An Expert System for Chemical Speciation of Individual Particles

An expert system that can rapidly and reliably perform chemical speciation from the low-Z particle EPMA data is presented. This expert system tries to...
40 downloads 15 Views 65KB Size
Anal. Chem. 2004, 76, 1322-1327

An Expert System for Chemical Speciation of Individual Particles Using Low-Z Particle Electron Probe X-ray Microanalysis Data Chul-Un Ro,*,† HyeKyeong Kim,† and Rene´ Van Grieken‡

Department of Chemistry, Hallym University, ChunCheon, KangWonDo, 200-702, Korea, and Department of Chemistry, University of Antwerp, Universiteitsplein 1, B-2610 Antwerpen, Belgium

An electron probe X-ray microanalysis (EPMA) technique, using an energy-dispersive X-ray detector with an ultrathin window, designated a low-Z particle EPMA, has been developed. The low-Z particle EPMA allows the quantitative determination of concentrations of low-Z elements, such as C, N, and O, as well as chemical elements that can be analyzed by conventional energy-dispersive EPMA, in individual particles. Since a data set is usually composed of data for several thousands of particles in order to make environmentally meaningful observations of real atmospheric aerosol samples, the development of a method that fully extracts chemical information contained in the low-Z particle EPMA data is important. An expert system that can rapidly and reliably perform chemical speciation from the low-Z particle EPMA data is presented. This expert system tries to mimic the logics used by experts and is implemented by applying macroprogramming available in MS Excel software. Its feasibility is confirmed by applying the expert system to data for various types of standard particles and a real atmospheric aerosol sample. By applying the expert system, the time necessary for chemical speciation becomes shortened very much and detailed information on particle data can be saved and extracted later if more information is needed for further analysis. In many environmental and geological applications, the characterization of particulate matter or materials having grainlike structure can be essential to understand the nature of samples. There are two major approaches in the analysis of particulate matter: bulk and single-particle analyses. By using bulk methods, only the average composition of particulate samples can be obtained. However, these samples can be a heterogeneous mixture of different types of particles, so the average composition and average diameter do not describe the variability of the particles in the sample. To describe the chemical microprocesses in (or on), for example, aerosol particles or sediment particles, detailed knowledge of the chemical composition and the morphology of individual particles on a micrometer scale is becoming increasingly * Corresponding author. Tel.: +82 33 248 2076. Fax: +82 33 256 3421. E-mail: [email protected]. † Hallym University. ‡ University of Antwerp.

1322 Analytical Chemistry, Vol. 76, No. 5, March 1, 2004

important. Microanalytical methods allowing quantitative analysis are thus required.1 The analysis of a relatively large number (at least several hundreds) of individual particles is required to provide relevant statistics on the population of particles in the sample. To determine possible sources for particulate air pollution or characterize particles suspended in water, even a greater number of individual particles may have to be analyzed.2 An advanced degree of analysis automation is thus necessary. Recently, a microanalytical technique, named “low-Z particle electron probe X-ray microanalysis” (low-Z particle EPMA), that uses an energy-dispersive X-ray (EDX) detector with an ultrathin window, has been developed.3,4 The low-Z particle EPMA allows the quantitative determination of concentrations of low-Z elements, such as carbon, nitrogen, and oxygen, as well as chemical elements that can be analyzed by conventional energy-dispersive EPMA (ED-EPMA), in individual particles of micrometer size. To briefly summarize the overall procedure of aerosol analysis using the low-Z particle EPMA, first, particles are usually sampled on Ag foil using a seven-stage May cascade impactor.5 The May impactor has, at a 20 L/min sampling flow, aerodynamic cutoffs of 16, 8, 4, 2, 1, 0.5, and 0.25 µm for stages 1-7, respectively. The seventh stage is not analyzed because the very small size of particles collected on the stage is not suitable for EPMA measurements. The sampling duration varies between about 1 (for stage 6) and 180 (for stage 1) min, to obtain a good loading of particles at the impaction slots. Some 300 particles for each stage sample are analyzed. The measurements are carried out on a JEOL 733 electron probe microanalyzer equipped with an Oxford Link SATW ultrathin window EDX detector. To achieve optimal experimental conditions such as a low background level in the spectra and high sensitivity for low-Z element analysis, a 10-kV accelerating voltage is chosen. The beam current is 1.0 nA for all the measurements. To obtain statistically enough counts in the X-ray spectra while limiting the beam damage effect on sensitive particles, a typical measuring time of 10 s is used. A cold stage of the electron microprobe allows the analysis of particulate samples at liquid nitrogen temperature (∼-193 °C), which minimizes contamination (1) Frustorfer, P.; Niessner, R. Mikrochim. Acta 1994, 113, 239-250. (2) Weinbruch, S.; Wentzel, M.; Kluckner, M.; Hoffmann, P.; Ortner, H. M. Mikrochim. Acta 1997, 125, 137-141. (3) Ro, C.-U.; Osan, J.; Van Grieken, R. Anal. Chem. 1999, 71, 1521-1528. (4) Ro, C.-U.; Osan, J.; Szaloki, I.; de Hoog, J.; Worobiec, A.; Van Grieken, R. Anal. Chem. 2003, 75, 851-859. (5) May, K. R. J. Aerosol Sci. 1975, 6, 1-7. 10.1021/ac035149i CCC: $27.50

