Séminaire LaMCoS (présentiel - INSA) : Prediction of numerically modeled earthquakes using supervised machine learning

 Piotr Klejment
Institute of Geophysics Polish Academy of Sciences

Résumé au format pdf

The stick slip cycle roughly describes the processes in a tectonic fault leading
to an earthquake and the granular fault gouge plays a fundamental role in
determining the frictional strength of the fault Numerical tools (like Discrete Element Method DEM) can provide useful information about the processes that govern fault slip and the machine learning can help analyze the huge datasets coming from the simulation Here, the DEM model was used to verify the influence of the micro parameters (independent variables) on the macro parameter (the dependent variable - global friction coefficient). Among the eight tested algorithms, Random Forest turned out to be the most effective with a very good base metric R² at the level of 93 %. In the
next step, the supervised machine learning was applied to show that features of the particles in a simulated sheared granular fault contain information regarding the global friction and the time of the next slip event at any point of the simulation.