Modelling three-dimensional dynamic stall of helicopter blades using computational fluid dynamics and neural networks

Abstract

Numerical simulation of three-dimensional dynamic stall (DS) of helicopter rotor blades has been undertaken using computational fluid dynamics (CFD) based on the Navier-Stokes equations. The CFD method has been carefully validated against experimental data before being used to study the topology of the three-dimensional DS vortex and the effects of yaw and rotation on its shape and trajectory. The three-dimensional unsteady viscous computations were found to give a wealth of results at the expense of significant amounts of CPU time. To alleviate this problem and to develop a faster model for the aerodynamic loads encountered during three-dimensional DS, a neural network (NN) was put forward. The NN was trained using both CFD and experimental data and was subsequently used as a method for interpolating DS loads between known states for which CFD data were available. The NN was found to work well for such interpolations, although its capability to extrapolate outside the training envelope was somehow limited. Nevertheless, the NN was found to be a very efficient technique for reconstructing three-dimensional DS and has demonstrated great potential as a method for reducing non-linear aerodynamics to a simple computational model.