MACHINE LEARNING BASED TRANSONIC FLOW COMPUTATIONS ON AIRCRAFT LIFTING SURFACES

Date

2024-6-11

Author

Gömeç, Fazıl Selçuk

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

230
views

45
downloads

Cite This

This study aims to reduce the computational cost of high-fidelity transonic flow simulations to the level of empirical tools using Machine-Learning methods. Therefore, high-fidelity wing optimizations and aeroelastic analyses of fighter aircraft can be affordable even in the conceptual design phase. The characteristics of pressure distributions over the various wing geometries are investigated. Deep neural networks having encoder-decoder architectures are created to model the wing pressure variation along chord and spanwise directions. The deep learning training is applied at Mach 0.9, and the angle of attack ranges between 7 and 12 degrees. Wing planform figures and low-fidelity datasets are the inputs of the deep learning algorithm. The low-fidelity dataset is created with 2D airfoil pressure distributions. The aspect ratio, leading-edge sweep, and taper ratio are represented in the wing planform figure. The airfoil characteristics, angle of attack, and tip twist are integrated as the low-fidelity dataset. $\Delta C_{p}$ distribution between the high and low fidelity dataset is the output of the deep learning model. A predictor algorithm is created to generate the high-fidelity pressure distributions using $\Delta C_{p}$ predictions. The size of training inputs, the number of feature maps, and interpolation performances are investigated. Pressure variations of different wing planforms are generated and compared. As the output of the thesis, the static pressure field on the suction and pressure sides of a fighter wing are accurately predicted with deep neural networks which are created by an encoder-decoder algorithm.

Subject Keywords

Aircraft design, Machine learning, CFD

URI

https://hdl.handle.net/11511/110086

Collections

Graduate School of Natural and Applied Sciences, Thesis