Advancing methane hydrate (MH)-based technologies for efficient energy storage and gas recovery fundamentally relies on precise regulation of their phase equilibrium and kinetics. Accordingly, hydrophobic modified calcium oxide (mCaO) is herein proposed as an additive to regulate MH behaviors through hydration heat and surface properties. Targeting a prolonged exothermic duration and minimal heat loss, the hydrophobic modification process was optimized via orthogonal experiments. The effects of CaO and mCaO on MH formation thermodynamics and kinetics are also elucidated, with both CaO and mCaO systems exhibiting thermodynamic inhibition. An established Antoine-like model predicts the phase equilibrium conditions with an error of less than 1.99%. In terms of kinetics, at low concentrations (<3 wt%), the mCaO systems exhibits severe kinetic inhibition, which is hypothesized to be driven by surface shielding from organic silane chains and steric hindrance. Conversely, at higher concentrations (>3 wt%), the mechanism reverses to promote nucleation, likely pointing towards the gradual exposure of active sites. Furthermore, based on a comprehensive evaluation index (E-r) coupling thermodynamic and kinetic parameters, customized energy conversion strategies are proposed: the low-concentration mCaO systems is optimal for hydrate recovery and flow assurance, while the high-concentration CaO systems favors high-efficiency solidified energy storage. This work provides novel insights for tailored management in MH recovery and storage.

Particle size distribution (PSD) strongly influences particle transport, collision dynamics, and vibration excitation in sand-laden gas flows, but non-intrusive PSD characterization in closed pipelines remains challenging. This study integrates CFD–DEM simulations with collision-induced vibration experiments to investigate particle transport and particle–tool impacts and to predict particle-size quantiles. Four natural quartz sand samples were characterized by microscopy to obtain number-based reference PSDs and bootstrap confidence intervals for D10, D50, and D90. Window-level time-domain, event energy, and band energy features were used to train random forest models for the three quantiles, from which a lognormal parametric cumulative PSD approximation was constructed. At a fixed particle mass flow rate and injection duration, the pronounced increase in total collision count for finer particles was governed mainly by their larger particle number. Constant-particle-number simulations nevertheless showed moderately higher collision event density and collision participation for finer particles, indicating an additional size-dependent transport contribution. Trend-level comparisons between simulated collision descriptors and measured vibration features supported qualitative mechanistic interpretation without implying a one-to-one force-to-acceleration mapping. The RF model yielded an aggregate window-level RMSEPSD of 9.60 ± 0.17%. In five-fold leave-one-FlowConditionID-out validation, it yielded MAE = 0.0479 ± 0.0110 mm, RMSE = 0.0652 ± 0.0150 mm, R2 = 0.749 ± 0.110, and RMSEPSD = 12.17 ± 2.72%; these results did not resolve a clear RMSEPSD advantage over the Energy/peak baseline. The method is limited to the four calibrated PSD groups and the apparatus and operating ranges examined here and provides a quantile-constrained parametric approximation rather than arbitrary pointwise PSD reconstruction.

Deepwater gas invasion is highly sudden, concealed, and characterized by complex pressure–flow coupling, and failure to accurately capture its pressure evolution can rapidly lead to uncontrolled blowout events. To enhance well-control reliability, this study develops a transient gas–liquid two-phase flow model for deepwater ultra-deep horizontal wells based on mass and momentum conservation, and employs numerical simulation to analyze the temporal and spatial evolution of bottomhole pressure, annular pressure, and wellhead casing pressure during gas invasion and shut-in. Results show that bottomhole pressure continuously decreases after gas invasion, with a faster decline at higher circulation rates, while longer horizontal sections generate higher baseline bottomhole pressure due to increased friction. As gas invasion progresses, the annular pressure profile shifts downward and the invaded gas zone expands to nearly three times its initial extent within 45 min; lifting the drill bit off bottom significantly increases gas influx and mud-pit gain and accelerates bottomhole pressure decline. Horizontal wells exhibit higher post-invasion bottomhole pressure and a distinct inflection point compared with vertical wells, enabling determination of the gas invasion location through the proposed multiphase-flow-based method. The shut-in process comprises overflow, pressure-stabilization, and rapid pressure-rise stages, with the stabilization stage representing the critical “golden window” for well killing. Delaying shut-in from 15 to 35 min increases wellhead casing pressure by 81

