Additive manufacturing source identification from internal void structures using X-ray computed tomography and point cloud deep learning

Additive manufacturing (AM) enables flexible, distributed production in modern supply chains; however, key challenges remain in ensuring part quality, traceability, and authenticity. This work introduces a novel deep machine learning framework to identify the printer of origin and other process settings from three-dimensional (3D) point cloud representations of internal void structures. To develop and demonstrate this fingerprinting method, we designed, fabricated, and measured 270 parts made with fused deposition modeling (FDM) on five identical printers. The parts were made from either polylactic acid (PLA) or polycarbonate (PC), four part designs that include simple cubes and industrial style parts, and six infill settings, yielding void fractions of 0.7% to 7.6%. The parts were measured using X-Ray computed tomography (CT) with voxel size of 21 μm, resulting in three-dimensional (3D) point cloud representations of internal voids. The deep learning model based on a fine-tuned PointNet++ architecture learned the machine- and process-specific patterns of these internal void structures. The model predicts the printer of origin for a given part with 100% accuracy. The model also predicts material (100% accuracy) and infill setting (95.8% accuracy). The model generalizes to predict the printer of origin for part designs and infill settings that do not appear in the training dataset. This work establishes AM source identification based on information from inside the part, with potential relevance to part traceability, authentication, and quality assurance.