Abstract Scope |
Although amorphous solids are among the most versatile, diverse, and widely used materials in the world, they are still poorly understood from a physical standpoint. The disordered nature of their structures and the limited understanding of their relationship to material properties have constrained efforts in understanding and designing functional materials with tailored properties. Recently, emerging data-driven approaches provide an unique pathway to address this issue. Here, by encoding atomic local structure into images, we reveal the hidden order from the disorder of amorphous solids in computer simulations. In this work, we utilize convolutional neural networks to model glasses with different stability and study their structural evolution. Our model successfully classifies the different yielding behaviors of amorphous solids. Moreover, we find a hidden structural crossover when the amorphous solids evolve from ductile to brittle, where the transition point is around the glass transition temperature. |