원심펌프 고장진단을 위한 OCSVM기반 특징에 관한 실험적 연구

Abstract

Purpose: This study aims to distinguish false and true alarms by extracting features that identify the difference between vibration due to rotational speed and centrifugal pump fault. Methods: The time domain statistical features and frequency domain physical features are extracted. Each feature is learned by OCSVM and visualized by the anomaly score. Results: Time-domain statistical features shows poor classification performance. In contrast, frequency-domain physical features can classify the fault status at low rotational speed. Conclusion: Since the centrifugal pump is a rotating machine, diagnosing failure using the frequency domain physical characteristics is practical. Further study will extract better features to diagnose failure at a high rotational speed with reasonable performance. Keywords: Centrifugal Pump, Fault Diagnosis, One Class Support Vector Machine, Anomaly Score, Feature Extraction