Deep Learning in Mechanical Metamaterials: From Prediction and Generation to Inverse Design

Advanced Materials, Journal Year: 2023, Volume and Issue: 35(45)

Published: June 19, 2023

Abstract Mechanical metamaterials are meticulously designed structures with exceptional mechanical properties determined by their microstructures and constituent materials. Tailoring material geometric distribution unlocks the potential to achieve unprecedented bulk functions. However, current metamaterial design considerably relies on experienced designers' inspiration through trial error, while investigating responses entails time‐consuming testing or computationally expensive simulations. Nevertheless, recent advancements in deep learning have revolutionized process of metamaterials, enabling property prediction geometry generation without prior knowledge. Furthermore, generative models can transform conventional forward into inverse design. Many studies implementation highly specialized, pros cons may not be immediately evident. This critical review provides a comprehensive overview capabilities prediction, generation, metamaterials. Additionally, this highlights leveraging create universally applicable datasets, intelligently intelligence. article is expected valuable only researchers working but also those field materials informatics.

Language: Английский

---

Unifying the design space and optimizing linear and nonlinear truss metamaterials by generative modeling

Nature Communications, Journal Year: 2023, Volume and Issue: 14(1)

Published: Nov. 21, 2023

The rise of machine learning has fueled the discovery new materials and, especially, metamaterials-truss lattices being their most prominent class. While tailorable properties have been explored extensively, design truss-based metamaterials remained highly limited and often heuristic, due to vast, discrete space lack a comprehensive parameterization. We here present graph-based deep generative framework, which combines variational autoencoder property predictor, construct reduced, continuous latent representation covering an enormous range trusses. This unified allows for fast generation designs through simple operations (e.g., traversing or interpolating between structures). further demonstrate optimization framework inverse trusses with customized mechanical in both linear nonlinear regimes, including exhibiting exceptionally stiff, auxetic, pentamode-like, tailored behaviors. model can predict manufacturable (and counter-intuitive) extreme target beyond training domain.

Language: Английский

---

Inverse design of nonlinear mechanical metamaterials via video denoising diffusion models

Nature Machine Intelligence, Journal Year: 2023, Volume and Issue: 5(12), P. 1466 - 1475

Published: Dec. 11, 2023

Language: Английский

---

Machine learning-based inverse design methods considering data characteristics and design space size in materials design and manufacturing: a review

Materials Horizons, Journal Year: 2023, Volume and Issue: 10(12), P. 5436 - 5456

Published: Jan. 1, 2023

This review offers a guideline for selecting the ML-based inverse design method, considering data characteristics and space size. It categorizes challenges underscores proper methods, with focus on composites its manufacturing.

Language: Английский

---

Multi‐Physical Lattice Metamaterials Enabled by Additive Manufacturing: Design Principles, Interaction Mechanisms, and Multifunctional Applications

Advanced Science, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 20, 2025

Abstract Lattice metamaterials emerge as advanced architected materials with superior physical properties and significant potential for lightweight applications. Recent developments in additive manufacturing (AM) techniques facilitate the of lattice intricate microarchitectures promote their applications multi‐physical scenarios. Previous reviews on have largely focused a specific/single field, limited discussion properties, interaction mechanisms, multifunctional Accordingly, this article critically design principles, structure‐mechanism‐property relationships, enabled by AM techniques. First, are categorized into homogeneous lattices, inhomogeneous other forms, whose principles processes discussed, including benefits drawbacks different fabricating types lattices. Subsequently, structure–mechanism–property relationships mechanisms range fields, mechanical, acoustic, electromagnetic/optical, thermal disciplines, summarized to reveal critical principles. Moreover, metamaterials, such sound absorbers, insulators, manipulators, sensors, actuators, soft robots, management, invisible cloaks, biomedical implants, enumerated. These provide effective guidelines

Language: Английский

---

Inverse-designed growth-based cellular metamaterials

Mechanics of Materials, Journal Year: 2023, Volume and Issue: 182, P. 104668 - 104668

Published: May 2, 2023

Advancements in machine learning have sparked significant interest designing mechanical metamaterials, i.e., materials that derive their properties from inherent microstructure rather than just constituent material. We propose a data-driven exploration of the design space growth-based cellular metamaterials based on star-shaped distances. These two-dimensional are periodically-repeating unit cells consisting material and void patterns with non-trivial geometries. Machine models exploiting large datasets then employed to inverse for tailored anisotropic stiffness. Firstly, forward model is created bypass growth homogenization process accurately predict given finite set parameters. Secondly, an used invert structure–property maps enable accurate prediction designs stiffness query. successfully demonstrate frameworks' generalization capabilities by chosen outside domain space.

Language: Английский

---

Perspective: Machine Learning in Design for 3D/4D Printing

Journal of Applied Mechanics, Journal Year: 2023, Volume and Issue: 91(3)

Published: Oct. 5, 2023

Abstract 3D/4D printing offers significant flexibility in manufacturing complex structures with a diverse range of mechanical responses, while also posing critical needs tackling challenging inverse design problems. The rapidly developing machine learning (ML) approach new opportunities and has attracted interest the field. In this perspective paper, we highlight recent advancements utilizing ML for designing printed desired responses. First, provide an overview common forward problems, relevant types structures, space responses printing. Second, review works that have employed variety approaches different ranging from structural properties to active shape changes. Finally, briefly discuss main challenges, summarize existing potential approaches, extend discussion broader problems field This paper is expected foundational guides insights into application design.

Language: Английский

---

AI-enhanced biomedical micro/nanorobots in microfluidics

Lab on a Chip, Journal Year: 2024, Volume and Issue: 24(5), P. 1419 - 1440

Published: Jan. 1, 2024

Although developed independently at the beginning, AI, micro/nanorobots and microfluidics have become more intertwined in past few years which has greatly propelled cutting-edge development fields of biomedical sciences.

Language: Английский

---

Inverse machine learning framework for optimizing gradient honeycomb structure under impact loading

Engineering Structures, Journal Year: 2024, Volume and Issue: 309, P. 118079 - 118079

Published: April 27, 2024

Language: Английский

---

Accelerating the design of lattice structures using machine learning

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: June 14, 2024

Abstract Lattices remain an attractive class of structures due to their design versatility; however, rapidly designing lattice with tailored or optimal mechanical properties remains a significant challenge. With each added variable, the space quickly becomes intractable. To address this challenge, research efforts have sought combine computational approaches machine learning (ML)-based reduce cost process and accelerate design. While these made substantial progress, challenges in (1) building interpreting ML-based surrogate models (2) iteratively efficiently curating training datasets for optimization tasks. Here, we first challenge by combining modeling Shapley additive explanation (SHAP) analysis interpret impact variable. We find that our achieve excellent prediction capabilities ( R 2 > 0.95) SHAP values aid uncovering variables influencing performance. second utilizing active learning-based methods, such as Bayesian optimization, explore report 5 × reduction simulations relative grid-based search. Collectively, results underscore value intelligent systems leverage methods key accelerating

Language: Английский

---