Alireza Khataee*,
Mehrangiz Fathinia and
Tannaz Sadeghi Rad
Research Laboratory of Advanced Water and Wastewater Treatment Processes, Department of Applied Chemistry, Faculty of Chemistry, University of Tabriz, 51666-16471 Tabriz, Iran. E-mail: a_khataee@tabrizu.ac.ir; ar_khataee@yahoo.com; Fax: +98 41 33340191; Tel: +98 41 33393165
First published on 20th April 2016
The removal of nalidixic acid (NAD) through the clinoptilolite nanorod (CN)-catalyzed ozonation process was modeled by three types of kinetic approaches. By using the glow discharge plasma (GDP) technique, natural clinoptilolite microparticles (NCMs) were successfully converted to CNs. The samples were characterized by scanning electron microscopy (SEM) and Brunauer–Emmett–Teller (BET) analysis. The impacts of ozone inlet flow rate, catalyst concentration, pH and NAD initial concentration were examined to find out the kinetic characteristics of the heterogeneous catalytic ozonation. Based on the intrinsic elementary reactions of the heterogeneous catalytic ozonation process, a novel kinetic model was developed and validated. An empirical kinetic model and an artificial neural network (ANN) model were established in order to appraise the accuracy of the proposed intrinsic kinetic model by a 3-layer feed-forward back-propagation network with the topology of 4:
14
:
1. The error functions and analysis of variance (ANOVA) were used to compare the performance of the three models.
One of the most detected pharmaceuticals in water and wastewater is NAD, which cannot be thoroughly degraded by typical wastewater treatment processes.3 NAD is a widely spread antibacterial agent, and its residual occurrence through natural water, soils and sediments is confirmed.4
Through recent years, advanced oxidation processes (AOPs) have been developed as effective water treatment methods. AOPs can generate highly reactive oxygen species, like hydroxyl radicals (E° = 2.6 eV), which can be applied for the mineralization of refractory organic pollutants.5 Particularly, as an AOP technique, the heterogeneous catalytic ozonation process has received considerable attention due to its promising results among the high-performance water treatment processes.6–9
One of the most commonly used catalysts in the heterogeneous catalytic ozonation is zeolite. Among zeolites, as an abundant and non-toxic compound, clinoptilolite has been utilized in different catalytic processes.10 In spite of the advantages of clinoptilolite, low specific surface area restricts its usage in heterogeneous catalytic ozonation.11 In order to overcome the limitations and increase the specific surface area of catalysts, distinct methods have been utilized.12,13 In recent decades, the nonthermal plasma technique, as a green method for generating nanomaterials and catalyst surface modification, has been utilized extensively.14–16 According to various researches' results, the morphology of a catalyst can be altered to nanostructures during the GDP technique. Thereby, the specific surface area of a catalyst can be improved.14,15 By using a nanocatalyst, the mineralization rate of refractory organic substances is increased considerably through the catalytic ozonation process for water and wastewater treatment.
Heterogeneous catalytic ozonation process involves free radical, catalyst surface reactions and mass transfer processes; thus, the kinetic modeling of the process is complicated. With the aim of studying the process variables and effective parameters, various kinetic models have been developed.17 However, the validation of the developed models was not completely investigated and evaluated using other kinds of models.
