This doctoral thesis aims to validate new methods that increase the efficiency of sensor-based sorting processes for waste streams consisting of plastics. It deals with set boundaries on aggregate level; the plant level is not considered. The used equipment is the experimental sensor-based sorting setup at the Chair of Waste Processing Technology and Waste Management at Montanuniversität Leoben and the used sensor technology near-infrared spectroscopy.
Increasing the sorting efficiency can be done by optimizing the identification of the mechanical discharge of particles. Data analytics is shown as a solution to achieve optimization, therefore this thesis focuses on using data-analytics-related methods.
For optimizing the identification of particles, research is conducted in the fields:
•Influence of surface roughness
•Influence of reflectors as background material
•Usage of machine learning approaches
For optimizing the mechanical discharge of particles, research is conducted in the fields:
•Correlations between the input parameters (input composition, throughput rate) and the output parameters (purity, recovery, yield, incorrect discharged particles) of a sensor-based sorting process
•Mathematical approaches to describe the optimal operation point of a sensor-based sorting machine to achieve a specific sorting result
It is stated that this outcome allows a sorting plant to increase purity by using machine learning approaches to optimize the identification or running the plant on the optimal operation point, both without substantially adapting the plant. Superordinate considered these solutions help to increase the amount of recycled plastic so that less plastic waste is thermally treated.