The Application Of Chemometrics To The Detection And Classification Of Ignitable Liquids In Fire Debris Using The Total Ion Spectrum

Current methods in ignitable liquid identification and classification from fire debris rely on pattern recognition of ignitable liquids in total ion chromatograms, extracted ion profiles, and target compound comparisons, as described in American Standards for Testing and Materials E1618-10. The total ion spectra method takes advantage of the reproducibility among sample spectra from the same American Society for Testing and Materials class. It is a method that is independent of the chromatographic conditions that affect retention times of target compounds, thus aiding in the use of computer-based library searching techniques. The total ion spectrum was obtained by summing the ion intensities across all retention times. The total ion spectrum from multiple fire debris samples were combined for target factor analysis. Principal components analysis allowed the dimensions of the data matrix to be reduced prior to target factor analysis, and the number of principal components retained was based on the determination of rank by median absolute deviation. The latent variables were rotated to find new vectors (resultant vectors) that were the best possible match to spectra in a reference library of over 450 ignitable liquid spectra (test factors). The Pearson correlation between target factors and resultant vectors were used to rank the ignitable liquids in the library. Ignitable liquids with the highest correlation represented possible contributions to the sample. Posterior probabilities for the ASTM ignitable liquid classes were calculated based on the probability distribution function of the correlation values. The ASTM ignitable liquid class present in the sample set was identified based on the class with the highest posterior probability value. iv Tests included computer simulations of artificially generated total ion spectra from a combination of ignitable liquid and substrate spectra, as well as large scale burns in 20’x8’x8’ containers complete with furnishings and flooring. Computer simulations were performed for each ASTM ignitable liquid class across a range of parameters. Of the total number of total ion spectra in a data set, the percentage of samples containing an ignitable liquid was varied, as well as the percent of ignitable liquid contribution in a given total ion spectrum. Target factor analysis was them performed on the computer-generated sample set. The correlation values from target factor analysis were used to calculate posterior probabilities for each ASTM ignitable liquid class. Large scale burns were designed to test the detection capabilities of the chemometric approach to ignitable liquid detection under conditions similar to those of a structure fire. Burn conditions were controlled by adjusting the type and volume of ignitable liquid used, the fuel load, ventilation, and the elapsed time of the burn. Samples collected from the large scale burns were analyzed using passive headspace adsorption with activated charcoal strips and carbon disulfide desorption of volatiles for analysis using gas chromatography-mass spectrometr