Automatic detection of cracked rotors combining multiresolution analysis and artificial neural networks

In the maintenance of motor driven systems, detection of cracks in shafts play a critical role. Condition monitoring and fault diagnostics detect and distinguish different kinds of machinery faults, and provide a significant improvement in maintenance efficiency. In this study, we apply the discrete wavelet transform theory and multiresolution analysis (MRA) to vibration signals to find characteristic patterns of shafts with a transversal crack. The feature vectors generated are used as input to an intelligent classification system based on artificial neural networks (ANNs). Wavelet theory provides signal timescale information, and enables the extraction of significant features from vibration signals that can be used for damage detection. The feature vectors generated for every fault condition feed a radial basis function neural network (ANN-RBF) and apply supervised learning designed and adapted for different fault crack conditions. Together, MRA and RBF constitute an automatic monitoring system with a fast diagnosis online capability. The proposed method is applied to simulated numerical signals to prove its soundness. The numerical data are acquired from a modified Jeffcott Rotor model with four transverse breathing crack sizes. The results demonstrate that this novel diagnostic method that combines wavelets and an artificial neural network is an efficient tool for the automatic detection of cracks in rotors.