Coupled Deep Neural Network Model with CFD for Predicting the Heat Transfer Coefficient in Fluidized Beds

Abstract

Coupled transport processes are very complex in fluidized beds. Several prior CFD and experimental studies for the estimation of the heat transfer coefficient were conducted under certain fixed operating conditions and material properties in previous decades. Given that experimental measurements are often expensive, there is an urgent need to reduce the simulation time of the existing CFD models. This study applies the CFD-aided deep neural network (CFD-DNN) architecture to estimate the volumetric heat transfer coefficient using available correlations for the Nusselt number (Nu_p). In addition, a comprehensive model to estimate the local heat transfer coefficient of particles under different Reynolds numbers (Re_p), Prandtl numbers (Pr_f), and voidage conditions is also developed. The computing time decreased by 7% for the 2D case and 18% for the 3D scenario, while the coefficient of determination (COD) value reached 0.9995 when calculating the volumetric heat transfer coefficient. The results show that the DNN model not only has superior learning capability but also reduces the computational time with adequate accuracy. Also, the CFD-DNN-based Nu model coupled with CFD gave a reasonable prediction of the volumetric heat transfer coefficient between gas–solid phases, demonstrating good applicability to a wide range of fluidization conditions. This model is adaptable and reliable for industrial applications regardless of the particle shape. Expanding to a broader range of particle sizes and large-scale simulations further enhances its applicability, improving drying efficiency in multiphase heat and mass transfer, particularly in the pharmaceutical industry.