On-line flow visualization in multiphase reactors using neural networks

The success of the radioactive particle tracking system (RPT) developed at the École Polytechnique (Montréal) and pplied to the study of particle motion in a variety of chemical reactors (three-phase fluidized bed, gas spouted bed and liquid fluidized bed) has motivated us to continue improving this technique (in terms of accuracy and resolution) and to apply it to new reactor types. Our goals are: (i) to enhance the original search location algorithm in order to permit on-line flow visualization and (ii) to extend RPT to very fast solids flows, such as those encountered in circulating fluidized beds (particle velocities higher than a few m s−1). The potential of neural networks for on-line and real-time visualization of particle movements in multiphase reactors will be illustrated. The original least-squares search location algorithm (Larachi et al., 1994) has been replaced with an enhanced algorithm which uses a three-layer feedforward neural network. The results obtained from the two algorithms for particle tracking in a three-phase fluidized bed reactor are compared. The RPT system employs 8 NaI(TI) scintillation detectors to study the movement of solid particles in chemical reactors. The performance of the system was investigated using particles containing the radioisotopes 46Sc (γ-ray energy 1005 keV), 99Mo (γ-ray energy 140 keV) and 198Au (γ-ray energy 412 keV). The three-dimensional spatial resolution was measured in empty and water-filled tubes, simulating highly diluted and dense media. The best results were obtained using 198Au with which the particle can be located to within 7 mm in water and 9 mm in air 100 times s−1.