Fiber-reinforcement is possibly seen as an effective solution to improve the performance of geopolymers. This new study provides an extensive experimental work on a typical carbon fiber-reinforced metakaolin-based geopolymer composites. Both mechanical and transport properties are investigated in relation with multi-scale analysis of microstructure. The focus is to analyze the effects of fiber reinforcement and confining pressure. Based on workability and mechanical performance, an optimal carbon-fiber (CF) volume fraction of 0.5% is first identified, resulting in an increase of 44.30%, 29.56%, and 52.60% respectively in flexural, compressive, and tensile strength. Microstructural observations, combined with porosity and permeability analyses, demonstrate that the CF network refines the pore structure and suppresses crack initiation and propagation. Transport property under hydrostatic compression is further evaluated, revealing stress path-dependent permeability evolution. X-ray micro-computed tomography and scanning electron microscopy analyses allow for the observation of progressive failure processes in triaxial compression tests, underscoring coupled effects of confining pressure and fiber reinforcement. Due to the interaction between CF and confining pressure, there is a transition of failure model from brittle splitting to ductile shearing. Mohr-Coulomb strength envelope fitting yields a cohesion of 12.85 MPa and internal friction angle of 49.23 degrees, exceeding typical values for most conventional cementitious and geopolymer materials. These findings provide valuable experimental background for design and analysis of engineering structures with fiber-reinforced geopolymers.
The excavation mechanism of Tunnel Boring Machine (TBM) cutter discs in rock tunnelling is highly complex. Because an extremely high contact pressure develops at the disc-rock interface, it triggers fractures in the rock and leads to chip formation. This study develops two predictive models that relate specific energy (SE) consumption in rock cutting to both the spacing-to-penetration (s/p) ratio and uniaxial compressive strength (UCS). The first model extends the Colorado School of Mines (CSM) framework by replacing its single contact-pressure distribution with alternative profiles, thereby revealing how the pressure profile governs the orientation and magnitude of disc-induced forces across different rock types. The second model is fully numerical. It uses contact-pressure profiles recorded in the laboratory with specialized testing machines (LCM and ILCM) for several rock types at reference penetrations, then generalizes these profiles across the full penetration range and computes the resulting cutting forces. The results are presented as characteristic curves that relate the SE to the s/p ratio and the UCS. The study shows that asymmetric and narrower profiles contribute to increased prediction uncertainty. The optimal s/p ratio is roughly 13-14 for low-strength rocks, drops to just under 10 for medium-strength rocks, and rises slightly above 10 for very high-strength rocks. A broader experimental pressure distribution profile better captures variations in contact pressure, enabling more stable and accurate predictions. Based on the performed analyses, knowledge of the contact pressure distribution is essential for identifying the optimal s/p ratio, which is associated with the minimum Excavation Specific Energy. This will thereby enable the proper design of the TBM cutter head and the effective management of the machine during tunnel advancement.
Rock cuttability is a paramount factor influencing the efficiency, cost, and safety of tunneling, drilling, and mining operations. However, accurately predicting cutting performance remains challenging due to the inherent microstructure heterogeneity of rocks, which is often oversimplified in conventional models. This study employs a grain-based rock cutting model (GB-RCM) combined with moment tensor analysis, which enables quantitative elucidation of the seismic source mechanisms during fragmentation and investigates the effects of microstructural heterogeneity and initial stress on the rock-cutting process for the first time. The model, calibrated against laboratory data for granite, explicitly represents its mineral composition and grain-size distribution, with a dimensionless heterogeneity index H systematizing the microstructural variation. Simulations reveal that heterogeneity dominates the cutting response, with the peak cutting force decreasing by over 50% as H increases, indicating significantly reduced rock resistance. Microcrack analysis shows that higher heterogeneity promotes granulation, while initial stress localizes damage. Crucially, the moment tensor calculation of simulated acoustic emission events provides profound physical insights into the fracture process. It confirms that the energy release is predominantly governed by shear mechanisms, and successfully classifies microcrack types (tensile, shear, implosive), linking them to microstructural features and stress conditions. The b-value is found to increase with heterogeneity, reflecting a shift towards small-scale fracturing. Furthermore, an improved specific energy (SEI) formula, which incorporates the efficiency of intergranular cracking, is proposed. The results demonstrate that SEI increases with initial stress and exhibits a non-monotonic relationship with heterogeneity. The integration of grain-based modeling with advanced seismic source quantification offers a powerful framework for understanding rock-tool interaction and optimizing excavation parameters.
