Prandtl

CFD surrogate modeling toolkit. Train fast aerodynamic surrogates — scikit-learn-like API.

import prandtl as pr

# Sample + learn + predict
X, Y = pr.sample(pr.analytical.cl_flat_plate, bounds=[(-5, 15), (0.01, 0.1)], n=100)
surrogate = pr.Surrogate(params=["alpha", "camber"], outputs=["CL"], method="gp")
surrogate.fit(X, Y)
Y_pred = surrogate.predict(X_test)

# Validate
from prandtl import metrics, cross_validate, learning_curve, residual_analysis

The Problem

CFD simulation: 40 minutes per run.
You need 100+ parameter combinations → 66 hours.

Prandtl: Learn from 100 runs → predict the rest in milliseconds, error < 0.2%.

Key Features

- **四大后端** Gaussian Process (GPyTorch) 适合小数据 + 不确定性。 MLP (PyTorch) 适合大规模 + ONNX 导出。 Random Forest / Gradient Boosting (scikit-learn) — 无需 PyTorch。 - **不确定性量化** `predict_with_uncertainty()` — GP 解析方差，RF 树集成方差。 - **验证套件** 交叉验证、学习曲线、残差分析 — 量化代理模型是否真正有效。 - **物理约束** 单调性、凸性、边界值、Sobolev 梯度约束 — 将领域知识直接注入训练。 - **主动学习** `ActiveLearner` — 最大标准差/随机采样策略，智能选择下一个采样点。 - **Co-Kriging 多保真** `CoKriging` — 结合廉价 + 昂贵仿真数据，构建多保真度代理模型。 - **GPU/CUDA 加速** `device='cuda'` 标志 — MLP 后端直接在 GPU 上训练。 - **ONNX 导出** 将训练好的代理模型部署到任何地方：边缘设备、实时控制环、云端。 - **CFD I/O** 一行代码解析 OpenFOAM forces 和 SU2 history — 从求解器输出到训练就绪。

Quick Example

import prandtl as pr
import numpy as np

# 1. Sample the parameter space
X, Y = pr.sample(
    pr.analytical.cl_flat_plate,
    bounds=[(-5, 15), (0.01, 0.1)],
    n=100, method="lhs", seed=42
)

# 2. Train a surrogate
surrogate = pr.Surrogate(
    params=["alpha", "camber"],
    outputs=["CL"],
    method="gp"
)
surrogate.fit(X, Y)

# 3. Predict on new points
X_test, Y_test = pr.sample(
    pr.analytical.cl_flat_plate,
    bounds=[(-5, 15), (0.01, 0.1)],
    n=30, seed=99
)
Y_pred = surrogate.predict(X_test)

# 4. Evaluate
report = pr.metrics(Y_test, Y_pred)
print(report)  # R² > 0.999 on smooth analytical functions

:octicons-arrow-right-24: Get Started{ .md-button .md-button--primary } :octicons-mark-github-24: GitHub{ .md-button }