强噪声条件下基于EMD-AE优选特征的离心泵多故障诊断方法  

Multi-fault diagnosis method for centrifugal pumps based on EMD-AE optimal selected features under strong noise conditions

作　　者：向明胜 冯坤[1,2] 贾韶辉[3] 赵衍 XIANG Mingsheng;FENG Kun;JIA Shaohui;ZHAO Yan(College of Mechanical&Electrical Engineering,Beijing University of Chemical Technology,Beijing 100029,China;State Key Lab of High-end Compressor and System Technology,Beijing University of Chemical Technology,Beijing 100029,China;Science and Technology Research Institute Branch Co,National Petroleum and Natural Gas Pipeline Network Group Co.,Ltd.,Tianjin 300457,China;College of Information Science and Technology,Beijing University of Chemical Technology,Beijing 100029,China)

机构地区：[1]北京化工大学机电工程学院,北京100029 [2]北京化工大学高端压缩机及系统技术全国重点实验室,北京100029 [3]国家石油天然气管网集团有限公司科学技术研究总院分公司,天津300457 [4]北京化工大学信息科学与技术学院,北京100029

出　　处：《振动与冲击》2024年第23期66-74,共9页Journal of Vibration and Shock

摘　　要：工业离心泵故障诊断中常常受到噪声的干扰,针对这一问题,提出一种强噪声条件下基于经验模态分解(empirical mode decomposition,EMD)和自编码器的优选特征方法。首先利用补偿距离评估技术确定出有效的时频特征,然后通过EMD处理,得到包含不同尺度和频率特性的模态分量。通过能量比变异系数确定出有效的分析分量,通过提取出所选分量的有效特征,拼接构造高维的深度特征。最后通过自编码器对深度特征做降维处理,进一步优选特征,得到最终的故障敏感特征,完成特征提取。选用支持向量机作为故障诊断模型,通过工业离心泵多故障数据进行对比试验。结果表明所提方法在信噪比为-5 dB、-7 dB和-10 dB强噪声干扰条件下,准确率较传统时频特征分别提高了6.13%、7.46%、12.00%。该方法有较强的抗噪声的能力,在噪声干扰下能有效提取表征设备状态的敏感特征。In fault diagnosis of industrial centrifugal pumps,noise interference is often encountered.Here,aiming at this problem,a feature selection method based on empirical mode decomposition(EMD)and autoencoder(AE)under strong noise condition was proposed.Firstly,effective time-frequency features were determined using compensation distance evaluation technique,and then modal components containing different scales and frequency characteristics were obtained through EMD processing.Effective analysis components were selected with energy ratio variation coefficients,and their effective features were extracted and spliced together to construct high-dimensional deep features.Finally,an autoencoder was utilized to reduce dimensionality of these deep features,further select features and obtain final fault-sensitive features for completing feature extraction.Support vector machine was taken as the fault diagnosis model and contrastive experiments were conducted with multi-fault data of industrial centrifugal pumps.The results showed that the proposed method can improve correctness rates by 6.13%,7.46%and 12%,respectively compared to traditional time-frequency features under conditions of strong noise interferences with signal-to-noise ratios of-5 dB,-7 dB and-10 dB,respectively;this method has a stronger anti-noise ability and can effectively extract sensitive features characterizing device status under noise interference.

关 键 词：强噪声 离心泵 经验模态分解(EMD) 优选特征 敏感特征 

分 类 号：TH311[机械工程—机械制造及自动化]