基于遗传算法圆周密封结构多目标优化

Multi-Objective Optimization of Circumferential Seal Structure Based on Genetic Algorithm

关键词:  圆周密封  热流固耦合  神经网络  遗传算法  多目标优化

Key words:  Circumferential seal  Thermal-fluid-structure coupling  Neural network  Genetic algorithm  Multi-objective optimization

摘要:

针对航空发动机密封性能的提升需求，基于圆周密封工况分析，建立了以泄漏量最小、密封环最大温升和最大变形最小为优化目标，以主、辅助密封带宽度，搭接头角度和长度及卸荷槽宽度为设计变量的多目标优化模型。采用拉丁超立方抽样方法得到了代表性样本库，通过热流固耦合分析确定对应目标函数值。利用RBF神经网络建立了高拟合精度的设计变量与目标函数映射关系代理模型，并结合第二代非劣排序遗传算法（NSGA-II），得到了考虑目标函数重要度的一组最优解。结果表明：与优化前相比，圆周密封泄漏量降低了17.69%，最大温升降低了11.88%，最大变形降低了38.10%，最大应力降低了31.02%。

Abstract:

Aiming at the demand of aero-engine sealing performances improvement， based on the analysis of the working conditions of circumferential seal， a multi-objective optimization model with the minimum leakage， the minimum maximum temperature rise and the minimum maximum deformation of the sealing ring as the optimization objectives， and the width of the main and auxiliary sealing belts， the angle and length of the lap joint and the width of the unloading grooves as the design variables was established. A representative sample database was obtained using the Latin hypercube sampling method， and the corresponding objective function values were determined by thermal-fluid-structure coupling analysis. The surrogate models of mapping relationship between design variables and objective functions with high fitting accuracy were established by RBF neural network method， and a set of optimal solutions considering the importance of the objective functions were obtained by combining with the Non-dominated Sorting Genetic Algorithm II （NSGA-II）. The results show that the leakage of circumferential seal is reduced by 17.69%， the maximum temperature rise is decreased by 11.88%， the maximum deformation is reduced by 38.10%， and the maximum stress is reduced by 31.02% by optimization.