In deep coal mining, the instability of roadway surrounding rock induced by complex geological structures poses critical threats to operational safety. Specifically, small-scale faults (offset <30 m) act as localized stress concentrators, yet their dynamic activation mechanisms and stress evolution remain poorly understood compared to regional tectonic structures. To address this, an integrated approach combining theoretical mechanical modeling, 3DEC numerical simulations, and in-situ microseismic (MS) monitoring was employed to investigate stress redistribution and fault slip dynamics. Field data and simulations reveal that mining advancement within a critical range of 15–20 m from the fault triggers concentrated MS events and peak energy densities (>10,000 J/m3), correlating with shear stress exceeding the critical slip threshold of 11.2 MPa. A logarithmic relationship between shear stress and mining duration is established to characterize the time-dependent stress accumulation and post-failure plateau. Furthermore, informed by Ripley's K-function spatial clustering analysis, a novel synergistic support strategy combining “Rib Anchor Cables + Roof Bolt-Mesh-Anchor” is proposed. Engineering application demonstrates that this active reinforcement system reduces roof displacement by 72%, effectively mitigating fault-induced instability. These findings elucidate the “unloading-induced slip” mechanisms of small-scale faults and provide actionable guidelines for stability control in tectonically complex coal seams.
Accurate lithology detection is a critical component of reservoir characterization. This study evaluates the performance of unsupervised machine learning algorithms—specifically K-means++ and Gaussian Mixture Models (GMM)—for lithology clustering using rigorously quality-controlled well log data from the SEAM dataset. We first assess clustering performance using 3-feature elastic data (Vp, Vs, and Density) and then investigate the impact of incorporating resistivity logs. Our numerical results demonstrate that GMM consistently outperforms K-means++ by better capturing data covariance and variability. For example, using elastic data in Well 3, GMM achieved superior cluster separation with a Silhouette score of 0.56 and a Davies-Bouldin Index (DBI) of 0.89, compared to K-means++ (Silhouette: 0.43, DBI: 1.12). Furthermore, integrating resistivity data significantly enhanced the discriminatory power for sand-shale mixtures, yielding the best overall clustering performance in Well 3 with a Silhouette score of 0.81 and a DBI of 0.61. Finally, independent validation using Rock Physics Models (RPMs) confirmed that the derived clusters are geologically realistic, falling accurately within established sand and shale trend boundaries. These findings highlight that combining GMM with multi-physics well logs (elastic and resistivity) provide a highly robust and geologically consistent framework for lithology classification.
The Neolithic settlement of Favella della Corte (Cosenza, southern Italy) provides an opportunity to investigate human-environment interactions during prehistory. To unravel the shallow paleo-environmental architecture of the site, we integrate geophysical investigations with sedimentological observations from an excavation trench. Geophysical datasets include a total-field magnetic-intensity areal survey, measurements of magnetic susceptibility collected along the excavation wall, and an electrical resistivity survey, performed with a low-density and sparse-electrode layout (six profiles). The available sedimentological observations identify a thin cultivated topsoil overlying silty sands and silty-sandy gravels. Because of the limited thickness of the topsoil, we neglect it in our analysis, adopting a first-order two-unit structure. This two-unit structure is also supported by susceptibility measurements, although some variability is present within the units. Additionally, the Total-Field Magnetic Intensity anomaly map reveals a linear feature and a terrace-like trend also consistent with such a two-unit susceptibility contrast. To estimate the first-order layer geometry and susceptibility contrast, we have inverted the total-field magnetic data by solving an optimization problem, employing a 3D prism-based interface model. The results allow us to interpret the linear feature and the terrace-like trend as a paleo-channel and an associated terrace system, possibly of anthropogenic origin. The presence of the paleo-channel is supported also by 2D electrical resistivity inversion models. These resistivity models provide additional constraints on the local hydrogeological setting, including a conductive horizon consistent with the water table. Coupling multiple geophysical techniques with targeted sedimentological analysis has proven to be an effective tool for paleo-environmental reconstruction at the Favella site.
Reservoir fluid identification is challenging in mature tight-sandstone wells where only a limited conventional log suite is available. This study presents a petrophysically guided workflow for classifying five reservoir interpretation classes from gamma ray (GR), spontaneous potential (SP), acoustic transit time (AC), and deep laterolog resistivity (Rt) logs in the Chang 4 + 5 reservoir of the Geng 60 block, Jiyuan Oilfield, Ordos Basin. The logs are converted into log-specific depth-frequency maps to retain local changes and interval-scale variations. Regionally calibrated acoustic porosity, a permeability proxy, and Archie-derived water saturation are incorporated as auxiliary petrophysical descriptors and as a training-stage soft regularization term, while a compact multi-scale context extractor integrates the available logging responses for classification. The dataset comprises 49 wells and 10,766 depth samples, divided by well into 34 training wells, 9 validation wells, and 6 predefined blind-test wells. The blind-test wells were excluded from model fitting, preprocessing-parameter estimation, and hyperparameter selection. On the six blind-test wells, the workflow achieved an accuracy of 92.24% and a Macro-F1 score of 0.898. The minority oil–water class attained an F1 score of 0.860. SHAP analysis showed that the predictions were associated mainly with Rt, AC, and GR, consistent with regional well-log interpretation. These results support the use of depth-frequency representation and locally calibrated petrophysical guidance for first-pass fluid interpretation in unseen wells within the study block under limited-log conditions. Application to other reservoirs requires local recalibration and independent validation.
The semi-airborne transient electromagnetic (SATEM) method has proven to be an effective geophysical tool due to its enhanced exploration capability compared to conventional ground and airborne methods. In clay and sulfide mineral deposits, the coupling of induced polarization (IP) effects and EM induction exhibits rapid decay at later times or even a sign reversal in EM signal. It is crucial to consider the IP effects in SATEM when doing the forward modeling. The frequency–time transformation method, which directly incorporates the Cole-Cole model in frequency-domain and offers excellent parallelization capability, has been widely adopted in studying IP effects. The SATEM method has a large number of receivers. To achieve high-accuracy simulation, fine meshes are used near the receivers for numerical method, which leads to enormous degrees of freedom (DoF). The direct solvers suffer from high computational complexity and significant memory consumption for 3D EM modeling problems. The flexible generalized minimum residual (FGMRES) iterative solvers require less memory and can employ the auxiliary-space Maxwell solver (AMS) preconditioner featuring significantly fast convergence behavior. However, it remains unclear whether this approach exhibits superior performance for modeling IP effects by using the Cole-Cole model. In this paper, we apply AMS method to solve the EM problem with IP effects for SATEM. We verify the numerical accuracy of the method by comparing our results with 1-D semi-analytical solutions for a polarizable half-space model. We further design a set of typical parameters to analyze the IP effects. Additionally, we also compare our method with direct solvers to reveal the efficiency of our iterative methods. Finally, we analyze the impact of IP effects on EM responses utilizing the complex model with topography. The proposed method demonstrates robust convergence across all tested models. The most significant advantage of our approach is the linear scaling of both computational time and memory consumption with respect to mesh size for SATEM with IP effects.