As a natural material, rock is inherently fractured with discontinuities exhibiting pronounced spatial variability in geometry and high nonlinearity in mechanical property-stress relationships. Each fracture involves multiple interdependent parameters that collectively form a high-dimensional, heterogenous feature space, making their identification and characterization a fundamental challenge in rock mechanics. Conventional Kalman-based ensemble smoother (ES(K)), based on linear covariance updates, is inadequate for capturing the complex influence of fracture networks on the overall mechanical behavior of rock masses. To address this challenge, this study develops a deep learning (DL)-based ensemble smoother (ES(DL)), in which a feature-weighting block is introduced to explicitly account for the heterogeneity of fracture-related parameters during the update process. By enabling adaptive reweighting of high-dimensional fracture features within the DL-based update operator, the proposed method enhances the representation of strongly nonlinear parameter-response relationships in rock fracture systems. Numerical experiments indicate that the proposed approach provides more accurate posterior estimation of high-dimensional and strongly heterogenous fracture parameters than the conventional ES(K) method, accompanied by a more effective reduction in parameter uncertainty. The feature-weighting enhanced ES(DL) is capable of reproducing the full stress-strain response of fractured rock and exhibits distinct advantages in the peak strength regime, where fracture-induced process nonlinearity is most pronounced, the method is examined through an engineering application involving rock specimens from the F115 fault damage zone, in which the proposed ES(DL) consistently captures fracture-governed mechanical behavior and associated parameter variability, providing clear evidence of its applicability and robustness for practical fractured rock masses under complex geological conditions.
Carbonate formations are attractive for CO2 geological storage owing to their global abundance and favorable petrophysical properties. However, the reactivity of carbonate rocks under CO2-enriched conditions induces dissolution-driven pore structure alterations, potentially compromising long-term storage integrity. While previous investigations have predominantly examined dissolution impacts on absolute permeability, critical knowledge gaps persist regarding concomitant changes in two-phase flow characteristics (relative permeability) and elastic mechanical behavior-essential parameters for evaluating reservoir performance and geomechanical stability. To address these limitations, we present an integrated numerical framework that combines lattice Boltzmann method (LBM) with finite element method (FEM). This approach enables systematic quantification of dissolution-induced modifications in pore geometry, CO2-water relative permeability relationships, and elastic moduli evolution using identical digital rock samples. This represents a critical advancement that overcomes experimental constraints in obtaining paired pre-/post-reaction measurements (due to the destructive nature of rock mechanics tests). Our findings reveal four distinct dissolution regimes (uniform, compact, wormhole, and hybrid patterns) governed by the interplay of Peclet (Pe) and Damko & uml;hler (Da) numbers. Reactive alteration induces dramatic relative permeability shifts, particularly in wormhole-dominated systems: the CO2 relative permeability shows a 14.2-fold enhancement, whereas the water relative permeability is reduced to 15% of its original value. These nonlinear responses necessitate revision of conventional Corey model parameters (exponents ng and nw) for predictive accuracy. Concurrently, dissolution reduces both bulk and shear moduli, with shear modulus demonstrating greater sensitivity to dissolution patterns than bulk modulus. Power-law correlations between elastic moduli and porosity emerge, invalidating traditional linear approximations. Notably, homogeneous dissolution (i.e., uniform pattern) preserves higher shear modulus retention and more stable Poisson's ratio compared to heterogeneous dissolution modes (i.e., compact and wormhole patterns). This analysis provides critical insights for optimizing CO2 injection strategies and developing predictive models integrating hydrodynamical-geomechanical coupling in carbonate reservoirs.
