Deliver Diverse Candidate Designs That Balance Every Metric

In multi-objective engineering optimization, EvoX does not produce a single "optimal solution", but rather a set ofdiverse candidate designs that balance every metric: the system continuously generates and predicts from a base airfoil combined with perturbation samples, explicitly preserving the different trade-offs among conflicting metrics such as the critical buffet lift coefficient and the cruise lift-to-drag ratio, and forming a set of candidates that can be compared and used for decision-making.

Why the Output Is a Set of Candidate Designs, Not a Single Answer

When objectives conflict with one another—for example, raising the critical buffet lift coefficient often comes with a lower cruise lift-to-drag ratio—there is no feasible solution that dominates on every metric at once. If the evaluation mechanism is dominated by a single metric, the search quickly converges to a compromise that "looks decent on average but stands out on no key metric", and may even discard deliverable design directions with a clear preference. EvoX instead incorporates multiple constraints and objectives into the same evaluation space and explicitly preserves diverse designs, turning the trade-offs themselves into information that can be presented to decision-makers.

Continuous Generation and Prediction: Using High-Fidelity Simulation Only Where It Matters

• Evaluation mechanism that balances multiple metrics: incorporates multiple constraints and objectives—such as the critical buffet lift coefficient and the cruise lift-to-drag ratio—into the same evaluation space, preventing screening from being dominated by a single metric.
• Continuous generation and prediction: continuously generates new candidates around a base airfoil and perturbation samples, using surrogate models to rapidly estimate metrics and alleviate the bottleneck of expensive high-fidelity simulation.
• Diverse output format: the output is a set of candidate designs covering different metric preferences and trade-off positions, rather than a single average-optimal or single-point answer.
• Comparability: each candidate retains comparable multi-dimensional metric results and its generation perturbation source, making it easy to reproduce, evaluate, and select.

From Scatter Points to Frontier: How Candidates Compare

Plotting two key metrics of each candidate as scatter points, those on the outer edge that are not simultaneously exceeded on every metric by any other candidate constitute thePareto frontier; points inside the frontier are dominated. Engineering reviews typically pick designs from the frontier first, then return to high-fidelity simulation for confirmation; the differences between points along the frontier are precisely the concrete manifestation of engineering orientations such as "buffet suppression first" or "cruise efficiency first".