Real‐Time Diagnosis of Polymer Compositions During Extrusion via Pressure Signals and Machine Learning

ABSTRACT One of the most critical challenges in mechanical plastic recycling is the presence of various materials mixed within the recycled polymer. Even after selective collection, sorting, and cleaning, polymers such as PP, PVC, PS, and PE inevitably remain mixed, causing incompatibility, immiscibility, and inhomogeneity. These issues lead to decreased and inconsistent properties in recycled polymer blends, limiting their use in high‐value applications. This study proposes a method for inline, real‐time diagnosis of polymer composition during extrusion processes, which are central to large‐scale mechanical recycling. The method trains a machine learning model to predict polymer compositions from pressure signals measured during extrusion. The underlying assumption is that extrusion pressure signals reflect both processing conditions and rheological properties of the polymer. Therefore, a change in composition uniquely determines the pressure signal under constant processing conditions. The proposed approach predicts polymer compositions, varying in 10 wt% increments, with 96% accuracy using only 10 s of pressure data. This enables real‐time adjustment of processing parameters or product classification in response to continuously varying compositions. Moreover, the method requires only a single pressure sensor, offering a cost‐effective and scalable solution for real‐time quality control in polymer recycling.