CFD analysis of 3D dynamic stall
Focusing on helicopter aerodynamics, it is known that the aerodynamic performance of the retreating side of a rotor disk is mainly dictated by the stall characteristics of the blade. Stall under dynamic conditions (Dynamic Stall) is the dominant phenomenon encountered on heavily loaded fast-flying rotors, resulting in an extra lift and excessive pitching moments. Dynamic stall (DS) can be idealised as the pitching motion of a finite wing and this is the focus of the present work which includes three main stages. At first, comparisons between available experimental data with CFD simulations were performed for 3D DS cases. This work is the first detailed CFD study of 3D Dynamic Stall and has produced results indicating that DS can be predicted and analysed using CFD. The CFD results were validated against all known experimental investigations. In addition, a comprehensive set of CFD results was generated and used to enhance our understanding of 3D DS. Straight, tapered and swept-tip wings of various aspect ratios were used at a range of Reynolds and Mach numbers and flow conditions. For all cases where experimental data were available effort was put to obtain the original data and process these in exactly the same ways as the CFD results. Special care was put to represent exactly the motion of the lifting surfaces, its geometry and the boundary conditions of the problem. Secondly, the evolution of the Ω-shaped DS vortex observed in experimental works as well as its interaction with the tip vortices were investigated. Both pitching and pitching/rotating blade conditions were considered. Finally, the potential of training a Neural network as a model for DS was assessed in an attempt to reduce the required CPU time for modelling 3D DS. Neural networks have a proven track record in applications involving pattern recognition but so far have seen little application in unsteady aerodynamics. In this work, two different NN models were developed and assessed in a variety of conditions involving DS. Both experimental and CFD data were used during these investigations. The dependence of the quality of the predictions of the NN on the choice of the training data was then assessed and thoughts towards the correct strategy behind this choice were laid out.