© 2004 American Chemical Society Published on Web 01/31/2004

and reduces beam damage to samples as well.6,7 Computercontrolled X-ray data acquisition for individual particles is carried out automatically in point analysis mode; i.e., the electron beam is focused on the center of each particle and X-rays are acquired while the beam stays on the single spot. The localization of the particles is based on inverse backscattered electron contrast. Morphological parameters such as diameter and shape factor are calculated by an image processing routine. These estimated geometrical data are set as input parameters for the quantification procedure. The net X-ray intensities for the elements are obtained by nonlinear least-squares fitting of the collected spectra using the AXIL program.8 The elemental concentrations of individual particles are determined from their X-ray intensities, by the application of a Monte Carlo calculation combined with reverse successive approximations.4,9 The quantification procedure provides results accurate within 12% relative deviations between the calculated and nominal elemental concentrations, when the method is applied to various types of standard particles such as NaCl, Al2O3, CaSO4.2H2O, Fe2O3, CaCO3, (NH4)2SO4, and NH4NO3.10-12 Since the quantitative concentrations of almost all chemical elements (Z > 5), can be obtained, chemical speciation, even for internally mixed individual particles, is possible. For example, in our previous studies,10,11 it was demonstrated that quantitative formula compositions of internally mixed individual atmospheric particles, such as (NH4)2SO4/NH4NO3 and CaCO3/ aluminosilicate mixtures, could be determined. This analytical procedure is useful to fully extract information on chemical compositions of individual particles from the low-Z particle EPMA data. To characterize environmental aerosol samples, the application of the low-Z particle EPMA technique begins with computercontrolled automatic measurement of some thousands of aerosol particles for each environmental sample. Since the number of particles is rather huge, the whole process to determine chemical species from all the particle X-ray data requires a lot of data interpretation time. Data interpretation procedure includes, first, the extraction of characteristic X-ray intensity values from raw X-ray spectral data, second, the calculation of atomic concentrations from the characteristic X-ray intensity values, and, third, chemical speciation on the basis of the atomic concentration data. The first two steps of data interpretation procedure can be performed in batch mode; i.e., several hundreds of particle data can be processed on a computer without being attended, and the batch-mode calculations can be finished in several hours. The most time-consuming step for data interpretation has been the chemical speciation from the resultant atomic concentration data. (6) Worobiec, A.; de Hoog, J.; Osan, J.; Szaloki, I.; Ro, C.-U.; Van Grieken, R. Spectrochim. Acta 2003, B58, 479-496. (7) Szaloki, I.; Osan, J.; Worobiec, A.; de Hoog, J.; Van Grieken, R.X-Ray Spectrom. 2001, 30, 143-155. (8) Vekemans, B.; Janssens, K.; Vincze, L.; Adams, F.; Van Espen, P. X-Ray Spectrom. 1994, 23, 278-285. (9) Szaloki, I.; Osan, J.; Ro, C.-U.; Van Grieken, R. Spectrochim. Acta 2000, B55, 1017-1030. (10) Ro, C.-U.; Osan, J.; Szaloki, I.; Oh, K.-Y.; Kim, H.; Van Grieken, R. Environ. Sci. Technol. 2000, 34, 3023-3030. (11) Ro, C.-U.; Oh, K.-Y.; Kim, H.; Chun, Y.-S.; Osan, J.; de Hoog, J.; Van Grieken, R. Atmos. Environ. 2001, 35, 4995-5005. (12) Osan, J.; Szaloki, I.; Ro, C.-U.; Van Grieken, R. Mikrochim. Acta 2000, 132, 349-355.

