The diversity of materials and complex geometric shapes of the intelligent manufactured components lead to sparse 3D point clouds at edge regions, resulting in non-uniform density distribution across complex mechanical parts. This study proposes a hybrid filtering method integrating density-adaptive SOR and ROR to remove multiple noise types of complex component point clouds. The method firstly merges voxels of the segmented point clouds based on density similarity, then establishes minimum neighborhood points for ROR using merged voxel size exponents, afterward determines search radius through neighborhood average distances, and calculates SOR standard deviations using scaling coefficients. The proposed method was tested using classical 3D point cloud models. Experiment results demonstrates that the edge retention of post-processed point clouds significantly exceed those achieved by fixed-parameter filtering methods, while effectively preserving detail information in sparse regions, and the noise removal rate is also improved. Robustness tests conducted under varying noise levels confirm the consistent performance across different noise intensities. This method establishes a technical foundation for online inspection in intelligent manufacturing systems requiring high-fidelity geometric reconstruction.