Predicting Properties from Near-Infrared Spectra with Machine Learning for Improved Polyolefin Differentiation

Area of Science:

• Polymer Science
• Spectroscopy
• Machine Learning

Background:

• Growing plastic production necessitates advanced recycling solutions.
• Current near-infrared (NIR) spectroscopy struggles to differentiate polyolefin subclasses due to spectral similarity.
• Effective differentiation of polyolefins like low-density polyethylene (LDPE) and high-density polyethylene (HDPE) is crucial for recycling.

Purpose of the Study:

• To develop a machine learning (ML) approach for predicting polyolefin properties directly from NIR spectra.
• To enable property-based sorting for enhanced plastic recycling efficiency.
• To link ML predictions to underlying polyolefin chemistry for improved understanding.

Main Methods:

• Utilized machine learning (ML) models to predict density, crystallinity, and short-chain branching from NIR spectra.
• Evaluated various ML models, identifying partial least squares regression for its accuracy and simplicity.
• Developed a method to identify key wavenumbers for property prediction, enhancing model interpretability.

Main Results:

• Partial least squares regression demonstrated high accuracy in predicting polyolefin properties from NIR spectra.
• Identified specific wavenumbers correlated with CH3 NIR vibrational absorption bands, linking spectral data to chemical structure.
• Confirmed that ML models effectively capture spectrum-structure-property relationships in polyolefins.

Conclusions:

• Machine learning combined with NIR spectroscopy offers a powerful tool for polyolefin differentiation.
• The developed method enhances understanding of polyolefin chemistry through spectral analysis.
• Findings support advancements in property-based sorting for more efficient plastic recycling.
• Meta_Description