Machine learning models for leading-edge inflatable airfoil aerodynamics

This repository contains the trained machine learning models for predicting the aerodynamic performance of leading-edge inflatable (LEI) kite profiles, an output of the MSc thesis of K.R.G. Masure. The models are based on Extra Trees regression and allow fast estimation of lift, drag, and moment coefficients directly from geometric parameters. The models were trained on a large set of two-dimensional CFD results (included in a separate Zenodo dataset) covering a range of profile geometries, angles of attack, and Reynolds numbers. Each model file corresponds to a specific Reynolds number: ET_re1e6.pkl – Reynolds number 1×10⁶ ET_re5e6.pkl – Reynolds number 5×10⁶ ET_re2e7.pkl – Reynolds number 2×10⁷ Predictions are nearly instantaneous once the models are loaded, making them suitable for rapid aerodynamic evaluations and for use as input in kite simulation frameworks, such as the vortex step method (VSM). Two Python files are included that (a) create an airfoil geometry from a set of parameters and (b) run one of the machine learning models using the defined set of parameters and plotting the lift, drag, and moment coefficient as a function of alpha.To run the Python files, two dependencies have to be installed: "numpy" and "matplotlib". Key features: Geometry-based input using non-dimensional parameters (thickness, camber position, camber height, reflex angle, camber tension, leading edge curvature) Fast and accurate aerodynamic predictions (R² > 0.98 across all Reynolds numbers) An open-accessible code base used to develop and train the models and the CFD training data itself..