Page 18 - IJB-8-1

Additive Manufactured Beta-Titanium Alloys

Figure 3. A conceptual framework for predicting manufacturability for powder bed fusion using machine learning.

also fabricated using Ti24Nb4Zr8Sn and the samples the input and output of L-PBF. Smaller data sets can
have Young’s modulus of 0.95 ± 0.05 GPa [50] . Porous be used to develop and train the machine learning
lattice structures are also fabricated using β-phase algorithms by measuring the part quality from different
Ti24Nb4Zr8Sn [51] . Ti35Zr28Nb was used to fabricate process parameters. Predictive models using genetic
porous lattice structures with FCCZ and FBCCZ programming and neural network modeling for process
unit cells. While the bulk material elastic modulus is planning purposes have been developed for L-PBF .
[57]
between 57.0 ± 1.8% GPa, the porous lattice structures Deep neural networks can also be used to interpret and
are able to achieve elastic modulus of between 1.1 ± classify melt pool images that can be used to predict the
0.1% GPa and 1.3 ± 0.1% GPa [52] . It was also concluded defects during the process [58,59] . Shin et al. developed an
that the structures have good corrosion behavior and artificial neural network to identify the optimum L-PBF
biocompatibility. Using Ti25Nb3Zr3Mo2Sn to fabricate process parameters to obtain defect-free Ti5Al5V5Mo3Cr
porous lattice structures by L-PBF, Liu et al. found (wt%) and showed the potential of using this approach to
out that martensitic phase transformation takes place reduce processing time and making the L-PBF process
during yielding under compression loading for the more cost effective . A conceptual framework on using
[60]
samples, which prevents the samples from layer-wise machine learning to predict manufacturability for PBF is
fracture [53] . Such phase transformation is also found to shown in Figure 3.
improve the tensile strength of β-Ti alloys [54] . Using Comprehensive reviews on machine learning for
EB-PBF, Liu et al. fabricated porous structures made AM are available [56,61-63] . Using data-driven artificial
of Ti24Nb4Zr8Sn and compared the effect of different
designed porosities [55] . intelligence, it is possible to predict and optimize process
parameters that can obtain desired part properties [64,65] .
3.3. Machine learning and artificial intelligence Furthermore, a conceptual framework that combines
statistical analysis, mathematical modeling, and machine
The fluctuations in L-PBF make process control and learning techniques has also been proposed .
[66]
prediction challenging. While physical simulations have
been developed to provide deeper understanding of 4. Summary
the process, it is still challenging for them to show the
full aspects of L-PBF. Machine learning and artificial In this perspective article, the feasibility of processing
intelligence can be used to enhance and complement β-Ti alloys using PBF has been shown using currently
simulations in achieving higher part quality . The available literature. It has been shown that these alloys
[56]
current practice to obtain defect-free parts, or commonly have potential to be better candidates for biomedical
known as the process parameters optimization, involves implants due to their lower elastic modulus and high
multiple rounds of experiments which are costly and mechanical strength. However, to realize their full
time intensive. Machine learning techniques potentially potential, future research in the area of new processing
can be applied to establish relationships between approaches and new designs are still needed.

4 International Journal of Bioprinting (2022)–Volume 8, Issue 1