E-Thesis 270 views 187 downloads
Multi-Scale Analysis of Reinforced Composites / MALEBOGO TSHEKO
Swansea University Author: MALEBOGO TSHEKO
DOI (Published version): 10.23889/SUThesis.70851
Abstract
The thesis centres on multi-scale modelling of heterogeneous solids, where the macroscopic behaviour is intricately linked to the microscopic structure. To manage the substantial memory and computational power demands of multi-scale modelling, a discrete Representative Volume Element (RVE) boundary...
| Published: |
Swansea
2025
|
|---|---|
| Institution: | Swansea University |
| Degree level: | Doctoral |
| Degree name: | Ph.D |
| Supervisor: | W. G. Wulf, and D. Peric. |
| URI: | https://cronfa.swan.ac.uk/Record/cronfa70851 |
| Abstract: |
The thesis centres on multi-scale modelling of heterogeneous solids, where the macroscopic behaviour is intricately linked to the microscopic structure. To manage the substantial memory and computational power demands of multi-scale modelling, a discrete Representative Volume Element (RVE) boundary approach based on a finite element model of the microstructure is employed. RVE refers to a small unit of a material that represents the whole material in terms of structure and properties. A new homogenisation method is explored in this research, and subsequently, the results obtained from this approach are rigorously compared to analytical methods using multiple verifying examples. This comparative analysis provides insights into the accuracy and reliability of the multi-scale modelling technique in predicting material behaviours across different scales.Homogenous, isotropic and transversely isotropic 3D solids are investigated. Then, a homogenisation tool employing the least squares method is used to determine the effective elastic properties of the composite. A more advanced tool, a Neural Network, utilises data from the least squares method to predict the elastic properties of the composite without the necessity of re-modelling the RVE and conducting additional tests.The trained neural network, which utilises data derived from least squared method plays a crucial role in predicting elastic properties of composites with very high accuracy. By integrating the neural network’s predictions with an optimisation algorithm, they can effectively tailor materials to meet specific criteria such as cost-efficiency and lightweight construction. This integrated approach not only enhances design precision but also reduces the need for extensive re-modelling of the RVE and repetitive testing. Ultimately, it fosters the development of innovative materials that strike an optimal balance between performance and resource utilisation in various engineering applications. |
|---|---|
| Keywords: |
Multi-scale, Composites, Homogenisation, Solid Mechanics,Machine learning, Neural Network, Optimisation, Isotropy, orthotropy, VolumeRepresentative Element, Matrix, Inclusion, Gmsh |
| College: |
Faculty of Science and Engineering |
| Funders: |
Botswana government |