In situ stress estimates obtained using the borehole overcoring method are subject to significant uncertainty arising from rock mass heterogeneity and testing procedures. However, comprehensive methods for evaluating the reliability of overcoring results remain limited, posing potential risks to engineering decision-making. To address this issue, this study proposes a stress estimation error metric and establishes a corresponding quality classification system for overcoring tests. The proposed metric is numerically examined through Monte Carlo simulations of in situ stress measurements, and the effects of key influencing factors are systematically analyzed. The results show that the stress estimation error follows a scaled chi distribution and increases with the level of strain measurement error. Its variability is governed primarily by the number of strain gauges, whereas the effects of rock elastic properties, hollow inclusion parameters, borehole size, and strain gauge orientation are comparatively minor within the investigated ranges. Based on these findings, a quantitative framework for evaluating the quality of overcoring test results is developed. This study provides a unified scalar index for evaluating the uncertainty of three-dimensional overcoring stress estimates. The proposed relative error index and quality classification scheme establish a useful quantitative basis for identifying potentially unreliable overcoring results, thereby improving the standardization, reliability assessment, and quality control of in situ stress determination.
Deep tunnels with elliptical cross sections are increasingly constructed in layered rock masses under high in situ stress, where tunnel response is strongly influenced by bedding orientation, stress direction, and cross-sectional geometry. However, closed-form analytical solutions remain limited for deep elliptical tunnels where the tunnel axis, bedding orientation, and principal stress directions are misaligned, particularly when assessing the initial instability tendency along the tunnel perimeter while accounting for anisotropic strength. A closed-form analytical solution for deep elliptical tunnels excavated in layered transversely isotropic rock under arbitrary three-dimensional far-field stresses was developed. This solution is then combined with anisotropic shear and tensile strength criteria to determine the Local Factor of Safety (LFoS) and to construct a weighted comprehensive stability index (WCSI) for comparative assessment of the initial instability tendency. Validation against numerical results shows close agreement for both perimeter stress and displacement distributions, with maximum stress and displacement errors below 2.16% and 2.79%, corresponding R2 values above 0.991 and 0.989, and an analytical runtime of about 0.014% of the FE wall-clock time for the same validation case. The results show that the alignment between the tunnel axis and the major principal stress is the dominant factor, while bedding orientation and aspect ratio also exert important influences. Configurations in the highest WCSI screening band are rare, accounting for only 0.89% of the orientation combinations considered. The proposed solution provides a rapid basis for pre-failure comparison of initial instability tendencies across tunnel orientations and aspect ratios, and for identifying critical perimeter sectors.
Geological prediction during long-distance tunnel boring machine construction is challenged by changing ground conditions, cross-line distribution differences, and sensor noise, which limits the reliability of static prediction models after deployment. This study develops a Bayesian continual learning framework for geological information prediction using real operational data from a twin-line Shenzhen intercity railway tunnel. Ring-level machine parameters were organized into sequential learning episodes to represent geological evolution, unknown geological states, cross-line transfer, and noisy data streams. The framework combines Bayesian inference with model updating strategies and separates epistemic uncertainty, which reflects limited model knowledge, from aleatoric uncertainty, which reflects data noise. In the static tests, the model achieved 84.76% accuracy for known geological states, whereas direct cross-line transfer reduced the accuracy to 45.40%. Unknown geological states increased the mean epistemic uncertainty by about 2.3 times, indicating that uncertainty can flag unfamiliar conditions but cannot recover prediction performance without updating. In the dynamic tests, experience replay reduced catastrophic forgetting and increased the final accuracy from 28.7% for naive fine-tuning to 77.2% during sequential geological evolution. For cross-line transfer, experience replay achieved 79.9% accuracy and the lowest epistemic uncertainty. Elastic weight consolidation was more effective for newly emerging geological states, reaching 82.1% accuracy while retaining higher uncertainty for unknown states than for known states. Under strong noise, all updating strategies maintained >85% accuracy, and aleatoric uncertainty dominated the total uncertainty. These results suggest that Bayesian continual learning can link uncertainty-based warnings with subsequent model updating, supporting uncertainty-informed geological perception and model maintenance during TBM construction.
