Because of different types of load, material properties deviation and construction errors, structures have initial defects inevitably. Therefore structural damages emerge easily and have strong randomness. At the same time, the ideal design model often has difference with structure in service. To most structures, the initial testing dates cannot be obtained, while this initial model is very important to structural damage detection. So the ideal model needs to revise. In this paper, elastic modulus, Poisson ratio and link section area are given as initial random defects and these defects obey normal distribution which can be constructed by Monte Carlo probabilistic design method. Firstly, the sensitivity parameters to structural response will be received by PDS technology from Ansys. Next, the square pyramid space grid models with random defects were obtained. Finally, given link element damage, using the method combined curvature mode difference with wavelet transform, the link element damage can be determined. Through analysis, the effects about the initial defects to damage detection will be obtained.