CRSS determination combining ab-initio framework and Surrogate Neural Networks. (March 2023)

• Record Type: Journal Article
• Authors: You, Daegun; Celebi, Orcun Koray; Mohammed, Ahmed Sameer Khan; Abueidda, Diab W.; Koric, Seid; Sehitoglu, Huseyin
• Is Part Of: International journal of plasticity. Volume 162(2023)
• Publication Date: 2023-03
• Subjects: Critical stress -- Dislocations -- Machine learning -- Surrogate Neural Network -- Wigner-Seitz cell; Plasticity -- Periodicals; Plasticité -- Périodiques; Plasticity; Periodicals; 620.11233
• DOI: 10.1016/j.ijplas.2023.103524
• Languages: English
• ISSNs: 0749-6419

Abstract

Highlights : A novel ab-initio framework of the CRSS for an extended dislocation slip is utilized to generate a large dataset including a wide range of hypothetical and real FCC materials. A novel minimum energy path approach to calculate the CRSS is established to derive a triangular trajectory of Shockley partials in a robust intermittent "zig-zag" motion. Learning from a large dataset, the CRSS variations with materials' fingerprints are revealed to be mediated by equilibrium core-widths of Shockley partials for an extended dislocation slip. Surrogate Neural Networks (SNN) model is developed with a large dataset of the CRSS in hypothetical FCC materials for the first time in literature, resulting in high accuracy 94% on real FCC materials as well, including metals and high entropy alloys (HEAs). Core-widths mediated dependencies of the CRSS on the materials' fingerprints such as stacking fault energies, lattice constant, and elastic moduli are distinctly revealed for the first time in literature, which is instantly demonstrated in the precise SNN as well. Abstract: Critical Resolved Shear Stress (CRSS), fundamentally linked to the dislocation glide stress, is a crucial measure in dictating plastic deformation in metallic materials. A recent ab-initio predictive model for dislocation glide stress in Face-Centered Cubic (FCC) materials is developed which accurately predicts available experimental data, considering the anisotropic continuum energy, the atomistic misfit energy, and the minimum energy path for the intermittent motion of Shockley partials. The CRSS of a material is predominantly controlled by six parameters, namely, lattice constant, unstable/stable stacking-fault energies, and three anisotropic elastic constants for cubic materials, which are inputs to the predictive model. In this work, a large material dataset is produced incorporating properties of real materials and generating hypothetical combinations, subsequently calculating the CRSS for each combination using the predictive model. The hypothetical combinations of properties are employed to train a machine learning-based Surrogate Neural Network (SNN), and the ones of real materials are utilized to validate the SNN model yielding a 94% accuracy for 1, 033 materials. The generated dataset is used to unravel the sensitivity of each material parameter to the predicted CRSS establishing a general trend for the FCC materials for the first time guiding the field in achieving superior mechanical properties. Graphical abstract: Image, graphical abstract