Abstract: Aiming at the issue of uncertainty in the initial damage evolution of wind turbine blade spars with wrinkle defects,Gaussian probability distribution function is used to determine the distribution information of seven factors. The stress data of wrinkle defects under tensile loading is used to improve Sobol’s algorithm through Latin hypercube sampling,which enables the acquisition of sample points and the training of a BP neural network. The relative entropy of strain residual energy density at maximum and non-maximum damage sites is calculated using Kullback-Leibler divergence. These values serve as sensitivity response indexes for matrix cracking and fiber fracture,respectively. The results show that loading amplitude is the most sensitive factor for both matrix damage and fiber fracture,followed by fiber content,matrix-to-fiber modulus ratio,and wrinkle height-to-width ratio. However,the order of the matrix-tofiber modulus ratio and wrinkle height-to-width ratio is different. The related result shows that the blade spar’s material properties and the defect feature’s appearance have different effects on the initial damage mode. Finally,the finite element model of GFPR laminate with wrinkle is established. The simulation results verify the accuracy of the global sensitivity analysis method.