In addition, there are a few studies which focused on the kinetic modeling of the heterogeneous catalytic ozonation process for contaminant decomposition.18,19 As an example, Valdés et al., in their research including the heterogeneous and homogeneous catalytic ozonation of benzothiazole in the presence of activated carbon, reported the contribution of the heterogeneous and homogeneous reactions in benzothiazole decomposition.20 They used the reactions in order to just develop a kinetic model for the degradation process. In this context, Zhang et al. investigated the degradation kinetics of a reactive azo dye by ferrous-catalyzed ozonation in a bubble column reactor. The obtained results indicated that the degradation process was well fitted by an irreversible second-order kinetic model.21 Pocostales et al. studied the ozonation of sulfamethoxazole catalyzed by a powder activated carbon. They proposed a mechanism that involved both homogeneous and heterogeneous reactions for the mineralization of the pharmaceutical. Also, regarding the proposed mechanism a mathematical kinetic model was developed using the corresponding mass balances of the main species present.22 Andreozzi et al. investigated the ozonation of pyruvic acid in aqueous solution, catalyzed by Mn(II) and Mn(IV) ions, at three different pH values (1.1, 2.0 and 3.0). A mathematical model of the process was developed taking into account the reactions occurring in the liquid phase and the ozone mass transfer from the gas bubbles. The reactions proposed were then used to establish two alternative kinetic models. The two kinetic models correlated with the experimental data with a fair accuracy at the lowest and at the highest pH values examined.23
In the present study, CNs were modified by the GDP technique and applied as a heterogeneous catalyst in ozonation of NAD. The characterization of the prepared catalyst was conducted by SEM and BET analysis. Based on the intrinsic elementary reactions of the heterogeneous catalytic ozonation process, an innovative kinetic model was established for the degradation of NAD. In addition, the accuracy and precision of the developed model were validated and evaluated using the experimental data and empirical and ANN modeling, respectively.
It should be mentioned that the GDP technique operating conditions were investigated and properly optimized in previous reports.14,24 Ar, O2 and N2 gases were tested as plasma-forming gases and plasma treatment time ranged from 15 to 60 min. The best results were obtained at a treatment time of 60 min, when using N2 as a feeding plasma-forming gas. So, in this work, after evacuating the reactor, N2 gas was introduced to the reactor at a pressure of 53.3 Pa. The duration of the GDP procedure was about 60 min. After that, the process was stopped and CNs were collected for use in the catalytic ozonation process as a catalyst.
Ozone inlet flow rate (L h−1) | Ozone dissolved concentration (mg L−1) | Ozone concentration in the inlet gas (mg L−1) | Ozone concentration in the outlet gas (mg L−1) |
---|---|---|---|
1 | 0.10 | 1.17 | 1.07 |
3 | 0.62 | 2.37 | 1.75 |
5 | 1.31 | 3.79 | 2.48 |
7 | 2.16 | 7.10 | 4.94 |
3 mL of the sample was taken at distinct time intervals. To suppress any ozone development or oxidation of hydroxyl radicals before the analytical determination of NAD, 1 mL of ethanol was added to the samples. By using a UV-visible spectrophotometer (Lightwave S2000, England), the absorbance at the maximum wavelength (λmax = 330 nm) was computed and the removal efficiency of the NAD was measured. By concentrating on other studies, it can be understood that the NAD removal efficiency (%) was determined as the percentage ratio of the removed drug concentration to the initial one. Also, the dissolved ozone concentration was determined by the indigo colorimetric method which was proposed by Bader and Hoigné.25
Regarding the kinetics of the oxidation process, the pseudo-first-order kinetics depicted in eqn (1) can be used to fit the experimental data of NAD removal efficiency in the various processes:
![]() | (1) |
![]() | ||
