Abstract: In order to accurately identify the location and length of surface cracks on wind turbine blades,a crack identification method based on wavelet packet decomposition-singular value decomposition-nuclear limit learning machine（WPT-SVD-KELM）is proposed.The crack identification experiment platform of a typical blade was built,modal experiments and pitch experiments of normal blades and cracked blades were carried out,and vibration signals of normal blades and cracked blades under different working conditions were obtained. The frequency response function is used to study the influence of the crack position on the amplitude-frequency response of the vibration signal,so as to accurately identify the position of the crack on the blade surface. The WPT-SVD is used to extract the timefrequency characteristics of the wind turbine blade surface crack vibration signal,and the parameter kris defined to characterize the change of the crack length. The characteristic parameters are imported into the optimized KELM to identify the length of surface crack on the wind turbine blade. The research results have important guiding significance for the identification of surface cracks on large wind turbine blades.