CIESC Journal ›› 2023, Vol. 74 ›› Issue (7): 2979-2987.DOI: 10.11949/0438-1157.20230454

• Process system engineering • Previous Articles     Next Articles

Surge diagnosis method of centrifugal compressor based on multi-source data fusion

Ye XU1(), Wenjun HUANG2, Junpeng MI2, Chuanchuan SHEN2, Jianxiang JIN2()   

  1. 1.China Petroleum and Chemical Corporation, Beijing 100029, China
    2.College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, Zhejiang, China
  • Received:2023-05-08 Revised:2023-06-25 Online:2023-08-31 Published:2023-07-05
  • Contact: Jianxiang JIN

多源信息融合的离心式压缩机喘振诊断方法

徐野1(), 黄文君2, 米俊芃2, 申川川2, 金建祥2()   

  1. 1.中国石油化工股份有限公司,北京 100029
    2.浙江大学控制科学与工程学院,浙江 杭州 310027
  • 通讯作者: 金建祥
  • 作者简介:徐野(1971—),男,高级工程师,xuye@sohu.com
  • 基金资助:
    浙江省“尖兵”研发攻关计划项目(2022C01047)

Abstract:

The centrifugal compressor is the key power equipment for oil refining and chemical production. Once it breaks down, it may cause major plant's accidents. Therefore, online equipment monitoring and fault diagnosis throughout the life cycle is conducive to the continuous and stable operation of plant. This paper proposes a surge diagnosis method for centrifugal compressor based on multi-source data fusion such as flow, pressure and vibration. First, the vibration data is decomposed into a specified number of sub band signals by using empirical wavelet transform and the signal is reconstructed according to the correlation order. Convolutional neural network is used to pre-diagnose the reconstructed vibration signal, flow signal and pressure signal, and the final diagnosis is made using the weighted D-S evidence theory for the normalized diagnosis results of the three signals. Through the experiment of surge simulation test on centrifugal compressor, the diagnosis accuracy can reach 97.25% by using the multi-source fusion fault diagnosis method proposed in this paper. Compared with the use of single sensor data, this method significantly improves the fault tolerance ability of diagnosis, and it has higher diagnosis accuracy compared with other methods.

Key words: centrifugal compressor, muti-source data fusion, surge diagnosis, empirical wavelet transform, convolutional neural network, weighted D-S evidence theory

摘要:

离心式压缩机是炼油化工生产的关键动力设备,其一旦出现故障将可能造成重大生产事故,因此全生命周期的在线设备监测和故障诊断有助于化工生产的连续稳定运行。提出了一种基于流量、压力和振动等多源信息融合的离心式压缩机喘振诊断方法,首先利用经验小波变换将振动数据分解成指定数目的子频带信号并按照相关性排序进行信号重构,将重构的振动、流量、压力信号分别利用卷积神经网络进行预诊断,并将三种信号的归一化诊断结果采用加权的D-S证据理论进行最终诊断。通过离心式压缩机喘振模拟实验数据验证,利用提出的多源融合的故障诊断方法进行诊断,诊断精度可达97.25%,与使用单一传感器数据相比该方法显著提升了喘振故障诊断的容错能力,与其他多源融合方法相比诊断精度更高。

关键词: 离心式压缩机, 多源信息融合, 喘振诊断, 经验小波变换, 卷积神经网络, 加权D-S证据理论

CLC Number: