Page 45 - ESAM-1-1

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

input parameters, such as material properties and process solutions. Furthermore, exploring design-for-sustainability
parameters, affect the final design outcomes would also be strategies, where Gen-AI models are guided to minimize
valuable. In addition, integrating physics-guided neural material usage, maximize recyclability, or enhance the
networks could help ensure that outputs are consistent life-cycle performance of lattice structures, is essential.
with established physical laws, enhancing both trust and Integrating life-cycle assessment into the optimization
interpretability. process can ensure that generated designs not only
meet performance requirements but also adhere to
While Gen-AI models excel in generating designs
based on learned distributions, they may struggle when environmental sustainability goals.
confronted with novel design requirements or unique While current Gen-AI frameworks consider general
material-process combinations not represented in the processing parameters, there is a need for more material-
training data. For example, a Gen-AI model trained on specific process optimization, particularly for complex
titanium lattice structures may not generalize well to high- materials like high-strength alloys or biocompatible
entropy alloys or biocompatible polymers. Enhancing polymers. 67-69 Understanding how Gen-AI can better
adaptability requires the development of transfer learning integrate AM process constraints, such as residual stress,
strategies, where models trained on one material system printability, and thermal gradients, will enhance the
can be fine-tuned for another with minimal additional data. manufacturability of generated lattice designs. Research
Utilizing active learning frameworks, where the model can should focus on developing material-specific Gen-AI
iteratively select new data points for labeling, can improve models that explicitly account for unique processing
generalizability to novel conditions. Incorporating behaviors, ensuring that generated designs are both optimal
multifidelity data, combining low- and high-resolution and manufacturable. In addition, integrating physics-based
data, could expand the learning scope when minimizing constraints into Gen-AI frameworks could ensure designs
computational costs. are feasible under realistic processing conditions.
Although Gen-AI significantly accelerates lattice Finally, transitioning from conceptual Gen-AI
design, hybrid approaches that combine Gen-AI with designs to practical industrial applications remains a
traditional computational techniques, such as topology challenge. The gap between model-generated designs and
optimization and FEM simulations, can offer more reliable manufacturable solutions often leads to inefficiencies and
and manufacturable solutions. For instance, while Gen-AI increased post-processing efforts. 70-72 Bridging this gap
can generate initial lattice configurations, topology requires close collaboration between AI researchers and
optimization or physics-based simulations can refine these industry practitioners to align model outputs with real-
designs to ensure they meet mechanical and manufacturing world manufacturing capabilities. Establishing validation
constraints. Developing hybrid frameworks where Gen-AI protocols that combine simulation, experimental
models propose initial designs, and conventional methods testing, and Gen-AI predictions is necessary to ensure
perform validation and refinement, would be a promising reliable outcomes. Moreover, developing standards and
direction. Creating feedback loops where insights from guidelines for the industrial implementation of Gen-AI
physical simulations are used to iteratively improve Gen- in lattice design will ensure safety, quality, and regulatory
AI-generated designs can enhance robustness. In addition, compliance, facilitating broader industrial adoption.
establishing multiobjective optimization models that In summary, while Gen-AI has demonstrated
balance trade-offs between mechanical performance, significant potential in revolutionizing lattice structure
process feasibility, and material usage would contribute to design for AM, overcoming its current limitations requires
more effective design strategies. 65,66 a multifaceted research approach. Future studies should
The sustainability of Gen-AI-driven lattice design prioritize enhancing data quality, model interpretability,
also warrants consideration. While AM offers significant adaptability, and sustainability, when also ensuring robust
advantages in reducing material waste, Gen-AI-driven integration with traditional design techniques and process-
processes can be computationally intensive and energy- aware considerations. By addressing these challenges,
consuming, especially when training large models or Gen-AI can be more effectively positioned as a cornerstone
running high-fidelity simulations. Promoting sustainable technology for next-generation, high-performance, and
AI-driven design approaches requires optimizing sustainable lattice structures in advanced manufacturing.
computational workflows to reduce energy consumption, Acknowledgments
for instance, by using more efficient neural network
architectures or leveraging cloud-based green computing None.

Volume 1 Issue 1 (2025) 10 doi: 10.36922/ESAM025110006