Machine learning optimized by sparrow search for co-design of heat treatment process, microstructure, and properties in ultra-high-strength maraging steels

This study deploys an integrated computational materials engineering (ICME) workflow to elucidate how aging schedules govern the microstructure and mechanical properties of a novel ultra-high-strength maraging steel. Thermodynamic calculations identify precipitate species and contents, while kinetic modeling captures nucleation and growth; in parallel, a machine-learning model tuned by the Sparrow Search Algorithm (SSA) assesses model accuracy, examines the normality of residuals, and delivers global predictions of mechanical properties across the design space. A vacuum-arc-melted martensitic steel was aged between 380 and 680 °C. Strength increases and then decreases with aging temperature, whereas ductility shows the opposite trend. The 480 °C condition provides the optimal overall balance, with a yield strength of 2160 MPa, ultimate tensile strength of 2220 MPa, elongation of 3.85 %, reduction of area of 33.7 %, and hardness of 62.2 HRC. In contrast, the 680 °C condition yields maximum plasticity, characterized by an elongation of 15.5 % and a reduction of area of 62.2 %. EBSD reveals an increase in characteristic grain size with temperature and a gradual decrease in the fraction of low-angle boundaries; XRD and SEM corroborate the formation of reverted austenite at higher temperatures. The SSA-assisted model achieves absolute prediction errors within ±10 % across all tested conditions, supporting its reliability. Integrating thermodynamics, kinetics, and data-driven prediction yields a coherent process–structure–property map and offers practical guidance for heat-treatment design in new maraging steels.