Physics-embedded deep learning framework for next-generation microstructure design to enhance fibre bridging in composites

Micromechanical finite element (FE) techniques have shown great promise in naturally predicting the key mechanism of fibre bridging, which can significantly improve toughness in fibre-reinforced composites. However, they remain too computationally expensive for practical microstructural design. I will build on the strengths of these micromechanical FE techniques to establish a new, scalable virtual design paradigm. To achieve this, I will develop a Physics-Embedded Deep Learning framework that embeds phase field theory within deep learning models. This innovative tool will enable rapid, high-fidelity virtual testing across a range of microstructures, thereby overcoming the computational barrier. The framework will be demonstrated by addressing the critical problem of transverse cracking. Its primary application will be the design and optimisation of novel twisted hybrid core fibre microstructures, engineered to maximise fibre bridging, enhance toughness, and significantly delay crack initiation and growth. The project builds on my prior contributions in modelling transverse cracking and developing deep learning methods, and will be further advanced by an ideal collaborative environment. It will be carried out at the University of Oxford under the supervision of Prof. Emilio Martínez-Pañeda, a world-leading expert in computational mechanics and phase field modelling, with a secondment at the University of Porto under Prof. Pedro Camanho, an internationally recognised authority on the mechanics of advanced polymer composites. For me, as a fellow, this project is a vital step toward research independence, advancing composite materials through the integration of computational mechanics and deep learning.