Abstract:The particle residence time in counter-current spray drying towers has a significant influence on the moisture content of the powder exiting the tower. Therefore, the reliability of predictions of residence time by numerical methods is highly desirable. A combined experimental and computational fluid dynamics investigation is reported for the prediction of the residence time distributions of glass beads with a narrow size range of 300-425 m in a counter-current tower with isothermal swirling flows of air. The particle-wall collision is taken into account using a rough-wall collision model. Overall, a reasonably good agreement is obtained between the measurements and predictions. Consideration of wall roughness results in greater axial dispersion of particles in the tower compared to a smooth wall assumption. The rough particle-wall collision is important for a reliable prediction of residence time distributions. In addition, analysis of the results infers that the clustering effect of particles on drag and particle-particle interactions are important and should be investigated in a future study.

This work studies the dispersion of solids in the cold isothermal operation of swirl counter current spray dryers. Residence time distributions (RTDs) of glass beads and detergent powder are obtained in a semi-industrial unit under varying Reynolds and injection positions and validated with the results of a novel Reynolds average Navier Stokes-discrete parcel method (DPM) framework. The simulations stress that the particle RTD is governed to a large extent by the interaction of the solids with the walls, which is usually simplified with the assumption of a particle-wall restitution coefficient. Since this is often unavailable experimentally, here, we propose an alternative combined model that integrates the computational fluid dynamics (CFD)-DPM model with reinforcement learning using a training set of experimental RTDs to extract an "effective pair of wall restitution coefficients". The method improves the accuracy of existing CFD platforms, reducing the errors in the mean residence time from 30−100% to <25%.

No abstract