Year: 2025
Type: conference-paper
Abstract: Calculation of lift and drag coefficients is an important issue in aeronautical engineering and fluid dynamics, but conventional experiments require a great deal of time and cost, while numerical calculations are hampered by computational complexity and accuracy due to unsteady flow and turbulence models. In this study, we propose a method to solve these issues by utilizing neural networks. In a previous study, errors occurred due to overlearning in machine learning, so two countermeasures were taken: increasing the number of training data and simplifying the structure of the model. As a result, both methods improved the accuracy quantitatively and qualitatively. In addition, as an approach to designing and developing new wing shapes, we constructed a neural network that solves the relationship between the lift coefficient and wing shape as an inverse problem. In this approach, the lift coefficient calculated by the forward network is input to the backward network to reproduce the wing shape. The results showed the possibility of properly modeling the relationship between the lift coefficient and blade shape, but further improvement in accuracy is needed. The results of this study provide a new approach for efficient and accurate blade shape design of gas turbines and wind turbines.
Authors: Seiya Shiihara, Arisa Masuda, Hana Hebishima
Institution: Fukuoka University
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4411367196