基于多尺度熵和神经网络的滚动轴承故障诊断  

Rolling Bearing Diagnosis Based on Multiscale Entropy and Neural Network

作　　者：张龙[1] 张磊[1] 熊国良[1] 黄文艺[1] 周继惠[1] 

机构地区：[1]华东交通大学机电工程学院,南昌330013

出　　处：《机械设计与研究》2014年第5期96-98,105,共4页Machine Design And Research

基　　金：国家自然科学基金资助项目(51205130;51265010);江西省教育厅资助科技项目(GJJ12318);江西省自然科学基金资助项目(20132BAB216029)

摘　　要：滚动轴承是机械设备中最常用的零部件之一,其运行状态直接影响整机性能。文章针对滚动轴承故障振动信号具有跨尺度复杂性的特点,提出了一种新的基于多尺度熵(Multiscale entropy,简称MSE)和BP神经网络的滚动轴承故障智能诊断模型。该模型首先利用MSE方法对滚动轴承不同健康状态下的振动信号进行特征提取,再将其作为BP神经网络的输入,实现网络训练,最后利用神经网络自动识别故障类型及故障程度。实验结果表明,该方法能有效地实现滚动轴承故障类型及程度的智能诊断,并具有对网络初始值不敏感及较低的误报率和漏报率等优点。Rolling bearing is commonly used in mechanical equipment, and its running state directly influences the performance of the whole machine. Considering the vibration signals of the faulty rolling bearing with the complexity of striding scales, a new rolling bearing fault diagnosis method was proposed based on multiscale entropy （ short for MSE） and BP neural network. Fauh features were firstly extracted from vibration signals using MSE, and then input to a BP neural network as feature vectors. After the network training, the BP network was used to simultaneously judge the bearing fault type and severity. The results on experimental data verify the effectiveness of the proposed approach.

关 键 词：多尺度熵 BP神经网络 滚动轴承 故障诊断 

分 类 号：TH165.3[机械工程—机械制造及自动化] TN91[电子电信—通信与信息系统]