Raman spectroscopy integrated with machine learning techniques to improve industrial sorting of Waste Electric and Electronic Equipment (WEEE) plastics.

Current industrial separation and sorting technologies struggle to efficiently identify and classify a large part of Waste of Electric and Electronic Equipment (WEEE) plastics due to their high content of certain additives. In this study, Raman spectroscopy in combination with machine learning methods was assessed to develop classification models that could improve the identification and separation of Polystyrene (PS), Acrylonitrile Butadiene Styrene (ABS), Polycarbonate (PC) and the blend PC/ABS contained in WEEE streams, including black plastics, to increase their recycling rate, and to enhance plastics circularity. Raman spectral analysis was carried out with two lasers of different excitation wavelengths (785 nm and 1064 nm) and varying setting parameters (laser power, integration time, focus distance) with the aim at reducing the fluorescence. Raman spectral data were used to train and test Discriminant Analysis (DA) and Support Vector Machine (SVM) algorithms in an iterative procedure to assess their performance in identifying and classifying real WEEE plastics. Analysis settings were optimized considering industry requirements, such as process productivity (classification rate, short measuring time for fast identification) and product quality (purity of the sorted polymers). Classification models were trained, in a first approach, only on the target WEEE plastics; and in a second approach, on all polymers expected in the WEEE stream, leading to a realistic overview of the potential scalability of the advanced sorting methods and their limitations. The best classification models, based on DA of Raman spectral data obtained with the 1064 nm laser at 500 mW and 1.0 s, led to classify PS and ABS with a purity up to 80 %.