Multifidelity Aerodynamic/Stealth Design Optimization Method for Flying Wing Aircraft

The design optimization considering both aerodynamic and stealth performances has been an important and challenging area for next-generation aircraft. One of the main difficulties is associated with the prohibitive computational cost of optimization with a large number of design variables. A two-rounds multi-fidelity aerodynamic/stealth design optimization method based on hierarchical Kriging (HK) model is developed in this paper by using the validated RANS solver and computational electromagnetics (CEM) methods based on the multilevel fast multipole algorithm (MLFMA) and physical optics (PO) algorithm. RANS method with fine grids and MLFMA method are used as high-fidelity CFD/CEM simulation. RANS method with coarse grids and PO method are served as low-fidelity CFD/CEM simulation. A low-fidelity optimization based on Kriging model is carried out to get lots of low-fidelity sample data which are used by HK model. The optimum shape is simulated by high-fidelity CFD/CEM methods to get the data used as high-fidelity sample points of the HK model. At the same time, some new initial high-fidelity sample points are generated by Latin hypercube sampling (LHS) method. The two kinds of high-fidelity sample points and lots of low-fidelity sample points are used to build initial HK model. Then an aerodynamic/stealth coupled design optimization of a flying wing aircraft with 108 design variables is carried out to validate effectiveness of the method. The objective is to reduce total drag of cruise condition and frontal RCS. Results indicate that with the method developed in this paper, the efficiency of aerodynamic/stealth optimization is improved significantly. The number of high-fidelity CFD/CEM simulations is reduced by seven times. Only 39 CFD/CEM simulations are used to get the optimum shape. The drag coefficient of optimized flying wing at a cruise condition is reduced by 3.5% and the average RCS in the frontal observation angle range is reduced by 44.3%, which validates the effectiveness of the developed method.