Dynamic analysis of complex and fine structure often involves external load with different time scales. Meanwhile, a local refined mesh is needed at the analysis zone. The classical explicit-implicit mixed method was used to deal with such problems with the interpolation process of boundary data which was likely to cause computational instability. At the same time, a single time step would cause waste of computing resources. An improved explicit-implicit mixed asynchronous time step computing method was proposed with multiple boundary mesh. Node segmentation was used to divide the finite element model into explicit and implicit partitions. A compatible format predictive correction Newmark integration scheme was adopted in different partitions. A larger time step was used for implicit partition and a smaller time step for explicit region. The prediction waveform for explicit partition can be transmitted completely in the boundary mesh without truncation. Sparse storage row compression CRS format was used in the implicit partition and node data can be obtained by sparse direct solver. The node acceleration data for explicit and implicit partition were solved sequentially and efficiently. Numerical examples showed that the accuracy of the proposed method was higher than that of the traditional explicitimplicit mixed method and the computation time for dynamics analysis was further reduced.