摘要
针对传统的电机故障诊断方法往往采用单一信号作为诊断依据,以及利用传统的BP神经网络进行故障诊断时存在的训练速度慢、易陷入局部极小值的缺点,提出了一种基于极限学习机和多源信息融合的电机故障诊断方法。首先将定子电流信号做陷波处理,滤除基波分量;然后对电流及振动信号进行小波包分解和重构,以各频带的小波包能量谱作为故障特征向量训练极限学习机模型;最后将训练好的极限学习机模型作为诊断决策分类器来判断电机的运行状态。实验结果表明,此方法能够准确地诊断电机的故障类型,具有运行速度快、故障诊断准确率高的特点,满足了系统在线实时诊断的要求。
According to the shortcoming of traditional method for motor fault diagnosis such as using single signal for the basis of diagnosis and using the BP neural network that has slow training speed and local minima,a new method is proposed for motor fault diagnosis based on extreme learning machine and multi-source information integration.Firstly,in order to remove the fundamental component,the stator current signal passes through a notch filter of 50 Hz.Secondly,the current and vibration signals are processed as wavelet packet decomposition and reconstruction and the wavelet packets energy spectrum of each band are used as fault feature vector to train the extreme learning machine model.Finally,the trained extreme learning machine model is used as classifier of diagnosis and decision to judge the operation state of motor.The experiments result shows that the method can not only accurately judge the fault type of motor,but also have the advantages of high running speed and accuracy to meet the requirement of online real-time diagnostics.
出处
《测控技术》
CSCD
2015年第8期12-15,共4页
Measurement & Control Technology
基金
"十二五"国家科技支撑计划项目资助(2012BAH32F06)
关键词
极限学习机
陷波处理
故障诊断
小波包能量谱
extreme learning machine
wave-trapping
fault diagnosis
wavelet packets energy spectrum