Aerodynamic inverse design using multifidelity models and manifold mapping

By Xiaosong Du, J. Ren, Leifur Þ. Leifsson. Aerospace Science and Technology, 2019

46 citations recorded by the source corpus.

Abstract Aerodynamic inverse design is proposed using multifidelity models and the manifold mapping (MM) technique. Aerodynamic inverse design aims at achieving a target performance characteristic, such as a pressure coefficient distribution of an airfoil or local lift distribution of a wing. Due to the high computational cost of accurate aerodynamic models and the large number of design variables, the overall cost of inverse design can be prohibitive. The MM-based optimization algorithm leverages the speed of the low-fidelity model to accelerate the optimization process, but refers back to the high-fidelity model to ensure an accurate solution. In this work, the MM technique is applied to the characteristic distribution under consideration in each application. In particular, the pressure coefficient distribution is modeled with the MM technique in the case of airfoil inverse design, and the sectional lift distribution in the case of wing design. The proposed method is tested and evaluated on six airfoil inverse design cases and one rectangular wing inverse design case. In the two-dimensional cases, parameterized with eight design variables, direct aerodynamic inverse design using pattern search required 700 to 1200 high-fidelity model evaluations, which took 300 to 700 hours in total. The MM-based design algorithm required less than 20 high-fidelity simulations and 1000 to 2000 low-fidelity evaluations, which took 30 to 90 hours. In the three-dimensional case, parameterized with three design variables, direct aerodynamic inverse design took around 50 hours, whereas the MM-based design needed around six hours.

Subfields: Computational Mechanics, Statistical and Nonlinear Physics, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/920a3eab69d5d276f28a88a5914272602bd85e0f

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

1. Ji-chao Li
2. Xiaosong Du
3. Joaquim R. Martins
4. Gaetan K. W. Kenway

In Statistical and Nonlinear Physics

1. Joaquim R. Martins
2. Ji-chao Li
3. Xiaosong Du

In Statistics and Probability

1. Ji-chao Li
2. Joaquim R. Martins
3. Xiaosong Du
4. Gaetan K. W. Kenway
5. Martin P. Bendsøe

Citations in the corpus

• Machine Learning in Aerodynamic Shape Optimization (2022, methodology): “Moreover, other multi-fidelity surrogate methods [111, 165, 166], such as manifold mapping, managed to accurately predict scalar and vector quantities for multilevel aerodynamic design optimizations [167, 168].”