Aerodynamic Shape Optimization of Turbine Blades Using a Design-Parameter-Based Shape Representation

Abstract

Currently, most shape optimization activities for 2D blade sections focus on modifying the blade shape locally to get an optimum one, which implicitly assumes that the global shape is near optimum. Moreover, the common design parameters in most cases are not the variables used in shape optimization, hence the designer does not have control over the parameters that he or she uses in the design. In this work, the turbine blade shape at any given radial location, is represented with the MRATD model (Modified Rapid Axial Turbine Design), which is a low-order representation that describes the blade profile using a maximum of 17 aerodynamic design parameters that are given (and used) by the turbine designer, e.g. the blade axial chord, stagger, maximum thickness, throat, uncovered turning, inlet and exit blade and wedge angles, LE and TE radii etc... This representation is used in an optimization scheme to sweep the design space and identify the design parameters that would accomplish a certain optimization objective (e.g. maximum adiabatic efficiency) subject to some constraints (e.g. fixed throat area or minimum TE radius or maximum TE wedge angle or metal angles etc...). The optimization scheme uses evolutionary optimization algorithm, Genetic Algorithm(GA) and, to save computing time, Artificial Neural Network (ANN) is introduced to approximate the optimization objectives and constraints; it is trained and tested using a relatively small number of high fidelity CFD flow simulations. This approach to geometry representation is used to carry out a sensitivity study of the effect of the different design parameters on the blade performance of a highly efficient subsonic turbine blade. Its impact on the design process is also demonstrated.

Performance Improvement of an Industrial Axial Turbine: Two-Stage Power Turbine Aerodynamic Optimization

“…The blade shape parametrization plays an important role in the optimization process. Many references have been presented variety of methods to parameterize the profile shape including Pritchard method, MRATD model, NURBS, Bezier, and B-spline curves [19][20][21][22]. In this study, the 2D profile shape at any given radius is parameterized using the five-point Bezier curves of both camber-line and thickness distributions.…”

Improved centrifugal compressor impeller optimization with a radial basis function network and principle component analysis

“…Moraal and Kolmanovsky [13] indicate that an ANN can produce better performance compared with other curve fitting techniques if the ANN is sufficiently trained. Feedforward neural networks (FFNNs) [6,14,15] with the back-propagation learning algorithm and the RBFN [3,16] are the most widely used ANNs to train performance maps for turbomachinery. An FFNN, which is one type of ANN, also called multi-layer perceptrons, consists of one input layer, one or several hidden layers, and one output layer.…”

An optimum design procedure for an aerodynamic radial diffuser with incompressible flow at different Reynolds numbers

“…NACA profiles may be selected [4,6], but also straight vanes can be adopted, especially when low solidity vaned diffusers are requested [2,19]. Examples of vaned diffuser geometries derived from a given blade loading distribution were also proposed [8], as well as optimized profile shapes [10,13].…”

“…Numerous shape optimization procedures have recently been proposed, most of them based on evolutionary techniques [9][10][11][12][13]. Excellent improvement in performance has been obtained, but rarely a new optimum design may readily be made with previous solutions when one or more design parameters are changed.…”