Applied Catalysis B: Environmental 64 (2006) 121–130 www.elsevier.com/locate/apcatb
Solar photo-Fenton treatment—Process parameters and process control W. Gernjak a,b,*, M. Fuerhacker b, P. Ferna´ndez-Iban˜ez a, J. Blanco a, S. Malato a b
a Plataforma Solar de Almerı´a-CIEMAT, Carretera Sene´s km 4, 04200 Tabernas (Almerı´a), Spain University of Applied Life Sciences, Department of Water, Air and Environment, Muthgasse 18, 1190 Vienna, Austria
Received 3 August 2005; received in revised form 29 November 2005; accepted 1 December 2005 Available online 6 January 2006
Abstract Photo-Fenton experiments were performed using alachlor as a model compound (initial concentration 100 mg/L) in a compound parabolic collector solar pilot-plant. Three process parameters were varied following a central composite design without star points (temperature 20–50 8C, iron concentration 2–20 mg/L, illuminated volume 11.9–59.5% of total). Under all experimental conditions, complete alachlor degradation, mineralisation of chloride and 85–95% mineralisation of dissolved organic carbon (DOC) was achieved. An increase in temperature, iron concentration and illuminated volume from minimum to maximum value reduced the time required for 80% degradation of initial DOC by approximate factors of 5, 6 and 2, respectively. When process parameter changes were made simultaneously, these factors multiplied each other, resulting in degradation times between 20 and 1250 min. Models were designed to predict the time necessary to degrade 50 or 80% of the initial DOC applying response surface methodology (RSM). Another model based on the logistic dose response curve was also designed, which predicted the whole DOC degradation curve over time very well. The three varied process parameters (temperature, iron concentration and illuminated volume) were independent variables in all the models. Mass balances of hydrogen peroxide consumption showed that the same amount of hydrogen peroxide was always needed to degrade a certain amount of DOC regardless of variations in the process parameters within the range applied. Possible applications of the models developed for automatic process control are discussed. # 2005 Elsevier B.V. All rights reserved. Keywords: Photo-Fenton; Advanced oxidation processes; Wastewater treatment; Solar energy; Response surface methodology
1. Introduction Although adopted as the best available technology, biological treatment can only be partially employed to treat waste water. In the case of non-biodegradable or toxic wastewater sources, alternative treatments have to be employed. Among the chemical oxidative treatments, advanced oxidation processes (AOP) are well known for their capacity for oxidising and mineralising almost any organic contaminant. Compendium reviews of these technologies are available [1–4], but most of the research reported has been performed at laboratory scale, not very much at pilot-plant scale and practically none at full scale. What keeps AOPs from going commercial are their comparatively high costs, although, generally, valid cost figures are difficult to provide, as they are subject to caseby-case differences depending on the particular waste problem, * Corresponding author. Tel.: +34 950387957; fax: +34 950365015. E-mail address:
[email protected] (W. Gernjak). 0926-3373/$ – see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.apcatb.2005.12.002
possible process integration and uncertainties inherent in such estimations for not fully developed processes. Attempts have been made to assess and compare AOPs on the basis of defined figures-of-merit, chemical parameters such as reaction constants [5,6], electrical energy per order [7] and financial parameters [6,8]. Although quantitative process assessment is difficult, several promising cost-cutting approaches have been proposed: (1) Using renewable energy instead electrical energy to drive the process [9–12]. Heterogeneous semiconductor photocatalysis and homogeneous photo-Fenton are the only AOPs which can use sunlight to produce hydroxyl radicals. Of the two, photo-Fenton is known to have higher reaction rates [13]. Furthermore, it uses non-toxic, easy-to-handle reagents. (2) Integration of the AOP as part of a sequence of various treatments, in which the AOP would typically be a pretreatment of non-biodegradable or toxic waste water. Once biodegradability has been achieved, the effluent can be
122
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
transferred to a cheaper biological treatment. The key is to minimise residence time and reagent consumption in the more expensive AOP stage by applying an optimised coupling strategy [14,15]. (3) Maximising AOP reaction rates, which leads to higher throughput and lower capital costs because of the smaller plant size needed [10,14,16]. (4) Improvement of plant components, e.g., the solar collector. The use of non-concentrating compound parabolic collectors (CPC) for this purpose was proposed in the early nineties [17], and this set-up has been tested successfully for solar photo-Fenton [12,18] as well as for heterogeneous photocatalysis [19–21]. Requirements for the solar collector have also been discussed [11,21,22]. (5) Improvement of plant operation and control strategies leading to more automation and lower operating costs. This study focuses on solar photo-Fenton degradation of alachlor as a model contaminant. Several previous studies have discussed the influence of iron concentration and its catalytic behaviour [10,14,16] and temperature [10,23]. However, to our knowledge, only one study examines the result of alternating time intervals with and without illumination [24]. To determine the effect of these process parameters on the degradation of alachlor, a central composite design without star points varying three factors (temperature, iron concentration and collector area) was performed, and the results were analysed by the response surface methodology [14,16,25]. Alachlor is soluble in water (240 mg/L, 25 8C) and moderately toxic to aquatic organisms: EC50 (48 h) water flea (Daphnia magna) 10 mg/L; TL50 (72 h) algae (Selenastrum capricornutum) 0.012 mg/L. Furthermore, it is a persistent herbicide with a half-life in soil and water of over 70 and 30 days, respectively. Its main application is grass and weed control for corn, cabbage, cotton and several other crops [26]. Alachlor is classified by the US Environmental Protection Agency (USEPA) as Type III, that is, toxic and slightly hazardous, and a priority substance (PS) by the European Commission (EC) within the scope of the Water Framework Directive (WFD, Directive 2000/60/EC). Furthermore, alachlor’s molecular structure can be regarded as that of a rather typical non-biodegradable contaminant, having an aromatic ring structure, aliphatic carbon and organically bound chlorine and nitrogen. Therefore, in addition to its importance as a contaminant, its use as a model compound in a generic study on the influence of process parameters is further justified. Finally, several parameters, which can be easily measured on-line, are correlated to process progress. Suggestions for the use of such data in improved process operation strategies are provided. 2. Experimental 2.1. Analysis All measurements were performed from samples filtered through 0.2 mm syringe-driven filters (Millipore Millex-GN).
