基于CEEMD和优化KNN的离心泵故障诊断方法  

Fault diagnosis method for horizontal centrifugal pump based on CEEMD and optimized KNN

作　　者：杨波[1,3] 黄倩 付强 朱荣生[1,3] YANG Bo;HUANG Qian;FU Qiang;ZHU Rong-sheng(National Research Center of Pumps,Jiangsu University,Zhenjiang 212001,China;China Nuclear Power Engineering Co.,Ltd.,Beijing 100840,China;Joint Laboratory of Intelligent Diagnosis Operation and Maintenance of Nuclear Power Pumps and Devices,Zhenjiang 212013,China)

机构地区：[1]江苏大学流体机械技术研究中心,江苏镇江212001 [2]中国核电工程有限公司,北京100840 [3]核电泵及装置智能诊断运维联合实验室,江苏镇江212013

出　　处：《机电工程》2022年第11期1502-1509,共8页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(U20A20292);江苏省重点研发计划资助项目(BE2018112)。

摘　　要：卧式离心泵实际测量中背景噪声含量较大,故障特征常被淹没,导致机械故障诊断效果较差,为了实时、精准地获得其运行状态,或对其进行故障诊断,提出了一种基于互补集合经验模态分解(CEEMD)和优化最邻近(KNN)算法的卧式离心泵机械故障诊断方法。首先,采集了卧式离心泵机械故障加速度信号,使用CEEMD对信号进行了一次分解,得到了本征模函数(IMF),采用相关系数法得到了IMF相关系数,确定了相关分量与不相关分量;其次,通过改进小波阈值去噪方法对不相关分量进行处理,提取了重构信号可分析的时频故障特征;最后,搭建了离心泵实验台,采用上述故障诊断方法对离心泵机械故障进行了分类诊断。研究结果表明:经CEEMD降噪后,信号评价指标信噪比(SNR)为2.2571,比原来的去噪方法提升了0.4381;优化后KNN分类对于卧式离心泵的机械故障诊断准确率可达96.7%,能够有效识别离心泵故障,达到智能诊断的目的。As the background noise content in the actual measurement of horizontal centrifugal pump was large,and the fault characteristics were often submerged,resulting in poor mechanical fault diagnosis effect.In order to obtain its operation state and fault diagnosis in current time and accurately,a mechanical fault diagnosis method of horizontal centrifugal pump based on complementary ensemble empirical mode decomposition(CEEMD)-optimized k nearest neighbor(KNN)was proposed.Firstly,the mechanical fault acceleration signal of horizontal centrifugal pump was collected,and the signal was decomposed once by CEEMD to obtain the intrinsicmode function(IMF).The correlation coefficient of IMF was obtained by the correlation coefficient method to determine the correlation component and uncorrelated component.Secondly,the uncorrelated component was processed by the improved wavelet threshold denoising method to extract the time-frequency fault features that could be analyzed by the reconstructed signal.Finally,a centrifugal pump experimental bench was built,and the above fault diagnosis methods were used to classify and diagnose the mechanical faults of the centrifugal pump.The results show that the signal-to-noise ratio(SNR)of the signal evaluation index after CEEMD noise reduction is 2.2571,which is 0.4381 higher than the original denoising method.After optimization,the accuracy of KNN classification for mechanical fault diagnosis of horizontal centrifugal pump can reach 96.7%,which can effectively identify faults and achieve the purpose of intelligent diagnosis.

关 键 词：叶片式泵 故障信号分解 互补集合经验模态分解 改进小波阈值降噪 优化最邻近算法分类 本征模函数 相关分量/不相关分量 

分 类 号：TH311[机械工程—机械制造及自动化] TH17