Geofluids & Geomechanics

Rock Rheology/Constitutive models are a critical component of geodynamical models, but are an under-appreciated source of uncertainty in Earth sciences (e.g., magmatic system, plate boundary interfaces, subsurface reservoirs with thermo-hydro-chemo-mechanical interactions). For instance, multiple DOE Energy Earthshots rely on effective control of permeability to optimize mass and heat transfer between fluid and rock within the subsurface (e.g., carbon sequestration, geologic hydrogen stimulation, energy storage, and enhanced geothermal). Central to each of these Earthshot applications is an operational understanding that is sufficient to enable the controlled evolution of fracture networks, as illustrated below for geothermal reservoirs. Additionally, the drilling, sensing, and reservoir stimulation technologies must be adaptable to a broad range of geological settings with diverse mechanical properties that vary with temperature, stress, and damage. 

Given computational limitations, it is not feasible to model the crystal/single fracture scale dynamics for large reservoirs over 10s of years. Thus, generalized constitutive models that can accurately represent the micro-scale processes in a representative volume element (REV) are crucial for advancing drilling, sensing, and reservoir stimulation technologies. Despite developing various theoretical models with damage, laboratory validation remains a significant challenge due to the difficulty in characterizing the complex damage state evolution within the samples at high temporal resolution. Recently, seismo-acoustics and X-ray tomography (and similar other methods) are increasingly being used for 3D sample characterization during deformation. However, there is limited work to date in directly incorporating this information into dynamic constitutive models. ​​

One approach that provides real-time data for these applications is seismo-acoustic monitoring tools. However, these tools still need to be integrated into rock constitutive modeling with quantitative physical linkages between, or interpretations of, subtle acoustic signals and different fracture formation processes. As part of the DOE Science Foundations for Earthshots Program, our new project - MARBLE-C ((Machine leArning Rheology of Brittle-ductiLe Experiments and Crust) proposes a new framework for constitutive modeling of rocks with distributed damage and fracture networks, predicting their response to mechanical, hydraulic, and chemical perturbations with direct constraints coming from acoustic (and other) data. We will incorporate energy conservation laws from Thermodynamics of Irreversible Processes (TIP) into a TIP-informed-Neural Network architecture (TIP-INN) to develop a representative volume element scale rheological model for stress-strain and transport property predictions.​ ​

​

To be applicable to field settings, we will focus on using seismo-acoustic features (both for active source/ambient noise and micro-seismic events) in the TIP-INN model directly as a proxy for the evolution of internal parameters of the fracture network system/material (e.g., total damage, damage distribution). Thus, seismo-acoustic data is not a posteriori check on the model but is directly incorporated into the rheology predictions, including the time evolution of internal variables.

​

 ​