Affiliations: College of Mechanical and Transportation Engineering, China University of Petroleum-Beijing, Beijing, China
Correspondence:
[*]
Corresponding author. Wei Liang, College of Mechanical and Transportation Engineering, China University of Petroleum-Beijing, 102249 Beijing, China. E-mails: [email protected] and [email protected].
Abstract: As the dynamic kernel of natural gas pipeline infrastructure, centrifugal compressor unit and its status of health hold a crucial impact on the regular operation of the pipeline system. In order to keep track of the state of the centrifugal compressor unit, monitoring parameters are taken as the indexes for state evaluation. However, with the increasing complexity of monitoring systems, how to select appropriate parameters among numerous monitoring parameters and to establish a rational index system becomes an urgent problem in the process of state evaluation. In traditional, the indexes are selected with expert experience, ignoring the correlation between the monitoring parameters, which results in information redundancy in state evaluation. Besides, the monitoring parameters of auxiliary systems are usually not taken into consideration. For improving the performance of the index system, namely to reduce the information redundancy and to broaden the coverage, a time lag based correlation analysis model (TLCAM) is proposed in this paper for index selection. In the TLCAM, Erdem correlation coefficient (ECC), which takes the lag difference into account, is used to depict the coupling correlation between the monitoring parameters. Meanwhile, a neighborhood extreme value method (NEVM) is proposed to calculate the ECC. Hereafter, the monitoring parameters are classified through incorporating the advantage of information entropy into the density-peak based clustering (CDP). Consequently, an index system with less redundancy and wider coverage is established for the state evaluation of centrifugal compressor unit. At the end, correlation analysis and sensitivity analysis are performed on the two index systems respectively obtained from expert knowledge and TLCAM to compare their performance. The result shows that the index system established by using TLCAM is more sensitive to some early faults and contains less information redundancy. Moreover, it covers more critical components of the whole unit.
Keywords: Centrifugal compressor unit, state evaluation, index selection, correlation analysis, neighborhood extreme value method (NEVM), density-peak based clustering (CDP)
