A convective heat‐transfer model for non‐spherical particles based on SuperDEM – CFD from direct numerical simulation

Abstract Predicting convective heat transfer for non‐spherical particles is challenging due to the coupled effects of shape and orientation. This study performs particle‐resolved direct numerical simulations for ellipsoids, cylinders, and cuboids across a wide range of aspect ratios (ARs) and attack angles. Regional thermal analyses identify crosswise sphericity as the key geometric factor governing stagnation‐zone transport, while overall sphericity accounts for the available surface area. To unify these coupled effects, a composite descriptor combining both sphericities is proposed, which successfully collapses data from all investigated geometries. Leveraging this descriptor, a mechanism‐based correlation and four machine‐learning models were developed. For unseen conditions, the mechanism‐based correlation achieved mean absolute percentage errors below 5%, whereas the shape‐specific correlations reduced the errors to below 3%. Machine‐learning models provided higher in‐domain accuracy but showed limited extrapolation capability, with back propagation neural network performing best for Reynolds‐number extrapolation but deteriorating at AR = 6.