Chemical kinetic and turbulent combustion model uncertainties can significantly affect the reliability of combustion simulations. Quantifying these uncertainties is therefore essential but remains computationally prohibitive, as classical uncertainty quantification (UQ) methods struggle to handle the high dimensionality and complexity inherent in turbulent combustion models. In this work, we develop a comprehensive framework to perform a two-stage UQ analysis based on the active subspace (AS) method, enabled by fast and accurate gradient evaluations using Fourier neural operator (FNO) surrogates. The framework is demonstrated using Reynolds-averaged Navier–Stokes (RANS) simulations of the Sandia/ETH CO/H 2 /N 2 jet flames. A small number of dominant subspaces in the combined space of turbulent combustion model parameters and a small number of low-dimensional kinetic active directions extracted from the detailed reaction mechanism are identified, capturing the majority of the model uncertainty and demonstrating that the underlying uncertainty manifold is intrinsically low-dimensional. Furthermore, the contributions of elementary reactions within the chemical kinetic uncertainty space are analyzed as functions of the scalar dissipation rate ( χ ), revealing clear connections to radical transformation processes and reaction pathways strongly associated with heat release. By incorporating a broader range of model uncertainties together with advanced neural operator surrogates, this work substantially extends the scope of existing combustion UQ methodologies. The proposed framework is general and readily transferable to other turbulent combustion configurations. • A physics-informed intelligent UQ framework for turbulent combustion modeling. • Identification of a low-dimensional manifold capturing model uncertainties. • A resolved interpretation of chemistry uncertainty across the flamelet state space. • Potential application to a wide range of turbulent combustion simulations.
Zhang et al. (Sun,) studied this question.