Rapid prediction of the flow fields of fluidized beds with the varying flow regimes by coupling CFD and machine learning

Abstract

Integrating Computational Fluid Dynamics (CFD) with machine learning to predict the complex dynamics in fluidized beds faces significant challenges. This study employed Multiphase Particle-In-Cell (MP-PIC) simulations coupled with a hybrid machine learning algorithm based on Proper Orthogonal Decomposition (POD) and Support Vector Regression (SVR) to analyze dataset construction, feature extraction, and regression modeling under varying operational conditions. Results demonstrate that, compared to the fixed bed, the bubbling bed demands fourfold pressure and sixfold particle volume fraction data, while the turbulent bed exhibits sixfold and twofold, respectively. Feature extraction rates differ by bed type: the fixed and turbulent bed share the highest rate for pressure (95 %), exceeding the bubbling bed (90 %); while the fixed bed leads in particle volume fraction (95 %), with the bubbling and turbulent bed at 75 %. The machine-learning-enhanced framework could achieve 4–5 orders of magnitude acceleration compared to the conventional numerical method while retaining accuracy.