From
Timetable to
Place Amphi Dorothy Hodgkin
Extraction of 3D weaving architecture of large composite parts from tomographic images
Traditional simulations of composite materials often rely on small-scale samples or simplified models that fail to capture the complexity of real woven structures.
This thesis addresses this limitation by developing automated segmentation methods for tomography images of large-scale 3D woven composites.
The goal is to generate high-fidelity numerical models from coarse-resolution CT images (above 140 μm), focusing on textile reinforcements in fan blades.
The first part introduces a variational tracking algorithm to follow yarns within the reinforcement.
It leverages three statistical properties: cross-sectional shape, disorientation, and spatial arrangement between neighboring yarns.
The method tracks yarn centers across successive cross-sectional planes orthogonal to their main directions.
It is computationally efficient, processing thousands of yarns in seconds per slice, making it suitable for industrial-scale components.
The second part exploits the approximate periodicity of the weaving.
Combined with Digital Volume Correlation (DVC), it enables segmentation of highly compacted regions such as the root.
A representative periodic cell is segmented and used to build an idealized model.
DVC maps this model onto the real structure, propagating segmentation across the full volume.
Overall, the thesis proposes efficient and scalable methods for automated textile reinforcement annotation in complex composite structures.
Composition du jury :
- Angulo Lopez Jesus, Directeur de recherche, Mines Paris - PSL - Rapporteur
- Hamila Nahiene, Professeur des universités, Bretagne INP - Rapporteur
- Laurin Frédéric, Directeur de recherche, ONERA - Examinateur
- Béatriz Marcotegui, Professeure des universités, MINES ParisTech - Examinatrice
- Frédéric Jurie, Professeur des universités, Université de Caen - Examinateur