New Features for Image-Based Classification of Refuse- Derived Fuel Particles in a Rotary Kiln Environment

Abstract Camera-based methods can be used to detect particles from refuse-derived fuels (RDF) in a combustion chamber environment, such as a rotary kiln, and thus analyze their flight and combustion behavior. In this work, we use camera-based tracking results to build up a new automatic classification of different RDF particles with the help of a machine learning method, i.e., to recognize which RDF fraction a particle belongs to based on its trajectory. For this purpose, new features are introduced for the tracking data, which enable a uniform description of the combustion and velocity behavior over the flight duration for all RDF fractions considered. Based on a data set with RDF trajectories in a real rotary kiln environment, it is shown that reliable classification can be achieved based on the new features.