InterPore2025

Deep Learning-Based Surrogate Model for Predicting Air Permeability of Technical Textiles

22 May 2025, 12:35

15m

Oral Presentation (MS22) Advances in Porous Materials: Design, Characterization, and Applications MS22

Speaker

Eleonora Bianca (Polytechnic of Turin)

Description

The permeability of technical textiles is of crucial importance for industrial applications. It plays a key role in the development of high-performance fabrics as well as in the production of composite materials. In this context, porosity and structural properties have a much stronger influence on permeability than fibre composition, so the geometric design of the fabric is a more decisive factor than the choice of material. Air permeability is critical to fabric comfort, drying efficiency and production processes (such as drying itself), where the structure of the fabric can significantly affect performance. However, quantifying air permeability, especially for tightly woven fabrics, remains a challenge. Therefore, predicting permeability during the design phase can optimise production processes and minimise the need for extensive experimental testing. Liquid Composite Moulding (LCM) is a process for the production of fibre-reinforced composites in which dry fibre reinforcements are impregnated with low-viscosity resin. This process is subject to Darcy’s law and is driven by the viscosity of the resin and the permeability of the material [1].
The aim of this project is to develop a surrogate model for permeability prediction based on deep learning techniques: in particular Fully Connected Neural Networks (FCNNs) and Convolutional Neural Networks (CNNs). These models are selected based on the desired output dimensionality: integral values or full-field spatial data. The model aims to predict the permeability from structural input parameters such as the weave type (plain weave, basket weave, filled rib and twill), cover factor (the ratio of the area covered by the fabric to the open area) and the aspect ratio (the shape ratio of the space between the yarns).
To achieve this goal, a training dataset was generated from CFD simulations performed for different geometries. These geometries differ in terms of weave pattern and yarn density and were generated using custom Python code implemented in the TexGen software API [2]. A semi-automatic process was developed to investigate a wide range of geometric parameters, including mesh generation and adjustment of simulation settings. Permeability was calculated using Darcy’s law under laminar flow conditions [3]. This approach made it possible to evaluate the permeability through the air velocity flowing through the fabric.
Finally, the CFD models were refined and validated against experimental results from the existing literature [4]. By combining CFD simulations with deep learning techniques, this study provides a powerful set of tools for the prediction of air permeability that facilitates the design and optimisation of technical textiles in various industrial applications.