Research On Surrogate-based Efficient Aerodynamic Optimization And Many-objective Problems

In modern aircraft design,aerodynamic shape optimization(ASO)method coupled with CFD(Computational Fluid Dynamics)and numerical optimization theory has become a very important tool to improve aerodynamic performance of aircraft.In this field,surrogate-based optimization method plays a significant role for its capability of achieving efficient global optimization and handling multi-objective problems.Under the sustained pursuit of excellent aerodynamic performance,the complexity of ASO is becoming prominent.Thereby,surrogate-based aerodynamic optimization(SBAO)method is faced with new problems and challenges.On the one hand,from the perspective of design space,adequate range(scale)of design variables is needed to meet the requirements of global optimization and innovative design.Meanwhile,sufficient number(dimension)of design variables is necessary for precise control of complex configurations.The increase of scale and dimension will result in the explosion of design space,which brings rigorous troubles to the efficiency of SBAO method.On the other hand,from the perspective of objective space,more design objectives are needed to cope with the stringent requirements of aircraft performance.However,the increase of objective space dimension will lead to a sharp decline in the property of multi-objective optimization algorithm.In addition,it becomes difficult to visualize the Pareto optimal solutions.Focusing on these issues above,surrogate-based efficient aerodynamic optimization and many-objective problems are studied in this paper.First,basic methods and key technologies are researched to construct the SBAO platform.Afterwards,the issues of design space scale,design space dimension and many-objective optimization are studied in depth.Some innovative design methods and thoughts are proposed,which expanded the value of SBAO method in engineering application.The main contents of this paper are as follows:1.Surrogate-based aerodynamic optimization platform is constructed,which is qualified to single-objective/multi-objective optimization of complex aircraft configurations.Basic components in SBAO are researched,including geometry parameterization,grid deformation,calculation of aerodynamic characteristics,single-objective/multi-objective optimization algorithms,surrogate model and infill criterion in surrogate-based optimization.In the meantime,key technologies in SBAO are analyzed and improved.For airfoil parameterization,the method featuring a linear superposition of basis functions is studied.Through fitting the 1300 standard aifoils in UIUC(University of Illinois Urbana Champaign)airfoil library,it can be found that,POD orthogonal basis vector method and CST method have a better description of design space with airfoil geometric features.For parameterization of 3D shape,multi-block FFD(Free-form Deformation)technique is developed,which has the ability to arbitrary deformation of complex aircraft configurations.IDW(Inverse Distance Weighted)grid deformation technique coupling quaternion is developed.Afterwards,IDW-TFI hierarchical hybrid deformation technique,sparse boundary grid strategy and MPI parallel computing method are proposed,which substantially improve the efficiency of deformation for large-scale multi-block structured grid.Kriging model and infill criterion are studied in depth.The identical expression of Kriging prediction is derived using maximum likelihood estimation,Bayesian inference and minimum variance unbiased estimation respectively.The influence of hyperparameter,complexity of original problem and density of samples on the accuracy of Kriging prediction is studied.Further,it is pointed out that,the main contradiction in SBAO method is the contradiction between the demand for high dimensional and large scale design space in global optimization with the sparsity of samples.To solve the problem of poor convergence property of EI infill criterion,a hybrid infill method mixed EI and MP criterion is put forward,which improves the convergence rate within the framework of EI criterion.2.In view of the difficulty to guarantee global and efficient optimization in large-scale design space,SBAO method with adaptive design space expansion is proposed.The influence of design space scale on SBAO is studied,which reveals that,it is difficult to guarantee global and efficient optimization in conventional SBAO method and then,SBAO method with adaptive design space expansion is proposed.In this method,the optimization is carried out in a dynamic design space and the samples in large-scale design space are allocated efficiently through adaptive expansion of design space boundaries.Through single-objective and multi-objective aerodynamic optimziations of NACA0012 airfoil and RAE2822 airfoil,it can be confirmed that,the proposed method with adaptive design space expansion can significantly improve the efficiency of optimization compared to conventional method with fixed design space.3.In order to handle the ?curse of dimensionality? faced by surrogate-based optimization method in high dimensional design space,SBAO method with effective design space reduction is proposed.The influence of design space dimension on SBAO is analyzed,which reveals the ?curse of dimensionality? in SBAO and then SBAO method with effective design space reduction is proposed.In this method,effective design space is extracted by effective samples and is further reconstructed by K-L transformation.Accordingly,the predicament of ?curse of dimensionality? is relieved through gradually reducing effective design space during the optimziation process.Finally,NACA0012 airfoil with 48 design variables and CRM(Common Research Model)configuration with 99 design variables are optimized,which confirms that the proposed method with effective design space reduction can distinctly improve the capability of SBAO method in high dimensional design space.4.In view of many-objective aerodynamic design optimization problem,many-objective particle swarm optimization(MaOPSO)algorithm for engineering design is proposed and SOM-based visualization method for high dimensional Pareto front is developed.In MaOPSO,the ideas of objective reduction,grouping and decomposition are fused and many-objective optimization problem is converted to a series of two-objective sub-optimization problems utilizing correlation analysis of objectives and experimental design method.Accordingly,the convergence of non-dominated solutions is significantly improved on the premise of certain diversity.Through DTLZ-2 test function with 3-10 objectives,it is verified that the overall performance of MaOPSO is better than that of MOPSO,MOEA/D and NSGA-III.In SOM-based visualization method,high dimensional objective vectors are mapped to two-dimensional planes through adaptive clustering technique.Decisions are made according to the SOM nephograms of each objective,which provides a global perspective for the designer.Afterwards,many-objective aerodynamic optimization of SC1095 rotor airfoil and NACA64A204 fighter airfoil are implemented applying MaOPSO and SOM.Excellent rotor airfoil and fighter airfoil are obtained which satisfies complex engineering design requirements.In addition,the design results of fighter airfoil are verified through wind tunnel test.