Vibration fault diagnosis based on Multi-scale EMD time-series similarity mining for hydroturbine

29th IAHR Symposium on Hydraulic Machinery and Systems

Xue, Y.G.^1 ; Wang, H.^2

Lanzhou Institute of Technology, Gansu Province, Lanzhou

730000, China^1

State Key Laboratory of Eco-hydraulics in Northwest Arid Region of China, Xi'an University of Technology, Xi'an, Shaanxi Province

710048, China^2

关键词: Characteristic curve;    Data mining algorithm;    Degree of similarity;    Discrete fourier transformation;    Non-stationary time series;    Turbine vibrations;    Vibration fault diagnosis;    Vibration faults;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/240/2/022016/pdf
DOI  :  10.1088/1755-1315/240/2/022016

来源: IOP

【 摘 要 】

Time-series similarity mining is an important method for vibration fault diagnosis of hydraulic turbine. In this paper, based on multiscale EMD frequency fuzzy neartude, a time-series similarity data mining algorithm is presented to solve the problem of similarity comparison between characteristic curves of vibration faults. Firstly, all high-dimension deformation data in time-series bank are pretreated by standardized multiscale EMD. The stationarity of series is promoted and the detailed information is reserved. Then, Discrete Fourier Transformation is carried out for the obtained multiscale IMF. Finally, the distances among time series are measured with fuzzy neartude of IMF component series. The degree of similarity among time series is also described. To test its effectiveness, the method is applied to the prototype hydraulic turbine vibration fault series. As its result shows, multiscale EMD tranquilizes the complex non-stationary time series. It conquers the problem of information loss in the process of data interception. At the same time, the method can help identify unit faults accurately, and classify different types of faults, it's discriminant accuracy is about 82.9%. Due to its low requirement for the number of data, and the efficiency in computing, the method is suitable for large-scale graphic series mining in hydraulic turbine fault diagnosis.