Research On The Fault Feature Extraction And Classification Techniques For Gearboxin Wind Turbine

With the proposed environmental protection slogans, the clean, inexhaustible energy: wind has attracted people’s attention. Now the wind resources are mainly used for wind power generation. Wind power generator as the link between the mechanical energy exchange energy and electricity plays a crucial role in wind power generation. If the generator problems, it will directly affect the normal operation of electric power system. This article mainly studies the mechanical fault of gearbox of wind power generator.Mechanical fault signal is not stationary, nonlinear characteristic. In view of such forward characteristic analysis of a multifractal spectra and approximate entropy based on the combination of quantitative. And the results of the quantitative analysis are used as inputs of fuzzy clustering for pattern recognition, and the application of this method to the fault diagnosis of gearbox.First of all, introduce the working principle of wind power generation and part. Needle common fault for gearbox of wind power generator are described, the vibration mechanism of gear box and paper.Secondly, the signals are a series of processings, including noise elimination are part of signal decomposition of signals, feature extraction, and the amount of. On the basis of empirical mode decomposition method is presented in combination of EEMD-ICA, so that the filtering effect is excellent; LMD decomposition, decomposition algorithm is better than that from the empirical mode, which overcomes the end effect in a certain extent; multifractal spectra and approximate entropy are extracted by combining characteristics can be expressed signal feature a more comprehensive.Then, the feature as the input of fuzzy C clustering, pattern recognition, and lay the foundation for the mechanical fault diagnosis.Finally, the methods used for the experimental platform for collecting data, through the analysis of the data to verify the effectiveness of the above proposed method. EEMD-ICA filter can make the signal to achieve a good filtering effect; multifractal combined with the approximate entropy can be more comprehensive to make quantitative analysis to absorb; fuzzy C clustering can be a better clustering effect.