Classifying Two Primary Bearing Defect Causes Via the Highest-Energy Node in Wavelet Packet Decomposition

Abstract

The current research focuses on the study of two main causes of bearing defects: load unbalance and bearing improper lubrication using Dspace 1104 card for three stator current signals acquisition. This study suggests a straightforward and effective technique for identifying and categorizing two different kinds of defects. It consists of introducing the current space vector (CSV) analysis technique to avoid loss of information between the three stator current signals; the resulting signal is then processed by wavelet packet decomposition (WPD) to calculate the energy of the final level WPD nodes. The node containing the highest energy values will be selected to train the Multilayer Perceptron Neural Network (MLP-NN) classifier implemented by round-robin cross-validation technique. The results confirm the efficiency of the proposed procedure in bearing causes defects classification with an average accuracy of 100% for the tests and 99.88% for the training.