Sand production monitoring is essential for production safety, equipment integrity, and sand-management decision-making in oil and gas wells. However, existing monitoring methods target different physical quantities and stages of the sand-production process. Sampling captures transported sand particles, acoustic and vibration methods record impact-induced dynamic responses, distributed fiber-optic sensing mainly provides distributed acoustic/vibration-related anomalies, and ER/erosion probes measure local material loss. Therefore, direct comparison using a single performance indicator may lead to misleading technology assessment. This review develops a measurement-target-based framework that links sand generation, transport-impact-erosion processes, signal formation, monitoring outputs, and scenario-based technology selection. Downhole, surface, and subsea monitoring routes are compared in terms of measured quantity, calibration requirement, uncertainty source, field maturity, and decision suitability. Particular attention is given to practical limitations, including DAS interpretation uncertainty, acoustic calibration transferability, ER-probe locality, and subsea reliability constraints. An uncertainty-aware target-specific evaluation framework and a scenario-based technology-selection matrix are proposed to support method combination and decision-oriented sand management. Finally, the review discusses how monitoring data can be coupled with sanding prediction, erosion-risk assessment, safe operating-envelope updating, and choke or intervention decisions. The analysis indicates that reliable sand management benefits from calibrated, cross-validated, and decision-oriented integration of physical sampling, dynamic-response monitoring, distributed diagnostics, erosion assessment, and production data.

Water exhibits a variety of anomalous behaviours. To date, many computer simulations and experimental studies have shown compelling evidence for the existence of a liquid-liquid phase transition between the high-density and low-density phases in the deeply supercooled regime and for a critical point that terminates the boundary between the two phases. A possible theoretical scenario concerning the origin of these anomalies is the two-state model of water, which posits that water is composed of two distinct and interconvertible local structures. Here we identify multidimensional reaction coordinates that can provide molecular-level evidence for the generic existence of two local structures undergoing interconvertible reactions in liquid water. By employing unsupervised deep learning on massive molecular dynamics simulations of an accurate and widely used water model, we found that near the high-density/low-density phase boundary, interconvertible reactions proceed via a full-loop reaction pathway with three saddle points, whereas away from it the reactions proceed via a semi-loop pathway with a single saddle point. These findings provide molecular-level evidence in support of the two-state water model and may offer physical insights into the origin of the liquid-liquid phase transitions more generally.

This study investigates the corrosion mechanisms of L80-13Cr stainless steel (SS) in highly mineralized produced water from oil and gas fields. The formation, evolution, and interaction of passive films and corrosion products under the effect of turbulent flow at 150 degrees C and 2 MPa PCO2 were systematically investigated. The results show that the average corrosion rate (vcorr) is higher under turbulent flow than under static conditions. Under static conditions, the outer Cr2O3 passive film transformed into FeCr2O4, while the inner layer evolved from Cr2O3 and Fe2O3 into a mixed carbonate deposit of CaCO3 and FexCayCO3 (x + y = 1). In contrast, turbulent flow increased the mass transfer of corrosive ions, promoting early nucleation and growth of FexCayCO3 mixed carbonates in both layers while delaying FeCr2O4 spinel formation. The mixed carbonate deposits increased layer thickness by filling cracks in the passive film and accumulating at the film/substrate interface, enhancing the long-term corrosion resistance of L80-13Cr SS. This study offers new insights into the corrosion mechanisms of L80-13Cr SS in high-temperature, high-pressure (HTHP), highly mineralized CO2 environments, with implications for materials selection in aggressive oil and gas field applications.

