MICROSCALE DOMAIN PERMEABILITY PREDICTION OF FIBER REINFORCEMENT STRUCTURES BASED ON THE LATTICE BOLTZMANN METHOD AND MACHINE LEARNING

The aim of this work was to evaluate the possibility of numerically based fluid permeability prediction for polymer composite fibrous microstructures using the lattice Boltzmann method (LBM). The work was carried out using the open-source lattice Boltzmann software library Palabos (parallel lattice Boltzmann solver). For validation purposes, we first compared the results of a simple geometry with an analytical solution before using the LBM to calculate the permeability of a real three-dimensional (3D) fibrous structure generated via X-ray computed tomography, which was used in an international virtual permeability benchmark exercise. It was found that within the relevant working range, the LBM simulations were in good agreement with both the semi-analytical models and the average benchmark results. In addition, several hundred artificially created 3D microscale geometries were generated, and the velocity fields in each were determined using Palabos, together with a statistically valid average value of their permeability in all three directions. These data were then used to train a convolutional neural network-based machine learning model for the purposes of quicker prediction of the permeability for different microstructural parameters. It was concluded that Palabos is a suitable and computationally efficient open source alternative for performing permeabilities calculations to generate simulation-based data for machine learning purposes.