Scholarlink

<p>To enhance the prediction accuracy of multiaxial fatigue life for critical safety components such as gear shafts in automotive drivetrains, this study proposes a hybrid framework that integrates experiments, simulations, and physics-informed machine learning. Focusing on commonly used steels for automotive gear shafts, a dataset including multiple parameters such as stress level, notch dimension, specimen diameter, and fatigue life is established through finite element simulations, whose validity has been verified by experimental research. Building upon a mechanistic analysis of key factors governing fatigue life, the neural network loss function is physically constrained and modified, thereby establishing a prediction model with improved physical interpretability. Compared with conventional approaches such as random forests, support vector machines, and standard neural networks, the proposed model achieves higher prediction accuracy while markedly enhancing both physical interpretability and extrapolation capability. This work provides a new perspective on multiaxial fatigue life prediction for automotive gear shafts.</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 Hybrid Framework for Multiaxial Fatigue Life Prediction Integrating Experiments, Simulations, and Physics-Informed Machine Learning

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Junlou Li,
  • Yan Ma,
  • Ziqiang Li,
  • Jianjiao Deng,
  • Hongbin Tang,
  • Yuhang Shen,
  • Lvfeng Jiang,
  • Yanan Hu,
  • Qianhua Kan,
  • Xuandong Li

摘要

To enhance the prediction accuracy of multiaxial fatigue life for critical safety components such as gear shafts in automotive drivetrains, this study proposes a hybrid framework that integrates experiments, simulations, and physics-informed machine learning. Focusing on commonly used steels for automotive gear shafts, a dataset including multiple parameters such as stress level, notch dimension, specimen diameter, and fatigue life is established through finite element simulations, whose validity has been verified by experimental research. Building upon a mechanistic analysis of key factors governing fatigue life, the neural network loss function is physically constrained and modified, thereby establishing a prediction model with improved physical interpretability. Compared with conventional approaches such as random forests, support vector machines, and standard neural networks, the proposed model achieves higher prediction accuracy while markedly enhancing both physical interpretability and extrapolation capability. This work provides a new perspective on multiaxial fatigue life prediction for automotive gear shafts.