Machine-Learning Based Multi-Scale Simulation for Polymer Melt Spinning Process

We develop an improved multi-scale simulation method for polymer melt spinning processes by replacing the microscopic simulator with a machine-learned constitutive relation (MLCR). For the MLCR method, the estimation of stress responses for shear deformations has previously been validated. In this study, we apply this method to uniaxial elongational deformations, as a necessary step towards predicting multi-deformation mode flows. Applied to the KGCG model, the MLCR method has a higher computational efficiency and a degree of accuracy, we expect to enhance the accuracy of the predictions for the KGCG model, by using more training data to learn the constitutive relations. The high computational efficiency and a degree of accuracy of the MLCR method has helped us to simulate complex calculation conditions more efficiently, e.g., varying the apparent Reynolds numbers. The MLCR method enables us to analyze the quantitative characteristics of the flows of the KGCG model for a polymer melt spinning application more efficiently. It is an important step to be able to handle industrially relevant applications of melt spinning, e.g., the analysis of flow-induced crystallization.