Scholarlink

<p>Efficient prediction for stress intensity factors (SIFs) of multiple interacting cracks on complex geometries is critical for structural safety. Here, a prediction framework for mixed-mode SIFs of two interacting cracks on an open cylindrical shell was developed based on machine learning approaches, such as gradient boosting regression tree and artificial neural network (ANN). Results show that ANN-based framework with outstanding extrapolation capability is more suitable for predicting non-monotonic relationships between SIFs and crack-pair geometries. Meanwhile, predictions by the ANN-based model effectively suppress artificial fluctuations induced by coarse-mesh finite element method (FEM) simulations, while reducing computation time by nearly four orders of magnitude comparable to direct fine-mesh FEM calculations. Phase diagrams of attractive or repulsive modes in two interacting cracks under different parameters are efficiently obtained by plotting the initial deflection angle. Our results show that the ANN-based machine learning framework can be extended to predict SIFs of complex crack networks in engineering structures.</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

Predictions of stress intensity factors for two interacting cracks on curved surfaces using machine learning approaches

  • 期刊论文

  • 发表日期：

    2026年5月21日

查看全文
  • Zhengyu Yuan,
  • Kaijin Wu,
  • Mengqi Liu,
  • Zhaoqiang Song,
  • Linghui He,
  • Yong Ni

摘要

Efficient prediction for stress intensity factors (SIFs) of multiple interacting cracks on complex geometries is critical for structural safety. Here, a prediction framework for mixed-mode SIFs of two interacting cracks on an open cylindrical shell was developed based on machine learning approaches, such as gradient boosting regression tree and artificial neural network (ANN). Results show that ANN-based framework with outstanding extrapolation capability is more suitable for predicting non-monotonic relationships between SIFs and crack-pair geometries. Meanwhile, predictions by the ANN-based model effectively suppress artificial fluctuations induced by coarse-mesh finite element method (FEM) simulations, while reducing computation time by nearly four orders of magnitude comparable to direct fine-mesh FEM calculations. Phase diagrams of attractive or repulsive modes in two interacting cracks under different parameters are efficiently obtained by plotting the initial deflection angle. Our results show that the ANN-based machine learning framework can be extended to predict SIFs of complex crack networks in engineering structures.