Computational Fluid Dynamics–Deep Neural Network (CFD-DNN) Surrogate Model with Graphical User Interface (GUI) for Predicting Hydrodynamic Parameters in Three-Phase Bubble Column Reactors

Bubble column reactors (BCRs) find extensive application in the chemical industry for facilitating three-phase reactions where most often solid-phase catalysts are employed alongside gaseous and liquid reactants. In these reactors, accurately identifying the prevailing hydrodynamic behavior is crucial for both design optimization and successful scale-up efforts. This research investigates the flow characteristics of solid–liquid–gas mixtures in cylindrical bubble columns through a comprehensive approach combining computational fluid dynamics (CFD) simulations and a CFD-based machine learning (ML) model. The three-phase Eulerian–Eulerian CFD simulations explored the effects of 5% and 20% solid loading with varying superficial gas velocities (0.1 and 0.2 m/s) on solid distribution. The obtained results align qualitatively with the experimental findings from the literature. A CFD–deep neural network (DNN) model is presented in a parallel study of a three-phase bubble column situation. By using a large data set obtained from CFD simulations, the deep neural network model is trained to accurately predict fluid dynamics closely resembling the CFD simulations. The ML architecture incorporates input data obtained from CFD results, thereby allowing the deep neural network to acquire knowledge and identify patterns in different fluid flow characteristics, such as velocity, pressure, and holdup. The tuning of hyperparameters is crucial for successful ML predictions (with R2 values 0.97 to 0.99) in the current CFD-DNN model. Additionally, a graphical user interface (GUI) on the basis of CFD-DNN has been developed for easy model handling. The CFD-DNN model showcases its capacity to forecast hydrodynamic parameters that were not used in the training process, hence diminishing computing expenses and simulation duration in comparison with conventional CFD techniques.