3D point cloud damage detection for automating metal repair processes through additive manufacturing

Damage detection in point clouds is a critical step for automating systems of metal component repair through additive manufacturing. However, current approaches face challenges due to the variability in part geometries and the diversity of damage types. This paper presents a novel damage detection algorithm that adapts to variations in damage location, geometry, and volume through a parameter-driven approach based on plane segmentation and clustering tasks. A synthesized dataset was generated for the validation, with ground truth established using efficient Boolean comparisons combining voxelization with spatial data structures. To optimize the performance of the algorithm, a Design of Experiments was conducted, analyzing parameter influences and guiding the development of optimal detection models. This approach demonstrated a remarkable performance across a wide range of complex damage scenarios, achieving higher precision, recall, and F1 scores than existing methods. The optimized version further enhances recall with minimal impact on precision, ensuring robust and balanced detection capabilities. The approach was also validated in a case study, demonstrating its operational reliability under realistic conditions. This work offers a generalizable solution for effective damage detection, setting a strong foundation for future advancements in automated repair systems.