Investigations On Bearing Vibration Detection And Fault Diagnosis Of Wind Turbine

Wind energy as a clean, widely distributed, abundant deposited, renewable energy, itefficient use will be a major significance for our country of single energy structure, theenergy distribution. Currently, an important reason for hindering development of windpower is that fault investigation and detection. According to the data, the troubleshootingand maintenance about wind power accounts for2%of wind power total expenditure,while wind turbine drive train bearing as an important source of the fault and sensorsdetect point is particularly important for turbine fault detection system.Aiming at the vibration of the wind turbines, this paper uses the EMD technique toextract fault feature and uses vector machine (SVM) to classify the diagnosis. Vibrationdata is from field data of Jilin Baicheng Longbei Wind Field, and the failure data is fromCase Western Reserve University in the United States. For data denoising, the waveletdenoising, the EMD de-noising, envelope demodulation denoising, the wavelet envelopedemodulation denoising, the EMD envelope demodulation denoising processingrespectively, are used to process field data. Finally the wavelet envelope demodulationmethod gets the best denoising effect. For feature extraction of the EMD false IMFproblems, on the basis of the traditional energy analysis method, this paper considers thecubic spline interpolation error and the effect of envelope fitting error. According to thecharacteristics of wind vibration acquisition sampling frequency, traditional energyanalysis method is improved to calculate using multi-layer. And divergence of spectrumparameters as an auxiliary identification method also used to identify false component,aiming to get more accurate fault characteristics. This method is compared with thecorrelation coefficient method and the KL divergence method has a better stability. Faultdiagnosis respectively Vapnik support vector machine (SVM) classification is adopted tofault characteristic signal recognition, the RBF kernel function utilized, parameteroptimization in terms of the overlapping grid authentication method, genetic algorithm,particle swarm optimization (pso) algorithm for parameters optimization, theclassification of the final effect is compared and analyzed.