Systematic assessment of data-driven approaches for wall heat transfer modelling for LES in IC engines using DNS data

Data-driven (DD) methods offer a promising pathway towards novel modelling solutions in fluid flow and heat transfer. In this study, we investigate the application of DD neural network (NN) on wall transfer context wall-modelled large-eddy simulation (WMLES) engines, focusing systematic evaluation criteria for successful model generation. High-fidelity input data training testing is generated by spatial filtering DNS wall-resolved LES fields several engine engine-like configurations. The NN-based models are constructed using different wall-adjacent cell schemes, while size complexity also varied. evaluated demonstrate improved performance with respect to classical functions, indicating potential engineering applications. particular, better results were obtained inclusions wall-normal Reynolds number from second cell. Such two-cell format appears good compromise between complexity. Both present NN literature reference approaches generally perform unburned regions than burned ones. near-wall flame fronts, an analysis dividing samples into “unburned”, “burned”, “flame boundary” zones exposing characteristics varying degree difficulty.