Quantitative identification of melamine flame retardant in polyolefins with hyperspectral imaging and machine...

Non-halogenated flame-retardant Melamine (MEL), an alternative to toxic halogenated flame retardants, has been recognized as a substance of very high concern by the ECHA. Therefore, identifying MEL in plastic waste is essential for ensuring safe handling and recycling. This study presents industrial in-line quantitative identification techniques for MEL in low-density polyethylene (LDPE) and polypropylene (PP) via short-waved infrared (SWIR) hyperspectral imaging combined with machine learning. LDPE and PP samples with varying MEL loadings were compounded and characterized through elemental analysis, ATR-FTIR, TGA, and DSC. Regression on the SWIR band area ratio and principal component one was applied to the SWIR spectra to compile predictive models. The models performed equally well, demonstrating a strong correlation between measured and predicted MEL concentrations. The model based on SWIR band area ratio ranged from 0.8 to 29.8 wt% MEL in LDPE (R²_LDPE = 0.975) and 1.4 to 26.3 wt% MEL in PP (R²_PP = 0.995), while the model based on principal component one ranged from 0.8 to 29.8 wt% MEL in LDPE (R²_LDPE = 0.978) and 0.9 to 26.3 wt% MEL in PP (R²_PP = 0.988). Further, the hyperspectral in-line models perform comparably to FTIR band ratio ranging from 1.11 to 29.78 wt% MEL in LDPE and 1.40 to 26.26 wt% MEL in PP. The proposed methods can facilitate real-time monitoring of MEL concentrations in plastic and enable detection below hazardous threshold limits, which is crucial in the recycling industries.