A parametric reduced-order fracture model is developed to predict deformation and stress fields induced by a planar fracture of arbitrary shape embedded in an isotropic elastic half-space using the distributed dislocation technique. The fracture front is represented as a smooth Jordan curve defined through cubic spline interpolation of parametric nodes, and fluid pressure within the fracture is parameterized along streamlines of an orthogonal curvilinear coordinate system conformal to the fracture front. This model serves as an efficient forward solver, enabling systematic evaluation of fracture parameters in inverse analyses. A joint inversion framework integrating tilt, strain, and fluid pressure data is proposed to simultaneously estimate fracture orientation and dimensions using the Differential Evolution global optimization algorithm. Numerical experiments are conducted to assess the accuracy and robustness of the proposed joint inversion method for fracture geometry mapping and to compare the capability of global and local optimization strategies for this highly nonlinear, multimodal inverse problem. Results demonstrate the complementarity between far-field (surface) and near-field (downhole) deformation, rotational (tilt) and translational (strain) responses, fluid pressure and fracture deformation, as well as surface (two-dimensional areal) and depth (one-dimensional linear) data coverage in resolving different aspects of fracture geometry. Inversions of synthetic data and field measurements show that the joint inversion significantly improves the accuracy and reliability of hydraulic fracture characterization.
As natural weak structures in layered rock masses, the bedding planes lead to complex mechanical responses and failure modes of the surrounding rock under tunnel excavation. To accurately reveal the progressive failure mechanism of layered surrounding rock during excavation, this study combines the ABAQUS user subroutine (VUSDFLD) developed in Fortran with the Cohesive Zone Model (CZM) to construct an improved implicit-explicit sequential analysis framework of the CZM-FDEM. This framework successfully resolves the problem of computational convergence during the excavation of the surrounding rock. In addition, the quantitative identity of the type and account of fractures, and reconstruction of spatial distribution features of tension-shear failure were conducted through Python. The failure mechanism of layered surrounding rocks was systematically analyzed. The results indicate that the bedding plane significantly weakens the structural integrity of the rock and induces anisotropic characteristic failure of the surrounding rock. The smaller the layer thickness, the more significant the interface stress concentration, which more easily triggers interlayer sliding and through-layer failure, forming larger fragmented rock blocks. Under these conditions, the number of shear fractures accounts more than 80% and plays a dominant role in the process of rock failure. When the lateral coefficient of in-situ stress lambda = 1, the increase of bedding inclination results in the transformation of failure modes from tensile-shear composite type to shear slip type. This drives the directional expansion of the excavation damage zone (EDZ). When lambda/=1, the directional effect of in-situ stress overrides the influence by bedding inclination, promoting preferential development of damage zones along the direction of the small in-situ stress. Moreover, the number of fractures increases significantly with the increase of lambda. The simulation results were compared with the published literature. The accuracy and reliability of the CZM-FDEM method in this study are verified in predicting the extent of the EDZ and the patterns of crack distribution.
CO2 capture, utilization, and storage in depleted petroleum reservoirs is one of the rapidly growing areas of research in the petroleum industry. Of these reservoirs, shale gas reservoirs are particularly promising, considering the potential for trapping the injected CO2 by sorption, in addition to other trapping mechanisms expected in other subsurface reservoirs. However, accurate characterization of the pressure transient behaviors during CO2 injection is imperative for the estimation of reservoir properties and CO2 storage capacity. The accuracy of standard models for capturing transient flow dynamics remains limited in representing the early-time interactions between fractures and the surrounding matrix. To address these limitations, this study proposes an Analytically Projected Embedded Discrete Fracture Model (AEDFM), in which fractures are projected onto all six faces of three-dimensional host matrix elements. This approach systematically incorporates the influence of surrounding matrix blocks, enabling a more accurate simulation of the CO2 pressure transient behaviors of multistage fractured horizontal wells (MFHWs) in shale gas reservoirs. The accuracy of the proposed model is validated against reference solutions. Utilizing the proposed AEDFM, we developed a workflow for CO2 storage capacity evaluation based on pressure transient analysis. Furthermore, we analyzed the impact of fracture and reservoir parameters on pressure response and CO2 storage performance. Based on pressure test data from a shale gas well in the Longmaxi formation, reservoir rock properties, such as permeability, were obtained through curve matching. Using the matched property values, the CO2 storage capacity was evaluated at different injection rates. Overall, the proposed model offers an efficient and accurate framework for analyzing transient flow and assessing CO2 storage capacity in shale reservoirs.
