PhD in Mechanical Engineering

This project fuses machine learning (ML) based inverse design approaches and topology optimisation (TO) to realise multiscale architected materials or metamaterials (MTM) that can undergo targeted non-linear response. You will develop a computational framework that can reveal novel Multiphysics (thermo-mechanical) MTM solutions for demanding terrestrial and non-terrestrial (Aerospace) applications accounting for sources of variability and uncertainty, for example, those arising from material, fabrication process, and boundary and loading conditions. By generating datasets from finite element simulations, ML models can learn the mapping between unit cell design parameters and homogenised properties. State-of-the-art approaches – such as tandem neural networks, video diffusion models, and reinforcement learning – will be explored to efficiently navigate these high-dimensional, nonlinear design spaces. To achieve robust property-to-design mapping, mixture density networks or MDN-based inverse generators will be employed to capture the multimodal distribution of the design space, enabling flexible inverse design sampling based on probability, property, geometric or printability criteria.