Macroscale Modelling of 3D-Woven Composites: Inelasticity, Progressive Damage and Final Failure

Doctoral thesis, 2022

Composites with 3D-woven reinforcement have been slowly making their way into different industrial applications. The interlacement of yarns, not only in-plane but also through-thickness, means that in many applications 3D-woven composites can outperform their laminated counterparts. In particular, this includes increased out-of-plane stiffness and strength, damage tolerance and specific energy absorption properties. The widespread adoption of 3D-woven composites in industry however, requires the development of accurate and efficient computational models that can capture the material behaviour.

In terms of computational efficiency, the most promising choice is to treat the material as a homogeneous and anisotropic solid. This is referred to as a macroscale model. Developing a macroscale model, which can predict how 3D-woven composites deform and eventually fail, is the main focus of this work. Particular attention is given to predicting the relevant non-linear behaviours that lead to energy absorption.

A framework for modelling the mechanical response of 3D-woven composites on the macroscale is presented. The proposed framework decomposes the stress and strain tensors into two main parts motivated by the material architecture. This allows for a convenient separation of the modelling of the shear behaviour from the modelling of the behaviour along each of the reinforcement directions. In particular, this division allows for a straightforward addition and modification of various non-linear phenomena observed in 3D-woven composites. As a next step, material modelling approaches are considered and added to the framework in order to capture these non-linear phenomena. This includes the use of a viscoelastic model as well as a combined elasto-plastic and continuum damage model to capture the development of permanent deformations and stiffness reduction mechanisms. Finally, an anisotropic phase-field model extension is developed in order to induce local softening and failure in a way which does not induce spurious mesh-dependencies in finite element analyses. The model predictions are compared to experimental tests and show good agreement.

The aim has been to develop a model that allows the constitutive relations to be identified directly from uniaxial cyclic stress-strain tests without the need for complex calibration schemes. However, characterising the out-of-plane behaviour is not trivial. Therefore, the current work also explores the use of high-fidelity mesoscale models as an additional source of data for model calibration and validation.