Scholarlink

<p>The study of transient heat transfer in composite materials is essential for engineering, particularly in aerospace and energy applications. Recently, Physics-Informed Neural Network (PINN), an explainable Artificial Intelligence (AI) model, has been applied to computational heat and mechanics. PINNs integrate physical principles into neural networks, eliminating the need for numerical integration, mesh discretization, and labeled training data. However, existing PINN methods primarily address single-domain problems and lack procedures for handling multi-domain problems involving with composite materials. This paper introduces a PINN model with multiple subnets for transient heat transfer in composite materials with multi-phase inclusions. Each subnet predicts the transient temperature field within its respective domain, while the continuity conditions link all subnets across material boundaries. This approach ensures the continuity of temperature and heat flux in the normal direction of material interfaces. This paper also examines the impact of collocation points and sampling methods on prediction accuracy. The multi-subnet PINN (Ms-PINN) model is applied to various inclusion properties, shapes, quantities, and three-dimensional cases, yielding results closely matching finite difference method predictions. This framework can be extended to multi-domain composites for other physical problems governed by partial differential equations.</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

A multi-subnets physics-informed neural network (Ms-PINN) model for transient heat transfer analysis in materials with heterogeneous microstructures

  • 期刊论文

  • 发表日期：

    2025年7月28日

查看全文
  • Chengcheng Shen,
  • Haifeng Zhao,
  • Jian Jiao

摘要

The study of transient heat transfer in composite materials is essential for engineering, particularly in aerospace and energy applications. Recently, Physics-Informed Neural Network (PINN), an explainable Artificial Intelligence (AI) model, has been applied to computational heat and mechanics. PINNs integrate physical principles into neural networks, eliminating the need for numerical integration, mesh discretization, and labeled training data. However, existing PINN methods primarily address single-domain problems and lack procedures for handling multi-domain problems involving with composite materials. This paper introduces a PINN model with multiple subnets for transient heat transfer in composite materials with multi-phase inclusions. Each subnet predicts the transient temperature field within its respective domain, while the continuity conditions link all subnets across material boundaries. This approach ensures the continuity of temperature and heat flux in the normal direction of material interfaces. This paper also examines the impact of collocation points and sampling methods on prediction accuracy. The multi-subnet PINN (Ms-PINN) model is applied to various inclusion properties, shapes, quantities, and three-dimensional cases, yielding results closely matching finite difference method predictions. This framework can be extended to multi-domain composites for other physical problems governed by partial differential equations.