Derived cetane number prediction of jet fuels and their functional group surrogates using liquid phase infrared absorption

Abstract

Derived cetane number (DCN) is an important fuel property relevant to ignition in compression ignition engines. The measurement of IR spectra of neat liquid fuels along with robust models for DCN potentially opens pathways for DCN sensing using miniaturized sensors for engine control. For neat liquid fuels, there are practical measurement difficulties because of strong absorption features in the mid-infrared. This study assesses the viability of using subsets of the IR spectrum, primarily 2400–800cm−1, which contains the fingerprint region, for generating models for DCN. IR absorption spectra in the liquid phase and associated DCNs were experimentally collected on a dataset including some real jet fuels, hydrocarbon components that span the range of chemical functional groups encountered in those jet fuels, and mixtures of those components. Four machine learning models were used to evaluate the association of liquid IR absorption over different spectral ranges to fuel DCN. Non-linear models, including Convolutional Neural Networks performed excellently. Regions likely to exhibit saturation can be excluded without problems in generating models. A spectral resolution of 14 cm−1 appears to be optimal for measurements. There is no loss in accuracy in limiting the modeling only to the 2400–800 cm−1 region. Although only a few real fuel mixtures were assessed in this way, models were able to yield good predictions for these fuels, even though the model was trained primarily on simpler surrogate mixtures exhibiting the same chemical functional groups.