Roughness analysis of Model I Fractures in HDR Geothermal Reservoirs using Convolutional Neural Networks

In enhanced geothermal systems (EGS), the stimulation and operation of hot dry rock (HDR) reservoirs often induce numerous Model I (tensile) fractures within the rock mass. The geometric characteristics of these fractures significantly influence both the heat exchange efficiency and mechanical stability of the reservoir. The Joint Roughness Coefficient (JRC) serves as a critical index for evaluating the shear resistance of rock discontinuities. This study investigates the influence of thermal treatment on the JRC of Model I fractures in Lu-Hui granite. Samples were subjected to a series of thermal treatments ranging from 25°C to 1000°C (25℃, 100℃, 200℃, …, 1000℃). This was followed by Brazilian splitting tests and high-resolution profilometry of the resulting fracture surfaces. A one-dimensional convolutional neural network (1D-CNN) model was developed to predict JRC values directly from two-dimensional profile coordinate inputs. Compared to conventional statistical methods based on the root mean square (RMS) height (Z ₂ ), the CNN-based approach demonstrated a high level of accuracy (88% ± 1.2%) and robust feature extraction capabilities. The results indicate that JRC values increase monotonically with temperature, exhibit minimal anisotropy between orthogonal directions, and that higher surface roughness enhances the heat exchange efficiency of fracture networks. These findings provide theoretical insights for the design and stability assessment of engineered geothermal reservoirs.