with a Langmuir–Hinshelwood
Photocatalysis
Formaldehyde
kinetic model, was then used to calculate variations in concentration at the reactor output. By adding the
Reactor modeling transfer flux between the bulk and the surface to the material balance equation, it is possible to distin-
Residence time distribution guish between mass transfer and chemical reaction limitation and determine the chemical kinetics.
Kinetics Experimental data from different initial concentrations were used to calculate the Langmuir–Hinshel-
wood kinetic constants. The dispersion model with chemical reaction was validated under various irra-
diation and flow rate conditions. The results show that the fixed bed photocatalytic reactor efficiently
degrades formaldehyde in an aqueous solution. The chemical constants of photocatalytic degradation
obtained for formaldehyde are necessary parameters if this technology is to become the basis of indus-
trial applications. This study provides a new tool, integrating mass transfer limitations and light distribu-
tion, to design photo-reactors.
Ó 2014 Elsevier B.V. All rights reserved.

⇑ Corresponding author. Tel.: +33 383 509 820; fax: +33 383 502 184.
E-mail address: fabien.gerardin@inrs.fr (F. Gérardin).

http://dx.doi.org/10.1016/j.cej.2014.03.067
1385-8947/Ó 2014 Elsevier B.V. All rights reserved.
122 A. Cloteaux et al. / Chemical Engineering Journal 249 (2014) 121–129

Nomenclature

a surface area of the packing material (m2 m3) r global apparent reaction rate (mol s1 m3)
C tracer concentration (g m3) rs surface reaction rate (mol s1 m2)
C formaldehyde concentration in the fluid (mol m3) R radius of the reactor (m)
Ck1, Ck tracer concentrations at the inlet and outlet of tank k RL radius of the lamp (m)
(g m3) Re Reynolds number
CP, CR, CT tracer concentrations at the outlet of the plug flow, the S integral of the experimental signal y over time
reactor and the tank (g m3) Sc Schmidt number
CR, CR0 formaldehyde concentrations at the reactor outlet at a t time variable (s)
given time and initial time (mol m3) ts mean residence time (s)
Cs formaldehyde concentration at the surface of the cata- u interstitial velocity (m s1)
lyst (mol m3) u0 superficial velocity (m s1)
d distance between measuring position and lamp axis (m) V total volume of liquid in the reactor (m3)
dp characteristic size of the packing material (m) Vd dead volume in the reactor (m3)
dTiO2 density of TiO2 loaded onto the packing (g m2) VR volume of liquid flowing through the reactor VR = V  Vd
D dispersion coefficient in the interstitial volume (m2 s1) (m3)
Dm molecular diffusion coefficient (m2 s1) X removal rate (%)
E residence time distribution y experimental signal
fe external resistance fraction ystat experimental signal at steady state
I fluence rate (W m2) z axial coordinate (m)
Imoy mean fluence rate over section (W m2)
Ireference reference light flux (W m2) Greek letters
Itransmitted transmitted light flux (W m2) d Dirac function
IN, IQi irradiance at point N and Q (W m2) Dt step time (s)
I0 irradiance at the surface of the lamp (W m2) e bed porosity
IQ fluence rate at point Q (W m2) e apparent extinction coefficient (m1)
J parameter of the tank-in-series model / transfer flux (mol s1 m3)
kL mass transfer coefficient (m s1) l dynamic viscosity of water (Pa s)
kLH kinetic constant of degradation relative to the mass of ln n order moment of a distribution (sn)
TiO2 (mol m2n g1 Wn s1) q density of water (kg m3)
KLH adsorption constant (m3 mol1) qapp apparent density of the packing material (kg m3)
l length (m) r 2
variance of the distribution (s2)
L length of the packed bed (m) s space time (s)
P parameter of the dispersion model sP, sR, sT space time of the plug flow, the fixed bed reactor and
Pe Péclet number Pe = udp/D the tank (s)
Pem Péclet number for molecular diffusion Pem = udp/Dm
Q liquid flow rate (m3 s1)
r radius (m)

1. Introduction were chosen for their large surface area and relatively high light
transmittance. Photocatalysis is a doubly heterogeneous process;
Photocatalysis is a technological solution allowing oxidative the reaction may be partially limited by species transfer between
degradation of a wide range of organic pollutants in the gas or liquid the surface [7–9] and within the bulk and by non-homogeneous
phase. The heterogeneous catalytic process is based on exciting a light absorption within the catalyst particles. The influence of this
semi-conductor (catalyst) by light at an appropriate wavelength parameter on free radical production yields has not yet been stud-
to generate oxidizing free radicals. These radicals can then react ied. Here, the average influence of light flux density (irradiance) on
with compounds present at or near the surface of the semi-conduc- the semi-quantitative production of oxidative radicals is taken into
tor or catalyst. Due to its performance and practicality, titanium account. In modeling photocatalytic reactors and assessing the
dioxide (TiO2) is the most commonly used photocatalyst. It can be chemical kinetics of pollutant degradation, the material balance
used in various ways, either in suspension or fixed to a reactor sur- is established based on the mass transfer phenomenon [10,11].
face, a packed bed, a monolith or a membrane [1–4]. Developing this balance requires some assumptions to be made
In this study, a fixed bed photocatalytic reactor is modeled and with regard to the reactor’s hydraulic behavior. Photocatalytic
its capacity to degrade formaldehyde in the aqueous phase is as- reactors are very often considered as ideal reactors; as continu-
sessed. Formaldehyde in gas or liquid phase is a particularly toxic ously stirred tank reactors [12,13] or as plug flow reactors
compound (carcinogen category 1) [5]. Its elimination is therefore [11,14]. However, very few in-depth studies of the real behavior
a priority in the treatment of industrial or hospital effluents. In this of photocatalytic reactors have been published. In the present
study, formaldehyde is degraded in a photocatalytic reactor con- study, a residence time distribution (RTD) was initially determined
taining a bed of TiO2-coated glass Raschig rings. This method of fix- to characterize fixed bed hydraulic behavior. A model was then
ing the catalyst to a support has several advantages: no filtration is built by integrating mass dispersion into the bed, mass transfer
required, unlike when TiO2 particles are in suspension; and opera- to the catalyst’s surface and the chemical kinetics of photocataly-
tors are not exposed to titanium dioxide nanoparticles. In some sis. As the chemical kinetics is dependent on light flux, information
applications, however, micrometric or nanometric suspensions of on the light distribution in the reactor is required. This can be cal-
catalyst particles may be of interest because they provide exten- culated by solving the radiative transfer equation [15–18]. In this
sive contact and large reactive surface areas [6]. Glass Raschig rings study, the light distribution in the reactor was estimated using a
 A. Cloteaux et al. / Chemical Engineering Journal 249 (2014) 121–129 123

