Vision based feedback control of weatherstrip coextrusion with predictive dimension modeling

Achieving geometric accuracy in weatherstrip co-extrusion is inherently challenging due to the complex nonlinear interactions among process variables and the significant time delay caused by the long vulcanization line. However, conventional approaches primarily focus on managing process parameters or fail to integrate dimensional inspection into closed-loop feedback systems. Therefore, an integrated control framework is required to regulate product dimensions while simultaneously accounting for complex process variables. This study proposes a data-driven control strategy comprising two models to manage both process initialization and steady-state production. For the startup phase, a Random Forest-based prediction model was developed to determine optimal initial screw speeds, effectively minimizing trial-and-error adjustments. For steady-state operation, a smart vision-based feedback framework utilizing an artificial neural network was implemented to proactively compensate for dimensional deviations despite transport delays. Field validation in a mass-production environment confirmed that the system successfully maintained dimensional errors within a strict tolerance range of - 0.1 to + 0.2 mm. These results demonstrate the practical feasibility of this integrated approach for autonomous quality management in industrial weatherstrip co-extrusion.