Page 38 - ESAM-1-4

Engineering Science in
Additive Manufacturing Machine learning for biomedical metal AM

A

B

Figure 11. Machine learning–driven additive manufacturing process optimization for nickel-based high-temperature. (A) SOM including input vectors.
(B) Identification of the optimal design window through contour plots of all design variables. 114
Abbreviations: SDAS: Secondary dendrite arm spacing; SOM: Self-organizing map.

particularly in the customized development of implants optimization identifies Pareto-optimal process parameter
for specific clinical application scenarios. When a single sets that achieve different balances of conflicting
performance indicator becomes the key design constraint, performance targets, and the workflow is illustrated in
focusing optimization on that specific target can ensure Figure 12.
the implant’s reliability and safety in that particular Aboutaleb et al. addressed the classic trade-off in
107
environment.
Ti-6Al-4V mechanical properties by inversely identifying
3.4. Multi-objective inverse optimization Pareto-optimal process parameter combinations that
simultaneously maximized the elastic modulus and UTS.
The service performance of biomedical metals often The m-APO framework outputs a set of Pareto-optimal
requires the synergistic satisfaction of multiple process parameters, and compared to time-consuming
performance indicators, which frequently exhibit trade-
off relationships. For instance, increasing strength may full-factorial experimental designs, it reduces the
computational cost by 51.8%, highlighting its significant
sacrifice ductility, while reducing the elastic modulus may
compromise fatigue performance due to the introduction value in rapid process development.
of excessive porosity. MOO algorithms generate Pareto Beyond bulk materials properties, MOO methods
fronts, enabling selection of process parameters based are equally suitable for guiding the design of complex
on clinical needs. 116,117 ML-driven multi-objective inverse structures with specific functional requirements. Meng

Volume 1 Issue 4 (2025) 16 doi: 10.36922/ESAM025440031