Chapter 16 Use of a Rule-Building Expert System for Classifying Particles Based on Sem Analysis

Abstract

Computer-controlled scanning electron microscopy (CCSEM) permits the characterization of size, shape, and composition of individual particles and thus provides a rich source of data to identify the origin of ambient airborne particles. The problem is how to make best use of this information. The general procedure has been to assign each particle to a class of similar particles based on its x-ray fluorescence spectrum. The initial efforts developed the class characteristics and classification rules in an empirical fashion. Recent studies have suggested that greater specificity and precision in the subsequent class balance analysis can be obtained if particle classes are more homogeneous. To obtain the classification of large numbers of particles in an efficient manner, the x-ray intensity data have been subjected to an agglomcrative hierarchical cluster analysis. The resulting groups of samples are then used as candidate examples for a rule-building expert system that provides a decision tree that can be used for subsequent particle classification. An error analysis approach based on jackknifing has also been developed. The use of this approach to characterize particle source emissions from a large coal-fired electric generating stations was employed. It was found that the samples collected in the stack using a dilution sampler could be distinguished from the particles collected in the plume emitted by the plant. It was also found that the plume could be readily distinguished from the particles in the ambient air upwind of the plant.