The increasing interest towards additive manufacturing (AM) is pushing the industry to provide new solutions to improve process stability. Monitoring is a key tool for this purpose but the typical AM fast process dynamics and the high data flow required to accurately describe the process are pushing the limits of standard statistical process monitoring (SPM) techniques. The adoption of novel smart data extraction and analysis methods are fundamental to monitor the process with the required accuracy while keeping the computational effort to a reasonable level for real-time application. In this work, a new framework for the detection of defects in metal additive manufacturing processes via in-situ high-speed cameras is presented: a new data extraction method is developed to efficiently extract only the relevant information from the regions of interest identified in the high-speed imaging data stream and to reduce the dimensionality of the anomaly detection task performed by three competitor machine learning classification methods. The defect detection performance and computational speed of this approach is carefully evaluated through computer simulations and experimental studies, and directly compared with the performance and computational speed of other existing methods applied on the same reference dataset. The results show that the proposed method is capable of quickly detecting the occurrence of defects while keeping the high computational speed that would be required to implement this new process monitoring approach for real-time defect detection.

对增材制造 (AM) 日益增长的兴趣正在推动该行业提供新的解决方案以提高工艺稳定性。监控是实现此目的的关键工具，但典型的 AM 快速过程动态和准确描述过程所需的大量数据流正在推动标准统计过程监控 (SPM) 技术的极限。采用新颖的智能数据提取和分析方法对于以所需的精度监控过程，同时将计算工作量保持在实时应用的合理水平至关重要。在这项工作中，提出了一种通过原位高速相机检测金属增材制造过程中缺陷的新框架：开发了一种新的数据提取方法，以有效地仅从高速成像数据流中识别的感兴趣区域中提取相关信息，并降低三种竞争对手机器学习分类方法执行的异常检测任务的维数。通过计算机模拟和实验研究仔细评估了这种方法的缺陷检测性能和计算速度，并直接与应用于同一参考数据集的其他现有方法的性能和计算速度进行了比较。结果表明，所提出的方法能够快速检测缺陷的发生，同时保持实现这种新的实时缺陷检测过程监控方法所需的高计算速度。