A machine-learning aided multiscale homogenization model for crystal plasticity: application for face-centered cubic single crystals

Artificial Neural Networks for Predicting Mechanical Properties of Crystalline Polyamide12 via Molecular Dynamics Simulations

“…In order to adopt the model into the continuum mechanics framework, we need to transform the data set into the tensor quantities utilized in finite deformation. To preserve material objectivity [5,30], we chose the energetic conjugates right Cauchy-Green tensor C and the second Piola-Kirchhoff (PK2) stress tensor S to represent the deformation state and material response, respectively. Each of these second-order tensors have nine components, but making use of the symmetry and Voigt notation we represent them in the vector form as below.…”

“…However, the cost of multiscale modeling may vary, especially when it is involved with molecular-scale simulations. Traditionally, in multiscale modeling of crystalline materials, this is overcome by using a technique called the Cauchy-Born rule [4,5], which is an approximation of molecular dynamics by using simplified molecular statics or by using the phase field crystal method for polycrystalline metals [6]. However, for semicrystalline or amorphous materials, due to the lack of a definite crystal structure, fullscale molecular dynamics must be utilized, which will significantly increase the cost of the multiscale simulation.…”

“…One common approach is to use features based on the material composition to obtain structure and electronic properties, such as predicting the band gap of semiconductor materials using convolutional neural networks [7] and predicting adsorption energies of high-entropy alloys through deep neural networks [8]. In a recent study, standard feedforward neural networks are used with EAM-based MD simulations to model crystal plasticity in a multiscale framework [5].…”