Vibration Fault Detection and Analysis of Rotating Machinery Using Neural Network Techniques

To improve the limitation of applying traditional fault diagnosis method to the diagnosis of multi-concurrent vibrant faults for turbo-generator set in power system, a novel approach combining the wavelet transform with self-organizing learning array (SOLAR) system is proposed. The effective eigenvectors are acquired by binary discrete orthonormal wavelet transform based on multi-resolution analysis (MRA). These feature vectors then are applied to a SOLAR system for training and testing. SOLAR system has three advantageous over a typical neural network: data driven learning, local interconnections and entropy based self-organization. The synthesized method of recursive orthogonal least squares algorithm (ROLSA) and improved Givens rotation is used to fulfill the combined network structure and parameter initialization. By means of choosing enough practical samples to verify the proposed network performance and the information representing the faults is inputted into the trained network, and according to the output result the type of fault can be determined. Simulation results and actual applications show that the method can effectively diagnose and analyze the multi-concurrent vibrant fault patterns of turbo-generator set and the diagnosis result is correct. The method can be generalized to other devices' fault diagnosis.