Experimental-data-driven thermal conductivity prediction and inverse composition design for alloys

Area of Science:

• Materials Science
• Computational Materials Science
• Data Science

Background:

• Predicting thermal conductivity in metals and alloys is crucial for material design.
• Existing methods often lack accuracy or efficiency for complex alloy systems.
• A comprehensive dataset and robust predictive models are needed.

Purpose of the Study:

• To develop a data-driven framework for predicting thermal conductivity of metals and alloys.
• To create an inverse design workflow for proposing alloy compositions with target thermal conductivity.
• To establish a reliable and efficient method for materials discovery.

Main Methods:

• Collected the largest experimental dataset of thermal conductivity (6259 points, 49 elements, 0-1400 K).
• Trained and benchmarked regression models using alloy composition and temperature as inputs.
• Developed an inverse-design workflow based on the trained forward model.

Main Results:

• Achieved high predictive accuracy (R² > 0.99, RMSE 6-9 W m⁻¹ K⁻¹).
• Demonstrated quantitative reliability for challenging alloys (Mg alloys, steel) across temperature ranges.
• Successfully proposed candidate alloys meeting specific thermal conductivity targets.

Conclusions:

• The data-driven framework accurately predicts thermal conductivity and enables inverse design of alloys.
• The inverse search identifies practical composition windows for experimental validation.
• This approach accelerates the discovery of materials with tailored thermal properties.