A method of convolutional neural network based on frequency segmentation for monitoring the state of wind turbine blades

Wind turbine blades are prone to failure due to high tip speed, rain, dust and so on. A surface condition detecting approach based on wind turbine blade aerodynamic noise is proposed. On the experimental measurement data, variational mode decomposition filtering and Mel spectrogram drawing are conducted first. The Mel spectrogram is divided into two halves based on frequency characteristics and then sent into the convolutional neural network. Gaussian white noise is superimposed on the original signal and the output results are assessed based on score coefficients, considering the complexity of the real environment. The surfaces of Wind turbine blades are classified into four types: standard, attachments, polishing, and serrated trailing edge. The proposed method is evaluated and the detection accuracy in complicated background conditions is found to be 99.59%. In addition to support the differentiation of trained models, utilizing proper score coefficients also permit the screening of unknown types.