![Proposed deep neural network architecture. For instance, Conv1 64-10∗\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$*$$\end{document}10 means the first convolutional layer with 64 filters of size 10×\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times $$\end{document}10. FC 512 denotes the fully-connected layer with 512 neurons](https://www.researchgate.net/publication/339483471/figure/fig7/AS:961715558699018@1606302265886/Proposed-deep-neural-network-architecture-For-instance-Conv1_Q320.jpg)
- A preview of this full-text is provided by Springer Nature.
- Learn more
Preview content only
Content available from Journal of Intelligent Manufacturing
This content is subject to copyright. Terms and conditions apply.
Journal of Intelligent Manufacturing (2020) 31:2003–2017
https://doi.org/10.1007/s10845-020-01549-2
Quality analysis in metal additive manufacturing with deep learning
Xiang Li1,2 ·Xiaodong Jia1·Qibo Yang1·Jay Lee1
Received: 24 October 2019 / Accepted: 14 February 2020 / Published online: 25 February 2020
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract
As a promising modern technology, additive manufacturing (AM) has been receiving increasing research and industrial
attention in the recent years. With its rapid development, the importance of quality monitoring in AM process has been
recognized, which significantly affects the property of the manufactured parts. Since the conventional hand-crafted features
for quality identification are generally costly, time-consuming and sensitive to noises, the intelligent data-driven automatic
process monitoring methods are becoming more and more popular at present. This paper proposes a deep learning-based
quality identification method for metal AM process. To alleviate the requirement for large amounts of high-quality labeled
training data by most existing data-driven methods, an identification consistency-based approach is proposed to better explore
the semi-supervised training data. The proposed method is able to achieve promising performance using limited supervised
samples with low quality, such as noisy and blurred images. Experiments on a real-world metal AM dataset are implemented
to validate the effectiveness of the proposed method, which offers a promising tool for real industrial applications.
Keywords Additive manufacturing ·Process monitoring ·Quality identification ·Deep learning ·Low-quality data
Introduction
In the recent years, additive manufacturing (AM) techniques
have been emerging as one of the most promising manufac-
turing technologies in a wide variety of application scenarios
(Kwon et al. 2018; Tapia and Elwany 2014;Huetal.2019;
Zhao and Guo 2019; Gonzalez-Val et al. 2019), such as auto-
motive, aerospace, robotics, electronics etc. AM which is also
popularly known as 3D printing, free-form fabrication and
rapid prototyping, has been receiving increasing academic
and industrial attention, since it is capable of manufacturing
highly sophisticated and fully functional 3D objects that are
difficult to be accomplished by conventional manufacturing
approaches (Everton et al. 2016). This novel technology has
the potential to largely reduce the existing constraints on the
geometrical design, accelerate the production time and min-
imize the manufacturing cost.
BXiang Li
xiangli@mail.neu.edu.cn
1Department of Mechanical and Materials Engineering,
University of Cincinnati, Cincinnati, OH 45221, USA
2Key Laboratory of Vibration and Control of Aero-Propulsion
System Ministry of Education, Northeastern University,
Shenyang 110819, China
Currently, different AM methods have been developed,
including material extrusion, sheet lamination, directed
energy deposition, material jetting, powder bed fusion etc.
Among them, the powder bed fusion (PBF) approach has
been one of the most popular methods in metal part addi-
tive manufacturing (Chua et al. 2017; Zhang et al. 2018).
Generally, metallic powder is spread over the previous lay-
ers during the AM process, and a laser is usually used as
the energy source to achieve the metallic bonding between
layers. In this study, the PBF process is investigated.
Despite the promising development of AM methods, the
main existing quality-related problems lie in the limited sta-
bility and repeatability of the mechanical properties of the
manufactured parts in production (Shevchik et al. 2019).
Generally, the part quality is directly influenced by the mate-
rial microstructure in the AM process, and the defects such
as porosity, residual stress and cracks are caused by a num-
ber of factors, including laser scanning speed, laser power,
hatch distance etc. Timely and accurate quality assessment
in AM process is of great importance in improving manu-
facturing quality and reducing industrial cost (Kwon et al.
2018). However, it is challenging for real implementation of
the assessment due to the high complexity of the involved
physics covering a wide range of practical factors.
123
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