The conventional way for chemical speciation from EPMA data has been the application of multivariate analytical methods, either statistical13-15 or artificial neural networks.16,17 These analyses have usefully been applied in atmospheric environmental research, especially for data obtained by using conventional EDX detectors. However, conventional EDX detectors have a limited ability to detect low-Z elements because of their thick beryllium entrance window. Since conventional ED-EPMA cannot detect low-Z elements, many types of particles cannot be directly analyzed, e.g., organic particles that are inferred only from their high Bremsstrahlung background without their characteristic X-ray information. In addition, many atmospheric aerosol particles just provide partial information on their chemical compositions, when their characteristic X-ray spectral data are examined. For example, particles in the form of nitrates, sulfates, ammonium salts, oxides, or mixtures including a carbon matrix cannot definitely be analyzed without direct information on carbon, nitrogen, and oxygen. Furthermore, in many occasions, individual environmental particles are observed to be internally mixed with two or more chemical species, and in this case, it is also important to be able to fully utilize quantitative elemental concentration data. The low-Z particle EPMA data contain much richer information on the chemical species of individual particles, albeit with lack of information on hydrogen. Since the low-Z particle EPMA can provide almost complete information necessary for chemical speciation, it is natural to apply a method that can fully extract chemical composition information contained in the data. In our previous study,10 it was demonstrated how this analytical aim could be achieved; it was shown that formula concentrations of major chemical species in individual environmental particles could be determined. For example, formula concentrations of ammonium sulfate and nitrate were analyzed for particles internally mixed with them. This approach to determine chemical species from the low-Z particle EPMA data of individual particles has successfully been applied for real atmospheric particle samples and proven to be valuable to understand the nature of the particle sample.11,18,19 However, this approach has been applied manually, and thus, extracting the chemical speciation from the elemental concentration data is a quite time-consuming process. In this paper, we present an expert system, which can perform chemical speciation from X-ray data obtained by the low-Z particle EPMA. By applying the expert system, the time necessary for the chemical speciation of several thousands of particles becomes shortened very much and also all the detailed information on particle data can easily be saved and extracted later if more information is needed for further analysis. (13) Hoornaert, S.; Van Malderen, H.; Van Grieken, R. Environ. Sci. Technol. 1996, 30, 1515-1520. (14) Jambers, W.; Van Grieken, R. Environ. Sci. Technol. 1997, 31, 1525-1533. (15) Osan, J.; de Hoog, J.; Worobiec, A.; Ro, C.-U.; Oh, K.-Y.; Szaloki, I.; Van Grieken, R. Anal. Chim. Acta 2001, 446, 211-222. (16) Hopke, P. K.; Song, X. H. Anal. Chim. Acta 1997, 348, 375-388. (17) Song, X. H.; Hadjiiski, L.; Hopke, P. K.; Ashbaugh, L. L.; Carvacho, O.; Casuccio, G. S.; Schlaegle, S. J. Air Waste Manage. 1999, 49, 773-783. (18) Ro, C.-U.; Oh, K.-Y.; Kim, H.; Kim, Y. P.; Lee, C. B.; Kim, K. H.; Osan, J.; de Hoog, J.; Worobiec, A.; Van Grieken, R. Environ. Sci. Technol. 2001, 35, 4487-4494. (19) Ro, C.-U.; Kim, H.; Oh, K.-Y.; Yea, S. K.; Lee, C. B.; Jang, M.; Van Grieken, R. Environ. Sci. Technol. 2002, 36, 4770-4776.

