Page 59 - MSAM-4-3

Materials Science in Additive Manufacturing Bead geometry prediction in laser-arc AM

the layer-by-layer deposition and solidification of metal Table 1. Overview of research on data modeling between
feedstock, MAM creates new possibilities for designing process parameters and layer geometries in metal additive
and producing intricate metallic parts, significantly manufacturing
2-4
enhances manufacturing productivity, propels innovation Process Models Inputs Outputs Ref.
in product design and processing routes, and expedites the LPBF RF and Laser power and Layer height, 27
overall evolution of industrial manufacturing. Among ANN scanning velocity width, and
5,6
MAM technologies, powder bed fusion (PBF) and directed penetration
energy deposition (DED) are the most widespread; wire- depth
7
feed DED, in particular, garners interest for its superior Powder-laser SVR Laser power, travel Layer height and 28
deposition rates and material efficiency. Wire-feed DED speed, and powder width
8,9
additive manufacturing utilizes lasers, electron beams, or feed rate
welding arcs as energy inputs. In comparison with laser-or Wire-laser DED NB Laser power, travel Layer height and 29
electron-beam-based AM, wire arc additive manufacturing speed, and wire width
feed rate
(WAAM) offers significant benefits in capital expenditure
and build rate. The typical deposition rate for laser or WAAM ANFIS Wire feed rate and Surface 30
10
travel speed
roughness
electron beam AM lies between 2 and 10 g/min, whereas
WAAM attains 50 – 130 g/min; 11,12 its energy efficiency WAAM SVR Wire feed rate and Layer height and 31
travel speed
width
of roughly 90% greatly surpasses the 2 – 5% achieved by WAAM XGBoost Current and travel Layer height, 32
laser-based AM. 13,14 Thanks to their favorable welding speed width, and area
characteristics, aluminum–copper (Al–Cu) alloys serve Abbreviations: ANFIS: Adaptive neuro-fuzzy inference system;
as ideal materials for WAAM fabrication. 15-17 The bead ANN: Artificial neural network; DED: Directed energy deposition;
geometry, which represents the basic fabrication unit LPBF: Laser powder bed fusion; NB: Naïve Bayes; RF: Random
in WAAM, exerts a direct influence on the dimensional forest; SVR: Support vector regression; WAAM: Wire arc additive
fidelity of the manufactured component. Inadequate bead manufacturing; XGBoost: Extreme gradient boosting.
quality requires additional finishing operations, thereby
27
increasing labor expenditure and leading to material laser power and scan speed as predictors, Le-Hong et al.
wastage. developed random-forest and artificial-neural-network
models to estimate layer geometry in laser PBF, reporting
Recently, the laser-arc hybrid approach to additive validation R values exceeding 90%. With laser power,
2
manufacturing has come to the forefront of scholarly travel speed, and powder feed rate as variables, Zhu et al.
28
interest. By steering the arc, the laser decreases its electrical employed support vector regression (SVR) to predict the
resistance and field strength, which in turn improves arc layer height and width of DED, reaching 93% accuracy.
stability and boosts the build rate. When the two energy Deploying a naïve Bayes approach, Liu and Kuo studied
18
29
sources are coupled, inverse bremsstrahlung further wire-laser DED; by inputting travel velocity, laser power,
augments laser-energy uptake, and the laser improves melt- and wire feed speed, they predicted bead dimensions,
pool convection as well as the homogeneity of elemental yielding R² values of 91% and 94%. Xia et al. employed
30
distribution. Even so, integrating laser and arc complicates a genetic-algorithm-tuned adaptive neuro-fuzzy inference
19
the multiphysics interactions, greatly raising the challenge system model to estimate surface roughness from wire
of forecasting weld-bead dimensions. Consequently, the feed and welding speeds, achieving validation R values
2
precise prediction of weld-bead geometry in laser-arc exceeding 90%. Oh et al. introduced an SVR model into
31
hybrid additive manufacturing (LAHAM) emerges as a the WAAM process, using key parameters such as welding
pivotal problem that demands prompt resolution. current and wire feed rate to accurately predict bead width
As an essential subdivision of artificial intelligence and height and to identify geometric non-uniformity
(AI), machine learning (ML) identifies patterns through defects that may arise under different operating conditions.
32
data analysis and leverages them for prediction and Šket et al. used an extreme gradient boosting (XGBoost)
decision-making tasks. During the past decade, ML has regression model, with current and travel speed as the two
20
found broad applications across numerous domains, such process parameters, to predict the geometric morphology
as medical diagnostics, forecasting material properties, of the weld bead.
22
21
intelligent manufacturing, self-driving vehicles, natural Despite these advances in predicting layer geometries of
24
23
language processing, and object detection. A synopsis of MAM based on process parameters, reliable estimation of
26
25
research forecasting the links between process parameters bead size in LAHAM still confronts significant challenges.
and layer geometry in MAM is provided in Table 1. Using The link between LAHAM processing parameters and
Volume 4 Issue 3 (2025) 2 doi: 10.36922/MSAM025220036