跳到正文
arXiv · 机器人与计算机视觉· Stefano Raimondo (Department of Mechanical Engineering, Politecnico di Milano, Milan, Italy), Matteo Bugatti (Department of Mechanical Engineering, Politecnico di Milano, Milan, Italy), Marco Grasso (Department of Mechanical Engineering, Politecnico di Milano, Milan, Italy)·· 14 小时前AI 评分45

Graph-augmented U-Net 提升激光粉末床熔融原位几何重建鲁棒性

Graph-augmented Segmentation for Enhanced In-Situ Geometry Reconstruction of Complex Shapes in Laser Powder Bed Fusion

AI 导读

Graph-augmented U-Net 架构首次探索将图神经网络集成至瓶颈层,以解决金属激光粉末床熔融中因图像外观差异导致的分布外泛化难题。该模型在涉及不同晶格、材料及制造设置的两个案例研究中,相比传统 U-Net 和主动轮廓分割方法实现了更高的重建精度与一致性。初步异常检测实验表明,这种鲁棒性提升并未牺牲对几何偏差的敏感度,为异构制造条件下的可靠原位检测提供了新策略。

来源:arXiv · 机器人与计算机视觉 · arxiv.org