Genetic Programming based Drag Model with Improved Prediction Accuracy for Fluidization Systems

Abstract

The drag coefficient plays a vital role in the modeling of gas-solid flows. Its knowledge is essential for understanding the momentum exchange between the gas and solid phases of a fluidization system, and correctly predicting the related hydrodynamics. There exists a number of models for predicting the magnitude of the drag coefficient. However, their major limitation is that they predict widely differing drag coefficient values over same parameter ranges. The parameter ranges over which models possess a good drag prediction accuracy are also not specified explicitly. Accordingly, the present investigation employs Geldart’s group B particles fluidization data from various studies covering wide ranges of Re and εs to propose a new unified drag coefficient model. A novel artificial intelligence based formalism namely genetic programming (GP) has been used to obtain this model. It is developed using the pressure drop approach, and its performance has been assessed rigorously for predicting the bed height, pressure drop, and solid volume fraction at different magnitudes of Reynolds number, by simulating a 3D bubbling fluidized bed. The new drag model has been found to possess better prediction accuracy and applicability over a much wider range of Re and εs than a number of existing models. Owing to the superior performance of the new drag model, it has a potential to gainfully replace the existing drag models in predicting the hydrodynamic behavior of fluidized beds.

Artificial Intelligence in Physical Sciences: Symbolic Regression Trends and Perspectives

“…Drag coefficient has a crucial role in gas-solid flows, as it provides an analytical view of the hydrodynamics therein [ 166 ]. At the industrial scale, Computational Fluid Dynamics (CFD) simulations are being employed towards this investigation [ 167 ].…”

“…Moreover, CFD simulations depend on the Euler–Euler or Euler–Lagrange models [ 167 ], which both of them need to possess efficient drag correlation models in order to take into consideration gas-particle interactions [ 168 ]. Current employment of SR techniques, are found in studies about the investigation of fundamental principles in the drag coefficient [ 169 ], construction of brand new drag correlations which can be used as input to CFD models [ 167 ] or a simple drag model [ 166 ] which have proven to outperform standard formulations.…”

Robust Data-Driven Soft Sensors for Online Monitoring of Volatile Fatty Acids in Anaerobic Digestion Processes

“…The output of GA is a value, whereas the output of GP is a computer program. The complicated structures of computer programs, mathematical expressions, and process system models in GP are represented with trees [35,36].…”