Non-history-based DEM model for predictions of numerical earthquakes

Stick-slip phenomena roughly describe the behavior of a tectonic fault. A simplified model of stick-slip events is often assumed in laboratory experiments and numerical simulations of laboratory earthquakes. This work proposes a more advanced approach. The Discrete Element Method (DEM) was used to generate a numerical model for simulating the laboratory earthquakes in which the granular layer was taken into account. The proposed model takes into account an irregular, random pattern of stress increase and decrease in such a system. At 5,000 selected, regularly spaced time points, the so-called “checkpoints”, 25 parameters were measured, describing the average state of all particles forming the numerical fault at a given moment. The created dataset was used to train the Random Forest algorithm, and then, as part of the tests, this algorithm was used to predict subsequent stick-slip events. The algorithm made predictions solely on the basis of information about the current parameters of the particles. Importantly, the predictions made did not use the history of previous stick-slip events. Feature Importance and SHapley Additive exPlanations (SHAP) were used to assess the contribution of individual particle physical parameters to the prediction results.