采用动态交互作用分析的叶轮机械优化算法研究

Study on Turbomachinery Optimization Algorithm Based on Dynamic Interaction Analysis

针对当前基于代理模型的优化算法因遭遇“维度灾难”而难以开展叶轮机械高维优化的难题,提出了自适应厘清变量间关系的动态聚合全局优化算法。该算法能在不消耗额外样本的条件下获取设计变量间的交互信息,并基于该信息一方面将高维问题分解为多个低维问题,实现高维问题的高效优化,另一方面为后续知识挖掘提供信息基础。在完成典型30维函数算例测试后,利用所提出的算法完成了包含28个设计变量的Rotor 37叶栅气动优化与设计空间知识挖掘。结果表明,优化后,Rotor 37叶栅设计工况效率相对参考设计提高了1.69%,且变工况性能相对参考设计亦有不同程度提高。在获得最优解的同时,所提出的算法还厘清了Rotor 37叶栅设计空间各变量间交互作用关系。进一步,结合算法所揭示的变量交互作用进行知识挖掘与CFD验证,证明对分离线附近型线进行微调可有效降低激波与边界层相互作用,从而提高跨声速叶栅的气动性能。由此,所提出的算法在求解高维大资源叶轮机械优化设计问题上的适用性和有效性得到验证。

Due to the curse of dimensionality,it is difficult to perform high-dimensional optimization of turbomachinery with current surrogate-based optimization algorithm.To address the problem,a dynamic aggregation global optimization algorithm,which adaptively clarifies the variable interactions,was proposed.The proposed algorithm obtains the information on interactions between design variables without consuming additional samples.Based on this information,the high-dimensional problem is decomposed into multiple low-dimensional problems for efficient optimization.In addition,such information also serves as the basis for subsequent knowledge mining.After finishing the test on a 30-dimensional mathematical benchmark problem,the proposed algorithm is used for optimization and design space knowledge mining of the Rotor 37 compressor blade with 28 design variables.After optimization,the isentropic efficiency of Rotor 37 blade is increased by 1.69%at design condition;the aerodynamic performance of optimized design at off-design conditions is also improved.While obtaining the optimal solution,the proposed algorithm also reveals the interactions among variables in the design space of Rotor 37 blades,and knowledge mining and CFD verification are further carried out based on the variable interactions revealed.It shows that,the fine tuning of the profile near the separation line will effectively reduce the interactions between the shock waves and boundary layer,thus improving the aerodynamic performance of the transonic blade.Therefore,the applicability and effectiveness of the proposed algorithm in optimization of expensive high-dimensional turbomachinery design are well demonstrated.