西南石油大学学报(自然科学版)

• 石油机械工程 • 上一篇    下一篇

泥浆泵液力端故障诊断技术研究

张志东1,2 *,艾志久1,郑伟3,李奔4,钟功祥1   

  1. 1. 西南石油大学机电工程学院,四川成都610500;2. 中国石油川庆钻探工程有限公司安全环保质量监督检测研究院,四川广汉618300;
    3. 中国石油长城钻探工程有限公司钻具公司,辽宁盘锦124000;4. 中国石油长城钻探工程有限公司装备部,北京朝阳100101
  • 出版日期:2015-10-01 发布日期:2015-10-01
  • 通讯作者: 张志东,E-mail:zhangzhidong_0109@126.com

Study on Fault Diagnosis Technology for Fluid End of Drilling Pump

Zhang Zhidong1,2*, Ai Zhijiu1, Zheng Wei3, Li Ben4, Zhong Gongxiang1   

  1. 1. School of mechatronic Engineering,Southwest Petroleum University,Chengdu,Sichuan 610500,China
    2. CCDC Safty Environment Quality Survellance &Inspection Research Institute,Guanghan,Sichuan 618300,China
    3. Drilling Tool Company of CNPC Greatwall Drilling Company,Panjin,Liaoning 124000,China
    4. Equipment Department of CNPC Greatwall Drilling Company,Chaoyang,Beijing 100101,China
  • Online:2015-10-01 Published:2015-10-01

摘要:

通过对现有的泥浆泵液力端故障诊断技术的分析研究,并结合泥浆泵的结构及工况特征,提出了振动信号统
计指标与神经网络相结合的液力端故障诊断方法。该方法选取振动信号的有效值、方差、峰值指标、脉冲指标、峭度指
标和裕度指标作为表征液力端振动信号的特征指标;采用动态数据采集仪、压电式加速度传感器采集振动测试信号,
并计算得出振动信号平均特征量;然后通过对振动信号特征指标的归一化处理,构建BP 网络和设置网络参数,将经
归一化处理后的时域统计指标作为训练样本,输入到构建的BP 网络中进行网络训练;经过训练,使BP 网络满足预定
的精度要求。现场应用诊断误差分别为:0.007 7,0.017 9,0.017 7,0.021 6,说明构建的BP 网络的性能能够满足故障
诊断要求。利用统计指标和BP 神经网络结合的故障诊断方法,对泥浆泵故障诊断具有较准确的识别效果,可应用于
泥浆泵液力端的故障诊断。

关键词: 泥浆泵, 液力端, 统计指标, 神经网络, 故障诊断

Abstract:

The fault diagnosis method of fluid end that combines the statistical indexes and neural network is proposed in this
paper based on the analysis of fault diagnosis methods for fluid end of mud pump and the features of structure and operating
condition of mud pump. Firstly,the following indexes of vibration signal are selected to illustrate the characteristic index:
effective value,variance,peak index,impulsion index,kurtosis value and margin index,etc. At the same time,vibration
testing signal is collected by using a dynamic data acquisition instrument and piezoelectric acceleration sensors and the average
index for vibration signal is caculated. In order to construct the BP neural network and set network parameters,the selected
characteristic indexes are normalized. As the training samples,then these indexes are put into the BP neural network for training,
after which the constructed BP neural network can entirely meet the set training precision requirement. The corresponding
diagnostic errors of different plungers in field are 0.007 7,0.017 9,0.017 7 and 0.021 6,which shows that the constructed BP
network can reach the demands of plunger fault diagnosis. Therefore,this method may accurately diagnose fault of fluid end
of mud pump,which can be applied to engineering practice.

Key words: mud pump, fluid end, statistical indexes, neural network, fault diagnosis

中图分类号: