Tailings thickening is a key unit operation in mineral processing, paste backfill, and tailings management systems, exerting a direct influence on water recovery efficiency, slurry transport behavior, and the stability of downstream dewatering and disposal processes. Throughout the entire thickening workflow, spanning free settling, compression settling, and high-concentration discharge, the evolution of rheological properties governs the formation of particle networks, the development of yield stress, and the resistance to flow and consolidation. These rheological responses are strongly coupled with particle size distribution, mineral composition, surface physicochemical characteristics, and operating conditions, leading to complex, stage-dependent thickening behavior. This review adopts a rheology-oriented perspective to analyze the thickening process across its distinct zones—clarification, free settling, hindered settling, and compression. Rheological behavior evolves from Newtonian to non-Newtonian and viscoplastic as solid concentration increases, affecting particle aggregation, settling, and network consolidation. Key parameters—shear yield stress, compressive yield stress, viscosity, and viscoelastic moduli—are central to understanding flocculation (e.g., Derjagin–Landau–Verwey–Overbeek, adsorption–bridging), settling, and compression. Integrating rheology with macroscopic models (e.g., Coe–Clevenger, Kynch, Buscall–White) and microscopic theories provides a unified framework for interpreting thickening mechanisms and guiding optimization. Advances in rheometry and online monitoring enable accurate slurry characterization, while numerical simulations incorporating rheology support the prediction of flow fields and solid–liquid separation. This review underscores the necessity of a rheology-based approach for designing flocculant dosing, optimizing equipment, and diagnosing failures. Future thickening technologies will rely on rheology combined with multi-scale modeling, intelligent monitoring, and artificial intelligence (Al)-based control for real-time regulation and sustainable tailings management.

The initial bearing age of a backfill is primarily determined by the roof collapse time, which critically affects its performance during the hydration stage. In this study, 96 h uniaxial compression creep and permeability tests were conducted to investigate the creep behavior of cemented backfill at different stress levels and initial bearing ages and their effects on strength and permeability. The results indicated that with an increase in the stress level, the cumulative axial strain of 28 d backfill increases from 0.962% to 1.647%. As the initial bearing age of the backfill increased, the axial strain curve in the creep deceleration stage gradually changed from concave to convex. Simultaneously, the strain rate curve of the backfill in the creep deceleration stage was characterized by a sawtooth-like fluctuation, and the creep time in this interval was shortened from 15–25 (3 d) to 0–10 h (28 d). Additionally, there were no evident macroscopic cracks on the surface of the backfill after the creep test, but its compressive strength and permeability exhibited significant differences. Static load exerted a significant strengthening effect on the 3-d-old backfill, with its compressive strength index increased by 19.95%, while the permeability index decreased by 30.90%. In contrast, static load caused damage to the backfill on the 28 d, leading to 31% decrease in compressive strength and 116.29% increase in permeability. Therefore, the stress level and initial load-bearing age had a significant impact on the mechanical properties of the backfill under static load. These findings provide valuable insights and practical references for on-site safety applications and strength design of backfills.

Backfill–rock composite structures (BRCSs) are crucial for the stability of underground mining areas. However, during the mining and backfilling cycles, they are subjected to coupled dynamic-static loading. Herein, to systematically investigate the mechanical properties of BRCSs under in situ mining and filling stress loading, true triaxial dynamic-static tests were conducted. First, the effects of the depth, cement–tailings (C/T) ratio by mass, and interfacial angle (IA) on the composite strength, deformation characteristics, and failure modes were systematically investigated. Subsequently, the evolution of acoustic emission (AE) signal parameters during BRCS failure was analyzed. Finally, a damage constitutive model was established based on the AE energy analysis. With increasing depth, C/T ratio, and IA, the peak strength and elastic modulus of the BRCS exhibited an upward trend, and the strain during the loading–unloading disturbance stages correspondingly increased. At a C/T ratio of 1:8, the specimens exhibited a rock-dominated load-carrying capacity with distinct brittle failure. Conversely, at a C/T ratio of 1:4, the specimens demonstrated a coupled backfill–rock load-carrying capacity, exhibiting ductile failure in the shallow regions and a transition to brittle failure in the deeper zones. AE signals were concentrated during loading–unloading disturbance, plastic yielding, and failure stages. The dominant failure mode was tensile-shear composite fracture, with the proportion of shear cracks gradually increasing with depth. The damage evolution process of a BRCS can be divided into three stages: initial, accelerated, and ultimate failures. This study provides an important theoretical basis and practical guidance for optimizing C/T ratio and enhancing stability assessment in backfilled mine designs.

The incorporation of supplementary cementitious materials (SCMs), such as blast furnace slag (BFS) and fly ash (FA), provides a promising strategy to optimize cemented paste backfill (CPB) by reducing operational costs, enhancing durability, and reducing the carbon footprint. This study systematically investigates the effects of partially replacing cement with BFS or FA on the autogenous self-healing behavior of the CPB system. Four binder blend ratios (i.e., cement/SCM mass ratio of 100/0, 80/20, 50/50, and 20/80) were evaluated based on crack closure observations, uniaxial compressive strength, hydraulic conductivity, and porosity-related parameters. Results show that the appropriate BFS contents promote self-healing efficiency at early and long-term healing stages compared with cement-only CPB, attributed to the secondary latent hydraulic reactions between BFS and calcium hydroxide, leading to the formation of calcium–(alumino)–silicate–hydrate (C–(A)–S–H) gels and microstructural densification. The pre-cracked Portland cement type I (PCI)/BFS 50/50 specimens exhibited the highest healing efficiency at 28 d of self-healing, with strength exceeding the uncracked control by 17.2% and hydraulic conductivity recovery reaching 81.8%, whereas at 90 d, superior long-term performance was observed for the PCI/BFS 80/20 mixture, achieving a 21.4% strength increase relative to the uncracked control and 96.2% recovery in hydraulic conductivity. In contrast, FA incorporation resulted in progressively reduced self-healing efficiency with increasing FA content due to its intrinsic physical characteristics and delayed pozzolanic reactivity. While the PCI/FA 80/20 mixture recovered strength comparable to the uncracked control, higher FA contents (50/50 and 20/80) exhibited significantly inferior strength recovery (13.3% and 23.2% lower than controls, respectively) and limited hydraulic conductivity recovery. Across both SCM systems, the formation and sufficiency of self-healing products, such as C–(A)–S–H, calcite, and ettringite, primarily govern the overall self-healing performance. These findings provide critical insights for designing CPB with improved autogenous self-healing capacity and optimized binder formulations for field applications.

Backfilling and grouting in the goaf are effective methods that can efficiently dispose of solid wastes including coal gangue (CG) and coal gasification slag (CGS). When backfilling is solely for solid waste disposal, the strength requirement for backfill materials is low. In view of this, a gangue and coal gasification slag-based backfill material (GCBM) was prepared, using a low content of alkali-activated slag (AAS) to adjust its mechanical properties. Considering three influencing factors (solid content, CGS content, and AAS content), single-factor experiments and optimization experiments based on response surface methodology (RSM) were conducted, with fluidity and strength as the optimization objectives. Finally, the hardening mechanism and microstructure of GCBM were analyzed. Test results show that the fluidity of GCBM is negatively correlated with solid content, CGS content, and AAS content; the strength is positively correlated with solid content (in a certain range) and AAS content, and first increases and then decreases with the increment of CGS content. The optimal mix-proportions obtained via RSM were as follows: 75.35wt% solid content, 24.13wt% CGS content, and 4.94wt% AAS content. Analysis of GCBM’s chemical composition and microstructure indicates that its main gel phases are calcium aluminosilic-ate hydrate (C–A–S–H), sodium aluminosilicate hydrate (N–A–S–H), and potassium aluminosilicate hydrate (K–A–S–H). The gels begin to adhere to and wrap inert solid particles when elements Si, Al, Ca, Na, and K dissolved in the alkali-activated system undergo heterogeneous nucleation on solid particle surfaces. The presence of multiple spherical pits on the fracture planes suggests that glass microspheres in CGS to some extent reduce GCBM’s strength. The specific surface area and pore structure of GCBM were analyzed, and its environmental safety was preliminarily verified. The results provide theoretical guidance for the large-scale, efficient backfilling disposal of solid wastes, especially CG and CGS.

High-volume fly ash (HVFA) binders are widely utilized as a mature method for cemented paste backfill in green mining, yet their performance remains highly sensitive to mix design. The fundamental coupling mechanism between the water-to-binder mass ratio (W/B) and sodium lignosulfonate (SL) content in pozzolan-rich HVFA systems remains insufficiently understood. In this study, HVFA pastes with varying SL contents (0–0.9wt%) and W/B (0.5–0.8) were characterized via rheometry, unconfined compressive strength (UCS) tests, and microstructural analyses, including zeta potential measurements. Results indicated that the absolute zeta potential magnitude increased from 11.88 to 27.08 mV as SL dosage rose from 0 to 0.9wt%, providing direct evidence for enhanced electrostatic repulsion. This surface modification significantly reduced yield stress and decreased the Relative Thixotropic Index (RTI) from 14.99% to 7.88% at a W/B of 0.5 with 0.3wt% SL. The effect of SL on 28-d UCS was non-monotonic, peaking at 35.72 MPa with 0.3wt% SL. The mercury intrusion porosimetry (MIP) analysis revealed a primary pore diameter shift from the harmful range (∼284 nm) to the refined range (183 nm), while X-ray diffraction (XRD) analysis confirmed enhanced calcium hydroxide consumption via pozzolanic reactions. The findings elucidate the dual role of SL as a physical dispersant optimizing particle packing and a chemical modulator governing hydration kinetics. These quantitative relationships provide a scientific basis for the performance-based design and intelligent pumping control of HVFA binders.

The microbial-induced carbonate precipitation (MICP) is a cementation and solidification method for sand with environmental advantages. However, the bonding performance of MICP for tailings with different particle sizes, as well as its applicability within conventional backfilling processes under varying cementing solution concentrations and bacterial addition levels, remains insufficiently understood. In this study, a uniform proportioning experiment considering the influence of cement–sand ratio (CSR), cementing solution concentration (CSC), and the volumetric ratio of bacterial solution to cementing solution (VRBC) on the uniaxial compressive strength (UCS) was conducted, and a series of microscopic analyses were used to demonstrate microbial mineralization behavior. Results show that the UCS of microbial blended tailings backfill (MBTB) exhibits a general trend of increasing and subsequently decreasing with rising CSC and VRBC. The UCS of optimally proportioned MBTB exceeds that of conventional backfill without microbial addition and maintains stable long-term strength. Comparative analysis indicates that a CSC of 0.5 mol/L and a VRBC of 1:1 yield the most effective microbial bonding performance. Although the cementing solution alone suppresses UCS, the subsequent incorporation of microbes significantly enhances strength, confirming the critical role of microbial mineralization and cementation within the backfill. The optimal UCS values for MBTB prepared with coarse and fine tailings are 2.44 and 1.55 MPa, representing increases of 49.69% and 23.03% relative to conventional backfill, respectively, demonstrating superior MICP efficiency in coarse-grained tailings. Microscopic analyses reveal substantial mineralized calcium carbonate distributed within the backfill, effectively filling interparticle pores, improving compactness, and enhancing mechanical behavior. Compared with microbial cyclic grouting approaches, the preparation of MBTB through conventional mixing not only ensures mechanical performance but also aligns more closely with practical backfilling operations, providing substantial engineering value.

With the increasing mining depth, heat hazards have become a critical challenge in deep underground operations. This study explores the incorporation of polyvinyl chloride (PVC) powder as a partial cement replacement in cemented backfill to improve thermal insulation and promote sustainable waste utilization. Five mix designs were prepared with 0, 5wt%, 10wt%, 15wt%, and 20wt% PVC, and their thermomechanical behaviors were systematically evaluated through uniaxial compressive strength (UCS) testing, thermal parameter measurements, energy evolution analysis, and microstructural characterization via scanning electron microscopy. The results showed that the UCS and energy absorption capacity first increased and then decreased with PVC addition, reaching an optimum at 10wt% PVC, which achieved an 87.5% higher strength and improved energy dissipation compared with the control. The thermal conductivity and specific heat capacity progressively decreased with increasing PVC content, with the maximum reductions of 23.0% and 40.2%, respectively, for 20wt% PVC. Microstructural analysis confirmed that moderate PVC addition reduced the porosity and enhanced the internal compactness, whereas excessive PVC likely inhibited calcium silicate hydrate gel formation and weakened the structural integrity. A PVC dosage of 10wt% was identified as the optimal replacement level, providing a favorable balance between strength and thermal insulation. This study provides new insights into sustainable backfill design and offers a practical strategy for mitigating thermal hazards in deep mining environments.

