Finding Dominant Parameters For Fault Diagnosis Of a Single Bearing System Using Back Propagation Neural Network

Bearing is a component that affects the performance of a rotating machine in order to be used properly. The purpose of using ball bearing is to reduce rotational friction and support radial and axial loads. Diagnosis of bearing vibration data is an important process to identify either the bearing is normal or defect. In this paper, we consider a single bearing system that consists of one normal or defect Fan End (FE) bearing in which an accelerometer is attached to both normal and defect FE bearing to capture the vibration data. These vibration data are grouped into several samples data. Five parameters are extracted from each sample to be used as the input in neural network model and class of target output is either normal or defect. Those five parameters are standard deviation, skewness, kurtosis, absolute mean and root mean square. Our objectives are to find the dominant parameters and the best model for fault diagnosis of single bearing system. It is important to find the dominant parameters since the selection of dominant parameter can reduce the pre-processing data phase. According to our observation that the dominant parameter is either of standard deviation, absolute mean or root mean square and the best BPNN model is the model with one input neuron and one neuron in hidden layer. Index Term -Back Propagation Neural Network; Fault Diagnosis; Bearing System; Dominant Parameters.