Dissolved organic carbon (DOC) and inorganic carbon (IC) were measured by a Shimadzu model 5050A TOC analyser. Ammonium concentrations were determined with a Dionex DX-120 ion chromatograph equipped with a Dionex Ionpac CS12A 4 mm 250 mm column. Chloride and nitrate concentrations were measured with a Dionex DX-600 ion chromatograph using a Dionex Ionpac AS11-HC 4 mm 250 mm column. Alachlor concentration was analysed using reversephase liquid chromatography with UV detector in an Agilent Technologies, series 1100 HPLC-UV with C-18 column (Phenomenex LUNA 5 mm, 3 mm 150 mm). Complete alachlor degradation means degradation below the detection limit of 0.1 mg/L. All other parameters are measured much above the quantification limit. For photometric measurements and recording of UV spectra, a Unicam-2 spectrophotometer was used. Iron determination was done by colorimetry with 1,10-phenantroline [27]. Hydrogen peroxide concentrations were analysed by iodometric titration. pH, oxidation reduction potential (ORP), temperature (T) and dissolved oxygen (DO) were measured on-line in the pilotplant by the corresponding WTW Sensolyt system electrodes. Global UV (300–400 nm) irradiation in the solar plant was recorded by a Kipp&Zonen CUV3 detector at the same 378 inclination as the reactor modules. That way incident UVradiation could be evaluated as a function of time of day taking into account cloudiness and other environmental variations. Experiments could thus be compared using a corrected illumination time t30W (Eq. (1)) [18]. t30W;n ¼ t30W;n1 þ Dtn
UV ; 30
Dtn ¼ tn tn1
(1)
where tn is the experimental time for each sample and UV is the average solar ultraviolet radiation measured during Dtn. In this case, t30W refers to a constant solar UV power of 30 W/m2 (typical solar UV power on a perfectly sunny day around noon). All measured parameters have an estimated relative error of about 2% (on-line analysis) to 5% (laboratory measurements). However, standard deviation of experiments repeated under the same experimental conditions (compare Table 1 and Fig. 2a) is 10–11% of the corresponding value with regard to t30W. This error can be attributed to the fact that the experiments are pilotplant experiments under solar conditions. While experimental conditions such as iron concentration can be replicated well, the intensity of solar radiation cannot, as it varies with time of day. So, the use of the corrected illumination time t30W partly compensates the effect of changing radiation, but it cannot completely avoid its influence. 2.2. Pilot-plant The pilot-plant, consisting of compound parabolic collectors exposed to sunlight, a reservoir tank, a recirculation pump and connecting tubing, was operated in batch mode. The collector consists of 20 Pyrex absorber tubes with an inner diameter of 46.4 mm. The reflectors are made of aluminium with a concentration factor of one. Collector area is 4.16 m2,
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
123
Table 1 Central composite design without star points Experiment
Centre 1 Centre 2 Centre 3 Cube 1 Cube 2 Cube 3 Cube 4 Cube 5 Cube 6 Cube 7 Cube 8
Process variables
Measured 2
cFe (mg/L)
T (8C)
A (m )
DOC50% t30W
11 11 11 2 2 2 2 20 20 20 20
35 35 35 50 20 50 20 50 20 50 20
2.50 2.50 2.50 0.83 0.83 4.16 4.16 0.83 0.83 4.16 4.16
45 54 46 141 703 50 308 12 110 5.9 39
(min)
Model 1 DOC80% t30W
75 92 77 252 1060 77 375 35 181 18 69
(min)
DOC50% t30W (min)
DOC80% t30W (min)
38 38 38 140 703 61 307 22 111 10 49
62 62 62 254 1058 91 378 46 192 17 69
DOC50% DOC80% Set-up, main DOC and regression results for t30W and t30W calculated with Model 1. Initial alachlor concentration was always 100 mg/L. A = 0.83, 2.50 and 4.16 m2 means that 11.9, 35.7 and 59.5% of the total volume were illuminated.
illuminated volume (Vi) when the collector is completely exposed to sunlight is 44.6 L and total volume (VT) 70–82 L, depending on how full the tank is. Total volume was 75 L for all experiments in this study. The pilot-plant is equipped with online measurement sensors for T, pH, ORP and DO. The plant also incorporates heating and cooling devices to control reaction solution temperatures during an experiment. A flow diagram of the plant is depicted in Fig. 1. 2.3. Experimental set-up and statistical evaluation Photo-Fenton experiments were performed as follows: alachlor was homogeneously dissolved in the pilot-plant with the collectors covered, the pH was adjusted to 2.7 with
sulphuric acid, pre-dissolved ferrous sulphate was added and the first sample was taken after 15 min of homogenisation. Then hydrogen peroxide was added, and after 30 min of dark Fenton reaction, a zero-illumination-time sample was taken, and illumination began. After the start of illumination, periodic samples were taken and hydrogen peroxide concentration was kept between 200 and 400 mg/L by addition simultaneously compensating consumption. The experimental set-up of the experiments performed within a central composite design without star points and main DOC results are given in Table 1. Apart from the central composite design, a dark Fenton control experiment was performed in a magnetically stirred, temperature-controlled, 5L flask with 20 mg/L iron at 50 8C.
