Multi-fidelity strategies for lean burn combustor design

In combustor design and development, the use of unsteady computational fluid dynamics (CFD) simulations of transient combustor aero-thermo-dynamics to provide an insight into the complex reacting flow-field is expensive in terms of computational time. A large number of such high-fidelity reactive CFD analyses of the objective and constraint functions are normally required in combustor design and optimisation process. Hence, traditional design strategies utilizing only high-fidelity CFD analyses are often ruled out, given the complexity in obtaining accurate flow predictions and limits on available computational resources and time. This necessitates a careful design of fast, reliable and efficient design strategies. Surrogate modeling design strategies, including Kriging models, are currently being used to balance the challenges of accuracy and computational resource to accelerate the combustor design process. However, its feasibility still largely relies on the total number of design variables, objective and constraint functions, as only high-fidelity CFD analyses are used to construct the surrogate model. This thesis explores these issues in combustor design by aiming to minimize the total number of high fidelity CFD runs and to accelerate the process of finding a good design earlier in the design process. Initially, various multi-fidelity design strategies employing a co-Kriging surrogate modeling approach were developed and assessed for performance and confidence against a traditional Kriging based design strategy, within a fixed computational budget. Later, a time-parallel combustor CFD simulation methodology is proposed, based on temporal domain decomposition, and further developed into a novel time-parallel co-Kriging based multi-fidelity design strategy requiring only a single CFD simulation to be setup for various fidelities. The performance and confidence assessment of the newly developed multi-fidelity strategies shows that they are, in general, competitive against the traditional Kriging based design strategy, and evidence exists of finding a good design early in the design optimisation process