摘要
为提高滚动轴承故障诊断的性能,结合故障敏感特征的选择,提出了一种基于小波包变换(WPT)和监督NPE的滚动轴承故障诊断模型。首先,WPT对原始振动信号进行处理,利用终端节点的单支重构信号得到多域统计特征,构成原始特征集。然后,为减少特征集中的冗余信息和干扰特征,提出一种基于朴素贝叶斯的故障敏感特征选择方法(FSNB)。为了进一步降低冗余信息和运算复杂度,提出一种基于类别标签的监督邻域保持嵌入(SNPEL)方法,实现对高维特征集的低维表示。最后,利用K近邻(KNN)算法实现滚动轴承的故障诊断。采用12种轴承故障数据来验证提出的故障诊断模型的性能,结果表明,提出的模型可以实现较高的故障诊断准确度和较好的适应性。
In order to enhance the performance of bearings fault diagnosis,a bearings fault diagnosis model based on wavelet packet transform(WPT)and supervised NPE,incorporating fault sensitive features selection,was proposed in this paper.First,the original vibration signals are processed by WPT,and multi-domain statistical features can be obtained by single branch reconstruction signals of terminal nodes,which construct origina l feature set. Then,in order to reduce redundant information and interference features,a fault sensitive features selection method based on naive bayes(FSNB)was proposed.In order to further reduce redundant information and computational complexity,a supervised neighborhood preserving embedding based on class label(SNPEL)was proposed to realize low-dimensional representation for high-dimensional feature set. Finally,KNN algorithm was applied to achieve bearings fault diagnosis. In this article,12 kinds of bearing fault data were employed to evaluate the performance of the proposed model,the results indicate that the proposed model can achieve higher accuracy of fault diagnosis and preferable adaptability.
作者
董飞
俞啸
丁恩杰
吴守鹏
DONG Fei;YU Xiao;DING En-jie;WU Shou-peng(School of Information and Control Engineering,China University of Mining and Technology,Jiangsu Xuzhou 221008,China;IOT Perception Mine Research Center,China University of Mining and Technology,Jiangsu Xuzhou 221008,China;School of Medicine Information,Xuzhou Medical University,Jiangsu Xuzhou 221009,China)
出处
《机械设计与制造》
北大核心
2020年第3期29-33,共5页
Machinery Design & Manufacture
基金
国家重点研发计划—矿山安全生产物联网关键技术与装备研发(2017YFC0804400,2017YFC0804401)
国家重点基础研究发展计划(973)—深部危险煤层无人采掘装备关键基础研究(2014CB046300)。
关键词
故障诊断
敏感特征
小波包变换
朴素贝叶斯
K近邻
Fault Diagnosis
Sensitive Features Selection
Wavelet Packet Transform
Naive Bayes
KNN
