Scholarlink

<p>Ceramic matrix composites in aerospace thermal components endure high-temperature fatigue loading where interfacial degradation is critical. However, existing identification methods relying on single hysteresis loop features often lack reliability, as they fail to fully capture the complex interfacial damage state, leading to inconsistent results. This study proposes a machine learning approach integrating multiple hysteresis loop features. Theoretical sample points covering the experimental data range of multiple hysteresis features were generated based on the shear-lag model and Latin hypercube sampling. An optimized artificial neural network established the mapping between features and interfacial shear stress. By incorporating component parameters derived from tensile curves and experimental hysteresis loop data tested at 1400&#xa0;°C into the data-driven model, the degradation law of the interfacial shear stress was identified. This enabled prediction of hysteresis features and fatigue life, with maximum prediction errors below 10% for all features. The method demonstrates superior accuracy over single-feature approaches, confirming its effectiveness for interfacial degradation assessment in high-temperature applications.</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

​Machine Learning-Based Identification of Interfacial Shear Stress under High-Temperature Fatigue Using Multiple Hysteresis Loop Features​​

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Xiao Han,
  • Zikai Zhou,
  • Zhikang Zheng,
  • Chao You,
  • Xihui Chen,
  • Gao Xiguang,
  • Yingdong Song

摘要

Ceramic matrix composites in aerospace thermal components endure high-temperature fatigue loading where interfacial degradation is critical. However, existing identification methods relying on single hysteresis loop features often lack reliability, as they fail to fully capture the complex interfacial damage state, leading to inconsistent results. This study proposes a machine learning approach integrating multiple hysteresis loop features. Theoretical sample points covering the experimental data range of multiple hysteresis features were generated based on the shear-lag model and Latin hypercube sampling. An optimized artificial neural network established the mapping between features and interfacial shear stress. By incorporating component parameters derived from tensile curves and experimental hysteresis loop data tested at 1400 °C into the data-driven model, the degradation law of the interfacial shear stress was identified. This enabled prediction of hysteresis features and fatigue life, with maximum prediction errors below 10% for all features. The method demonstrates superior accuracy over single-feature approaches, confirming its effectiveness for interfacial degradation assessment in high-temperature applications.