Page 19 - IJAMD-2-2

International Journal of AI for
Materials and Design AI-driven material development for AM

structural homogeneity and stability. Mechanical property within PSP relationships, allowing for improved process
optimization, on the other hand, primarily focuses on the parameter selection to minimize defects and enhance
enhancement of properties, including strength, hardness, mechanical performance. The model showed strong
ductility, and fracture toughness, which are typically agreement with experimental results, suggesting reliable
realized through appropriate microstructural and process prediction accuracy.
control. In addition, Gaussian process-based models have been
Recent advancements in AI-driven methodologies explored to optimize process parameters for improved
80
have enabled the establishment of correlations between mechanical performance. Tapia et al. developed a
processing parameters and microstructural characteristics Gaussian process-based surrogate modeling framework
in AM. By leveraging ML models trained on experimental to predict melt pool depth in LPBF of 316L stainless steel,
and computational datasets, AI enables more accurate thereby identifying processing windows that enhance
predictions of microstructural evolution and mechanical part quality. Their approach enabled the classification of
properties under varying process conditions. For example, conduction-mode and keyhole-mode melting regimes,
Awd et al. integrated mechanistic ML with physics-based which directly affect the microstructure and, consequently,
63
models to predict and optimize the fatigue strength of the mechanical properties of AM components.
60
AM-fabricated metamaterials. Their approach combines Chernyavsky et al. introduced a Heteroscedastic
electronic structure calculations, stochastic process Gaussian process (HGP) model to predict the amorphicity
modeling, and microstructural characterization to of a Zr-based glass-forming alloy fabricated through
establish process-structure-property (PSP) relationships. LPBF. This model effectively establishes a quantitative link
Through μ-CT imaging and defect quantification, between LPBF conditions and microstructural evolution.
they demonstrated how AI-driven methodologies can Figure 8 illustrates the predictive capability of the HGP
enhance fatigue damage modeling, enabling more model, covering amorphicity distribution (Figure 7A),
accurate predictions of material performance under cyclic uncertainty quantification (Figure 7B), and prediction
loading. Similarly, Liu et al. employed Gaussian process accuracy (Figure 7 C). These results highlight the model’s
62
regression (GPR) to model the complex relationships robustness in not only delivering accurate amorphicity
between processing parameters, microstructure, and predictions but also in assessing dataset reliability and
mechanical properties in laser powder bed fusion (LPBF)- identifying underlying physical mechanisms governing
fabricated AlSi10Mg. GPR was utilized to predict density glass formation in AM.
variations and microstructural characteristics based The integration of AI-driven methodologies in alloy
on key process parameters. Their study demonstrated performance optimization has significantly advanced
that GPR effectively captures non-linear dependencies the understanding and prediction of PSP relationships.

A C

B

Figure 8. Predicted amorphicity distributions and associated uncertainty for alloys fabricated through laser powder bed fusion. (A) HGP model predictions
for mean values of amorphicity and its total uncertainty. (B) Position-resolved aleatoric and epistemic uncertainties predicted by the HGP model. (C) Two-
dimensional contour maps of HGP model predictions for mean values of amorphicity and its total uncertainty. Reproduced from Chernyavsky et al. 60
Abbreviation: HGP: Heteroscedastic Gaussian process.

Volume 2 Issue 2 (2025) 13 doi: 10.36922/IJAMD025100007