Summary: | Optimisation of multistage axial compressors is a time-demanding challenge in large need of effective methods. Traditionally, the design has been based on "trial-and-error" procedures, which count on the skills and experience of the designer to suggest redesigns that are likely to yield improvements. While design by "trial-and-error" usually yield acceptable solutions, automatic shape optimisation will assist in finding the optimal design and shorten the design cycle. This thesis considers issues that have to be solved before automatic shape optimisation of compressor blades can be performed, as well as a demonstration of the developed methods suitability for optimising 2D compressor blades. In paper A the geometry parameterisation of compressor blades is studied. Non-Uniform Rational B-Splines (NURBS) are used for the basic geometry description. The profile shape is then described in terms of traditional parameters, e.g. throat and maximum thickness, and an optimisation algorithm is used to adjust the NURBS parameters until the requested values of the traditional parameters are obtained. In paper B the presence of numerical noise in the objective function is investigated. The occurrence of numerical noise is related to the discretisation of the partial differential equations describing the flow. The truncation error in the flow equations together with the discontinuous change in the grid as the geometry is modified will produce a small amplitude "wave-pattern" imposed on the objective function. Two optimisation techniques, a response surface model and a gradient based method, are used to study how the solution is affected by the numerical noise when minimising the pressure drop in an axisymmetric contraction. In paper C the geometry definition using implicit design variables (paper A) is combined with a response surface model (paper B) to optimise the performance of a 2D compressor blade for a jet engine. The results show that the geometry modification strategy using implicit design variables can be used for optimisation. However, more experience is necessary before a definite conclusion can be drawn to claim that using these design parameters are superior to using the control points of the NURBS. === Godkänd; 2001; 20070314 (ysko)
|