Investigation of particle descriptors for size characterisation of solid waste particles for treatment process control

Abstract

Real-time control of the particle size distribution of coarsely shredded mixed solid waste has a large potential for improving the performance of mechanical processing plants. In addition to controllers and actuators, online metrology for particle size distributions is needed. In this work, 2D-images of single waste particles from the material fractions wood and plastic are investigated. The materials were gained by handsorting of shredded mixed commercial waste and the individual particles were described through different descriptors, which were used in regression models for particle size determination. It is shown that univariate models are not very likely to perform well due to the overlapping of the descriptor values for different particle size classes. Though, using a Partial Least Squares regression, that considers many different descriptors, an accuracy of over 70 % was reached in most of the considered particle size classes for detecting the correct particle size in the investigated material fractions. Therefore, the potential of the method was proven, while further research is needed, to reach an accuracy level that is suitable for process control. Additionally, the evaluated particle size class must be combined with the particle weight to determine the effects of assigned particles in a particle size distribution.