Detection and Discrimination of Phenol and Primary Alcohols in Water

Electronic nose methodology was used for discriminating between samples of n-primary alcohols and phenol in water. An electronic nose device developed...
0 downloads 0 Views 300KB Size
Download PDF
Environ. Sci. Technol. 2006, 40, 6058-6063

Detection and Discrimination of Phenol and Primary Alcohols in Water Using Electronic Noses VIRGINIA DIZ,† MIRYAN CASSANELLO,‡ AND R . M A R T I N N E G R I * ,† Instituto de Quı´mica Fı´sica de Materiales, Ambiente y Energı´a (INQUIMAE), Departamento de Quı´mica Inorga´nica, Analı´tica y Quı´mica Fı´sica, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, Buenos Aires, Argentina, and PINMATEsDepartamento de Industrias, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, Buenos Aires, Argentina

Electronic nose methodology was used for discriminating between samples of n-primary alcohols and phenol in water. An electronic nose device developed at our laboratory was applied to detect the vapors from the samples. The device consists of 10 nonspecific gas sensors based on tindioxide, whose electrical conductivity change when exposed to the vapors. The whole set of responses constitutes a fingerprint associated with the sample for the particular array of sensors. Using this method, we could differentiate among n-primary alcohols, pure phenol, and aqueous dilutions of phenol. Dilutions of primary alcohols and phenol of 1, 3, 5, and 10% (v/v) were assayed. The fingerprints of the dilutions are similar to that of the respective pure (not diluted) alcohol, but with lower signal intensities. Principal component analysis and cluster analysis were performed in order to explore the feasibility of discriminating among the samples. Dilutions of phenol were successfully discriminated from those of primary alcohols. Discrimination of n-octanol and n-decanol from the other alcohols, including phenol, was also achieved. The results suggest that it is possible to discriminate among dilutions of phenol in water and to give an approximate gross estimation of phenol concentration in water.

Introduction The detection of volatile organic compounds (VOCs) dissolved in water through their vapors is not an easy task. Traditional methodologies normally require collecting a sample, for instance with bag samplers, canisters, or passive tubes (1), to process them afterward in the laboratory by gas chromatography for separation, with mass spectroscopy for quantification and/or identification of the components. These methodologies allow the accurate analysis of the dissolved VOC and determination of the chemical nature of the components in the samples. The use of an electronic nose to have an approximate indication of the dissolved VOC constitutes an interesting * Corresponding author fax: 54-11-4576-3341; e-mail: [email protected]. † Instituto de Quı ´mica Fı´sica de Materiales, Ambiente y Energı´a (INQUIMAE), Departamento de Quı´mica Inorga´nica, Analı´tica y Quı´mica Fı´sica. ‡ PINMATEsDepartamento de Industrias. 6058

