Prediction of particle mixing time in a rotary drum by 2D DEM simulations and cross-correlation

Abstract

A new concept for the prediction of the mixing time of a large particle system from the mixing time of a small particle system and a scale-spanning cross-correlation are presented. By way of example, the considered large system is a 3D rotary drum, and the used small system is a 2D rotary drum. At both levels, data for the change in mixing degree with time are created by DEM simulation. The cross-correlation is developed from a learning set which consists of 17 sets of simulation data and represents a variation of revolution frequency. The prediction of mixing time through the cross-correlation is excellent within the parametric range of the learning set (R2 = 0.92). Beyond the parametric range of the learning set, good predictions are obtained for binary mixtures of particles with different density (R2 = 0.86), but not in case of different size. Whether the additional parameter will affect the axial mixing in the rotary drum is considered to be the key for such parameter expansion. Advantages of interpretability, expandability and relatively high accuracy for the limited data size compared to machine learning approaches are seen in the cross-correlation method, which may promote its application for the fast and cheap prediction of industrial mixers.