Hydrate-wax-asphaltene multi-solid-phase coupled deposition and plugging in deepwater oil and gas production/transport systems is a critical flow-assurance safety challenge. However, the microscopic interfacial forces between hydrates and the wax and asphaltene fractions in crude oil remain poorly understood. In this work, a self-designed high-pressure visual micromechanical apparatus was employed to, for the first time, comparatively investigate the microscopic interaction forces between CH4 hydrate particles in three oil-dominated systems: pure oil, waxy oil, and asphaltenic oil. The effect of heavy hydrocarbon components on the interfacial cohesion behavior of hydrate particles was systematically explored. The results show that, at subcoolings of 1–10 °C, the microscopic force between hydrate particles in the oil phase ranges from 1.99 to 6.32 mN·m-1, which is significantly lower than that in the gas phase by about 74.7%–79.1%. Lower subcooling and longer contact time favor the formation of liquid bridges between particles, thereby increasing the microscopic interaction force. In wax-containing oils, the microscopic force between hydrate particles further decreases to 1.2–3.8 mN·m-1, corresponding to a reduction of about 29.0%–42.1% relative to the pure oil system. In asphaltene-containing oils, when the asphaltene concentration is 0.2 wt%, the microscopic force is reduced by approximately 20.9%–36.8% compared with pure oil; however, when the asphaltene content increases to 0.5 wt%, the reduction in microscopic force is only about 8%. Wax and asphaltenes markedly weaken the cohesive adhesion between hydrate particles by modifying interfacial wettability and mass transfer processes. The presence of heavy components slows down hydrate formation and decreases the surface hydrophilicity of hydrate particles: wax crystals form an oil-wet layer on the particle surface, while asphaltene molecules adsorb at the interface to form a hydrophobic film, both of which suppress water-bridge formation and direct particle-particle adhesion. At higher asphaltene loadings, however, precipitated asphaltene solids may create new adhesive bridges with hydrate particles. From the perspective of microscopic interaction forces. This study elucidates, from the perspective of microscopic interaction forces, how heavy oil fractions in the oil phase influence hydrate particle aggregation and deposition, providing a useful reference on interfacial behavior for deepwater hydrate prevention and inhibitor development.

In the development of deepwater gas fields, hydrate flow risks in gas-dominated pipelines pose a significant threat to flow assurance. At present, there is no mature risk assessment model capable of accounting for hydrate behavior under long-distance coupled multiphase flow conditions. This study establishes a multiphase flow pressure-temperature distribution model and couples it with a hydrate formation and deposition kinetics model. This framework systematically reveals the spatial-temporal evolution of hydrate deposition layer thickness and the associated variations of multiphase flow parameters during flow obstruction. Results demonstrate that the model can reliably capture the non-uniform growth of hydrate layers along the pipeline wall as a function of time and distance. Furthermore, it shows that high-risk zones for flow blockage gradually shift downstream with time, and that the tendency of hydrate blockage decreases as the pipeline inclination increases. This work bridges the limitations of conventional models that neglect the coupled mechanisms of multiphase flow and hydrate formation-deposition kinetics, thereby providing both theoretical support and a predictive tool for the safe and efficient development of deepwater gas fields.

Deepwater drilling is conducted in structurally complex formations and harsh marine environments, which greatly intensifies the difficulty of maintaining well control. Reliable prediction of the shut-in pressure response in the wellbore is therefore essential for safe and efficient operations. In this work, published laboratory data are reanalyzed to derive empirical correlations that describe how drilling-fluid density changes with temperature and pressure. On this basis, the thermal expansion behavior of the drilling fluid is incorporated, and segmental temperature-field models are formulated for the seawater column, choke line, and formation intervals. A corresponding gas-expansion model and a shut-in wellhead-pressure prediction model suitable for deepwater conditions are then developed. Numerical examples are used to investigate the evolution of temperature, pressure, and fluid properties under different geothermal gradients. The simulations indicate that drilling-fluid density decreases exponentially with temperature and increases approximately linearly with pressure. In the choke line, the annular temperature continuously drops under the cooling effect of seawater, whereas in the formation interval it decreases in the upper part and increases toward the bottom, ultimately approaching the geothermal profile. When the thermal expansion of drilling fluids is taken into account, the geothermal gradient exerts a strong influence on the shut-in behavior: larger gradients enhance fluid expansion, leading to higher wellhead casing pressures and faster recovery of bottomhole pressure, which places stricter requirements on well-control design. The proposed method provides a useful theoretical reference for pressure prediction and safety evaluation during deepwater drilling shut-in.

