Evaluation of Feature Extraction Techniques for Intelligent Fault Diagnostics of High-Pressure LNG Pump

The Liquefied Natural Gas (LNG) receiving terminal is designed to deliver a specified gas rate into a pipeline network, and High Pressure LNG pumps are crucial equipment because they determine the total supply capacity of natural gas in the terminal. Therefore, condition of HP-LNG pumps are regularly monitored and managed based on Condition Based Maintenance (CBM) technique. In general CBM system is composed of a number of functional capabilities such as data acquisition, signal processing, feature extraction, diagnostics, prognostics and decision reasoning. In this paper, a comparative study on evaluation of the performance of feature extraction techniques is carried out for intelligent fault diagnostics of HP-LNG pump using real industrial data. In order to estimate the abilities of feature extraction techniques, three methods such as Principal Component Analysis (PCA), Liner Discriminant Analysis (LDA) and Distance Evaluation Technique (DET) are employed and tested for the features based fault diagnostics. The accuracy of fault classification performance is estimated by using One-Against-All Multi-Class SVMs (MCSVMs) technique. The result shows that DET has a better capability than other conventional techniques as a feature extraction technique for fault diagnostics of HP-LNG pump.