In conventional cemented paste backfill (CPB), ordinary Portland cement (OPC) is the primary binder; however, it has drawbacks such as high costs and carbon emissions, and low durability. Granulated blast-furnace slag, a byproduct of ironmaking, has emerged as a promising sustainable additive. In this review, three slag-based binders—slag–cement blends (SCB), alkali-activated slag (AAS), and alkali-sulfate-activated slag (ASAS)—are discussed, focusing on their hydration mechanisms, rheological characteristics, mechanical properties, microstructure, sulfate resistance, and heavy metal solidification capabilities. SCB–CPB exhibits enhanced fluidity and late-stage strength compared to OPC–CPB, albeit with reduced early-stage strength. Although AAS exhibits superior comprehensive properties, its application is hindered by the high cost and corrosiveness of alkali activators. In contrast, ASAS emerges as a balanced solution, offering early- and late-age strength, second only to AAS, while being the most cost-effective and lowest-carbon option. Moreover, the future prospects of slag-based binders in CPB are discussed, providing valuable guidance for their formulation and application. These findings offer valuable insights for the further development and implementation of cost-effective and environmentally friendly slag-based binders in CPB applications.

The sustainable management of coal-based solid waste and effective CO₂ sequestration are critical challenges for the mining industry. To address this, a novel aluminum nanoparticle-modified CO₂-carbonated backfill (ANCB) material was developed that synergistically enhances mechanical properties with carbon capture functionality. The effects of varying aluminum nanoparticles (Al-NPs) concentrations (0.02wt%–0.1wt%) were investigated on the unconfined compressive strength (UCS), CO₂ adsorption capacity, hydration and carbonation reactions, and microstructural evolution of the ANCB. Results demonstrate that low-dose Al-NPs (0.02wt%–0.06wt%) enhance UCS by up to 73.8%, with early-age strength development accelerated by promoting nucleation and hydration kinetics. Notably, CO₂ adsorption was strongly correlated with this strength gain, with an optimal concentration of 0.06wt% achieving a balanced enhancement of both properties. Microstructural and spectroscopic analyses (scanning electron microscopy coupled with energy dispersive spectroscopy (SEM-EDS), X-ray diffraction (XRD), and Fourier transform infrared spectroscopy (FTIR)) revealed that Al-NPs promote the formation of a denser calcium aluminosilicate hydrate (C–A–S–H) gel, facilitate CaCO₃ precipitation, and release reactive [Al(OH)₄]⁻ ions that drive pozzolanic reactions. However, excessive Al-NPs led to agglomeration and microstructural heterogeneity, impairing performance. This study establishes the dual role of Al-NPs in advancing the multifunctionality of backfill materials. The ANCB system thus presents a scalable, carbon-negative strategy for underground mining, effectively bridging the gap between waste valorization and in-situ CO₂ sequestration.

Mine backfilling is a critical geotechnical operation for underground stability and waste management, while its environmental performance is increasingly constrained by the high carbon footprint of cementitious binders, particularly ordinary Portland cement (OPC). This study investigates the use of CO₂-mineralized steel slag as a reactive supplementary binder for mine backfill to develop a carbon-reducing mine backfill (CRMB), while simultaneously enhancing mechanical performance and reducing carbon emissions. The results show that moderate carbonation (∼50% carbonation degree) significantly improves backfill performance, with CRMB2 achieving a 24.5% increase in 28 d compressive strength compared with the uncarbonated system and slightly outperforming OPC under identical conditions. Mechanistic analyses demonstrate that CO₂ mineralization induces the formation of highly reactive CaCO₃ and silica gels, which reprogram hydration pathways by accelerating silicate and aluminate reactions. Furthermore, the availability of carbonate species promotes the formation of stable carboaluminate phases, contributing to sustained strength development at later ages. Nanoindentation also confirms that both low- and high-density calcium silicate hydrate (C–S–H) phases exhibit enhanced stiffness at moderate carbonation levels. From an environmental perspective, CRMB reduces embodied CO₂ emissions by up to ∼60% relative to OPC, resulting in superior strength-to-emission efficiency and demonstrating its effectiveness as a carbon-sequestering supplementary binder for mine backfill.

