Assessing Melt Flow Rate in Post-Consumer Polypropylene via Near-Infrared Hyperspectral Imaging

TL;DR

This study shows that near-infrared imaging can predict melt flow rate in recycled polypropylene, helping improve recycling quality.

Contribution

The novel use of near-infrared hyperspectral imaging for predicting melt flow rate in post-consumer polypropylene is demonstrated.

Findings

Tree-based models achieved R2 = 0.85 for white samples and R2 = 0.61 for clear samples in predicting MFR.

Median spectral representations outperformed pixel-wise aggregation in model performance.

Balanced accuracies of 0.82–0.92 were achieved for binary classification of MFR thresholds.

Abstract

Mechanical recycling of polypropylene (PP) is constrained by the heterogeneous properties of post-consumer feedstocks. Melt flow rate (MFR) is a key property relevant to processing, and it varies widely across packaging grades, which limits the quality and substitutability of recyclates. This study evaluates near-infrared hyperspectral imaging (NIR-HSI) for predicting MFR in post-consumer PP packaging. Eighty-two rigid PP samples (46 white, 36 clear) with MFR values between 2 and 108 g 10 min−1 were collected from an Austrian material recovery facility. Thirteen different linear and non-linear regression models were examined using median and pixel-wise aggregated spectral representations across the samples. Tree-based models consistently achieved best performances with R2 = 0.85, RMSE = 12.4 g 10 min−1 on white samples and R2 = 0.61, RMSE = 14.0 g 10 min−1 on clear samples. On the combined sample set, R2 = 0.66 and RMSE = 17.3 g 10 min−1 were reached. Informative spectral regions correspond to typical bands of PP. Binary classification at different thresholds (6, 12, 30, 60 g 10 min−1) were also examined and achieved balanced accuracies of 0.82–0.92. Median spectral representations consistently outperformed pixel-wise aggregation. Results demonstrate that NIR-HSI can support grade-directed sorting of post-consumer PP, particularly for opaque white samples, though heteroscedasticity at high MFR values and irreducible outliers represent inherent limitations.

Figures