Fig. 1. Flow diagram of pilot-plant.
124
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
All statistical evaluation and calculations were done by multivariate linear regression and Levenberg–Marquardt nonDOC50% linear curve fitting using Origin v7.03 software. t30W and DOC80% t30W values given in Table 1 were obtained by linear interpolation between the two points adjacent to the limit value. DOC80% For Cube 2 and Cube 4 experiments, t30W was obtained by linear extrapolation from the last three measured points. This method provides sufficient accuracy (relative error 3–5%) compared to the overall repeatability of the experiments as detected by the repetition of the centre points of the factorial design (relative error 10–11%, compare Fig. 2a and Table 1). 3. Results and discussion 3.1. Degradation of alachlor Fig. 2a–c show the DOC degradation of alachlor versus t30W for the experiments performed (for experiment details refer to Table 1). Several qualitative points can be deduced from these graphics. First of all, DOC degradation was confirmed under all the experimental conditions tested in the photo-Fenton experiments, even at the rather low iron concentration of 2 mg/L (see Fig. 2b). Repeatability of the results is confirmed (see Fig. 2a).
Fig. 2. (a–c) DOC degradation curves measured and predicted with Model 2; cube points.
Referring to the values for 50 and 80% degradation given in Table 1 for the centre experiments, the standard deviations are 4.9 and 9.3 min, respectively, which corresponds to 10 and 11% of the mean values. The beneficial effect of increased temperature and iron concentration is confirmed as well (see Fig. 2b and c). Furthermore, Fig. 2b and c show that by reducing the illuminated area from 4.16 to 0.83 m2 (uncovering only part of the CPC) the reaction rate decreases with respect to t30W. But while the illuminated area is reduced five times, the real treatment time increases only by a factor of 2.5 instead of 5, as might be expected if all the reactions were to be induced by photochemical processes (at least as a rate-limiting step involved in the recycling of ferrous iron). Observing Eq. (1), this means that only about half the photons are necessary with less illuminated area. This suggests that an important part of the reactions are thermally induced in the dark. Several possibilities could explain the difference in the number of photons needed for degradation depending on the relationship between dark and illuminated reactor volume. Either intermediates are formed under illumination, which boost the reaction further after leaving the illuminated reactor zones (e.g., hydroquinones/quinones maintaining the catalytic iron cycle [28]), or intermediates are formed in the dark, which then react quickly under illumination (e.g., organic acids forming photo-active complexes with ferric iron). A combination of both explanations is also possible. A dark Fenton control experiment performed at 20 mg/L iron and 50 8C yielded the highest reaction rates in the experimental region investigated. Fig. 2b shows that although degradation was confirmed, the reaction was considerably slower than the corresponding experiments under illumination. Furthermore, it seems that DOC degradation cannot be achieved to the same extent as under illumination and intermediates produced in the degradation process slowed down the reaction further. Therefore, it may be concluded that illumination is necessary for high DOC degradation to be achieved. From a practical point of view, this means that lowering the ratio of illuminated to total volume is beneficial because fewer photons are required, meaning smaller collector area and lower capital costs. The limitations of this approach to optimise the effect of promoting incident photons would depend on the residence time in dark and illuminated zones in the reactor system. Quantification of the effects of temperature, iron concentration and variation of collector area will be dealt with later in this paper. The stoichiometric demand for hydrogen peroxide necessary to completely oxidise 100 mg/L of alachlor can be calculated with Eq. (2) as 12.6 mM. As can be seen in Fig. 7, hydrogen peroxide consumption had to be two to three times higher if 80% DOC degradation was to be achieved. The dark Fenton experiment consumed even more hydrogen peroxide in accordance with Eqs. (3)–(9), where reactions involved in the catalytic iron cycle are described. Sychev and Isaak [29] have reported the reaction rates given. The dark reactions reducing ferric iron consume hydrogen peroxide molecules and produce only a less active peroxyl radical instead of a hydroxyl
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
125
radical (Eq. (4)). This peroxyl radical can reduce another ferric iron ion forming oxygen (Eq. (6)). So all together, two ferrous iron ions would be recycled at the expense of one hydrogen peroxide molecule, but without generating any oxidising species. C14 H20 ClNO2 þ 34H2 O2 ! 14CO2 þ NH4 Cl þ 42H2 O Fe2þ þ H2 O2 ! Fe3þ þ OH þ OH
(2)
ð5376 M1 s1 Þ (3)
Fe3þ þ H2 O2 ! Fe2þ þ HO2 þ Hþ Fe2þ þ HO2 ! Fe3þ þ HO2
ð12 102 M1 s1 Þ (4)
ð0:721:5 106 M1 s1 Þ (5)
Fe3þ þ HO2 ! Fe2þ þ O2 þ Hþ ð0:332:1 106 M1 s1 Þ (6) Fe
2þ
þ OH ! Fe
3þ
þ OH
8
ð2:65 10 M
1 1
s Þ
(7)
½Fe3þ ðOH Þx ðH2 OÞy þ hn ! Fe2þ þ ðx 1ÞOH þ yH2 O þ OH
(8)
½Fe3þ ðRCO2 Þ þ hn ! Fe2þ þ R þ CO2 "
(9)
Fig. 3 shows other parameters measured in a typical degradation experiment. As can be observed, mineralisation of the compound is not only confirmed by the decrease in DOC, but also by the complete transformation of organic chlorine into chloride and organic nitrogen into nitrate and ammonium. Both heteroatoms are attached in the aliphatic side chains of the aromatic ring. Typically, attack on the aromatic ring is faster compared to attack on aliphatic side chains. This explains the gradual mineralisation of the nitrogen. Chloride release begins immediately, indicating that chloride substitution can take place at any stage of the degradation process. Nevertheless, complete release is only achieved at the end of the experiment. It can be clearly seen that alachlor conversion into oxidised intermediates does not mean direct mineralisation of DOC, as almost all alachlor is converted before DOC begins to decrease.
