To contribute to the targeted 10 million tons per year of recycled plastic in Europe by 2025 and to improve the mechanical sorting degree of polyethylene (PE) products, density prediction models were developed from Fourier transform infrared-attenuated total reflectance (FTIR-ATR) and Raman spectroscopic data. State-of-the-art sorting in mechanical recycling provides separated polymer classes, however an improved classification with specific chemical and physical features such as density or melt flow rate has not been developed yet.
Applying multivariate data analysis (MVDA) on the spectral datasets of 10 different PE materials, one FTIR-ATR and two Raman spectra based partial least square (PLS) density models were developed. However, whereas all three models are applicable to predict PE density accurately, the Raman models have shown some advantages. Firstly, less principle components (PC) are needed and secondly the density can be assessed with higher accuracy, due to the more robust cross-validated PLS model. Moreover, the obtained PC-s indicate that in the FTIR-ATR model the CH3/CH2 ratio, while in the Raman model the CH2, CH and the crystalline C–C bands can be correlated with the PE density. The most accurate PLS model was obtained from the 1500-1000 cm−1 Raman shift region. The developed models could improve the density based mechanical separation of PE and consequently increase the quality of recycled and reprocessed PE products.