The formation pressure system of deep and ultra-deep carbonate reservoirs is complex, with widespread fracture development and a narrow safety pressure window. The use of conventional cyclic well control methods often leads to overpressure and leakage alternation. The bullheading well control method is a commonly used technique to address blowouts in such oil and gas reservoirs. However, during the bullheading method, gas-liquid counter-current two-phase flow is frequently encountered, yet it remains inadequately studied. This gap in scientific understanding may result in potential hazards. This study carried out experiments to investigate gas-liquid counter-current two-phase flow at various inclination angles and viscosities. The research revealed the transition behaviors of the flow and developed a set of criteria to identify shifts in the flow patterns. A model for two-phase flow in bullheading well control within the wellbore was developed. The discrepancy between the predicted and observed casing pressures during the well control phase was found to be within 15%. In addition, the study explored the effects of flow rate, density, and viscosity of the well control fluid on wellbore pressure and the distribution of gas holdup. The results indicate that a higher fluid flow rate leads to a shorter well control duration but causes an increase in casing pressure at the wellhead. On the other hand, an increase in fluid density accelerates the reduction in casing pressure, ultimately lowering the pressure at the wellhead. Increasing the viscosity of the well control fluid slows down the casing pressure decline, leading to a higher wellhead casing pressure, while the required backflow time decreases. The research suggests that appropriately increasing the density and viscosity of the well control fluid can improve the efficiency of bullheading operations for well control.

Elucidating the multi-field, multi-phase coupling mechanisms in submarine hydrate reservoirs is essential for releasing their energy potential. Current research gaps remain in understanding the permeability-stress coupling characteristics of hydrate reservoirs. This study conducted permeability-stress coupling tests on hydrate-bearing clayey-silty sediments (HBCS). The seepage characteristics and mechanical response of HBCS under seepage-stress coupling effects were investigated. The results indicate that under stress conditions, the seepage flow of HBCS follows a logarithmic relationship with axial strain, with its rate of increase decelerating as axial strain grows. Apparent permeability decreases rapidly at first and then more gradually with increasing axial strain. Hydrate saturation (Sh) and seepage pressure difference (PS) exhibit a coupling effect on apparent permeability. At low-Sh levels, particle migration induced by seepage reduces permeability. At high-Sh levels, permeability is affected by two competing mechanisms (pore expansion and particle migration) driven by seepage forces. Unlike triaxial tests for HBCS, under seepage conditions, the plastic deformation capacity of HBCS is enhanced. The stress-strain curve exhibits strain hardening and plastic flow, and lacks obvious strain softening. Shear strength decreases with increasing Sh and PS, while the secant modulus E50 rises with both parameters. During hydrate exploitation, production efficiency correlates with Sh. In low-Sh reservoirs, a “slow and steady” production strategy is recommended to prevent sand production and wellbore blockage. In high-Sh reservoirs, increasing PS can promote pore expansion and fracture propagation, thereby enhancing production efficiency. This study provides a reference for engineers to understand the multi-field and multi-phase coupling characteristics during the exploitation of hydrate reservoirs.

The complex architecture and stacking patterns of deepwater turbidite channel sandbodies introduce significant uncertainty in injector-producer connectivity. This uncertainty affects both the mechanisms and the quantitative evaluation of the waterflood sweep. In this study, a representative reservoir in the Niger Delta Basin is selected as a case study. Injector-producer well groups are first classified into three connectivity patterns-coeval, cross-stage, and hybrid based on geological and seismic constraints. Time-lapse seismic data are then interpreted to delineate sweep morphology and to infer the controlling mechanisms associated with each pattern. Coeval connectivity exhibits a relatively uniform and continuous front advance with minimal barriers. Cross-stage connectivity shows fragmented swept regions with pronounced bypassing, and localized preferential breakthrough caused by discontinuous sandbodies and pervasive barriers. Hybrid connectivity is characterized by intermediate behavior, combining features of both patterns. To translate these mechanistic differences into quantitative metrics for development evaluation, an oil-water relative permeability ratio correlation for low viscosity oil is established that remains valid across the full water cut range, thereby overcoming the limitations of conventional semi-log linear correlations at both low and ultra-high water cut stages. Based on this framework, a production data-driven predictive model for waterflood sweep efficiency is derived using production data and steady state flow theory. The model is validated across well groups representing different connectivity patterns. Field application yields a consistent ranking of sweep efficiency: coeval > hybrid > cross-stage, with group average values of 0.86, 0.80, and 0.70, respectively. These results agree with the mechanistic interpretation derived from time-lapse seismic analysis. The proposed methodology provides a practical quantitative framework for evaluating injector-producer connectivity and comparing development strategies in deepwater turbidite channel reservoirs.