Fig. 3. Typical degradation experiment (Cube 6). Alachlor conversion and formation of inorganic ions.
Fig. 4. Typical degradation experiment (Cube 6). HPLC chromatograms at 225 nm wavelength of detection showing the formation of intermediates. Alachlor is the chromatogram before H2O2 addition and 0 min refers to the chromatogram after the Fenton reaction in the dark.
Fig. 4 shows HPLC chromatograms recorded at 225 nm wavelength of the same experiment. At least 12–15 peaks can be distinguished as intermediates, however neither identifying these compounds nor investigating the degradation pathway was within the scope of this study as this was done before by Pen˜uela and Barcelo´ [30]. The disappearance of the peaks in the chromatograms at 225 nm during degradation also confirms the disappearance of aromatic compounds, which are at least slightly absorbing at this wavelength. Fig. 5b shows UV absorbance by the samples at 254 and 300 nm. Although absorption at 254 is commonly used as an
Fig. 5. Typical degradation experiment (Cube 6). (a) DO, H2O2 consumption, ORP; (b) UV absorption (1 cm pathlength), pH.
126
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
indicator for the content of aromatic compounds (DIN 38404C3), in the case of photo-Fenton, this is not the only cause of absorption at 254 nm, which is also due to dissolved iron and hydrogen peroxide. So to avoid at least the interference of hydrogen peroxide, it is advisable to observe absorption at 300 nm. The drop in the aromatic compound content correlates to the absorption of the solution as depicted in Fig. 5. Conversion of aromatic compounds to oxidised molecule fragments, which then lose their aromaticity, can usually lead to detoxification without need of complete mineralisation. Typical early intermediate degradation products, such as phenols, hydroquinones and quinones, are also known to be considerably toxic and light absorbing in the ultraviolet region. Consequently, the increasing absorption at the beginning of the reaction can most probably be attributed to such intermediates, while subsequent decreasing absorption reflects the dearomatisation of these compounds. In another previous study by our group, we showed that acute toxicity determined with the Vibrio Fischeri bioluminescence test showed a profile similar to absorption at 300 nm. First, toxicity rose higher than the toxicity of alachlor itself, and then decreased as DOC began to decrease [31]. 3.2. Influence of process parameters—Model 1 To ensure effective detoxification, wastewater detoxification must intrinsically involve process assessment. Measurements of biodegradability enhancement, decrease in toxicity, chemical oxygen demand or dissolved organic carbon are among the most frequently applied figures-of-merit. In this work, DOC degradation was chosen for process evaluation, because other figures-of-merit can be estimated based upon this measure, if empirically determined values are available for the given waste water and oxidation process [12,31]. DOC degradation curves are usually sigmoidal because in the initial degradation stages, the pollutant is transformed into oxygenised intermediates but without a loss of carbon dioxide resulting in initially stable DOC. When degradation proceeds, DOC decrease accelerates until it slows down again in the final stages. This particular behaviour impedes calculating rate constants based on simple rate equations. Alternatively, process efficiency can be compared based on a given DOC decrease [9,12]. According to a previous study [31], detoxification of an alachlor solution can be ensured once DOC degradation reaches 50–80% of the initial value. It was therefore attempted to DOC50% DOC80% develop a model that could predict the time (t30W ; t30W ) required for these levels of degradation. Response surface methodology (RSM) [25] has recently been applied by several authors to modelling tasks related to photoFenton [14,16]. We attempted the same mathematical approach in our work, but fitting degradation to second order polynomial equations was unsuccessful. If all parameters were included, the resulting models were over-fitted and the response surfaces folded, and gave local minima and maxima where their occurrence was not logical from a physical–chemical perspective. If the number of parameters were limited by forward and backward selection of parameters, the models simply were not able to predict the target variables satisfactorily.
