Enhancing Mechanical Properties and Design of Metallic Glasses Through Nanoscale Heterogeneities and Machine Learning Optimization

Metallic glasses (MGs) are a type of metal alloy that has a disordered atomic structure similar to glass. They possess unique properties such as high strength, elasticity, and corrosion resistance. However, their susceptibility to catastrophic failure through shear banding has limited their widespread use. Recently, the local ordering in MGs has been identified within the amorphous structure, which can influence the physical and mechanical properties of these materials. This dissertation investigates the effect of local ordering and nanoscale heterogeneities on shear band behaviors and optimizes MG designs accordingly. Firstly, a mesoscale shear transformation zone (STZ) dynamic model is employed to investigate the deformation behaviors of MGs. The presence of nanoscale heterogeneity results in a Hall-Petch-like relationship between yield stress and spatial correlation length of heterogeneity. Secondly, dynamic mechanical responses of MGs are studied via experiments and simulations on thin film MGs with various nanoscale heterogeneities. The strain rate sensitivity transition is attributed to a shift in deformation mechanisms from structure-dictated strain localization to stress-dictated strain percolation into a shear band. Finally, a data-driven design framework is developed using artificial neural networks (ANN) and a genetic algorithm (GA) to optimize dual-phase MG designs. The ANN models are trained using simulation data from the STZ dynamic model, and the GA is used to guide the development of new dual-phase MGs with improved mechanical properties. Additionally, the data-driven design framework is extended to optimize material performance under both mechanical and electrochemical processes, minimizing material degradation under the simultaneous action of wear and corrosion based on hierarchical ANN models trained on multiphysics simulations. This dissertation integrates machine learning with multiscale and multiphysics models to uncover the mechanisms that explain the emergence of mechanical behaviors of MGs and explores a massive design space for property optimization. These findings provide valuable insights into the material structure-property relationships, enabling informed decision-making for material design.