While horizontal wells for gas hydrate exploitation allow for a larger effective contact area between the horizontal well and the reservoir, they also face severe wellbore stability problems. Two key issues are the dynamic drilling fluid safety density window and the limiting extension length of the horizontal section. This study established a horizontal well drilling model for hydrate reservoirs and a model for analyzing key parameters of the horizontal section based on the basic data of hydrate reservoirs at station SH2 in the South China Sea. The factors influencing the variation of the dynamic drilling fluid safety density window and the ultimate extension length of a horizontally drilled well section were analyzed using this model. The results showed that: The stress distribution in the surrounding rock of the well wall in the hydrate formation is dynamically changing due to changes in formation mechanical parameters resulting from hydrate decomposition after drilling fluid intrusion into the reservoir. This ultimately results in the safety density window for drilling fluids in horizontal wells drilled in marine hydrate reservoirs, which also changes dynamically over time. Drilling fluid physical properties play a vital role in limiting borehole extension length. As the drilling fluid injection temperature, rate and density increase, the ultimate extension length of the horizontal section decreases. Therefore, reducing the drilling fluid injection temperature, rate and density is an effective means to extend the borehole extension length and avoid wall stability problems. This study can provide theoretical and technical support for the safe and efficient drilling and extraction of natural gas hydrate in the future sea area.

Accurate and timely detection of string leakage in ultra-deep gas wells is vital for operational safety. Current methods largely depend on instrumental measurements, with limited intelligent approaches based on multi-source time-frequency data. Based on wellbore multiphase flow theory and considering temperature-pressure coupling effects, this study establishes a high-fidelity dynamic dataset (leak depth: 500-7500 m; size: 0.1-30 mm). A time-frequency dual-stream gated attention network is proposed, integrating parallel multi-scale Convolutional Neural Network (CNN) and Fast Fourier Transform (FFT) to capture local anomalies and frequency-domain signatures. Bidirectional Long Short-term Memory (BiLSTM) with adaptive multi-head attention highlights transient signals, while gated residual units remove feature redundancy, addressing the issue of leak signal decoupling and quantitative inversion in a complex noise environment. The model achieves highly accurate and robust regression predictions, with Root Mean Square Error (RMSE)≈30 m and Mean Absolute Error (MAE)≈0.09 mm for location and size, respectively (location: R2 > 0.99; size: R2 > 0.99). Through feature Integrated Gradients analysis, annulus head pressure is identified as the dominant feature (20 %-30 % contribution). The Gated Residual Network (GRN) module effectively filters nonlinear redundant information, reducing MAE by 24 %-58 % compared to Multilayer Perceptron (MLP) and Residual Network (ResNet). Furthermore, the time-frequency design lowers MAE by 70 %-84 % in location and 79 %-85 % in size compared to mainstream sequential models (e.g., TCN, Transformer). Under feature scarcity (1-2 features) or limited data (20 % samples), performance remains robust (location R2 > 0.99; size R2 > 0.97), outperforming single models by 47 %-79 % in MAE. Additionally, Light Gradient Boosting Machine (LightGBM) proves effective as a cost-efficient localization solution in resource-constrained scenarios (R2 > 0.995).

