Physics-informed neural network to retrieve tectonic stresses from satellite and stress orientation information

Many geological applications require the current stress information at some specific location, for instance in the contexts of fault reactivation, nuclear waste disposal, geothermal energy, carbon capture and storage, mining, or underground production operations. When large areas are involved, it can be useful to consider stress orientation from the World Stress Map for calibration purposes, but one still needs to run some geomechanical models to extract complete stress and displacement fields. 

In our recent study published in Scientific Reports, https://doi.org/10.1038/s41598-023-50759-0, we show how to automatically fit an elastic model with a physics-informed deep neural network approach, to match satellite information and stress orientation. This method is particularly interesting as its nearly completely bypasses the need for explicit boundary condition inputs and yields comprehensive distributions of material properties, displacements, and stress tensors.