Optimizing spunbond nonwovens for filter media production using a novel approach of machine learning and fiber/fluid simulations

S. Gramsch*, A. Sarishvili, A. Schmei├čer, Fraunhofer Institute for Industrial Mathematics ITWM, Germany

Nonwoven filter media are getting more and more important due to their versatile properties. Nonwovens are classified by their production process. Mainly, there are three different nonwoven production processes: dry-lay processes, wet-lay processes, and extrusion processes. In this paper, we focus on the simulation of spunbond processes that belong to the class of extrusion processes.

The Fraunhofer Institute for Industrial Mathematics ITWM has developed a simulation tool called FIDYST that simulates the fiber dynamics in the air and the laydown process of the fibers on a transport belt of spunbond production processes. Hereby, the final web quality depends mainly on the air stream, but also on the material properties of the fibers. Simulating an existing spunbond process is state of the art. Optimizing or designing a new spunbond production process with specific properties of the resulting filter media requires a huge amount of simulation runs. Hereby, performing the corresponding CFD simulations is a very time-consuming task. Additionally, the fiber dynamics simulations include stochastic forces due to the turbulent air stream. Hence, a statistically significant number of representative fibers must be simulated in order to get realistic results.

In this paper, we present a novel hybrid informed learning approach that generates an optimal design of experiments for the simulation runs based on machine learning techniques. Furthermore, the general simulation method for spunbond processes and its application to filter media production is demonstrated...

Session: F9 - Filter Media - Modelling, Artificial Intelligence, Machine Learning
Day: 24 October 2019
Time: 09:00 - 10:15 h

Learn more at FILTECH 2019 - Register Now!