Comparison of Infrared Spectroscopic Methods in Predicting DCN of Jet Fuels and Their Blends Using Chemometric Tools

Abstract

Abstract Fuels are made up of thousands of chemical components, but these can be grouped into far fewer chemical functional groups. Fuel properties can be correlated to simpler fuel representations, potentially as simple as the composition of their functional groups. Machine learning (ML) models have been built to predict fuel properties from spectroscopic measurements (preferably in liquid phase). Derived cetane number (DCN) is one such property, which is an indicator of ignition quality in compression-ignition engines. Attenuated total reflectance (ATR) spectroscopy has advantages over transmission-based approaches to overcome challenges associated with the narrow pathlength liquid cells (< 10μm) needed to avoid saturation in strong absorption bands in the mid-infrared. In our previous work, we demonstrated the use of Fourier transform infrared (FTIR) spectra (through transmission) in predicting the DCN of jet fuels after eliminating saturated regions. Convolutional neural network (CNN) models performed the best with a mean percentage error of 3% with an R2 score of 0.931 over a spectral region that includes the fingerprint region. Although those models have shown high predictive capability, since they ignored information from regions likely to show saturated features, it is intuitive that prediction capabilities could be improved further with access to the information in these regions. To evaluate the impact of these saturation regions on DCN estimation, we compare the prediction accuracies and important regions corresponding to DCN obtained from models trained on FT-ATR spectra, which include the regions excluded in the previous work. In this work, uncorrected ATR spectral measurements were used in developing calibration models for predicting DCN of jet fuels and their blends. ATR spectra of jet fuels like F-24, an alcohol-to-jet fuel, Jet-A, etc., and neat hydrocarbons and their mixtures, which in their assembly, represent the chemical functional groups present in jet fuels, were collected using an analytical grade FTIR spectrometer. Spectral measurements between 4000–650cm−1 were used in this analysis, and important spectral regions associated to functional groups correlating to DCN were also identified.