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 19, 2006

alternative to avoid the rather cumbersome and slow procedure of sampling and separate analysis (4). Instead of determining accurately the concentration of one or many of the compounds that are present in the sample, the e-nose methodology tries to discriminate among different samples by associating a fingerprint to each one. Thus, the objective is to generate a fingerprint of each sample and to use pattern recognition methods to discriminate among them. The main application is in-lab analysis of multicomponent samples (wines, food, perfumes, etc). The work pursuing field studies for environmental applications using sensors arrays is rather scarce (5-13). Most of the studies are based on collecting gas samples with Tedlar bags, or sniffing vapors with a pump or from the headspace of a vial to feed them into the place where the sensor’s array is located. Primary alcohols and phenol are some of the VOCs commonly found in wastewaters, for instance from polymeric resin production, chemical industries, and oil refineries. Primary alcohol concentrations vary with process conditions within the range of 10-3 to 10-2 M, while phenol concentrations can reach significantly large values in some cases. Phenol is highly toxic for the microorganisms employed for the biological wastewater treatments. Hence, its concentration should be reduced below a certain level before feeding the treatment plant. If the concentration is higher than the threshold, the wastewater should be pretreated for example by stripping or extraction (2, 3). Moreover, if the concentration is too high, a thermal decomposition method becomes the best option. In this context, it is quite profitable to have a fast and low cost method to decide whether a pretreatment is required before feeding the wastewater to the biological plant, to avoid destroying the culture. In addition, a rapid indication to aid operators in deciding if a thermal decomposition could be adequate would also be beneficial. However, the detection at an industrial effluent presents several difficulties. Having a portable device that allows real time or, at least, rapid monitoring of the effluent characteristics would be convenient. Moreover, as discussed above, to discriminate in situ and continuously if an effluent contains primary alcohols or phenol over a given concentration would be very useful. In this work, we have studied the response of an e-nose array pursuing the continuous and semiquantitative detection of phenol dissolved in water and the discrimination among it and different primary alcohols that could also be present in the wastewaters. The system is simple and robust and can be easily improved by adding accessories and modifying the geometry to decrease the time required and the limit of detection toward the desired level, at the expense of a low increase in instrumentation complexity. In this system, the vapors are not inhaled; the liquid samples are introduced directly into a sealed chamber where the sensors are placed, and the signals are collected after saturation of the chamber with the vapors. This approach has the advantages of instrumental simplicity, less influence of external environmental factors such as variations in wind speed or distance odor source to sampling site, which are frequently found in environmental monitoring and that largely contribute to the data-scatter. On the other hand, the main drawback of our approach is that the time required for attaining steady-state, the detection limit, and the limit of resolution (7) are larger. However, they can be significantly improved by simple accessories that increase the temperature and/or reduce the pressure in the chamber. The aim of this work is to illustrate the capability of our specific approach in the field of the e-nose methodology to 10.1021/es052322e CCC: $33.50

 2006 American Chemical Society Published on Web 09/01/2006

detect dissolved primary alcohols and phenols in water through their vapors, in concentration ranges frequently found in water industrial effluents. In addition, we report preliminary studies addressing the ability of discriminating among water solutions of primary alcohols and phenol. That is, we explore not only the possibility of detecting dissolved alcohols, but also of discriminating among different alcohols by means of their fingerprint, using pattern recognition methods. Dilutions of primary alcohols, phenol, and its oxidation derivatives are discriminated using two multivariate data analysis methods: Principal Component Analysis (PCA) and Cluster Analysis (CA). These methods are nonsupervised (no previously recorded training sets are required, as in the case of Artificial Neural Networks) and succeeded in providing an excellent discrimination among solutions of phenols and primary alcohols in water, using a relatively small database.

TABLE 1. Results of the Cluster Analysis

system MeOH EtOH PrOH BuOH OcOH DecOH dilutions of MeOH dilutions of EtOH dilutions of PrOH phenol dilutions of phenol

expected number of classification number of samples group according analyzed assigned by CA to to figure 4 samples the expected group A A A A B B A A A C C