Rock engineering in cold regions is persistently challenged by the coupled effects of freeze-thaw (F-T) cycling and temperature. Revealing the damage accumulation and fracture-evolution mechanisms under such conditions is essential for stability assessment. In this study, a micro-macro fracture mechanics model is established by explicitly coupling the F-T cycle number and temperature. The initial damage D0, friction coefficient & micro;, and mode-I fracture toughness KIC are formulated as variables that evolve with both F-T cycles and temperature. The freezing stress field is decoupled into effective skeleton contractile stress Pti, effective frost heave stress Pfi, and effective cohesive stress Pci, thereby building a quantitative link between microscopic damage accumulation and macroscopic mechanical response. The results indicate that, in both frozen and thawed states, F-T cycling causes nonlinear degradation of macroscopic strength. Moreover, the rock strength in the frozen state is markedly higher than that in the thawed state, due to low-temperature strengthening and cohesion effects. This study primarily validates the proposed model with experimental data of green sandstone and anhydrite rock in the thawed state, and systematically clarifies the micro-macro strength evolution in the frozen state. The proposed framework provides a theoretical basis and quantitative tools for stability analysis and protection design of cold region rock engineering.
Reinforcing fractured rock masses remains challenging due to irregular fracture surfaces and orientations, stress-induced deformation, and the tendency of conventional cementitious grouts to shrink and generate interfacial cracks after curing. These limitations reduce bond strength, weaken stress transfer, and compromise the longterm performance of grouted anchoring systems. To address these limitations, expansive mortar grouting has been introduced as an alternative reinforcement approach that improves interfacial bonding and shear resistance along rock discontinuities. This study combines laboratory experiments and numerical simulations to investigate the pull-out mechanical behavior of expansive mortar grouted anchors and the mechanical response of fractures with different orientations within the rock mass. The results indicate that the average ultimate pull-out force of expansive mortar grouted anchors reaches 91.63 kN, representing a 53.7% increase compared with ordinary mortar grouted anchors 59.60 kN. The pull-out force-displacement curve shows higher initial bond stiffness and a nonlinear hardening trend, demonstrating greater stress transfer and deformation coordination in the expansive mortar grouted anchoring system. The failure mode shifts from shear slip at the anchor-mortar interface and adjacent mortar scraping in ordinary systems to shear slip at the mortar-rock interface in expansive mortar systems. Furthermore, the radial expansive stress generated by the expansive mortar significantly increases the normal compressive stress and shear resistance of fractures that are not perpendicular to the anchor axis, thereby improving fracture resistance to failure. This work provides insight into the effective reinforcement of fractured rock masses using expansive mortar grouted anchors, particularly in conditions where fracture orientation and grout-rock interface stability are critical to advancing tunneling and rock grouting engineering applications.
Hydraulic fracture height containment is a persistent problem in layered continental shale formations and is commonly interpreted using stress-contrast-based criteria. However, fractures frequently remain height-limited even under weak stress shielding, indicating that additional mechanical controls remain insufficiently understood. In this study, the role of lithofacies-related mechanical heterogeneity in governing hydraulic fracture height growth is systematically investigated using continental shale from the Fuxing area. Quantitative mineralogical analysis, thin-section observation, and nanoindentation measurements reveal pronounced mechanical contrasts between clay-rich shale and shelly limestone interlayers, particularly in elastic stiffness and deformation behavior. True triaxial hydraulic fracturing experiments conducted under unified stress conditions demonstrate that fractures in clay-rich shale preferentially propagate along bedding planes, resulting in height containment, whereas shelly limestone promotes the development of vertically oriented dominant fractures when sufficient net pressure is sustained. Numerical simulations further indicate that although interlayer stress contrast remains the primary control on fracture height growth, lithofacies mechanical mismatch plays a decisive regulatory role under weak stress shielding conditions. The coupled effects of injection rate and fluid viscosity are shown to enhance vertical fracture penetration by maintaining fracture net pressure and reducing energy dissipation at lithofacies interfaces. These results extend conventional fracture height theories by explicitly incorporating lithofacies-dependent mechanical heterogeneity and provide a rock-mechanics-based framework for understanding fracture height evolution in strongly heterogeneous continental shale formations.