Analytical Chemistry, Vol. 76, No. 5, March 1, 2004

1323

Figure 1. Overall scheme of the expert system.

DESCRIPTION OF EXPERT SYSTEM When a program utilizes the computational techniques that try to simulate human reasoning to solve complex problems, it is called an expert system. With the advent of fast, powerful, and inexpensive computers in the past decades, many applications of expert systems to solve specific problems have appeared in various research fields.20-24 Our expert system tries to mimic the logic used by experts in order to determine chemical species of individual particles using the low-Z particle EPMA data. The expert system is implemented by macroprogramming that is done by using the MS Visual Basic interpreter available in MS Excel software. The expert system runs on IBM-PC compatible computers and uses input and output files in the format of MS Excel files. The overall scheme of the expert system is shown in Figure 1. The inputs to this expert system are the concentration data for individual particles that are given as MS Excel files. The concentration data are the outputs of our iterative Monte Carlo calculation program that obtains atomic concentrations of particles from X-ray spectral data. The outputs of the expert system are the chemical species and formula concentrations of each particle, particle groups with similar chemical compositions, and distributions of particle groups in the different size ranges. The explanation for the routines shown in Figure 1 is provided more in detail below. (A) Input Routine. In this routine, information is provided on the working directory, directories where input and output files are, input and output file names, and samples and particles to be processed. The input files contain atomic concentration data for all the particles. The output files contain the list of chemical species in individual particles, the list of all chemical species whose formula fraction are >10%, the information on particle class groups classified on the basis of their chemical species, and the number of particles assigned in each class group for all stage samples (each stage of a cascade impactor collects differently sizesegregated particles). (B) Chemical Speciation Routines (EX_CLASS). In this part of the program, the speciation of particles is performed based on their atomic concentration data. This part is composed of four (20) Harrington, P. B.; Street, T. E.; Voorhees, K. J.; di Brozolo, F. R.; Odom, R. W. Anal. Chem. 1989, 61, 715-719. (21) Koutny, L. B.; Yeung, E. S. Anal. Chem. 1993, 65, 148-152. (22) Stiber, N. A.; Pantazidou, M.; Small, M. J. Environ. Sci. Technol. 1999, 33, 3012-3020. (23) Zhang, W.; Chait, B. T. Anal. Chem. 2000, 72, 2482-2489. (24) Ferreira, M. J. P.; Rodrigues, G. V.; Emerenciano, V. P. Can. J. Chem. 2001, 79, 1915-1925.

