On-machine, optical imaging of the layer is bound to play an important role in the development of future in-process monitoring solutions for fused filament fabrication. In addition to detecting layer internal defects, layer imaging offers the possibility to retrieve the outer contours of a layer and thus three-dimensionally reconstruct the side walls of a part by vertically stacking contours belonging to sequential layers. The current open challenges for layer imaging relate to the compromise between imaging resolution and extension of covered area, and to the complexity of image interpretation, exacerbated by optically difficult materials, complex filament patterns and non-uniform environmental conditions. In this work we propose an original optical imaging solution for layer contour identification and stacking. Our solution overcomes the current challenges related to range/resolution and image interpretation by making use of digital twins, i.e. geometric simulations of the layer, based on the current manufacturing process parameters. Thanks to the use of digital twins, our vision system is able to: use high-resolution imaging only where it is needed, by following the predicted positions of the layer contours while acquiring a sequence of images; algorithmically interpret each image by concentrating only on interesting regions as suggested by the simulated prediction, thus reducing the risk for false positives; and finally monitor contour quality by performing a more accurate comparison between the measurement and a reference contour obtained by a simulation which replicates the actual fingerprint of the process, as opposed to simply relying on the edge obtained by slicing the CAD model.

The in-process monitoring capability of the computer vision system is demonstrated through the fabrication of acceptable and defected instances of a test geometry. Fabrication issues are effectively and efficiently detected by the system and alarms are triggered that can be used to command early termination of the build process or to initiate corrective actions.

在机器上，该层的光学成像必将在熔丝制造的未来过程监控解决方案的开发中发挥重要作用。除了检测层的内部缺陷外，层成像还可以检索层的外部轮廓，从而通过垂直堆叠属于连续层的轮廓在三维上重建零件的侧壁。层成像的当前开放挑战涉及成像分辨率和覆盖区域的扩展之间的折衷，以及由于光学困难的材料，复杂的灯丝图案和不均匀的环境条件而加剧的图像解释的复杂性。在这项工作中，我们提出了一种用于层轮廓识别和堆叠的原始光学成像解决方案。我们的解决方案通过基于当前制造工艺参数的数字孪生模型，即层的几何模拟，克服了与范围/分辨率和图像解释相关的当前挑战。由于使用了数字双胞胎，我们的视觉系统能够：在获取一系列图像的同时跟踪图层轮廓的预测位置，从而仅在需要的地方使用高分辨率成像；从算法上讲，通过仅将注意力集中在模拟预测所建议的感兴趣区域上来解释每幅图像，从而降低了误报的风险；最终通过在测量值和通过模拟获得的参考轮廓之间进行更精确的比较来监视轮廓质量，该仿真可以复制过程的实际指纹，

通过制造可接受的和有缺陷的测试几何体实例，证明了计算机视觉系统的进程内监视功能。系统可以有效地检测到制造问题，并触发警报，这些警报可以用于命令过早终止构建过程或启动纠正措施。