Page 39 - ESAM-1-4

Engineering Science in
Additive Manufacturing Machine learning for biomedical metal AM

Figure 12. Workflow for machine learning-based multi-objective inverse optimization of process parameters. Image created by the authors.
Abbreviations: DNN: Deep neural network; EL: Elongation; SVM: Support vector machine; UTS: Ultimate tensile strength.

et al. used the NSGA-II to solve for the optimal LB-PBF With the deepening of the green manufacturing
118
process parameters that would simultaneously maximize concept, process optimization is no longer limited to part
energy absorption and minimize density in bio-inspired performance. Peng et al. aimed to identify pareto-optimal
121
lattice structures. The inverse optimization approach process parameters that would minimize specific energy
determined the specific laser parameters and scanning consumption while maximize powder usage rate. They
strategies that achieved different compromises between constructed predictive models from process parameters
these competing structural performance objectives. to these two economic/environmental indicator and
used the NSGA-II algorithm for multi-objective inverse
Regarding the synergistic control of geometric optimization. The inverse optimization framework solved
accuracy, precise control over the deposition morphology for the laser power and scanning parameter combinations
is crucial when manufacturing customized biomedical that achieved different trade-offs between economic and
metals with complex geometries. Cai et al., working environmental performance indicators.
119
with wire-laser DED, inversely derived optimal process
parameter combinations that would simultaneously The principal value of multi-objective inverse
control the deposited layer height and width. They optimization lies in its ability to transform process
established a forward prediction model using SVR and development from an experience-dependent trial-and-
coupled it with the NSGA-II algorithm for inverse MOO. error approach into a precise, informed decision-making
They successfully inversely derived the optimal process process grounded in systematic trade-offs. By analyzing the
parameter combinations that could synergistically meet shape and distribution of the Pareto front, researchers can
these two-dimensional requirements from the desired quantify the conflict level between different performance
geometries. indicators. This clear trade-off landscape provides a
critical foundation for final decision-making. In clinical
In the parameter inverse identification for coordinating applications, the optimal solution can be selected from a
density and controlled porosity, this represents a particularly set of Pareto-optimal process parameter sets based on the
classic contradiction in the biomedical field. Biomedical specific implantation site and patient needs, thereby best
metals often require both high densities to ensure addressing the immediate clinical priorities.
mechanical performance and specific surface or internal
porosity to promote osseointegration. Heiss et al., during 4. ML-driven quality control and
120
the LB-PBF manufacturing of biodegradable Zn alloy, monitoring
innovatively used principal component analysis to reduce
the dimensionality and quantify the multi-dimensional 4.1. Overview
pore morphology characteristics, subsequently creating a The AM technology, particularly for biomedical metals (e.g.,
clear process map. The map effectively inversely identified titanium alloys, cobalt-chromium alloys, and biodegradable
the process window capable of simultaneously achieving magnesium alloys), has garnered significant attention due
high bulk density and ideal pore morphology, successfully to its unique advantages in producing complex porous
producing demonstrator parts that achieved a balance structures and enabling personalized customization.
between mechanical performance and biological function. However, physical variability during manufacturing,

Volume 1 Issue 4 (2025) 17 doi: 10.36922/ESAM025440031