Glass formation is frequently observed in metallic alloys.Machine learning has been applied to discover new metallic glasses.However,the incomplete understanding of glass formation hinders descriptor selection and mat...Glass formation is frequently observed in metallic alloys.Machine learning has been applied to discover new metallic glasses.However,the incomplete understanding of glass formation hinders descriptor selection and material property representation.Here,we use X-ray diffraction spectra,the essential tool for identifying amorphous structure,as an intermediate link.By representing spectra as images,we train generative models to produce high-fidelity spectra for all alloys in multicomponent alloy systems.Training with spectra from a tiny fraction of the total alloys is sufficient for accurate spectra generation,enabling the identification of compositional regions with a high probability of glass formation.The shift from numerical to image-based representation unlocks the potential of machine learning in the design of glass-forming alloys.Furthermore,our approach is applicable to a wide range of materials and spectroscopic techniques.We anticipate that this strategy will accelerate materials discovery across previously unexplored compositional and processing spaces.展开更多
基金supported by the National Natural Science Foundation of China(grant nos.52331007,52192602,T2222028,52471189).The AI-driven experiments,simulations and model trainingwere performed on the robotic AI-Scientist platform of Chinese Academy of Sciences.
文摘Glass formation is frequently observed in metallic alloys.Machine learning has been applied to discover new metallic glasses.However,the incomplete understanding of glass formation hinders descriptor selection and material property representation.Here,we use X-ray diffraction spectra,the essential tool for identifying amorphous structure,as an intermediate link.By representing spectra as images,we train generative models to produce high-fidelity spectra for all alloys in multicomponent alloy systems.Training with spectra from a tiny fraction of the total alloys is sufficient for accurate spectra generation,enabling the identification of compositional regions with a high probability of glass formation.The shift from numerical to image-based representation unlocks the potential of machine learning in the design of glass-forming alloys.Furthermore,our approach is applicable to a wide range of materials and spectroscopic techniques.We anticipate that this strategy will accelerate materials discovery across previously unexplored compositional and processing spaces.
