Implementation of Artificial Intelligence-Based Fault Classification and Anomaly Detection: A Case Study on Hydraulic Centrifugal Pumps

Monitoring systems have been used in hydraulic centrifugal pumps for fault detection and predictive mainte-nance, which are critical to prevent costly unplanned shutdowns. Machine learning-driven monitoring systems that are based on pump vibration data have widely become popular in this context. The success of these techniques crucially depends on the quality of the vibration data received from the pump data acquisition system. In this regard, the paper first implements a proof-of-concept custom data acquisition unit mountable on the centrifugal pumps to log data related to the vibration, pressure, voltage, current, and temperature. Second, advanced machine learning techniques including Random Forest (RF), Multi-Layer Perceptron Classifier (MLPClassifier), Autoencoders (AEs), and Voting Classifier (VC) are applied to detect anomalies and classify 11 known faults. Performance metrics of the models are compared based on several tests on different experimental datasets. The accuracy of the implemented system has ranged from 98.5% to 100% regarding the classification of faults and detection of anomalies.