EPSS Colloquium fall-2024

Generalized constitutive models are crucial for understanding various solid Earth problems ranging from volcano-magmatic systems, tectonics, subduction zones, and landslides. Despite the development of various theoretical models with damage, laboratory validation remains a significant challenge due to the difficulty in representing the damage state evolution. In this talk, I will present a new framework to address this challenge by incorporating energy conservation laws from Thermodynamics of Irreversible Processes (TIP) into a TIP-informed-Neural Network architecture along with seismic/acoustic waveform features (TIP-INN). The talk will focus on two primary set of results : (a) ) Unsupervised feature selection in acoustic waveforms --Can features in the micro EQs as well as active source coda waveform determine the stress-strain quantitatively? Using acoustic data for Carrara marble spanning the brittle-ductile transition and dimensionality reduction methods, such as UMAP (Uniform Manifold Projection), we show that the acoustics have distinct diagnostic features related to the stress-strain-damage state of the system; (B) Theoretical framework for TIP-INN - I will describe the theoretical framework underlying the constitutive modeling and also present results for a new general TIP-based forward rheology solver that we have developed for thermo-visco-elastic material models (with multiple damage internal parameters). Finally, I will present some present preliminary results towards development of the full TIP-INN model framework including out new Neural network architecture to implement TIP-INN.