simple model based on the apparent absorption/extinction coeffi- was used as a fluorescent tracer. Fluorescein solution in water
cient of the coated packed bed. The novelty of this work relates (5  107 m3) was pulse-injected at the reactor inlet. Part of the
to the model developed, which considers the hydraulics, the light flow was diverted from the reactor outlet to a spectrofluorimeter
distribution, and the mass transfer in the reactor, as well as the (Shimadzu spectrofluorimetric detector: flow rate 5  108 m3 s1,
chemical kinetics. A fixed bed reactor is a good technological op- cell 1.2  108 m3). Fluorescence was observed using excitation
tion for air or water treatment processes, although it may suffer and emission wavelengths of 490 nm and 515 nm, respectively. The
from mass transfer and light transmission problems. However, a molecular diffusion coefficient for fluorescein is 4.3  1010 m2 s1
fixed catalyst saves the cost of the filtration stage (required in slur- [20].
ry technology) and avoids emission of titanium dioxide nanometric
particles. The results predicted by the model proposed in this study 2.2. Method to measure dissolved formaldehyde concentration
were compared with experimental results for various concentra-
tions, fluence rates and flow conditions. Once validated, the model The analytical method used to measure the formaldehyde con-
built using the principles of chemical reaction engineering should centration was based on the Hantzsch reaction (reaction of a b-
form a major element in designing and scaling photo-reactors. diketone with an aldehyde). Li et al. [21] selected acetoacetanilide
The information on the chemical kinetics of TiO2-coated packing as b-diketone allowing rapid synthesis (10 min) at room tempera-
provided can be used to consider other ways of implementing this ture. Fluorescence spectrometry was implemented to analyze the
photocatalytic medium. resulting cyclic compound, which has an emission peak at
470 nm upon excitation at 370 nm. The formaldehyde solution un-
2. Material and methods der study (106 m3) was mixed with 5  106 m3 of ammonium
acetate solution at 4 mol L1, 2  106 m3 of acetoacetanilide solu-
2.1. Experimental set-up tion at 0.2 mol L1, and 2  106 m3 of ethanol. After a 10-min
reaction time at room temperature, emission at 470 nm was mea-
The reactor was composed of a tube of borosilicate glass of in- sured (Shimadzu spectrofluorimetric detector: flow rate 1.7 
ner diameter 2.25  102 m filled with photocatalytic packing over 108 m3 s1, cell 1.2  108 m3). A calibration range was set up
0.42 m. The reactor was illuminated by 6 UV-A lamps (Philips TL-D using a stabilizer-free 16% formaldehyde solution. The response
18W BLB) placed around it at an angle of p/3 Radian and 0.1 m was linear for the formaldehyde concentrations within the selected
from the reactor axis. The emission spectrum for these lamps concentration range.
was centered on 365 nm [19]. The closed-loop system included a
photocatalytic reactor and tank (cf. Fig. 1). The reactor was filled 2.3. Preparation of coated packing material
with uncoated packing (5  103 m glass Raschig rings) to a depth
of 0.42 m for the hydraulic study. The flow rate was set to The packed bed material comprised 5-mm external diameter, 3-
5.0  107 m3 s1 (1.8 L h1). At this flow rate, the interstitial mm internal diameter, 5-mm high glass Raschig rings. The surface
velocity was 2.0  103 m s1 and the regime was laminar, with area, a, of these Raschig rings is 700 m2 m3 and their apparent
a Reynolds number of 10. density qapp is 730 kg m3. TiO2 was deposited on the packing by
The system was studied under both open-loop and closed-loop immersing the previously cleaned rings in a 1 g L1 TiO2 P25 De-
conditions. In open-loop conditions, the reactor was continuously gussa (CAS 13463-67-7, BET surface area 35–65 m2 g1, particle
supplied with fresh water, while in closed-loop conditions, the size 21 nm) suspension for 1 min. The rings were then dried for
solution leaving the reactor was collected in a tank for re-pumping 1 h at 100 °C. After drying, the rings were baked at 500 °C for 1 h
to the top of the reactor. The total volume of solution in the system [22]. TiO2 loading of was measured by weighing three samples be-
was 2.0  104 m3 (0.2 L). The 5-mm diameter pipes connecting fore and after coating, and found to be 0.18 ± 0.03 g m2. This cor-
the closed-loop components were approximately 1 m long and ac- responds to a mass of TiO2 in the photocatalytic bed of
counted for a further volume of 2.0  105 m3 (0.02 L). Fluorescein 0.021 ± 0.004 g. Krypton adsorption was used to determine the

Fig. 1. Experimental set-up: (1) fixed bed reactor, (2) lamps, and (3) tank.
124 A. Cloteaux et al. / Chemical Engineering Journal 249 (2014) 121–129

