Additive manufacturing has been increasingly applied. As one of the most commonly used technologies, fused deposition modeling (FDM) still faces the challenge of instable performance. The appearance of the printed part is an important feature to assess its quality. As FDM processes usually take a long time, it is very important to timely identify the defects to avoid unnecessary waste of time and cost. At current stage, this identification work is usually done by the operators. However, it is difficult to realize continuous monitoring for multiple printers and identify surface defects shortly. With the advanced artificial intelligence techniques, a vision-based adaptive monitoring system is proposed in this article to achieve online monitoring with high efficiency and accuracy. The system design is introduced for common FDM printers that allows one camera to move to different angles and capture the images of the printing part. A heuristic algorithm is then proposed to achieve adaptive shooting position planning according to the part geometries. Furthermore, a convolutional neural network-based model is designed to achieve efficient defect classification with high accuracy. A series of experiments have been conducted to illustrate the effectiveness of the proposed system.

中文翻译：

---

基于 CNN 的熔融沉积建模自适应表面监测系统


增材制造的应用日益广泛。作为最常用的技术之一，熔融沉积建模（FDM）仍然面临着性能不稳定的挑战。打印部件的外观是评估其质量的重要特征。由于FDM工艺通常需要很长时间，因此及时识别缺陷非常重要，以避免不必要的时间和成本浪费。目前阶段，这种识别工作通常由运营商来完成。然而，实现多台打印机的连续监控并快速识别表面缺陷是很困难的。本文利用先进的人工智能技术，提出了一种基于视觉的自适应监测系统，以实现高效、准确的在线监测。该系统设计针对常见的FDM打印机，允许一台相机移动到不同的角度并捕获打印部分的图像。然后提出一种启发式算法，根据零件几何形状实现自适应拍摄位置规划。此外，设计了基于卷积神经网络的模型，以实现高精度的高效缺陷分类。进行了一系列实验来说明所提出系统的有效性。