Hybrid CFD-neural networks technique to predict circulating fluidized bed reactor riser hydrodynamics

Abstract

Circulating fluidized bed (CFB)-based co-pyrolysis is a promising technology for producing synthetic fuel and chemical co-products from biomass and waste feedstocks. Computational fluid dynamics (CFD) tools provide valuable insights for understanding gas-solid flow hydrodynamics, troubleshooting performance issues, and optimizing reactor operations. CFD simulations are computationally expensive and time-consuming; therefore, in this study, we developed an artificial neural network (ANN)-based machine learning model from a wide CFD simulation campaign. The CFB riser axial solid holdup profile was predicted using a two-fluid model and validated using the experimental data. Finally, a dataset connecting the input features of the model parameter-based simulations with their axial solid holdup was constructed for training the ANN. The performance of the developed model was later compared with the experimental data. The developed ANN model exhibited low mean square error values of order of 10(-3) for both validation datasets to predict axial solid holdup. The ANN model performed well with an interpolated solid circulation rate of 45 kg/m(2)s and 62 kg/m(2)s and an extrapolated solid circulation rate of 30 kg/m(2)s and 80 kg/m(2)s.