BET [23] surface area of uncoated and coated packing: and variance. The space time s (s) is V/Q, where V (m3) is the total
0.004 m2 g1 and 0.011 m2 g1, respectively. volume of liquid in the reactor and Q (m3 s1) is the liquid flow
rate. Space time and mean residence time are used to reveal the
2.4. Modeling reactor hydraulics presence of dead volumes and short circuits. There is a dead vol-
ume in the reactor, when ts < s. The ratio of the dead volume, Vd,
Reactor hydraulics can be studied using a tracer and compared to the total volume, V, is 1  t s =s.
with different models. Parametric models can be used, when ideal For the dispersion model, the RTD E is obtained by solving the
reactors such as the continuously stirred tank reactor or the plug differential equation subject to the boundary conditions and the
flow reactor are not deemed appropriate for describing a real reac- initial condition described in Appendix A. The term uC(0) in the
tor. The single parameter models based on dispersion and tanks-in- first boundary condition is equal to zero under open-loop condi-
series were selected for this study. tions. The partial differential equation can be solved using the
pdepe function in MatlabÒ. This function solves initial boundary va-
2.4.1. Dispersion model lue problems for systems of parabolic and elliptic partial differen-
The plug flow model can be improved by introducing a disper- tial equations in a space variable and time. The pdepe solver
sion term. This term is based on Fick’s diffusion law with an empir- converts the partial differential equations into ordinary differential
ical dispersion coefficient substituted for the diffusion coefficient. equations using second-order accurate spatial discretization. The
Dispersion represents molecular diffusion and the hydrodynamic ordinary differential equations can then be solved using a function
effect associated with a non-uniform velocity profile. The disper- based on numerical differentiation formulas.
sion coefficient depends on the fluid, the hydraulic regime and For the tanks-in-series model, the RTD is given by Eq. (2) for an
the geometry of the reactor. For a given packed bed, the dispersion integer J. The mean residence time ts is then equal to the space
coefficient, D, is calculated with the correlation proposed by Belfi- time s and the variance is given by:
ore et al. [24] (cf. Appendix A). The P value, used in the dispersion r2 1
model is related to the dispersion coefficient: P = uL/D. At P values ¼ ð7Þ
s2 J
greater than 100, there is no significant difference between the dis-
persion model and the plug flow model [25]. The material balance 2.4.4. Study under closed-loop conditions
for the tracer, the boundaries conditions and the initial conditions The signal at the reactor outlet y is monitored over time under
are given in Appendix A. experimental conditions. Signal S(t) is y(t)/ystat where y(t) is the
signal at the outlet of the reactor at time t and ystat is the value
2.4.2. Tanks-in-series model at steady state. The closed-loop system is composed of the reactor,
Fluid dynamics in a real reactor can be represented by a series the tank, the connecting pipes and any pumps. The following
of J continuously stirred tank reactors (CSTR) of the same volume. assumptions were made for modeling purposes:
The material balance for tank k in a transient state is:
– The reactor volume is equal to the volume of liquid actually
s dC k
C k1 ¼ C k þ ð1Þ flowing through the experimental bed (VR = VVd). The space
J dt
time is sR. The concentration at the reactor outlet is CR, which
where Ck1 and Ck (g m3) are the tracer concentrations at the inlet is iteratively computed using each model’s open-loop equation.
and outlet of tank k, s (s) is the space time in the series and J is the – The tank, with space time sT, is considered to be a CSTR. The
total number of CSTR. The RTD E in the time-domain can be de- step time is Dt, the concentration at the tank outlet, CT, is calcu-
duced from the material balances: lated from Eq. (8). This equation is obtained by discretizing the
 J J1 material balance on the tank.
J t exp ðJt=sÞ
EðtÞ ¼ ð2Þ
s ðJ  1Þ! C R ðtÞ þ sDTt C T ðt  DtÞ
C T ðtÞ ¼ ð8Þ
1 þ sDTt
2.4.3. Study under open-loop conditions – The influence of the connecting pipes is represented by a plug
The signal at the reactor outlet y is monitored over time in flow reactor, which induces a delay equal to space time sp with-
experimental conditions. This signal is normalized to verify that: out altering the shape of the signal. The concentration at the
Z 1 plug flow outlet is CP.
EðtÞdt ¼ 1 ð3Þ – The presence of the pumps is assumed not to alter the signal.
0

Eqs. (4) and (5) are used to obtain the RTD E based on experi- The calculation algorithm is as follows:
mental signal y.
Z 1 – Initialization of concentrations CR, CT and CP.
S¼ yðtÞdt ð4Þ – At each time step:
0
 t = t+Dt
 Solve reactor model equation for a time interval Dt with:
EðtÞ ¼ yðtÞ=S ð5Þ
initial condition C(t, z) provided by the previous iteration
The experimental and modeled RTD are characterized by their z = 0 condition provided by C(t, 0)=CP(tDt).
moments. The n order moments of a distribution E are:  Calculate concentration at reactor outlet CR(t) = C(t, L).
Z 1  Calculate CT(t) from CR(t) and CT(t  Dt) (Eq. (8)).
ln ¼ t n EðtÞdt ð6Þ  Calculate concentration at plug flow outlet CP(t) = CT(tsP).
0

The mean residence time t s is equal to l1. The variance of the 2.5. Kinetic study
distribution r2 is equal to l2  l21 . In Section 3.1, the experimental
and modeled RTD are compared using the distance calculated by The kinetic study initially followed the photocatalytic degrada-
the least squares method and considering the mean residence time tion of formaldehyde with different initial concentrations. The data
 A. Cloteaux et al. / Chemical Engineering Journal 249 (2014) 121–129 125

