Page 44 - ESAM-1-1

Engineering Science in
Additive Manufacturing Gen-AI for lattice structure design

Figure 5. Overview of the roles of Gen-AI in accelerating lattice structure design and innovation
Abbreviation: Gen-AI: Generative artificial intelligence.

scalability and accuracy. Besides, hybrid approaches that is imbalanced or lacks diversity in lattice types, material
combine the strengths of Gen-AI with conventional design types, or process conditions, the generated designs may
techniques are gaining attention: (1) enhancing topology be suboptimal or non-generalizable. This limitation is
optimization, Gen-AI can accelerate iterative optimization particularly critical in material-specific lattice design,
cycles by proposing initial design configurations that meet where variations in material behavior, such as differences
predefined constraints; this reduces computational overhead between metallic and polymeric lattices, can significantly
and expands the design space exploration; (2) refining influence the feasibility and quality of generated designs.
parametric models, Gen-AI models can learn from existing To mitigate model bias, future research should focus
parametric data and suggest novel geometries that go on expanding high-quality databases that cover diverse
beyond traditional rule-based limitations; and (3) coupling lattice geometries, materials, and process parameters. This
with physics-based simulations, Gen-AI surrogate models can be achieved through high-throughput simulations,
can replace computationally intensive simulations, offering standardized experimental protocols, and synthetic
rapid but approximate predictions, which can later be fine- data augmentation strategies. Moreover, promoting the
tuned using high-fidelity simulations. development of open-access datasets could facilitate
broader model training and validation. Integrating
6.2. Perspectives physics-informed learning approaches could also guide
In conclusion, Gen-AI has redefined the possibilities in model predictions, ensuring that generated designs adhere
lattice structure design for AM, bridging the gap between to fundamental physical and material constraints.
innovative conceptualization and manufacturable reality. Another major limitation is the lack of interpretability
By enabling efficient exploration of vast design spaces, in Gen-AI models. 63,64 The decision-making processes
automating optimization processes, and enhancing within complex neural networks, such as GANs or
simulation workflows, Gen-AI stands as a cornerstone diffusion models, are often opaque, making it challenging
for the next generation of advanced lattice structures. for engineers to understand why certain lattice geometries
Future research should focus on enhancing multiobjective are proposed. This “black-box” nature can reduce trust
optimization capabilities, improving the interpretability and hinder the adoption of Gen-AI in critical applications,
of AI models, and ensuring the seamless integration of particularly in sectors like aerospace and biomedical
Gen-AI into industrial AM workflows. engineering, where design validation is stringent.

Despite the transformative potential of Gen-AI in Addressing interpretability challenges requires the
lattice structure design, several inherent challenges and development of explainable AI frameworks tailored for
limitations must be addressed to ensure its robust and lattice design. For instance, incorporating visualization
effective application in AM. One significant challenge is techniques that highlight which features influence design
the bias inherent in Gen-AI models, which stems primarily decisions could enhance understanding. Implementing
from the limitations of the training dataset. 61,62 If the dataset sensitivity analyses to understand how variations in

Volume 1 Issue 1 (2025) 9 doi: 10.36922/ESAM025110006