Fig. 3 Linear relationship of NAD degradation during heterogeneous catalytic ozonation process at (a) different ozone inlet flow rates (experimental conditions: [CNs] = 6 g L−1, pH = 7 and [NAD]0 = 20 mg L−1); (b) various catalyst concentrations (experimental conditions: ozone inlet flow rate = 7 L h−1, pH = 7 and [NAD]0 = 20 mg L−1); (c) different pH of solution (experimental conditions: ozone inlet flow rate = 7 L h−1, [CNs] = 2 g L−1 and [NAD]0 = 20 mg L−1); (d) various initial NAD concentrations (experimental conditions: ozone inlet flow rate = 7 L h−1, [CNs] = 2 g L−1 and pH = 7), fitted by pseudo-first-order model. (e) Comparison of the calculated NAD removal efficiency via developed intrinsic kinetic model (eqn (17)) with experimental removal efficiency. |
Fig. 3a shows that kapp increases gradually with the ozone inlet flow rate increasing up to 7 L h−1. Also, with an increase of ozone inlet flow rate from 1 L h−1 to 7 L h−1, the dissolved ozone concentration altered from 0.10 to 2.16 mg L−1. Also it is worth mentioning that the increase in the concentration of the ozone molecules leads to an increase in the ozone decomposition on the catalyst surface, which leads to the production of reactive oxygen species such as hydroxyl radicals which are responsible for NAD degradation in the catalytic ozonation process.30,31 The obtained results are compatible with the literature studying ozone inlet flow rate influence on catalytic ozonation.32,33
Fig. 3b illustrates that the pseudo-first-order rate constants increase with increasing catalyst concentration from 0.4 g L−1 to 8 g L−1. It has been well mentioned that more surface active sites of the catalyst are provided by a high concentration of catalyst, and thus more ozone molecules decomposed and converted to hydroxyl radicals.30
Fig. 3c shows that kapp continuously increases with increasing pH value. The cause of this phenomenon can be related to the production of more hydroxyl radicals at a higher pH of the solution due to the faster decomposition of ozone.34,35 Hence, producing more hydroxyl radicals leads to an increase of the NAD removal efficiency.36
Fig. 3d shows that the pseudo-first-order removal rate constant of NAD decreases systematically on increasing the NAD initial concentration. This is because of the fact that under the optimum conditions, the flow rate of ozone is constant and also the amount of ˙OH generated by ozone decomposition is constant. Thus, the limited concentration of ˙OH is not sufficient to degrade all of the pollutants thoroughly and finally the removal efficiency is reduced by increasing the NAD concentration.37
Reaction no. | Elementary reaction | Kinetic constant (M−1 s−1) | Ref. |
---|---|---|---|
a IP: refers to intermediate products, DP: refers to degradation products. | |||
1 | O3 + NAD → DP + O3˙− | ||
2 | O3˙− + H+ ↔ HO3˙ | 5.2 × 1010, k− = 3.7 × 104 | 22,31,38 |
3 | HO3˙ → ˙OH + O2 | 5.2 × 1010 | 22,31,38 |
4 | O3 + OH− →HO−2 + O2 | 70 | 22,31,38 |
5 | O3 + HO−2 → HO2˙ + O3˙− | 2.8 × 106 | 22,31,38 |
6 | ![]() |
3.2 × 105, k− = 2.0 × 1010 | 22,31,38 |
7 | O2˙− + O3 → O3˙− + O2 | 1.6 × 109 | 22,31,38 |
8 | 2HO2˙ → O2 + H2O2 | 5.0 × 109 | 22,31,38 |
9 | 2HO3˙ → 2O2 + H2O2 | 5.0 × 109 | 22,31,38 |
10 | ˙OH + HO3˙ → O2 + H2O2 | 5.0 × 109 | 22,31,38 |
11 | ˙OH + NAD → IP | ||
12 | ˙OH + IP → DP | ||
13 | O3 + ˙OH → HO2˙ + O2 | 2 × 109 | 22,31,38 |
14 | H2O2 + ˙OH → HO2˙ + H2O | 2.7 × 107 | 22,31,38 |
15 | HO−2 + ˙OH → HO2˙ + −OH | 7.5 × 109 | 22,31,38 |
Reaction no. | Elementary reaction | Kinetic constant (M−1 s−1) | Ref. |
---|---|---|---|
a S: refers to catalyst surface. | |||
16 | NAD + S → NAD − S | ||
17 | O3 + S → O3 − S | ||
18 | O3 − S + OH− − S → HO−2 + O2 | 70 | 22,31,38 |
19 | HO−2 + O3 → HO2˙ + O3˙− | 2.8 × 106 | 22,31,38 |
20 | ![]() |
3.2 × 105, k− = 2.0 × 1010 | 22,31,38 |
21 | O2˙− + O3 → O3˙− + O2 | 1.6 × 109 | 22,31,38 |
22 | O3˙−+ H + ↔ HO3˙ | 5.2 × 1010, k− = 3.3 × 102 | 22,31,38 |
23 | HO3˙ → ˙OH + O2 | 1.1 × 105 | 22,31,38 |
24 | ˙OH + O3 → HO2˙ + O2 | 2 × 109 | 22,31,38 |
25 | HO−2 + ˙OH → HO2˙ + OH− | 7.5 × 109 | 22,31,38 |
26 | ˙OH + NAD → IP | ||
27 | ˙OH + IP → DP |
