基于全局灵敏度分析的某自动装填机构轻量化设计

摘要

An optimization strategy composed of a global sensitive analysis method and the surrogate model technology was proposed for the case of large scale parameters and expensive computing cost in the lightweight design of an auto loading system.A global sensitive analysis based on Morris trajectory was carried out to determine the 14 important parameters from 32 ones.Simulations were performed on the Latin Hypercube Sampling points of important parameters, and the surrogate model of the responses about the important parameters was established based on RBF NN (radial basis function neural networks).The optimal lightweight solution was obtained by using the Multi-island Genetic Algorithm, with which the weight of the loading mechanism was reduced by 21.8%.A numerical example shows that the surrogate model including only the important parameters can achieve more accurate prediction results.The optimization strategy was proved to be efficient in the lightweight design of complex systems.%针对某自动装填机构轻量化设计中出现参数多、模型计算量大等问题，提出将全局灵敏度分析与代理模型技术相结合的优化策略。通过基于 Morris 轨迹的全局灵敏度分析从32个系统参数中确定14个关键参数，基于拉丁超立方采样技术及径向基函数神经网络技术（Radial Basis Function Neural Networks，RBFNN）建立系统响应关于关键参数的代理模型，用多岛遗传算法对系统参数进行优化求解，致机构重量下降21．8％。数值检验结果表明仅含关键参数的代理模型预测精度较高，证明该方法在多参数复杂系统结构轻量化设计中的有效性。