The Finite Element Method (FEM) is a widely adopted numerical technique in civil and geotechnical engineering for modelling the geometry and mechanical behaviour of structures and subsurface materials subjected to various types of loads. The FEM program numgeo (www.numgeo.de), developed by the Chair of Geotechnics at Bauhaus-Universität Weimar, is an open-access tool used by thousands of engineers and researchers worldwide. One prominent application of numgeo is the modelling of offshore wind turbine foundations. Current FEM-based predictions typically involve the simulation of millions of load cycles, such as those generated by sea waves or the rotation of wind turbine rotors, which are transmitted from the superstructure to the supporting soil. The interaction between soil and structure is therefore accounted for through these simulations, with the largest models requiring several hours of computational time to complete. numgeo is written in modern Fortran, a language still widely used for performance-oriented scientific computing, though no longer commonly taught in standard curricula. The program currently supports shared-memory parallelisation via OpenMP, enabling efficient use of multi-core CPUs for specific computational routines. However, other components of the code still execute serially, presenting opportunities for further optimisation. This project has two main objectives (to be worked on by 2 students). The first is to explore and implement advanced parallelisation strategies to improve computational performance. This includes enhancing existing OpenMP capabilities, integrating GPU acceleration via OpenACC and potentially rewriting performance-critical components using CUDA for direct execution on NVIDIA GPUs. The second objective is to improve the user interface and workflow through the integration of open-source AI tools. The goal is to link numgeo with existing mesh generation tools and develop a dedicated pre-processor that allows intuitive model setup. The AI component should be trained and optimised to enable users to define key features of the intended numerical model (such as geometry, boundary conditions, material properties, and loading scenarios) through natural language input or a guided interface. It should then automatically generate a consistent and solvable finite element model that can be directly processed and computed by numgeo. This not only reduces manual pre-processing effort but also broadens accessibility for less-experienced users. Such integration has the potential to significantly accelerate the modelling workflow, improve reproducibility, and foster the use of advanced simulation techniques in geotechnical practice. |
