Machine learning based health assessment model for high pressure output pumps in LNG terminals

• Published in: IOP conference series. Earth and environmental science Vol. 461; no. 1; pp. 12085 - 12090
• Main Authors: YuyunZeng, Zhu, Wenbo, Yang, Jingjie, Xie, Guangyao, Liu, Jingquan
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.04.2020
• Subjects: Anomalies; Feature extraction; High pressure; Learning algorithms; Liquefied natural gas; Machine learning; Machinery condition monitoring; Maintenance management; Mathematical models; Pressure; Pumps; Statistical analysis; Terminals; Time domain analysis; Wavelet analysis
• ISSN: 1755-1307; 1755-1315
• DOI: 10.1088/1755-1315/461/1/012085

High pressure output pumps are one of the critical equipment in LNG terminals. Since the health condition of high pressure output pumps has a direct influence on production capability of the terminal, health assessment for these pumps in real time plays an important role on guaranteeing efficient productivity of LNG terminals. Using condition monitoring data, a machine learning based health assessment model for high pressure output pumps is proposed. Health features are constructed based on time domain statistical analysis and wavelet packet decomposition, and a SVR model is trained to calculate a health index from extracted features. Actual operating data in Qingdao LNG terminal are used for model validation. Results show that the calculated health indices are sensitive to faults and anomalies of the pumps, and are good indicators of pump health status. The proposed model also shows capability of early warning for some sudden failures, which can be valuable in the operation and maintenance management of LNG terminal equipment.