obtained were used to determine the kinetic constants of the mod-

el. Experiments were then conducted under other irradiation con-
ditions and with different flow rates to validate the model and
kinetic constants.
This study was conducted using the reactor described in Sec-
tion 2.1, supplied with a formaldehyde solution with concentration
1.6  102, 3.2  102, 1.7  101 or 3.4  101 mol m3, prepared
from a stabilizer-free 16% formaldehyde solution. The liquid flow
rate was 5  107 m3 s1 (1.8 L h1). The total solution volume
was 2.0  104 m3 (0.2 L). The bed porosity was 0.62, thus the vol-
ume of liquid in the reactor was 1.05  104 m3 (0.105 L). The solu-
tion in the tank was continuously aerated by bubbling at
8.5  107 m3 s1 (3 L h1). The formaldehyde concentration at
the reactor outlet was monitored for 1 h after starting irradiation.
It was confirmed that the formaldehyde was not degraded by
UV-A irradiation alone.
The hydraulic study was used to build a model considering
the photocatalytic reaction, as described in Section 3.2. For a
given flow rate and at a given light flux, the kinetic constants
were fitted to ensure that the model-predicted concentration
Fig. 2. Definition of the points (in black) and of the distances (in blue) used in light
variation at the reactor outlet corresponded to the experimental
distribution calculation. (For interpretation of the references to color in this figure
data. legend, the reader is referred to the web version of this article.)
Experiments were conducted for different light fluxes and flow
rates to validate the model and the kinetic constants. The influence
 The irradiance, IN (W m2), at point N was calculated using
of light flux was initially studied with a formaldehyde concentra-
the following equation:
tion of 1.7  101 mol m3 and a liquid flow rate of 5  107
m3 s1. Experiments were subsequently conducted at flow rates I 0  RL
IN ¼ ð10Þ
of 2.5  107 m3 s1, 5.0  107 m3 s1 and 1.0  106 m3 s1 Li N
(0.9 L h1, 1.8 L h1 and 3.6 L h1) for the fixed initial concentration
where I0 is the irradiance at the surface of the lamp and RL is
and light flux.
the radius of the lamp.
 The irradiance at point Q was calculated taking the absorp-
2.6. Calculating reactor light distribution tion of the reactor wall and of the packed bed into account
(Eq. (11)).
The chemical kinetics of photocatalytic degradation of formal- IN  Li N
dehyde depends on the incident light flux or irradiance. Therefore, IQi ¼ 0:95   expðe  NQÞ ð11Þ
Li Q
the reactor light distribution is required. The following assump-
tions were made to calculate the light distribution. – The fluence rate at point Q is the sum of the irradiance deter-
mined for each lamp.
– Side effects at reactor boundaries were neglected. This renders X
IQ ¼ IQi ð12Þ
the system independent of the z coordinate.
i
– Irradiance for an isolated lamp considered as a line source is
governed by a 1/d law [26], where d (m) is the distance between – The fluence rate at the surface of the lamp was measured with a
the measuring position and the lamp axis. UV radiometer (Gigahertz Optik X11-XD-9511, 315–400 nm)
– The packed bed is made of glass rings coated with titanium and was found to be 90 W m2.
dioxide. The submerged packed bed is assumed to have an
apparent extinction coefficient e (m1) independent of the The apparent extinction coefficient of the packed bed was
wavelength [19]. In the fixed bed, light is transmitted, scattered experimentally determined. The transmitted light flux for different
or absorbed. For a length l (m), the transmitted light flux is cal- packed bed heights was measured using a UV radiometer. These
culated from the flux for a non-absorbent medium Ireference (Eq. data were then fitted with Eq. (9) to obtain the value of the appar-
(9)) [27,28]. ent extinction coefficient. Based on this experiment, the apparent
extinction coefficient of the packing was found to be 66 m1.
Itransmitted ¼ Ireference expðelÞ ð9Þ
– Reflection by the borosilicate glass reactor wall was neglected. 3. Results and discussion
For a wavelength between 340 nm and 400 nm, the absorbance
of borosilicate glass is almost independent of the wavelength 3.1. Modeling reactor hydraulics
and is equal to 5% for 3 mm thickness glass [19].
The RTD was performed several times under experimental con-
Irradiation at each point Q of the reactor was calculated using ditions. The mean residence time was found to be between 198 and
the following method (cf. Fig. 2): 205 s, with a standard deviation of approximately 50 s for the six
experiments (cf. Table 1). The mean residence time was therefore
– For each lamp centered at point Li, 201 s. The space time was 210 s, thus the average dead volume
 The coordinates of the intersection point N between the line in the reactor represents 4.1% of the liquid volume.
(QLi) and the circle of radius R matching with the reactor The RTD was calculated for a space time of 201 s since the
wall were determined. models did not take the dead volume into account. The mean
experimental and model residence times were therefore equal.
126 A. Cloteaux et al. / Chemical Engineering Journal 249 (2014) 121–129

Table 1 4.5
Mean residence time and standard deviation for experimental and modeled RTD. Experiment
4 Dispersion model P = 39, delay 40 s
Mean residence time Standard deviation
Tanks-in-series model J = 21, delay 40 s
ts (s) r (s)
3.5
Experiment 1 205 49
Experiment 2 205 51 3
Experiment 3 198 49
Experiment 4 200 50 2.5

S(t)
Experiment 5 203 50
Experiment 6 197 49 2
Average of experiments 201 50
Dispersion model (s = 201 s, 201 42.9 1.5
P = 39)
Tanks-in-series model 201 44.4 1
(s = 201 s, J = 21)
0.5

