peterdsharpe/NeuralFoil

NeuralFoil is a practical airfoil aerodynamics analysis tool using physics-informed machine learning, exposed to end-users in pure Python/NumPy.

What it solves

NeuralFoil provides a rapid, reliable method for analyzing airfoil aerodynamics. It replaces traditional tools like XFoil, which can suffer from non-convergence issues, "jagged" predictions, and high computational costs during large-scale batch analysis or gradient-based shape optimization.

How it works

It is a hybrid of physics-informed machine learning and analytical models. The core is a neural network trained on nearly 8 million XFoil runs, using CST (Kulfan) parameterization to describe airfoil geometry. The tool is available as a standalone Python/NumPy implementation for fast runtime execution, or integrated into AeroSandbox for advanced features like compressible aerodynamics (transonic/supersonic), post-stall analysis, and control surface deflections.

Who it’s for

Aerodynamics engineers, aircraft designers, and researchers who need fast, differentiable, and robust airfoil performance predictions for design optimization, flight simulation, or real-time control systems.

Highlights

  • Extreme Speed: Up to 30x faster than XFoil for single analyses and ~1000x faster for multipoint batch analyses.
  • Robustness: Guaranteed to return an answer without convergence failures and provides $C^\infty$-continuous outputs, making it ideal for gradient-based optimization.
  • Self-Assessing Confidence: Includes an analysis_confidence metric to flag out-of-distribution queries or sensitive flow conditions.
  • Generalization: Capable of accurate predictions even for airfoils not present in the training set, including those with extreme control surface deflections.

Related

  • Project
    peterdsharpe/AeroSandbox

    A Python package for aircraft and engineered system design and optimization using automatic differentiation and differentiable physics models.

  • Project
    google/brax

    A fast, fully differentiable physics engine written in JAX for large-scale rigid body simulation and reinforcement learning training.

  • Project
    MiroMindAI/MiroFlow

    MiroFlow is an open-source research agent framework for multi-step internet research and complex reasoning tasks, such as future event prediction.

  • Project
    microsoft/FLAML

    A lightweight Python library for efficient automated machine learning and hyperparameter tuning that optimizes model performance under resource constraints.

  • Project
    ServiceNow/PipelineRL

    A scalable asynchronous reinforcement learning framework for LLM agents that uses in-flight weight updates to maximize GPU utilization while maintaining on-policy data freshness.