基于改进变分模态分解排列熵和极限学习机的汽轮发电机转子故障诊断方法  

Turbine Generator Rotor Fault Diagnosis Method Based on Improved Variational Mode Decomposition Permutation Entropy and Extreme Learning Machine

作　　者：冯培燕[1] 张少波[2] 

机构地区：[1]苏州工业园区职业技术学院机电工程系,江苏苏州215123 [2]华北电力大学机械工程系,河北保定071003

出　　处：《电机与控制应用》2016年第11期86-91,共6页Electric machines & control application

基　　金：国家自然科学基金项目(11072078)

摘　　要：针对实际转子振动信号的非线性、非平稳性引起的故障类型难以准确识别的问题,提出了一种基于改进变分模态分解(VMD)排列熵和极限学习机的转子故障诊断方法。首先,为克服VMD中惩罚因子和分解个数按经验选择的问题,提出一种基于人工化学反应算法的改进VMD方法,将其用于振动信号分解,得到若干个不同尺度的固有模态分量(IMF);随后计算蕴含主要故障特征信息的前几个IMF的排列熵值;最后将得到的前几个排列熵值作为特征矢量,输入到建立的极限学习机中实现不同状态下转子振动信号的模式识别。将提出方法应用于汽轮发电机转子试验台采集的数据,结果表明:提出的方法能有效实现不同运行状态下的转子振动信号的辨识,提高了模式识别精度。Aim at the nonlinear and non-stationary of the actual rotor vibration signal as well as the difficulty of rotor fault type identification, a rotor fault diagnosis method based on the improved VMD permutation entropy and extreme learning machine was proposed. Firstly, to overcome the empirical selection of punishment factor and the number of decomposition in VMD, an improved VMD based on the artificial chemical reaction algorithm was proposed to decompose the vibration signal and obtain several intrinsic mode components （IMFs）. Then permutation entropy value of intrinsic mode components containing the main fault characteristic information was computed. Finally, permutation entropy was regarded as eigenvector and was input to extreme learning machine ; pattern recognition of the rotor vibration signals under different condition of could be realized. The proposed method was applied to the rotor experiment data, the analysis results showed that the proposed method could effectively identify rotor vibration signal under different running status and improved the pattern recognition accuracy.

关 键 词：变分模态分解 人工化学反应算法 排列熵 极限学习机 故障诊断 

分 类 号：TM307.1[电气工程—电机]