A data-driven method to optimize soot kinetics based on uncertainty quantification and the active subspace approach

This paper introduces a novel approach integrating uncertainty quantification (UQ) and data-driven techniques that aim to optimize soot particle size distributions (PSDs) using an existing soot kinetic model. Leveraging the active subspace (AS) method, the influential parameters governing the overall soot production and several representative PSDs are identified. Gradient descent techniques are employed to optimize the kinetic parameters simultaneously with reference to experimental measurements of burner stabilized stagnation (BSS) flames. The optimization process is rigorously validated against experimental data and the response surface predictions, demonstrating robustness and generalization capabilities across different cases. It is found that while the soot volume fraction was adequately predicted, the iterative UQ-assisted gradient descent technique can improve the prediction of PSDs but fails to fully reproduce the experimentally observed bimodality. This confirms the need for future improvements in the sectional kinetics model. In this regard, the analysis performed points at the need of distinguishing the coagulation kinetics of liquid-like and solid primary particles. With such future improvements, whose implementation is guided by the combined UQ and data-driven approach, soot modeling may advance into a data-driven era, minimizing reliance on expert knowledge alone.