Scholarlink

<p>Crevice corrosion is a critical degradation mechanism for stainless steels in aggressive service environments, governed by highly nonlinear interactions between material and environmental stressors. Traditional predictive frameworks, relying heavily on empirical formulas or isolated electrochemical parameters, often fail to capture these complex coupling effects, limiting their utility for generalized risk assessment. This study establishes a robust explainable machine learning framework based on 333 curated experimental points to develop a classification model for predicting crevice corrosion. Among five supervised learning algorithms, the random forest classifier exhibited superior performance, achieving a prediction accuracy of 0.96 and exceptional stability. Beyond prediction, SHapley additive explanations and a novel environmental severity index (ESI) are employed to elucidate underlying mechanisms, revealing a critical nonlinear threshold effect. The analysis demonstrates that in non-mild environments (ESI &gt; 0.3), the marginal benefit of the pitting resistance equivalent number for corrosion inhibition diminishes sharply. This finding quantitatively identifies environmental management as more decisive than material upgrades under high-risk conditions. Finally, 3D probability contour maps are generated to provide engineers with a quantitative tool for material selection and failure prevention.</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

Data-Driven Prediction of Crevice Corrosion Failure in Stainless Steel

  • 期刊论文

  • 发表日期：

    2026年7月21日

查看全文
  • Junpeng Liu,
  • Zhongheng Fu,
  • Zhiyu Han,
  • Wei Wang,
  • Bolei Song,
  • Jiahong Wei,
  • Wenkai Cao,
  • Weihua Li

摘要

Crevice corrosion is a critical degradation mechanism for stainless steels in aggressive service environments, governed by highly nonlinear interactions between material and environmental stressors. Traditional predictive frameworks, relying heavily on empirical formulas or isolated electrochemical parameters, often fail to capture these complex coupling effects, limiting their utility for generalized risk assessment. This study establishes a robust explainable machine learning framework based on 333 curated experimental points to develop a classification model for predicting crevice corrosion. Among five supervised learning algorithms, the random forest classifier exhibited superior performance, achieving a prediction accuracy of 0.96 and exceptional stability. Beyond prediction, SHapley additive explanations and a novel environmental severity index (ESI) are employed to elucidate underlying mechanisms, revealing a critical nonlinear threshold effect. The analysis demonstrates that in non-mild environments (ESI > 0.3), the marginal benefit of the pitting resistance equivalent number for corrosion inhibition diminishes sharply. This finding quantitatively identifies environmental management as more decisive than material upgrades under high-risk conditions. Finally, 3D probability contour maps are generated to provide engineers with a quantitative tool for material selection and failure prevention.