ABSTRACT

Surface monitoring is an essential part of quality assurance for additive manufacturing (AM). Surface defects need to be identified early in the AM process to avoid further deterioration of the part quality. In this paper, a rapid surface defect identification method for directed energy deposition (DED) is proposed. The main contribution of this work is the development of an in-situ point cloud processing with machine learning methods that enable automatic surface monitoring without sensor intermittence. An in-house software platform with a multi-nodal architecture is developed. In-situ point cloud processing steps, including filtering, segmentation, surface-to-point distance calculation, point clustering, and machine learning feature extraction, are performed by multiple subprocesses running simultaneously. The combined unsupervised and supervised machine learning techniques are applied to detect and classify surface defects. The proposed method is experimentally validated, and a surface defect identification accuracy of 93.15% is achieved.

摘要

表面监测是增材制造（AM）质量保证的重要组成部分。需要在增材制造过程中及早发现表面缺陷，以免零件质量进一步下降。本文提出了一种用于定向能量沉积（DED）的快速表面缺陷识别方法。这项工作的主要贡献是利用机器学习方法开发了一种现场点云处理技术，该方法可实现自动表面监测而无需传感器间断。开发了具有多节点架构的内部软件平台。原位点云处理步骤（包括过滤，分割，面到点距离计算，点聚类和机器学习特征提取）由同时运行的多个子过程执行。结合了无监督和监督机器学习技术来检测和分类表面缺陷。实验验证了该方法的有效性，表面缺陷识别精度达到93.15％。