Considering Table 2, reactions 1 to 15 are suggested for homogeneous ozonation of NAD. According to reaction 1, ozonide ion radical is formed by the direct reaction of NAD with ozone molecule. The ozonide ion radical provokes a radical mechanism that ends with hydroxyl free radical production (reactions 2 and 3). As can be seen from reactions 4 and 5, ozone reacts with hydroxyl ions and ionic form of hydrogen peroxide in order to generate the hydroxyl radicals. The hydroperoxide radical (HO2˙) is changed to a superoxide ion (O2˙−) by shifting reaction 6 to the right at pH = 7. Finally, via reactions of 10 to 15, hydroxyl radicals commence reacting with NAD, intermediates, ozone and hydrogen peroxide to yield further free radicals, intermediates and so on.
- equilibrium conditions are obtained between the bulk and adsorbed concentrations of NAD;
- in order to calculate the concentration of unstable reactive species such as radicals, the steady state approximation (SSA) is used;
- the reaction between ozone and hydroxyl radical which occurs in homogeneous phase is neglected;
- the recombination of reactive oxygen species with each other is of low probability due to the presence of organic compounds.
The overall reaction rate for the pseudo-first-order homogeneous reaction and second-order catalyst surface reaction of NAD can be represented as eqn (2):
![]() | (2) |
The total concentration of the adsorbed hydroxyl radicals according to SSA is given in eqn (3):
![]() | (3) |
Eqn (3) can be rearranged to eqn (4):
![]() | (4) |
As can be seen from eqn (4), the concentration measurement of [NAD] seems to be difficult.43 Thus, by considering the assumption given in eqn (5) and eqn (4), eqn (4) can be altered to eqn (6).
[NAD]ads = [NAD]0 − [IP] ⇒ k26[NAD]ads + k27[IP] = k26[NAD]0 − k26[IP] + k27[NAD]0 − k27[NAD]ads ≈ (k26 + k27)[NAD]0 | (5) |
By substituting eqn (5) in (4), eqn (6) is obtained:
![]() | (6) |
In order to state the concentration of the radical species, eqn (7)–(10) are used by applying SSA:
![]() | (7) |
![]() | (8) |
![]() | (9) |
![]() | (10) |
The concentration of [HO2−] can be obtained by eqn (11):
![]() | (11) |
By considering the fact that the ozone inlet flow rate and the specific surface area of catalyst are constant during the experiments, the concentration of OH− remains constant, too. Consequently, instead of k18 [OH−], k′18 can be used and eqn (12) is obtained:
![]() | (12) |
By replacing eqn (12) in (10), eqn (10) in (9), eqn (9) in (8), eqn (8) in (7) and finally eqn (7) in (6), the expression for the ˙OH concentration is obtained from eqn (13):
![]() | (13) |
The expression for the total degradation rate of NAD can be written as eqn (14):
![]() | (14) |
Eqn (14) can be altered to a pseudo-first-order rate equation by considering eqn (15) as a kinetic rate constant:
![]() | (15) |
Eqn (15) can be simplified to eqn (16) by considering α = khomo, β = 2kheterok′18 and γ = k26 + k27 as the constants of eqn (15):
![]() | (16) |
Eqn (16) elucidates the dependency of the pseudo-first-order kinetic constant on dissolved ozone concentration and initial NAD concentration. Constants of eqn (16) such as α, β and γ were computed by using a least squares curve fit method for the given dissolved ozone concentration and initial NAD concentration and which are given in Table 4. Finally, by combining eqn (1) and (16) and by considering the equation constants, the expression for predicting the NAD removal efficiency (RE) can be defined as eqn (17):
![]() | (17) |
In order to assess the performance of the obtained kinetic model (eqn (17)), the dispersion diagram of the predicted NAD removal efficiency calculated from eqn (17) against the experimental removal efficiency is indicated in Fig. 3e. The correlation coefficient (R2) is 0.980 which implies a significant correlation between the experimental and theoretical NAD removal efficiency calculated by using the developed kinetic model (eqn (17)). Based on the obtained R2, it can be proposed that the obtained model can notably represent the real trend and the mechanism of the heterogeneous catalytic ozonation process. It can be deduced that the proposed mechanism and assumptions are valid. Also, the process is modeled by empirical modeling based on nonlinear regression analysis and ANN in the following sections, to evaluate further the efficiency of the developed model (eqn (17)).
