We address this challenge by using a combination of imaging sensors to identify key chemical compounds in material streams. By means of hyperspectral (HSI) reflectance spectroscopy in the near- and mid-infrared range certain alloys, ceramics, and plastics can be identified and classified. Laser-induced Fluorescence (LiF) spectroscopy enables the detection of rare earth elements (REEs) and low-reflective black plastics among others. In order to increase the range of waste classes characterized by our system, we propose to add a rapid, non-destructive and cost-efficient Raman sensor within the project RAMSES-4-CE. This module will be integrated into an existing sensor system, comprising LiF and HSI, developed during the EIT inSPECtor project.
For industrial applications, the requirements for a sensor-based sorting system implies high measurement speed (up to 1 m/s) for inline high throughput processing, as well a high spatial resolution (about 2 mm) for the identification of shredded and non-shredded recycling materials. Thus, the data generated by the individual sensors must be processed, integrated and analysed immediately. For this purpose, fast data processing tools based on machine learning will be developed. Together, smart real time integration of imaging and spectroscopic sensors using machine learning methods will allow for a robust analysis of complex, temporally and spatially variable industrial waste streams on a conveyor belt.