1324

Analytical Chemistry, Vol. 76, No. 5, March 1, 2004

routines where input data are read (ReadInput routine), the concentration data are preprocessed (SortElements routine), particles are roughly classified based on the chemical elements present in them (FindCases routine), and chemical speciation of particles is performed (ClassList routine). (B-1) SortElements Routine. In this routine, chemical elements with >1% in atomic fraction are searched and information on the names and concentrations of the elements is saved for each particle. The chemical elements are sorted in decreasing order of their concentrations. (B-2) FindCases Routine. Since it is known which chemical elements exist in the particle data, first, the particle data are distributed to different decision trees, according to the most abundant chemical element. And then the second most abundant chemical element is used to distribute the data to one level lower tree. Most of particle data are finally distributed when three or four chemical elements are used for the distribution. After distributing the particle data, ClassList routine is applied where the chemical speciation is performed. (B-3) ClassList Routine. This routine mostly includes a knowledge database for performing chemical speciation based on the pattern of elemental concentrations in the particle data. Since hydrogen cannot be detected, the low-Z particle EPMA has a limitation on the specific speciation of organic particles. And thus the particles are just classified either as organic or carbon-rich when the sum of the carbon and oxygen contents of the particles is larger than 90% in atomic fraction. When the carbon content is, somewhat arbitrarily, 3 times larger than the oxygen content, carbon-rich particles are differently identified from the organic particles. The low-Z particle EPMA is, of course, more powerful for the speciation of inorganic particles. It is well known by many other works, as well as by the application of the low-Z particle EPMA to aerosol particle characterization, that there exist various types of major inorganic particles in the atmosphere, to name a few: aluminosilicates, silicon dioxide, calcium carbonate, iron oxide, marine originated such as sodium chloride and sodium nitrate, ammonium sulfate particles, etc. In addition, the chemical composition of a particle is never exactly the same as that of others; it is also rare to see particles composed of only one pure chemical species and also particles with two or more chemical species have different compositions. And thus, this expert system tries to differentiate between inorganic species and also obtain information on the mixture state for internally mixed particles. Inorganic species are mainly divided into two chemical groups such as aluminosilicates and inorganic salts. The aluminosilicates group includes aluminosilicate species, composed of mainly aluminum and silicon oxides, often either with one or more minor elements, such as K, Na, Mg, Ca, S, Cl, Fe, and some transition metals. Inorganic salts group includes carbonate, sulfate, nitrate, and chloride combined with various cations such as ammonium, sodium, magnesium, potassium, and calcium. Internally mixed inorganic salt particles can be analyzed and the relative abundance of two or more chemical species in individual particles is determined. Just for example, an X-ray spectrum of a particle internally mixed with sulfate and chloride combined with sodium, magnesium, and calcium is shown in Figure 2. The atomic concentration of each element calculated by the iterative Monte Carlo calculation method is also shown.

Figure 2. X-ray spectrum of a particle internally mixed with sodium, magnesium, calcium, chloride, and sulfate species.

This particle is chosen from an Asian Dust sample collected during an Asian Dust event in ChunCheon, Korea, on April 7, 2000. Since the particle was collected on Ag foil, characteristic X-ray peaks of Ag also appeared. This particle is most probably a reacted sea salt with calcium species. The reaction of sea salt with SOx species must have happened while the particle traveled over the Yellow Sea before arriving to ChunCheon. The reasons why this is a reacted sea salt particle are as follows: first, the ratio of magnesium to sodium is 0.11, which is very close to that of seawater, 0.126,18 and second, remaining chloride from genuine sea salt still exists. The chemical form of the calcium species is uncertain but probably originated from a soil, either as gypsum or calcium carbonate reacted with SOx during long-range transport. Since it is not possible to exclude a calcium chloride for calcium chemical species, this particle is regarded to have sodium, magnesium, and calcium as its cations and sulfate and chloride as its anions. To calculate the relative abundance of chemical species existing in the particle, the equivalent atomic fractions of anions and cations are calculated. In this case, the equivalent atomic fraction of anions is 28.5% () 2 × 10.1% + 8.3% for sulfate and chloride) and that of cations is 28.4% () 14.8% + 2 × 1.6% + 2 × 5.2% for sodium, magnesium, and calcium) (see Figure 2). Since the equivalent atomic fractions of the cations and anions are very similar, the calculation proceeds. If the ratios of the atomic fractions are different (i.e., by >15%), then the calculating routine gives a warning in the output. For this particle, the calculated contents of Na2SO4, CaSO4, NaCl, CaCl2, MgSO4, and MgCl2 are 37, 26, 15, 11, 8, and 3%, respectively. The output of this calculation contains the list of chemical species in decreasing order of their formula concentrations and a particle ID code (Na2SO4-CaSO4NaCl-CaCl2) that is composed of all chemical species with its formula concentration of >10%. Up to now, a knowledge database of the expert system for the chemical speciation of individual particles has been built by using 13 data sets for two Asian Dust, 4 urban, and 7 marine aerosol particle samples collected in ChunCheon, Seoul, and Cheju island, Korea, respectively (overall ∼20 000 particles). The details on instrumental conditions of data acquisition and procedure of data

