关键词: 经验模态分解, Dijkstra 算法, 风机叶片, 暂态缺陷, 声纹信号

Abstract: Due to the nonlinear characteristics of the acoustic emission signal of wind turbine blades in the propagation path,it is difficult to reflect the vibration characteristics of wind turbine blades through residual voiceprint signals,resulting in low accuracy of temporary defect voiceprint recognition of wind turbine blades. Therefore,a wind turbine blade temporary defect voiceprint recognition method based on dense convolution and Dijkstra is proposed. Frame the voiceprint signals of wind turbine blades,apply Dijkstra algorithm to calculate the shortest path between voiceprint signal frames starting from a certain frame,and align the original voiceprint signals of wind turbine blades. Extract sound sub-signals from the original voiceprint signal using empirical mode decomposition algorithm,and use sub-signals that conform to the modal components as modal components to obtain residual voiceprint signals of wind turbine blades, which directly reflect the vibration characteristics of wind turbine blades. Determine strongly correlated modal vectors based on local minima of modal component energy and reconstruct voiceprint signals. Combine the reconstructed voiceprint signal with dense convolution to construct a recognition model,and optimize it through a combination of cross entropy loss function and loss function to output the recognition results of wind turbine blade defects. The experimental results showed that the designed method identified the frequencies of voiceprint signals for normal blades,cracked blades,and worn blades,which were 3 800 Hz,4 017 Hz,and
3 972 Hz,respectively,which were in line with the actual situation;the designed method can accurately identify single and multiple defects with high recognition accuracy in defect recognition of different degrees of defects.

Key words: empirical mode decomposition,Dijkstra algorithm,fan blades,temporary defect,voiceprint signal