Page 37 - ESAM-1-1

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

high-performance applications ranging from aerospace design strategy available. Conventional approaches such as
components to biomedical implants. 4,8,9 AM’s freedom of topology optimization, parametric modeling, and physics-
design has pushed lattice design to a new level, allowing based simulations have long been employed to generate
engineers to lightweight parts (minimize material and and refine lattice geometries. 31,32 These methods, while
weight) when maintaining or even improving mechanical effective, often require significant computational resources
performance. 10-12 Such lattices provide high surface area and manual intervention to achieve optimal results. What
and tunable stiffness for improved biological integration. distinguishes Gen-AI from traditional computational
The materials used for AM-built lattice structures design methods is its scalability, adaptability, and capability
encompass both metallic and polymer-based materials. 13,14 to uncover non-intuitive solutions within vast design
Metallic lattices, often fabricated from materials like spaces. Traditional methods often rely on iterative manual
titanium, stainless steel, or aluminum alloys, are widely adjustments or rule-based algorithms, which can be time-
used in aerospace, biomedical implants, and automotive consuming and limited in scope. In contrast, Gen-AI
components due to their high strength, durability, and rapidly evaluates numerous design permutations, identifies
thermal resistance. 15-19 On the other hand, polymer-based optimal solutions, and adapts to complex, multiobjective
lattices, crafted from materials such as nylon, polylactic constraints. This ability accelerates the design process and
acid (PLA), or thermoplastic polyurethane (TPU), are fosters innovation by unveiling lattice configurations that
advantageous for applications requiring lightweight, may not be immediately apparent through conventional
20
flexible, or biocompatible structures. The choice of approaches. In essence, AM offers the capability to fabricate
material significantly influences the design strategy, highly complex lattice structures, while Gen-AI serves as
performance outcomes, and process considerations in the intelligent engine that designs these structures for peak
lattice fabrication. Therefore, understanding material- performance. The synergy between AM and Gen-AI not
specific design requirements is essential for optimizing the only streamlines the design-to-production workflow but
performance and manufacturability of lattice structures in also pushes the boundaries of what is achievable in lattice
diverse application scenarios. architecture. 33,34
In recent years, Gen-AI has emerged as a transformative Figure 2 presents a typical example of lattice
approach in automating and enhancing lattice design for structure generation using Gen-AI, demonstrating
AM. 23-25 Gen-AI in design, often referred to as generative the transformation from initial Gaussian noise to final
design in engineering contexts, utilizes sophisticated implicit and mesh representations suitable for AM.
35
machine learning algorithms to autonomously explore an Lattice structures have emerged as a transformative design
extensive array of design possibilities that satisfy predefined strategy in various industries due to their unique ability to
performance objectives. 26,27 Unlike conventional machine balance lightweight characteristics with high mechanical
learning models that primarily focus on prediction or performance. However, optimizing lattice architectures
classification tasks, Gen-AI models actively generate for specific functionalities, such as energy absorption,
new data instances, enabling the creation of novel and biomedical customization, mechanical strength, and
optimized design solutions. 28,29 thermal management, remains a complex challenge. 36,37

Common Gen-AI techniques applicable to lattice In applications requiring energy absorption and
design include generative adversarial networks (GANs), impact resistance, Gen-AI can optimize lattice parameters
variational autoencoders (VAEs), and diffusion models, like cell size, shape, and wall thickness to achieve
as illustrated in Figure 1. Each of these models operates controlled deformation under impact, enhancing safety in
on distinct mechanisms: GANs employ a competitive automotive, aerospace, and protective gear applications. 38,39
framework between a generator and a discriminator to In the medical field, customized lattice orthoses and
create realistic designs; VAEs encode input data into a implants benefit from generative design, where Gen-AI-
latent space and decode it to generate variations; and driven models, combined with patient-specific 3D scans,
diffusion models progressively refine random noise into enabled the fabrication of ergonomic, lightweight orthoses
structured designs. These models empower engineers via AM. 40,41 For lightweight strength and mechanical
to set specific design goals, such as maximizing stiffness optimization, Gen-AI excels in creating lattice structures
when minimizing weight, and allow the algorithms to that minimize weight without compromising strength,
autonomously propose optimal lattice geometries. 30 which is especially valuable in aerospace and automotive
While Gen-AI has emerged as a transformative approach sectors. 42,43 Finally, in heat dissipation and thermal
to lattice structure design, it is not the only computational management, Gen-AI facilitates the design of lattices

Volume 1 Issue 1 (2025) 2 doi: 10.36922/ESAM025110006