Introducing an artificial neural network energy minimization multi-scale drag scheme for fluidized particles

Particles under fluidization conditions tend to clog and aggregate, and form meso-scale structures that significantly affect gas-solid transport phenomena. In the last decade, resolution of multi-scale particle structures has been attained by using advanced sub-grid models, such as the Energy Minimization Multi-Scale (EMMS) scheme. The current work aims to develop an ANN (Artificial Neural Network) to better resolve the effect of such structures. The ANN is developed, trained and validated using data gen-erated by a custom-built FORTRAN code that solves the EMMS equations for a wide variety of gas-particle mixture properties (1 < dp* < 10). The model is tested in the simulation of a pilot-scale CFB carbonator. The difference in the predictions of the CFD model incorporating the ANN-EMMS drag scheme compared to the conventional EMMS drag scheme is 11.29% in terms of pressure drop (dP/dz) in average and less than 1% in terms of CO2 concentration at the exit of the reactor. (c) 2020 Elsevier Ltd. All rights reserved.

Automated summary

Particles tend to clog and aggregate under fluidization conditions, forming meso-scale structures that affect gas-solid transport phenomena. The use of an ANN-EMMS drag scheme in CFD modeling shows improved accuracy in predicting pressure drop and CO2 concentration compared to conventional methods.