The synergistic CO₂ sequestration via solid waste backfilling in goafs can simultaneously address the issues of CO₂ emissions, accumulation of coal-based solid wastes, and safety hazards in goafs under China’s coal-dominated energy structure. In this study, a modified magnesium-coal-based all-solid-waste carbon-sequestering backfill material (MFCC, prepared from modified magnesium slag (MMS), fly ash (FA), coal gangue (CG), and coal gasification slag (CGS)) was fabricated. The fluidity of the fresh slurry was characterized using the mini slump test, and its carbonation curing performance was investigated via uniaxial compressive strength (UCS), carbonation depth (CD), X-ray diffraction (XRD), scanning electron microscopy (SEM), thermogravimetry-differential thermogravimetry (TG-DTG), and computed tomography (CT) tests, aiming to achieve the synergistic goals of high-value utilization of solid wastes and CO₂ sequestration. The results indicate that the fresh MFCC slurry exhibits excellent fluidity with a mini slump ranging from 121.5 to 135 mm. The fluidity increases with the rise in CGS content, which fully meets the requirements for industrial pipeline pumping. During the carbonation curing process, the UCS of the material increases continuously with the extension of curing age, with the 28-d UCS ranging from 7.36 to 8.71 MPa, which fully meets the strength design requirements for coal mine backfilling engineering. Microscopic analyses reveal that the filling and cementation effects of hydration and carbonation products on pores render the material’s microstructure denser, significantly reducing pore volume and connectivity, which is the key reason for the strength improvement. After 28 d of carbonation curing, when the CGS content is 20wt%, the UCS reaches a maximum value of 8.71 MPa, and the CO₂ uptake also attains a peak of 13.94%. In summary, after carbonation curing, the MFCC material not only exhibits excellent mechanical properties but also enables the simultaneous realization of resource utilization of solid wastes and efficient CO₂ sequestration, thus holding broad application prospects in backfilling engineering.

Kimberlite beneficiation is constrained by high circulating loads, inefficient liberation, and substantial fine waste generation. This study evaluates the effectiveness of microwave pre-treatment in improving comminution and dense media separation performance by subjecting coarse (16–31.5 mm) and overall (6.7–31.5 mm) feed classes to controlled microwave irradiation using an industrial 2.45 GHz, 15 kW system. Treated samples were subsequently crushed in a single-roll crusher and processed through a laboratory dense media separation workflow to quantify changes in product size distribution, concentrate recovery, circulating load, and waste generation. Microwave pre-treatment produced consistent improvements in downstream performance. At low energy inputs, concentrate mass fraction increased by up to 75% for the coarse feed and 68% for the overall feed, indicating enhanced liberation of dense mineral phases. Circulating load decreased by 6%–10%, reducing the volume of material requiring reprocessing and lowering the potential for diamond damage during repeated crushing. Waste (<1 mm) and fine waste (<0.3 mm) generation declined by 5%–10%, demonstrating that microwave exposure can suppress excessive fines production and support more sustainable tailings management. Overall, the results show that controlled microwave pre-treatment can enhance mineral liberation, improve separation outcomes, and reduce waste generation, representing a promising pathway for increasing the efficiency and sustainability of kimberlite beneficiation.

