CFD based aerodynamic design optimization using Bayesian inference and Kriging surrogate model

Date

2024-1-18

Author

Sunay, Yunus Emre

Metadata

Show full item record

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

Item Usage Stats

172
views

464
downloads

Cite This

The advancement of computer hardware technology significantly contributes to resolving complex engineering challenges that require solution of complex, highly nonlinear equations. The fidelity of these analyses is closely tied to the capabilities of computing systems and improves with advancements in computing power. However, even with high-fidelity numerical simulators, obtaining a solution can be time-consuming, often requiring several hours. This study aims to integrate high- fidelity analytical tools into the design cycle and thereby to reduce the overall duration of the design process through enhanced computational power. An optimization tool based on Bayesian inference process is developed to find the optimum design. In this study, deterministic numerical simulations are employed to construct a Kriging surrogate model, which represents a stochastic relationship between inputs and outputs, independent of the underlying physics of the simulations. Computational Fluid Dynamics, an expensive high-fidelity tool, is replaced by the Kriging surrogate model in Bayesian optimization. Latin Hypercube Sampling, which has randomness and homogeneity, is used to initialize the samples in the design space; however, it is not possible to increase the number of samples for iterative processes. Furthermore, this study investigates a hybrid methodology for parallel incremental sampling, which uses randomness and even distribution of LHS. This method is applicable for any initial quantity of samples to target number of samples. The optimization tool, which contains hybrid parallel algorithms for sampling, is tested with well-known Rosenbrock function and applied for shape optimization of delta wing using meshless CFD.

Subject Keywords

Kriging surrogate model, Bayesian optimization, Meshless computational fluid dynamics, Vortex particle method, Parallel design strategies

URI

https://hdl.handle.net/11511/109171

Collections

Graduate School of Natural and Applied Sciences, Thesis