Geometric and Material Nonlinearity

We have helped various research groups around the world adapt ParaFEM to solve large strain plasticity problems using a parallel “implicit” strategy. Regrettably, until now, that capability has not yet made it back into the open source distribution.

We are pleased to report that Pankaj Pankaj and Francesc Levrero at the University of Edinburgh have written a new set of subroutines and a ParaFEM driver program that can solve both geometrical and material nonlinear problems. The stress-strain laws are provided through Abaqus UMATs (user subroutines) allowing users to adapt the program for their own materials via an industry standard “plugin”.

Using 1024 cores on the Hartree Centre’s BlueGene/Q, a test problem with 1 million degrees of freedom takes 600 seconds to solve. This is the total run time for one hundred load increments, where the material is undergoing large strains and plastic deformation. Using a commercial finite element package, and a new workstation equipped with an Intel quad-core processor, the same problem takes 122,000 seconds. A journal paper which describes the implementation is being prepared. The new source code will be released when the paper has been accepted for publication.

In practical terms, this new capability will mean that ParaFEM can serve a broader range of scientific domains, particularly biological, medical and materials science. In these application areas, more realism is needed to accurately reproduce emergent properties such as flow, damage, failure, collapse and yield.

We would like to thank the Hartree Centre for access to their facilities, which were used to develop and test the new software.