kapp = μi(Xi)θi ⇒ μia(X1)θ1 × (X2)θ2 ×, …, × (Xn)θn | (18) |
kapp = 0.254[pH]0.707[ozone inlet flow rate]0.525[NAD]−0.447[catalyst]0.414 | (19) |
![]() | ||
Fig. 4 The effect of (a) ozone inlet flow rate, (b) catalyst concentration, (c) pH, and (d) initial NAD concentration on pseudo-first-order kinetic constant (kapp). (e) Comparison of the calculated NAD removal efficiency via developed empirical kinetic model (eqn (20)) with experimental values. For experimental details, refer to Table 5. |
Parameter | Range | μ | θ | R2 |
---|---|---|---|---|
pH | 3–9 | 0.0260 (±0.0104) | 0.7070 (±0.1357) | 0.980 |
Ozone inlet flow rate (L h−1) | 1–7 | 0.0298 (±0.0072) | 0.5250 (±0.1012) | 0.971 |
[NAD]0 (mg L−1) | 10–50 | 0.31050 (±0.1006) | −0.4470 (±0.0831) | 0.951 |
Catalyst concentration (g L−1) | 0.4–8 | 0.0526 (±0.0124) | 0.4140 (±0.1758) | 0.950 |
Finally, by incorporating the presented eqn (1) and (19), (20) can be formed as an empirical kinetic expression for calculating NAD removal efficiency:
REpredicted = 1 − exp(−0.254[pH]0.707[ozone inlet flow rate]0.525[NAD]−0.447[catalyst]0.414 × t) | (20) |
In order to validate the empirical model (eqn (20)), the experimental NAD removal efficiencies were plotted against the calculated ones and the results are depicted in Fig. 4e. The high correlation coefficient (R2 = 0.998) confirms that the calculated removal efficiency and the experimental data are in good agreement. So, it can be stated that the empirical modeling of the process is a simple and effective method for defining the interdependence of the kinetic rate constants of the process with the main operational parameters.
![]() | (21) |
Yt = ϕ(Dt + Bt) | (22) |
X1, …, XM are the data of the input, Wt1, …,WtM are the weights of neuron t, Dt is the output of the linear combiner due to the input data, Bt is the bias, ϕ is the transfer function and Yt is the neuron output data.
In this section, modeling of the heterogeneous catalytic ozonation process is demonstrated by using an ANN. The samples were divided into training, validation and test subsets of 69, 15 and 15 samples. In order to figure out the validating ability of the model, the validation and test sets were randomly selected from the experimental data. By normalizing the input data in the range of −1 to 1, the pre-processing was performed and all the data (Yi), containing the training, validation and test sets, were scaled to a new value Ynorm by using eqn (23):47
![]() | (23) |
![]() | (24) |
![]() | (25) |
purelin(x) = x | (26) |
As described in eqn (27), the ANN model performed a nonlinear functional assessment from the previous observation to the future values (Yt):45
Yt = f(Xi,Wti) + Bt | (27) |
A schematic of a typical ANN with four neurons in the input, fourteen neurons in the hidden layer and one neuron in the output is represented in Fig. 5a. Ozone inlet flow rate (L h−1), catalyst concentration (g L−1), pH and initial NAD concentration (mg L−1) were the input variables to the feed-forward neural network. Also, the output variable was the NAD removal efficiency.
