Hydroacoustic optimization with using 3D viscous-based Noise-GAN

In recent years, noise pollution has significantly affected marine organisms, necessitating the implementation of certain restrictions and safety protocols. The primary objective of these restrictions is to reduce the noise produced by human-operated vehicles in aquatic environments. For this reason, hydroacoustical studies are increasingly being integrated into design processes. This study aims to introduce an innovative approach to the design of hydrofoils, which are regarded as a critical component in hydroacoustic design. The focus of this approach is to develop an advanced optimization tool by integrating machine learning with hydroacoustic performance calculations. This study presents the 3D viscous-based Noise-GAN method, which innovatively combines Generative Adversarial Networks (GAN) algorithms with hydroacoustic performance calculations, enhanced by 3D viscous-based performance calculators. In contrast to the inviscid-based versions, this method, which incorporates 3D and viscous effects, allows for a comparative analysis of the impacts of these effects on the optimization process. Particularly, the performance of optimal geometries obtained through both 3D and 2D solvers will be compared, elucidating the role of 3D effects in the optimization process. This study addresses the drawbacks of 2D profile solutions in the optimization process, which generally offer a rapid solution in the field of machine learning for shape optimization. The effects have been examined at three different angles of attack (AoA). Thus, the positive and negative impacts on the optimization process under challenging environmental conditions have been identified. Additionally, cavitation constraints have been incorporated into the optimization process, ensuring that only profiles devoid of cavitation risk under the specified conditions are considered. Through the utilization of GAN algorithms, innovative profile geometries that do not present cavitation hazards at various angles of attack have been developed. The performance of the obtained optimal geometries has been compared to the widely utilized NACA0009 profile. By comparing the performance of the newly derived geometries with that of a profile with average performance, meaningful insights have been drawn. The results from the 3D viscous-based Noise-GAN method have been presented alongside the outputs derived from the 2D viscous-based method and the performance results of the NACA0009 profile under different Angle of Attack (AoA) conditions in this study.