4 5 3 2 4 4 2 3 2 2 8

4 5 3 2 3 4 2 3 2 2 7

Materials and Methods Chemicals. All the alcohols were analytical grade and Milli-Q water was used. Primary alcohols (methanol, ethanol, nbutanol, n-propanol, n-octanol, n-decanol) and phenol were used as purchased (Merck) without further distillation. Phenol was purified by re-crystallization under vacuum and stored in the presence of silica gel. Sample Preparation. Samples were prepared immediately before the respective electronic nose measurement. The dilutions of primary alcohols in water are referred by their respective v/v percentages. In the case of phenol, the percentages are w/v. Phenol was purified and dehydrated according to standard protocols in order to obtain the solid form. The solutions of phenol in water were prepared immediately after purification by weighing the solid form. However it is worth mentioning that in the displayed figures the term “pure phenol” refers to the liquid form of phenol obtained at about 35 °C, to compare with the other liquid pure alcohols. The range of alcohols and phenol concentration is in the order of 1-10% w/v, except, of course, when analyzing the pure alcohols. These concentrations can be found in industrial wastewaters (1, 2). Although the concentration in the vapor phase depends on the Henry’s constant (see for example Teja et al. (14) for the case of methanol in water) those concentrations correspond to partial pressures in the order of 10-3 to 10-4 atm (103 to 102 ppm in the vapor phase), depending on the alcohol. These values are well above the sensors’ detection limits, which are in the order of a few ppm, according to the technical data sheets provided by Figaro Inc. The detection limit of our device is higher and depends on the instrument, for example the size of the chamber. The lowest vapor concentration detected with the device used in the present work was about 10 ppm, with signal-to-noise ratio of about 3. This suggests that low concentrations in the liquid phase can be detected if the dissolved compound is volatile enough to have a high vapor pressure. In this work, the purpose was focused in discrimination and in analyzing high concentrations such as those found in nontreated effluents. Hence, the detection limit was not searched for. Additionally, Nicolas and Romain (7) have indicated that it is worthwhile to distinguish between the limit of detection and the limit of resolution for pairwise comparisons of vapors, as the concentration levels below which a vapor can no longer be reliably recognized from another one on the basis of its response pattern. Concentrations for the limit of resolution are significantly higher than those of the limit of detection. Electronic Nose Measurements. The developed electronic nose consists of a sample chamber and an array of commercial gas sensors (Figaro Inc.) based on polycrystalline tin dioxide. The total number of sensors used in this work is 10. The sensors are nonspecific; one sensor may provide the

same signal for different gases in different concentrations or proportions. However, the manufacturer suggests different target gases for the sensors used here: TGS 831 (sensor 1, chlorofluocarbons), TGS 813 (sensor 2, combustible gases), TGS 825 (sensor 3, H2S), TGS 832 (sensor 4, halocarbons), TGS 880 (sensor 5, water vapor in cooking process), TGS 826 (sensor 6, NH3 and amines), TGS 816 (sensor 7, combustible gases), TGS 882 (sensor 8, alcohols), TGS 823 (sensor 9, xylene and toluene), and TGS 800 (sensor 10, gasoline exhausts). A detailed description of the system can be found elsewhere (15-17). The detection method responsible for the signals presented here is based on changes in voltage in an electrical circuit containing the sensor element; thus, signals are indicated in volts. These signals increase when increasing the electrical conductivity of the active sensing material (tin dioxide). An open flask with the sample (200 µL) was placed into the closed chamber of the e-nose for each measurement. Sensors’ signals were continuously recorded after closing the chamber, up to a steady-state situation for which the sensor signals reached a plateau. The values of the sensor’s signals at the plateau were used for the analysis after subtraction of the corresponding baseline. Thus, a set of 10 signals {S1 ,..., S10} was obtained for each experiment. Each signal is indicative of the electrical conductance increase in the respective sensor due to the presence of alcohol vapors in the chamber. Some typical sensors’ signals (not all of them for the sake of clarity) are shown in the Supporting Information. The whole set of sensors’ signals is referred to as a “fingerprint”. The measurement time is about 20 min with the device used in the present work. However, this time has been reduced five times by using new models developed in our laboratory just by adjusting the geometry of the chamber. Thus, the same can now be performed in about 2-4 min (18, 19). Principal Component Analysis (PCA). The e-nose methodology uses multivariate data analysis to discriminate among groups of signals and, as a consequence, to discriminate between different samples. There are several wellknown methods of multivariate data analysis (20); the most popular being the Principal Component Analysis (PCA) (21), an unsupervised method that is useful for graphic visualization and data discrimination. PCA was performed with the whole group of obtained data. Thus, for every data group {S1,...,S10}, the so-called principal components, {PC1, PC2,...,PC10}, were calculated. We present an extensive description of this feature extraction method in the Supporting Information. Cluster Analysis (CA). Once the set of projections in the first two principal directions, {PC1, PC2}, (vectors) were obtained, they were analyzed using another unsupervised multivariate method, the so-called Cluster Analysis (CA). CA VOL. 40, NO. 19, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6059

