Improving the efficiency of airfoil design is key to the creation of more fuel-efficient airplanes, reducing their carbon footprints. Unfortunately, currently component optimization is inaccessible due to its reliance on iterative Computational Fluid Dynamics (CFD), which is expensive, computationally costly, and difficult to use, resulting in inefficient designs. Rapid advancements in machine learning techniques open new avenues to streamline this process. We investigated the simulation runtime and accuracy of generating airfoil designs through the construction of a series of machine learning models. Models utilizing ensemble methods, and K-Nearest Neighbors Bagging, as well as the multilayer perceptron, produce highly accurate and efficient predictions for familiar data, though error increases substantially when assessed with completely unseen data.
Our Projectboard.
Grace Yu and Akira Yoshiyama
Improving the efficiency of airfoil design is key to the creation of more fuel-efficient airplanes, reducing their carbon footprints. Unfortunately, currently component optimization is inaccessible due to its reliance on iterative Computational Fluid Dynamics (CFD), which is expensive, computationally costly, and difficult to use, resulting in inefficient designs. Rapid advancements in machine learning techniques open new avenues to streamline this process. We investigated the simulation runtime and accuracy of generating airfoil designs through the construction of a series of machine learning models. Models utilizing ensemble methods, and K-Nearest Neighbors Bagging, as well as the multilayer perceptron, produce highly accurate and efficient predictions for familiar data, though error increases substantially when assessed with completely unseen data.
Our Projectboard.