Scholarlink

<p>The ductile-to-brittle transition temperature shift (Δ<i>T</i><sub>41J</sub>) and the change of the upper shelf energy represent important parameters for assessing neutron irradiation embrittlement. <i>T</i><sub>41J</sub> and USE are determined from a transition curve of absorbed Charpy V-Notch energy (CVE) obtained from Charpy test results. In this study, we constructed a CVE prediction model using the extreme gradient boost (XGBoost) machine learning (ML) method for non-irradiated materials in the Reactor Embrittlement Archive Project database. The XGBoost model demonstrated the ability to predict CVE with an accuracy of the root mean square error (RMSE) and the coefficient of determination (<i>R</i><sup>2</sup>): RMSE = 19.5&#xa0;J and <i>R</i><sup>2</sup> = 0.92 for the training data and RMSE = 23.9&#xa0;J and<i> R</i><sup>2</sup> = 0.87 for the test data. The RMSE of our model is comparable to other ML models that predict CVE for other materials, which is around 20&#xa0;J. In addition, we interpreted the CVE prediction model using the Shapley additive explanations method (SHAP), which estimates the SHAP value that quantifies the contribution of explanatory variables to the predictions of the constructed model. The SHAP values revealed that the CVE prediction model was developed based on factors, such as test temperature, product form information, chemical content, such as S, Cu, and V-notch orientation (for plate and forging).</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

Charpy V-notch Energy Prediction for Un-irradiated Reactor Pressure Vessel Steel Using the Extreme Gradient Boost Method

  • 期刊论文

  • 发表日期：

    2025年6月5日

查看全文
  • Yoshinori Hashimoto,
  • Akiyoshi Nomoto,
  • Mark Kirk

摘要

The ductile-to-brittle transition temperature shift (ΔT41J) and the change of the upper shelf energy represent important parameters for assessing neutron irradiation embrittlement. T41J and USE are determined from a transition curve of absorbed Charpy V-Notch energy (CVE) obtained from Charpy test results. In this study, we constructed a CVE prediction model using the extreme gradient boost (XGBoost) machine learning (ML) method for non-irradiated materials in the Reactor Embrittlement Archive Project database. The XGBoost model demonstrated the ability to predict CVE with an accuracy of the root mean square error (RMSE) and the coefficient of determination (R2): RMSE = 19.5 J and R2 = 0.92 for the training data and RMSE = 23.9 J and R2 = 0.87 for the test data. The RMSE of our model is comparable to other ML models that predict CVE for other materials, which is around 20 J. In addition, we interpreted the CVE prediction model using the Shapley additive explanations method (SHAP), which estimates the SHAP value that quantifies the contribution of explanatory variables to the predictions of the constructed model. The SHAP values revealed that the CVE prediction model was developed based on factors, such as test temperature, product form information, chemical content, such as S, Cu, and V-notch orientation (for plate and forging).