Cohesive elements have extensively been employed inside a FEM code in order to predict damage initiation and progression in composite materials. Nevertheless, the use of cohesive elements increases the size of problem that needs to be solved. Thus, they have been used sparingly in areas where fracture is expected to occur. As the computational power increases and with the introduction of the new generation of supercomputers, it is essential to possess tools capable of exploiting this potential.

Commercial packages such as ABAQUS do not scale well on parallel supercomputers. The implementation steps of a Cohesive Zone Model (CZM) into the open source FE package: ParaFEM (http://parafem.org.uk) are thoroughly presented. These steps are: (1) a study on the CZM existing in ABAQUS and its key parameters, (2) the generation of a Fortran UEL of that particular CZM, (3) the integration of the latter into the source code of ParaFEM as a Cohesive Interface (zero thickness) Element Subroutine, (4) verification of its accurate functionality with the aid of well established sample tests, (5) and finally the culmination of scalability tests and direct comparison between the two programs. ParaFEM uses MPI for message passing and has been proven to scale on up to 64,000 cores for problems with >1 billion unknowns.

The purpose of this research is to improve the particular CZM, rendering it into a computational model that will be apt to capture damage in a more realistic and physical manner. More specifically, the simulation of damage initiation and propagation in Carbon/Fibre reinforced composites emphasizing on mixed-mode conditions on a humongous scale. Furthermore, a superlative scalability is expected for ParaFEM in comparison with ABAQUS. Simulation of real and artificially generated microstructures on a very large scale has not yet been attempted. Thus, three different interfaces will be modelled (matrix/fiber, matrix/matrix & fiber/fiber) with cohesive interface elements. The mesh geometry along with the diverse interfaces parameters are of special interest and therefore crucial for the delivery of meaningful results from the simulations.