Hybrid CFD-ML Method Integrated with Optimization Techniques to Predict the Hydrodynamics of Gas-Solid Fluidized Bed

Abstract

This study addresses the challenging task of predicting the hydrodynamics of fluidized bed reactors, with a particular focus on bubbling beds involving fine Geldart A particles having limited successful simulations in existing literature. Our investigation explores into the influence of various parameters, such as height, width, velocity, particle diameter, initial height, solid fraction, particle density, drag scaling factor, restitution coefficient, specularity coefficient, and mesh size, on both solid volume fraction and a modified drag accounting for interparticle forces neglected in conventional drag models. Throughout the simulations, data was systematically generated and subsequently utilized to train four machine learning models: Decision Trees (DT), Extreme Learning Trees (ELT), Gaussian Process Regression (GPR), and Support Vector Machines (SVM). Optimization techniques, including Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), were employed to enhance model performance. Notably, the GPR-GA model exhibited superior results, achieving an R2 of 0.99 and 0.97 for training and testing, respectively. To facilitate user-friendly application and practical implementation, a Graphical User Interface (GUI) was developed. This GUI empowers users to predict the hydrodynamics of gas-solid fluidized bed reactors by inputting specific parameters and leveraging the optimized machine learning models. This comprehensive approach contributes valuable insights and tools for understanding and predicting the intricate dynamics of fluidized bed reactors.