Sensor-based characterization of anthropogenic material systems: developing characterization methods and novel applications for optimizing the mechanical recycling of lightweight packaging waste

Abstract

Mechanical recycling of post-consumer plastic packaging is characterized by a high lack of transparency due to the high effort of manual material flow characterization. As a consequence, optimizing collection processes in a targeted manner as well as adaptively designing and operating sorting and processing plants are often not possible, and confidence in secondary raw materials is often missing. This dissertation demonstrates how sensor technology in mechanical recycling can evolve from a sorting technology towards a key technology for enabling value-chain-wide transparency, and what optimization potentials can be derived from this enabled transparency. In the first part of the dissertation, a systematic literature review was conducted, which introduces a unified terminology, provides a comprehensive overview of the current state of research and identifies ten essential future research potentials. In the second part, novel characterization methods for extracting mass-based material flow compositions from area/volume-based sensor data were developed. Machine learning models were successfully trained to determine binary compositions of plastic flakes and post-consumer plastic packaging with a measurement uncertainty of 1.2 vol% and 2.4 wt%, respectively, across different material flow presentations and compositions using near-infrared (NIR) sensors. Based on the developed characterization methods, two novel sensor technology applications were demonstrated at industrial scale in the third part. First, a sensor-based quality monitoring of plastic pre-concentrates from sorting plants was developed using inline NIR sensors. The results showed that for a PET tray fraction as an example, sensor-based quality control of mass-based product purities is possible with a measurement uncertainity of 0.31 wt%. In comparison with state-of-the-art sampling-based quality analyses, it was shown that more than 350 kg of a 600 kg PET tray pre-concentrate bale would need to be analyzed manually to achieve a comparable measurement accuracy. Second, inline NIR sensors were used for sensor-based process monitoring of an industrial sensor-based sorting (SBS) unit. Using artificial neural networks, the process data was used to develop a process model that can predict the sorting behavior of the SBS unit across different sorting scenarios with a mean absolute error of 3.0%. In summary, the dissertation demonstrates the promising potentials of sensor technology in optimizing the circular economy. Based on the generated transparency, future process improvements can be implemented in individual process stages and across value chains, thus increasing the quantity and quality of recycled materials and the resulting ecological and economic benefits. Published by RWTH Aachen University, Aachen