Research on Model Reconstruction Methods for Partial Point Cloud Information Loss

Objective With the increasing demand for improved assembly performance, conducting assembly performance analysis based on measured data has become increasingly important. At the present stage, assembly performance analysis requires model refactoring to predict assembly errors. The refactoring of a model of point cloud inverse reconstruction can reflect real physical information, which solves the deviation between simulation and actual results. However, large-sized and structurally complex parts face challenges such as blocks, which result in incomplete point cloud information. Therefore, the overall surface point cloud of the part cannot be fully obtained, preventing the refactored model from accurately representing real physical information. Therefore, research on model reconstruction methods without local point clouds is essential. This study proposes a Laplace mesh transformation method to deform the theoretical triangular mesh based on the scanned area. The corresponding region of the theoretical model is replaced with the scanned point cloud to obtain a complete point cloud of the part.Methods The theoretical model and the scanned area point cloud were used as data sources. The triangular mesh formed by the scanned region point cloud and the surface of the theoretical model constituted a two-dimensional manifold reference domain, ensuring that any point selected from the reference domain had a corresponding mapping in the triangular mesh. Then, the Laplacian mesh transformation was applied to achieve mesh mapping. This process utilized the Laplacian operator in the vertex differential equation of the discrete mesh model to express local features regarding the differential properties of the discrete scalar field on the surface. The reconstructed differential properties preserved the original meaning of the scalar fields, ensuring that the transformation process maintained the local feature details of the initial surface to the maximum extent. In other words, the mapped points of the triangular mesh on the reference domain could be projected onto the two-dimensional manifold triangular mesh without changing the mesh topology. Finally, the scanned point cloud carried on region partition, and the Interactive Closest Point (ICP) algorithm was utilized to match the scanned area with the theoretical model. The theoretical model, driven by the matching regions, was adjusted to obtain the surface points in the absence region of the point clouds of actual parts. These points replaced the corresponding region points, resulting in the parts’ complete surface point cloud. The primary difference between the proposed method and traditional local point cloud reconstruction approaches was that the point cloud fitting surface was no longer used for substitution. Instead, the original model was driven to adapt. Although the local model was efficiently reconstructed, the absence regions of the point cloud were ensured to transform in alignment with the actual model, improving the reconstructed model’s fidelity.Results and Discussions This study employed the outer cylinder of the landing gear as an experimental object to verify the proposed model reconstruction method. A comparative analysis between the traditional method and the proposed approach revealed that direct substitution in conventional methods resulted in misalignment at the junction of the scanned area and the absent point cloud region, leading to inconsistencies with the actual part. In contrast, the model reconstructed using the proposed method exhibited a smooth transition area, ensuring surface continuity. The deviation cloud map comparing the theoretical model with the complete point cloud indicated a maximum deviation of 0.802 mm, representing the largest error in 3D-printed part manufacturing in the middle of the outer cylinder. The model reconstructed using the traditional localized point cloud data approach exhibited a maximum deviation of 0.798 mm, whereas the model reconstructed using the proposed method had a maximum deviation of 0.719 mm, primarily in the middle region of the cylinder segment. The improvement in accuracy was not substantial due to the limited intermediate information in the scanned area. Additional control experiments were conducted by reducing the absent point cloud region. In these experiments, the maximum deviation for the model reconstructed using the traditional method remained at 0.798 mm, while the proposed method yielded a maximum deviation of 0.436 mm. The deviation in the outer cylinder was significantly reduced to 0.236 mm due to increased scanned data in the middle region of the outer cylinder, whereas the traditional method failed to reduce the deviation. When the point cloud coverage reached about 90%, another control experiment was conducted. The maximum deviation for the model reconstructed using the traditional method was 0.717 mm, while the proposed method reduced the maximum deviation to 0.145 mm. These findings indicated that increased scanned area decreases the reconstructed model’s overall deviation, validating the proposed approach’s rationality.Conclusions Compared to traditional direct replacement methods, the results showed that the proposed method effectively addresses surface misalignment issues and better preserves the reconstructed model’s surface continuity. In addition, the overall accuracy of the reconstructed model significantly improves as point cloud coverage increases. When the point cloud coverage reached about 90%, the maximum deviation of the reconstructed model was reduced to 0.145 mm, confirming the feasibility of the proposed method.