0
Parameters P and J were adjusted to obtain the minimum distance 0 200 400 600 800 1000 1200
between the modeled and experimental distributions, calculated Time (s)
using the least squares method. The optimum value for the disper-
Fig. 4. Experimental and modeled signals at reactor outlet under closed-loop
sion model P is 39 or a dispersion coefficient D = 2.2  105 m2 s1. conditions.
Under these experimental conditions (u = 2.0  103 m s1,
Dm = 4.3  1010 m2 s1), the correlations of Belfiore et al. [24] The signals calculated using the dispersion model and the
gives a value of P equal to 42 and a dispersion coefficient of tanks-in-series model are identical under closed-loop conditions.
2.0  105 m2 s1. The optimized values of P and D therefore agree Overall, the system behavior under these conditions is quite well
with the values predicted by Belfiore et al. [24]. For the tank-in- modeled. In particular, modeled signal oscillations are synchro-
series model, the optimum number of tanks is J = 21. Fig. 3 shows nized with oscillation of the experimental signal. However, the
the experimental distribution, the distribution calculated using the amplitude of the first peak is underestimated. In the tank-in-series
dispersion model for P = 39 and the distribution calculated using model, parameter J is not explicitly linked to the hydraulic condi-
the tanks-in-series model for J = 21. tions. The hydraulic behavior cannot therefore be predicted with
The maximum experimental RTD value occurs slightly earlier this model for other conditions. In the dispersion model, parameter
than for the modeled distributions. This is probably due to the P can be calculated by correlations (cf. Section 2.4.1) with other
presence of a dead volume in the reactor, which reduces the vol- hydraulic conditions. Hence, studying the reactor under other con-
ume of liquid circulating and results in a mean residence time less ditions does not require additional RTD experiments. For this rea-
than the space time. Presence of a dead volume is confirmed by the son, the dispersion model was used to model the photocatalytic
tail on the experimental distribution (Fig. 3). reactor in the following section.
Under closed-loop conditions, the total liquid volume in the
system is 2.0  104 m3 (0.2 L). The reactor liquid volume is 3.2. Reactor modeling
1.05  104 m3 (0.105 L). The liquid volume in the connecting
pipes (1 m of 5-mm diameter pipe) is roughly 2.0  105 m3 3.2.1. Material balance with surface reaction
(0.02 L). The volume in the tank is therefore 7.5  105 m3 The dispersion model was used with an estimated dispersion
(0.075 L). The space time in the tank and the plug flow are respec- coefficient to model the system with chemical reaction at the sur-
tively 150 s and 40 s for a flow rate equal to 5.0  107 m3 s1 face. Both the dispersion model and the calculated value of D have
(1.8 L h1). Fig. 4 illustrates the experimental signal at the reactor been validated and can be applied for different flow rates without
outlet under closed-loop conditions. The signal variation was cal- requiring a new experimental RTD. The partial differential equa-
culated using the dispersion model for P = 39 and using the tion (Eq. S1) has been modified to include the chemical reaction
tanks-in-series model with J = 21. In both cases, the space time in at the surface of the catalyst, by removing the Dirac function and
the reactor was 201 s. The results are plotted in Fig. 4. adding a reaction term. The boundary conditions are those used
in the system with recirculation. Initially, the concentration in
0.012 the reactor is uniform. The photocatalytic reaction takes place at
Experiment
Dispersion model P = 39 the surface of the coated packing material at a surface reaction rate
0.01 Tanks-in-series model J = 21 of rs. Langmuir–Hinshelwood monomolecular kinetics are used in
this model [29]. As the dissolved oxygen concentration is constant,
the kinetic law is:
0.008
K LH C s ðt; r; zÞ
rs ¼ dTiO2 In ðrÞkLH ð13Þ
E(t)

0.006 1 þ K LH C s ðt; r; zÞ
where rs (mol s1 m2) is the surface reaction rate, dTiO2
0.004 (g m2) is the density of TiO2 loaded onto the packing, I(r)
(W m2) is the fluence rate at radius r, n is the order relative to I,
kLH (mol m2n g1 Wn s1) is the kinetic constant of degradation rel-
0.002
ative to the mass of TiO2, KLH (m3 mol1) is the adsorption constant,
and Cs (mol m3) is the compound concentration at the surface of the
0 catalyst.
0 100 200 300 400 500 600
Time (s) To estimate the concentration gradient between the bulk and
the catalyst’s surface, the external resistance fraction, fe, was
Fig. 3. Experimental and modeled RTD of reactor. calculated. The external resistance fraction compares the apparent
 A. Cloteaux et al. / Chemical Engineering Journal 249 (2014) 121–129 127

global reaction rate and the mass transfer rate between the bulk 55
and the surface of the catalyst. It is defined as follows: 50
r dp C  C s 45
fe ¼ ¼ ð14Þ
kL C C

Fluence rate (W/m²)

40
where r (mol s1 m3) is the global apparent reaction rate, kL 35
(m s1) is the mass transfer coefficient and C (mol m3) the concen- 30
tration in the fluid. The transfer coefficient depends on the Reynolds
25
number, Re, and the Schmidt number, Sc.
20
qu0 dp
Re ¼ ð15Þ 15
l
10
l 5
Sc ¼ ð16Þ
qDm 0
0 0.002 0.004 0.006 0.008 0.01 0.012
In fixed bed cases, Onda et al. [30] proposed Eq. (17) to calculate Radius (m)
the transfer coefficient in the liquid phase.
 1=3 Fig. 5. Light distribution in reactor for an apparent extinction coefficient of 66 m1.
q 0:49 0:49 0:50
kL ¼ 0:021ðadp Þ Re Sc ð17Þ
lg  
C R ðtÞ
If fe < 0.05, the external resistance is negligible and the concen- XðtÞ ¼ 1   100 ð22Þ
C R0
tration within the bulk, C, equals the concentration at the surface
of the catalyst, Cs. Otherwise, the model must consider the transfer The variation in removal rate is plotted in Fig. 6 for these four
flux, / (mol s1 m3), which is: experiments. After 1 h, the removal rates obtained for the experi-
ments at CR0 = 1.6  102 mol m3, 3.2  102 mol m3,
/ ¼ akL ðC  C s Þ ð18Þ 1 3 1 3
1.7  10 mol m and 3.4  10 mol m were 63%, 61%, 52%
In the dispersion model case, the concentration depends only and 44%, respectively. The external resistance fraction was found
on coordinate z and time. It is assumed that n = 1. The validity of to be approximately 0.033 with a transfer coefficient kL of
this assumption is tested in Section 3.2.4. The kinetics of photoca- 2.5  105 m s1. A low external resistance fraction indicates that
talysis is governed by a 1st order variation law related to the flu- the reaction at the catalyst surface is not fast enough compared
ence rate, which is dependent on r. The mean fluence rate over to the external mass transfer to create a concentration gradient be-
section Imoy (W m2) is applied to avoid the whole model depend- tween the surface and the bulk. Under these experimental condi-
ing on r. This approximation is possible because the chemical tions, the external transfer resistance is negligible. Formaldehyde
kinetics in the fluence rate is first order. Eq. (13) becomes: degradation is therefore limited by chemical kinetics. Moreover,
the lower removal rates for higher initial concentrations correlated
K LH C s ðt; zÞ
r s ¼ dTiO2 Imoy kLH ð19Þ with a Langmuir–Hinshelwood model.
1 þ K LH C s ðt; zÞ As the external resistance fraction (fe) is close to the limit value,
Writing of the material balance for an infinitesimal volume it was decided to retain the mass transfer term (Eq. (18)) to ensure
based on external transfer gives the following Eq. (20): model validity under other flow rate and concentration conditions.
The concentration at the reactor outlet, CR, was determined using
@Cðt; zÞ @Cðt; zÞ @ 2 Cðt; zÞ the algorithm described in Section 2.4.4 and partial differential
e ¼ akL ðCðt; zÞ  C s ðt; zÞÞ  eu þ eD ð20Þ
@t @z @z2 Eq. (20). Constants kLH and KLH are two model adjustment variables,
The material flux between the bulk and the surface equals the which can be used to minimize the least squares difference be-
reaction rate at the surface (Eq. (21)). tween the experimental results and the model-predicted results

