Scalable computational modelling of concrete ageing and degradation

The typical lifespan of concrete structures ranges from tens to hundreds of years. During such a long period of time many external factors including weather conditions, loading history or environmental pollution. play a crucial role in concrete health and serviceability state. Prediction (via the means of computer simulation) of the long-term material properties of concrete can thus provide valuable insights and lead to better reusability of construction components.

Several very complex multi-physics models were developed in the past decades for this purpose. While these models usually include a wide range of phenomena, the numerical problem which has to be solved poses major challenges and significantly increases required computational time. This makes a predictive simulation of any larger-scale structure non-feasible. On the other hand, commercial codes (ABAQUS, ANSYS, etc.) either lack the material models for a more accurate creep prediction or provide custom material routines which are not computationally optimised. In addition, a specific model and discretisation approach often requires a very specific choice of solvers and preconditioners in order to achieve good parallel scaling properties, so much required for execution on modern HPC infrastructures.

In this thesis a 3-D material model for a reinforced concrete based on the micro-prestress solidification theory (MPS) of Bažant, continuum damage mechanics and the temperature and humidity model of Kunzel is efficiently implemented in the finite-element software FEniCS. A high-performance code for the assembly of residual and tangent operators is automatically derived using automatic differentiation capabilities (AD) of FEniCS. Seamless parallel integration with the linear algebra solvers suite PETSc then offers a wide range of solvers.

The combination of AD, code generation techniques (e.g. FEniCS), and parallel performance of PETSc solvers for predictive modelling of concrete degradation is not present in the existing literature. It is believed that the results presented here allow the study of reusability and degradation of concrete components also for larger structures, where the conventional existing approaches cannot provide a reasonable computation time.