Application of CFD zooming for preliminary design of a low emissions combustor.

The design of low emissions combustors is particularly challenging as there is a requirement to deliver designs that meet a large number of performance, emissions and operability (often conflicting) objectives. There is an increasing need for combustor preliminary design and performance tools which can be used in the early phases of the design process for rapid design space exploration thereby reducing the risk and cost in the long term. Although both reduced order models and higher fidelity tools have been widely used for preliminary design independently, significant benefit can be derived from using a multi-fidelity modelling approach to address the limitation of reduced order model (accuracy) and high fidelity CFD (time and cost). To the author’s best knowledge there is no information in the public domain related to the coupling of reduced order models with higher fidelity 3D CFD multi-fidelity modelling tools for low emissions gas turbine combustion systems. Such a tool has a potential to offer a good compromise between modelling accuracy and computational expense. In this PhD research, a novel multi-fidelity zooming combustor preliminary design method is proposed. The method uses design outcomes of an existing reduced order model based design tool to construct CFD models for a series of RANS simulations. A case study for the design of a Lean Direct Injection Partially Premixed combustor was conducted to identify the limitations of an existing reduced order modelling approach. Dedicated CFD simulations were performed to demonstrate that improved methods/models/correlations can be derived from these higher fidelity simulations to refine the existing reduced order model. The main research contributions are summarised below: External aerodynamics – Performance is sensitive to inlet velocity profiles, the effect of which cannot be reflected in ROMs, realistic compressor outlet profiles is needed instead of generic turbulent pipe flow profiles. – Performance maps were generated from CFD which include more degrees of freedom and suggest a different ‘optimum locus’ than 1D correlations. Fuel injector initial conditions – The Sauter Mean Diameter calculated from correlations in the ROM is not suitable to be used as injection initial condition. Detailed correlations on jet breakup were used to generate representative droplet size and velocity for different nozzle designs and conditions. – Swirler flow split correlations does not account for flow turning in the venturi and the pre-mixer, coarse mesh CFD was sufficient to generate more accurate flow splits among different stages. Reacting flow – The initial 10 fuel nozzle ports design from the ROM was not sufficient for good mixing quality at the main stage, which resulted in higher flame temperature. The number was increased to 16, which provides more uniform flame distribution at the circumferential direction. – Three of the four methods used to generate the time delay provides consistent results. The time delay was used as an input of the ROM thermoacoustic analysis model. – The reactor layout can be better customised for emissions prediction with extra zones within the pilot injector and the dilution zone to account for reaction and recirculation. – Combustor cooling design was refined without modifying the variables of ROM, in which circumferential distribution was not captured. Simplified re-fining method was developed at less computational expense compared to complete Conjugate Heat Transfer simulations with the radiation model. Based on these findings, the reduced order design tool could be refined once the data from all parametric study cases are extracted and incorporated in the model, which is recommended as the future development of the work. The CFD model constructed could also be used to initiate higher fidelity Large Eddy Simulation.