Artificial intelligence (AI) is a rapidly advancing field with significant potential for analyzing and investigating large datasets. However, most AI methods are purely data-driven, posing challenges in application areas where data is scarce but substantial knowledge of physical principles exists. In such cases, AI methods face two key limitations: they require extensive data for training, and the resulting models often lack interpretability.

One possibility to overcome these limitations is integrating fundamental physical principles (i.e., model-based approaches) into AI methods, creating what is known as hybrid AI. The research group aims to further develop and optimize hybrid AI systems to enhance their efficiency and broaden their applicability in scientifically driven fields. The group mainly uses the non-intrusive reduced basis method, which, unlike conventional AI techniques, produces explainable models while requiring significantly less data, as illustrated for several large-scale geothermal applications.

The project applies AI methods to assess sensitivities, uncertainties, and optimal experimental design. A key demonstration of the economic and societal impact of this research is its application to induced seismicity in geothermal energy. Geothermal energy plays a crucial role in the energy transition, however, many projects are halted due to induced seismicity. Hybrid AI techniques are employed to integrate physics into the monitoring of induced seismicity, enabling better prediction of seismic events by improving the system understanding—especially regarding the non-critical state. Additionally, optimal experimental design methods are used to enhance the placement and effectiveness of monitoring stations.

The developed methodologies are demonstrated using datasets from two geothermal projects: Pohang and Soult-sous-Forêts.