Predicting the performance of different electric vertical take-off and landing (eVTOL) vehicle designs is crucial for manufacturers and hobbyists. The maximum flight time (endurance) and maximum flight distance (range) of these vehicles depend on various design and operational parameters such as structure, motors, propellers, battery pack, payload, and velocity. While sophisticated physics-based models can estimate and optimize these parameters, they require advanced knowledge and extensive computational resources, limiting their widespread use.
This project aims to develop and implement a framework that utilizes machine learning to create simpler and more accessible surrogate models. These models are trained on performance estimates generated by complex physics-based models and can accurately mimic their results. By doing so, we provide an efficient and easy-to-use tool for predicting eVTOL vehicle performance without the need for extensive computational power.
- Training Data: The training data consists of 47,000 eVTOL vehicle designs simulated using physics-based models.
- Machine Learning Models: Various models are employed, including decision trees, K-Nearest Neighbors, linear regression, and a multi-layer perceptron neural network.
- Features and Targets: The models use intuitive vehicle design and operational parameters (e.g., propeller size, payload mass, drag coefficient, motor, and battery parameters) as features, and predict endurance and range as target variables.
- Performance: The surrogate models demonstrate excellent performance, with several models achieving a mean average percentage error of less than 2% on hold-out datasets. Benefits
- Computational Efficiency: These machine learning-based surrogates are much more computationally efficient compared to physics-based models.
- Ease of Use: The models are easy to understand and apply, making them accessible to a broader audience, including those without advanced technical expertise.
The project is divided into the following sections:
- Import Data
- Exploratory data analysis and visualizations
- Feature engineering to transform data
- Building Models and evaulating performance
List of libraries used for this project:
matplotlib==3.7.1
pycaret==3.1.0
pandas==1.5.3
seaborn==0.12.2
joblib==1.3.2
Use the package manager pip to install the dependencies from the requirements.txt
pip install -r requirements.txt
Follow the two notebooks to recreate and evaluate the experiments
Distributed under the Apache License 2.0 License. See LICENSE for more information.