Session: DAC-02-01-Artificial Intelligence and Machine Learning for Challenging Real-World Problems in Design Automation

Paper Number: 90068

90068 - Effect of Optimal Geometries and Performance Parameters on Airfoil Latent Space Dimension 

Although learning low dimensional airfoil manifolds can facilitate aerodynamic optimizations, the properties of these latent spaces are not well understood. This paper investigates airfoil manifolds to provide greater insight into the effects of optimized geometry and data set features on latent spaces. Specifically, we investigate if optimized geometries occupy lower dimensional manifolds than non-optimized geometries. We additionally examine the effect of including target optimization conditions as data set features for a range of latent space sizes. We explore these areas using the UIUC airfoil database and a subset of these airfoils optimized with CBGAN and CEBGAN generative models. Lower dimensional airfoil manifolds are learned using both autoencoders and Principal Component Analysis (PCA) models. The performance of these models are also compared to each other in ranges of training sample sizes and latent dimension size using MSE between the original testing samples and the reprojected data constructed from the models as a metric. The results of this study suggest that optimized geometry does not always lie in a lower dimensional latent space as the two data sets were observed to have similar intrinsic dimensionalities. This study also demonstrates that including input parameters used in airfoil coordinate generation as data set features does not necessarily decrease the latent space dimensionality. 

Presenting Author: Alec Van Slooten University of Maryland

Presenting Author Biography: Alec Van Slooten received his B.S. in Mechanical Engineering from the University of Connecticut. He is a current Ph.D. student at the University of Maryland, College Park. His current research interests include machine learning, optimization, and design.

Authors:

Alec Van Slooten University of Maryland
Mark Fuge University of Maryland