基于紧致型小波神经网络的往复泵故障诊断-北京大学学报(医学版).PDFVIP

150 基于紧致型小波神经网络的往复泵故障诊断 2013 年10 月 文章编号：1006-1355(2013)05-0150-05 基于紧致型小波神经网络的往复泵故障诊断 1 2 1 赵志华 ，吴 力 ，殷海双 （1. 东北石油大学 电气信息工程学院，黑龙江 大庆 163318； 2. 大庆油田天然气分公司 培训中心，黑龙江 大庆 163412 ） 摘 要：为了对往复泵的故障进行正确诊断，提出基于紧致型小波神经网络的往复泵故障诊断方法。以往复泵单 个泵缸内的压力信号作为系统特征信号通过小波包分解来提取故障特征向量，同时将此特征向量作为小波神经网络 的输入，利用小波神经网络对故障做进一步的精确实时诊断。通过对往复泵液力端多故障诊断实例的检验表明，该系 统故障诊断正确率达到94 % 以上。 关键词：振动与波；小波神经网络；往复泵；故障诊断；小波包 中图分类号：TP306 文献标识码：A DOI 编码：10.3969/j.issn. 1006-1335.2013.05.032 FaultFault DiagnosisDiagnosis TechnologyTechnology ofof ReciprocatingReciprocating PumpsPumps BasedBased onon CompactCompact WaveletWavelet NeuralNeural NetworkNetwork ZHAO Zhi-hua 1, WU Li 2 , YIN Hai-shuang 1 ( 1. Academy of Electric Information Engineering, Northeast Petroleum University, Daqing 163318, Heilongjiang China; 2. Training Center, Natural Gas Branch of Daqing Oilfield, Daqing 163412, Heilongjiang China ) AbstractAbstract :: A method of fault diagnosis for reciprocating pumps was proposed based on the compact wavelet neural network. In this method, the pressure signal of a single cylinder of the reciprocating pump was used as the characteristic signal of the system to extract the feature vector of the faults by means of the wavelet packet decomposition. At the same time, this feature vector was employed as the input signal of the wavelet neural network to determine the type of the fault. The examples of faults diagnosis at the fluid end of the reciprocating pump show that the correctness rate of the system fault diagnosis can exceed 94 %.