Abnormal Sound Detection Of Drain Hole For Wind Turbine Blade

The blade is a crucial part of wind turbine for acquiring wind energy.Due to long-term operation in harsh natural environment,it is easy to cause the cracking and falling of the main beam,surface skin,bonding materials,resulting in heavy economic losses and casualties.It is significant to carry out on-line defect detection for wind turbine blade.Compared with the contact detection methods such as acoustic emission,vibration and fiber Bragg grating,the non-contact acoustic detection method with sensors at the bottom of the tower has the advantages of flexible installation and convenient maintenance.However,the movement of drain hole for blade will produce abnormal sound randomly in practical,which causes serious interference to blade defect detection.Accordingly,this paper presents an abnormal sound detection method of drain hole for wind turbine blade.The specific research contents and results are as follows:(1)The characteristics and mechanism of abnormal sound for drain holes are studied.The causes and characteristics under complex working conditions for abnormal sound are analyzed combined with the structure of drain hole.The simplified motion model between rotation blade and receiver microphone is established according to the geometric size of the blade and the position of the microphone.The variation of the received frequency for abnormal sound is analyzed according to Doppler effect and the correctness of the conclusion is verified by simulation.(2)Aiming at the location of abnormal sound for drain hole under complex background noise,an adaptive location method based on multi band spectral subtraction is proposed.The spectrum characteristics of complex background noise and signal is explored and the periodic characteristic index is designed based on the spectral centroid.The location method of blade signal which may contains abnormal sound for drain hole are proposed.An improved spectral subtraction method is established to remove the strong interference in the mixed signal combined real-time signal-to-noise ratio and multi band strategy.The effectiveness of the method is verified by simulation based on real measured data.(3)Aiming at the randomness and diversity of abnormal sound of drain hole,a feature extraction method based on correlation empirical mode decomposition(EMD)and morphological filtering is proposed.EMD is conducted to decompose the abnormal sound of drain hole adaptively.The target IMF is selected based on the correlation coefficient between the decomposed intrinsic mode function(IMF)and the original signal.The reconstruction criteria for IMF is designed considering the frequency band expansion.On this basis,morphological filtering is introduced to optimize the time-frequency ridge characteristics of abnormal sound,and the recognition model based on support vector machine is established.The simulation results show that the method can effectively characterize and identify the abnormal sound events.(4)Aiming at the strong subjectivity and incompleteness of manual feature extraction,the recognition method of abnormal sound of drain hole based on convolution neural network is proposed.The non-positive convolution kernel group structure is integrated to extend the depth of the network.The ability of nonlinear transformation and feature extraction is improved.The cross entropy loss function is optimized with weight coefficient index to strengthen the fitting degree of a small number samples.The batch normalization layer is introduced to adjust the data distribution between layers to improve the training performance of the network.The experimental results show that the optimized convolutional neural network model has higher accuracy and faster convergence speed to recognize the abnormal sound of drain hole.