treatment are given in earlier papers.3,4,10,11,18,19 In Table 1 of Supporting Information, frequently encountered chemical elements used for distributions of particle concentration data to corresponding decision trees are listed with their possible chemical compositions. When particles are distributed based on their chemical elements, corresponding routines are used to specify the chemical compositions of the particles to produce their particle ID codes. This part can easily be updated and expanded if particles with the pattern of atomic concentration data unknown to the existing knowledge database of the expert system are encountered when data sets from new samples are tried for the chemical speciation. (C) Particle Grouping. The particle ID code is composed of the names of chemical species in decreasing order of their formula concentrations. For example, if a particle were a reacted sea salt and would contain Na and Mg as its cations and NO3- and SO4-2 as its anions, its particle ID code could be either “NaNO3-Mg(NO3)2-Na2SO4-MgSO4” or “Na2SO4-MgSO4-NaNO3-Mg(NO3)2” among many other possible ID codes, depending on their formula concentrations. Even though these particle ID codes are different, it is clear that they represent particles with similar chemical compositions. In this part of the expert system, the two particles are grouped as (Na,Mg)(NO3,SO4) particle group by checking their particle ID codes. (D) Bookkeeping. For our analysis, a May cascade impactor has usually been used for aerosol sampling.5 This impactor has seven stages; the cutoff diameters of its 1-7 stages are 16, 8, 4, 2, 1, 0.5, and 0.25 µm, respectively. The seventh stage is not analyzed because the very small size of particles collected on it is not suitable for EPMA measurements. Some 300 particles for each stage are usually measured by the low-Z particle EPMA. This bookkeeping routine calculates how many particles are assigned to individual particle groups for all stage samples. By analyzing or plotting bookkeeping results, it is easy to understand the characteristics of aerosol samples. (E) Checking the Classified Results. The investigation of X-ray spectra of particles can be very helpful when it is necessary to check whether all the particles are correctly classified. For this purpose, X-ray spectra are generated and stored for all the Analytical Chemistry, Vol. 76, No. 5, March 1, 2004

1325

Table 1. Result of Chemical Speciation of Standard Particles Using the Expert System standard particles

number of correctly identified particles

algae lichen CaCO3 Fe2O3 Na2CO3 NaCl NH4NO3 (NH4)2SO4 Al2O3 CaCl2 CaSO4 K2CO3 K2SO4 KCl KNO3 MgSO4 Na2SO4

20 21 13 21 14 19

sum

19 18 18 20 19 19 15 18 19 19 292

number of particles incorrectly identified

6 18 1 4 2 1 (organic) 1 5 2 1 41

identified particle groups. If one spectrum looks different from those of particles in the same particle groups, the particle of the spectrum has to be treated in a different way; i.e., a corresponding routine for the chemical speciation of the particle has to be updated to correctly classify the particle. (F) Utilities. Since information on particle groups, formula compositions, and atomic and formula concentrations of each particle is available, it is possible to extract further detailed information on aerosol samples. For example, it is important to analyze contents of minor chemical elements for aluminosilicatecontaining particles of Asian Dust samples, because the information can reflect different sources of the samples. In addition, it is easy to calculate how many times each chemical species is encountered for all the analyzed particles and what the overall weight percent composition of each chemical species is since the size of each particle can be obtained in EPMA measurement and the density of each particle can be estimated by quantitatively elucidating the chemical compositions. The weight percent data of each chemical species are expected to provide a link between single-particle analysis and bulk analysis results. To meet possible specific needs in the characterization of various types of environmental aerosol particle samples, corresponding utility programs can be implemented. APPLICATION TO STANDARD PARTICLES The expert system established using 13 data sets with overall ∼20 000 particles was tested by applying it to the concentration data of various types of particles with known chemical compositions. Seventeen types of standard particles included various inorganic particles as well as algae and lichen (see Table 1). It was attempted to identify overall 333 particles, ∼20 particles for each chemical species. The data of the standard particles were collected either on Al or on Ag foils. Details on the data set of the standard particles are given elsewhere.4,15 The performance of the expert system is very good; among 333 particles, 292 particles were correctly identified. Since algae and lichen particulates are biogenic particles, they were classified either as organic or as carbon-rich particles, depending on their carbon and oxygen 1326 Analytical Chemistry, Vol. 76, No. 5, March 1, 2004

