Ensuring the consistent quality of polymer recyclates is one of the key challenges for closing the loop in plastics. Traditional characterization methods often provide limited or qualitative information and fail with dark-colored, multilayer, or contaminated materials. To address this, NETZSCH has developed Proteus Now Quantify, a machine-learning-based software extension for differential scanning calorimetry (DSC).

Quantify uses the thermal fingerprint of polymers to identify and quantify compositions of complex recyclate blends with high precision. The foundation of this approach is a carefully curated dataset of defined polymer mixtures, which required significant effort in preparation, curation, and validation. Behind the software lies not only the AI model but also the systematic generation and cleaning of reference data to ensure robust predictions across a wide range of recycling streams.

This contribution will present the development process, including the data challenges, the iterative improvement of the machine learning models, and the feedback from alpha testers and industry partners.