We present a methodology for assessing similarity between the 13C Nuclear Magnetic Resonance (NMR) spectrum of a novel experimental fuel of interest and a set of spectra from reference hydrocarbon fuels. The methodology is aimed at the application of 13C NMR spectra for building empirical models for property prediction. Using data from a set of complex biofuels, our results show the methodology is effective in finding relevant differences among fuels that may be difficult to determine by visual inspection of the 13C NMR spectra. An important consideration when applying empirical models for predicting the property values of an experimental fuel is to determine how closely the 13C NMR spectrum of the new fuel resembles those used to fit the model. Empirical models tend to produce more accurate predictions when used for interpolation rather than for extrapolation. However, determining the closeness between a new observation and the dataset used to fit a predictive model is not always straightforward, and no single metric can capture the similarities and differences present in a high-dimensional space. This problem is made more difficult because the 27 dimensions comprising different integration regions of a 13C NMR spectrum, each a descriptor of a particular carbon type, define a space too large to be fully explored. The problem is exacerbated by the relatively small number of samples in the datasets commonly available to create empirical models. Our approach searches for outliers in spaces with few dimensions and provides an overall view, in a low-dimensional space, of the position of the experimental fuel with respect to others in the set. The most and least similar fuels from a particular set are determined based on the linear correlation with the spectrum of the experimental fuel, and we provide a measure of extrapolation that can be used in some higher-dimensional spaces. Our results show that fuels of similar end use tend to cluster together in this type of low dimensional space representation.

Similarity, outlier detection, and representation in reduced dimensionality space of 13C nuclear magnetic resonance spectra of fuels used for building empirical models

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.