| With the increasingly higher demands of the aerodynamic and overall performance of modern aircrafts,the aerodynamic design optimization methods have been facing higher and higher requirements and challenges.Aiming at the efficient global design problem of aerodynamic shapes of aircrafts,in order to satisfy the increasingly higher requirements of various aerodynamic characteristics and to design the aerodynamic shapes satisfying the engineering requirements,in this thesis we made a detailed study on surrogate-based aerodynamic design optimization（SBADO）technique and uncertainty-based robust aerodynamic design optimization（RADO）method,including their related key techniques,theories,primary challenges and so on.We focused on the high-dimensional surrogate-based optimization problem,many-objective optimization challenge,multi-fidelity/variable-fidelity surrogate modeling（MF/VF-SM）technique and efficient RADO method,and proposed effective solution approaches and ideas,to improve the optimization efficiency and capability of complex aerodynamic design problems.Finally,the developed methods and researches in this thesis were investigated and validated by a series of complex aerodynamic shapes optimization examples of different types.The main contributions are as follows:1,An efficient surrogate-based design optimization platform for complex aerodynamic configuration design of aircrafts was developed.Some key technologies of the platform were studied in depth.Firstly,the transition prediction technique based on SST?-Re?t model and the grid generation method for natural laminar flow（NLF）airfoil computation were studied to provide accurate and reliable high-speed NLF airfoil numerical evaluations.Then,we studied the constitution characteristics and deformation capability of three kinds of typical airfoil parameterization methods and multi-block control frame free form deformation（FFD）parameterization methods for three-dimensional complex configurations to effectively parameterize different types of airfoils and three-dimensional complex shapes as well as to improve the design efficiency.The research on the current popular SBADO method indicates that the hybrid infill strategies have better robustness and performance than the single infill sampling criteria,and Kriging surrogate model has strong flexibility and adaptability but encounters the serious"curse of dimensionality"problem.In this thesis,to ameliorate the"curse of dimensionality"problem in the global aerodynamic shape optimization,a supervised nonlinear dimension-reduction surrogate modeling（SN-DRSM）method was proposed.This method integrates the nonlinear dimension-reduction technique into the surrogate modeling process,and establishes the accurate mapping from the high-dimensional inputs to the output,which avoids the low accuracy and instability of current linear dimension-reduction methods（e.g.,PCA）or unsupervised dimension-reduction methods applied in surrogate models.Further,this method improves the efficiency of SBADO in high-dimensional aerodynamic shape optimization applications and ameliorates the"curse of dimensionality"problem.An efficient global aerodynamic optimization method for complex high-dimensional aerodynamic configurations was developed and applied to three standard test cases proposed as part of the AIAA aerodynamic design optimization discussion group（ADODG）.Results demonstrate that the developed method provides a better performance and an appreciably higher efficiency compared to the Kriging-based SBADO method.2,An efficient RADO method considering multi-parameter uncertainty for engineering complex aerodynamic shapes design was established.Considering the great challenges for aerodynamic shape design due to the adverse effect of uncertainties on aerodynamic characteristics,we firstly studied the state-of-the-art in RADO methodology,highlighted the key techniques and primary challenges in RADO,as well as provided the beneficial directions for future researches.The most important technical block of RADO,i.e.,aerodynamic UQ technique,was studied in depth.To avoid the prohibitive computational cost and a limited application range arising from the traditional UQ methods,an efficient UQ method based on PCE was developed,and it was demonstrated to be very suitable for complex aerodynamic uncertainty and sensitivity analysis.Aiming at the low-dimensional aerodynamic UQ,the PCE approach based on a two-step adjustment strategy was proposed,which appreciably improves the reconstruction efficiency and prediction accuracy of the full PC.For medium-and high-dimensional aerodynamic UQ,a sparse PC reconstruction algorithm based on an adaptive forward-backward selection（AFBS）method was proposed.The AFBS method significantly improves the efficiency and accuracy of sonic boom UQ and aerodynamic UQ of NLF airfoil,as compared to the classic least angle regression（LAR）algorithm.An efficient RADO method based on PCE was established.A class of high-speed and high-lift NLF airfoils with superior aerodynamic characteristics was successfully designed using this method,and it was compared to the classic NLF airfoils applied in the Global Hawk unmanned aerial vehicle（UAV）.It was demonstrated that the developed RADO method can provide more effective designs and greater advantages for engineering application when compared to deterministic optimization methods.3,A global approximation and aerodynamic optimization method based on the AMF-PCK surrogate model was built.To improve the generation capability of multi-fidelity surrogate models,an adaptive multi-fidelity polynomial chaos-Kriging（AMF-PCK）surrogate model was proposed,which significantly improves the metamodeling efficiency and approximation accuracy of multimodal and highly nonlinear landscape,as well as appreciably improves the efficiency of SBO algorithm.In order to further improve the efficiency of multi-fidelity surrogate modeling,a multi-fidelity adaptive sequential sampling strategy based on leave-one-out-cross validation-Voronoi-maximin scaled distance（LOOCV-Voronoi-MSD）was developed.Compared with the traditional single-fidelity sequence sampling approach,the proposed method can adaptively enrich the high-and low-fidelity samples sets separately until the accuracy criterion is satisfied.The developed techniques were investigated by several benchmark examples and transonic aerodynamic modeling applications,and were validated comprehensively by comparing its performances with those of universal Kriging,hierarchical Kriging and OPC-Kriging methods.It is demonstrated that the proposed method is more efficient and more accurate for global approximation than the other three.An aerodynamic optimization framework based on the proposed AMF-PCK metamodel was developed,and was then applied to both transonic deterministic aerodynamic optimization and RADO applications.Results reveal that the developed method provides a significant improvement of efficiency and reliability compared to Kriging-based optimization method.4,A many-objective RADO method for complex aerodynamic shape design was established.An efficient objective-reduction strategy was proposed to deal with the many-objective and complex constraints optimization problems for the high-speed rotor airfoil design over a wide range of Mach numbers.Considering the highly sensitive aerodynamic performance of high-speed rotor airfoils under transonic flow conditions,a many-objective robust optimization method was developed.An efficient many-objective robust optimization framework based on the developed AMF-PCK surrogate model was proposed to mitigate the computational burden.Further,the developed many-objective robust optimization method was utilized to optimize the high-speed rotor airfoils of 7%thickness.The aerodynamic performances of the optimized airfoils were compared with those of the classic OA407 airfoil at high,medium and low Mach numbers,respectively.Results validate the improved effectiveness and reliability of the proposed many-objective robust optimization method for complex aerodynamic optimization problems. |
PDF Full Text Request