Purpose: This study aims to differentiate between false and true alarms by extracting features that distinguish vibration resulting from rotational speed and centrifugal pump fault.
Methods: We extracted time-domain statistical features and frequency-...
Purpose: This study aims to differentiate between false and true alarms by extracting features that distinguish vibration resulting from rotational speed and centrifugal pump fault.
Methods: We extracted time-domain statistical features and frequency-domain physical features.
Each feature was learned by OCSVM and visualized through the anomaly score.
Results: Time-domain statistical features showed poor classification performance. In contrast, frequency-domain physical features successfully classified fault status at low rotational speed.
Conclusion: Considering that the centrifugal pump is a rotating machine, diagnosing fault using the frequency domain physical characteristics is practical. Further study should focus on extracting improved features for diagnosing failure at a high rotational speed with acceptable performance.