Efficient Optimisation & Surrogate Learning for Fluid Dynamics Designs

Simulation-driven engineering design poses two challenges: optimising over expensive black-box functions defined on high-dimensional geometry spaces, and learning solution operators for complex physical systems such as multiphase flows that need non-standard discretisation. We address both.

For chemical reactor design, we formulate geometry optimisation as a multi-fidelity black-box problem and solve it with a GP surrogate and cost-adjusted acquisition function that adaptively allocates evaluations across computational fluid dynamic fidelities (CFD), recovering high performing geometries under a fixed compute budget. For two-phase nozzle flows, where adaptive mesh refinement produces irregular, geometry-dependent discretisations, we learn a neural operator that maps AMR cell-density fields to a compact latent via a Fourier Neural Operator, and decode transient flow trajectories. Together, these works establish physics-informed representation learning as a unifying principle for tractable, data-efficient simulation-driven design.

Bio: I am an Assistant Professor in Chemical Engineering at the University of Manchester, specialising in Digital Manufacturing and AI. I am building a research group focused on machine learning for engineering systems, combining computational fluid dynamics, optimisation, and advanced manufacturing to accelerate the design of next-generation chemical technologies. Previously, I was a Research Associate in the Department of Chemical Engineering at Imperial College London.