Application of machine learning to reconstruct flamelet tables based on an unsteady flamelet progress variable approach

This study explores the use of machine learning (ML) techniques, specifically artificial neural networks (ANN) and gradient boosting(GB), to replace conventional flamelet look-up tables in turbulent combustion modeling. The approach is based on the unsteadyflamelet/progress variable (UFPV) formulation, which serves as the foundation for the underlying flamelet manifolds. A central focusis placed on the transformation of high-dimensional tabulated chemistry data into efficient ML surrogates suitable for integrationinto computational fluid dynamics (CFD) simulations. Accordingly, various data pre-processing strategies and training methodologiesare evaluated, and the predictive performance of the resulting ML models is thoroughly assessed. To demonstrate the generalityand robustness of the proposed framework, the methodology is applied to multiple fuels, including n-dodecane (C12H26), n-heptane(C7H16), and oxymethylene ether (OME34). Beyond predictive accuracy, the study emphasizes key practical advantages of theML-based approach, namely reduced memory footprint and lower computational overhead by eliminating runtime interpolation,quantifying its potential to enhance the efficiency of combustion simulations.