基于代理模型的批产化卫星动力学试验舱内响应预示方法

A method for internal response prediction based on surrogated model for batch production satellites in dynamics test cabin

摘要: 文章提出基于代理模型的航天器结构动力学试验舱内响应预示方法，分别采用前馈神经网络（FNN）和Kriging代理模型，利用正弦扫频试验所获得的外部实际测点数据对内部测点响应进行预示；并通过组件级产品和系统级航天器实例，对比2种方法的预示结果，讨论分析代理预示方法的有效性及样本数量对预示精度的影响。研究结果对后续批产化航天器动力学试验测点剪裁具有一定指导意义。

Abstract: In this paper, an internal response prediction method based on surrogate model was proposed for spacecraft structural dynamics test cabin. The feedforward neural network (FNN) and the Kriging model were applied respectively so as to predict the internal measurement point response using actual external measurement data obtained in the sinusoidal sweep frequency test. The prediction results by the two methods were compared through the examples of a component-level product and a system-level spacecraft. The effectiveness of the surrogate prediction method and the influence of sample number on the prediction accuracy were discussed and analyzed. This study may provide a reference for measurement point tailoring in the following dynamics test for batch production spacecrafts.