Exploring and Predicting Fiber Laydown - A Dual Approach to Spunbond Process Optimization using Informed Machine Learning

This study extends previous work on fiber laydown behavior in spunbond processes by combining simulation, scientific visualization, and machine learning (ML). The fiber laydown process is simulated using Fraunhofer ITWM's FIDYST simulation tool across a Latin Hypercube of material and process parameters. Three output parameters - deposition spread in machine and cross-machine directions (σ1 and σ2) and a stochasticity parameter A - are analyzed. Scientific visualization techniques are applied to uncover nonlinear relationships between input factors and deposition behavior. A multi-objective regression model is trained to predict the output parameters, with conformal prediction used to provide confidence intervals. The results demonstrate that ML informed by physical simulation supports faster and more interpretable optimization of complex nonwoven processes.