Skip to Main Content
The high-capacity turbine-generator set has been widely used in power system as an important power supply and its operating condition is under mal-condition, so keeping it running in safety status is essential. A novel approach using wavelet neural network is proposed for transient vibration signal processing and fault pattern classification. In signal acquiring, the occurrence of transient signal makes the waveform nonstationary, especially during the start-up of turbo-generator. By means of wavelet transform, the transient signal can be decomposed into series of wavelet subspaces, each of which covers a specific octave frequency band in time-frequency. The effective eigenvectors are acquired by orthonormal wavelet transform based on multi-resolution analysis, which is called feature extraction. These feature vectors are applied to the neural network for training and testing. The neural network has three advantageous: data driven learning, local interconnections and good convergence property. The improved training algorithm based on recursive orthogonal least squares is utilized to accomplish network parameter initialization. By means of proper samples selection and network parameter adjustment, the fault pattern can be determined from the network output values. The simulation results and applications show that the proposed method is effective and the diagnosis result is correct.