Application of machine learning techniques to predict rupture propagation and arrest in 2-D dynamic earthquake simulations

Abstract

SUMMARY Simulating dynamic earthquake rupture propagation is challenging due to uncertainties in the underlying physics of fault slip, stress conditions and the fault frictional properties. A trial and error approach is often used to determine the unknown parameters describing rupture, though running many simulations usually requires human review to determine how to adjust parameter values and is thus inefficient. We leverage machine learning approaches to reduce computational cost and improve the ability to determine reasonable stress and friction parameters. Two models have been developed using neural networks and the random forest to predict if a rupture can break 2-D geometrically complex fault. We train the models using a database of 1600 dynamic rupture simulations computed numerically. Fault geometry, stress conditions and friction parameters vary in each simulation. Both models distinguish the underlying complex data patterns that reflect the physics of the rupture. For example, our models identify in a quantitative fashion, how higher normal and shear stress components and low static and dynamic friction can be tied to the probability of rupture propagation. Both models are efficient in predicting rupture propagation such that 400 unseen examples are predicted in a fraction of a second, leading to potential applications of dynamic rupture that have previously not been possible due to the computational demands of physics-based rupture simulations.