Scholarlink

<p>The optimization of functionally graded metal matrix composites (FGMMCs) has long been hindered by the complex interplay between reinforcement distribution, process constraints, and multifunctional performance objectives. Current data-driven and empirical frameworks like ANN-GA hybrids, CNN-DQN models and XGBoost regressors cannot guarantee physical validity, as well as generalizability, and, as a result, provide non-manufacturable or unstable predictions. To overcome these shortcomings, this study presents a Physics-Informed Neural Operator (PINO)-assisted inverse design workflow of the Al6061 hybrid composites that combines Principal Component Analysis (PCA) to reduce dimensionality of the gradient fields and Bayesian Optimization (BO) to optimize the performance of the material-synthesizing system. The model is trained and tested on a high-fidelity dataset based on experimental and simulated material property data such as tensile strength, hardness and fatigue life. The proposed framework, implemented in Python and based on PyTorch and GpyOpt, can explore the reduced coefficient space in an intelligent manner to find the best reinforcement architectures, keeping manufacturing constraints. The results reveal a 3.6% increase in predictive accuracy (R<sup>2</sup> = 0.987) over state-of-the-art approaches, with less than 3% deviation from Finite Element Analysis (FEA) benchmarks. Moreover, optimized designs exhibit a 22% improvement in stiffness-to-weight ratio and a 17% gain in fatigue life compared to reference Al6061–SiC composites. The synergy of physics-informed learning and probabilistic optimization performs a dual function of speeding up design convergence, while still rendering the resulting designs interpretable and feasible from a practical standpoint. The proposed framework represents a new direction for autonomous, data-efficient, high-fidelity material design, thereby setting the stage for new intelligent FGMMC engineering and adaptive manufacturing studies.</p> Graphical Abstract <p>Graphical representation of an AI-based framework for optimizing FGMMC reinforcement gradients and mechanical performance.</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

AI-optimized design of hybrid functionally graded metal matrix composites for additive manufacturing and multifunctional engineering applications

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • L. Natrayan,
  • Kathi Venkataramana,
  • K. Vijetha,
  • Seeniappan Kaliappan,
  • Ramya Maranan,
  • Anand Rajendran

摘要

The optimization of functionally graded metal matrix composites (FGMMCs) has long been hindered by the complex interplay between reinforcement distribution, process constraints, and multifunctional performance objectives. Current data-driven and empirical frameworks like ANN-GA hybrids, CNN-DQN models and XGBoost regressors cannot guarantee physical validity, as well as generalizability, and, as a result, provide non-manufacturable or unstable predictions. To overcome these shortcomings, this study presents a Physics-Informed Neural Operator (PINO)-assisted inverse design workflow of the Al6061 hybrid composites that combines Principal Component Analysis (PCA) to reduce dimensionality of the gradient fields and Bayesian Optimization (BO) to optimize the performance of the material-synthesizing system. The model is trained and tested on a high-fidelity dataset based on experimental and simulated material property data such as tensile strength, hardness and fatigue life. The proposed framework, implemented in Python and based on PyTorch and GpyOpt, can explore the reduced coefficient space in an intelligent manner to find the best reinforcement architectures, keeping manufacturing constraints. The results reveal a 3.6% increase in predictive accuracy (R2 = 0.987) over state-of-the-art approaches, with less than 3% deviation from Finite Element Analysis (FEA) benchmarks. Moreover, optimized designs exhibit a 22% improvement in stiffness-to-weight ratio and a 17% gain in fatigue life compared to reference Al6061–SiC composites. The synergy of physics-informed learning and probabilistic optimization performs a dual function of speeding up design convergence, while still rendering the resulting designs interpretable and feasible from a practical standpoint. The proposed framework represents a new direction for autonomous, data-efficient, high-fidelity material design, thereby setting the stage for new intelligent FGMMC engineering and adaptive manufacturing studies.

Graphical Abstract

Graphical representation of an AI-based framework for optimizing FGMMC reinforcement gradients and mechanical performance.