Large-scale artificial intelligence (AI) models are increasingly shaping safety, efficiency, and sustainability in the mining industry. This paper reviews the development, applications, and challenges of domain-specific large AI models in coal mining. These models integrate heterogeneous multimodal data—text, images, video, audio, design data, point clouds, and time series—within multi-layered architectures encompassing infrastructure, data resources, algorithms, application services, and security. Application platforms supporting knowledge services, visual analysis, and intelligent scheduling demonstrate practical improvements in operational decision-making. Despite these advances, deployment faces challenges including fragmented data, limited labeled datasets, few-/zero-shot scenarios, industry-specific adaptation, robustness and interpretability, weak causal reasoning, edge computing limitations, cost–benefit trade-offs, and compatibility issues. Overcoming these barriers requires coordinated progress in data governance, model design, and industry standardization.

The extractive metallurgy sector is undergoing rapid digital transformation driven by Industry 4.0, advanced sensing, and artificial intelligence (AI). While machine learning has been widely adopted for predictive control and optimization, the role of generative artificial intelligence in metallurgical engineering remains inadequately characterized in the literature. This paper critically reviews the state of generative AI for extractive metallurgy, focusing on practical industrial applications rather than purely theoretical AI methods. We synthesize peer-reviewed research, industrial case studies, and emerging applications across comminution, flotation, hydrometallurgy, pyrometallurgy, ore sorting, and plant reliability. The review identifies five key generative AI models applicable to metallurgy: generative diffusion models, flow-based models, variational autoencoders, generative pre-trained transformers, and generative adversarial networks. Generative AI presents a transformative opportunity for extractive metallurgy, offering solutions for optimized process control, enhanced mineral recovery, predictive maintenance, and improved sustainability. While generative AI offers significant potential, its deployment requires rigorous validation, physics-informed modeling, and hybrid human AI workflows in metallurgical plants.

Driven by the global energy transition and industrial intelligence, the mining industry is evolving towards smarter and more efficient methods. In mineral processing, particularly flotation, traditional techniques rely heavily on human experience, facing challenges due to complexity and variability. This study proposes an intelligent control system based on machine vision for spodumene flotation. It introduces an improved YOLOv11-M model with real-time foam detection and decision optimization, enhancing flotation efficiency. The research utilizes a dataset of over 100000 foam images and deep learning to detect foam states. Innovations include using EfficientNetV2 for feature extraction, the C3k2_LGP module for enhanced frequency perception, and the Saga-PIoU loss function for better robustness under complex conditions. Experimental results show improvements in mean average precision (mAP) (by 1.9%), precision (0.3%), and recall (2.5%). YOLOv11-M outperforms other models, with a significant frames per second (FPS) increase (135.3) and improved accuracy. A semi-industrial trial demonstrated YOLOv11-M’s ability to enhance flotation recovery and grade. Not only did the grade improve, but the flotation process’s stability was also significantly enhanced. The foam velocity distribution became more reasonable, and the fluctuations in grade and recovery were significantly reduced, with standard deviations decreasing by 65% and 90%, respectively. These findings indicate that YOLOv11-M not only improves the efficiency and stability of the flotation process but also provides an intelligent, automated solution for the industry, with the potential for widespread application in large-scale mining flotation processes. The open-source code and dataset will be released at: https://github.com/users/ytyyty368-arch.

Unsafe driving behaviors significantly contribute to accidents in open-pit mining operations. Conventional monitoring systems often fail to capture the temporal continuity and semantic ambiguity of such behaviors. To address these challenges, an event-level abnormal driving behavior recognition method based on a vision-language model (VLM) is proposed. The method integrates Temporal-Aware Low-Rank Adaptation (T-LoRA) with event-level supervised fine-tuning to enable efficient vertical-domain customization of the VLM toward complex open-pit operating environments. By explicitly enhancing temporal modeling of visual sequences, the proposed method enables accurate event-level recognition of abnormal driving behaviors, precise localization of their onset and duration, and the generation of interpretable semantic descriptions. Experiments conducted on a real-world mining truck dataset demonstrate that the proposed approach achieves an event-level F1-score of 0.923, while maintaining a mean absolute error for temporal localization below 0.8 s. These results indicate that the proposed method effectively captures continuous behavioral patterns and improves both recognition accuracy and temporal precision, providing a reliable solution for intelligent safety monitoring. Furthermore, the system supports proactive intervention through event-level early warnings, contributing to safer and more efficient mining operations.

