Chemical boundary conditions for the classification of aerosol particles using computer controlled electron probe microanalysis Willemien Anaf n , Benjamin Horemans, Rene ´ Van Grieken, Karolien De Wael Department of Chemistry, University of Antwerp, Universiteitsplein 1, 2610 Antwerp, Belgium article info Article history: Received 8 May 2012 Received in revised form 12 September 2012 Accepted 22 September 2012 Available online 30 September 2012 Keywords: Classification Airborne particulate matter Single particle analysis EPMA abstract A method for the classification of individual aerosol particles using computer controlled electron probe microanalysis is presented. It is based on chemical boundary conditions (CBC) and enables quick and easy processing of a large set of elemental concentration data (mass%), derived from the X-ray spectra of individual particles. The particles are first classified into five major classes (sea salt related, secondary inorganic, minerals, iron-rich and carbonaceous), after which advanced data mining can be performed by examining the elemental composition of particles within each class into more detail (e.g., by ternary diagrams). The CBC method is validated and evaluated by comparing its results with the output obtained with hierarchical cluster analysis (HCA) for well-known standard particles as well as real aerosol particles collected with a cascade impactor. The CBC method gives reliable results and has a major advantage compared to HCA. CBC is based on boundary conditions that are derived from chemical logical thinking and does not require a translation of a mathematical algorithm output as does HCA. Therefore, the CBC method is more objective and enables comparison between samples without intermediate steps. & 2012 Elsevier B.V. All rights reserved. 1. Introduction Automated chemical analysis of individual aerosol particles with computer controlled electron probe microanalysis (CC- EPMA) produces a vast body of numerical data which has to be translated to a physical reality. The analytical scientist can rely on his ability of logical reasoning and decision making. However, his limited processing power makes him rapidly outnumbered when he is analyzing huge particle numbers. Luckily, the discipline of chemometrics has facilitated his task by offering some mathema- tical tools for data mining. Clustering analysis is a popular chemometric technique which has been frequently used for the classification of aerosol particles in groups of chemical similarity [1–3]. Since clustering is based on mathematic algorithms, it is necessary to interpret the chemical relevance of the extracted clusters [4,5]. This interpretation depends on the experience of the user, and therefore introduces some degree of subjectivity and irreproducibility. Expert systems can overcome such problems by mimicking the human ability of decision making. Such computer systems combine a knowledge base and an inference engine (computer programme). Ro et al. [6] developed an expert system for the automatic recognition and classification of chemical species in individual aerosol particles. As an output, the system gives an overview of the chemical species present in every single particle and the class to which the particle belongs, based on the chemical composition. Every particle is labelled with an ID code, composed of the formula of the chemical species with an abundance of more than 10%. Since particles are rarely in a pure form but contain some internal degree of heterogeneity; this leads to a very specific and complex particle classification scheme. Other chemical-based classification methods start from the net count rates from energy-dispersive X-ray microanalysis, combining the chemical composition with morphological information [7–9]. The chemical criteria are sim- ple, giving a very basic classification output. In the present article, an alternative classification method is presented, which can be considered as an elementary system based on an introductory knowledge base. The idea of the system is different from the system developed by Ro et al. [6]. In first instance, it uses chemical boundary conditions (CBC) for the automated classification of particles into five major classes (sea salt related, secondary inorganic, minerals, iron-rich and carbo- naceous). In this way, particles are easily and rapidly grouped into general classes. Afterwards, the elemental composition of parti- cles within each class can be examined into more detail, depend- ing on the goals and needs of the research. Contents lists available at SciVerse ScienceDirect journal homepage: www.elsevier.com/locate/talanta Talanta 0039-9140/$ - see front matter & 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.talanta.2012.09.051 n Corresponding author. Tel.: þ32 32653411; fax: þ32 32652376. E-mail addresses: willemien.anaf@ua.ac.be (W. Anaf), benjamin.horemans@ua.ac.be (B. Horemans), rene.vangrieken@ua.ac.be (R. Van Grieken), karolien.dewael@ua.ac.be (K. De Wael). Talanta 101 (2012) 420–427