基于RBF神经网络的补燃火箭发动机频率预测  

FREQUENCY PREDICTION OF STAGED COMBUSTION ROCKET ENGINE BASED ON RBF NEURAL NETWORK

作　　者：杜飞平[1] 谭永华[2] 陈建华[1] 

机构地区：[1]西安航天动力研究所,西安710100 [2]航天推进技术研究院.西安710100

出　　处：《机械强度》2015年第6期1190-1194,共5页Journal of Mechanical Strength

摘　　要：为避免建立或修改计算量巨大的发动机有限元模型,研究探讨径向基函数(radial basis function,RBF)神经网络在液体火箭发动机频率预测中的应用。以某型高压补燃液氧/煤油火箭发动机为研究对象,在考虑喷管内外壁材料差异的基础上,利用刚度和质量等效原则,建立了喷管有限元模型。然后采用分布参数法建立了补燃循环火箭发动机的有限元模型,同时利用模态试验数据修正有限元模型。根据结构灵敏度分析理论,选择不同的结构参数组合作为训练样本训练神经网络,并利用训练好的神经网络预测发动机结构频率。研究结果表明,RBF神经网络能较好地预测液体火箭发动机结构频率,预测误差在1.0%以内。同时该方法具有收敛速度快的优点,可广泛应用于火箭发动机数值仿真领域。In order to avoid creating or modifying the huge amount of finite element model of the rocket engine,application of the RBF（ radial basis function） neural network which was used to predict structural frequency of the liquid rocket engine was analyzed. A high pressure staged combustion LOX（ liquid oxygen） / kerosene rocket engine was taken as research object. By considering the material differences between the inner and outer wall of the nozzle,the precise nozzle finite element model was established using the stiffness and mass equivalence principle. Then the finite element model of staged combustion rocket engine was established using the method of distributed parameter and the finite element model was modified by modal experimental data.The difference structural parameters which were chosen according to the structural sensitivity analysis were considered as training samples. Then the RBF neural network was trained using these samples,and the structural frequency was predicted by the trained RBF neural network. The results show that the RBF neural network can predict structural frequency of the rocket engine accurately and the prediction error was less than 1. 0%. This method also has a fast convergence pace and it can be widely used in the field of the structural simulation of the rocket engine.

关 键 词：径向基函数神经网络 补燃火箭发动机 数值仿真 结构分析 性能预测 

分 类 号：V434[航空宇航科学与技术—航空宇航推进理论与工程]