A fully coupled thermal-hydraulic-mechanical (THM) finite element model is presented for fractured geothermal reservoirs. Fractures are modelled as surface discontinuities within a three-dimensional matrix. Non-isothermal flow through the rock matrix and fractures are defined and coupled to a mechanical deformation model. A robust contact model is utilised to resolve the contact tractions between opposing fracture surfaces under THM loadings. A numerical model has been developed using the standard Galerkin method. Quadratic tetrahedral and triangular elements are used for spatial discretisation. The model has been validated against several analytical solutions, and applied to study the effects of the deformable fractures on the injection of cold water in fractured geothermal systems. Results show that the creation of flow channelling due to the thermal volumetric contraction of the rock matrix is very likely. The fluid exchanges heat with the rock matrix, which results in cooling down of the matrix, and subsequent volumetric deformation. The cooling down of the rock matrix around a fracture reduces the contact stress on the fracture surfaces, and increases the fracture aperture. Stress redistribution reduces the aperture, as the area with lower contact stress on the fracture expands. Stress redistribution reduces the likelihood of fracture propagation under pure opening mode, while the expansion of the area with lower contact stress may increase the likelihood of shear fracturing

This paper presents the potential of applying physics-informed neural networks for solving nonlinear multiphysics problems, which are essential to many fields such as biomedical engineering, earthquake prediction, and underground energy harvesting. Specifically, we investigate how to extend the methodology of physics-informed neural networks to solve both the forward and inverse problems in relation to the nonlinear diffusivity and Biot’s equations. We explore the accuracy of the physics-informed neural networks with different training example sizes and choices of hyperparameters. The impacts of the stochastic variations between various training realizations are also investigated. In the inverse case, we also study the effects of noisy measurements. Furthermore, we address the challenge of selecting the hyperparameters of the inverse model and illustrate how this challenge is linked to the hyperparameters selection performed for the forward one.

A fracture network arrangement is quantified across an isolated carbonate platform from outcrop and aerial imagery to address its impact on fluid flow. The network is described in terms of fracture density, orientation, and length distribution parameters. Of particular interest is the role of fracture cross connections and abutments on the effective permeability. Hence, the flow simulations explicitly account for network topology by adopting Discrete‐Fracture‐and‐Matrix description. The interior of the Latemar carbonate platform (Dolomites, Italy) is taken as outcrop analogue for subsurface reservoirs of isolated carbonate build‐ups that exhibit a fracture‐dominated permeability. New is our dual strategy to describe the fracture network both as deterministic‐ and stochastic‐based inputs for flow simulations. The fracture geometries are captured explicitly and form a multiscale data set by integration of interpretations from outcrops, airborne imagery, and lidar. The deterministic network descriptions form the basis for descriptive rules that are diagnostic of the complex natural fracture arrangement. The fracture networks exhibit a variable degree of multitier hierarchies with smaller‐sized fractures abutting against larger fractures under both right and oblique angles. The influence of network topology on connectivity is quantified using Discrete‐Fracture‐Single phase fluid flow simulations. The simulation results show that the effective permeability for the fracture and matrix ensemble can be 50 to 400 times higher than the matrix permeability of 1.0 · 10⁻¹⁴ m². The permeability enhancement is strongly controlled by the connectivity of the fracture network. Therefore, the degree of intersecting and abutting fractures should be captured from outcrops with accuracy to be of value as analogue.