Accurate determination of polyethylene (PE) and polypropylene (PP) content in polyolefin blends using machine learning-assisted differential scanning calorimetry (DSC) analysis

Polyethylene (PE) and polypropylene (PP) are among the most recycled polymers. However, these polymers present similar physicochemical characteristics and cross-contamination between them is commonly observed, affecting the quality of the recyclates. With the increasing demand for recycled plastics, understanding the composition of these materials is crucial. Numerous techniques have been introduced in the literature to determine the composition of recycled plastics. An ideal technique should be accessible, cost-efficient, fast, and accurate. Differential Scanning Calorimetry (DSC) emerges as a suitable technique since it analyzes the thermal behavior of compounds under controlled time and temperature conditions, entitling the quantitative determination of each component, e.g., in PE/PP blends. Nevertheless, the existing predictive methods lack accuracy in estimating the composition of PE/PP blends from DSC analysis since the composition of this blend affects its overall crystallinity. This study advances the state-of-the-art regarding this quantification using DSC by implementing a non-linear calibration curve correlating the evolutions of crystallinity with blend composition. Additionally, a machine-learned (ML) model is introduced and validated, achieving high accuracy for the composition determination, presenting an overall mean absolute error as low as 1.0 wt%. Notably, this ML-assisted approach can also quantify the content of subcategory polymers, enhancing its utility.