Machine-Learning Models for Loss and Deviation Angle of Compressor Cascades

We present deep artificial neural network models of deviation angle and loss coefficient for typical compressor cascade blade rows trained from data obtained from Computational Fluid Dynamics (CFD). Computations based on Reynolds-averaged Navier-Stokes equations with a two-equation turbulence model are first validated against available experimental data and then used to generate data representing a large regime of flow and blade conditions for a typical blade row. Flow conditions and design parameters considered include incidence angle, stagger angle, solidity, Mach number, and Reynolds number for the NACA 65(12)-10 typical compressor airfoil. Artificial Neural-network models for deviation angle and loss coefficient with those parameters as inputs are then trained based on the computed data. The performance of the neural network models is then assessed and compared with typical empirical design models. The artificial neural network models are valid over a broader range of flow conditions and geometry variations and can capture effects that the conventional models do not consider.