FIGURE 1. Radar plots of different samples of pure alcohols. Each vertex of the radar corresponds to one sensor of the array: (a) methanol, (b) ethanol, (c) n-propanol, (d) n-butanol, (e) n-octanol, (f) n-decanol, (g) phenol, and (h) oxidation derivatives of phenol. The scales range from 0 to 1 V for a, b, c, and d, and from 0 to 0.5 V for e, f, g, and h. In the case of phenol and its oxidation derivatives, the measurement corresponds to the compounds in the liquid form. classifies the data inputs into a given number of groups, referred to as clusters. The number of desired clusters is previously indicated. In the present work, the number of clusters was fixed to three, termed A, B, and C, in Table 1 and Figure 4, associated with samples of the short-chain C1-C4 primary alcohols (A), octanol and decanol (B), and samples with phenol (C) (more details of the algorithm and method can be found in the Supporting Information).

Results and Discussion Signals and Fingerprints of Pure Alcohols. The typical signals of the gas sensors after a liquid sample is introduced in the chamber are shown in the Supporting Information. All sensors showed a good response to analyzed vapors, although some were more sensitive to phenol and others were more sensitive to the primary alcohols. 6060

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 19, 2006

The output of the array is a group of signals {S1,...,SN}, which are represented as radar plots in Figure 1, where each vertex is associated with one defined sensor of the array. The shape and the intensity are similar for methanol and ethanol (although not identical) and then vary through the series. That is, the pattern distribution changes and the signals of the sensors slightly decrease in the series of primary alcohols. Figure 1 also shows the fingerprints of pure primary alcohols, pure phenol (liquid), and a mixture of quinones (liquid oxidative forms). These fingerprints are clearly different among them and are also different from those of the primary alcohols. These results indicate that the different alcohols can be discriminated by the array. In other words, even in the case that the individual signals of some particular sensors are the same for different alcohols, the whole set of signals, {S1,...,S10}, displays differences. This feature constitutes the strength of

FIGURE 2. (a) Radar plots of pure ethanol and ethanol dilutions in water. (b) Radar plots of n-propanol and its dilutions in water. Percentages in v/v. the methodology and is the basis for going to more difficult tasks which involve detecting and discriminating among dilutions of alcohols in water, as discussed in the next section. Dilutions of Methanol, Ethanol, Propanol, and Phenol in Water. The radar plots of some dilutions in water of ethanol and n-propanol are shown in Figure 2. The signals of the dilutions are well above the baseline with a good signal-tonoise ratio (>100) even for the more diluted samples. For example, the signals obtained for the dilutions of ethanol are about 16% of those for pure ethanol. Hence, the array is able to detect the presence of ethanol even in the most diluted solution examined. The shape of the radar plot, and thus the signal pattern provided by the sensors, does not change with the concentration of the substance, even if the intensity of the responses is markedly different. Similar results were obtained for the other prepared dilutions (methanol, propanol, and phenol), with lower differences between the pure alcohol and its dilutions as the signals of the pure sample decrease. Only in simple cases where different compounds have significantly different odor patterns can the products be discriminated by visual inspection of a radar plot, as in Figure 1. In most practical situations, the radar plots display very small differences and mathematical methods are necessary to establish whether the differences are statistically significant. Therefore, in those cases, the last step of the analysis is to try to discriminate mathematically among the groups of signals obtained using pattern recognition methods such as Principal Component Analysis or Cluster Analysis. Principal Component Analysis and Cluster Analysis. More than 96% of the total data variance is contained in two directions of the “vector” space referred to as the two first principal components, {PC1, PC2}. Therefore, a bi-dimensional plot of {PC1, PC2} is well representative of the data and allows a simpler visualization of the results. Hence, the discrimination between samples can even be achieved by visual inspection of the {PC1, PC2} points in a 2-D PCA plot, referred to as a “PCA-map”. In Figure 3, we have defined three groups referred to as (1), (2), and (3). Group (1) is the group of pure primary alcohols up to butanol, that is, short-

