With the advent of Fourier transform infrared (FTIR) instrumentation, IR techniques have seen increasing application in the area of environmental analysis. Environmental analysis applications of FTIR include the characterization of hazardous waste mixtures and the speciation and quantitation of airborne vapors and gases both in ambient air, vehicular emissions and the workplace. The most commonly applied method of spectral identification uses forward searching of infrared spectral libraries. This method succeeds where the compound and the resulting spectrum is pure, or where the spectrum of a commercial mixture is available. The method fails in the case of the spectra of mixtures that are not stored as mixtures in the library. Methods of identifying the compounds in the spectra of mixtures, especially in environmental mixtures, have been pioneered by our group, based on the work of Woodruff, et al, and Herget, et al. This approach was achieved through the use of a three level rule structure for each peak, giving increasing """"""""goodness"""""""" scores for each peak in successively narrower frequency windows. At the heart of the programs that are used to accomplish either compound or compound class identification are PAIRS and PAWMI. This proposal is aimed at research that will enable these programs to self-train and self- optimize. Specifically, the hypothesis is: A self-training, self-optimizing expert system can be developed to identify the components of mixtures of environmental significance using infrared spectrocopy. 1. Optimization of weighting factors for peak goodness based on frequency of occurrence of peaks in each wavenumber window for given training sets. 2. Optimization of peak window widths for rules containing three peak position windows for each expected absorption. 3. Incorporation of a program to deconvove overlapping peaks. 4. Incorporation of the automated rule generator, the automated rule optimizer, the deconvolution and peak picking programs, and the mixture interpretation program into one expert system applicable to both condensed and vapor phase species. Each of the above specific aims involves basic research into the use of expert systems for spectroscopic analysis. This research will be conducted using appropriate training sets of environmental significance.
