Supervised machine learning approach to estimating the convective heat transfer coefficient for heated oblate spheroidal particles

An accurate prediction of the convective heat transfer coefficient for heated non-spherical particles is important in many engineering applications. In this work, a supervised machine learning (ML)-based model is proposed to predict the Nusselt number (Nu) of heated oblate spheroidal particles in laminar flow. A computational fluid dynamics (CFD) dataset of Nusselt numbers is obtained from COMSOL Multiphysics simulations performed at different Reynolds numbers, aspect ratios, and surface temperatures. Three supervised ML models are trained and compared — gradient boosting regressor (GBR), random forest (RF), and ridge regression — and cross-validation and model evaluation are performed using standard performance metrics (R² and RMSE). Among the three models, the GBR model achieves the best predictive performance by capturing the nonlinear joint associations of geometric and thermal factors. Hyperparameter optimization is implemented to improve model performance. An accurate predictive model is suggested as an alternative to classical empirical correlations to estimate Nu for oblate spheroids, establishing a theoretical and data-driven foundation relevant to industrial processes like gas-solid reactions, aerosols, and fluidized beds.