Experimental acoustic study of small horizontal axis wind turbines based on computational fluid dynamics and artificial intelligence approaches

In modern wind turbine design, two significant challenges arise: achieving optimal aerodynamic performance while minimizing acoustic noise emissions. However, the extensive numerical computations required for accurate evaluation often hinder the implementation of multi-objective optimization strategies. This paper introduces an innovative approach to address this issue, leveraging a combination of neural network-based reduced order modeling and a multi-objective genetic algorithm. This methodology aims to optimize the aerodynamic and aero-acoustic characteristics of an S8xx-series airfoil, including the trailing edge serration geometry. Utilizing Class-Shape Transformation to parameterize various serrated airfoil geometries, the method minimizes the need for costly computational fluid dynamics (CFD) simulations. Instead, a feed-forward neural network (NN) is trained with a minimal dataset to predict airfoil behavior within a specified range. Comparisons between CFD results and NN predictions validate the accuracy of the neural network. Significantly, this approach substantially reduces optimization time by approximately 95%, maintaining high levels of accuracy. In conducting multi-objective optimization for both the airfoil and serration shapes, the study demonstrates notable improvements: a 5 to 7% enhancement in aerodynamic performance alongside a simultaneous 1-4% reduction in noise compared to benchmark airfoils. Then in the second step, experimental methodology is employed to investigate the aeroacoustic attributes of a small horizontal axis wind turbine with optimized blades. Conducted within a semi-anechoic chamber, this investigation meticulously positions both original and optimized geometry models to measure sound pressure levels (SPL) across various rotational speeds and positions. The results reveal subtle enhancements in aerodynamic performance with the optimized serrated blade configuration, accompanied by a remarkable reduction in noise levels across the frequency spectrum, culminating in an impressive overall sound pressure reduction of approximately 10 dB. Additionally, intriguing observations highlight the impact of turbine rotational speed on noise production, particularly in the downstream domain. Notably, the noise emission reduction for the serrated optimized blade is more dispersed in the plane of rotation compared to the original blade, which exhibited nearly uniform noise distribution. Overall, these findings offer valuable insights into the intricate interplay between aerodynamics and aeroacoustics in the context of small wind turbines with optimized blades.