Unraveling pore evolution in post-processing of binder jetting materials: X-ray computed tomography, computer vision, and machine learning
Unraveling pore evolution in post-processing of binder jetting materials: X-ray computed tomography, computer vision, and machine learning
复制标题
DOI：
10.1016/j.addma.2020.101183
复制
发表时间：
2020-04
影响因子：
11
通讯作者：
Yunhui Zhu;Ziling Wu;W. Douglas Hartley;J. Sietins;Christopher B. Williams;Hang Z. Yu
中科院分区：
文献类型：
--
作者：
Yunhui Zhu;Ziling Wu;W. Douglas Hartley;J. Sietins;Christopher B. Williams;Hang Z. Yu
金属增材制造中的质量控制优先考虑先进检测方案的开发，以表征加工和后加工过程中的缺陷演变。这涉及到检测内部缺陷和分析宏观样品中大型复杂缺陷数据集的巨大挑战。在这里，我们提出了一个检测管道，它集成了（i）快速的微型X射线计算机断层扫描重建，（ii）自动化3D形态分析，以及（iii）基于机器学习的大数据分析。X射线计算机断层扫描和自动化计算机视觉导致被检查的宏观体积的整体缺陷形态数据库，基于该数据库，采用机器学习分析来揭示对缺陷特征的全球演变的定量见解，而不是定性的人类观察。我们通过检查粘合剂喷射增材制造的后处理中的全球规模的孔隙演变来展示该管线，从铜的绿色状态到烧结状态，以及到热等静压状态。该管道被证明是有效的检测和处理与大量（105）的孔隙在宏观体积的信息。随后的主成分分析和聚类分析提取的关键形态描述符和分类检测到的孔隙分为四个形态组。通过量化的演变（i）的重量的孔形态参数和（ii）的孔隙数量和体积分数的每一个分类组，新的认识发展的影响，烧结和热等静压孔分解，收缩，并在后处理过程中的粘合剂喷射平滑。
Quality control in metal additive manufacturing prioritizes the development of advanced inspection schemes to characterize the defect evolution during processing and post-processing. This involves grand challenges in detecting internal defects and analyzing large and complex defect datasets in macroscopic samples. Here, we present an inspection pipeline that integrates (i) fast, micro X-ray computed tomography reconstruction, (ii) automated 3D morphology analysis, and (iii) machine learning-based big data analysis. X-ray computed tomography and automated computer vision result in a holistic defect morphology database for the inspected macroscopic volume, based on which machine learning analysis is employed to reveal quantitative insights into the global evolution of defect characteristics beyond qualitative human observations. We demonstrate this pipeline by examining the global-scale pore evolution in post-processing of binder jetting additive manufacturing, from the green state, to the sintered state, and to the hot isostatic pressed state of copper. The pipeline is shown to be effective at detecting and processing the information associated with a large number (∼105) of pores in macroscopic volumes. The subsequent principal component analysis and clustering analysis extract the key morphological descriptors and categorize the detected pores into four morphological groups. By quantifying the evolution of (i) the weight of pore morphology parameters and (ii) the pore number and volume fraction of each categorized group, new understandings are developed regarding the effects of sintering and hot isostatic pressing on pore decomposition, shrinkage, and smoothing during post-processing of binder jetting.