Aerodynamic Optimization Method Based On High-Fidelity Flow Field Prediction

Aerodynamic design is an important part of aircraft design,which has a significant influence on overall performance of aircraft.With the improvement of aircraft performance requirements,aircraft configurations become more complex,which put forward higher requirements for the refined design of the aerodynamic layout.The traditional aerodynamic engineering estimation model is not suitable in many occasions nowadays,so high-fidelity numerical simulation is imperative.Traditional evolutionary optimization algorithms usually cost thousands of function evaluations,which is unacceptable for problems with time-consuming simulation models.The approximationbased optimization method was proposed in order to achieve a balance between accuracy and efficiency in aerodynamic design.This paper studies on modifying Sequence Approximate Optimization method in order to improve its efficiency in aircraft aerodynamic optimization design problems.An improved Sequence Approximate Optimization method for aerodynamic optimization design is proposed,which can effectively utilize the large amount of flow field data generated by the high-fidelity CFD simulation model to improve the aerodynamic optimization efficiency.A modified Optimal Latin Hypercube Experimental Design method for constrained design space is proposed.Feasible and infeasible domains are differentiated by weighting and the uniformity and the number of feasible sampling points are considered comprehensively.Consequently,the sampling criteria suitable for constrained Optimal Latin Hypercube Experimental Design are constructed and optimize using evolutionary algorithms.The proposed design of experiment method effectively solves the uniform sampling problem of constrained domain in aerodynamic optimization problems and avoids the sampling and CFD simulation calculation of infeasible sampling points.A hybrid approximation model combining Radial Basis Function and Polynomial Chaos Expansions is proposed,which significantly improves the accuracy of approximate model.On this basis,an analytical solution method for the sensitivity index of the hybrid approximation model is proposed,which significantly reduces the cost of sensitivity analysis calculation.A Field Approximation Model for the aerodynamic optimization problem is proposed.With the adoption of Field Approximation Model,the Sequential Approximate Optimization method could be deeply coupled with the aerodynamic optimization problem and the results of high-fidelity CFD simulation are fully utilized.As a result,the number of time-consuming aerodynamic CFD simulation calls is greatly reduced and the efficiency of Sequential Approximate Optimization method could be greatly improved in engineering optimization problems with high-fidelity aerodynamic CFD simulation models.The inaccurate search strategy is introduced and the optimization performance of sequential sampling strategy is significantly improved.The elite archive storing a certain number of high-quality sample points is adopted to implement the adaptive sampling strategy based on inaccurate search.Under this circumstance,the misguidance to the optimization process caused by imprecise approximation model in the early stage and the waste of computational resources caused by excessive searching could be avoided.The time cost of stochastic algorithm optimization and the CFD simulation of infeasible sampling points could be saved.Therefore,the convergence efficiency of the Sequential Approximate Optimization method is further improved.The aerodynamic optimization method based on high-fidelity flow field prediction is proposed specifically for the application in aerodynamic optimization problems.The proposed optimization method is applied to several aerodynamic optimization design cases.Through comparing with results of conventional sequential approximate optimization method and stochastic evolutionary algorithm,the validity and efficiency of the proposed method are verified.