Sunwoong Yang
About Me 본문
About Me
Ph.D. candidate at Seoul National University in South Korea
Majoring in Aerospace Engineering
Interested in aerodynamic vehicle design using machine learning/deep learning
For more details, please click below icons:
Education
Ph.D. candidate - Seoul National University (Mar. 2020 ~ Present)
Department of Aerospace Engineering
Advisor: Prof. Kwanjung Yee
Master - Seoul National University (Mar. 2018 ~ Feb. 2020)
Department of Mechanical and Aerospace Engineering
Thesis: Planform Optimization of Unmanned Combat Aerial Vehicle Considering Longitudinal Stability and Low-observability Using Variable-fidelity
Advisor: Prof. Kwanjung Yee
Bachelor - Seoul National University (Mar. 2014 ~ Feb. 2018)
Department of Mechanical and Aerospace Engineering
Research Areas
Computational fluid dynamics (CFD)
- Performance improvement of unmanned combat aerial vehicle through CFD analysis
- Automation of the pre-processing & processing & post-processing for CFD-coupled optimization
Shape optimization
- Design shape optimization of various aerodynamic vehicles
- Multi-objective optimization for Pareto solutions
Bayesian optimization
- Shape optimization considering surrogate model uncertainty
Data mining
- Design rule extraction through various univariate/multivariate data mining techniques (e.g., analysis of variance, self-organizing maps, decision tree, rough set theory)
Surrogate modeling
- Efficient performance prediction through various surrogate models (e.g., Gaussian process, neural networks)
- Optimization leveraging surrogate models
Multi-fidelity modeling (or data fusion)
- Efficient surrogate modeling (e.g., hierarchical kriging, multi-fidelity deep neural networks) using various data with different fidelity (i.e., low-fidelity data from coarse mesh + high-fidelity data from dense mesh)
- Noise reduction technique considering characteristics of multi-fidelity data
- Optimization leveraging multi-fidelity surrogate models
Inverse design
- Efficient shape design with predefined target performance distribution
- Proposal of novel inverse design optimization framework through two-step deep learning approach (multi-layer perceptron and variational autoencoder)
Generative modeling
- Generative modeling in airfoil inverse design process via variational autoencoder
- Generative modeling in the reconstruction of high-dimensional data via β-variational autoencoder with convolutional layers
- Leveraging generative modeling for reduced-order modeling
Reduced-order modeling (ROM)
- Proposal of an efficient framework for predicting high-dimensional data via β-variational autoencoder with convolutional layers
- Comparison of various dimensionality reduction techniques (principal component analysis & autoencoder & variational autoencoder & β-variational autoencoder)
Feature extraction
- Extraction of interpretable physical parameters from the given image dataset via β-variational autoencoder
Uncertainty quantification (UQ)
- UQ using Gaussian process and its extension to Bayesian optimization
- UQ using neural networks considering engineering applicability: deep ensembles
- Calibration of uncertainty estimated by deep ensembles
- Comparison of Gaussian process and deep ensembles considering prediction accuracy & estimated uncertainty accuracy & computational efficiency