Development of chemometric methodologies for supervised class-based discovery experiments using GC×GC-TOFMS: application to aerospace fuel analysis

Abstract

Comprehensive two-dimensional gas chromatography with time-of-flight mass spectrometry detection (GCÃ GC-TOFMS) is a prominent instrumental technique which produces information rich datasets for the analysis of complex volatile mixtures. Due to this fact advanced data analysis methods are required and chemometric supervised discovery approaches excel at by exploiting class-based experimental design to extract useful chemical information. Analysis of aerospace kerosene-based fuels benefit from these approaches as the samples often contain thousands of unique compounds and features-of-interest may be minute in concentration. This dissertation describes several studies pertaining to the development and application of supervised discovery experiments (tile-based Fisher ratio (F-ratio) and 1v1 analysis) to fuels, focused onextracting pertinent information. First, tile-based F-ratio results were used alongside measures of peak shape consistency to statistically determine which m/z were pure (comprised of a single compound) for each hit for accurate relative quantitation. Next, building on the ideas of the first project, tile-based F-ratio results are used to statistically determine which m/z are consistent between classes to enable the extraction of a purified mass spectrum representing the differences between classes for reliable compound identification. Next, a class-based experiment is set up to target the identities of extractable polar contaminants in rocket fuel using solid-phase extraction. This approach is then applied to multiple fuels for the purpose of inter-batch analysis for errant fuel detection. The extraction behavior of polar compounds within a fuel matrix is then investigated with tile-based 1v1 analysis. Differences in extraction profiles are investigated for several for two different stationary phases as a function of load volume. Finally, partial least squares (PLS) is combined with tile-based variance ranking and RreliefF to improve modeling of carbonaceous deposits in thermal fuel lines, a measure of thermal stability, for a dataset of rocket fuels. The results are then used to identify the compounds that are highly correlated with increased carbon deposition.