DRLFoil (Deep Reinforcement Learning Foil) is a library developed to provide an optimization framework where you can specify lift coefficient and dimensional constraints. The project began development in the context of a Final Degree Thesis, and it is currently in the alpha stage, so all contributions and feedback are appreciated!
The key features of DRLFoil are:
-
High-Speed Optimization: The models can find the optimal airfoil not in seconds or tenths of seconds, but in milliseconds.
-
Low Resource Requirements: As the neural networks used are very small, the models can be operated on any computer and even embedded systems (with appropriate adaptations).
-
High Flexibility with Constraints: The models are trained to accommodate box constraints, cl targets, and a wide range of Reynolds numbers.
DRLFoil is currently under development, as is the documentation. If you would like to contribute to its development, please contact me or submit a pull request. The first release is coming soon!
DRLFoil works with three main libraries:
-
Neuralfoil: This library works with Physics-Informed Neural Networks trained with XFoil, making the iterations significantly faster than with the aforementioned software.
-
Stable Baselines 3: This is one of the most popular Reinforcement Learning libraries, providing the most important algorithms and tools for developing models.
-
Gymnasium: An OpenAI framework that simplifies the creation of environments for deployment. It is fully compatible with SB3.
As it was developed by an aerospace engineer for other engineers, one of the main objectives was to make it easy to use, removing unnecessary steps and making tools as accessible as possible. Optimizing an airfoil is as easy as writing a few lines of code:
import drlfoil
from drlfoil import BoxRestriction
optimization = drlfoil.Optimize('onebox', cl = 0.8, reynolds = 1e7, boxes=[BoxRestriction(0.4, 0.0, 0.4, 0.15)])
optimization.run()Examples can be found in the tutorials folder.
It is highly recommended to use Python 3.10.11 as it was the version used during development. However, subsequent versions of Python should not be incompatible with the libraries used.
If you want to install the requirements, here is the command:
pip install git+https://github.com/DRL-Geometry-Optimization/DRLFoilIf you want to install requirements, the command is the following:
pip install -r requirements.txt