Modeling Char Particle Oxidation Rate in a Turbulent Mixing Layer with Machine Learning

Accurate modeling of the burning rate of char particles in particle-laden flows is essential. However, because of the strong inhomogeneity and nonlinearity of the process, accurately resolving the surface burning rate of char particles remains challenging. In this study, an eXtreme Gradient Boosting (XGBoost)-based framework is developed to reformulate the conventional char oxidation rate model, namely the Baum&Street (B&S) model, resulting in a modified model referred to as the XGB-B&S model. In this model, a correction term βturb is incorporated and formulated using the particle Reynolds number together with a dimensionless temperature. A turbulent mixing layer with char particle combustion is simulated by means of particle-resolved direct numerical simulation with three-dimensions, generating a high-fidelity dataset for model training and validation. To assess the predictive capability of the XGBoost model, its results are benchmarked against those obtained from an Artificial Neural Network model. The comparison indicates that XGBoost provides better overall accuracy, as reflected by a larger coefficient of determination (R2) and smaller values of both the root mean square error and the mean absolute error. Finally, the XGB-B&S model is validated against the test dataset. The R2 between the XGB-B&S predictions and the PR-DNS results is significantly higher than that between the conventional B&S model and the PR-DNS results, confirming the strong predictive capability of XGBoost for modeling char particle oxidation rate.