A machine learning framework for drop-in volume swell characteristics of sustainable aviation fuel
Shane T. Kosir, Joshua S. Heyne, J. Graham
Abstract
A machine learning framework has been developed to predict volume swell for 10 non-metallic materials submerged in neat compounds. The non-metallic materials included nitrile rubber, extracted nitrile rubber, fluorosilicone, low temp fluorocarbon, lightweight polysulfide, polythioether, epoxy (0.2 mm), epoxy (0.04 mm), nylon, and Kapton. Volume swell, a material compatibility concern, serves as a significant impediment for the minimization of the greenhouse gas emissions of aviation. Sustainable aviation fuels, the only near and mid-term solution to mitigating greenhouse gas emissions, are limited to low blend limits with conventional fuel due to material compatibility issues (i.e. O-ring swell). A neural network was trained to predict volume swell for non-metallic materials submerged in neat compounds. Subsequent blend optimization incorporated nitrile rubber volume swell predictions for iso- and cycloalkanes to create a high-performance jet fuel within ‘drop-in’ limits. The results of this study are volume swell predictions for 3 of the 10 materials -nitrile rubber, extracted nitrile rubber, and polythioether- with holdout errors of 12.4% or better relative to mean volume swell values. Optimization considering nitrile rubber volume swell achieved median specific energy [MJ/kg] and energy density [MJ/L] increases of 1.9% and 5.1% relative to conventional jet fuel and an average volume swell of 6.2% v/v which is within the range of conventional fuels. Optimized solutions were heavily biased toward monocycloalkanes, indicating that they are a suitable replacement for aromatics. This study concludes that cycloalkanes can replace aromatics in jet fuel considering volume swell and other operability requirements while significantly reducing soot and particulate matter emissions.