Loading...

A Machine Learning-Based Atomistic-Continuum Multi-Scale Modeling of Perfect and Defective Ni-Based Superalloy in Elastoplastic Regions

Kianezhad Tajanaki, Mohammad | 2021

665 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54747 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Khoei, Amir Reza
  7. Abstract:
  8. In this paper, a machine learning-based atomistic-continuum multi-scale scheme is introduced to model the materials' geometrically and materially nonlinear behavior. The kinematic and energetic consistency principles are employed to link the atomistic and continuum scales. In order to establish the kinematic consistency principle, the periodic boundary condition is implemented for the atomistic RVE. The Ni-based superalloy, including 0 to 3% porosity, is considered for the models. Several parameter analysis is done to distinguish the proper atomistic RVE to be used in multi-scale models. The data set, including the stress-strain samples, is generated through molecular dynamics analysis considering a number of load cases. The data set is divided into several groups by the K-means algorithm regarding the strain to enhance the regression accuracy. Subsequently, a feedforward neural network is trained for each group. Due to the high capabilities of Bayesian regularization in optimization, it is used to optimize the values of unknown parameters in neural networks to minimize the error. The material properties of coarse-scale are modeled in the nonlinear finite element framework, in which the stress tensor and tangent modulus are obtained from trained neural networks. Finally, several multi-scale examples are solved to exhibit the capability of the proposed machine learning-based multi-scale technique. Verified multi-scale results with MD outcomes are capable of capturing the sensitive phenomena, including rupture and stress concentration.
  9. Keywords:
  10. Machine Learning ; Neural Network ; Elastoplastic ; Multiscale Modeling ; Nickel-Base Superalloy ; Nonlinear Finite Element Analysis ; Atomistic Representative Elementary Volume (REV) ; Elastoplastic Modeling

 Digital Object List

 Bookmark

...see more