Fault Diagnosis and Isolation of an Electro-Pump using Neural Data Fusion

In this paper, the authors present successful experimental testing by Data Fusion method to detect and isolate five different physical faults on a 3-phase 380-V, WYE connected, two-pole 5.5 hp, 50 Hz, squirrel-cage induction motor connected to a water pump under its rated load. The tested faults included a bearing outer race, broken rotor bar, stator winding short circuit, unbalanced power source, and dry running faults of the connected water pump. The faults were applied separately in this paper; however, common mode faults testing results will be published in our future article. Temporal features are extracted from three-phase current and voltage signals using independent components analysis method or ICA; then features are fused adopting a neural network with the reference of the motor's healthy condition. Remarkable reliability analysis of the testing results is presented in this paper. Results indicate that neural data fusion method is a reliable non-intrusive diagnostic motor testing under normal loading.