Fluid flow in rock fractures, as one of the most important hydrogeological phenomena, plays a determining role in subsurface engineering applications, including geothermal energy systems, groundwater resource management, carbon storage, in-situ mining, an‎d prediction of induced earthquakes. Mining an‎d tectonic activities alter stress equilibrium an‎d fracture geometry, directly impacting hydraulic behavior, flow distribution, an‎d preferential fluid pathways. The geometric complexity of natural fractures—including surface roughness, variable apertures, contact points, an‎d multiple intersections—imposes serious limitations on simplified models such as the cubic law an‎d Darcyʹs law. This study aims to achieve a deeper understan‎ding of fluid flow dynamics in rough an‎d intersecting rock fractures by employing a comprehensive combination of physical, numerical, an‎d machine learning approaches. In this study, a novel method for producing artificial fracture physical samples was first developed, based on the integration of 3D printing technology, photogrammetry, an‎d concrete casting. This method enables the production of repeatable samples with precise control over geometric parameters. eva‎luation of the quality of produced samples using the Joint Roughness Coefficient index an‎d advanced statistical parameters demonstrated a suitable match between artificial an‎d real samples. Numerical modeling was performed based on the precise solution of the Navier-Stokes equations using a developed code. Comprehensive validation of numerical models with laboratory data for geometries with varying degrees of roughness showed high agreement. The design of laboratory equipment ensured boundary conditions identical to those in numerical simulations. Extensive parametric studies on single fractures revealed that increasing the fracture scale reduces the effect of roughness on outlet flow rate. The transition of the flow regime from linear Darcy to nonlinear Forchheimer was identified using the critical Reynolds number, which is highly dependent on roughness, dimensions, an‎d fracture aperture. Results indicated that the equivalent hydraulic aperture is always less than the physical aperture, an‎d relative permeability decreases with increasing Reynolds number. For the first time, a new sigmoidal relationship was developed to describe relative permeability as a function of Reynolds number, which accurately combines linear an‎d nonlinear behaviors. Using design of experiments an‎d machine learning algorithms, high-accuracy predictive models for flow rate an‎d hydraulic aperture were developed. Analysis of intersecting fractures showed that geometric parameters such as intersection angle, intersection location, fracture length, an‎d surface roughness have significant effects on flow distribution an‎d permeability. Larger intersection angles led to reduced mixing an‎d permeability, while increasing fracture length facilitated flow. Studies on multiple intersecting fractures, by comparing regular, irregular, an‎d scaffold geometries, demonstrated that increasing network complexity has a significant impact on velocity patterns, fluctuations, an‎d vorticity. In regular geometries, increasing the number of fractures led to reduced fluctuations an‎d greater flow stability, whereas in irregular geometries, geometric complexities remained dominant. The results of this research, by providing new empirical relationships, machine learning models, an‎d better understan‎ding of physical mechanisms, offer practical tools for the design an‎d optimization of engineering systems related to fractured rock environments.

عليرضا باغبانان

امين ازهري

محمود بهنيا , احمدرضا پيشه وراصفهاني , حسين جلالي فر