An artificial neural network as a model for chaotic behavior of a three-phase fluidized bed

Non-linear hydrodynamic behavior of bubble motion and that of particle motion in a three-phase fluidized bed have been modeled by resorting to an artificial neural network (ANN). The experiments were performed in a transparent acrylic resin column with an inner diameter of 0.184 m and a height of 2.0 m. Subsequently, the ANN was trained with the time-series data comprising temporal intervals, each of which was the period between two sequential signals of bubbles or particles from an optical transmittance probe. By successively adapting its output to input, the ANN has regenerated time-series data at any superficial gas velocity, Ug, thereby yielding the bifurcation diagrams of both bubble and particle motion. These diagrams exhibit complex behavior over a wide range of Ug, thus demonstrating that the ANN is capable of predicting and modeling non-linear dynamics of three-phase fluidized beds often behaving chaotically.