A Machine Learning Approach to Model Oxidation of Toluene in a Bubble Column Reactor

Abstract

A machine-learned (ML) subgrid-scale (SGS) modeling technique is introduced for efficient and accurate prediction of reactants and products undergoing parallel competitive reactions as seen in a bubble column. The model relies on data generated from a simple substitute problem with a small number of features. The machine-learned model corrects the errors in concentration and concentration gradients at cell faces arising from using linear interpolation and showed good accuracy for a mesh that barely covers the concentration boundary layer with minimal computational overhead. The present model, thus, offers a significant performance bonus when applied to near spherical, ellipsoid, and dimple-ellipsoidal bubbles.