contents. Since hydrogen content cannot be determined, it is impossible to differentiate algae and lichen particulates. The expert system could not identify ammonium nitrate particles because the expert system did not include any knowledge database for particles with only nitrogen and oxygen concentrations at the time of the test run. Now the expert system can identify ammonium nitrate particles after update. The remaining 23 particle data, which could not be classified, were checked by investigating their concentration data and X-ray spectra. Without any exception, the data themselves deviate from the theoretical values and their spectral patterns are surely different from those of correctly identified particles. In other words, the expert system works very well, and the particle data with atypical concentration patterns for specific chemical species are easily identified. The atypical particle data seem to be acquired for impurity particles or for particles that were bounced off by electron beam during X-ray data acquisition. APPLICATION TO ENVIRONMENTAL AEROSOL PARTICLES The expert system was applied to the data obtained from an Asian Dust sample. The sample was collected on April 7, 2000 in ChunCheon, Korea, during a severe Asian Dust event. The May cascade impactor was used for the collection of the sample, and some 200 particles on each stage were analyzed (total number of analyzed particles, 1100). Even though detailed characterization of the sample is given elsewhere,25 in this study, the emphasis is on how the expert system works. Among overall 1100 particle data, 145 cases with different chemical compositions, i.e., different particle ID codes, were identified, as shown in Table 2 of Supporting Information. Among the 145 cases, the most abundant chemical compositions are SiO2-Al2O3, followed by SiO2-Al2O3C, SiO2, organic, etc. Since the knowledge database of the expert system was built with 13 data sets including 2 other Asian Dust sample data, the identification rate is very high (94.7%). They are further grouped into 58 groups (see Table 3 of Supporting Information). The most abundant particle group is formed by aluminosilicates (560 particles; notated as AlSi in Table 3 of Supporting Information), followed by SiO2 (118 particles), internal mixtures of aluminosilicate and carbonaceous species (68 particles; notated as AlSi-C in Table 3 of Supporting Information), organic (64 particles), carbon-rich (36 particles), CaCO3 (23 particles), etc. All the X-ray spectra and concentration data for the identified particles were examined manually, and we did not observe any systematic malfunction of the expert system. The output of the expert system also includes the number of particles assigned in each particle group for each sample stage, which is important to understand the nature of the collected Asian Dust sample; the information is omitted here for brevity. It took ∼1 h to classify the particle data using the expert system on an IBM-PC compatible computer. As shown in Tables 2 and 3 of Supporting Information, there are many different types of identified particles, and this complexity in chemical compositions of aerosol samples can be manipulated fast and reliably with the application of the expert system. CONCLUSIONS Since the low-Z particle EPMA data contain very rich information on chemical speciation for individual aerosol particles, an (25) Ro, C.-U.; Hwang, H. J.; Kim, H.; Van Grieken, R., in preparation.

expert system has been developed in order to fully extract information contained in the low-Z particle EPMA data. The expert system tries to mimic the logics used by experts and worked very well when it was applied to data for standard particles and a real aerosol sample. Currently, we are working on the analysis of bottom and fly ashes from municipal solid waste incinerators using the low-Z particle EPMA technique. The attempt to apply the expert system to perform chemical speciation for ash data was not successful because the knowledge database of the expert system was not built from any ash data. By updating the knowledge database using samples with new characteristics, it is expected that the expert system will get more general application scope for aerosol characterization.

ACKNOWLEDGMENT This research was supported by the Ministry of Environment, Republic of Korea (Eco-technopia 2001, 16-018) and also a research grant from Hallym University, Korea. SUPPORTING INFORMATION AVAILABLE Additional tabular information as noted in the text. This material is available free of charge via the Internet at http:// pubs.acs.org.

Received for review September 30, 2003. Accepted December 30, 2003. AC035149I

Analytical Chemistry, Vol. 76, No. 5, March 1, 2004

1327