Base Airfoil and Continuous Generation of Perturbation Samples

In airfoil optimization,the base airfoilserves as a stable reference that satisfies the task constraints, anchoring the parameterization, perturbation range, and evaluation criteria. Without a baseline, the search space tends to drift without order, and candidate airfoils become difficult to compare fairly.

Around the base airfoil, applycontrolled perturbationsto the thickness distribution, camber, or control point parameters, continuously generating perturbation sample families in batches or online. Local fine-grained search and global diversity coverage must be balanced simultaneously, so the perturbation intensity, direction, and sampling distribution should remain controllable.

Each round, samples are fed in batches into theprediction/evaluationstage, which outputs the performance of multiple candidates on the objectives. These results then enterevolutionary selection, update, and resampling, which in turn adjusts the perturbation generation for the next round.

This forms a closed loop ofbaseline—perturbation—prediction/evaluation—evolutionary feedback—continuous generation: the base airfoil provides a stable anchor, perturbation samples provide diversity, and evolutionary feedback makes multi-objective exploration sustainable and scalable.

Four Control Points in the Loop

Base airfoil selection: First confirm the baseline satisfies the task constraints, then define the parameterization space and perturbation range to avoid unregulated drift in the search space.
Continuous perturbation sample generation: Expand samples in batches or online—small local perturbations for fine search, large perturbations for global coverage.
Combining multi-sample prediction with evolution: Batch evaluation outputs the performance of multiple candidates, serving as the basis for evolutionary selection, update, and resampling.
Continuous closed loop: Adjust perturbation intensity, direction, or sampling distribution based on evolutionary feedback, so that the base airfoil and perturbation samples jointly support stable, scalable multi-objective exploration.

Engineering Tip

If each round applies only small perturbations in the same direction, the population tends to converge prematurely; if perturbations are entirely random and large, the evaluation budget may be wasted. A more reliable strategy is to mix local and global perturbations and use feedback to determine the focus of the next sampling round.

Airfoil Perturbation Generation Workflow

The diagram uses the base airfoil as a stable anchor, applying controlled perturbations to the thickness, camber, or control point parameters to continuously generate the sample families needed for local fine search and global coverage. Samples are fed in batches into prediction/evaluation and evolutionary selection, and then feedback is used to adjust the intensity and direction of the next round of perturbations, forming a closed loop of baseline, perturbation, evaluation, and evolutionary feedback.

Question 1

In the "Base Airfoil and Continuous Generation of Perturbation Samples" loop, what is the main role of the base airfoil?

It serves as a stable anchor for the parameterization, perturbation range, and evaluation criteria

It is only used for visualization and does not affect the search

It provides the single optimal solution, avoiding any perturbation

It replaces the evolutionary algorithm and directly outputs the final airfoil

Question 2

When continuously generating perturbation samples, what is a more reasonable approach to balance local fine search and global diversity coverage?

Generate samples only in the first round and never update afterward

Always apply only small perturbations in the same direction

Apply entirely random large perturbations with no baseline

Adjust perturbation intensity, direction, and sampling distribution in batches or online

Question 3

After perturbation samples are fed in batches into the prediction/evaluation stage, what is the main output?

It outputs only the training loss and provides no basis for selection

The performance of multiple candidates on the objectives, used for evolutionary selection, update, and resampling

It outputs only a single hard-coded label

It directly replaces the geometry of the base airfoil

Question 4

How does evolutionary feedback support "continuous" generation in the loop?

Use feedback to adjust perturbation intensity, direction, or sampling distribution, then supplement the next round of samples

Feedback is used to delete the base airfoil

Feedback requires stopping all sample generation

Feedback only logs results and does not change sampling

Question 5

If optimization starts from a supercritical base airfoil, which of the following best reflects the "baseline—perturbation—prediction/evaluation—evolutionary feedback" loop?

Randomly generate many airfoils but perform no evaluation or selection

Use a single preset airfoil to cover all conditions without generating samples

Apply a single small perturbation to the thickness distribution and adopt it directly

Each round generates a family of airfoil samples, which enters evolution after evaluation/prediction, and results are used to supplement the next round of perturbation samples

Case Study: Continuous Perturbation Optimization of a Supercritical Base Airfoil

Read the scenario and answer the questions.

The team has selected a supercritical base airfoil that satisfies the cruise lift-to-drag ratio and volume constraints, and wants to build a continuous perturbation sample generation mechanism around it and use evolutionary computation for multi-objective optimization. The objectives include maximizing the lift-to-drag ratio, minimizing drag divergence, and maintaining the structural thickness constraint. Currently, 256 airfoil samples can be generated per round, and evaluation is time-consuming.

1. Design one round of the "base airfoil—perturbation samples—prediction/evaluation—evolutionary feedback" process in the loop, and explain why the base airfoil cannot be replaced at will.

Answer:
Process: Using the base airfoil as an anchor, apply controlled perturbations to the thickness distribution, camber, or control point parameters to generate a batch of airfoil samples; feed the samples in batches into prediction/evaluation to obtain performance measures such as lift-to-drag ratio, drag divergence, and thickness constraints; then proceed to evolutionary selection, update, and resampling, and adjust the next round's perturbation intensity, direction, or sampling distribution based on feedback. The base airfoil cannot be replaced at will because it defines the parameterization space, perturbation range, and evaluation criteria; replacing it would cause the search space to drift, make candidates incomparable, and cause the loop to lose its stable reference.

2. If evolution stalls for several consecutive rounds, how would you use evolutionary feedback to adjust perturbation generation?

Answer:
You could increase perturbation intensity or expand the sampling distribution to introduce more global diversity; adjust the perturbation direction to conduct targeted exploration of sensitive parameters such as thickness and camber; or mix local fine perturbations with global perturbations and increase sample batches or online supplementation to avoid premature convergence of the population. If objectives conflict, retain diverse candidates on the Pareto front and use feedback to determine the focus of the next sampling round.