Parameterized 3D Representation of Gas Turbine Blades for Machine Learning Applications

The author proposes a parametric method of 3D representations for gas turbine blades. It is based on the use of Bernstein polynomials terms as descriptors in the design of 2D blade profiles and further automatic creation of parametric 3D shapes of blades. This method helps to reduce the number of geometric design variables used in the optimization of the aerodynamic shape of blades, thereby reducing the dimensionality of the feature space when using machine-learning models and accelerating blades design processes. Then, BladeGAN and AutoBlade machine learning models are described, which, by adding a layer of rational B-splines, provide more accurate 3D shape control when prototyping gas turbine engine blades. It is shown that the author's AutoBlade model learns twice as fast as BladeGAN, with comparable quality, which justifies the expediency of using autoencoders for optimization of parameters of gas turbine blades.