A tight conglomerate is mechanically heterogeneous because gravel particles, the matrix, and their contacts can perturb hydraulic-fracture paths. This study presents an exploratory comparison of three 2.0 m × 2.0 m × 1.0 m reconstituted conglomerate specimens, identified as specimen 97 (Condition A), specimen 98 (Condition B), and specimen 99 (Condition C). One meter-scale specimen was tested under each formulation condition using the same applied boundary stresses, nominal pumping rate, and fluid viscosity. Post-fracturing surface observations, inlet-pressure histories, and laboratory acoustic-emission (AE) event locations were examined. After event-quality control, DBSCAN-based spatial filtering was performed in Z-score-standardized coordinates, and three-dimensional alpha shapes were constructed from the retained event coordinates in meters to calculate geometric AE-event-envelope volumes. The external surface-fracture traces were more spatially distributed in specimen 97 and more localized in specimen 99. The available DBSCAN–alpha-shape calculation outputs yielded geometric AE-event-envelope volumes of 0.2832, 0.2460, and 0.0937 m3 for specimens 97, 98, and 99, respectively. Because only one specimen was tested under each condition, the specimen formulations and gravel characteristics were not identical, the gravel–matrix interfacial properties were not measured directly, and the pumping histories differed, the observations were interpreted as descriptive case differences rather than as a statistically validated single-factor relationship.
Rough-walled fractures in conglomerate reservoirs promote near-wellbore proppant deposition, nonuniform flow, and insufficient distal support, making proppant-schedule screening difficult using small-scale smooth-slot tests alone. This study develops a benchmark-constrained and cost-aware hierarchical screening workflow by integrating a 20 m rough-wall physical experiment, transient Fluent simulations, and archived short-time EDEM sensitivity records. The benchmark experiment used a 20 m × 4.5 m × 10 mm artificial rough-wall fracture and ten operating conditions involving pumping rate, fluid viscosity, proppant size, and sand concentration. In the Fluent model, wall roughness was treated as a regularized roughness representation, and the carrier fluids were modeled using Newtonian constant viscosities measured from laboratory calibration. The experimental effective propped area ranged from 25.5% to 65.1%. Within single-factor comparison subsets, medium viscosity improved support continuity, pumping-rate gains became limited near 0.20 m3/min, particle size affected the balance between distal coverage and bed stability, and 300 kg/m3 sand concentration caused blockage. Image-segmentation-based comparison showed that Fluent captured the main wedge-shaped deposition morphology and screening-level geometric trends. The archived EDEM records indicated that grid-resolution refinement and mixed particle-size representation substantially increased computational cost. A Case 10 mesh-sensitivity check further confirmed that mesh refinement did not alter the first-order deposition morphology. The proposed workflow uses Fluent for whole-domain rapid screening and reserves EDEM/CFD-DEM for targeted short-time sensitivity checks.