Scholarlink

<p>Multiaxial fatigue poses a significant threat to structural integrity in aerospace components. This study proposes a data-driven method for multiaxial fatigue life prediction across five engineering materials under six different loading paths. The methodology integrates comprehensive feature engineering with algorithm optimization, establishing tailored prediction models for each material through systematic evaluation of four machine learning algorithms (MLP, SVR, Random Forest, XGBoost). Results demonstrate exceptional prediction performance, with <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> values consistently exceeding 0.95, over 95% of predictions falling within the two-fold error band, and fewer than 30% of data points exceeding 50% relative error across all materials. The analysis reveals distinct material-algorithm affinities, with XGBoost outperforming for most materials while MLP excels in data-limited scenarios. Furthermore, we identify approximately 150 samples as the critical threshold for model performance saturation. This work provides both a validated methodological framework and practical insights for data-driven fatigue prediction in engineering 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

A Data-Driven Method for Multiaxial Fatigue Life Prediction Using Machine Learning

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • Z. R. Wu,
  • H. Lei,
  • C. W. Gui,
  • S. L. Shi,
  • M. L. Chen

摘要

Multiaxial fatigue poses a significant threat to structural integrity in aerospace components. This study proposes a data-driven method for multiaxial fatigue life prediction across five engineering materials under six different loading paths. The methodology integrates comprehensive feature engineering with algorithm optimization, establishing tailored prediction models for each material through systematic evaluation of four machine learning algorithms (MLP, SVR, Random Forest, XGBoost). Results demonstrate exceptional prediction performance, with \(R^{2}\) R 2 values consistently exceeding 0.95, over 95% of predictions falling within the two-fold error band, and fewer than 30% of data points exceeding 50% relative error across all materials. The analysis reveals distinct material-algorithm affinities, with XGBoost outperforming for most materials while MLP excels in data-limited scenarios. Furthermore, we identify approximately 150 samples as the critical threshold for model performance saturation. This work provides both a validated methodological framework and practical insights for data-driven fatigue prediction in engineering applications.