Bearing Defect Diagnosis: An Approach for Manufacturing Industries Using Wavelet Transform Based Features

Abstract

To maintain reliability in the manufacturing units, industries have concentrated their attention on the condition based maintenance. Fault detection and diagnosis are the two of three condition based maintenance mainstays. Bearing is one of the most important and essential components of the rotating machines. Hence, the researchers have shown their interest in bearing fault detection and diagnosis from the last few years. They mainly use bearing vibration as fault characteristics for fault detection and use various optimization techniques for bearing fault classification. As the process of fault classification involves few neural network (NN) geometry and parameters, it’s not quite easy. There is no derived formulation to select the optimal values for the network parameters. These parameters have the direct impact on reliability of fault diagnosis calculation. This paper investigates the effects of neural network geometry and parameters on rolling element bearing fault diagnosis. The vibration signals are recorded for healthy bearing, bearing with inner race fault, outer race fault and rolling element fault and surface roughness fault produced using an EDM (Electrical Discharge Machining). The vibration features are calculated using wavelet transform feature extraction methods. Rolling element bearing faults are classified as same using back-propagation neural network (BPNN) and the results are simulated.