FIGURE 3. 2-D PCA map of the two first principal components, PC1-PC2, where each point is associated with one measurement. Here, the sensors’ signals were not normalized. The black symbols correspond to “pure” alcohols while the open symbols correspond to dilutions of the alcohols in water. 1 Phenol; 2 MeOH; [ EtOH; ` propanol; * BuOH; b OcOH; 9 DecOH; 0 dilutions of phenol; 4 dilutions of MeOH; ] dilutions of EtOH; " dilutions of propanol. Group (1): short-chain alcohols (C1-C4) without dilution. Group (2): long-chain (C8,C10) alcohols and also pure phenol and dilutions of phenol in water. Group (3): dilutions of the short-chain (C1-C3) alcohols. chain alcohols (C1-C4) without dilution. Group (2) is associated with the long-chain (C8,C10) primary alcohols and also pure phenol and dilutions of phenol in water. Group (3) is associated with the dilutions of the short-chain (C1-C3) primary alcohols. The visual inspection in Figure 3 indicates that group 1 is well defined and separated from the others. When comparing the samples of groups 2 and 3, many sample points can be well assigned to these groups, except for the measurements of MeOH dilutions (open triangles), which are located in an ambiguous region in the PCA-map (they should be included in group 3, but they may also be grouped into 2). In other words, the discrimination between dilutions of primary alcohols (group 2) and dilutions of phenol (group 3) is not clear by the above procedure. For these reasons, a new analysis was performed by preprocessing the data prior to their introduction in the principal component analysis. Preprocessing consists of “normalizing” the sensor signals by converting the input data vectors {S1,....,S10} into new

10 vectors {S1′,....,S10′}, with Sk′) Sk/x∑i)1 Si2. This normalization is performed for each vector. The result of the new PCA is a new set of vectors, {PC1′,...,PC10′}, with the total number of vectors always being equal to the total number of measurements. This data treatment is currently reported in the electronic nose literature (4, 5, 16), mainly in those cases where the dispersion due to concentration variations makes discrimination between the samples of relevance difficult for the specific problem of interest. For instance, in the present study, we are particularly interested in discriminating between the phenol dilutions and the primary alcohol dilutions. Since this target is more relevant for us than to discriminate between “pure” and diluted alcohols, we decided to normalize the data and to perform the associated PCA. More than 95% of the variance is again contained in the {PC1′,PC2′} vectors. Figure 4 shows the results of the PCA with the normalized data. Three groups are now well discriminated in the PCA map, indicated as A, B, and C in Figure 4: the short-chain primary alcohols and their dilutions (group A), the long-chain primary alcohols (octanol and decanol, group B), and finally phenol and its dilutions (group C). Thus, the desired target, discrimination between dilutions

VOL. 40, NO. 19, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6061

FIGURE 4. 2-D “normalized” PCA, referred in the text as the {PC1′,-

2 PC2′} vectors. Sensor signals, Sj, normalized by x∑i10 )1Si before performing the analysis of principal component. Each point is associated with one measurement. The black symbols correspond to “pure” alcohols while the open symbols correspond to dilutions of the alcohols in water. The ellipsoids correspond to the classification provided by Cluster Analysis. The symbols correspond to those of Figure 3.

