Accelerating Computational Fluid Dynamics

The primary goal of the research is accelerating computational fluid dynamics (CFD) using a combination of novel numerical methods, hybridisation with experiments, and using parallelisation strategies on heterogeneous back-end hardware (including clusters of GPU's and CPU's). Automatic and intelligent mapping between high and low fidelity CFD solvers will be used to accelerate computationally intensive, high resolution (possibly eddy-resolving) simulations. Novel high order spatial discretisation schemes will be implemented in order to reduce the mesh density and hence computational cost for a high level of accuracy. An unstructured mesh solver, specifically for turbomachinery, will be developed as a platform for the developments within the research. Unstructured grids allow automatic meshing of highly complex geometries with minimal user intervention. This will enable real-geometry features in turbomachines such as shrouds, cooling holes, glands, seals, bleed-offs etc. to be simulated, allowing a greater understanding of the interactions between the resulting flow physics. This improved understanding can be incorporating into design and used to improve efficiency, component lifetime and plant flexibility. Capturing such complex geometries is either impossible or a very time consuming processes on multi-block structured grids.

With regards to application, the research is primarily targeted at increasing the operational flexibility of conventional power generation plants. However, the philosophy behind these novel numerical techniques can be applied to many other disciplines. As the penetration of intermittent renewable sources into the energy mix increases, conventional power generation technologies (e.g. combined cycle power plants) must be operated more flexibly. This will require frequent start-up and shut-down cycles, rapid ramp-up rates, and increased performance a reduced load. Plant monitoring will become essential in order to improve operational flexibility of the plant. Real-time monitoring using instrumentation and measurements only provide data at discrete spatial and temporal points. Hybrid tools will be developed to automatically tune and accelerate CFD simulations, in real-time, based-on experimental monitoring data from the plant. Hybridising experimental monitoring with CFD combines both the high spatial resolution (from CFD) and high accuracy (from experimental measurements). Feeding know experimental data intelligently into CFD simulations will significantly enhance the convergence rate to steady state. The acceleration of the CFD solver gained from novel parallelisation strategies and numerical methods discussed previously, coupled with the acceleration provided by known measurements from real-time experiments, will result in extremely fast CFD simulations. Therefore, CFD simulations hybridised with experiments will provide a complete picture of the flowfield and/or temperature field within a gas/steam turbine in real-time. These improved plant monitoring capabilities can be coupled with active control in order to significantly improve the plant flexibility. For example, the local temperature of the outer casing a steam turbine can be actively controlled in order to minimise thermal stresses and increase the start-up rate.