Automated Segmentation via Quantile-Based Thresholding and Shape Modeling of Geometric Features in Additive Manufactured Parts

Abstract

Compared to subtractive manufacturing, Additive manufacturing (AM) technologies are often utilized for better resource efficiency and faster fabrication of structures with complex geometries. AM can be utilized to fabricate parts with complex freeform designs. However, the part geometry might also contain irregularities due to energy fluctuations during deposition and severe thermal gradients. Typically, the first step for geometric evaluation of fabricated parts is the effective segmentation of features of interest from geometric measurements (i.e., point clouds) of as-deposited parts. These features can range from local distortions caused by fluctuations in process parameters to a decomposition of the overall geometry into multiple freeform shape components for detailed characterization and analysis. However, manual segmentation of large point cloud data is time-consuming and often inaccurate. This thesis proposes an unsupervised quantile-based framework for automatic segmentation of geometric features caused by freeform designs or excessive material accumulation. The implementation of the segmentation process is shown for four Wire Arc Additive Manufacturing (WAAM) printed cylindrical components and one Stereolithography (SLA) printed dental part. Beyond segmentation, a parametric shape model is introduced to model each segmented feature with an asymmetric, quadrant-aware ellipse. Additionally, the deviation at each point for any particular segment is captured by a power model. Finally, principal components analysis (PCA) of all the fitted parameters (ellipse scales and power model parameters) is conducted for dimensionality reduction. Case studies involving the WAAM printed cylindrical components and complex SLA-printed teeth structures demonstrate the robustness and interpretability of both the segmentation results and the shape modeling approach.