Real-Time Crankshaft 3D Metrology via Adaptive Boundary Extraction and Registration-Constrained Cylinder Fitting

The crankshaft metrology process for pipelinecompressor workpieces, as a critical link in new-energy vehicle manufacturing, must simultaneously satisfy high accuracy, full-surface coverage, and takt-time efficiency. At present, the industry still relies predominantly on traditional contact CMM workflows, which are inefficient and provide incomplete coverage on complex free-form surfaces. Existing non-contact solutions, however, have not been widely adopted because nonuniform sampling and missing end faces lead to instability and convergence failure. To this end, we propose a real-time 3D pointcloud metrology scheme that dynamically models the crankshaft's spatial features. For pin holes, by combining normal-consistency estimation with curvature-compensated, density-adaptive boundary cues, the circle center and radius are robustly recovered under end-face incompleteness; for pins, a registration-aware cylinder fitting guided by FPFH and weighted-ICP registration is performed, and design-prior constraints are incorporated to obtain an accurate perpendicularity reference model. In addition, dynamic-radius denoising and voxelization with a minimum-point constraint are introduced in preprocessing to preserve weak features. Benefiting from elastic feature capture, the method exhibits excellent performance in benchmark tests across multiple part models and achieves substantial efficiency gains relative to CMM.