An Intelligent and Robust Fault Diagnosis System for Identification of Centrifugal Pump Defects in Frequency Domain Using Corrupted Vibration and Current Signatures

Historically, fault diagnosis techniques in industries were experience based. These techniques, however, are very tedious and involve a lot of human error. Therefore, intelligent methods need to be developed for the sustained operation of vital equipment like centrifugal pumps (CPs). In the present investigation, multiple independent and coexisting hydraulic and mechanical faults in a CP are attempted to be classified. The faults include blockages (discharge and suction), dry runs, impeller cracks and CP cover plate damages. The blockage faults are considered with varying severities. The current and vibration signatures are collected in time-domain by experimentally simulating the faults on the CP. These signatures are later converted into frequency domain. Support vector machine (SVM) classifier in conjunction with the Gaussian RBF kernel is used to develop the expert system for the fault diagnosis. To inspect the algorithm’s robustness, noisy/corrupted fault data is used to test the algorithm. The prediction accuracies thus obtained are compared with the non-corrupt data’s classification performance.