One disadvantage was that the model target values have very high relative errors for fast experiments. This is due to least square minimisation, which also takes into account absolute differences. To counter this effect, the model calculation was DOC50% DOC80% ; t30W ) to put directly weighted with the target value (t30W additional weight on the fast experiments, but the approach yielded poor results nevertheless. The main reason for the failure of this methodological approach is probably the wide range of results that the model DOC50% DOC80% must cover. For the fastest experiment t30W and t30W were 6 and 18 min, respectively, for the slowest one 703 and 1060 min. Therefore, the polynomial function approach to the problem proved invalid. We then tried a search for functions which seemed appropriate to describe the problem in a mechanistic approach, given the knowledge existing about the photo-Fenton process and the expected influence of T, Fe and the relationship of collector surface (or illuminated volume) to total volume. After carefully examining the data and attempting several types of functions, we decided that the target function should be a product of functions of the process parameter. To be able to model the curvature in the n-dimensional space we selected the potential function. The resulting equation was Eq. (10), where C, pFe, pT and pA are the four parameters that have to be optimised, while cFe, T and A are the iron concentration, the temperature and the collector surface. This equation was then DOC50% DOC80% used to model t30W and t30W . DOC50% DOC80% or t30W ¼ C cPFeFe T pT A pA t30W
(10)
A second degree polynomial for three factors, including linear, quadratic, cross-product terms and offset, has 10 parameters that have to be adjusted. The advantage in this aspect of Eq. (10) is obvious, as for a given data set for n observations it leaves more degrees of freedom, which for a given Pearson’s coefficient results in a higher Fisher’s value in ANOVA analyses, i.e., a more plausible model. Furthermore, polynomial functions tend to have poor extrapolation qualities, which is another reason for searching for alternative functions more closely related to the physical–chemical behaviour of the system. To distinguish this approach from another described later, we refer to these results as Model 1. The results of parameter optimisation are given in Eqs. (11) and (12). The results of the model applied to the experimental results are given in Table 1. The model results are accurate, except that the relative error is considerable in very fast experiments. Note that the effect of changing a process factor can be estimated directly when the value of the exponent is known. DOC50% t30W ¼ 220; 200 c0:800 T 1:765 A0:515 Fe
(11)
DOC80% t30W ¼ 167; 000 c0:740 T 1:558 A0:638 Fe
(12)
As described, the results of these models are valid for constant process variables, i.e., those which do not change, such as temperature, iron concentration and collector area. The latter would not change in a real case plant either, of
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
127
course, while temperature obviously changes in a solar collector, if no external temperature control is applied. The same is possible for the iron concentration, if there should be precipitation due to high pH or the presence of phosphate, for example. 3.3. Influence of process parameters—Model 2 It would be desirable to have a dynamic model which is capable of predicting the reaction speed at every moment of the degradation process. As mentioned, DOC degradation curves have a sigmoidal form. Of the common curves fitting sigmoidal tendencies, the Boltzmann function and the logistic dose response curve are outstanding for their simplicity. Both have only four parameters to adjust. The problem with the Boltzmann function is that its curvature is symmetric at both sides of the inflexion point, which is not necessarily the case in a DOC degradation curve. So this function must be discarded in favour of the logistic dose response curve, which is commonly used to describe dose response curves in pharmacology. The four parameters of the equation (Eq. (13)) are A1 and A2, which are the initial and the final DOC values (DOCi, DOCf), t1/2, the time when degradation is half-way between DOCi and DOCf and p, an exponent largely determining the curvature and the slope of the curve. DOC in Eq. (13) refers to the measured DOC value at any time during degradation. DOC A1 A2 ¼ þ A2 i 1 þ ðt30W =t1=2 Þ p DOC
(13)
As we used normalised values (DOC/DOCi), A1 was always one, except for the experiments in which 20 mg/L iron was used at 50 8C because these were the only ones in which DOC degradation during the Fenton reaction before illumination was remarkable. Consequently, A1 was set at 0.82, the average value of both experiments at zero illumination time (see Fig. 2c). At the same time, it was assumed that the DOCf is always 5% of DOCi and A2 is 0.05, thus resulting a non-linear fitting problem with only two parameters. This assumption was made to optimise modelling between 20 and 80% DOC degradation,
Fig. 6. DOC/DOCi values calculated by logistic dose response curve modelling of the real data for each experiment against measured DOC/DOCi from all experiments (see text).
which is considered the most relevant region because detoxification takes place somewhere in this phase of degradation. We then fitted each experiment, which gave an excellent coefficient of determination (square of Pearson’s coefficient) of higher than 0.99 for each experiment and also when all measured samples were plotted against their calculated value (see Fig. 6). This confirms the adequacy of the equation for this problem. The fitted parameters are given in Table 2. Then we fitted t1/2 the same way as described above for DOC50% DOC80% t30W and t30W (see Eq. (10)) and the result was Eq. (14). No logical reason for any correlation between p and the influencing process variables could be found. So we started out again with a second degree polynomial including crossproducts of the variables. This time the results of the regression were logical and consistent. Forward selection was applied to find the optimum selection of parameters in the multivariate linear regression process [32] based on the criteria of maximising not only coefficient of determination (square of
Table 2 Logistic dose response curve parameters modelled from experimental data Experiment
Fitted values 1/2
t Centre 1 Centre 2 Centre 3 Cube 1 Cube 2 Cube 3 Cube 4 Cube 5 Cube 6 Cube 7 Cube 8
(min)
42 53 45 139 689 48 272 15 107 5.1 37
Fitted values with Model 2 p
t1/2 (min)
p
DOC50% (min) t30W
DOC80% t30W (min)
3.02 3.39 3.13 2.93 4.27 3.86 4.91 1.77 3.51 1.26 2.82
36 36 36 140 688 55 272 22 110 8.9 44
3.17 3.17 3.17 2.95 4.24 3.86 4.92 1.76 3.55 1.27 2.82
38 38 38 145 706 57 278 24 114 9.6 45
62 62 62 247 1022 85 383 58 177 33 79
DOC50% DOC80% These parameters were modelled from process parameters as described in the text and regression results for t30W and t30W calculated with Model 2.
