Machine Learning-Augmented Boundary Value Problem Solvers for Optimizing Melt Spinning Processes

Melt spinning is a crucial step in the production of technical textiles, known for their strength, durability, wrinkle resistance, and moisture-wicking properties, which make them prevalent in daily use. Traditional numerical models, often referred to as digital twins of physical processes, are employed for optimization but face limitations in complex and computationally intensive scenarios. The results from the simulations based on these models are difficult to interpret without domain expertise, making it challenging to identify patterns, correlations, and causal relationships within the findings. This paper aims to enhance the efficiency of classical numerical models in melt spinning by incorporating machine learning into the optimization pipeline. This integration seeks to expedite the optimization process while minimizing the time and domain expertise needed. We further demonstrate how numerical models and machine learning synergize in scientific machine learning to enhance the optimization and analysis of melt spinning processes, using boundary value problem solvers as a case study.