Enhancing mechanical properties of Al-Li alloys via multi-objective active learning optimization

Designing Al-Li alloy compositions to simultaneously optimize strength, ductility, and stiffness remains a formidable challenge due to their conflicting nature. This study presents a Multi-Objective Active Learning Optimization (MALO) framework to efficiently explore and optimize the high-dimensional composition space of Al-Li alloys. A curated dataset containing 88 Al-Li compositions with corresponding tensile strength, elongation, and elastic modulus was expanded to 176 samples via Gaussian noise augmentation. A total of 81 features, including elemental physicochemical properties and binary formation energies, were extracted and input into machine learning models. In each active learning iteration, candidate alloys on the predicted Pareto front were first identified, followed by λ-EI evaluation to select the top three for experimental validation, resulting in nine tested alloys from ~28 million candidates. The best-performing alloy exhibited a specific strength of 266 MPa·cm³/g, an elongation of 5%, and a specific modulus of 30.5 GPa·cm³/g. SHAP analysis identified Cu and Li contents as key drivers of property optimization, and revealed a synergistic effect between the two elements. This framework enables efficient, data-driven multi-property optimization with limited experimental resources, providing an efficient strategy for multi-property optimization within the targeted Al-Li alloy design space.