Bearing Fault Diagnosis Based On W P T And Automatic Reconstruct The Raw Signal With FF And BRA Of Artificial Neural Networks

Abstract

The intricate signal comprised of a machine component, that has several sources of vibration generation, leads to further investigation in signal processing and fault classification of ball bearing to improve maintenance systems efficiency. In this paper, we present an approach to extract time-domain features by wavelet packet WPT technique that is used to obtain the restored signal automatically from the original signal analyzed previously relying on robust energy changing of the WPT’s coefficients. The training and testing phases are conducted by fed the features as samples into Bayesian Regularization Algorithm of neural networks BRANNs. The effectiveness of proposed method is evaluated by comparison with FFNNs. The observation results from the experimental data proved that the time feature and BRANNs are very effective for classifying the ball bearing conditions as IRF, ORF, BSF and healthy bearing. Consequently, this approach has the capability to realize categorization for the signals of rotating machines in fault diagnosis systems.