Adaptive sampling of pressure distribution features for the airfoil statistical analysis

Runze Li, Yufei Zhang, Haixin Chen

Abstract

In the area of supercritical wing design, a variety of principles, laws and rules have been summarized by scholars who perform theoretical and experimental analyses. The applicability of these rules is usually restricted by the airfoil samples investigated. With the advance of computational fluid dynamics and computational intelligence, such work can be better conducted on computers. The present paper proposes an output space sampling method to generate airfoil samples that have specified pressure distributions or meet certain special requirements. The well-selected and distributed samples are then utilized for statistical studies to obtain more reliable or more universal aerodynamics rules that can be used as guidance in the process of supercritical airfoil design. The capabilities of the output space sampling method in regard to filling the space and exploring the boundaries are also improved. The output space sampling method is employed to generate supercritical airfoil samples with different requirements, allowing the relationships between the shock wave location and the drag divergence Mach number as well as the drag creep characteristic to be revealed.

Introduction: The Mapping Problem in Aerodynamics

Aerodynamic design often relies on "rules of thumb" derived from limited experimental or numerical datasets. However, these rules can be biased if the samples do not cover a wide enough range of possible flow conditions. The core challenge lies in a complex mapping problem: while engineers can easily vary the input space (the geometric coordinates of an airfoil, $x$), they are primarily concerned with the output space (the resulting flow features, $y$, and performance metrics, $Y$).

Mapping relationship between geometry, flow features, and performance.

Because the relationship between shape and flow is governed by non-linear Navier-Stokes equations, it is difficult to "set" a desired pressure distribution and determine the exact shape required without extensive trial and error. This research introduces a sampling method designed to efficiently "fill" the output space to uncover universal aerodynamic rules.

Methodology: Adaptive Output Space Sampling (OSS)

The proposed Adaptive Output Space Sampling (OSS) method is built on three pillars:

  1. RBF Surrogate Models: Radial Basis Function (RBF) networks approximate the high-dimensional relationship between geometry ($x$) and flow features ($y$).
  2. Target Shooting Criterion (TSC): This innovation places "targets" in the desired output space. A "potential energy" function then attracts new samples toward unexplored or sparsely populated regions.
  3. Multi-objective Optimization: Using Differential Evolution (DE), the system searches for new geometries that simultaneously improve surrogate model accuracy and hit output space targets.

For evaluation, the CFL3D open-source solver was used to perform Reynolds-Averaged Navier–Stokes (RANS) simulations, ensuring the physical accuracy of generated samples.

Datasets and Validation

Rather than using a static dataset, the study generates a custom set of supercritical airfoils:

  • Initial Set: 400 airfoils derived from baseline cruise drag optimization.
  • Generated Set: Thousands of samples were generated via OSS, specifically targeting variations in shock wave location ($X_1$) and suction peak ($M_{w,L}$).
  • Validation: CFD methods were validated against classic RAE2822 experimental data.

Key Insights and Results

Sampling the "Result," not the "Cause"

Traditional sampling fills the input space (geometry), often leading to redundant results in the output space (performance). By sampling the output space directly, the researchers found a more diverse array of physically meaningful designs. The OSS-2 method achieved a diversity measure of 0.71, significantly higher than standard Latin Hypercube Sampling (0.40).

Refining Korn’s Equation

Korn’s equation is used to estimate the technology level ($\kappa$) of an airfoil. The authors discovered that the shock wave location ($X_1$) is a critical missing variable. They proposed an augmented version:

$$M_{\infty,DD} + 0.1C_L + (t/c)_{max} + 0.065X_1 = 0.97$$

This formula accounts for the shock wave location, providing a more universal fit ($R^2 = 0.932$) for supercritical airfoils.

The Drag Creep Limit

A clear threshold for "drag creep" (the slow increase in drag before divergence) was identified. If the suction peak ($M_{w,L}$) exceeds 1.25, the drag creep characteristic deteriorates rapidly, providing a concrete constraint for future optimization.

Conclusion and Future Work

While the OSS method is efficient, it still requires running a CFD solver for every new sample. Future work may involve integrating faster, machine learning-based solvers to reduce computational costs. Additionally, extending this approach to 3D wings will involve more complex flow features like spanwise flow and tip vortices. Identifying the "right" output features currently requires aerodynamic expertise; automated feature discovery remains a promising area for future research.

Summary updated

Abstract

In the area of supercritical wing design, a variety of principles, laws and rules have been summarized by scholars who perform theoretical and experimental analyses. The applicability of these rules is usually restricted by the airfoil samples investigated. With the advance of computational fluid dynamics and computational intelligence, such work can be better conducted on computers. The present paper proposes an output space sampling method to generate airfoil samples that have specified pressure distributions or meet certain special requirements. The well-selected and distributed samples are then utilized for statistical studies to obtain more reliable or more universal aerodynamics rules that can be used as guidance in the process of supercritical airfoil design. The capabilities of the output space sampling method in regard to filling the space and exploring the boundaries are also improved. The output space sampling method is employed to generate supercritical airfoil samples with different requirements, allowing the relationships between the shock wave location and the drag divergence Mach number as well as the drag creep characteristic to be revealed.
BibTeX
@inproceedings{li2021adaptive,
  author = {Runze Li and Yufei Zhang and Haixin Chen},
  title = {{Adaptive sampling of pressure distribution features for the airfoil statistical analysis}},
  booktitle = {Chinese},
  year = {2021},
  doi = {10.1016/j.cja.2021.10.028},
}

Loading related content...

Figure