摘要
针对转子系统早期碰磨故障,提出了一种基于时域和时频域联合特征提取和分析的方法,并采用该方法对简单转子模型进行了故障诊断。基于BP神经网络和影响函数法,建立了滑动轴承单盘转子碰磨故障动力学模型,并对转子系统碰磨故障进行了数值模拟;分别采用统计学和小波包分解方法,对振动信号的时域和时频域特征进行了提取,综合两者建立了碰磨故障的特征空间,并采用支持向量机(SVM)模型对比分析了基于时域、时频域和综合两者特征空间的故障诊断效果,在此基础上,通过引入可分度函数,将正常振动信号与故障信号同时考虑,对各特征的可分度进行了分析和排序;根据特征分析结果,将特征空间分为高可分度区域和低可分度区域,分别针对单特征和组合特征对碰磨故障进行识别。研究结果表明:单特征的碰磨故障识别率与其可分度函数值呈正相关;组合特征识别效果要优于单特征,且高可分度区域内的组合特征识别效果要明显优于低可分度区域,针对本文所建碰磨故障样本空间,高可分度区域内随机三特征组合平均故障识别率达到90%以上。文中提出的故障特征提取和分析方法可为复杂故障的识别提供参考。
A method of combined feature extraction and analysis in time and time-frequency domain is put forward for the diagnosis of early rubbing fault of rotor system is proposed and the implement of this method is given on a simplified rotor system.The dynamic model of rubbing fault related to single disk rotor with sliding bearing is established based on BP neural network and influence function method,followed with the numerical simulation of rotor system rubbing fault.The statistical method and wavelet packet decomposition are applied to extract features of the signal in time domain and time-frequency domain,based on which the feature space of rubbing fault is established preliminarily.Comparative analysis of the fault diagnosis performance based on features in time,time-frequency and coupled two domains is carried out using the SVM model.By introducing the distinguishing function,the normal vibration signal and fault signal are considered simultaneously and the divisibility of each feature is analyzed according to which features are further sorted.With the results of sorted features,the feature space is divided into two regions labelled with high and low distinguishing degree respectively.The SVM models according to the single feature and combined features in each region are established separately to diagnose the rubbing fault of rotor system.The results show that the recognition rate in term of single feature is positively correlated with the value of distinguishing function.The recognition effect of combined features is better than that of single feature,meanwhile the recognition effect of combined features selected in high distinguishing region is obviously better than that in low distinguishing region.In view of the rubbing fault samples in this paper,the mean fault recognition rate of random three-feature combination in high distinguishing region reach above 90%.In summary,the fault feature extraction and analysis method in this paper can provide a reference for the identification of complex faults.
作者
赵柄锡
冀大伟
袁奇
李浦
葛庆
ZHAO Bingxi;JI Dawei;YUAN Qi;LI Pu;GE Qing(School of Energy and Power Engineering,Xi’an Jiaotong University,Xi’an 710049,China;Shaanxi Engineering Laboratory of Turbomachinery and Power Equipment,Xi’an Jiaotong University,Xi’an 710049,China;Technology Development Branch,Turbine Company of Shanghai Electric Power Station Equipment Co.Ltd.,Shanghai 200240,China)
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2020年第1期75-84,共10页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(11872289)
西安交通大学自然科学基金重点资助项目(ZRZD2017025)
国家重点研发计划资助项目(2017YFB09602103)
关键词
碰磨故障诊断
时域分析
时频域分析
可分度函数
rubbing fault diagnosis
time domain analysis
time-frequency domain analysis
distinguishing function