of “phenols” and “primary” alcohols is now achieved. Moreover, the separation in the three groups, A, B and C, was obtained not only by visual data inspection in the PCA map, but also when a Cluster Analysis (CA) was performed. In fact, when using the set {PC1′,PC2′} as inputs for the CA, and with the only indication that those data must be classified into three clusters, the agreement between the results of CA (which groups the inputs into clusters) and the groups shown in Figure 4 is about 95% (37 samples assigned as expected from Figure 4 over a total of 39 samples). The results of CA are detailed in Table 1. Therefore, the drawbacks found when using nonnormalized inputs are solved by using normalized inputs into PCA: phenol solutions are well discriminated from primary alcohol solutions and the measurements can be classified not only by visual inspection in the PCA map, but also mathematically by a well-defined method such as cluster analysis. The normalization also allows a good discrimination between phenol dilutions and the largest primary alcohols (n-octanol and n-decanol). Finally, it is worth noting that the price paid by normalizing the data is that the concentration effects are substantially diminished. For instance, points associated with the alcohols dilutions (open points in Figure 4) are much closer to those of the pure alcohols, in contrast to the results obtained with the raw data (see Figure 3). Therefore, the concentration effects (through the vapor pressure of the samples), must be analyzed with the nonnormalized data as described in the next section. Dependence with the Vapor Pressure, Pv. We observed that the first principal component of the nonnormalized set, PC1, is very much dependent on the sensors’ signal intensities, which increase with vapor concentration. Moreover, a linear relationship holds between the first principal component, PC1, and the square root of the quadratic sum of signals, 10 Si2 (shown in the Supporting Information). This square x∑i)1

root is proportional to the “area” under the radar plot. That is, PC1 linearly increases with the intensity of the signals, as does the area. This correlation does not mean that the intensity is the only factor influencing PC1, but it represents a strong contribution. Therefore, as the intensities of the sensors’ signals are dependent on the vapor concentration which is given by its partial pressure, we analyzed the dependence of PC1 with the vapor pressure of each sample, Pv.

6062

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 19, 2006

FIGURE 5. PC1 vs. the natural logarithm of the vapor pressure associated with each sample (Pv indicates the vapor pressure in mmHg). The sensors’ signals were not preprocessed (no normalization was applied). The symbols correspond to those of Figure 3. The plot of PC1 vs Pv (Figure 5) shows that for the most volatile alcohols a saturation effect with lack of sensitivity is observed in the response of the sensors at the highest vapor pressures. For the dilutions and samples of octanol and decanol, having much lower values of Pv, PC1 follows Pv. This explains why in the nonnormalized PCA map (Figure 3) the points associated with the methanol and ethanol dilutions are close to the points associated with pure octanol and decanol: they are the samples of pure alcohols with the lowest vapor pressures. Nevertheless, the vapor pressure is not the only factor influencing the arrays response: the signal of each sensor is influenced also by specific alcohol-sensor interactions, as clearly shown by the different shapes of the radar plots associated with the pure alcohol’s samples and also by the results of Figure 4 and Table 1, which show that octanol-decanol are clearly discriminated with respect to dilutions of methanol-ethanol. To discriminate among dilutions of methanol, ethanol, and propanol is difficult because they all have relatively high vapor pressures and similar shapes of the radar plots. Similarly, to discriminate between dilutions of octanol and decanol will also be hard since they have comparable low pressure vapor and fingerprints. It is likely to distinguish dilutions of long-chain alcohols from dilutions of the C1-C4 alcohols, since their fingerprints are very different (Figure 1). This is reflected in the clear discrimination between octanoldecanol and the other alcohols illustrated in Figure 4. The PC1-PC2 plot in the case of phenol, its dilutions in water, and the oxidized phenol (Figure 6), suggests that it is possible to discriminate among the dilutions, as they are far enough from the saturation region and above the detection limits, defined by a signal-to-noise-ratio equal to 3 in agreement with the criteria adopted by Nicolas and Romain (7) and Doleman and Lewis (22). That is, our results suggest that it is possible to discriminate among dilutions of phenol in water and to give an approximate gross estimation of phenol concentration in water. To discriminate among dilutions of high volatility alcohols, like methanol and ethanol, is more difficult with this set, due to an effect of saturation of the sensors. For more concentrated solutions, the sensors’ response has low variability, which causes difficulties for the discrimination. The developed system has some relevant characteristics for environmental analysis of liquid effluents: it neither requires adaptation of the sample nor separation of the alcohols from water, it is not destructive and it uses only few hundred microliters of sample. It does not even need to volatilize the whole samples. Therefore, the approach can

