Data‐Driven Models and Accurate Transfer Methods for Screw Extrusion Process of Medical Catheters Considering Viscous Fluid Flow Behavior

ABSTRACT A strategy for building data‐driven models for the screw extrusion process of octagonal polyformaldehyde (POM) medical catheters is proposed, aiming to address the challenges faced by manufacturing catheters, such as a lot of time and materials wasted in adjusting machines, difficulties in characterizing melt viscosity, and difficulties in applying expert experience to new conditions. In this strategy, the data set of extrusion parameters for manufacturing octagonal POM catheters is collected. A neural network is trained based on the data set, and a function to guide adjusting screw speed is given. The prediction errors of the trained model are almost within 5%. The given function can help to quickly approximate the possible screw speed. Further, the apparent viscosity of the melt is calculated by a flow theory, and the neural network is trained based on the calculated viscosity. Training results proved that the pressure measured near the screw head can characterize the apparent viscosity of the melt, and that the screw speed is the main parameter influencing melt viscosity. Finally, the data‐driven model for selecting screw speed is changed and transferred to another condition where the die size is different. Extrusion experiments prove that the prediction errors of the transferred model were basically within 5%.