基于小波系数功率谱的潜油电泵偏磨故障诊断  

Partial friction fault diagnosis of electric submersible pump based on power spectra of wavelet coefficients

作　　者：姚诚[1,2] 刘广孚[3] 李忠国[4] 陶凤阳[3] 

机构地区：[1]中国石油大学(华东)石油工程学院,东营257061 [2]中国石化胜利油田胜利采油厂,东营257051 [3]中国石油大学(华东)信息与控制工程学院,东营257061 [4]江苏科技大学机械工程学院,镇江212003

出　　处：《仪器仪表学报》2011年第8期1757-1762,共6页Chinese Journal of Scientific Instrument

基　　金："十二.五"国家科技重大专项(2011ZX05024-002-009)资助项目

摘　　要：潜油电泵偏磨故障的诊断对于油田的安全生产具有重要意义。根据潜油电泵碰磨的理论分析,观察了振动信号小波系数的特点,提出了采用小波分解系数功率谱的层内最大值作为特征参数进行偏磨诊断的思想。通过交叉验证,以5次测试的平均识别率作为评价指标,使用改进粒子群算法对支持向量机的参数进行了优化。在径向基宽度为-0.310 9,惩罚参数C为429.127 8时,平均识别率最高为90%。研究结果表明,在潜油电泵偏磨故障的诊断中,小波系数功率谱优于小波分解后层内系数的傅里叶变换最大值和层内系数的方差。使用小波系数功率谱参数可以成功地实现潜油电泵偏磨诊断。Diagnosing the partial friction fault of Electric Submersible Pump （ESP） is very important in oil field. According to the theoretic analysis of ESP collision and friction, after observing the wavelet coefficient characters of vibration signal, a diagnosis idea is proposed, which uses the maximums of wavelet coefficient power spectra in every layer as the features to diagnose the partial friction of ESP. With cross-validation, the average recognition rate of five tests is taken as the evaluation indicator, and improved panicle swarm optimization algorithm is used to optimize the parameters of support vector machine （SVM）. The highest average recognition rate is as high as 90% when radial basis function width is -0. 310 9 and penalty parameter is 429. 127 8. Research results indicate that the wavelet coefficient power spectra are more competent for partial friction diagnosis than the maximums of wavelet coefficient Fast Fourier Transform （FFT） and the variances of wavelet coefficients in every layer. Wavelet coefficient power spectrum parameters can achieve good performance in partial friction fault diagnosis of ESP.

关 键 词：振动 小波 谱估计 粒子群优化 支持向量机 偏磨 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]