Abstract: 

Finite element simulations involving many loading cycles with nonlinear material models can lead to high computational costs. In this work, we propose an approach to speed up plasticity simulations by taking large time increments in multiaxial loading. Specifically, we train a Machine Learning (ML) model to give the incremental solutions of evolution equations for solution-dependent state variables. We incorporate knowledge from the underlying reference plasticity model to guide the ML-based model and reduce the amount of needed training data. The resulting explicit material model can take much larger time steps than existing implicit material models.

The trained ML-based material model has been implemented in Abaqus through a user-defined material subroutine (UMAT) and evaluated in finite element simulations against both a cyclic plasticity model implemented as a UMAT and the built-in Abaqus plasticity model. We have assessed the approach in terms of accuracy and computational efficiency for cyclic loading simulations. The results demonstrate the potential of the ML-based material model to reduce simulation time while maintaining accuracy.

PUT PDF DOCS HERE

Presenters: 

Nasrin Talebi Doctoral Student - Chalmers University of Technology Nasrin Talebi is a doctoral student in the Department of Mechanical Engineering at Chalmers University of Technology. Her project focuses on fatigue crack initiation in anisotropic railway rails. She investigates the influence of long-term accumulation of plastic deformations on the risk of fatigue crack initiation in rails under mixed traffic loading. This is achieved using machine learning-based plasticity modeling, finite element simulations of wheel over-rollings, and fatigue analysis.