Quantification of aluminium trihydrate flame retardant in polyolefins via in-line hyperspectral imaging and machine learning for safe sorting

The extensive use of aluminium trihydrate (ATH) flame retardant in plastics poses challenges and hazards in plastic waste recycling, thus it is crucial to accurately identify ATH. This study demonstrates the effectiveness of an industrial in-line shortwave infrared (SWIR) hyperspectral imaging system and principal component analysis (PCA) for detecting and quantifying ATH in low-density polyethylene (LDPE) and polypropylene (PP). The samples were characterized by elemental analysis, ATR-FTIR, DSC, and TGA. A quantitative estimation model was developed by analysing spectra with varying ATH concentrations. PCA and SWIR band area ratio were fitted to estimate the ATH concentration. The PCA model outperformed the SWIR band area ratio model and achieved good predictions between measured and predicted ATH concentrations ranging from 22.9 to 1.6 wt% (R²_LDPE = 0.95) in LDPE and from 24.0 to 2.5 wt% in PP (R²_PP = 0.94). The obtained in-line control tool is relevant to the recycling industries, enabling real-time assessment of additives.