Typical steady CFD is performed using the Reynolds-Averaged Navier-Stokes equations, however accuracy of the flowfield is strongly dependent on the turbulence model used and can be case dependent. Large Eddy Simulation produces accurate flow data consistently for a wide variety of complex flows. Although LES requires careful case setup and solution times are orders of magnitude larger than RANS, the accuracy, data detail and cost of LES relative to physical testing makes its use attractive for many applications. Modern Computational Fluid Dynamics (CFD) solvers utilise High Performance Computing (HPC) to reduce simulation turnaround time. However, the pre-processing (mesh generation) and post-processing (data extraction + visualisation) have now become bottlenecks requiring significant manual intervention and time. Modelling of flow features such as separation in gas-turbine internal cooling ducts lend themselves towards automation.
The project would involve generation of modules to automate mesh generation, run steady and unsteady modelling of such flows, extract data, perform machine learning and link these within a knowledge-based system. This would allow LES to be consistently deployed with minimal human intervention for a range of flows. Applicants would benefit from experience in fluid dynamics, CFD, HPC, Fortran/C++/Python.