akL ðCðt; zÞ  C s ðt; zÞÞ ¼ ars ð21Þ

Eq. (19) is substituted in Eq. (21) and solved for concentrations 100
Cs. 0.016 mol/m3
0.032 mol/m3
80 0.17 mol/m3
3.2.2. Light distribution in reactor
0.34 mol/m3
Removal rate (%)

The light distribution in the reactor, illuminated by six lamps,

was calculated using the method described in Section 2.6. This is 60
illustrated in Fig. 5, where the entire packed bed is shown to be
irradiated, including its center. The fluence rate is 35.0 W m2 at
the center and 53.8 W m2 at the outer boundary. Fluence rate 40
only varies slightly with radius. This is another argument justifying
the use of the average fluence rate in the kinetic law. Under exper-
20
imental conditions, the mean fluence rate over the section is
42.1 W m2.
0
0 0.2 0.4 0.6 0.8 1
3.2.3. Determining the kinetics of photocatalysis
Degradation was performed for four initial formaldehyde Time (h)
concentrations. The removal rate was calculated for the outlet Fig. 6. Comparison of experimental results (markers) and model predictions (lines)
reactor concentration, CR, at a given time and the initial concentra- for different initial concentrations (kLH = 2.5  108 mol m2 s1 g1 W1 and
tion, CR0: KLH = 3.6 m3 mol1).
128 A. Cloteaux et al. / Chemical Engineering Journal 249 (2014) 121–129

under the same conditions. Optimal adjustment was obtained with 100
kLH = 2.5  108 mol m2 s1 g1 W1 and KLH = 3.6 m3 mol1. The 2,5e-7 m3/s
removal rates for CR0 = 1.6  102 mol m3, 3.2  102 mol m3, 5,0e-7 m3/s
1.7  101 mol m3 and 3.4  101 mol m3 are 63%, 62%, 52% 80 1,0e-6 m3/s
and 43% respectively. Fig. 6 illustrates the variation in removal rate

Removal rate (%)

calculated using the model for different initial concentrations.
60

3.2.4. Influence of fluence rate on removal rate

In Section 3.2.3, degradation was shown to be limited by the 40
chemical kinetics under experimental conditions. Experiments
with different mean fluence rates were carried out to validate
the 1st order variation law with respect to light flux. The low con- 20
version per pass observed in Section 3.2.3 is another reason for
using the mean fluence rate in the kinetic law. The mean fluence
rates in the reactor were 11.4, 21.0, 27.8 and 42.1 W m2, with ini- 0
0 0.2 0.4 0.6 0.8 1
tial concentrations and flow rates of 1.7  101 mol m3 and
5  107 m3 s1, respectively. Fig. 7 illustrates the mild influence Time (h)
of fluence rate on the formaldehyde removal rate in the reactor. Fig. 8. Comparison of experimental results (markers) and model predictions
The variations obtained using the model under the same condi- (lines) for different flow rates (kLH = 2.5  108 mol m2 s1 g1 W1 and
tions are also plotted in Fig. 7. KLH = 3.6 m3 mol1).
These results show that the model predicts the formaldehyde
concentration at the reactor outlet fairly well. The assumption of
a 1st order power law with respect to the mean fluence rate was can be considered to effectively agree with the experimental re-
validated, as were the kinetic constants determined in sults. The model results plotted in Fig. 8 show that the removal rate
Section 3.2.3. increases slightly as the recirculation rate decreases. This observa-
tion is consistent with the fact that there is no transfer limitation. If
transfer were limited, the removal rate would be improved by
3.2.5. Influence of flow rate on removal rate increasing the recirculation rate. Under the experimental condi-
The validity of the model was tested for different flow rates. The tions retained, this study provided no evidence of transfer limita-
influence of flow rate was taken into account in the model through tion. Thus, formaldehyde degradation would be limited by the
Eq. (S4), calculating the dispersion coefficient, and through Eq. (17), mass transfer if the chemical kinetics was faster or if the mass
calculating the transfer coefficient. The experimental flow rates transfer coefficient, kL, was lower. Gérardin et al. [31] showed a
were 2.5  107 m3 s1, 5.0  107 m3 s1 and 1.0  106 m3 s1 mass transfer limitation in a photocatalytic reactor with supported
(0.9 L h1, 1.8 L h1 and 3.6 L h1), with an initial concentration of catalyst based on a similar gas-phase model.
1.7  101 mol m3 and an average fluence rate in the reactor of
40 W m2. For experiments at both 2.5  107 m3 s1 and
5.0  107 m3 s1, the removal rate after 1 h of irradiation was 4. Conclusion
52%. In the experiments at 1.0  106 m3 s1 the removal rate
was slightly lower (48%). The model predicts removal rates after The aim of this study was to model a fixed bed photocatalytic
1 h of irradiation of 54%, 52% and 52% for flow rates of 2.5  107 reactor designed to degrade formaldehyde, which is toxic to hu-
m3 s1, 5.0  107 m3 s1 and 1.0  106 m3 s1, respectively. mans, and to determine the chemical kinetics of the degradation
Thus, there is a difference of a few percent between the exper- reaction. To do this, the hydraulics of a fixed bed reactor was char-
imental data and the calculations. However, considering a mea- acterized under open-loop and closed-loop experimental condi-
surement uncertainty of approximately 5%, the model predictions tions. The dispersion model satisfactorily modeled the reactor’s
hydraulic behavior. The material balance was established based
on the dispersion model and taking both the mass transfer phe-
100
nomenon and the surface chemical reaction into account. Experi-
40 W/m²
mental data obtained for different initial concentrations were
27 W/m²
used to determine the Langmuir–Hinshelwood apparent kinetic
80 20 W/m²
constant. The model predicts a variation in concentration with
11 W/m²
time for different irradiation and flow rate conditions based on
Removal rate (%)