Dynamic disturbances from blasting or impact events induces transient stress wave propagation in underground rock masses, causing rapid and complex stress redistribution. In faulted rock masses, pre-existing discontinuities trigger wave reflection, transmission, and scattering at the fault interface, generating fault-geometry-dependent spatiotemporal stress concentrations. Quantitatively characterizing these transient wave-interface interactions remains challenging due to the rapid and mode-dependent energy partitioning, and the difficulty in capturing full-field stress distributions across complex interfaces with sufficient temporal and spatial resolution. This study develops an integrated experimental framework using 3D-printed transparent models and dynamic photoelasticity to achieve the direct quantitative visualization of stress wave propagation. A key innovation is the implementation of an automated digital workflow-incorporating pixel-level skeletonization and global phase unwrapping-which overcomes the inherent phase ambiguity and fringe overlap in transient photoelasticity. This technical breakthrough enables high-resolution, full-field reconstruction of principal stress differences (PSD) from single-shot fringe patterns. Systematic investigations of under-fault and fault-intersected tunnels reveal that faults act as critical regulators of dynamic stress, governing redistribution through reflection-induced shielding and guided propagation. Results show that under-fault configurations exhibit distributed stress amplification, whereas fault-intersected configurations produce localized, roof-dominated concentrations. A PSD-based framework is established to link transient stress evolution with failure criteria, providing a robust tool for evaluating dynamic fault slip and tunnel instability. These findings provide direct quantitative evidence of geometry-controlled wave-interface coupling and establish a new experimental foundation for investigating dynamic failure in discontinuous rock masses.
Understanding the interface behavior of rocks and structural materials is essential for assessing and ensuring structural stability in geotechnical and mining engineering applications like grouting and backfilling. The smallscale study can provide a refined analysis, including capturing the local stress concentration and microcrack propagation behavior in the vicinity of the interface, which is easily averaged and masked in large samples. This work utilizes a customized micromechanical tester, integrated with digital image correlation (DIC) technology, to investigate the interface shear behavior of small-scale analog cement-rock systems under various normal confinements. The results demonstrate a transition from brittle to ductile behavior of the interfaces with increasing confinement, associated with the microcrack evolution pattern of the localized band, embodied as a shift of separation dilation to overall dilation of the bonded sample. The dilation process is approximately divided into three stages based on the confinement magnitude, however, the DIC measurements do not show any correlation between the size of the localized band and the confinement. The strength envelope of the cement-rock interface satisfies the Mohr-Coulomb (MC) failure criterion with a tensile cutoff, transitioning to a cohesionless interface dominated by friction after bonding failure. The variations in peak force and shear behavior induced by the shear velocity effect manifest at the cement-rock interface. In terms of fatigue behavior in shear, the evolution and transformation of energy between elastic and plastic forms is largely affected by the shearing displacement. In total, the small-scale interface study offers insights into the constitutive behavior and modeling of weakly discontinuous interfaces, including parameter selection, rate effect, and energy evolution, with potential applications in assessing risks in underground geotechnical/mining engineering projects.
During deep coalbed methane production associated with CO2-hybrid fracturing, CO2 exists in multiple phases under high temperature and pressure, forming a complex gas-water flow system. To investigate this behavior under one reservoir-relevant condition, high-temperature and high-pressure low-field nuclear magnetic resonance experiments were conducted on a representative deep coal sample from the Benxi Formation (Ordos Basin, China) at 363.15 K and an effective stress of 20.8 MPa. CO2 and N-2 were used as injection gases for comparison. Displacement responses vary across T-2-defined pore-scale intervals. In micropore-associated intervals, the signal evolution reflects fluid redistribution that may be related to adsorption and exchange effects. The mesopore-associated interval shows the largest signal variation. The macropore-associated upper-end T-2 response is 43%-67% lower during CO2 displacement than during N-2 displacement. Kinetic analysis shows a rapid signal reduction in macropores at the early stage and a similar to 1.5-fold increase in mesopore water mobilization during the phase-transition stage. In the gas stage, mesopore variation becomes limited, indicating reduced large-scale displacement and increased capillary-controlled redistribution. Water-locking behavior is quantified using a water-locking index (WLI) and a hydraulically corrected index (HCWLI). In the CO2 system, mesopore WLI values are approximately 4.0%, 8.1%, and 5.8% for P1-P3, respectively. In contrast, N-2 shows weaker and more dispersed responses, with mesopore WLI around 2.5% in P2 and below 1% in other stages. The overall displacement efficiency of N-2 is 50.8%-56.2% of that of CO2. The early-stage response is consistent with a possible dissolution-influenced process. Overall, displacement behavior reflects the combined effects of phase behavior, capillary forces, and fluid-rock interactions.
