Page 63 - IJAMD-1-2

International Journal of AI for
Materials and Design
AI-assisted ML monitoring in additive auxetics

by red circles. These locations are consistent with those ii. Advanced DL architecture: To address the high
predicted by FEM and the DL model. Consequently, the computational costs associated with evaluating a
Mnet provides accurate predictions of effective strain large design space of auxetic structures, an advanced
fields for diverse auxetic designs, matching well with both DL architecture, the modified MNet, was employed.
numerical analysis and experimental observations. This DL model achieved over 90% accuracy and

However, certain discrepancies still exist between FEM demonstrated evaluation speed approximately
4
predictions and ML measurements. Two potential causes 10 times faster than FEA, using a relatively limited
are hypothesized: dataset of 387 training samples.
iii. Experimental validation: The model’s predictions
i. Material property deviation: Specimens fabricated were validated using experimental techniques
through additive manufacturing may have limited incorporating ML-aided non-contact evaluation.
particle uniformity, whereas FEM assumes perfect The results showed good agreement, particularly in
homogeneity, potentially leading to differences in areas with high-stress concentrations. This validation
intensity distribution. underscores the effectiveness of the data-driven model
ii. Precision of fabrication: The precision of the 3D in accurately predicting strain fields.
printer may be limited, with DLP-printed specimens iv. Comparison with the DIC technique: The ML-based
exhibiting weakened interlayer adhesion, introducing structural reliability evaluation technique was
variation in mechanical properties. FEM analysis, on calibrated and validated against the DIC technique.
the other hand, assumes perfect bonding with every While DIC provides accurate strain field assessments,
portion of the structure. its high cost and complex setup limit its applicability.
Addressing these potential issues could improve the The ML method, through careful mapping between
agreement between predicted and measured values. ML intensity and strain fields, demonstrated excellent
An investigation into the relation between normalized agreement with DIC measurements, highlighting its
intensity and structural Poisson’s ratio was conducted for the potential as a cost-effective SHM technique.
three designs. The results, shown in Figure S5, indicate that v. Practical applicability: The study demonstrates the
auxetic structures with high NPR exhibit high intensity at feasibility of simultaneous prediction, fabrication,
the same global deformation. This trend is due to increased and direct evaluation of structural reliability in
rotational deformation within the unit cell in high NPR various intricate structures, exemplified by auxetic
structures, leading to increased localized strain and thus metamaterial designs. The proposed ML methodology
high-intensity values. Given the favored NPR characteristics shows significant potential for practical applications
of auxetic structures in design applications, the ML method requiring high sensitivity to deformation, including
demonstrates high sensitivity for the designs. Therefore, aircraft design, pipe monitoring, and sensors.
future designs of various auxetic structures could benefit In conclusion, the research provides a robust framework
from the application of ML methods for real-time for effective strain prediction and structural reliability
monitoring and effective verification of structural reliability. assessment using ML-based methods. The results highlight
the practicality and efficiency of these techniques in real-
4. Conclusion world applications, paving the way for advanced SHM
This study presented a comprehensive framework for systems that are both cost-effective and highly sensitive to
predicting the effective strain in auxetic structures, validated structural deformations.
through ML-aided non-contact reliability evaluation
of 3D-printed specimens. The findings underscore the Acknowledgments
potential of ML-based structural reliability assessments as Not applicable.
cost-effective and deployable solutions for SHM applications,
particularly for intricate designs such as auxetic structures. Funding

The key contributions of the study include: This work was financially supported by the National
i. Unique parametrization with Bézier curves: The use Research Foundation of Korea (NRF) (RS-2023-00222166).
of Bézier curves for parametrizing auxetic structures Conflict of interest
ensured smooth surfaces, efficiently avoiding stress
concentrations associated with sharp corners. This Seunghwa Ryu is an Editorial Board Member of this
technique enhances the overall structural integrity journal, but was not in any way involved in the editorial
and reliability of the designs. and peer-review process conducted for this paper, directly

Volume 1 Issue 2 (2024) 57 doi: 10.36922/ijamd.3539