ORIGINAL ARTICLE
Figure from article: GPU-Accelerated Mesh-Based...
 
KEYWORDS
TOPICS
ABSTRACT
In-time quality control remains a major challenge in additive manufacturing (AM) owing to the high computational cost of deformation reconstruction and difficulty in handling real-image observations during printing. Therefore, this paper presents a deformation-reconstruction framework for image-based quality control in AM, which integrates free-form deformation, differentiable rendering, and an adaptive loss formulation to enable efficient and robust mesh-to-image alignment. This framework supports both synthetic multi-view inputs and real-image observations through a unified workflow. In the synthetic test case, multi-view target images were used to evaluate the reconstruction accuracy and computational performance of the proposed framework. In the real-image test case, a segmented top-view photograph was used as the target observation, and the deformation was guided by a combined loss formulation to improve robustness against real-image imperfections. The proposed framework achieved a mean runtime of 3.81 s, whereas the Central Processing Unit (CPU)-based baseline model required 692.65 s. The former method exhibited a speedup of 181.80 times, while maintaining high reconstruction accuracy. In the synthetic test case, the framework achieved Chamfer and Hausdorff distances of 0.0111 and 0.0207, respectively. In the real-image test case, the framework achieved a normalized cross-correlation value of 0.9729, together with a Chamfer distance of 0.0013 and Hausdorff distance of 0.0035, demonstrating the successful deformation of the nominal mesh toward the observed target geometry. This study provides an effective approach for deformation reconstruction under realistic conditions and demonstrates the feasibility of periodic near-real-time quality control in AM
Journals System - logo
Scroll to top