aviation-challenge

Machine learning algorithm utilized to aid aviation challenge team, taking a few test runs as training and extrapolating that data to find candidates for the best combination of airplane parameters to achieve the best score for the simulation.

Project Structure

• data.ipynb - Data preprocessing and cleaning pipeline
• ml.ipynb - Machine learning model training and evaluation
• test.ipynb - Monte Carlo optimization to find best aircraft configurations
• y_test.csv - Actual scores from test set
• y_pred.csv - Predicted scores from trained model
• formatted_data.csv - Cleaned training data (generated)
• results.csv - Optimized aircraft configurations (generated)
• model.pkl - Trained Random Forest model (generated)

Aircraft Parameters

The system optimizes the following aircraft design parameters:

Engine & Configuration

• Horsepower (400-1000 HP in 25 HP increments)
• Extra Propeller (0 or 1)
• Retractable Landing Gear (0 or 1)

Weight Parameters

• Empty Weight (calculated based on other parameters)
• Payload Weight (1700-6000 lbs in 300 lb increments)
• Fuel Load (fixed at 300 lbs for 40 min flight)
• Total Weight (must not exceed 8000 lbs)

Wing Design

• Wing Semi-Length (25-60 ft in 5 ft increments)
• Wing Root Chord (2.0-6.5 ft in 0.3 ft increments)
• Wing Tip Chord (2.0-6.5 ft in 0.3 ft increments)
• Wing Area (calculated)
• Aspect Ratio (calculated)

Performance Metrics

• Wing Loading (Total Weight / Wing Area)
• Power Loading (Total Weight / Horsepower)

Methodology

1. Data Preprocessing (data.ipynb)

• Loads raw CSV data with aircraft configurations and scores
• Removes unnamed columns and invalid entries
• Filters for valid score entries only
• Calculates derived parameters (wing area, aspect ratio, etc.)

2. Model Training (ml.ipynb)

• Uses Random Forest Regressor with hyperparameter optimization
• Performs 5-fold cross-validation with 100 random search iterations
• Achieves ~81% R² score on test set
• Average prediction error: ~8.3%
• Maximum prediction error: ~16.1%

3. Optimization (test.ipynb)

• Runs 1,000,000 Monte Carlo simulations
• Tests random combinations within parameter constraints
• Filters results to top 20% of maximum observed score
• Returns ranked list of optimal configurations

Usage

Requirements

pandas
scikit-learn
numpy
scipy

Running the Pipeline

1. Prepare Data:

   jupyter notebook data.ipynb

   Ensure your raw data is in data.csv format.

2. Train Model:

   jupyter notebook ml.ipynb

   This will generate model.pkl and performance metrics.

3. Find Optimal Configurations:

   jupyter notebook test.ipynb

   This generates results.csv with top-performing aircraft configurations.

Model Performance

The trained Random Forest model achieves:

• R² Score: 0.809 (81% variance explained)
• Mean Absolute Percentage Error: 8.28%
• Maximum Percentage Error: 16.08%

Results

The optimization process identifies aircraft configurations achieving scores ≥80% of the maximum observed score. Results are sorted by predicted score and include all design parameters plus calculated performance metrics.

Key Features

• Automated parameter calculation: Derives wing area, aspect ratio, empty weight, and performance metrics
• Constraint enforcement: Ensures total weight ≤ 8000 lbs
• Monte Carlo optimization: Explores vast parameter space efficiently
• Performance filtering: Returns only high-scoring configurations
• Reproducible pipeline: End-to-end workflow from raw data to optimized designs

Applications

This tool assists aviation challenge teams by:

• Reducing the number of physical test flights needed
• Identifying promising parameter combinations quickly
• Providing quantitative guidance for design decisions
• Enabling rapid iteration on aircraft configurations

Notes

• The model assumes a fixed longitudinal arm of 13 ft for all configurations
• Fuel load is standardized at 300 lbs (40-minute flight capacity)
• Wing lateral arm is fixed at 2 ft
• Sweep angle is set to 0° for all configurations
• Empty weight calculation includes adjustments for extra propeller and retractable landing gear