Abstract No.F180202-231

Author name(s): Seyun HWANG, Jeeyeon HEO, Kyutack HONG, Janghyun LEE

Company: Inha University, Korea

This study discusses a fault detection method for an air blower installed in oil & gas process plants using pattern recognition technique accompanied with statistical analysis. Fault diagnosis is performed based on the time series signal measured from various sensors installed in the equipment. Considering the fault type of the compressor and the repair history, signals which can be an indicator of failure are selected. Multivariate statistical features are extracted from the time series and multidimensional features are projected in two-dimensional space using Principal Component Analysis (PCA), one of the dimensional reduction algorithms. During the diagnosis, features with high correlation with failure are selected from the multi-dimensional features to identify the possibility of failure. Using the projected two multidimensional features, we classify the fault types using Naive Bayesian classification. Finally, a case of fault diagnosis by applying the measured sensor data to the proposed fault diagnosis system is presented.

KEY WORDS: pattern recognition; fault diagnosis; cause diagnosis; data processing; feature extraction; condition based maintenance

All Rights Reserved. If you need the entire text, please contact the author or intl.exchange@ssnaoe.org