The application of support vector regression and virtual sample generation technique in the optimization design of transonic compressor

Compressor blade design influences aero-engine performance mainly through its total pressure ratio and efficiency. As a volume-based geometric parameterization, Free-form Deformation (FFD) brings three-dimensionality that is essential to blade design. However, the manipulation of control points with respect to simple numeric parametric perturbation renders the use of design space low-efficient. Therefore, an improved FFD with ‘wiser’ control point lay-out is expected to identify those more important design variables. This paper proposes novel design variables that are assigned to grouping of control points’ displacement in FFD lay-out. In short, the approach is realized as: (i) establish a library of sufficient blade shape samples; (ii) filter the database with geometric constraints; (iii) extract dominant modes via (POD) Proper Orthogonal Decomposition; (iv) construct new design variables and apply them in optimization. With geometric constraint filtering, problem-oriented information is injected. With POD, the dominance of new selected geometric parameters in problem description is assured. Perfunctory details of displacement data of each control point in the lay-out can be replaced by grouped data as new design variable candidates. As a proof-of-concept study of the new approach, compressor blade Rotor 37 is selected to be the good platform of testing the feasibility of POD-aided FFD as a global and flexible yet economic geometric parameterization. Result demonstrates the feasibility of proposed POD-aided FFD approach that helps conduct an optimization involving displacement of 6 × 4 × 3 control points with as few as five new design variables, while still being capable of bringing optimization effect in three test cases.

Zhong Li

Papers

An approach of Proper Orthogonal Decomposition-aided Free-form Deformation with application in compressor blade design