A CONVOLUTIONAL NEURAL NETWORK METHODOLOGY WITH A MOMENTUM-FLUX-BASED LOSS FUNCTION FOR PREDICTING AERODYNAMIC FLOW AROUND AIRFOILS

Date

2023-12-7

Author

Deniz, Mustafa Mert

Metadata

Show full item record

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

Item Usage Stats

155
views

142
downloads

Cite This

The selection of critical components like aircraft wings can be time-consuming due to the high computational costs associated with flow simulations. Deep machine learning techniques can predict flow domain and desired parameters and coefficients at a significantly lower cost. However, A deep learning prediction often ignores the physical processes that form the flow. Suppose a deep learning network that is informed about the physics underlying the trained scenario can be developed. In that case, this algorithm can offer a tool that can more accurately predict the flow around objects. A model problem involving flow around an airfoil is proposed to test this idea. Lift and drag forces on this airfoil, together with the flow domain around the airfoil, are the prediction parameters. A convolutional neural network model is developed as the prediction tool. The loss function is enhanced by a new conservation of momentum-based loss function to improve the fidelity of the model. The addition of conservation of momentum improves the lift and drag predictions around the aircraft significantly. The flow database is prepared with compressible CFD runs. To improve the accuracy, the CFD solver and the loss function calculation utilize the same flux function. The predictions after the training showed improved lift and drag estimations compared to simple loss functions in a fraction of the time and cost of CFD calculations.

Subject Keywords

Deep Learning, Convolutional Neural Networks, HLLC Riemann Solver, Flow Prediction, Preliminary Design

URI

https://hdl.handle.net/11511/107794

Collections

Graduate School of Natural and Applied Sciences, Thesis