Optimization of the serrated airfoil for both aerodynamics and aero-acoustics using a neural network-augmented genetic algorithm

Abstract

The massive numerical computations for an accurate evaluation of the aerodynamic parameters inhibit multi-objective optimization of wind turbine blade sections from aerodynamic and acoustic aspects. This paper presents a novel approach to mitigate the problem. The method exploits the combination of a neural network-based reduced order modeling and a multi-objective genetic algorithm to optimize an S8xx-series airfoil and accordingly the trailing edge serration geometry for best aerodynamic and aero-acoustic characteristics. The Class-Shape Transformation (CST) was used to parameterize various serrated airfoil geometries. Using the modest amount of computational fluid dynamics (CFD) simulation data, a feed-forward neural network (NN), was trained to predict the airfoil behavior within a certain range. It has been worked out that this method can reduce the overall optimization time by a remarkable amount of about 1/20 using the same hardware while preserving accuracy. In multi-objective optimization of both the airfoil and the serration shapes, 5%–7% improvement in aerodynamic performance and simultaneously 1%–4% reduction in noise was pointed out compared to the benchmark airfoil.