The number of the nodes was changed from 1 to 19, in order to determine the optimum number of hidden neurons. A mean squared error (MSE) function was applied in order to compare the performance of the ANN for the various configurations. Eqn (28) defines the MSE properly:45,49
![]() | (28) |
The relationship among the network error and the number of neurons in the hidden layer for the train, validation and test data is shown in Fig. 5b. It can be clearly seen that the MSE is minimum at about 14 neurons for the train, validation and test data. Thus, as the best topology for modeling approach, a network with 14 neurons in the hidden layer is selected due to its best behavior without over-fitting.
In order to compute the train, validation and test errors, all output results were altered to the original scale and compared with the experimental values. The comparisons among the calculated and the experimental values of the variables of the output for the train, validation, test and all sets of data are represented in Fig. 6a–d. The reported R2 values of Fig. 6 are 0.991, 0.994, 0.994 and 0.991, which prove the fact of the small differences present among the experimental and predicted values of the output variable for the various subsets. It can be concluded that the developed topology (4:
14
:
1) can appropriately predict the NAD removal efficiency. Also, this topology can predict the processes' suitable generalization ability.
![]() | (29) |
M is the number of target data, Yi,pred and Yi,exp are the prediction of the model and the experimental response and i is the data index. The MSE, MAE and R2 values are represented in Table 6 for each developed model. As can be observed from Table 6, the R2 values for intrinsic kinetic, empirical kinetic and ANN models are 98.0%, 99.8% and 99.1%, respectively. The largest deviation is observed for the ANN model compared to the other two models. For further investigation of the developed models' performances, analysis of the variance was performed. The results are indicated in Table 7. By considering the ANOVA results, it can be deduced that there are no considerable differences between the three models for predicting the NAD removal efficiency. Thus, the three developed models have the capability for anticipating the NAD removal efficiency. By investigating the results, it is evident that the intrinsic kinetic model is developed depending on the precise elementary reactions which are involved in heterogeneous catalytic ozonation (see Section 3.5) and also that the model assumptions are valid. Therefore, the heterogeneous catalytic ozonation process can be well modeled by the three proposed models. By considering the fact that all kinetic models have advantages and disadvantages, they can be selected by regarding the operating conditions of the system and their performances. The kinetic model based on the intrinsic elementary reactions can be used for modeling heterogeneous catalytic ozonation with different reactor configurations and can be developed for scaling up the existing objectives. In order to investigate the influence and the contribution of the operational parameters, the empirical model is the preferred method. By using the empirical method, the notable main factors are identified, so the complexity of the system is lessened. Of course, it is necessary to mention that the ANN method is a nonlinear and simple mathematical model. By using ANN, it is not necessary to explain the reactions occurring in the system mathematically.
Developed model | REpredicted | R2 | MSE | MAE |
---|---|---|---|---|
a R2: correlation coefficient, MSE: mean squared error, MAE: mean absolute error. | ||||
Intrinsic | ![]() |
0.980 | 0.129 | 0.026 |
Empirical | REpredicted = 1 − exp(−0.254[pH]0.707[ozone flow rate]0.525[NAD]−0.447[catalyst]0.414 × t) | 0.998 | 0.041 | 0.017 |
Artificial neural network | A three-layered feed-forward back-propagation neural network with 4![]() ![]() ![]() ![]() |
0.991 | 4.212 | 0.015 |
This journal is © The Royal Society of Chemistry 2016 |