Simulation on hydrodynamics of non-spherical particulate system using a drag coefficient correlation based on artificial neural network

Abstract

Fluidization of non-spherical particles is very common in petroleum engineering. Understanding the complex phenomenon of non-spherical particle flow is of great significance. In this paper, coupled with two-fluid model, the drag coefficient correlation based on artificial neural network was applied in the simulations of a bubbling fluidized bed filled with non-spherical particles. The simulation results were compared with the experimental data from the literature. Good agreement between the experimental data and the simulation results reveals that the modified drag model can accurately capture the interaction between the gas phase and solid phase. Then, several cases of different particles, including tetrahedron, cube, and sphere, together with the nylon beads used in the model validation, were employed in the simulations to study the effect of particle shape on the flow behaviors in the bubbling fluidized bed. Particle shape affects the hydrodynamics of non-spherical particles mainly on microscale. This work can be a basis and reference for the utilization of artificial neural network in the investigation of drag coefficient correlation in the dense gas-solid two-phase flow. Moreover, the proposed drag coefficient correlation provides one more option when investigating the hydrodynamics of non-spherical particles in the gas-solid fluidized bed.

Hydrodynamics of Irregular-Shaped Graphite Particles in Coaxial Two-Phase Jet Flow

“…We can see that the profiles of higher sphericity particle exhibit flat and uniform since their interlock forces are less than those of lower sphericity particles. 38 , 39 Lower sphericity ultralight particles are sensitive to followability by the gas phase rather than controlled by their own inertia leading to the complex transport characteristics.…”

“…The most important phenomenon observed is that the peak values of ψ = 0.63 and 0.72 are approximately 3.0 times larger than those sphericities located at near central regions (see Figure a). It is demonstrated that the drag force for the higher irregular-shaped degree is far greater than the interlock force between particles and their own centrifugal force. − …”

“… 36 Regarding nonspherical particles, a new drag force coefficient, which is a function of the particle Reynolds number and particle sphericity using the artificial neural model for nonspherical particle in a dense gas–particle two-phase flow, was proposed. 37 , 38 Meanwhile, effects of particle shapes on the hydrodynamics on the microscale level are revealed. 39 − 41 Until now, the effects of nonspherical particle shapes on particle dispersions in swirling two-phase flows have not been reported, especially for ultralight density particles.…”

“…Most drag force correlations are always dependent on the spherical particle with homogeneous and heterogeneous empirical attributions that were derived from the existing drag equations . Regarding nonspherical particles, a new drag force coefficient, which is a function of the particle Reynolds number and particle sphericity using the artificial neural model for nonspherical particle in a dense gas–particle two-phase flow, was proposed. , Meanwhile, effects of particle shapes on the hydrodynamics on the microscale level are revealed. − Until now, the effects of nonspherical particle shapes on particle dispersions in swirling two-phase flows have not been reported, especially for ultralight density particles. The aim of the present study is to explore the effects of irregular shape on the hydrodynamics of ultralight expanded graphite turbulent flows.…”

Characteristics of Gas–Solid Mixture Flows through a Packed Moving Bed of Coarse Particles

“…Applying more complicated methods, such as quadratic expressions or artificial neural networks (ANNs), can achieve a better prediction by considering nonlinear behavior . The ANN is one machine learning method that can effectively predict systems without needing to use casual models and has been applied to various applications. − Whereas the parameters from the ANN do not have physical meaning, the quadratic expressions make it possible to interpret the interactions between independent variables. A least absolute shrinkage and selection operator (LASSO) method, another machine learning-based technique, can be applied to the quadratic equations to achieve high prediction accuracy and interpretability by excluding minor coefficients. − …”