Abstract:In the approach of metric-based geometric parameterization, the concept of database filtration and proper orthogonal decomposition manipulation quantitatively enhances the generation of new geometric parameters so that the problem-oriented information is reflected in aerodynamic design. Compared with the conventional approach that applies only geometric filtration, this paper incorporates performance filtration so that the problem-oriented information is no longer limited to geometric constraints, thereby enabling the guidance of optimization direction in a more informed manner. This paper focuses on the pressure distribution of samples and their relationship with geometric variations. A variational autoencoder is employed to extract the dominant features of the aerodynamic potentials that an initial geometry can attain in optimization. The authors believe that the aerodynamic potential is reflected in how the pressure distribution differs from that of the initial geometry. Then, samples have to meet the performance filtration criterion based on pressure distribution proposed by designers according to their aerodynamic understanding of the optimization problem. Finally, the remaining collection of qualified samples generates new geometric parameters via proper orthogonal decomposition. The conventional and new approaches were validated in a two-dimensional airfoil case of drag coefficient optimization; the lift-to-drag ratio improved by 17.07% using the traditional method, and by 38.20% using the new approach. The pressure distribution shows that the performance filtration approach produces more supercritical properties than the geometric filtration approach. This implies that the performance filtration approach can capture more intricate details related to the aerodynamic performance of the airfoil.

A Manifold-Based Airfoil Geometric-Feature Extraction and Discrepant Data Fusion Learning Method

Learning with Geometry: Including Riemannian Geometric Features in Coefficient of Pressure Prediction on Aircraft Wings

MHA-Net : Multi-source Heterogeneous Aerodynamic Data Fusion Neural Network Embedding Reduced-dimension Features

Scaling Function Learning: A sparse aerodynamic data reconstruction method for generalizing aircraft shapes

Knowledge-embedded meta-learning model for lift coefficient prediction of airfoils

Deep Learning for Multi-Fidelity Aerodynamic Distribution Modeling from Experimental and Simulation Data

Multi-fidelity deep neural network surrogate model for aerodynamic shape optimization

Aerodynamic Optimization Framework for a Three-Dimensional Nacelle Based on Deep Manifold Learning-Assisted Geometric Multiple Dimensionality Reduction

Physically Interpretable Feature Learning of Supercritical Airfoils Based on Variational Autoencoders

Parsimonious airfoil Parameterisation: A deep learning framework with Bidirectional LSTM and Gaussian Mixture models

Geodesic Based Semi-supervised Multi-manifold Feature Extraction

Physically Interpretable Feature Learning and Inverse Design of Supercritical Airfoils

Deep Learning Based Adaptive Deformation of Aerodynamic Shape for Ducted Propellers

Multi-fidelity Metamodeling in Turbine Blade Airfoils Via Transfer Learning on Manifolds

Aerodynamic data predictions based on multi-task learning

Plane Cascade Aerodynamic Performance Prediction Based on Metric Learning for Multi-Output Gaussian Process Regression

Prediction and optimization of airfoil aerodynamic performance using deep neural network coupled Bayesian method

Relative Assessment of Selected Machine Learning Techniques for Predicting Aerodynamic Coefficients of Airfoil

Transferable Machine Learning Model for the Aerodynamic Prediction of Swept Wings

A Novel Metric-Based Geometric Parameterization Approach with Performance Filtration

Deep learning for airfoil aerodynamic-electromagnetic coupling optimization with random forest