基于广义回归神经网络的转子系统质量不平衡辨识  

Identification of Mass Unbalance in a Rotor System based on Generalized Regression Neural Network Model

作　　者：陈东超[1] 顾煜炯[1] 徐婧[1] 赵鹏程[1] 

机构地区：[1]华北电力大学国家火力发电工程技术研究中心,北京102206

出　　处：《汽轮机技术》2015年第6期440-443,共4页Turbine Technology

基　　金：国家自然科学基金资助项目(51075145);中央高校基本科研业务费专项资金资助项目(2014XS32;2015XS88)

摘　　要：提出了一种基于广义回归神经网络(GRNN)模型的转子系统质量不平衡定量辨识方法:采用拉丁超立方抽样和转子系统的有限元模型获取训练样本,随后采用优化的GRNN模型建立不平衡参数与各测点振动响应的对应关系,降低反演优化迭代过程中冗繁的有限元计算;构建用于求解质量不平衡参数的目标函数,借助建立的GRNN关联模型,采用PSO算法寻找满足符合目标函数的全局最优解,从而达到质量不平衡定量识别的目的。此外,为了降低噪声的影响,采用零相位带通滤波器进行实测振动信号的去噪滤波。仿真结果表明该方法能有效地辨识出转子系统中的质量不平衡参数,辨识结果的准确度较高。A novel method of mass unbalance quantitative identification in a rotor system is proposed based on generalized regression neural network （GRNN） model. Firstly, training samples are obtained based on Latin hypercube sampling （LHS） and finite element model of the rotor system. And then GRNN is adopted to constructure association models which set up a relation between the unbalance parameters and correspoding vibration dynamic responses and thereby avoiding the time- consuming finite element calculation at every iterative step of optimization. For identifying unbalance, the objective function for solving unbalance parameters are constructed and then GRNN association models and particle swarm optimization algorithm are applied t6 enhance the global searching ability. In order to reduce the effects of noise, zero phase bandpass filters have been used to implement noise-elimination. Results show that the proposed method can identify unbalance parameters effectively.

关 键 词：转子系统 质量不平衡 定量辨识 广义回归神经网络 

分 类 号：TK268.1[动力工程及工程热物理—动力机械及工程]