Accurate identification of rock types from well logging data is essential for detailed reservoir characterization and evaluation. However, current machine learning and deep learning methods rely heavily on data patterns without incorporating fundamental physical principles; this limits their ability to distinguish rock types with similar geological origins and overlapping well log responses, often yielding predictions that contradict rock physics and leading to unreliable geological interpretations. To address this issue, this study proposes a Physicsinformed Graph Attention Network (PiGAT) model that embeds classic rock physics relationships as constraints within a data-driven framework. The model transforms one-dimensional well log sequences into a graph structure to capture spatial relationships between geological layers. It employs a Graph Attention Network (GAT) to dynamically learn complex response patterns between layers, effectively preserving key features at lithological boundaries. A hybrid loss function incorporates multi-scale rock physics constraints into the model's training process, including density-porosity mass balance, Kozeny-Carman (K-C) equation for permeability-porosity trends, and steady-state Darcy's law for fluid transport consistency. The model also uses a dynamic weighting strategy based on homoscedastic uncertainty to adaptively optimize multiple task losses, guiding it toward solutions that are both physically consistent and closely aligned with the data. Experiments on the North Sea well logging dataset demonstrate that the PiGAT model achieves an F1-score of 0.9528, significantly outperforming baseline models. Ablation studies confirmed that incorporating physical constraints is critical for improving model generalization and preventing overfitting. The PiGAT framework proposed in this study provides an effective solution for addressing complex geological problems. The multi-constraint strategy presented here is not limited to lithology identification from well logs; it also provides a methodological reference for multi-task well log inversion problems related to rock mechanics, such as caprock integrity evaluation and wellbore stability analysis.
This study presents a 3D polygon-based Genetic Discrete Fracture Network (G-DFN) modeling framework that generates high-fidelity fracture networks by explicitly simulating fracture nucleation, growth, and arrest processes. Fracture growth is governed by Charles' law, while vertex-level abutment rules and a sub-stepping scheme ensure realistic arrest behavior and improved resolution near terminations. These mechanisms enable fracture scaling properties and topological patterns-such as Y-nodes and T-shaped abutments-to emerge naturally, yielding networks that are more consistent with field observations. To link input parameters with observations, a calibration scheme is developed that uses direct constraints from trace map data and an iterative cyclic update procedure to address coupled parameter effects. Synthetic verification showed that the scheme can recover governing parameters and reproduce fracture scaling and topology within stochastic variability, while application to field trace maps demonstrated its ability to generate site-specific, geologically plausible networks. Overall, the framework provides a process-informed and data-integrated approach for DFN modeling, offering high-fidelity realizations of fracture networks.
Quantifying the evolution of permeability in deeply buried porous rocks remains one of the most challenging issues in geomechanics. In this study, we propose a micromechanics-based model that integrates damage theory with poroelasticity to capture the full hydro-mechanical response of porous rocks. The representative volume element (REV) is conceptualized as three interacting continua: an isotropic elastic solid skeleton, randomly distributed spherical pores, and oriented micro-crack clusters. Using Hill-Mandel homogenization, we derive macroscopic constitutive equations that incorporate both pore-fluid and crack-fluid pressure, enabling the prediction of the stress-strain path and the anisotropic Biot tensor without additional phenomenological assumptions. Porosity and crack density are updated incrementally based on the local stress field, while the overall permeability is computed as the sum of contributions from a pore network and a crack channel system. A segmented damage index is introduced to characterize the transition from intact elasticity to distributed cracking and, ultimately, to localized failure, allowing permeability to either increase or decrease within a single computational step. The proposed framework successfully reproduces laboratory triaxial flow-through data for sandstone, granite, coal, and shale, achieving a coefficient of determination greater than 0.93. Global variancebased sensitivity analysis reveals that initial porosity, mean pore-throat radius, geometric tortuosity, and crack intensity factor are the most influential parameters governing permeability evolution. Furthermore, the model is applied to simulate excavation-induced permeability changes around a deep tunnel, accurately predicting permeability enhancements of 3-5 orders of magnitude near the excavation boundary. This work provides a robust theoretical basis for permeability prediction in deep geological settings, with significant implications for energy extraction, underground storage, and geohazard mitigation.
