Prediction of the Effective Properties of Textile Composites based on X-ray Computed Tomography Data

Modelling the geometry and the mechanical properties has been an important part of the research in composite materials. Accurate models are necessary to understand the mechanical behaviour and predict structural response of composite parts and structures. This thesis aims at bridging the gap between the highly detailed 3D imaging capabilities of X-ray computed tomography (CT) and the modelling of composite materials, by developing a modelling approach that uses models of composite materials generated directly from X-ray CT images. Creating a material model from a 3D CT image involves a number of steps: from the physical X-ray imaging of a material sample, to the processing of the acquired image, and to the construction of a model of the internal structure of the sample. A proper choice of acquisition parameters is necessary to maximise the attenuation contrast between the constituents of the material, and can be achieved through the analysis of energy-dependent attenuation functions of the material’s constituents. Extraction of the information required for creating a model of a composite sample is achieved through the processing of a 3D CT image. Creating a model of a fibre-reinforced textile composite requires knowledge of the local fibre orientations and the distribution of the material components, which are the matrix and the reinforcement. Structure tensor method is adopted for the calculation of the local fibre orientations from the image. It is shown that the precision of the structure tensor in calculating fibre orientations depends on the noise level in the image, voxel size and resolution of the image, local fibre packing density. A method to estimate the precision of the orientations computed with structure tensor in a particular CT image is formulated, which is based on the structural anisotropy parameter derived from the eigenvalues of the structure tensor. Methods to segment a 3D image into the material components (matrix/reinforcement) is developed, which use either unsupervised classification based on clustering, or supervised classification with Gaussian mixture models. The segmentation is done using so-called feature vector, which includes local average grey value, structural anisotropy, and, optionally, local fibre orientation angle. The proposed modelling approach is validated with three different types of textile composite materials. The predicted properties are: the effective Young's modulus; non-linear material response due to the non-linearity of the component's properties and due to the accumulated damage; permeability of a composite preform.