Abstract: With complicated structure and severe operating environment, hydro-turbine generating units are often affected by hydraulic, mechanical and electric coupling factors, which degenerates the sensitivity of vibration fault features extracted using traditional methods. To overcome this problem, we present a new vibration feature extraction method based on CL3 adaptive multiwavelets (CL3-AMW) and local tangent space alignment (LTSA). This method improves the sensitivity through a three-step procedure: high dimensional vibrant feature set adaptive construction, feature selection and feature fusion. It has been applied to vibration signals collected from a rotating machinery system and hydro-turbine generating units, and compared with other methods. The results show the sensitivity of vibration fault features extracted by our method are more sensitive than those by other methods, thus providing useful data for more accurate fault diagnosis of hydro-turbine generating units.

Key words: vibration signal, feature extraction, CL3 adaptive multiwavelet, LTSA, hydro-turbine generating unit