60
fixed Langmuir–Hinshelwood parameters. The results of this study
show that a fixed bed photocatalytic reactor can efficiently degrade
formaldehyde in aqueous solution. The novelty of this work lies in
40 the development of a model which takes hydraulics, light distribu-
tion, chemical kinetics and mass transfer in the reactor into ac-
count. Photocatalysis is a heterogeneous process, therefore the
20 light transmission problem and the mass transfer limitation must
be considered whatever the reactor design used. In the literature,
these issues are rarely discussed. However the light transmission
0 and mass transfer phenomena must be understood by engineers
0 0.2 0.4 0.6 0.8 1
if efficient photocatalytic reactors are to be designed. The chemical
Time (h) kinetic laws proposed in the literature can involve different influ-
Fig. 7. Comparison of experimental results (markers) and model predictions
encing parameters. The presence of co-pollutants [32], the influ-
(lines) for different mean fluence rates (kLH = 2.5  108 mol m2 s1 g1 W1 and ence of pH [33] or the dissolved oxygen concentration [34] are
KLH = 3.6 m3 mol1). sometimes considered. The model proposed in this study could
 A. Cloteaux et al. / Chemical Engineering Journal 249 (2014) 121–129 129

be extended by applying these more complex kinetic models. It of mass transfer and chemical reaction steps in the photodegradation process,
J. Photochem. Photobiol. Chem. 177 (2006) 212–217.
would also be interesting to test the model under experimental
[12] A.V. Emeline, V. Ryabchuk, N. Serpone, Factors affecting the efficiency of a
conditions in which the reaction is limited by mass transfer. This photocatalyzed process in aqueous metal–oxide dispersions: prospect of
study contributes to the photo-reaction engineering field since it distinguishing between two kinetic models, J. Photochem. Photobiol. Chem.
models the photocatalytic reactor based on chemical reaction engi- 133 (2000) 89–97.
[13] I.J. Ochuma, O.O. Osibo, R.P. Fishwick, S. Pollington, A. Wagland, J. Wood, et al.,
neering principles. The model proposed in this study should help to Three-phase photocatalysis using suspended titania and titania supported on a
design and scale reactors to treat liquid effluents containing form- reticulated foam monolith for water purification, Catal. Today 128 (2007) 100–
aldehyde, such as industrial or hospital waste water. A Photo-reac- 107.
[14] R.M. Alberici, W.F. Jardim, Photocatalytic destruction of VOCs in the gas-phase
tor with a fixed catalyst is a highly advantageous technological using titanium dioxide, Appl. Catal. B Environ. 14 (1997) 55–68, http://
option for effluent treatment because there is no need for filtration. dx.doi.org/10.1016/S0926-3373(97)00012-X.
In addition, human exposure to, and environmental emission of, [15] O.M. Alfano, R.L. Romero, A.E. Cassano, Radiation field modelling in
photoreactors—I. Homogeneous media, Chem. Eng. Sci. 41 (1986) 421–444.
titanium dioxide nanometric particles are avoided. This study [16] O.M. Alfano, R.L. Romero, A.E. Cassano, Radiation field modelling in
shows that transfer limitation in a packed bed reactor can be pre- photoreactors—II. Heterogeneous media, Chem. Eng. Sci. 41 (1986) 1137–
vented by using a suitable recirculation rate. However, it will only 1153.
[17] M. Pasquali, F. Santarelli, J.F. Porter, P.-L. Yue, Radiative transfer in
be possible to use this type of heterogeneous photocatalytic reac- photocatalytic systems, AIChE J. 42 (1996) 532–537.
tor to treat effluents containing low concentrations of suspended [18] G.B. Roupp, J.A. Nico, S. Annangi, R. Changrani, R. Annapragada, Two-flux
matter since suspended matter could foul the photocatalytic radiation-field model for an annular packed-bed photocatalytic oxidation
reactor, AIChE J. 43 (1997) 792–801, http://dx.doi.org/10.1002/aic.690430324.
packed bed, thereby reducing its efficacy.
[19] M. Faure, Purification de l’air ambiant par l’action bactéricide de la
photocatalyse, INPL (2010).
Appendix A. Supplementary material [20] C.T. Culbertson, S.C. Jacobson, J. Michael Ramsey, Diffusion coefficient
measurements in microfluidic devices, Talanta 56 (2002) 365–373, http://
dx.doi.org/10.1016/S0039-9140(01)00602-6.
Supplementary data associated with this article can be found, in [21] Q. Li, P. Sritharathikhun, S. Motomizu, Development of novel reagent for
the online version, at http://dx.doi.org/10.1016/j.cej.2014.03.067. Hantzsch reaction for the determination of formaldehyde by