FIGURE 6. PC1-PC2 plot for phenol and its dilutions in water. The sensor signals were not normalized. Percentages in w/v. The term “pure phenol” indicates its liquid form. be easily optimized for control purposes in industries or online monitoring of highly concentrated wastewaters. Naturally, interference effects will complicate the analysis in a real sample where all the alcohols may be present in different proportions. This work shows that it is possible to discriminate among wastewater effluents having phenol and primary alcohols and suggests that the discrimination of long-chain alcohols from C1-C4 alcohols is also possible. The study of interference effects requires the systematic analysis of a large number of samples with composition mixtures and will be carried out as the next step. Each real-world wastewater sample constitutes a different complex matrix and, it is clear that the system will need to be calibrated and checked in each case, until a diagnosing procedure is established. The e-nose technology does not pursue the goal of identification of the compounds in a sample, but to diagnose if a given sample is in or out of a required standard, or in other cases to discriminate between samples; for this reason, it relies on pattern recognition and classification procedures. The presented results are of relevance to establish the presence of primary alcohols and phenol in wastewaters, including threshold limits for alarms in control of effluents. A slightly improved version of our device is able to perform a measurement in about 2-4 min (19), allowing detection and monitoring in real situations. For instance, the system has the potential to be applied for controlling phenols and primary alcohols levels during extraction and separation procedures, such as those described, for instance, by Materna et al. (23) and Hubb et al. (24).

Acknowledgments R.M.N. and M.C. are members of the Consejo Nacional de Investigaciones Cientı´ficas y Te´cnicas (CONICET, Argentina). V.D. is a teaching assistant at the University of Buenos Aires. This work was supported by the University of Buenos Aires (UBACyT 2004-2007, Projects X267 and X180) and Fundacio´n Antorchas (Argentina).

Supporting Information Available 10 Signals of the electronic nose, a plot of x∑i)1 Si2vs PC1, and details about Principal Component Analysis and Cluster Analysis. This material is available free of charge via the Internet at http://pubs.acs.org.

Literature Cited (1) Pratt, G. C.; Bock, D.; Stock, T. H.; Morandi, M.; Adgate, J. L.; Ramachandran, G.; Mongin, S. J.; Sexton, K. A field comparison of volatile organic compound measurements using passive organic vapor monitors and stainless steel canisters. Environ. Sci. Technol. 2005, 39, 3261-3268.

