Analysis of mixtures containing isomeric hydrocarbons using process mass spectrometry

The feasibility of process mass spectrometry (PrMS) as a method for the rapid analysis of a piperylene stream is assessed beginning with a focus on resolving binary mixtures of isomeric hydrocarbons. This application was suggested by an industrial member of the University of Tennessee\u27s Measurement and Control Engineering Center (MCEC). The piperylene stream consists mostly of cis- and trans-l,3-pentadiene (totaling over 55% of the stream) along with cyclopentene (~20%) and 2-methyl-2-butene (~10%), plus others at lower levels: cyclopentane, cis- and trans-2-pentene, n-pentane. isopentane, and 2,2-dimethylbutane. This particularly challenging PrMS application involved the need for resolution of hydrocarbon isomers which were constituents in the stream. This required the design and construction of a vaporization inlet since these components are liquids at standard temperature and pressure. It was found that parameterization (selection of which m/z signals to monitor in the deconvolution of the mixture mass spectra) was critical to the accuracy and precision of the mixture component concentration estimates. Use of the entire mass spectrum (an intuitive approach to parameterization) was found to be non-optimum since inclusion of some m/z signals deteriorated the accuracy and precision. A brute force approach to parameterization (whereby the accuracy is assessed using all possible combinations of ion intensities) was found to afford the most accurate concentration estimates but was useful for small datasets due to the large number of calculations required. The brute force method of parameterization was suspected of over-fitting the data due to the relatively small datasets utilized. In response to the large amount of calculation time required for the brute force method, other faster parameterization methods were evaluated. These faster parameterization methods included a genetic algorithm, an empirical algorithm and algorithm (process stream evaluator) provided with the instrument data acquisition system. The genetic and empirical algorithm provided parameterizations faster than the brute force method. However, the accuracy and precision of concentration estimates were not as good. The success or performance of mixture deconvolution was measured using validation plots and correlation coefficients (r-squared). Validation plots are graphs of estimated versus true mole fraction of gravimetrically prepared standard samples and the correlation coefficient is a number from 0 to 1 describing the accuracy and precision of estimated concentrations where a correlation coefficient of 1 is a perfect fit to the Y = X line. An r² of 0.9963 was obtained from analysis of binary mixtures of n-pentane / isopentane using the brute force method of parameterization. Similarly, correlation coefficients of 0.9054 and 0.9986 were obtained for binary mixtures of cis- / transpiperylene and cis- / trans-2-pentene. The concentrations of these samples ranged from approximately 0.10 to 0.90 mole fraction. The lower correlation coefficient for the piperylene system was attributed to the greater similarity between the pure component mass spectra (reference spectra) of these isomers and their relatively greater chemical reactivity. The similarity of the reference spectra was measured using the Drahos-Vekey similarity index equation and the reactivity was compared by considering the double bonds in the molecule. Finally, a nine-component standard mixture similar in composition to a piperylene stream grab sample was analyzed by GC-MS and PrMS. The additional complexity incurred by increasing the number of components to a total of nine required that the stereoisomer pairs each be treated as one quasi-component. In other words, to achieve a reasonable accuracy and precision each pair of cis- / trans- isomer reference spectra were averaged and considered as one component leaving a total of components. Upon reducing the mixture complexity to a total of seven components it was found that cis- / trans-piperylene, 2-methyl-2-butene and cyclopentene were among the major components in the piperylene stream that could be quickly analyzed with reasonable accuracy and precision by PrMS. The remaining components were at levels that appeared to be present below the limit of detection