Computed tomography is a crucial non-destructive testing technique for acquiring point cloud data and reverse reconstructing workpieces with complex internal structures. Point cloud segmentation is vital in reverse engineering， as its accuracy affects the quality of the reconstructed model. To address reverse reconstruction needs， this paper introduces a slice segmentation method based on the principal axis for part segmentation， involving two stages： coarse and fine segmentation. Initially， parallel equidistant point cloud slices are generated along the principal axis， classified by calculating slice features. Similar slices are then merged and clustered for coarse segmentation results. These results serve as input for the fine segmentation stage， where sub point cloud features are calculated and classified. The sub point clouds are verified， merged， and under-segmented to achieve the final segmentation result. Subsequently， a geometry-topology tree structure is proposed for fast， accurate model reconstruction， representing the model unambiguously. The experimental results demonstrate that the proposed method achieves effective segmentation and high applicability， with an average IoU of 0.879 7. Additionally， the model reconstructed using a geometric-topological tree fully retains the object's main structural information， which is highly valuable for practical engineering applications.