Research On Key Technologies Of Supercritical Airfoil Optimization Design

Airfoil design is a basic research in aircraft design which affects the aerodynamic efficiency,handling quality and even safety of the aircraft.The supercritical airfoil used for transonic flight is an important kind of basic airfoil for design.Due to the long design cycle and high cost,traditional methods for airfoil design,such as wind tunnel tests,have been gradually replaced by a hybrid method which combines computational fluid dynamics softwares,machine learning theory and optimization theory.Since airfoil design problems often involve highly nonlinear objectives,diverse constraints,and multiple local optima,it is still a challenge for traditional optimization methods to find the global optimum with high confidence.In this thesis,the supercritical airfoil RAE2822 is used as the basic airfoil for optimization design,and the key technologies with emphasis on optimization methods are studied in depth.The detailed contents are summarized as follows:(1)This thesis compares among the common methods of each key technology in airfoil design,and selects the appropriate ones for airfoil optimization.As for parameterization methods,the thesis proposes a hybrid airfoil parameterization method combining a linear Hicks-Henne which ensures uniform sampling in the sampling stage and a high-accuracy CST(Class Shape Transformation)in the model training stage;as for surrogate models,a multi-output Gaussian Process(MOGP)considering the correlation between lift coefficient and drag coefficient is proposed to be used for accuracy improvement.Based on the above,a general optimization framework for airfoil design is given.(2)In order to solve the airfoil optimization problem with highly nonlinear objectives and multiple local optima,an improved weed algorithm(IWO_DE/Ring)is introduced into this filed for the first time in the thesis.Experiments on several standard test functions and the airfoil optimization problem show that the algorithm has stable performance on search for the global optimum in highly nonlinear design space due to its balance between global exploration and local exploitation.In the premise of an initial surrogate model with high accuracy,IWO_DE/Ring is likely to search near the true global optimum of the airfoil optimization problem,resulting in stable and quick convergence to an optimal solution with high confidence in the dynamic optimization process and a significant improvement in the aerodynamic performances of the optimal airfoil compared with the original RAE2822.(3)The above-mentioned supercritical airfoil optimization based on the improved weed algorithm is devoted to finding the global optimum of the surrogate model.It needs a large number of initial samples to train an accurate surrogate model and thus to ensure the reliability of the found global optimum.This results in serious efficiency decrease.To solve this problem,the thesis studies a Bayesian forward selection technique based on probability of improvement(PI),and applies it to the airfoil optimization.This technique utilizes the prediction value as well as the prediction variance of the surrogate model to find the next sample with the highest probability of improvement over the existing samples and ignore the area with lower probability to be the true global optimum.So the sampling is highly targeted.Experiments on an analog function and the airfoil optimization problem show that the sampling in each iteration of the dynamic optimization process is devoted to finding the true global optimum in the design space.Compared with the results of supercritical airfoil optimization based on the improved weed algorithm,it can converge to a final optimal airfoil with better aerodynamic performances even using a small number of(initial)samples.