Fault Signal Feature Extraction Of Large-Scale Wind Turbine Based On Blind Source Separation

Recently, with rapid development of domestic wind power and more attention hasbeen paid to green clean energy, besides, because of poor working condition ,once windturbines go wrong, which will cause such difficulties as high cost and difficult to mend.Therefore, people begin to focus on safety problem of wind turbines, and it is more andmore important to accurately analysis and classifies the faults of wind turbine.In this paper, the mechanical principles of large-scale wind turbine is started, blindsource separation is used. This paper builds the mathematical model of blind sourceseparation, analysis of the detachable separation conditions and the uncertainty of theresults, by studying the objective function and optimization algorithm. This paper mainlyfocuses on JADE algorithm, which has such advantages as fast convergence and goodrobustness. This algorithm is achieved by programming. Algorithm performances excellentproperties in signal separation in the case of pure noise-free. After adding the white noiseand impulse noise to the modulation signal, in the case of almost impossible to identify thesource signal in the observed signal, Algorithm can extract the signal waveform of thesource from strong background noise; the results show the stability and effectiveness of thealgorithm.In fault feature extraction of wind turbine, this paper designs the overall program offeature extraction based on blind signal separation. By analyzing the main components'fault characteristics, this paper gets simplified formula of bearing's characteristicfrequency and optimizes the measuring point for wind turbine. This program collects dataon two wind turbine bearing failure were mainly which are rolling bearing wear and motorside inner ring crack fault by vibration analyzer. This paper first analysis the signals intime domain and frequency domain, Find that fault signal, especially early fault signal arealmost completely submerged in the machine running signal and difficult to identify; Evenin the spectrum, the fault signal characteristics is still not obvious. In this paper, blindsource separation and spectrum analysis technology are combined to extract the fault signalfeature extraction on the measured signal; the results show that even if the fault signal is submerged in background noise and the machine running signal, it's still able to extract thefault feature. The results show that the feature extraction program is effective.