摘要
提出了一种基于Zernike不变矩特征和神经网络分类器的轴心轨迹自动识别方法。通过对原始Zernike矩特征进行二次提取和处理,获得了对轴心轨迹识别更为敏感的矩特征量,降低了后续神经网络分类器设计的难度。仿真研究表明,基于Zernike矩的轴心轨迹识别方法,其识别精度优于常用的几何矩方法。将所提方法应用于汽轮发电机组和高速离心压缩机组轴心轨迹的自动识别,并结合频谱能量分布特征进行故障诊断,结果表明,引入轴心轨迹特征可以有效地提高旋转机械故障诊断的精度。
This paper presents an automatic recognition method of the shaft orbit based on the Zernike moments and neural network. By re-extracting and re-processing the original Zernike moments,the more sensitive moment features for the shaft orbit identification were obtained,which simplified the designing of the succeeding neural network classifier. The simulation result shows that the accuracy of the Zernike moment based method is superior to the Hu's geometric moment invariants based one. The proposed method was applied to automatically recognize the shaft orbit of a turbo-generator and a high-speed centrifugal air-compressor,and the recognition results were incorporated with the spectrum energy distribution features to diagnose the faults. It is shown that the accuracy of fault diagnosis is improved effectively by considering the orbit features,and the method is feasible.
出处
《振动.测试与诊断》
EI
CSCD
北大核心
2009年第2期141-145,共5页
Journal of Vibration,Measurement & Diagnosis
基金
国家"八六三"高技术研究发展计划资助项目(编号:2007AA04Z421)
国家自然科学基金资助项目(编号:50875048)
江苏省基础研究计划资助项目(编号:BK2007115)
关键词
轴心轨迹
不变矩
神经网络
旋转机械
故障诊断
shaft orbit moment invariant neural network rotating machinery fault diagnosis