# Non-linear analysis of concrete and masonry structures

By Jiri Brozovsky
I have been working on modelling of non-linear static behaviour of reinforced concrete and masonry. Developed program code has used to analyse of number of concrete and masonry structures to help our research projects. Everything have done in 2D only (for plane stress problem). Uncracked concrete (and components of masonry) has been modelled as elastoplastic material (using Chen-Chen's plasticity and failure conditions and Ohtani-Chen's hardening rule). Smeared crack model has been used for cracked concrete.Variant of Bazant's crack-band model has been used to minimize of localization in cracked concrete (without this there can be differences between results from different finite element meshes for the some structure). Reinforcement has been modelled as elastoplastic material (with or without hardening). I have used finite element method for numerical analysis (with four-node isoparametric elements). So material (concrete) state changes affects only material stiffness matrix.Reinforcement has been modelled as smeared (e.g. used as an addition to material stiffness matrix). Non-linear solution has been done using Newton-Raphson method and linear equations have been solved using Conjugate Gradient Method. How is it related to parallel computing? It's very slow to be done on usual PC or workstation (requires lot of work when stiffness matrices are created, solution of linear equations requires lot of time, too).It can be used for small structures and simple models, but not for larger ones. There are lot of things that can be easily parallelized (assembling of stiffness matrices or solution of linear equations, for example). The code has been parallelized using MPI and has been ran on Beowulf. Parallelization has done in quite simple way: f.e. nodes are divided between processes and every process creates their own part of global stiffness matrix (and of global load vector). So every process works only with elements related to process's nodes. It is fine as long as non-linear material law is deterministic (and it is). Parallelization of solution of linear equations is straightforward, too. I have been traditionally using Conjugate Gradient Method (with Jacobi preconditioner) so the most complicated operations are the vector-matrix and matrix-vector multiplications. Parallelization of these operations is trivial. Number of structures was analyzed. One of the simplest (and its load-displacement diagram for two different finite element meshes) is shown on figure.