Page 56 - MSAM-2-1

Materials Science in Additive Manufacturing Data imputation strategies of PBF Ti64

data. The study also demonstrates a practical approach compared to those produced at a speed of 1150 mm/s,
to combining multiple imputation models to enhance the and the shape was similar to that of heat-treated
statistical confidence of the imputed data, which can be samples, as documented by Vilaro et al. [11,42] . As the scan
useful in other domains as well. speed increased to 1150 mm/s, the quantity of β-phase
nanoparticles reduced, and only few white particles were
4. Visualizing relationships between dispersed across the fine acicular α (α′) grain boundaries.
variables in imputed dataset There are several factors that can contribute to the
Heatmaps for each variable was used to visualize the SOM relationship between laser power and Young’s modulus in
trained, produced by linking each variable value to a node SLM of metals. One possible explanation is related to the
on the map grid (Figure 14). The relationships between the changes in microstructure and grain size that can occur
variables can be determined based on visual analysis, by as a result of varying laser power levels. High laser power
comparing the locations of the red and blue regions that can lead to rapid melting and solidification of the material,
correspond to high and low values respectively. Based on resulting in smaller grain sizes and higher dislocation
the heatmaps, the following observations are made from densities, which can contribute to an increase in Young’s
the imputed dataset, some of which are well established modulus.
relationship:
It is noted from the heatmap that there is a slight
(i) Porosity is inversely related to Young’s modulus and correlation between laser power and Young’s modulus.
yield strength. However, the effect of laser power on the Young’s modulus
(ii) Scan speed is inversely related to microhardness and in SLM Ti64 is not straightforward and depends on several
macrohardness. factors. At low laser powers, the material experiences less
(iii) Exposure duration is directly related to macrohardness. thermal input and solidifies with a finer microstructure,
(iv) Ultimate tensile strength, yield strength, and resulting in a higher Young’s modulus due to the increased
elongation are directly related. strength of the material. However, as the laser power
(v) Energy density and scan speed are inversely related. increases, the material is heated to a higher temperature,
(vi) Laser power and Young’s modulus are slightly directly resulting in coarser microstructures due to increased grain
related. growth and leading to a decrease in the Young’s modulus.
These relations provide insight into the process- Furthermore, excessive laser power can result in porosity
property relationships in SLM Ti64 and can help users and defects in the material, which can significantly reduce
[10]
determine the process parameter window to obtain certain the Young’s modulus . Therefore, the laser power should
desired material properties. For example, to obtain a be optimized to achieve the desired microstructure and
specimen with higher hardness, a lower scan speed, higher avoid porosity formation to ensure that the built parts
energy density and longer exposure duration should be have the required Young’s modulus for their intended
used. A higher laser power is also likely to result in higher application.
Young’s modulus. The inverse relationship between the Overall, it is important to consider the complex
energy density and scan speed are found to be consistent interplay between multiple process parameters and
with the energy density equation. material characteristics that can affect mechanical
The inverse relationship between scan speed and properties in SLM. This also highlights the importance
microhardness in SLM Ti64 can be explained by several of including scan strategies and microstructures in the
factors. First, high scan speeds can also lead to incomplete dataset for better generalization of the process-structure-
melting, resulting in the formation of unmelted or partially properties relationship of SLM Ti64.
melted particles, which can act as a source of defects 5. Conclusions
[41]
and lower the microhardness . Second, there could be
a change in microstructure of the printed Ti64 from a In this study, three model-based imputation techniques,
coarser equiaxed grains to a finer columnar grains as the kNN, MICE, and GINN imputations, were used to
scan speed increases. When the scanning speed is slow, the impute missing values in the Ti6Al4V dataset, which
laser’s slower movement increases both the energy input contained various process parameters and material
and stability of the molten pool. The elevated temperature properties obtained from multiple sources available in the
in the molten pool creates adequate energy and nuclei for literature. The results of the imputations were evaluated
the epitaxial growth of columnar grains in the building using graphical checks and statistical summaries to
direction. Wang et al. observed the coarsening of acicular compare the imputed data with the original distribution
structures in the samples produced at a speed of 250 mm/s before imputation. Among the three techniques, GINN

Volume 2 Issue 1 (2023) 15 https://doi.org/10.36922/msam.50