Time frequency distribution for vibration signal analysis with application to turbo-generator fault diagnosis

With the growing demand for improving reliability and performance of turbo-generator set as an important power supply in modern electric industry, the vibration signal processing and fault diagnosis techniques have been developed for dynamic feature extraction and pattern recognition. A new approach using wavelet transform and fuzzy network modeling is presented for vibration fault diagnosis of turbo-generator set. The wavelet transform decomposes the vibration signal in time domain into a two-dimensional function in time-scale plane, which analyzes the low-frequency component of signal with a wide duration function and analyzes conversely high-frequency component with short-duration function. The wavelet fuzzy method can generate dilation and translation parameters, and the network output vector is a linear combination of fuzzy wavelet basis functions. The approximation procedure can achieve better approximation of accuracy order than normal neural network and fuzzy system. The experiment results and analysis approve that the proposed approach is effective, improving the accuracy and reliability of fault diagnosis technology for rotating machinery vibration.