In-situ layer-wise certification for direct laser deposition processes based on thermal image series analysis
In-situ layer-wise certification for direct laser deposition processes based on thermal image series analysis
复制标题
DOI：
10.1016/j.jmapro.2021.12.041
复制
发表时间：
2022-03
影响因子：
6.2
通讯作者：
M. N. Esfahani;M. Bappy;L. Bian;Wenmeng Tian
中科院分区：
文献类型：
--
作者：
M. N. Esfahani;M. Bappy;L. Bian;Wenmeng Tian
在直接激光沉积（DLD）工艺中，工艺不确定性会导致最终产品出现缺陷，这会严重影响产品质量、机械性能和增材制造（AM）部件的可靠性。因此，质量控制和认证是至关重要的，在更广泛地采用DLD工艺。原位热历史包含过程质量和缺陷发生的关键信息。本文提出了一种新的分层异常检测方法，利用热图像序列分析在原位DLD过程认证。利用图像配准来表征逐层热历史中的动态，并且使用高斯过程（GP）模型来表征图像配准操作无法解释的变化分量。从配准建模和GP模型中提取多个新的逐层特征。薄壁试件和圆柱形试件的算例验证了该方法的有效性。当与基准方法比较时，所提出的方法对于薄壁试样显示出相当的结果，并且对于圆柱形试样，其显著优于基准方法。此外，所提出的方法的平均计算时间是显着短于平均逐层构建时间，使所提出的方法，以促进in-situatrial异常检测和过程控制。
In direct laser deposition (DLD) processes, process uncertainty leads to defects in the final product, which can significantly compromise product quality, mechanical properties, and reliability of the additively manufactured (AM) parts. Therefore, quality control and certification are of critical importance in the broader adoption of DLD processes.In-situthermal history contains critical information of process quality and defect occurrences. This paper proposes a new layer-wise anomaly detection method forin-situDLD process certification by leveraging thermal image series analysis. Image registration is leveraged to characterize the dynamics in the layer-wise thermal history, and Gaussian process (GP) models are used to characterize the variation component which is left unexplained by the image registration operation. Multiple new layer-wise features are extracted from the registration modeling and the GP models. Both a thin wall specimen and a cylindrical shaped specimen are used in the case study to demonstrate the effectiveness of the proposed method. When comparing with the benchmark method, the proposed method shows comparable results for the thin wall specimen, and it significantly outperforms the benchmark method for the cylindrical shaped specimen. In addition, the average computational time of the proposed method is significantly shorter than the average layer-wise build time, enabling the proposed method to facilitatein-situanomaly detection and process control.