Predicting the Thermal Conductivity of Structural Materials Under Lead-Bismuth Corrosion Based on Machine Learning

Area of Science:

• Materials Science
• Nuclear Engineering
• Corrosion Science

Background:

• 316L stainless steel and T91 steel are crucial for lead-cooled fast reactors (LFRs).
• Lead-bismuth eutectic (LBE) corrosion degrades their thermal conductivity, impacting reactor safety and efficiency.
• Limited experimental data and predictive tools exist for LBE-corroded steel thermal conductivity.

Purpose of the Study:

• To experimentally determine thermal conductivity of steels after LBE corrosion.
• To develop machine learning models for predicting thermal conductivity under coupled LBE corrosion conditions.
• To provide a data-driven tool for material selection and safety assessment in LFRs.

Main Methods:

• Experimental acquisition of thermal conductivity data from corroded steels.
• Development of three machine learning models using material composition and corrosion parameters.
• Optimization of a Gradient Boosting Regression model for predictive performance.

Main Results:

• Machine learning models were established to predict thermal conductivity.
• The optimized Gradient Boosting Regression model demonstrated competitive predictive accuracy.
• Low overall prediction error was achieved by the best-performing model.

Conclusions:

• A preliminary data-driven tool was created for estimating thermal conductivity of corroded 316L stainless steel and T91 steel.
• The tool supports material selection, thermal design, and safety assessment for LFRs.
• Further specimen-level validation is needed for broader engineering application.