Revisiting surface tearing extrusion flow instabilities through machine learning

Artikel i vetenskaplig tidskrift, 2026

Polymer extrusion instabilities such as sharkskin, stick–slip, melt fracture, and surface tearing in filled composites are major manufacturing defects. They decrease surface quality, reduce mechanical performance, and increase scrap rates, making their detection essential for product quality and process efficiency. Here, we used machine learning algorithms to analyze extrudate surface patterns in wood–fiber polypropylene composites captured through an inline optical visualization system. Two compositions with distinct surface characteristics were processed in a single-screw extruder under controlled screw-speed ramps to capture several extrusion regimes. Supervised and unsupervised learning approaches classified surface instabilities, producing cluster maps that matched conventional Fourier-transform analysis while providing spatially resolved insight into surface transitions. Machine learning also detected subtle local features such as minute changes in surface tearing and bubble containing patterns, even in the absence of distinct frequency or wavenumber shifts. Spatially adaptive image classification can reveal flow variations that conventional averaged spectral methods cannot, enabling improved analysis and automation in polymer processing.