spectrophotometry and fluorometry, Anal. Sci. 23 (2007) 413–417.
[22] L.L.P. Lim, R.J. Lynch, S.-I. In, Comparison of simple and economical
References photocatalyst immobilisation procedures, Appl. Catal. Gen. 365 (2009) 214–
221.
[1] M.L. Sauer, D.F. Ollis, Acetone oxidation in a photocatalytic monolith reactor, J. [23] S. Brunauer, P.H. Emmett, E. Teller, Adsorption of gases in multimolecular
Catal. 149 (1994) 81–91, http://dx.doi.org/10.1006/jcat.1994.1274. layers, J. Am. Chem. Soc. 60 (1938) 309–319.
[2] A. Mills, N. Elliott, I.P. Parkin, S.A. O’Neill, R.J. Clark, Novel TiO2 CVD films for [24] L.A. Belfiore, J.J. Way, L. Zhang, Transport Phenomena for Chemical Reactor
semiconductor photocatalysis, J. Photochem. Photobiol. Chem. 151 (2002) Design, Wiley Online Library, 2003.
171–179. [25] J. Villermaux, Génie de la réaction chimique: conception et fonctionnement
[3] H. Choi, E. Stathatos, D.D. Dionysiou, Photocatalytic TiO2 films and membranes des réacteurs, 2ème ed., Lavoisier Tec & Doc, Paris, 1982.
for the development of efficient wastewater treatment and reuse systems, [26] O.M. Alfano, R.L. Romero, A.E. Cassano, Radiation field modelling in
Desalination 202 (2007) 199–206. photoreactors—I. Homogeneous media, Chem. Eng. Sci. 41 (1986) 421–444,
[4] P. Chin, D.F. Ollis, Decolorization of organic dyes on Pilkington Activ™ http://dx.doi.org/10.1016/0009-2509(86)87025-7.
photocatalytic glass, Catal. Today 123 (2007) 177–188, http://dx.doi.org/ [27] J.C. André, J.F. D’allest, Industrial photochemistry VIII: light repartition in
10.1016/j.cattod.2007.01.069. heterogeneous cylindrical photoreactors, J. Photochem. 36 (1987) 221–234.
[5] IARC, Formaldehyde, 2-Butoxyethanol and 1-tert-Butoxypropan-2-ol, World [28] J. André, M. Roger, A. Said, J. Villermaux, Photochimie industrielle. X:
Health Organization, Lyon, France, 2006. Répartition de la lumière dans un milieu absorbant et diffusant à l’intérieur
[6] M.F.J. Dijkstra, H. Buwalda, A.W.F. de Jong, A. Michorius, J.G.M. Winkelman, d’un réacteur plan semi-infini, Entropie 23 (1987) 87–92.
A.A.C.M. Beenackers, Experimental comparison of three reactor designs for [29] G. Scacchi, M. Bouchy, J.-F. Foucaut, O. Zahraa, Cinétique et catalyse, Lavoisier
photocatalytic water purification, Chem. Eng. Sci. 56 (2001) 547–555, http:// Tec & Doc, 1996.
dx.doi.org/10.1016/S0009-2509(00)00259-1. [30] K. Onda, E. Sada, Y. Murase, Liquid-side mass transfer coefficients in packed
[7] D.F. Ollis, Kinetic disguises in heterogeneous photocatalysis, Top. Catal. 35 towers, AIChE J. 5 (1959) 235–239, http://dx.doi.org/10.1002/aic.690050220.
(2005) 217–223, http://dx.doi.org/10.1007/s11244-005-3827-z. [31] F. Gérardin, A. Cloteaux, M. Guillemot, M. Faure, J.C. André, Photocatalytic
[8] M. de los M. Ballari, R. Brandi, O. Alfano, A. Cassano, Mass transfer limitations conversion of gaseous nitrogen trichloride into available chlorine—
in photocatalytic reactors employing titanium dioxide suspensions: I. experimental and modeling study, Environ. Sci. Technol. 47 (2013) 4628–
concentration profiles in the bulk, Chem. Eng. J. 136 (2008) 50–65, http:// 4635, http://dx.doi.org/10.1021/es400588m.
dx.doi.org/10.1016/j.cej.2007.03.028. [32] M.S. Vohra, S.M. Selimuzzaman, M.S. Al-Suwaiyan, NH+4-NH3 removal from
[9] M. de los M. Ballari, R. Brandi, O. Alfano, A. Cassano, Mass transfer limitations simulated wastewater using UV-TiO2 photocatalysis: effect of co-pollutants
in photocatalytic reactors employing titanium dioxide suspensions: II. external and pH, Environ. Technol. 31 (2010) 641–654.
and internal particle constrains for the reaction, Chem. Eng. J. 136 (2008) 242– [33] R.-A. Doong, C.-H. Chen, R. Maithreepala, S.-M. Chang, The influence of pH and
255, http://dx.doi.org/10.1016/j.cej.2007.03.031. cadmium sulfide on the photocatalytic degradation of 2-chlorophenol in
[10] M.F.J. Dijkstra, E.C.B. Koerts, A. A.C.M. Beenackers, J.A. Wesselingh, titanium dioxide suspensions, Water Res. 35 (2001) 2873–2880, http://
Performance of immobilized photocatalytic reactors in continuous mode, dx.doi.org/10.1016/S0043-1354(00)00580-7.
AIChE J. 49 (2003) 734–744, http://dx.doi.org/10.1002/aic.690490317. [34] A. Mills, J. Wang, D.F. Ollis, Dependence of the kinetics of liquid-phase
[11] A. Bouzaza, C. Vallet, A. Laplanche, Photocatalytic degradation of some VOCs in photocatalyzed reactions on oxygen concentration and light intensity, J. Catal.
the gas phase using an annular flow reactor: determination of the contribution 243 (2006) 1–6, http://dx.doi.org/10.1016/j.jcat.2006.06.025.