Aviation Turbine Fuel Thermal Conductivity: A Predictive Approach Using Entropy Scaling-Guided Machine Learning with Experimental Validation

[Image: see text] Although typical aircraft fuel thermal management analysis relies upon temperature-dependent thermodynamic and transport properties of aviation turbine fuel, the variation in properties associated with compositional variation in fuels and the subsequent impacts on system performance are not well established. With this in mind, the present work aimed to develop a predictive model of aviation turbine fuel thermal conductivity which utilized only compositional (hydrocarbon) and state (temperature and pressure) inputs and had errors within the bounds of typical uncertainty of the associated test data (3%). A novel modeling approach was developed to predict thermal conductivity using pseudo-component entropy scaling techniques with a machine learning-developed intermediate step in the overall model. Simple hyper-parameter optimization techniques were developed to promote model stability, computational efficiency, and long-term repeatability of the novel architecture. Validation data were gathered which included four fuel samples (3 JP-5 and 1 F-24), which underwent two-dimensional gas chromatography compositional testing and temperature-dependent density, viscosity, thermal conductivity, and specific heat testing. Model performance on the validation data set assembled from the literature data and present efforts showed an average deviation of 1% and an absolute average deviation of 2.5%. Model outputs outside the validation range are well-behaved and are expected to perform well on a large range of liquid hydrocarbon mixtures with the overall process expected to be well suited to prediction of other properties.