生成AIを用いた2次元翼の空力騒音予測

抄録

Machine learning and generative AI were performed to flow fields generation and aerodynamic noise estimation for optimization of fan design. CNN is suitable for design optimization, and the estimation results using machine learning data based on 2D airfoil data show the same trend as actual fan performance. It reveals that fan performance evaluation using machine learning can contribute to reducing analysis data and design time. The results show that fan performance evaluation using machine learning can contribute to reducing analysis data and design time. When estimating aerodynamic noise based on flow field information using generative AI, it was found that if the data is reduced too much, the estimation accuracy is not sufficient, as small changes in the flow field are important. Under the conditions of this study, the latent variable vector for the generated AI was rather large 64. It was found that care should be taken with the size of the collapsed data when estimating images and when estimating physical phenomena such as aerodynamic noise.