Page 38 - ESAM-1-1

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

A B

Figure 1. Schematic diagrams of generative artificial intelligence (Gen-AI) for lattice design. (A) Architectures of typical Gen-AI models used for lattice
structure design; (B) typical lattice structures. Reproduced with permission from Bogusz et al. and Gongora et al. 22
21
with high surface-area-to-volume ratios, optimizing heat LPBF) and conducting mechanical tests (compression,
transfer in components like heat exchangers and thermal tensile, and fatigue) alongside microstructural analysis
shields. 44-46 Across these diverse applications, Gen-AI (scanning electron microscopy, X-ray diffraction, and
enables the exploration of vast design spaces to meet microcomputed tomography). Although it provides high-
complex performance targets, pushing the boundaries of fidelity data, this method is labor-intensive and costly. To
what is achievable with lattice structures. overcome these limitations, synthetic data augmentation
The success of Gen-AI models in lattice design relies methods, such as geometric transformations, noise
heavily on the quality and diversity of training datasets, addition, and latent space sampling in models like VAEs,
which are generally generated through simulation, are employed to expand datasets. However, synthetic data
experimental, and synthetic approaches. Simulation- may require post-processing validation to ensure physical
driven data generation commonly involves parametric feasibility. Together, these strategies aim to enhance dataset
modeling of basic lattice geometries (e.g., octet, diamond, robustness, enabling Gen-AI models to generate diverse
gyroid, and triply periodic minimal surface [TPMS]) and reliable lattice designs.
using computer-aided design (CAD) software, with As discussed above, Gen-AI is emerging as a
variables like cell size and strut diameter systematically transformative tool in the field of lattice structure design,
varied. These models undergo high-throughput finite particularly within the context of AM. 47,48 Gen-AI
element method (FEM) simulations to compute properties offers the potential to revolutionize how complex
such as stiffness, strength, and energy absorption, lattice geometries are conceptualized, optimized, and
though this process can be computationally intensive. manufactured, enabling more efficient, innovative,
Experimental data collection complements simulations and performance-driven designs. Despite its growing
24
by fabricating lattice samples using AM techniques (e.g., relevance, there is a notable scarcity of comprehensive

Volume 1 Issue 1 (2025) 3 doi: 10.36922/ESAM025110006