Page 120 - MSAM-3-2

Materials Science in Additive Manufacturing
Figure 5. Architecture of the back propagation neural network. generated. The mechanical properties of these lattices
A B
Figure 6. Probability density of (A) P-Honeycomb cell and (B) G-Honeycomb cell in the dataset.
Volume 3 Issue 2 (2024)
Hybrid lattice structures design with AI
along the X- and Y-direction were then predicted based on
the hybrid lattice design. Figure 10 depicts the correlation
between the actual properties of the hybrid lattice and the
predictions made by the BPNN model.
Notably, the model exhibited optimal performance
in predicting the modulus along the X-direction within
a mid-range of values, suggesting a robust capability
to capture moderate variations in material stiffness.
However, it showed tendencies to overestimate modulus
values below 12 MPa and underestimate those exceeding
20 MPa, highlighting potential limitations in accurately
predicting extreme values (Figure 10A). Similarly, for the
modulus along the Y-direction, the model demonstrated
overestimations for low values and underestimations for
high values, indicating a systematic bias in some areas
of the property space (Figure 10C). Regarding Poisson’s
ratio predictions, the model showed improved accuracy
within a mid-range of values, aligning with its proficiency
in capturing moderate variations (Figure 10B and 10C).
Nonetheless, slight underestimations and overestimations
were detected for low and high values, respectively,
suggesting regions for refinement in accurately predicting
extreme ratios.
These findings proved the performance of the BPNN
model and provided valuable insights into its strengths
and limitations in predicting the mechanical properties
of hybrid lattice structures based on their topologies.
Further analysis and refinement of the model may enhance
its predictive capabilities and broaden its applicability in
material design and engineering.
3.4. Validation of BPNN
To further validate the capability of the trained BPNN,
a dataset comprising five random hybrid lattices was
7 doi: 10.36922/msam.3430