128
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
Pearson’s coefficient), but also the Fisher’s value. Eq. (15) was the result (Fisher’s value of 135). Regression results for Eqs. (14) and (15) are given in Table 2. It also lists the values DOC50% DOC80% calculated with Model 2 for t30W and t30W : T 1:740 A0:576 t1=2 ¼ 197; 000 c0:795 Fe
(14)
p ¼ 0:0431 T þ 0:203 A 0:000929 cFe T 0:0235 cFe A þ 0:00237 T A þ 4:97 (15) Model 2, the final equation for the DOC degradation curve as a function of illumination time, temperature, iron concentration, collector surface and irradiation intensity implicitly included in the illumination time (see Eq. (1)), results from inserting Eqs. (14) and (15) into Eq. (13). The results calculated for all samples measured are plotted against the measured values in Fig. 2. The fit is very good for most experiments, except, similar to the above modelling problems, in fitting very fast experiments, probably due to the extremely wide intervals of the reaction rates observed. The partial derivative with respect to illumination time of the resulting equation represents the DOC degradation rate when the other parameters are constant (except changes in irradiation intensity, which are taken into account). If the other parameters are not constant, the degradation curve can be calculated by parts. A complete degradation curve can thus be reconstructed for varying process parameters. This would present no problem should online prediction of the degradation curve be necessary, as long as temperature and radiation intensity are measured on-line and information about changes in iron concentration is made available to the control system. So, this is a possibility for on-line prediction of process progress and for making decisions about when to end the process and whether to transfer or discharge the treated effluent. 3.4. Determination of process progress by easy-to-measure on-line parameters As mentioned above, detoxification semantically implies that at one point of the degradation process the waste water is no longer toxic and can be disposed of to a subsequent treatment or the environment. Measuring chemical oxygen demand, DOC or toxicity is expensive, time-consuming and slow. This justifies the search for alternative assessment criteria that are cheaper and faster. One approach is to model the process based on process variables describing the system as above. Another approach is to measure alternative parameters directly. We measured the absorbance at various wavelengths and hydrogen peroxide consumption off-line, and DO, ORP and pH on-line. Fig. 5 shows the same parameters for the experiment shown in Fig. 3, which may be considered paradigmatic. ORP measurements gave little information in this first screening because of their many influencing parameters. The most interesting effect was that the ORP reflects whether iron ions present are ferric or ferrous very well (see Fig. 5 for the
step between 150 and 00 , where Fenton reaction takes place converting ferrous in ferric iron). This could be an indirect indicator of a lack of hydrogen peroxide as in this case the ferric/ferrous iron relationship is changed by Eqs. (8) and (9) from equilibrium in the presence of hydrogen peroxide, while Eq. (3) cannot take place. pH measurements are mainly useful because depending upon initial pollutant concentrations and the mineralisation processes taking place (acids or bases produced), the pH can be modified during treatment. So on-line pH measurement can avoid precipitation of iron, if pH tends to rise in a particular system. In our experiments, pH changed only slightly during the treatment reflecting the formation of chloride, inorganic nitrogen species and carboxylic acids as intermediates. DO measurements were of interest, as several different phenomena were present in the DO profile. First, DO decreased, indicating the incorporation of DO in the reaction mechanism by Eqs. (16) and (17) (Dorfman mechanism) [33]. The resulting peroxyl radical can then further participate in the reaction mechanism, e.g., by Eq. (5) generating an additional hydrogen peroxide molecule. When degradation proceeded, the ratio of reactants (particularly hydrogen peroxide) to pollutant changed and the reaction of the radicals generated (by Eq. (3)) with hydrogen peroxide (Eq. (18)) was favoured, leading to the mere decomposition of hydrogen peroxide into water and oxygen (Eq. (19)). Consequently, the DO profile is a reaction progress indicator. Furthermore, at the stage at which massive oxygen production took place, a lack of hydrogen peroxide can again be perceived in the DO profile, because DO concentration decreases from its supersaturation range by degassing to the atmosphere, if no new oxygen is supplied into the solution simultaneously by the reaction:
R þ O2 ! RO2
(16)
RO2 þ H2 O ! ROH þ HO2
(17)
OH þ H2 O2 ! H2 O þ HO2
(18)
2H2 O2 ! 2H2 O þ O2
(19)
Light absorbance by the solution at 254 nm can be used in wastewater treatment to estimate the content of aromatic substances (DIN 38404-C3). However, this approximation is applicable to waste water with a rather constant composition such as urban waste water, as it depends largely on the molecular extinction coefficient of the present organic substances at the chosen wavelength. In the particular case of photo-Fenton, it is advisable to measure at somewhat higher wavelengths for two reasons. First, if hydrogen peroxide is present, its absorption influences measurement below 300 nm, and second, if the treated waste water contains aromatics many of the typical quinone/hydroquinone reaction intermediates formed can be detected by measurement at higher wavelengths as that is where they typically absorb. Indication of these intermediates is especially desirable, as they are known to be highly toxic. Fig. 5 clearly shows that the value again has a
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
129
represents the hydrogen peroxide consumption and %DOC the share of initial DOC degraded (between 0 and 1). H2 Ocons ¼ 1110 %DOC5 2100 %DOC4 þ 1430 %DOC3 2 400 %DOC2 þ 58:9 %DOC 0:0325
Fig. 7. H2O2 consumption vs. the measured DOC/DOCi values of all experiments performed, including the Fenton experiment. The curve fits show the H2O2 consumption for photo-Fenton and Fenton (a) points are marked according to temperature; (b) points are marked according to iron concentration.
