Analysis And Application Research Of Noise From Wind Turbine Blades

Due to the increasingly serious environmental problems and the shortage of non-renewable resources,how to efficiently use wind energy for power generation has become an important re-search content.As the main equipment for wind power generation,wind turbines are mostly built in places with harsh environments such as seasides and mountains.The blades are easily damaged due to being exposed to the outdoors for long term.As the demand for wind power generation increases,the size of the blades continues to increase.At the same time,it also increases the risk of failure.Blade failure is one of the main threats to the safe and stable operation of wind turbines.This dissertation focuses on the fault detection of wind turbine blades using emitted noise.In order to analyze the difference between acoustical signals emitted from faulty and normal blades,numerical simulations are carried out over the COMSOL Multiphysics simulation platform.When there is a small hole in the blade,the power level of abnormal sound from the damaged blade is higher than that of normal blade.Thus,the blade damage can be early warned when the increase of the emission acoustic energy occurs.Additionally,all the natural frequency values tend to decrease.For the first low-order modes,the natural frequency values of the damage blade decrease a little while those corresponding to high-order modes decrease much more.According to the acoustical field directivity,the maximum sound pressure level is emitted along the blade tip direction where it is an optimal placement direction for a microphone array.Although Supervisory Control And Data Acquisition(SCADA)systems can be utilized for detection of damage in wind turbine blades,it has not yet spread to all wind farms and is suitable for detection of blade late fault.Thus,this dissertation proposes a blade fault detection scheme based on spatial-temporal processing implemented to data received by a contactless microphone array.Considering the weak abnormal signal emitted from the damaged blade in an early stage,direction of arrival(DOA)and beamforming techniques are utilized to improve signal to noise ratio(SNR)of the received signal.Among the algorithms analyzed in this dissertation,numerical and experimental results have shown that the combination of the DOA estimation algorithm based on sparse Bayesian learning(SBL)and the conventional beamforming(CBF)algorithm can get the best signal enhancement performance.Further,time-frequency analysis is combined with Support Vector Machine(SVM)for detec-tion of wind turbine blade faults.Although the SVM-based framework performs well in detection of damaged blades,the man-made extracted features are not perfect and complete.Thus,the classi-fication algorithm based on Convolutional Neural Networks(CNN)is proposed in this dissertation for the CNN can automatically extract features from inputs.The structure of CNN refers to Deep Complex Unet(DCUnet)whose the deconvolution layer is replaced by a fully connect layer.The experimental results from three wind farms have verified the effectiveness of the proposed blade damage detection method.