Multi-scale mechanical modelling of highly heterogeneous woven composites

Abstract

The multi-scale nature of woven composites can be clearly revealed by the strong dependency
of the mechanical behaviour on morphological features of lower length scales.
Geometric irregularities, induced during the manufacturing stages in the yarn architecture,
alter the meso-scopic material stress field, often dominating the overall material response.
In this dissertation, a numerical investigation supported by experimental data is performed
to shed light on the failure mechanics of complex 3D weave composites, followed by a novel
computationally efficient modelling approach that addresses the high computational burden
and non-compliant material homogenisation length scale issues that typically arise with
macro-scale modelling. In the first part of the dissertation, a high fidelity meso-scalemodelling
framework is formed utilising current state-of-the-art textile modelling technologies. Starting
from the textile manufacturing process simulation, an in-house multi-chain beam kinematic
solver is used to provide an accurate material representation. The acquired realistic “as-woven”
material geometry is then mapped into a voxelised domain to complete the finite element
mesh description. Complementing the model assembly, an appropriate constitutive model
describes the response of eachmaterial, while the damage regime is relying on a continuum
damage mechanics approach. Themeso-scale model capability is evaluated against a comprehensive
set of experimental data from tensile tests on in-plane pre-sheared specimens,
focusing closely on the physics of damage and its relation to the material architectural features.
Having gained a valuable insight on the mechanical behaviour, the second part of the dissertation
discusses the development of the novel macro-scale modelling approach. By employing a
three-dimensional Voronoi tessellation, amacroscopic description of the material is achieved
in a set of collectively exhaustive and mutually exclusive polyhedral cells. The yarn surface
points extracted from a high-definition meso-scale geometry input file are in this case the
Voronoi generator seed points. The output off the tessellation is then reconstructed so that
each cluster of small Voronois belonging to a yarn segment becomes a single well-defined,
arbitrary shape polyhedron. To exploit the reduced complexity of the reconstructed geometry,
a non-standard polyhedral element formulation based on the smoothed finite element
method is employed, directly translating the polyhedral cells into finite elements. In respect
to the distinctive cell morphology consisting of yarn and matrix volumetric partitions, a bimaterial
homogenisation constitutive model is constructed describing the material response
based on the local meso-structure and the mechanical properties of the two constituents.
Inaccuracies in predicted results fromconventional homogenisation techniques typically arise
from smearing out heterogeneities on the yarn level. The capability of the newly proposed
framework to overcome such deficiencies in a computationally efficient manner is investigated
through benchmark representative volume element models.Model capability on capturing
the elastic and damage material response, along with the corresponding computational cost
are illustrated and reveal the framework’s potential to successfully handle problems of the
intermediate length scales, more specifically coupon, structural element, and component
detail simulations.

Date of Award	4 Jul 2023
Original language	English
Awarding Institution	University of Bristol
Supervisor	Stephen R Hallett (Supervisor) & Bassam El Said (Supervisor)