Advantages of Machine Learning Methods in Aerodynamic Blade Optimization

Abstract The present paper describes the advantages of using Machine Learning methods within the aerodynamic optimization of blades, highlighting the benefits of such techniques in terms of both design time, and expected performance. The case study considered is the NASA Rotor 37. The automatic parameterization of entire datasets through the use of variational autoencoders, a specific type of Neural Network, is explained and discussed. The autoencoder latent parameters describe the blade 3D geometry and can be used as an alternative to the standard geometric parameters in describing the shape of each sample. The main advantage is that autoencoders enable an automatic parameterization of 3D geometries, thus overcoming the limits imposed by manual parameterization. The performance prediction (efficiency, pressure ratio) is carried out through a specifically developed Neural Network, properly trained. The aerodynamic optimization is performed using a Genetic Algorithm: by acting on the latent parameters, the algorithm generates new optimized blades, automatically meshed, verified through CFD simulation, and added to the starting database, with a re-train of the artificial intelligence algorithms. This loop is carried out several times until efficiency is maximized. Finally, a comparison with a classical optimization based on standard geometric parameters is reported and the results are deeply discussed and analyzed.