Swelling of clay-sulfate rocks is a major geotechnical hazard, which is encountered in various underground engineering projects such as tunneling or geothermal drilling. It is crucial to understand the underlying hydro-mechanical-chemical (HMC) processes of swelling phenomena in order to plan appropriate countermeasures. The swelling processes can be described by coupled partial differential equations (PDEs). Their solution can be approximated numerically by means of a discretization scheme, e.g. using the finite element method (FEM), the preferred numerical approach in solid mechanics. However, FEM still suffers from some shortcomings, e.g. the robustness of the solution depends strongly on the mesh quality, the discretization method and the order of the polynomials used to approximate the unknown fields.

The physics-informed neural network (PINN) is a meshless method that can approximate solutions to PDEs. Overall, PINN is a new and promising method for solving PDEs governing the coupled behavior of geo-materials and other complex systems. They offer a data-efficient and physics-based approach to model complex phenomena, and have the potential to advance our understanding of the behavior of swelling rocks in geotechnical engineering. The planned research has three main objectives: (1) developing a unified, meshless and robust PINN framework to model the coupled HMC processes in swelling clay-sulfate rocks; (2) advance our ability to model complex coupled processes by the PINN approach in general; (3) advance our present understanding of the coupled HMC processes in swelling of clay-sulfate rocks.

The study sites Staufen and Freudenstein tunnel provide comprehensive data sets including hydraulic, chemical and mechanical data to parameterize the models, as well as heave observations at the land surface and tunnel floor to scrutinize the PINN models.

Further information: https://gepris.dfg.de/gepris/projekt/533825365?language=en