Optimization-Embedded Active Multi-Fidelity Surrogate Learning for Multi-Condition Airfoil Shape Optimization

Active multi-fidelity surrogate modeling is developed for multi-condition airfoil shape optimization to reduce high-fidelity CFD cost while retaining RANS-level accuracy. The framework couples a low-fidelity-informed Gaussian process regression transfer model with uncertainty-triggered sampling and a synchronized elitism rule embedded in a hybrid genetic algorithm. Low-fidelity XFOIL evaluations provide inexpensive features, while sparse RANS simulations are adaptively allocated when predictive uncertainty exceeds a threshold; elite candidates are mandatorily validated at high fidelity, and the population is re-evaluated to prevent evolutionary selection based on outdated fitness values produced by earlier surrogate states. The method is demonstrated for a two-point problem at $Re=6\times10^6$ with cruise at $α=2^\circ$ (maximize $E=L/D$) and take-off at $α=10^\circ$ (maximize $C_L$) using a 12-parameter CST representation. Independent multi-fidelity surrogates per flight condition enable decoupled refinement. The optimized design improves cruise efficiency by 41.05% and take-off lift by 20.75% relative to the best first-generation individual. Over the full campaign, only 14.78% (cruise) and 9.5% (take-off) of evaluated individuals require RANS, indicating a substantial reduction in high-fidelity usage while maintaining consistent multi-point performance.

💡 Research Summary

The paper introduces an optimization‑embedded active multi‑fidelity surrogate learning framework for simultaneous multi‑condition airfoil shape optimization. The authors address the prohibitive cost of high‑fidelity Reynolds‑averaged Navier‑Stokes (RANS) simulations by coupling inexpensive low‑fidelity panel‑method evaluations (XFOIL) with a data‑driven transfer model based on Gaussian Process (GP) regression. In this transfer model, the high‑fidelity response is learned as a non‑parametric function of both the design variables and the corresponding low‑fidelity outputs, avoiding the restrictive linear scaling assumptions of classic co‑Kriging or additive/multiplicative correction models.

A key novelty is the uncertainty‑triggered high‑fidelity sampling strategy. The GP provides a predictive variance that quantifies local model confidence. Whenever this variance exceeds a pre‑defined threshold, a costly RANS simulation is launched for that candidate design, and the new data point is used to update the GP. This active learning loop ensures that high‑fidelity evaluations are performed only where the surrogate is unreliable, dramatically reducing the total number of RANS runs.

The surrogate framework is embedded within a hybrid global evolutionary optimizer (HyGO). An elite‑consistency mechanism forces elite individuals to be validated at high fidelity each generation; after validation the entire population’s fitness values are recomputed with the updated surrogate. This prevents “fitness drift” that can arise when the optimizer continues to select based on outdated surrogate predictions.

To handle multiple operating points, the authors construct independent multi‑fidelity surrogates for each flight condition (cruise at α = 2° maximizing L/D, and take‑off at α = 10° maximizing C_L). Decoupling the surrogates allows condition‑specific uncertainty structures and refinement schedules, which is crucial because the correlation between low‑ and high‑fidelity models can vary significantly across regimes.

The airfoil geometry is parameterized using a 12‑parameter Class‑Shape‑Transformation (CST) representation (six coefficients for the upper surface, six for the lower). Parameter bounds are derived from a broad database of legacy airfoils and expanded by 15 % to allow exploration beyond known families. Geometric feasibility constraints (thickness limits, trailing‑edge angle, monotonicity, etc.) are enforced via a death‑penalty approach.

The optimization problem is tackled at Reynolds number 6 × 10⁶. Over the full campaign, only 14.78 % of cruise candidates and 9.5 % of take‑off candidates required RANS evaluations, representing an ≈85 % reduction in high‑fidelity cost compared with conventional approaches. The final design improves cruise efficiency by 41.05 % and take‑off lift by 20.75 % relative to the best individual from the initial generation. Convergence analysis shows the GP’s mean prediction error dropping from ~5 % initially to below 1 % as the campaign progresses, with high‑fidelity calls concentrating around promising regions.

In summary, the work contributes three main advances: (1) an active multi‑fidelity strategy where uncertainty directly governs high‑fidelity sampling, (2) condition‑wise decoupled surrogate architectures for multi‑point aerodynamic design, and (3) a comprehensive demonstration on a two‑condition CST‑based airfoil problem, including detailed cost and convergence analyses. The authors suggest future extensions to incorporate additional fidelity levels (e.g., LES, DNS), experimental data, and three‑dimensional wing optimization.