Premium standalone sand screens must limit sand production while preserving productivity, yet most erosion- and plugging-centred studies do not quantify how a stabilised filter-bed transfers hydraulic loading into stresses on woven wires. This study quantifies sustained normal and shear loading on a plain-weave standalone screen beneath stabilised sand filter-beds using an immersed boundary computational fluid dynamics-discrete element framework under increasing imposed pressure drop. Screen-surface stresses were evaluated over the loaded screen area for the whole screen and for interior and perimeter reporting zones using area-time-weighted distributions. The results show a clear monotonic strengthening of the stabilised loading state as pressure drop increased. Typical loading rose in both the normal and shear components, with the normal component remaining dominant throughout. The loaded-area fraction increased only modestly, whereas the mean stress over the loaded area increased much more strongly. This indicates that higher pressure drop amplified stress intensity within already engaged regions more than it expanded the area carrying load. The upper tail of the stress distribution also strengthened, which shows that increasing pressure drop intensified not only the typical loading state but also the most severe loading regime. Hotspot maps further showed persistent wire-scale organisation within each stabilised window, together with perimeter-associated amplification in the normal upper tail under the present configuration. These findings provide mechanics-based loading descriptors that can support screen qualification procedures and operating-envelope assessment under stabilised filter-bed loading.

Understanding fluid-particle interactions is critical in petroleum engineering, particularly for controlling drilling fluid performance and mitigating fluid loss. Numerical methods, such as the coupled computational fluid dynamics discrete element method, enable a detailed investigation of these interactions without relying on extensive experimental testing. Traditional particle-sizing guidelines, including empirical bridging rules, provide only partial insights into the pore-scale mechanisms governing filter cake formation and permeability evolution. In contrast, numerical modelling directly resolves how the particle size distribution and solid concentration influence the filter cake structure and flow behavior. This study employed a coupled numerical simulation framework to examine filter cake formation for drilling fluids containing unimodal and bimodal particle size distributions across a range of solid concentrations. The key descriptors analyzed included the filtration rate, filter cake porosity, permeability, and pore size distribution. The results show that bimodal particle mixtures exhibit a concentration-dependent transition in permeability behavior. At lower solid loadings, bimodal systems maintain substantially higher permeability than unimodal systems because of the persistence of large, connected pore pathways formed by coarse particles. As the solid concentration increases, finer particles progressively infiltrate and occlude these pathways, leading to a marked permeability reduction and convergence toward unimodal behavior. Pore-size distribution analysis revealed that permeability is governed primarily by the connectivity and continuity of large pore throats rather than by bulk porosity. These findings demonstrate that bimodal distributions require sufficient fine content to achieve effective fluid loss control, providing pore-scale numerical guidance for optimizing drilling-fluid particle-size selection strategies.

This study introduces MicroGraphNets, a deep learning framework for automating the microscopic characterization of wettability in porous media using graph neural networks. The framework predicts rock surface roughness, fluid/fluid interfacial curvatures, and contact angles at 3-phase contact lines from segmented multiphase micro-computed tomography images. This is achieved by converting these images into sets of surface and interfacial points, with their intersection defining the 3-phase contact line points. Specialized geometrical training graphs are constructed from these points to predict each property, leveraging surface and interfacial normal vectors as input features for constructing surface and interfacial graphs. To address the unique challenge that arises from the coexistence of all phases around 3-phase contact lines, distinct node types assigned to each phase were embedded as node features for constructing contact angle graphs. To predict the properties, the framework employs a message-passing graph neural network with three modules: an encoder for initial feature embeddings, a processor for aggregating neighboring embeddings and propagating messages, and a decoder for final property prediction. This approach effectively captures node and edge relationships, facilitating accurate regression of surface and interfacial properties. Validation includes testing on unseen samples and a synthetic droplet test against analytical solutions. Time-resolved analysis was performed to demonstrate the scalability and efficiency of the framework on large datasets. MicroGraphNets demonstrates superior accuracy and efficiency compared to traditional deep learning methods, showcasing its potential for predicting microscopic surface and interfacial properties of porous media.

This paper presents a hybrid deep learning framework that combines graph neural networks with convolutional neural networks to predict porous media properties. This approach capitalizes on the capabilities of pre-trained convolutional neural networks to extract n-dimensional feature vectors from processed three dimensional micro computed tomography porous media images obtained from seven different sandstone rock samples. Subsequently, two strategies for embedding the computed feature vectors into graphs were explored: extracting a single feature vector per sample (image) and treating each sample as a node in the training graph, and representing each sample as a graph by extracting a fixed number of feature vectors, which form the nodes of each training graph. Various types of graph convolutional layers were examined to evaluate the capabilities and limitations of spectral and spatial approaches. The dataset was divided into 70/20/10 for training, validation, and testing. The models were trained to predict the absolute permeability of porous media. Notably, the proposed architectures further reduce the selected objective loss function to values below 35 mD, with improvements in the coefficient of determination reaching 9%. Moreover, the generalizability of the networks was evaluated by testing their performance on unseen sandstone and carbonate rock samples that were not encountered during training. Finally, a sensitivity analysis is conducted to investigate the influence of various hyperparameters on the performance of the models. The findings highlight the potential of graph neural networks as promising deep learning-based alternatives for characterizing porous media properties. The proposed architectures efficiently predict the permeability, which is more than 500 times faster than that of numerical solvers.