Fault pattern classification of turbine-generator set based on artificial neural network

By combining wavelet analysis and fuzzy theory, a new approach is presented for vibration fault diagnosis of rotating machine. The wavelet transform has become a powerful alternative for the analysis of nonstationary signals whose spectral characteristics are changing over time, since the widely used spectral analysis method provides only the frequency contents of the signals without providing the time localizations of the observed frequency components. A novel method based on the statistic rule is brought forward to determine the threshold of each order of wavelet space and the decomposition level adaptively, increasing the signal-noise-ratio and improving the performance of a traditional fault diagnosis method. The fuzzy wavelet basis functions can be specified by experts as traditional fuzzy systems. Furthermore, the architecture of wavelet fuzzy network can provide at least the same order of approximation error as neural networks. The improved least squares algorithm is employed to achieve the network parameters and the robustness of neural network is discussed. The practical diagnosis process for rotor vibration demonstrates that the wavelet fuzzy network can provide an effective way to diagnosis faults for turbo-generator set in power system, increasing the accuracy of the fault diagnosis for rotating machinery.