Page 39 - IJAMD-1-1

International Journal of AI
for Material and Design ML for quality improvement in L-PBF

careful consideration of the specific characteristics and Akbari et al. introduced a comprehensive framework
requirements of the data and the prediction task at hand. for benchmarking ML for melt pool characterization.
37
34
Wang et al. established a correlation between key process A substantial collection of melt pool characteristics data
parameters and quality indicators at two levels: the was acquired from L-PBF experiments and employed for
mesoscale level and the macroscale level. Specifically, the developing the ML model. Various ML models were built
35
mesoscale level pertains to porosity or relative density, as to predict melt pool characteristics. Among these models,
well as melt pool geometries, while the macroscale level NNs and gradient boosting consistently demonstrated
refers to mechanical properties. This approach allows for a superior predictive performance on regression tasks.
comprehensive understanding of the relationship between Furthermore, the study emphasized the critical importance
process parameters and quality indicators, enabling of selecting appropriate feature engineering techniques to
informed decision-making and process optimization. enhance the accuracy of the predictions.

3.1.1. Melt pool characteristic The research conducted by Lee et al. focuses on the
application of advanced data analytics for understanding
The characteristics of the melt pool exert a significant and predicting the formation of melt pools within a
impact on the quality of metal 3D-printed products. The physics-based context. Their investigation reveals that
38
shape and geometry of the melt pool can influence the the dimensions of the melt pool, namely its depth, width,
microstructure of the material. The utilization of a well- and height, are closely related to various physical factors.

defined and stable melt pool yields a more uniform and Specifically, the depth of the melt pool correlates with laser
fine-grained microstructure. In addition, the geometric penetration, the width is tied to fluid convection, and the
structure of the melt pool plays a critical role in determining height corresponds to the mechanical properties of the
the occurrence of defects such as lack of fusion, balling, melted powders.
or keyhole porosity. Irregular melt pool shapes are often
36
associated with defects that undermine the mechanical To predict the geometry of the melt pool, the researchers
properties of the printed component. Therefore, accurate employed six optimized ML algorithms: Bayesian ridge
prediction of melt pool characteristics is essential, and it (BR), kernel ridge (KR), LR, NN, random forest (RF),
can lead to enhanced optimization of the L-PBF process. and SVM. As depicted in Figure 10, the graph indicates

Figure 10. Accuracy analysis of the five targets (width [w], depth [d], area within substrate [A ], height [h], and area based on height [A ]) for the six
h
sub
optimized machine learning algorithms. 38
Volume 1 Issue 1 (2024) 33 https://doi.org/10.36922/ijamd.2301