Solving the issue of hydrate formation and blockage in pipelines during deepwater oil and gas development and hydrate-based green energy extraction is a crucial flow assurance safety concern. The micro-forces between hydrate particles and wetted surfaces are key parameters in studying the aggregation or detachment behavior of hydrates on pipe walls. Current research primarily focuses on carbon steel surfaces, with a lack of studies on the micro-forces between hydrates and wetted sand grain surfaces in oil and gas pipeline flow systems and hydrate extraction processes. In this study, a visual particle micro-force measurement device was designed to investigate the micro-forces between hydrates and wetted sand grain surfaces in a high-pressure system. The results showed that at subcooling degrees of 1, 3, 5, 7, and 10 degrees C, the micro-forces between hydrate particles and wetted sand grain surface droplets ranged from 4242.46 to 8315.27 mN & sdot;m 1. The micro-force increased by 196.88%, 217.99%, 219.92%, 209.92%, and 218.47% compared to the micro-forces between hydrate particles and wetted carbon steel surfaces. Subcooling and contact time are strong functions of the micro-force between hydrate particles and wetted sand grain surfaces. The enhancement of the micro-force on the wetted sand grain surfaces is primarily attributed to the increased supporting force provided by the hydrate shell. The increase in micro-force is governed by the superposition of multi-scale adhesion forces. Factors such as the hydrophilic nature of sand grain surfaces, pore effects, blockage of pore channels, localized pressure concentration, and mechanical interlocking effects contribute to the increased micro-force between hydrate particles and the wetted sand grain interface. This study lays a foundation for elucidating the coupling deposition mechanism of hydrates and sand grains during deepwater oil and gas production, and further provides a theoretical reference for the safe, efficient, and continuous exploitation of hydrate-based green energy. (c) 2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-ncnd/4.0/).

Cementing slurry serves as a critical safety barrier for deep well cementing. In the gelation stage, cement slurry often forms microstructural pore channels, significantly increase the risk of well-bore seal failure and gas leakage during cyclic injection and production. To address these challenges, a deep learning-based adaptive characterization identification of cement microstructure is proposed. This study employs environmental scanning electron microscopy (ESEM) and semi-supervised image segmentation to simulate and characterize cement microstructures from a microscopic perspective. Imaging cement microstructures is challenging due to blurred edges, irregular feature patterns, and limited annotated data. To overcome these limitations, we propose a novel image segmentation model named Cement Segmentation Network (CemSNet), which integrates a hybrid adaptive attention mechanism. By combining channel and spatial self-attention, CemSNet effectively captures irregular micropore distributions and enhances edge definition, significantly improving the precision of segmenting fine-edge details and small features. To address the scarcity of labeled training data, a semi-supervised learning strategy is adopted. This approach leverages consistency regularization and pseudo-labeling to utilize unlabeled data, thereby enhancing model robustness. With limited labeled samples, the proposed method achieves a Mean Intersection over Union (MIoU) of 81.46 %, a Mean Dice coefficient of 89.28 %, and a pixel accuracy of 96.24 %, demonstrating high segmentation performance. Furthermore, based on segmentation results and physical modeling, this study reveals that both porosity and permeability decrease significantly over time, with rapid reductions in early stages and slower declines thereafter. Lower water-cement ratios accelerate this reduction, leading to denser microstructures. These findings offer a new perspective for evaluating the sealing performance of wells cement.

CO₂ hydrate capture technology, a key component of carbon capture, utilization, and storage (CCUS), is a crucial pathway for achieving carbon peak and carbon neutrality goals. However, the high randomness of hydrate nucleation and slow growth rates limit its industrial application. This study proposes a non-contact magnetic-driven gas–liquid interface disturbance system, where a Helmholtz coil generates a magnetic field that drives magnetic particles on the interface to rotate continuously, creating controllable disturbances. The spatial uniformity and temporal periodicity of the rotating field are validated via COMSOL simulation, and the liquid-phase flow and CO₂ mass transfer under different field frequencies (f = 10, 20, 40 Hz) and particle numbers (N = 10, 20, 40) are analyzed using particle image velocimetry (PIV) and numerical simulations. Experiments combining magnetic disturbance with dynamic promoters in a high-pressure reactor show that under optimal conditions (f = 40 Hz, N = 40), hydrate induction time decreases from 204.5 ± 71.5 min (static) to 38.1 ± 17.4 min, the time to reach 90% gas storage shortens by 43.0%, and the average formation rate rises to 0.060 ± 0.006 mmol·mL−1·min−1. Mechanism analysis indicates that interface disturbance shifts mass transfer from diffusion-dominated to convection-diffusion synergistic control. Introducing the dimensionless interface disturbance mass transfer number (IDT) provides a unified and predictive description of the formation kinetics: the power-law correlations reproduce the induction time (R2 = 0.852) and the growth rate (R2 = 0.960) well, and the negative power-law exponent of the induction time (Pearson r = −0.87) confirms that the induction time decreases monotonically as the interfacial-disturbance intensity increases.

