Page 57 - MSAM-2-1

Materials Science in Additive Manufacturing Data imputation strategies of PBF Ti64

imputation gave the closest distribution to the original Conflict of interest
dataset and was the most accurate method, achieving the
lowest RMSE. A median approach was used by taking the The authors declare that they have no known competing
median of the imputed values from the three models. It financial interests or personal relationships that could have
was found that the median approach further improved the appeared to influence the work reported in this paper.
imputation accuracy by achieving RMSE of 0.026. Author contributions
Data mining of the imputed SLM Ti64 dataset Conceptualization: Guo Dong Goh
using SOM identified correlations between the process Data curation: Jia Li Janessa Thong, Jia Jun Seah
parameters and material properties. These correlations Formal analysis: Jia Li Janessa Thong, Jia Jun Seah
can be utilized to help users identify suitable process Funding acquisition: Wai Yee Yeong
parameters for specimens with certain desired properties. Investigation: Sheng Huang, Jia Li Janessa Thong, Jia Jun Seah
While the material properties of monotonic yield strength Methodology: Xi Huang
and elongation at break are important, there are many other Project administration: Wai Yee Yeong
properties that could be of interest in the field of additive Resources: Wai Yee Yeong
manufacturing. For example, fatigue strength, fracture Supervision: Wai Yee Yeong
toughness, creep resistance, and corrosion resistance are Validation: Guo Dong Goh
all important material properties that could be explored.
In addition, exploring the relationship between process Writing – original draft: Jia Li Janessa Thong
parameters and microstructure, such as grain size, Writing – review & editing: Guo Dong Goh
could also provide valuable insights for optimizing the Data availability
manufacturing process. The presented approach can also
be applied to other databases to obtain new knowledge Data are available on request.
from the database. References
However, a major limitation of the imputation methods
is that a large proportion of missing data would lead 1. Liu S, Shin YC, 2019, Additive manufacturing of Ti6Al4V
to more inaccurate data imputation, and more manual alloy: A review. Mater Des, 164: 107552.
checking of individual imputed values to ensure validity https://doi.org/10.1016/j.matdes.2018.107552
would be required. Imputation is not always appropriate 2. Elsayed M, Ghazy M, Youssef Y, et al., 2019, Optimization
and may introduce bias or lead to incorrect conclusions if of SLM process parameters for Ti6Al4V medical implants.
the missing data is non-random or missing not at random. Rapid Prototyp J, 25: 433–447.
There is no specific quantitative threshold or limit to how https://doi.org/10.1108/rpj-05-2018-0112
much imputation can be performed before the results
become meaningless. The validity of imputed data depends 3. Roudnicka M, Bigas J, Vojtech D, 2020, Tuning porosity
on the accuracy of the imputation models, the quality of the and mechanical properties of Ti6Al4V alloy additively
original data, and the degree and pattern of missingness. manufactured by SLM. In: Key Engineering Materials.
Vol. 865. Trans Tech Publications, Switzerland, p1–5.
In summary, imputation using the median approach was 4. Popovich A, Sufiiarov V, Borisov E, et al., 2015,
found to be the most accurate method for imputing missing Microstructure and mechanical properties of Ti-6Al-4V
data in the Ti6Al4V dataset, and the data mining approach manufactured by SLM. In: Key Engineering Materials.
using SOM identified correlations between process Vol. 651. Trans Tech Publications, Switzerland, p677–682.
parameters and material properties. However, the study
also highlights the need for more standardized testing and 5. Thijs L, Verhaeghe F, Craeghs T, et al., 2010, A study of the
microstructural evolution during selective laser melting of
reporting, and the limitations of data imputation methods Ti-6Al-4V. Acta Mater, 58: 3303–3312.
when dealing with a large proportion of missing data.
https://doi.org/10.1016/j.actamat.2010.02.004
Acknowledgments 6. Kuo C, Su C, Chiang A, 2017, Parametric optimization of
None. density and dimensions in three-dimensional printing of
Ti-6Al-4V powders on titanium plates using selective laser
Funding melting. Int J Precis Eng Manuf, 18: 1609–1618.
This research is supported by the National Research https://doi.org/10.1007/s12541-017-0190-5
Foundation, Prime Minister’s Office, Singapore under its 7. Pal S, Lojen G, Kokol V, et al., 2018, Evolution of metallurgical
Medium-Sized Centre funding scheme. properties of Ti-6Al-4V alloy fabricated in different energy

Volume 2 Issue 1 (2023) 16 https://doi.org/10.36922/msam.50