SAFprop: A Dual Spectroscopic Approach for Predicting Properties of Sustainable Aviation Fuels †

Abstract

This study introduces a novel method for predicting sustainable aviation fuel (SAF) properties by combining liquid-phase attenuated total reflectance -Fourier transform infrared (ATR-FTIR) spectroscopy with confocal Raman spectroscopy. While both techniques have been individually applied to fuel analysis, their combined use in liquid-phase measurements represents a significant advancement. The research provides a comprehensive, experimentally measured spectral dataset of pure components, surrogate blends, and SAFs, moving beyond synthetic spectra used in previous studies. This dataset is utilized to compare the performance of ATR-FTIR and Raman spectroscopy both independently and in combination. Convolutional neural networks (CNNs) are trained to predict nine key fuel properties from spectral data, including derived cetane number (DCN), molecular weight (MW), hydrogen-to-carbon (H/C) ratio, density at 20 • C, net heat of combustion (NHC), threshold sooting index (TSI), flash point, freezing point, and kinematic viscosity (KV) at -20 • C. Multiple fusion strategies for combining the spectral data were systematically evaluated, including convolution, concatenation, and point-wise addition, which generally improved prediction accuracy over single-modality models. Specifically, point-wise addition reduced DCN prediction error by 18% relative to the FTIR-only model. Concatenation yielded improvements of 14% for flash point and 25% for freezing point compared to the Raman-based model, and outperformed Raman by 11% for H/C ratio predictions. The prediction errors were generally within or close to the reproducibility limits specified by relevant ASTM standards. The findings offer valuable insights into the suitability of these spectroscopic techniques for rapid, accurate, and cost-effective pre-screening of SAFs.