typical profile that could be used to estimate process progress. The remnant absorption is due to iron complexes. Depending on pH, ferric iron forms different aquo complexes involving more or fewer hydroxyl radicals. These iron complexes differ in their light absorption properties [4], which interferes with this parameter, making it a qualitative measure. Hydrogen peroxide was measured off-line but could theoretically be measured on-line. Such sensors exist (Alldos Eichler GmbH, Prominent Dosiertechnik GmbH), although they are not very commonly employed. Fig. 7 shows DOC degradation as a function of hydrogen peroxide consumption. It can clearly be seen that the amount of degradation is strongly correlated to the amount of hydrogen peroxide consumed. It should be remarked that within the range of the parameters investigated, no influence of any of the selected process variables (iron concentration, temperature and collector area) on the amount of hydrogen peroxide consumption needed for degradation could be detected. On the contrary, Fig. 7 shows that Fenton degradation needed more hydrogen peroxide to reach the same degradation level. This is in accordance with the fact that, contrary to the dark Fenton reaction (Eq. (4)), in photo-Fenton, transformation of ferric to ferrous iron takes place mainly without hydrogen peroxide consumption (Eqs. (8) and (9)). The consumption of hydrogen peroxide (mmol/L) as a function of DOC degradation (between 0 and 1) can be estimated with a polynomial function (Eq. (20), coefficient of determination (square of Pearson’s coefficient) of 0.94, standard deviation of error 5.1 mmol/L), where H2O2cons
(20)
The theoretical hydrogen peroxide consumption for complete mineralisation of 100 mg/L Alachlor is 12.6 mmol/L (Eq. (2)). Calculated with Eq. (20), 55% of DOC mineralisation takes place before this. This correlation could be used for process control. It should be noted that the data and correlation shown are only valid for the case in hand, because hydrogen peroxide consumption depends on many parameters, mainly the type and amount of wastewater contamination. So similar empirical data will have to be obtained for different cases before such a correlation can be established. Fig. 7 shows furthermore, that extensive DOC degradation needs considerably higher amounts of hydrogen peroxide than moderate DOC degradation (e.g., 11.3, 25.2, 46.5 and 66.2 mmol/L for 50, 80, 90 and 95% DOC degradation, calculated with Eq. (20)). So apart from merely extending treatment time (and associated costs) increased reagent consumption has to be included in the economic considerations for making a decision as to when to stop treatment and/or with a view to possible combination of AOPs with subsequent biological treatment. 4. Conclusions Hundred milligrams per litre of alachlor solutions were mineralised with solar photo-Fenton treatment over a wide range of varying process variables for iron concentration, temperature and collector area per volume. Response surface methodology was applied to assess the influence of process parameters on the reaction rate. The use of second degree polynomials, including cross-product terms, yielded invalid results. Consequently, mechanistic modelling was attempted and the equation employed for modelling DOC degradation became a potential function. Prediction was further improved by modelling the DOC degradation curve as a function of illumination time with the ‘‘logistic dose response’’ curve. This produced an analytical expression as a function of time, irradiation intensity, iron concentration, temperature and solar collector area per volume. Given the analytical expression, the DOC degradation rate (first derivative) can be calculated for any moment of a real process with changing process variables, as long as feed waste water and process variable values are available. If on-line process parameter information is available, an on-line control system can be automated based on this information. Further parameters measurable on-line were also investigated. ORP, DO, pH and UV absorption of the solution produced only qualitative information about the degradation process progress. On the contrary, hydrogen peroxide measurements revealed that DOC degradation could be predicted by a polynomial relationship with hydrogen peroxide consumption as the only independent variable. That means that
130
W. Gernjak et al. / Applied Catalysis B: Environmental 64 (2006) 121–130
none of the process variables varied changed the amount of oxidising reagent necessary to reach a certain level of DOC decrease. Furthermore, it was shown that a multiple of the amount of reagent necessary for high DOC degradation levels must be added compared to what is required for moderate levels. This is important to process economic assessment and the possibility of a subsequent biological treatment. It should be noted that the models and correlations employed here were calculated for these experimental data and serve only as a model case. Due to the complexity of the system and depending on the waste water, the correlations are likely to be somewhat different in other cases. Nevertheless, the modelling approach is proposed as a methodology for obtaining models also useful for other cases. Acknowledgements The authors wish to thank the European Commission for financial support through the CADOX Project (contract no. EVK1-CT-2002-00122). Mr. Gernjak wants to thank the Austrian Academy of Sciences for a grant under the DOC Programme. The authors wish to express their gratitude to Mrs. Eva Augsten for the laboratory analysis and Mrs. Deborah Fuldauer for the correction of the English style. References [1] [2] [3] [4] [5]
[6] [7] [8] [9]
P.R. Gogate, A.B. Pandit, Adv. Environ. Res. 8 (2004) 501. P.R. Gogate, A.B. Pandit, Adv. Environ. Res. 8 (2004) 553. O. Legrini, E. Oliveros, A.M. Braun, Chem. Rev. 93 (1993) 671. A. Safarzadeh-Amiri, J.R. Bolton, S.R. Cater, J. Adv. Oxid. Technol. 1 (1996) 18. V. Augugliaro, C. Baiocchi, A. Bianco Prevot, E. Garcı´a-Lo´pez, V. Loddo, S. Malato, G. Marcı´, L. Palmisano, M. Pazzi, E. Pramauro, Chemosphere 49 (2002) 1223. S. Esplugas, J. Gime´nez, S. Contreras, E. Pascual, M. Rodrı´guez, Water Res. 36 (2002) 1034. M.E. Sigman, A.C. Buchanan III, S.M. Smith, J. Adv. Oxid. Technol. 2 (1997) 415. A. Goi, M. Trapido, Chemosphere 46 (2002) 913. H. Fallmann, T. Krutzler, R. Bauer, S. Malato, J. Blanco, Catal. Today 54 (1999) 309.