Accurate groutability prediction is essential not only for mine water-inrush prevention, but also for reducing excessive cement consumption and the associated carbon footprint of grouting operations. However, field geological datasets are often small, which limits the reliability and generalization of data-driven models. A geology-informed prediction framework is presented for predicting unit grouting amount (uga) from three routinely measured borehole variables: water inflow rate, hydraulic pressure, and groundwater level. A generative augmentation strategy was employed to expand the training data from 44 to 880 samples, and the input variables were organized into a physically informed feature sequence. Based on this representation, a hybrid deep-learning model integrating a bidirectional temporal convolutional network, a bidirectional gated recurrent unit, and an attention mechanism was developed, with its hyperparameters optimized using the Crested Porcupine Optimizer. The results demonstrate that data augmentation improved the test performance, with the coefficient of determination (R²) increasing from 0.8851 to 0.9293 and the root mean square error (RMSE) decreasing by 21.6%, effectively alleviating overfitting. External validation yielded R² values of 0.9982, 0.9884, and 0.9272 under identical, similar, and distinct geological settings, respectively. Distribution-shift analysis further highlighted the importance of hydrogeological-mechanism consistency for external generalization. Interpretability analysis using SHapley Additive exPlanations and Accumulated Local Effects indicated that borehole water inflow rate was the dominant predictor, contributing 75.4% to the model output, while high unit grouting amounts were associated with jointly elevated fracture connectivity and hydraulic pressure. The proposed framework provides practical support for accurate uga prediction and material-efficient grouting in intelligent mining.

The difficulty in accurately quantifying fracture features at underground excavation faces constrains stope blastability evaluation and refined blasting operations. An intelligent workflow—fracture recognition, indicator quantification, blastability classification, and scheme matching—was developed by integrating a You Only Look Once (YOLO)-based segmentation model with a comprehensive cloud model. Multimodal data augmentation was designed to emulate harsh underground imaging conditions, expanding 119 labeled images to 474 for model training. With DeepLabV3 and YOLOv8l-seg as reference models, YOLOv5-seg variants were benchmarked using Dice, pixel accuracy (PA), mask intersection over union (Mask IoU), and box intersection over union (Box IoU). YOLOv5x-seg was then selected and further enhanced by replacing PANet with a bidirectional feature pyramid network (BiFPN) and embedding a convolutional block attention module (CBAM), yielding YOLOv5x-CBF with consistent metric gains. Fracture trace length was quantified within the effective face region and integrated with powder factor and blasting advance. These indicators were then used to construct a comprehensive cloud model for uncertainty representation, three-class blastability classification, and “class–scheme” mapping. Field-scale trials on two stopes demonstrated improved blasting performance, with blasting advance increased by 18%–30% and powder factor reduced by 8%–13%. These results confirm the practicality of the proposed method and provide a field-deployable technical reference for intelligent blasting in underground metal mines.

Rockburst has become a major hazard constraining safe production and high-quality capacity release in China’s coal mines. During deep mining of near-vertical seams within the Tianshan seismic belt, the coupling of nonlinear coal-rock deformation responses with complex geological conditions markedly elevates rockburst risk. To meet the strategic demand for intelligent, safe, and efficient mining in rockburst-prone seams, this study integrates geophysics, spatial statistics, big data mining, and deep learning to investigate a steeply dipping coal mine in Xinjiang, China, and systematically analyze the relationship between microseismic activity parameters and mining-induced disturbances. On this basis, a temporal fusion feature identification method for microseismic indicators is proposed. By embedding temporal constraints into a deep-learning framework, an enhanced temporal fusion transformer (TFT) is developed to predict multiple microseismic indicators. Furthermore, an intelligent rockburst prediction and early-warning approach driven by fused microseismic parameters is established and validated in field applications. Results indicate that hazard risk in the sandwiched rock pillar and the B₆ roof areas increases with working-face advance, and the localized damage in the sandwiched rock pillar is more severe than that in the B₆ roof. To strengthen feature extraction, a WFTBlock is introduced by combining continuous wavelet transform, Fourier transform, and timestamp alignment to reveal the periodic evolution of spectral and phase characteristics in indicator sequences. The final multi-parameter TFT model is trained jointly with fused features and temporal inputs. Compared with the long short-term memory (LSTM) baseline, the proposed model reduces root mean square error (RMSE) by 47.9% and improves coefficient of determination (R²) by 54.5%, demonstrating substantially enhanced predictive accuracy. Overall, the proposed framework provides technical support for safe and efficient mining of steeply dipping seams and the secure development of key energy bases along the Belt and Road Initiative.

