Fractures impact the geological systems remarkably. So, their effects are included within the mathematical simulation models. Literature simulation schemes either lose the facture characteristics or are effort and timeconsuming. So HybridFracture schemes are developed to overcome these drawbacks.
In the dissertation, a generalized HybridEmbedded Fractures (HEF) scheme is developed. It establishes a hierarchical classification based on fracture length's relation to uniform gridcell lengths. Tall and mediumlength fractures are detected from images, classified using MachineLearning (ML) techniques, segmented based on the ML clusters intersections, and used to construct objectoriented based datastructures to simplify dealing with the fracture characteristics. A DeepLearning (DL) design is set up for future work that is supposed to extract the fracture attributes from images directly.
Several fracture characteristics are utilized to generate physically more accurate REF matrixfracture flux exchange parameters that validate better when compared to DiscreteFractureNetwork (DFN) scheme outcomes for similar conditions.
A generalized REF scheme that splits each fractured gridcell into subgrid matrices according to the fractures cutting them is designed. It extends the HEF concepts to nodalgrid based simulators and utilizes tree datastructure settings. The shortlength fractures, from the hierarchical classification used, refer to the fractures and pores varying in scale from nanometers to micrometers. They form another smaller multiscale system and contribute significantly to the multiscale physical processes. Fractional order derivatives are used to deal with this scale, utilizing their additional degree of freedom, which is the order of the fractional order derivatives.
A hybrid cellcentered Physics Preserving Averaging (PPA) scheme is introduced to discretize the fractional order derivatives. Each fractional order derivative is expanded to leftside and rightside derivatives. The PPA scheme discretizes one of them using the original GrunwaldLetnikov (GL) formula, and the other using a shifted GL formula. The original part preserves more physical properties, and the shifted part maintains the stability of the system. PPA scheme also creates symmetrical coefficient matrices that help significantly when converting to higher dimensions, or applying to MultiPoint Flux Approximation configurations, or Multiple Interacting Continua configurations. Additional future work expansions are discussed using porescale network analysis and inverse solution methods.
Date of Award  May 2019 

Original language  English (US) 

Awarding Institution   Physical Science and Engineering


Supervisor  Shuyu Sun (Supervisor) 
