Scholarlink

<p>This research introduces a new multi-modal generative AI framework for accelerating and optimising the design of next-generation intelligent materials, which has traditionally been a time-consuming and costly discovery process. The framework includes an AI-based text-to-structure translator, a variational autoencoder (VAE)–based molecular structure generator, and a generative adversarial network (GAN)–based macroscopic property predictor. Integration of physics-informed constraints and interpretability ensures the physical feasibility of the designed molecules and provides insights about structure–property correlations. The high performance of the proposed AI design framework is validated with case studies on shape-memory polymers and self-healing materials, which are two crucial functionalities for soft robots and biomedical devices. The AI-designed materials are found to outperform the state-of-the-art on fundamental properties: 11.8% increase in glass transition temperature, 15.4% increase in tensile strength, 10.2% enhancement in shape recovery ratio, 8.2% improvement in self-healing efficiency, and a healing time decreased by 41.8%, which were validated through molecular dynamics simulation. The knowledge graph revealed and predicted novel correlations between molecular substructures and macroscopic properties through domain-specific competition rules. This work is the first large-scale application of AI to materials science, which significantly reduces the time and cost of discovering new intelligent materials and paves the way to rapidly innovate aerospace, biomedical engineering, and soft robotics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Generative AI designs the next generation of smart materials from pixels to products

  • 期刊论文

  • 发表日期：

    2025年1月14日

查看全文
  • Muhammad Shafiq,
  • Kalpana Thakre,
  • Raji Pandurangan,
  • Regella Venkata Satya Lalitha

摘要

This research introduces a new multi-modal generative AI framework for accelerating and optimising the design of next-generation intelligent materials, which has traditionally been a time-consuming and costly discovery process. The framework includes an AI-based text-to-structure translator, a variational autoencoder (VAE)–based molecular structure generator, and a generative adversarial network (GAN)–based macroscopic property predictor. Integration of physics-informed constraints and interpretability ensures the physical feasibility of the designed molecules and provides insights about structure–property correlations. The high performance of the proposed AI design framework is validated with case studies on shape-memory polymers and self-healing materials, which are two crucial functionalities for soft robots and biomedical devices. The AI-designed materials are found to outperform the state-of-the-art on fundamental properties: 11.8% increase in glass transition temperature, 15.4% increase in tensile strength, 10.2% enhancement in shape recovery ratio, 8.2% improvement in self-healing efficiency, and a healing time decreased by 41.8%, which were validated through molecular dynamics simulation. The knowledge graph revealed and predicted novel correlations between molecular substructures and macroscopic properties through domain-specific competition rules. This work is the first large-scale application of AI to materials science, which significantly reduces the time and cost of discovering new intelligent materials and paves the way to rapidly innovate aerospace, biomedical engineering, and soft robotics.