Self-training, self-optimizing expert system for interpretation of the infrared spectra of environmental mixtures.
Ying-S; Levine-SP; Tomellini-SA; Lowry-SR
Anal Chem 1987 Sep; 59(17):2197-2203
The intIRpret computer program for the identification of the principal components of chemical mixtures of hazardous wastes based on analysis by infrared spectrometry was described. The program was comprised of five subroutines including interferogram analysis and peak selection, automated knowledge acquisition, system optimization, interpretation, and final processing for subtracting spectral similarities. IntIRpret differed from the earlier program for automated waste mixture identification (PAWMI) by the automatic entering of peaks and rules through the software. The PAWMI program utilized operator chosen and entered rule peaks whereas the intIRpret program used rule peaks that were chosen by the first subroutine and weighted for frequency of occurrence, intensity, and cross terms. The method was tested using 62 organic compounds commonly identified at hazardous waste sites and 67 mixtures of four components using the same compounds. The intIRpret analyses showed a 40 percent decrease in false positive results and a 24 percent decrease in false negative results relative to similar analyses made using the PAWMI program. The intIRpret false positive rate was 19.4 percent, and the rate of false negatives was 1.2 percent. Problems with the intIRpret model centered on uncertainty of peak resolution due to peak shifts in solution, structural similarities of different compounds, and the inability of the first subroutine to recognize peaks appearing as unresolved shoulders or in poorly resolved envelopes.
NIOSH-Publication; NIOSH-Grant; Aromatic-hydrocarbons; Analytical-methods; Occupational-hazards; Infrared-spectrophotometry; Analytical-chemistry; Organic-solvents; Waste-treatment; Environmental-contamination; Information-processing;
Environmental & Indust Health 1420 Washington Heights Ann Arbor, Mich 48109-2029