Combustor Design Optimisation

Research Theme: Computational Design

Gas turbine combustor design requires consideration of many, often contradictory, objectives. It is only recently that the computational power has become available to use automated techniques to investigate these design objectives. This project will look at coupling methods of different computational cost into one tool.

Motivation

The motivation for this project was to investigate whether gas turbine combustor design could be improved using automated optimisation techniques.

Objectives

  • Investigate the design space using a low order method and automated optimisationtechniques
  • Generate a representative CFD solution of the combustion chamber
  • Use the automated optimisation techniques to couple the low order methodand the CFD into one tool

Method

LOTAN is used to determine the low order objective function values. This tool was developed by S. Stow to determine the acoustic properties of gas turbine combustion chambers. It can also determine the average flow properties in each modular section of the combustion chamber.

The low order objective functions used are:

  • Reduce NOx
  • Reduce CO
  • Decrease the growth rate of the most unstable mode
  • Increase the frequency of the most unstable mode
  • Reduce the pressure loss

The test rig used is based on the RB211-DLE industrial gas turbine. The rig is located in the Engineering Department at Cambridge University, and the LOTAN geometry is based on this setup.

The optimisation technique used is the Tabu Search method. This combines a derivative-based local search method with a stochastic global search method. This method has been shown to work well for CFD optimisation.

The MCUNEWT CFD solver is an in-house RANS code. The CFD solution shown in the images is a three-dimensional solution, including fuel injectors. Future work will investigate whether a two-dimensional solution will provide greater flexibility to the multi-fidelity tool.

Findings

It has been shown that the pollutant objective functions depend only on a few geometry design variables, whilst the acoustic objective functions can be influenced by any of the design variables. In addition, the influence of any one design variable on the acoustic variables is not constant, but varies throughout the design space.

The importance of an accurate flame transfer function has also been shown, by showing how two flame transfer functions developed for the same combustion rig can result in very different results.

Acknowledgements

Support for this project was provided by Rolls Royce plc and the EPSRC.

Selected Publications

  • S.G. Wyse , G.T. Parks. and R.S. Cant, "Towards theMultiobjective Optimisation of Gas Turbine Combustors", Proceedings ofthe ASME Turbo Expo 2006, GT2006-90950