Page 55 - IJAMD-1-2

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

exceptional impact absorption, and augmented fracture mechanical properties by achieving more uniform stress
resistance. They reduced the Poisson’s ratio while amplifying distribution and reducing stress concentrations.
7
strain sensitivity compared to conventional materials. 8,9 This study aims to leverage an AI-driven framework to
This heightened sensitivity is particularly pronounced predict effective strain within complex auxetic structures
in auxetic structures that leverage rotating motion and employ ML materials to directly monitor strain fields.
under tensile stress to amplify their NPR behavior. 10,11 Auxetic designs incorporating ML were realized through
Such sensitivity is, especially beneficial in fields such 3D printing processes, and the predictive model’s accuracy
as structural health monitoring (SHM) and sensor was validated by quantifying ML intensity. The observed
technology, where precise monitoring of structural changes linear relationship between ML intensity and effective
is crucial. However, the distinctive deformation responses strain, alongside comparisons with actual ML images,
of auxetic structures can lead to localized stress and strain underscores the potential of 3D printing to produce
concentrations, which must be addressed during the design complex auxetic structures embedded with ML particles
process to prevent potential microcrack formation and for instantaneous verification of strain fields. Moreover,
failure over prolonged use. Hence, accurately measuring the application of ML in digital light processing (DLP)-
the spatial distribution of strain in both the design and based functional 3D printing demonstrates its broad
application phases of auxetic materials is crucial. 12-15 potential across structural applications, confirming its
feasibility for sensing and SHM. This research illustrates
To evaluate deformation without causing material damage
or alterations, non-contact methods have been developed to a comprehensive approach where prediction, fabrication,
and direct verification of structural properties, focusing
visualize strain fields. These methods encompass thermal, on the effective strain, are achievable simultaneously. It
16
acoustic, 17,18 optical, 19,20 and luminescent 21-23 techniques, showcases the versatility and practical applicability of
each offering distinct advantages and challenges. Among this innovative methodology in fields requiring acute
these, mechanoluminescence (ML) materials have emerged sensitivity to deformation, such as civil engineering,
34
as a novel non-contact method. ML materials can visualize aerospace engineering, and monitoring deformation in
35
stress and strain fields and track crack paths in response pipes and sacrificial layers. 36,37
to mechanical stimuli, 24,25 offering the potential to map
the spatial distribution of local mechanical responses. 2. Methods
12
3+
2+
Specifically, SrAl O : Eu , Dy (SAOED), a representative
4
2
material used in ML composites, demonstrates excellent 2.1. Generation of auxetic structures
reproducibility and a linear response to stress. This 2.1.1. Parametrization and design of auxetic structures
capability allows ML to provide strain distribution In this study, we designed auxetic structures with
information comparable to traditional methods such as NPR characteristics utilizing Bézier curves, which are
digital image correlation (DIC) by quantifying light parametric curves defined by a set of control points,
26
intensity without the use of speckle patterns. Given the offering continuous and smooth curves. The Bézier curves
12
high costs and extensive pre-processing required for DIC characterize the void domain within the unit cell of auxetic
setup and measurements, ML-aided evaluations of localized structures (Figure 1A) and are defined by Equation I:
strains are expected to offer a cost-effective and deployable
n
solution in SHM for various structures. n   ni
−
( − t
B ( ) = ∑   it ) t i P , ∈ t  0,1 , (I)

i
i
The predicted strain distribution from finite element i =  
0
analysis (FEA) 27-29 and those derived from ML-based SHM
methods can be effectively combined to enhance the design Where t is a continuous value between 0 and 1, P is the
i
reliability of auxetic structures. With advancements in set of control points, and denotes the binomial coefficient.
additive manufacturing, it is now possible to fabricate highly We employed four control points (n = 4) to balance design
38
complex auxetic designs. However, the growing complexity capability and cost for exploring design variables. For
of these structures, which encompass a vast design space, the design of a single closed curve, a total of eight Bézier
necessitates even more efficient prediction and design curves were utilized, as shown in Figure 1B. These eight
techniques. To meet these demands, recent developments Bézier curves were comprised two symmetrical shifts of
have introduced artificial intelligence (AI). By integrating AI two curves generated in the first quadrant, satisfying C
1
39
with FEA, various physical fields, such as stress fields, strain and C continuities.
2
fields, and crack behaviors, as well as diverse design spaces, Subsequently, unit cells were constructed by rotating
can be explored efficiently. 30-33 Consequently, through the and symmetrically shifting the generated closed Bézier
analysis of physical fields, structural engineers can enhance curves, defined by four design variables l , l , x, and y, as
a
b
Volume 1 Issue 2 (2024) 49 doi: 10.36922/ijamd.3539