< Back to previous page

Project

Unravelling in-situ constituent properties in fibre-reinforced composites: digital volume correlation and machine learning applied to synchrotron computed tomography images

Models are only as good as their input. Imagine that those inputs are lab- or operator-dependent, prone to large scatter, specimen size-dependent and measured in unrepresentative conditions. This would inevitably lead to unreliable model predictions as well as a lack of understanding of the features controlling the behaviour. Unfortunately, this situation applies to the state of the art in micromechanical modelling of fibre-reinforced composites. These issues impede composite material innovations in areas such as automotive, aerospace and wind energy. This project therefore develops a new test methodology that enables reliable measurements of the in-situ properties of the three constituents of fibre-reinforced composites: fibre, matrix and interface. This methodology exploits the latest advances in computed tomography to scan particle-filled fibre-reinforced composites at voxel sizes below 100 nm. Digital volume correlation (DVC) tracks these particles in the 3D images captured at increasing applied loads, allowing us to reach 2-4 µm 3D strain resolutions and thus to extract the constituent properties with a high accuracy. Due to the huge size of the acquired data, we will exploit machine learning techniques to efficiently identify regions of interest to extract strain maps and to reverse engineer the constituent properties. Since we will test composites with up to a thousand fibres rather than a single fibre, these properties are measured in conditions representative for real composites. These features empower us to develop fundamental knowledge on the in-situ constituent properties. This knowledge will create a breakthrough in micromechanical modelling, which would catalyse the design and implementation of improved composites, as next-generation materials for sustainable development.
Date:1 Oct 2021 →  Today
Keywords:polymer matrix composites, computed tomography, digital volume correlation, machine learning, computational materials science
Disciplines:Polymer composites, Computational materials science, Destructive and non-destructive testing of materials, Machine learning and decision making