Machine Learning Assisted Design of High Thermal Conductivity and High Strength Mg Alloys

Traditional trial-and-error experimental approaches are insufficient to address the trade-off between strength and thermal conductivity in magnesium (Mg) alloys. This study utilized a data-driven machine learning method to efficiently design and develop Mg–Mn–RE series alloys, achieving a superior synergy of thermal conductivity and strength. Various machine learning algorithms were employed to construct models that predict the relationship between Mg alloy composition, process type, and thermal conductivity/ultimate tensile strength. By analyzing the predicted performance of each model, the most accurate model, CatBoost, was selected and further optimized. Furthermore, by combining the Particle Swarm Optimization algorithm with the optimized CatBoost model, various compositions of the alloy with high thermal conductivity and strength were designed. An optimized alloy composition with Mg–1.6Mn–1.5Ce–1La (wt pct) was selected for experimental verification and mechanism exploration. The alloy exhibited excellent comprehensive thermal and mechanical properties, with a thermal conductivity of 139.2 W/(m K) and an ultimate tensile strength of 398.6 MPa. The total error between the experimental and predicted values was only 7.2 pct, confirming the accuracy of the machine learning model. The high strength was primarily attributed to the fine dynamic recrystallized grains and the density of nanoscale precipitates, while the high thermal conductivity resulted from the reduced concentration of solute atoms as extensive α-Mn phases precipitated from the α-Mg matrix.