Machine Learning Models for Propeller Mass, Power, and Efficiency Prediction from Basic Flight Parameters Using BEM-based Dataset
August 15, 2026
This repository contains trained machine learning models and the synthetic dataset.
The models predict propeller power, efficiency, and mass directly from basic
inputs:
- Diameter (m)
- Thrust (N)
- Speed (m/s)
- Blade count (2, 3, 4)
- Material (Carbon, GFRC, Composite)
- Altitude (m)
No detailed propeller geometry (chord, twist, airfoil) is required.
---------------------------------------------------------------------
CONTENTS of the .rar
---------------------------------------------------------------------
1) Data:
propeller_BEMT_5000_datapoints.csv
5,000 synthetic propeller designs generated using a validated BEM model.
Columns: diameter, thrust, speed, blades, material, altitude, power,
efficiency, mass, etc.
2) Models:
model_power.pkl - Random Forest for power prediction (R² = 0.994)
model_eff.pkl - Random Forest for efficiency prediction (R² = 0.899)
model_mass.pkl - Random Forest for mass prediction (R² = 0.935)
scaler.pkl - StandardScaler for numeric features
encoder.pkl - OneHotEncoder for material
feature_names.json - List of features in the correct order
3) README.md
---------------------------------------------------------------------
MODEL PERFORMANCE
---------------------------------------------------------------------
On synthetic test set (20% hold out):
Target R² MAPE
Power 0.994 8.3%
Efficiency 0.899 6.4%
Mass 0.935 13.9%
The models are most reliable in the mid‑range operating envelope (diameters 15–60 in, thrust 50–5000 N, speed 30–100 m/s), with errors increasing significantly at the extremes (diameters <10 in, thrust <10 N, speed <10 m/s).
---------------------------------------------------------------------
REQUIREMENTS
---------------------------------------------------------------------
Python 3.8+
numpy
pandas
scikit-learn
joblib
Install with:
pip install numpy pandas scikit-learn joblib
---------------------------------------------------------------------
Please enjoy!
The models predict propeller power, efficiency, and mass directly from basic
inputs:
- Diameter (m)
- Thrust (N)
- Speed (m/s)
- Blade count (2, 3, 4)
- Material (Carbon, GFRC, Composite)
- Altitude (m)
No detailed propeller geometry (chord, twist, airfoil) is required.
---------------------------------------------------------------------
CONTENTS of the .rar
---------------------------------------------------------------------
1) Data:
propeller_BEMT_5000_datapoints.csv
5,000 synthetic propeller designs generated using a validated BEM model.
Columns: diameter, thrust, speed, blades, material, altitude, power,
efficiency, mass, etc.
2) Models:
model_power.pkl - Random Forest for power prediction (R² = 0.994)
model_eff.pkl - Random Forest for efficiency prediction (R² = 0.899)
model_mass.pkl - Random Forest for mass prediction (R² = 0.935)
scaler.pkl - StandardScaler for numeric features
encoder.pkl - OneHotEncoder for material
feature_names.json - List of features in the correct order
3) README.md
---------------------------------------------------------------------
MODEL PERFORMANCE
---------------------------------------------------------------------
On synthetic test set (20% hold out):
Target R² MAPE
Power 0.994 8.3%
Efficiency 0.899 6.4%
Mass 0.935 13.9%
The models are most reliable in the mid‑range operating envelope (diameters 15–60 in, thrust 50–5000 N, speed 30–100 m/s), with errors increasing significantly at the extremes (diameters <10 in, thrust <10 N, speed <10 m/s).
---------------------------------------------------------------------
REQUIREMENTS
---------------------------------------------------------------------
Python 3.8+
numpy
pandas
scikit-learn
joblib
Install with:
pip install numpy pandas scikit-learn joblib
---------------------------------------------------------------------
Please enjoy!
