Scholarlink

<p>The continuous cooling diagram (CCT-diagram) is a convenient representation of how austenite decomposes into different microstructures during cooling. It can be applied for designing suitable cooling paths in order to achieve desired mechanical properties. Since the decomposition of austenite depends on the chemical composition of the steel, understanding how different components affect the time and temperature of the transformation provides a valuable tool in choosing a suitable composition. Several linear and non-linear methods have been presented earlier for predicting CCT-diagrams. These include linear regression models, machine-learning methods, and different neural network-based models. Generally, these methods do not provide any uncertainty measures around their predicted curves, large learning data sets may be needed and they may have problems with extrapolation. In this study rigid parametric modeling is combined with a non-parametric Bayesian approach to fit model parameters. This approach yields an interpretable and simple parametric curve function and accounts for complex interrelations present in the alloying elements. Special property of the method is that it produces probabilistic uncertainty estimates for the CCT-curve for a given steel. This reflects representativeness of the composition of that particular steel in the learning set.</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

Bayesian Parametric Curve Prediction for Ferrite Onset Temperature During Cooling of Steel with Multiple Output Gaussian Process Prior

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月26日

查看全文
  • Juho Luukkonen,
  • Aarne Pohjonen,
  • Seppo Louhenkilpi,
  • Jyrki Miettinen,
  • Erkki Laitinen,
  • Mikko J. Sillanpää

摘要

The continuous cooling diagram (CCT-diagram) is a convenient representation of how austenite decomposes into different microstructures during cooling. It can be applied for designing suitable cooling paths in order to achieve desired mechanical properties. Since the decomposition of austenite depends on the chemical composition of the steel, understanding how different components affect the time and temperature of the transformation provides a valuable tool in choosing a suitable composition. Several linear and non-linear methods have been presented earlier for predicting CCT-diagrams. These include linear regression models, machine-learning methods, and different neural network-based models. Generally, these methods do not provide any uncertainty measures around their predicted curves, large learning data sets may be needed and they may have problems with extrapolation. In this study rigid parametric modeling is combined with a non-parametric Bayesian approach to fit model parameters. This approach yields an interpretable and simple parametric curve function and accounts for complex interrelations present in the alloying elements. Special property of the method is that it produces probabilistic uncertainty estimates for the CCT-curve for a given steel. This reflects representativeness of the composition of that particular steel in the learning set.