(2) Jiang, H.; Fang, Y.; Fu, Y.; Guo, Q-X. Studies on the extraction of phenol in wastewater. J. Hazard. Mater., B 2003, 101, 179190. (3) Poulopoulos, S. G.; Voutsas, E. C.; Grigoropoulou, H. P.; Philippopoulos, C. J. Stripping as a preatreatment process of industrial oily wastewater. J. Hazard. Mater., B 2005, 117, 135-139. (4) Gardner, J. W.; Bartlett, P. Electronic Noses: Principles and Applications; Oxford University Press: Oxford, 1999. (5) Romain, A. C.; Nicolas, J.; Andre, P. In situ measurements of olfaction pollution with inorganic semiconductors: limitations due to the influence of humidity and temperature. Semin. Food Anal. 1997, 2, 283-296. (6) Bourgeois, W.; Romain, A. C.; Nicolas, J.; Stuetz, R. The use of sensor arrays for environmental monitoring. Interests and limitations. J. Environ. Monit. 2003, 5, 852-858. (7) Nicolas, J.; Romain, A. C. Establishing the limit of detection and the resolution limits of odorous sources in the environment for an array of metal oxide gas sensors. Sens. Actuators, B 2004, 99, 384-392. (8) Albert, K. J.; Myrick, M. L.; Brown, S. B.; Dale, L. J.; Milanovich, F. P.; Walt, D. R. Field-deployable sniffer for 2,4-dinitrotoluene detection. Environ. Sci. Technol. 2001, 35, 3193-3200. (9) Stuetz, R. M.; Fenner, R. A.; Engin, G. Characterisation of wastewater using an electronic nose. Water Res. 1999, 33, 442-452. (10) Dewettinck, T.; Van Hege, K.; Verstraete, W. The electronic nose as a rapid sensor for volatile compounds in treated domestic wastewater. Water Res. 2001, 35, 2475-2483. (11) Patel, S. V.; Mlsna, T. E.; Fruhberger, B.; Klaassen, E.; Cemalovic, S.; Baselt, D. R. Chemicapacitive microsensors for volatile organic compound detection. Sens. Actuators, B 2003, 96, 541553. (12) Wanekaya, A. K.; Uematsu, M.; Breimer, M.; Sadik, O. A. Multicomponent analysis of alcohol vapors using integrated gas chromatography with sensor arrays. Sens. Actuators, B 2005, 110, 41-48. (13) McCorkle, D. L.; Warmack, R. J.; Patel, S. V.; Mlsna, T.; Hunter, S. R.; Ferrell, T. L. Ethanol vapor detection in aqueous environments using micro-capacitors and dielectric polymers. Sens. Actuators, B 2005, 107, 892-903. (14) Teja, A. S.; Gupta, A. K.; Bullock, K. R.; Chai, X.-S.; Zhu, J. Y. Henry’s Constants of Methanol in Aqueous Systems containing Salts. Fluid Phase Equilib. 2001, 185, 265-274. (15) O’Connell, M.; Valdora, G.; Peltzer, G.; Negri, R. M. A practical approach for fish freshness determinations using a portable electronic nose. Sens. Actuators B 2001, 80, 149-154. (16) Branca, A.; Simonian, P.; Ferrante, M.; Novas, E.; Negri, R. M. Electronic nose based discrimination of a perfumery compound in a fragrance. Sens. Actuators, B 2003, 92, 222-227. (17) Monge, M. E.; Bullone, D.; Giacomazza, D.; Bernik, D. L.; Negri, R. M. Detection of flavour release from pectin gels using e-noses. Sens. Actuators, B 2004, 101, 28-38. (18) Lovino, M.; Cardinal, M. F.; Zubiri, D. B. V.; Bernik, D. L. Electronic nose screening of ethanol release during sol-gel encapsulation. A novel noninvasive method to test silica polymerisation. Biosens. Bioelectron. 2005, 21, 857-862. (19) Langenheim, M.; Lovino, M.; Monge, M. E.; Mizrahi, D.; Negri, R. M. Trends towards an Electronic Nose-Electronic Tongue Fusion Applied to Discrimination of Wines. Presented at6 CzechArgentine Biennale, “Interdisciplinary aspects of HumanMachine Co-existence and Co-operation”, Prague, July 2-5, 2005. (20) Jurs, P. C.; Bakken, G. A.; McClelland, H. E. Computational Methods for the Analysis of Chemical Sensor Array Data from Volatile Analytes. Chem. Rev. 2000, 100, 2649-2678. (21) Jollife, I. T. Principal Component Analysis; Springer-Verlag: New York, 1986. (22) Doleman, B. J.; Lewis, N. S. Comparison of odor detection thresholds and odor discriminablities of a conducting polymer composite electronic nose versus mammalian olfaction. Sens. Actuators, B 2001, 72, 41-50. (23) Materna, K.; Milosz, I.; Miesiac, I.; Cote, G.; Szymanowski. Removal of phenols from aqueous streams by the cloud point extraction technique with oxyethylated methyl dodecanoates as surfactants. Environ. Sci. Technol. 2001, 35, 2341-2346. (24) Huub, H. J.; Cox, T. S.; Zarook, M. S.; Deshusses, M. A. Thermophilic biotrickling filtration of ethanol vapors. Environ. Sci. Technol. 2001, 35, 2612-2619.

Received for review November 17, 2005. Revised manuscript received May 31, 2006. Accepted June 12, 2006. ES052322E VOL. 40, NO. 19, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6063