Page 89 - MSAM-3-4

Materials Science in Additive Manufacturing Super-resolution method for L-PBF

technique that builds solid parts by layer-by-layer of the melt pool can clearly reflect the current melting
accumulating metal materials. It can be used to produce state. Incorrect identification of the melt pool, such as
2
high-precision integrated built parts for industries such as interference from spatter and plume, will lead to error
automotive, aerospace, and medical. Ensuring consistency discrimination of the melt pool state, thereby affecting
3
in part quality is one of the research emphases in the L-PBF the subsequent feature extraction process. Collected
5
field. Many studies have attempted to ensure building melt pool images suffer from noise contamination, arc
accuracy through melt pool monitoring and process control. 4 light interference, edge blurring, and low resolution (LR).
The melt pool is a liquid region formed due to the Research has shown that interference from strong arc light
interaction of laser and powder, regarded as the fundamental can cause blurring at the edges of melt pool contour, and
unit constituting a part. The dynamic characteristics of the this effect can be reduced by installing appropriate filters
5
17,18
19
melt pool contain abundant information about the current and dimmers. Sampson et al. pointed out that process
melting state and process parameters, and it influences the parameters and exposure time would affect the quality
final quality of the part. An increasing array of melt pool of the melt pool image. To overcome this issue, they
6
monitoring techniques has been realized, including high- conducted trial-and-error experiments to determine the
speed camera imaging, photodiodes, infrared thermal optimal exposure time under different laser powers. Barua
20
imaging, and pyrometers. Photodiode measures the near- et al. pointed out that the radiation from the melt pool
4
infrared radiation intensity within a specific field of view itself interferes with external light sources, and the key
centered on the melt pool and it can achieve high temporal decision on which filter to use depends on the transmission
resolution monitoring of the dynamic characteristics of the efficiency of the filter for the required wavelength, which
21
melt pool. Infrared thermal imaging technology captures needs to be determined through experiments. Zhang et al.
7
the infrared radiation emitted by the melt pool and can noted that current real-time monitoring equipment faces
provide visual feedback on the states of the melt pool and challenges in balancing resolution and the field of view for
heat conduction. High-speed camera imaging is one of the monitoring the building process, which means enhancing
8
most commonly used methods for monitoring the L-PBF image quality would require additional investments in
process, providing a visual means to observe phenomena hardware. Furthermore, intense vibration caused by bright
such as melt pool, spatter, and plume. The melt pool spatter, plume, and shadow effects can affect the extraction
9
22
images can not only serve as an important feedback signal of melt pool features from the images. Projection errors
during the machining process but also as an important induced by optical setups will also result in unfavorable
23
means to study the dynamics of the molten pool and reveal outcomes. The above-mentioned problems will impact
the molten state and process. For example, by capturing the monitoring of the L-PBF process.
clear melt pool images, it is possible to intuitively observe In recent years, with the ongoing progress in ML, the
the vibration state of the melt pool when defects appear. transformation of LR images into high-resolution (HR)
This includes defects such as balling and discontinuities, images using ML is increasingly regarded as a feasible
10
pores, 11,12 and warpages caused by dynamic changes in the approach. Known as super-resolution (SR), it offers
13
melt pool. Thus, a series of L-PBF monitoring platforms a novel method for obtaining high-quality melt pool
based on high-speed camera imaging technology have images without requiring additional hardware expenses,
been established by many research teams. Research on achieving higher precision detection through software
14
melt pool state monitoring, prediction, and classification algorithms. The current SR methods can be categorized as
24
based on high-speed imaging technology has made interpolation-based, reconstruction-based, and learning-
25
26
significant progress. 15 based. Interpolation methods are the simplest and least
27
Current research primarily focuses on extracting computational techniques in SR. However, the effectiveness
features from melt pool images captured by high-speed of these methods is limited and can lead to image blurring
cameras, including highly popular machine learning and aliasing. Reconstruction-based methods treat SR as
(ML) methods. However, multiple factors can interfere the inverse process of image degradation. They involve
16
with melt pool monitoring, including the small melt considerations of both the frequency domain and spatial
pool volume, vulnerability to mirror-like reflections, the domain motion estimation of the image. These methods
presence of abundant metal powder spatter and plume are computationally complex and have a high requirement
in the vicinity, and variations in hardware conditions. for the distribution of sample data. Deep learning (DL) has
Nevertheless, research regarding the assurance and become the dominant force in the field of image processing,
evaluation of the melt pool image quality is limited. For significantly advancing the development of learning-based
additive manufacturing (AM), high-quality melt pool SR techniques. The SR Convolutional Neural Network
images are very important. The morphological changes (SRCNN) is the first to introduce DL into the field of

Volume 3 Issue 4 (2024) 2 doi: 10.36922/msam.5585