Structural fractures and karst cavities are widely developed in deep and ultra-deep carbonate reservoirs, providing preferential pathways for rapid fluid migration while increasing the risk of severe drilling fluid loss. To investigate the lost-circulation mechanism in fractured-vuggy formations, a hydro-mechanically coupled gas–liquid two-phase seepage model was established by considering the multiple-media characteristics of matrix, fractures and cavities, as well as rock deformation and fluid compressibility. We hypothesize that gas–liquid property differences and hydro-mechanical changes in conductivity jointly control drilling fluid loss, with the gas–liquid property contrast exerting the stronger effect under the simulated conditions. In the model, flow in the matrix and fractures is described by Darcy’s law, while high-velocity flow in cavities is characterized using the Forchheimer non-Darcy equation. The coupling between the seepage field and stress field is achieved by incorporating the effective stress relationship, using the Kozeny–Carman porosity–permeability evolution model and the Goodman fracture deformation model. The coupled equations were implemented in COMSOL. Model validation confirms the reliability of the proposed model in predicting drilling fluid loss. The fracture–vug system significantly enhances fluid exchange between the wellbore and formation. Pressure propagates rapidly along fractures and vugs at the early stage and subsequently diffuses into the surrounding matrix, while the loss rate generally decreases with time. After 120 min, hydro-mechanical coupling increased the loss rate from 1.15 × 10−3 to 1.23 × 10−3 m3/s and the cumulative loss volume from 11.41 to 12.06 m3. Compared with the single-phase model, the gas–liquid two-phase model predicted a 4.82-fold higher loss rate. Fracture aperture, vug size, bottomhole pressure differential, and rock mechanical properties are the principal factors controlling loss intensity and pressure propagation. Through effective stress variations, hydro-mechanical coupling modifies porosity, permeability, and fracture aperture, thereby affecting formation conductivity and dynamic loss behavior. These results provide theoretical guidance for lost-circulation mechanism analysis, risk assessment, and plugging optimization in deep fractured-vuggy carbonate formations.

Leakage in ultra-deep gas well production tubing threatens tubing integrity and operational safety, yet the acoustic signatures of leakage-hole morphology, particularly threaded-connection leakage, remain poorly understood. This study proposes a data-driven acoustic framework for identifying leakage morphology. A three-dimensional tubing–annulus model covering no-leak, five equal-area conventional hole shapes, and threaded leakage was established. Compressible large-eddy simulation coupled with a direct acoustic approach was used to obtain leakage-induced pressure fluctuations, and laboratory experiments were conducted to validate the numerical model. Acoustic responses were analyzed in the time, frequency, and time–frequency domains. Thirteen acoustic features were used for K-means clustering, and several models based on one-dimensional convolutional neural networks (1D-CNNs) were developed. Noise robustness was evaluated by applying additive white Gaussian noise to the test signals at signal-to-noise ratios of 0–20 dB. The results show that leakage morphology significantly affects fluctuation amplitude, spectral distribution, and time–frequency evolution. Circular and square holes produce stronger fluctuations than slit-shaped, triangular, and irregular holes, whereas threaded leakage exhibits more oscillatory, localized, and non-uniform time–frequency responses. K-means clustering provides useful but incomplete separability among complex morphologies. Under noise-free conditions, the dual-branch multi-scale 1D-CNN achieves Accuracy, Macro-Precision, Macro-Recall, and Macro-F1 values of 98.06%, 98.18%, 98.06%, and 98.06%, respectively. At 0 dB, its Accuracy and Macro-F1 remain at 78.4% and 75.9%, respectively, demonstrating greater robustness than the single-branch models. These findings support the acoustic diagnosis of leakage morphology and tubing-integrity monitoring in ultra-deep gas wells.