A new method for intelligent fault diagnosis of hydroelectric generating unit

There are considerably economical and social merits in the condition monitoring and fault diagnosis of hydroelectric generating unit (HGU). After the analysis on shortages in conventional techniques of signal processing and fault diagnosis, a new method for intelligent fault diagnosis of HGU based on compound feature extraction and radial basis function neural network (RBFNN) is proposed. Vibration or pressure pulsation signals from different parts of HGU are decomposed into different frequency bands via wavelet transform. Relative energy features are extracted after denoising. The influences of the process parameters' variations on the stability state are evaluated and quantified via correlation analysis, and relationship symptoms are obtained. Compound feature containing abundant fault information with several parameters is then formed and input into RBFNN based diagnosis system to determine the fault type and severity degree. Results of engineering application show that this proposed method can identify the faults relevant to the stability of HGU feasibly and efficiently.