Practical application of fuzzy technology for experiment data analysis of rotating machine

The combination of wavelet transform and soft computing technique has led to the development of wavelet fuzzy network. Based on rigorous multi-resolution wavelet analysis, a fuzzy inference system is proposed to improve fault diagnosis performance for turbo-generator set. To obtain the accurate signal characteristics, the statistic rule is utilized to determine each order threshold of wavelet subspace and decomposition level adaptively, increasing the signal-noise-ratio. The effective eigenvectors are acquired by wavelet transform and the fault patterns are classified by fuzzy inference system. The mathematics model for turbo-generator fault diagnosis is established and the improved least squares algorithm is used to complete network parameters initialization and the network robustness is discussed. Different from wavelet neural network, the fuzzy wavelet basis functions can be specified by experts as traditional fuzzy system. By means of sample training phase, the wavelet fuzzy network indicates its good convergence property of training error. Furthermore, the architecture of wavelet fuzzy network can provide at least the same order of approximation error as neural network. The experiment results demonstrate that the proposed method can effectively diagnose vibration accident for turbo-generator.