Turbomachinery Blade Design Using a Navier-Stokes Solver and Artificial Neural Network

Abstract

This paper describes a knowledge-based method for the automatic design of more efficient turbine blades. An Artificial Neural Network (ANN) is used to construct an approximate model (response surface) using a database containing Navier Stokes solutions for all previous designs. This approximate model is used for the optimization, by means of Simulated Annealing (SA), of the blade geometry which is then analyzed by a Navier-Stokes solver. This procedure results in a considerable speed-up of the design process by reducing both the interventions of the operator and the computational effort. It is also shown how such a method allows the design of more efficient blades while satisfying both the aerodynamic and mechanical constraints. The method has been applied to different types of 2D turbine blades of which three examples are presented in this paper.

Performance Optimization in a Centrifugal Pump Impeller by Orthogonal Experiment and Numerical Simulation

“…Considerable effort has already been invested in studying the performance optimization in pump [4][5][6][7] and turbomachines [8][9][10]. There are various methods to optimize the design geometry, including global optimization algorithms based on heuristic algorithms and gradient-based methods.…”

Multi-Level Multi-Objective Multi-Point Optimization System for Axial Flow Compressor 2D Blade Design

“…The blending functions are defined based on the distance to the nearest wall and on the flow variables by Eqs. (13,14).…”

“…Pierret and Van den Braembussche [14] presented a multilevel method for the automatic design of more efficient turbine blades. An artificial neural network (ANN) has been used to construct an approximate model using a database containing Navier-Stokes solutions for all previous designs.…”

An asynchronous metamodel-assisted memetic algorithm for CFD-based shape optimization

“…In order to reduce its wall-clock time, surrogate evaluation models (also known as metamodels) can be used. MetamodelAssisted EAs (MAEAs), in which the metamodels are trained separately from the evolution which is exclusively based on them, can be found in Bull (1999), Pierret and Van den Braembussche (1999), but are beyond the scope of this paper. This paper is concerned with EAs (MAs, in fact) assisted by on-line trained metamodels, in conformity with the method presented in Karakasis and Giannakoglou (2006), Giannakoglou et al (2001).…”