Machine learning methods for predicting tensile strength, elongation, and low-temperature impact energy of annealed ductile iron

Ductile iron is widely utilized in industries such as wind power and high-speed rail due to its superior mechanical properties, machinability, damping capacity and cost-effectiveness. Machine learning has the potential to reveal intricate relationships within data, enabling more precise predictions and designs of material properties. In this study, a dataset was developed that includes the composition of ductile iron, annealing process parameters, and corresponding performance metrics. Six machine learning algorithms were applied, utilizing the contents of C, Si, Mn, Mg, Re, and Ni, along with annealing temperature and time as inputs. The output parameters were tensile strength, elongation, and impact energy at −40 °C. The models were trained using 5-fold cross-validation, and the best-performing models for each property were identified, with correlation coefficients exceeding 0.87 and average relative errors of 2.17%, 10.53%, and 13.66%. The generalization ability of the models was validated by comparing predictions with external data, which showed relative errors within 10%. These data-driven models provide an efficient design tool for developing novel ductile irons with combined high strength, enhanced plasticity, and superior low-temperature toughness, significantly reducing experimental costs while accelerating industrial deployment in critical sectors including wind power and high-speed rail.