Abstract

Aerodynamic shape optimization of axial gas turbines in three dimensional flow is addressed. An effective and practical shape parameterization strategy for turbine stages is introduced to minimize the adverse effects of three-dimensional flow features on the turbine performance. The optimization method combines a genetic algorithm (GA), with a Response Surface Approximation (RSA) of the Artifcial Neural Network (ANN) type. During the optimization process, the individual objectives and constraints are approximated using ANN that is trained and tested using a few three-dimensional CFD ow simulations; the latter are obtained using the commercial CFD package Ansys-Fluent. To minimize three-dimensional effects, the stator and rotor stacking curves are taken as the design variable. They are parametrically represented using a quadratic rational Bezier curve (QRBC) whose parameters are directly and explicitly related to the blade lean, sweep and bow, which are used as the design variables. In addition, a noble representation of the stagger angle in the spanwise direction is introduced. The described strategy was applied to optimize the performance of the E/TU-3 axial turbine stage which is designed and tested in Germany. The optimization objectives introduced the isentropic efficiency and the streamwise vorticity, subject to some constraints. This optimization strategy proved to be successful, fexible and practical, and resulted in remarkable improvements in stage performance.