A Nearly Zero-CostLot-by-Lot Inspection of RecycledPlastics: Prediction of Mechanical Properties from Viscosity Evolutionduring Melt Kneading

Driving mechanical recycling with minimal energy consumption has become increasingly urgent. However, recycled plastics derived from household plastic waste, which accounts for approximately half of all plastic waste, are contaminated with non-plastic substances and mixed polymers. These contamination levels vary significantly from lot to lot, limiting their use to low-grade applications, where consistent quality is less critical. This study highlights that all recycled plastics undergo melting, kneading, and pelletizing processes. By predicting the mechanical properties of recycled products based on melt viscosity (auxiliary data obtained during kneading without additional costs), we propose a nearly zero-cost, lot-by-lot inspection method. Pre-production prediction of pellet properties during kneading enables the classification and extraction of high-quality, uniform recycled plastics tailored to specific applications. To validate this approach, we predict the tensile properties and Charpy impact energies of 23 lots of household polypropylene (PP) waste. Leveraging a bidirectional recurrent neural network, we developed a system to classify pellets prior to production based on predicted mechanical properties, achieving over 85% accuracy. This innovative analytical method provides a cost-effective solution for upcycling household waste, contributing to sustainability within the circular economy.