SOLVING FOR ASEISMIC AND SEISMIC SLIP IN SDOF SYSTEM USING PHYSICS INFORMED NEURAL NETWORKS

Abstract

Dynamic systems encompass a broad range of phenomena that describe the evolution of states over time through a mathematical framework where rich interaction of quantities is established. In this dissertation, we explore the dynamic system of earthquake nucleation using a single degree of freedom spring-block slider model. This model exhibits stick-slip behavior, closely mimicking the cyclic occurrence of earthquakes. Our research focuses on understanding the detailed functioning of neural networks in solving a coupled ordinary differential equations (ODE) problem representative of this physical process. This study can be viewed as a part of the broader effort in explainable AI, where we aim to comprehend how neural network architecture choices impact ODE solutions. In addition to benchmarking neural network-driven solutions against conventional simulations of governing ODEs, we have compared the numerical solutions of these equations with outputs from both feedforward neural networks and physics-informed neural networks. This comprehensive comparison enhances our understanding of the efficacy and accuracy of neural network approaches in modelling aseismic and seismic slip.