The acoustic emission (AE) Kaiser effect method is widely used for in-situ stress measurements because of its nondestructive nature, operational efficiency, and low cost. A key step in this method is the identification of the Kaiser point. However, traditional manual approaches require further improvement, indicating the importance of developing intelligent identification methods. In this study, an intelligent Kaiser point identification method was proposed based on a dual-branch gated recurrent unit (GRU) deep learning framework and phase-space reconstruction (PSR). In the proposed framework, AE waveform data was processed via PSR and principal component analysis to generate chaotic feature representations, which are then fused with the original waveform data through a dual-branch GRU architecture for classification. The classification results were then used for Kaiser point identification. The proposed model achieved an accuracy of 90.7% and an area under the curve of 0.9357 on the test set, indicating good discrimination ability between the Felicity and Kaiser areas. Compared with representative deep learning baseline models, including long short-term memory (LSTM), bidirectional LSTM, temporal convolutional network, and Transformer, the proposed model exhibited superior overall performance. The ablation results confirmed the positive contributions of both the original waveform and chaotic feature branches. The interpretability analysis revealed that the model relied primarily on a limited number of salient waveform segments and chaotic feature groups, both of which were physically relevant. Compared with the traditional manual method, the proposed method identifies the Kaiser point more stably and accurately. The application of the method to a porphyry copper mine in western China demonstrates its practicability in engineering applications.

The prediction of rock failure, a key fundamental research for addressing mining safety issues (such as mine slope stability and rockburst), faces challenges with traditional methods due to their complex generalization and computational processes that struggle to describe the entire failure process. Consequently, 12 prediction models integrating ensemble learning and optimization algorithms were established to predict rock peak stress and failure time using strain, elastic modulus, density, mass, and confining pressure as inputs. Five-fold cross-validation was used to optimize hyperparameters, significantly improving the model’s generalization ability, robustness, and stability. Dataset was established through rock mechanics experiments, with strain increments configured at 0.008‰, 0.01‰, and 0.012‰ in the test set. The cross-validation optimized particle swarm optimization eXtreme gradient boosting (CV-PSO-XGBoost) model performed best under a strain increment of 0.01‰, and its stress prediction achieved coefficient of determination R² = 0.904, mean absolute error (MAE) = 4.315, and root mean square error (RMSE) = 5.435; while the failure time prediction demonstrated R² = 0.811, mean absolute percentage error (MAPE) = 7.842%, and MAE = 30.343. Finally, SHapley Additive exPlanations (SHAP) analysis showed strain and stress significantly impact the model, with strain positively predicting failure time, aligning with traditional rock failure models, validating reliability. This study provides insights into the research on rock strata stability in mining.

To study the deep rock strength, this paper proposes a five-parameter deviatoric function to modify the deviatoric function of the Hoek–Brown (HB) criterion, introduces an intelligent optimization algorithm (IOA) to determine the material parameters, thereby constructing a modified three-dimensional (3D) HB criterion, namely MMCHB criterion. The MMCHB criterion avoids the defects of the traditional HB criterion, which neither considers the Intermediate principal stress (IPS) nor meets the smoothness requirement, and overcomes the shortcomings of parameter determination based on conventional methods, which can lead to a single deviatoric plane envelope shape. This modified criterion can be degenerated into the HB criterion under triaxial compression and tension. The proposed criterion is verified using true triaxial test data for six types of intact rock, and the modified 3D HB criteria are selected for comparative study. The results show that the proposed criterion under the IOA has the best prediction error for the six rock types, ranging from 1.6636% to 3.4023%. Overall, the MMCHB criterion outperforms the existing modified 3D HB criteria in prediction. Based on the proposed MMCHB criterion, an intelligent prediction system is developed, which provides a new approach for intelligent prediction of deep rock strength and dynamic construction of rock material parameters.