| description abstract | A shape-preserving response prediction (SPRP)-based variable-fidelity model is developed in this paper to improve the computation efficiency in helicopter rotor computational fluid dynamics (CFD). In helicopter rotor CFD, aerodynamic coefficients such as the thrust coefficient vary with azimuth angle. However, both the kriging-based and correction-based variable-fidelity surrogate models are constructed by building a scalar modification on the corresponding average value of the performance parameters, which will lead to possible loss of correlation and subsequently affect the accuracy of the established surrogate models. In this study, firstly, the design of experiments method is used to establish a sample database consisting of the numerical results of both low-fidelity and high-fidelity models. Then, the SPRP method is used to build the scaling function, and a generalized regression neural network is used to construct the surrogate of the scaling function. The SPRP method applies a vectorial modification to establish the surrogate model, which enables the model to correct the flow-field information of the low-fidelity model at all azimuth angles. Finally, three test cases are investigated, including two trimming cases and one optimization case, which all are very time-consuming problems. It is found that half of the computational cost can be reduced using the new method rather than the traditional trimming method. The numerical results show the potential of the present method in helicopter CFD. | |
