HIGHLY OPTIMIZED & COMPUTATIONALLY EFFICIENT COHESIVE ELEMENTS FOR MICROSTRUCTURAL MODELLING OF CFRCs

UoM administered thesis: Master of Philosophy

  • Authors:
  • Athanasios Tsamos

Abstract

The predictive capability of simulation for advanced materials is limited only by the available computational power. More specifically, composites exhibit a very convoluted response under loading due to the complexity and certain randomness that can be found in their microstructures. Despite the fact that there are available tools, both analytical and numerical, that can be employed in order to simplify their analysis macroscopically through various homogenization procedures, these are rendered inadequate in predicting damage evolution and eventually fracture under cyclic loading or after an impact. This project focuses on finite element modelling using cohesive elements. However, the use of cohesive elements increases the size of problem that needs to be solved. Therefore, they are used moderately and sporadically in areas where fracture is expected to occur. Heading towards the so called “Virtual Testing of Composite Materials”, one can possibly introduce these special elements everywhere in a mesh and therefore perform the modelling of the damaging mechanisms on a microstructural level with however a colossal increase of the problem size. This work contributes towards this goal by giving the right directions and in-depth the necessary theoretical background in the field to aid other researchers. That is, a comprehensive literature review on the microstructural properties and damage mechanisms present in fibrous composite materials, the traditional modelling techniques along with a literature review on modelling damage specifically with cohesive elements followed by notable case studies on the latter subject. Moreover, a brief literature review on high performance computing is included as well since this project aims towards massive simulations with cohesive elements (Virtual Testing of Composite Materials). Furthermore, a thorough nonetheless easy to ‘read-though’ guide on the formulation of two decohesion elements, suitable for modelling carbon fibre composites is given as well. However, the main and most significant contribution of this work, is the actual generation/programming of two highly computationally optimized decohesion elements with a well established cohesive zone model suitable for modelling advanced composite materials along with the verification procedures with state-of-the-art test problems. Although the programmed cohesive elements do not incorporate any novel cohesive zone model (CZM), they manage to perform the simulations faster and with better scalability in comparison with the industry’s standard simulation software’s in-built decohesion elements, due to the fact that they have been programmed with very sophisticated techniques in order to minimize their computational impact (Reduced Floating Point Operations) on an element level and exhibit improved scalability (Reduced data transfer: Memory Input/Output Operations) on overpopulated computing nodes (more efficient exploitation of available computational resources). As it is shown, that is because of the introduction of an analytical integration scheme alongside with mathematical and symbolical reduction techniques instead of a ‘do’ loop based numerical integration, typically employed by FE software packages. More specifically, the programmed 8-node (brick) cohesive element is on average x1.7 faster and the programmed 6-node (wedge) cohesive element is on average x1.9 faster in comparison with the equivalent in-built decohesion elements of the industry’s standard FE software. Additionally, it is shown that the programmed decohesion elements out-scale the software’s in-built ones on a scalability performance test on an overpopulated computing node (many cores sharing the same memory resources - workstation). Despite the fact that for linear problems the setup time is only a small fraction of the overall simulation time as most of the time is spent in the solver, for non-linear problems where on every increment the stiffness matrix must be updated, this is a beneficial improvement which is the main contribution from this work. Thus, these highly optimized cohesive elements can be proven an attractive tool for any engineer looking forward to perform very large composite material simulations either within the Abaqus environment or after certain modifications, in another more sophisticated and more computationally efficient software/solver. Furthermore, the optimised core (element kinematics) of the programmed codes can be used in conjunction with, and benefit, possibly other CZMs suitable of modelling other materials, simply by just replacing the material response within the generated codes if needed. The systematic application of advanced composite materials in the sensitive aerospace sector (transportation) is probably only inhibited by the lack of serious computational models for the latter. Obviously, experimental modelling and testing is far more expensive, and thus prohibitive, in comparison to computational modelling and testing. Therefore, more sophisticated engineering codes will unlock the possibility of such enormous simulations. It is a fact that the use of these more efficient materials in contrast to the conventional heavy materials that are currently being used will enact a safer, inexpensive air travel and ultimately will have a beneficial impact on the environment as well.

Details

Original languageEnglish
Awarding Institution
Supervisors/Advisors
Award date1 Aug 2018