Page 9 - ESAM-1-1

Engineering Science in
Additive Manufacturing ML in MAM monitoring and control through images

A B

Figure 1. Schematic diagram of two types of metal additive manufacturing: (A) Laser power bed fusion and (B) direct energy deposition

before they arise. By combining image-based monitoring 2. Experiment methodology for image data
24
with ML-driven control, MAM processes can achieve acquisition
greater precision, reliability, and scalability, paving the way
for broader industrial adoption. In MAM, the effect of laser on metal materials is complicated.
Components produced through MAM experience a rapidly
The field has been significantly advanced by the evolving thermophysical metallurgical procedure that
previously reviewed publications on in situ monitoring and involves radiation, Marangoni convection, plasma eruption,
control in MAM. 7,25-31 A recurring theme in earlier review material melting and solidification, and heat conduction. This
papers, such as those by Herzog et al. and Grasso et al., dynamic process generates rapidly changing temperature
26
7
was concentrated on ML for fault identification or sensor fields, visible light, ultraviolet light, acoustic waves, and IR
signal processing. Besides, other review papers, such as radiation. Rich information about the final parts’ quality
Everton et al. and Chua et al., have not fully discussed condition is obtained through image-based monitoring,
33
32
real-time process control or the newest monitoring offering the potential for real-time identification and online
technology. Meanwhile, review articles such as Xia et al. optimization to enhance the quality of final components.
25
and Tang et al. only addressed specific MAM techniques Various sensing techniques, such as visible light cameras,
34
like WAAM and DED. Moreover, Wu et al. concluded a high-speed cameras, thermal imaging, photodiode
35
review on image-based monitoring in LPBF emphasized the sensing, acoustic emission sensing, and X-ray imaging, are
image collection process over ML applications. In contrast deployed to capture intricate physical signals during the
to prior studies, this review paper critically evaluates a AM process. These captured signals are then utilized to
range of in situ image monitoring methods, encompassing generate various types of images, as detailed in Figure 2.
optical, acoustic, and infrared (IR) thermography sensing. Specifically, the optical camera includes the industrial
Advanced ML techniques for anomaly detection are also camera, high-speed camera as well as the IR camera. The
introduced. Furthermore, this review paper explores sampled images from industrial cameras are shown in
cutting-edge ML applications in process control strategies Figures 2A and B. The images collected from the high-speed
based on image-based monitoring. By addressing these camera are displayed in Figure 2C and D. IR image provides
gaps, this work not only bridges the limitations of previous additional information about temperature, as illustrated
reviews but also establishes a strong foundation for in Figure 2E and F. In addition to optical camera, X-ray
potential future research in the field. imaging can be used to capture the internal structures of the
This review is organized as follows. A summary of parts, as shown in Figure 2G and H. Acoustic and spectral
the image types obtained in MAM and associated in situ imaging techniques are also applied for defect detection, as
monitoring technologies is provided in Section 2. Section depicted in Figures 2I and J. Subsequently, ML algorithms
3 offers a critical analysis of ML applications for handling are employed to process the converted images and extract
image-based data with an emphasis on ML-assisted deep features. This is followed by the establishment of
anomaly detection. Subsequently, Section 4 introduces mathematical models linking feature quantities to controlled
frameworks of ML-based closed-loop control systems as parameters. This section predominantly delves into the
well as their implementation. Section 5 further presents primary image-based data collection methods that support
existing challenges in recent studies and respective future in situ monitoring within the realm of MAM.
improvements, followed by Section 6 where the key In situ monitoring systems play a significant role
findings from the paper are concluded. in ensuring the part quality in MAM. Imaging-based

Volume 1 Issue 1 (2025) 3 doi: 10.36922/esam.8548