A Defect Diagnosis in Bearings of a Centrifugal Pump using Vibration Analysis

Abstract

This paper presents an overview of defect diagnosis in bearings of a centrifugal pump. The data obtained using vibration-based condition monitoring (VCM) technique was recorded at regular intervals. The analysis provided using conventional methods would then be used for pump fault prognosis and trend pump conditions. Having studied the conventional method of analysing results off-line, the research uses a VCM system to predict bearing faults on-line. Several techniques for pattern recognition were considered, including Feed Forward type Neural Network (FF-NN) and Recurrent Neural Networks (RNN). The author decided to adopt the Artificial Neural Network (ANNs) to propose a solution and classify bearing faults. Since bearing faults don’t begin to appear before prolonged pump operations, the faults on bearings were simulated using a test-rig pump where an electrical discharge machine (EDM) would generate pit marks on bearings and the vibration signals thus collected be fed into the neural networks. An easy method of designing neural network models is by using the MATLAB Neural Network Toolbox. To carry out the analysis, only MATLAB models that are specifically functional to vibration signals are chosen for pump bearing fault diagnosis.