Machine learning assisted models for convective boundary heat transfer in turbulent fire simulations along wall/ceiling/floor

Accurate modeling of convective boundary heat flux is vital for the predictive modeling of turbulent fire along a solid surface like walls, floors and ceilings. High grid-resolution near the surface is required to produce accurate modeling of the heat flux on the surface. This requirement is not feasible in engineering modeling studies of fires along a surface. Wall models are thus needed to reconstruct boundary heat flux so that the grid requirement is not so restrictive. In this work, we examine the feasibility and potential of using machine learning to reconstruct the convective boundary heat flux in fire modeling. High-fidelity large-eddy simulations of several turbulent wall fire propagating along a vertical wall, a ceiling, and a floor are conducted to produce the training data for machine learning. A temperature gradient correction factor is introduced to compensate for the loss of accuracy of temperature gradient when discretized on a coarse grid. The machine learning model is used to train the model for the correction factor. The performance of the trained model is assessed for the different fire cases. Good performance is observed.