Abstract: Static response (displacement, strain, etc.) can hardly be recorded by a group of sensors installed on the structure in the inversion analysis of practical problems, while the dynamic characteristics (frequency, mode) and dynamic response (acceleration, velocity, dynamic displacement) of the structure can be easily acquired by sensors in practical problems. In this paper, the objective function of the inversion analysis model is constructed based on frequency residuals and modal assurance criteria. Combining the advantages of dynamic extended finite element method in frequency domain and artificial bee colony intelligent optimization algorithm, the extended finite element method avoids re-meshing in each iteration by introducing discontinuous displacement approximation and can reflect the number, location and size of defects by changing the level set function. In each iteration, the artificial bee colony intelligent optimization algorithm uses global and local searches. The probability of finding the optimal solution increases greatly and avoids local optimum. At the same time, by introducing topological variables, the number of flaws is incorporated into the inversion analysis process. The number of flaws can be intelligently inverted in the iteration process. Then, the inversion analysis model of multiple flaws (voids, cracks) in the structure is established. The analysis of several examples shows that the inversion analysis model can accurately detect the number, location and size of circular flaws, elliptical flaws, or crack in the structure. The result also shows the good robustness of the algorithm.