Machine learning models for characterizing macroscopic properties of diesel/ biodiesel surrogate fuels

Resumo

Accurate determination of fuel properties of complex mixtures over a wide range of pressure and tem-
perature conditions is essential to utilizing alternative fuels. Obtaining thermophysical properties of complex fuels

is important for design and analysis but difficult to measure/predict, especially in the range of extreme conditions

operations. Molecular dynamics (MD) simulations have been widely used to characterize physicochemical proper-
ties of fuels, including transport properties at supercritical conditions. Although MD simulations provide molecular

details that can be potentially be used to predict fuel properties accurately, they are generally too expensive in terms
of computational costs. In addition, MD predictions also need to be validated against experimental measurements,

which can be even more costly, especially in extreme conditions. Accordingly, it is not feasible to establish com-
plete and detailed fuel property databases consisting of a wide range of pressure and temperature conditions using

MD simulations or experimental measurements. Machine learning (ML) has great potentials to discover from data
the relation between inputs and outputs of complex systems. ML can be a powerful tool to predict fuel properties
from chemical compositions of the fuel mixture and/or chemical structures of the fuel molecules. Those models
can be trained using the database from MD simulations and/or experimental measurements in a data-fusion-fidelity

approach. The present work aims to construct cheap-to-compute ML models to act as closure equations for pre-
dicting the physical properties of diesel/biodiesel surrogate fuels. Here, Gaussian Process (GP) and probabilistic

generative models are adopted. GP is a popular non-parametric Bayesian approach to build surrogate models
mainly due to its capacity to handle the aleatory and epistemic uncertainties. Generative models have shown the
ability of deep neural networks employed with the same intent. In this work, machine learning analysis is focused
on a particular property, the fuel density, but it can also be extended to other physicochemical properties.