Despite the wide adoption of hydrocarbon-based fuels, significant challenges are still present in modeling their combustion phenomena. The complexity of the oxidation kinetics is exacerbated by the multitude of different components present in common liquid fuels.

In this study, a workflow for the automatic generation of kinetic models for the oxidation of real fuels is developed. The modeling focuses on the oxidation kinetics of alkanes, the main component of most liquid fuels. The study aims to achieve the following objectives:

• Accurate modeling of the low-temperature oxidation phenomenon

• Modeling of complex real-fuel mixtures, including diesel and jet fuels

• Generation of computationally lightweight models, suitable for complex simulation models

• Fast and straightforward model generation, limiting user inputs and manual modeling

This work is subdivided into four main parts.

In the first section, a methodology for the automatic generation of detailed kinetic models is developed based on the Mamox++ framework. Low-temperature oxidation submechanisms for alkanes are automatically compiled based on a predefined set of reaction pathways and rate rules.

The second section concerns the development of an automatic model lumping procedure. A methodology for the computation of the rates of lumped reactions and thermochemical properties of lumped species is developed. Lumped parameters computation is based on knowledge extracted from oxidation simulations adopting detailed models.

In the third section, a database of lumped models is generated based on the procedures developed in the previous sections. Features are extracted from the database and compiled into datasets for use in the following steps.

In the last section, a machine learning-based methodology for the generation of low-temperature oxidation lumped models for real fuels is presented. The correlation between the database parameters and the fuel functional group distribution is modeled via a machine learning algorithm. Machine learning predictions are then adopted to reconstruct kinetic models for real fuels.