Page 8 - ESAM-1-1

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

and then producing parts layer by layer. Since its inception applications. MAM-created parts frequently have
7
in the 1980s, AM has evolved from a rapid prototyping process-induced flaws such as porosity, cracks, and
tool to a full-fledged manufacturing technology, deformation caused by residual stresses. These flaws lead
8,9
revolutionizing industries such as aerospace, automotive, to unpredictable discrepancies in physical characteristics
healthcare, and consumer goods. The ability to produce both within and between parts, compromising the
1
complex geometries with minimal material waste and quality, consistency, and reliability of the components.
the potential for mass customization have made AM a To overcome these obstacles and guarantee MAM
cornerstone of modern manufacturing. Metal additive products’ dependability, in situ process monitoring has
2
manufacturing (MAM) stands out as a groundbreaking become indispensable. 10,11 Monitoring key parameters
technology within the realm of AM, facilitating the such as temperature, melt pool dynamics, and material
creation of intricate geometries with metal materials. deposition rates enables the timely detection of anomalies
MAM operates by depositing metal material (powder or and deviations from desired outcomes, aiding in quality
wire) layer by layer with controlled high-energy sources, control, fault detection, and process optimization. In
selectively fusing them to craft complex 3D objects. This recent years, image-based monitoring techniques have
3
process primarily encompasses two main techniques gained popularity for their non-intrusive nature and
based on energy sources and material forms, such as laser ability to provide visual insights into the MAM process.
powder bed fusion (LPBF) and direct energy deposition These techniques offer detailed information on layer
(DED), as shown in Figure 1A and B, respectively. LPBF, formation, defects, and overall part quality. 12-14 The
4,5
also known as selective laser melting (SLM), produces efficiency and effectiveness of MAM procedures could be
highly accurate and intricate components by selectively greatly increased by combining machine learning (ML)
melting metallic powders and fusing particles in layers algorithms with image-based monitoring to facilitate
with a high-energy laser. On the other hand, DED is automated defect detection, optimize process parameters,
the process of directly depositing molten material onto and enhance overall production quality. For instance, ML
a substrate, usually with the help of an electron or laser models such as convolutional neural networks (CNNs)
beam. DED is particularly suited for repairing or adding and deep belief networks (DBNs) have been employed to
material to existing components for large-scale structures. analyze image data for real-time defect detection, including
Notably, DED primarily encompasses wire arc additive porosity, cracks, and uneven layer deposition. 6,15-18 By
manufacturing (WAAM) and powder-based DED, while leveraging these ML approaches, researchers can achieve
commonly used LPBF includes electron beam melting precise monitoring of the MAM process, enabling early
(EBM) and SLM. A brief overview of MAM procedures identification of defects and ensuring consistent part
and the corresponding construction features is given in quality. 19-21
Table 1. 6
In addition to monitoring, the utilization of ML
While MAM offers tremendous potential, several approaches to manage and improve MAM operations
obstacles hinder its widespread adoption in industrial is growing. The dynamic nature of MAM, characterized
by complex interactions between process parameters,
Table 1. Comparison of technical concepts and features of necessitates advanced control strategies to ensure
common MAM processes consistent product quality. ML models, such as recurrent
neural network and reinforcement learning (RL), have
SLM DED EBM WAAM been utilized to optimize process parameters, including
Energy source Laser (single) Laser Electron Electric arc laser power, scan speed, and material feed rate, based on
(Multi) beam real-time image data. 22,23 These models enable adaptive
Focal spot 30–200 μm 0.5–3 mm 200–500 μm 1–3 mm control by predicting optimal parameter adjustments to
diameter mitigate defects and improve part quality. For example,
Output power 50–1000 W >1000 W >2 kV >1000 W ML algorithms can analyze image data to detect deviations
Material form Powder Powder Powder Wire in melt pool behavior and automatically adjust process
Working Inert gas Inert gas Vacuum Inert gas parameters to maintain optimal conditions. By lowering
environment (10 Pa) the possibility of errors and increasing overall production
-4
Build accuracy ±0.1 mm ±0.5 mm ±0.4 mm ±2 mm efficiency, this closed-loop control method improves the
Post-processing Almost no Less Almost no Much MAM process’s stability and dependability. Furthermore,
Abbreviations: DED: Direct energy deposition; EBM: Electron beam the integration of ML with physics-based models allows for
melting; MAM: Metal additive manufacturing; SLM: Selective laser more accurate predictions of process outcomes, enabling
melting; WAAM: Wire arc additive manufacturing. proactive control measures to address potential issues

Volume 1 Issue 1 (2025) 2 doi: 10.36922/esam.8548