Scholarlink

<p>The results of dynamic simulation play an important role in guiding the analysis and design of multibody systems. However, with the increase of degree of freedom and nonlinearity in flexible multibody systems, traditional analysis methods have poor computational efficiency when conducting dynamic analysis, which poses a challenge to real-time dynamic analysis. Therefore, this paper proposes a fast dynamic analysis method for tensegrity flexible multibody systems via a machine learning framework. The proposed method utilizes the efficient characteristics of neural network and the stability advantage of generalized-α scheme to build neural network training framework, and embeds the discretized differential algebraic dynamic equations into the loss function to guide the training process of the neural network, thus obtaining high-precision neural network model and achieving fast dynamic analysis process. In addition, two numerical examples are used to test the properties of the proposed method, and the results indicate that compared with traditional implicit integration method and physics-informed neural network, the proposed method performs well in numerical accuracy and computational efficiency, and has the potential to achieve real-time simulation.</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 novel machine learning method for real-time dynamic analysis of tensegrity flexible multibody systems

  • 期刊论文

  • 发表日期：

    2025年4月15日

查看全文
  • Ningning Song,
  • Mingji Wang,
  • Xinwei Wang,
  • Haijun Peng

摘要

The results of dynamic simulation play an important role in guiding the analysis and design of multibody systems. However, with the increase of degree of freedom and nonlinearity in flexible multibody systems, traditional analysis methods have poor computational efficiency when conducting dynamic analysis, which poses a challenge to real-time dynamic analysis. Therefore, this paper proposes a fast dynamic analysis method for tensegrity flexible multibody systems via a machine learning framework. The proposed method utilizes the efficient characteristics of neural network and the stability advantage of generalized-α scheme to build neural network training framework, and embeds the discretized differential algebraic dynamic equations into the loss function to guide the training process of the neural network, thus obtaining high-precision neural network model and achieving fast dynamic analysis process. In addition, two numerical examples are used to test the properties of the proposed method, and the results indicate that compared with traditional implicit integration method and physics-informed neural network, the proposed method performs well in numerical accuracy and computational efficiency, and has the potential to achieve real-time simulation.