数据与计算发展前沿 ›› 2025, Vol. 7 ›› Issue (6): 101-110.

CSTR: 32002.14.jfdc.CN10-1649/TP.2025.06.010

doi: 10.11871/jfdc.issn.2096-742X.2025.06.010

• 技术与应用 • 上一篇    下一篇

基于深度学习的Taylor杆碰撞数值模拟算法

杨沁蒙1(),聂宁明1,2,周纯葆1,2,王彦棡1,2,*()   

  1. 1.中国科学院计算机网络信息中心，北京 100083
    2.中国科学院大学，北京 100049
  • 出版日期:2025-12-20 发布日期:2025-12-17
  • 通讯作者: 王彦棡
  • 作者简介:杨沁蒙，中国科学院计算机网络信息中心，助理研究员，主要研究方向为人工智能模型应用，自然语言处理，跨学科应用等。
    本文主要承担工作为数据处理，实验设计，模型构建及学习训练。
    YANG Qinmeng is an assistant researcher at the Computer Network Information Center, Chinese Academy of Sciences. His research interests include applications of artificial intelligence models, natural language processing, and interdisciplinary applications.
    In this paper, he is responsible for data processing, experimental design, model construction and training.
    E-mail: qmyang@cnic.cn|王彦棡，中国科学院计算机网络信息中心，研究员，主要研究方向为人工智能算法与应用软件。
    本文主要承担工作为思路凝练和论文的指导。
    WANG Yangang is a professor at the Computer Network Information Center, Chinese Academy of Sciences. His research interests include artificial intelligence algorithms and application software.
    In this paper, he is responsible for refining the research ideas and the guidance of the paper.
    E-mail: wangyg@sccas.cn
  • 基金资助:
    高性能计算可扩展多级并行开发及毁伤效应数值仿真大规模算例验证(O2021A14243)

Algorithm for Taylor Bar Collision Data Simulation Based on Deep Learning

YANG Qinmeng1(),NIE Ningming1,2,ZHOU Chunbao1,2,WANG Yangang1,2,*()   

  1. 1. Computer Network Information Center, Chinese Academy of Sciences, Beijing 100083, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
  • Online:2025-12-20 Published:2025-12-17
  • Contact: WANG Yangang

RichHTML

3

PDF

0

摘要：

【目的】在超高速碰撞问题求解过程中，通常面临物质变形大、模拟难度大、计算成本高等问题。近年，随着深度学习的发展，融合物理知识神经网络方法的出现，为超高速碰撞问题的高效求解带来了新思路。【方法】本文设计并实现了针对超高速碰撞过程精确模拟算法，利用神经网络学习算子间复杂的非线性映射关系与变化趋势，融合物理先验知识，学习预测与真实算子间的差距，设计损失函数，实现算法对各个算子的精确模拟。【结果】通过Taylor杆碰撞算例进行算法测试，测试结果显示对算子拟合的误差率低于20%。【局限】超高速碰撞问题还有更多的案例，通过学习更多的相关案例可以提高算法的应用范围。【结论】本文实现了基于深度学习神经网络并结合物理知识的Taylor杆碰撞模拟算法，获得良好预测效果。

关键词: 深度学习, PINN, Taylor杆碰撞, 算子, 数据驱动, 科研范式

Abstract:

[Objective] In the process of solving ultra-high-speed collision problems, challenges such as significant material deformations and high computational costs are commonly encountered. In recent years, with the development of deep learning and the emergence of neural networks that incorporate physics knowledge, new ideas have been brought for the efficient solution of such problems. [Methods] This paper designs and implements an accurate simulation algorithm for ultra-high-speed collision problems, neural networks are used to learn the complex nonlinear mapping relationships between operators, thoroughly understanding the trends in changes among these operators. Based on this, physical prior knowledge is integrated, and the algorithm learns to predict and quantify the discrepancy between the predicted and real operators. A loss function is designed to achieve accurate simulation of each operator. [Results] Focusing on the Taylor bar collision case in ultra-high-speed collision problems, the error rate of operator fitting is less than 20%. [Limitations] There are many more cases in ultra-high-speed collision problems. By learning from a broader range of related cases, the applicability of the algorithm can be enhanced. [Conclusions] This paper designs a Taylor bar collision simulation algorithm based on deep learning neural networks combined with physical knowledge, and achieves good predictive performance.

Key words: deep learning, PINN, Taylor bar collision, operator, data-driven, scientific paradigm