[10] T. Krutzler, H. Fallmann, P. Maletzky, R. Bauer, S. Malato, J. Blanco, Catal. Today 54 (1999) 321. [11] S. Malato, J. Blanco, A. Vidal, C. Richter, Appl. Catal. B: Environ. 37 (2002) 1. [12] S. Malato, J. Blanco, A. Vidal, D. Alarco´n, M.I. Maldonado, J. Ca´ceres, W. Gernjak, Sol. Energy 75 (2003) 329. [13] R. Bauer, H. Fallmann, Res. Chem. Intermed. 23 (1997) 341. [14] V. Sarria, S. Kenfack, O. Guillod, C. Pulgarin, J. Photochem. Photobiol. A 159 (2003) 89. [15] S. Esplugas, D.F. Ollis, J. Adv. Oxid. Technol. 2 (1997) 197. [16] E. Oliveros, O. Legrini, M. Hohl, T. Mu¨ller, A.M. Braun, Chem. Eng. Proc. 36 (1997) 397. [17] E.K. May, R. Gee, D.T. Wickham, L.A. Lafloon, J.D. Wright, Design and Fabrication of a Prototype Solar Receiver/Reactors for the Solar Detoxification of Contaminated Water, NREL Report, Industrial Solar Technology Corp., Golden, Colorado, USA, 1991. [18] S. Malato, J. Ca´ceres, A.R. Ferna´ndez-Alba, L. Piedra, M.D. Hernando, A. Agu¨era, J. Vial, Environ. Sci. Technol. 37 (2003) 2516. [19] M. Kositzi, I. Poulios, S. Malato, J. Ca´ceres, A. Campos, Water Res. 38 (2004) 1147. [20] S. Malato, J. Blanco, M.I. Maldonado, P. Ferna´ndez-Iba´n˜ez, A. Campos, Appl. Catal. B: Environ. 28 (2000) 163. [21] S. Malato, J. Blanco, M.I. Maldonado, P. Ferna´ndez, D. Alarcon, M. Collares, J. Farinha, J. Correia, Sol. Energy 77 (2004) 513. [22] J. Blanco, S. Malato, P. Ferna´ndez, A. Vidal, A. Morales, P. Trincado, J.C. Oliveira, C. Minero, M. Musci, C. Casalle, M. Brunotte, S. Tratzky, N. Dischinger, K.H. Funken, C. Sattler, M. Vincent, M. Collares-Pereira, J.F. Mendes, C.M. Rangel, Sol. Energy 67 (2000) 317. [23] G. Sagawe, A. Lehnard, M. Lubber, G. Rochendorf, D. Bahnemann, Helv. Chim. Acta 84 (2001) 3742. [24] F. Herrera, C. Pulgarin, V. Nadtochenko, J. Kiwi, Appl. Catal. B: Environ. 17 (1998) 141. [25] G.E.P. Box, Statistics for Experimenters, John Wiley & Sons, New York, USA, 1978. [26] C.D.S. Tomlin, The Pesticide Manual, 11th ed., British Crop Protection Council, Farnham, UK, 1997. [27] APHA, AWWA, WEF, Standard Methods for the Examination of Water and Waste Water, 20th ed., United Book Press Inc., Maryland, USA, 1998 . [28] R. Chen, J.J. Pignatello, Environ. Sci. Technol. 31 (1997) 2399. [29] A.Y. Sychev, V.G. Isaak, Russ. Chem. Rev. 64 (1995) 1105. [30] G.A. Pen˜uela, D. Barcelo´, J. Chromatogr. A 754 (1996) 187. [31] M. Hincapie´, M.I. Maldonado, I. Oller, W. Gernjak, J.A. Sa´nchez-Pe´rez, M.M. Ballesteros, S. Malato, Catal. Today 101 (2005) 203. [32] J.F. Hair, R.E. Anderson, R.L. Tatham, W.C. Black, Multivariate Data Analysis, Prentice Hall International, New Jersey, USA, 1998. [33] C. von Sonntag, P. Dowideit, X. Fang, R. Mertens, X. Pan, M.N. Schuchmann, H.P. Schuchmann, Water Sci. Technol. 35 (4) (1997) 9.
