Optimisation & AI

This research area focuses on the intersection of mathematical design and intelligent computing to enhance the performance and efficiency of thermo-fluid systems. By employing Topology Optimization, we systematically determine the optimal distribution of material and fluid pathways within a given design space, allowing for the discovery of non-intuitive, high-performance geometries that minimize pressure drop or maximize heat transfer. Parallel to this, we integrate AI and Machine Learning in Fluid Dynamics to accelerate traditional CFD workflows through the development of surrogate models, reduced-order modeling, and data-driven turbulence closures. These synergistic approaches enable us to navigate vast design spaces rapidly, providing transformative solutions for aerospace, automotive, and energy-thermal management systems.

1. Topology Optimization

We apply mathematical optimisation techniques to systematically design high-performance fluid systems. By integrating the level-set method with adjoint-based optimisation and OpenFOAM, we perform multi-objective topology optimisation for both drag reduction and heat transfer enhancement. Our research focuses on identifying the "optimal shape" that minimises pressure drop while maximising thermal dissipation in heat exchangers and aerodynamic components. Through the use of Adaptive Mesh Refinement (AMR), we ensure that the evolving boundaries during the optimisation process are always resolved with maximum accuracy, leading to innovative designs that transcend traditional engineering intuition.

2. AI/ML in Fluid Dynamics

We are at the forefront of integrating artificial intelligence and machine learning into the fluid mechanics workflow. Our research focuses on data-driven fluid dynamics, where we use ML algorithms for Reduced-Order Modelling (ROM) to accelerate expensive CFD simulations. By training neural networks on high-fidelity DNS data, we develop surrogate models that can predict complex flow fields in real-time. Additionally, we apply AI-based image processing to experimental flow data (AI-Velocimetry) to enhance resolution and extract hidden features from noisy measurements, effectively creating a "digital twin" of our experimental and numerical setups.

Page updated

Google Sites

Report abuse