Aiming at the engineering problems such as insufficient crack detection accuracy and limited real-time performance in the complex geological environment of underground mine tunnels, an underground mine tunnel crack-segmentation network (UMTC-net) integrating multi-scale feature perception and adaptive attention mechanism was proposed. This network can realize cross-scale feature extraction of crack images from local texture to global structure through hierarchical cascading of Swin Transformer module groups. Meanwhile, a scaling cosine attention mechanism encoded by relative positions in logarithmic space was introduced to effectively suppress the interference of abnormal pixels. In addition, a codec framework based on dynamic patch merging/expansion was constructed, which solved the problems of ambiguous boundary positioning of fine cracks and high false detection rate in complex backgrounds in traditional methods. The results show that the UMTC-net has an accuracy of 85.15%, an average intersection-union ratio of 85.78%, and an F₁ value of 83.27% in the Crack 500 dataset, and an accuracy of 87.51%, an average intersection-union ratio of 79.98%, and an F₁ value of 86.95% in the MineTunnelCrack-2000 dataset. It exhibits stronger robustness, especially in low light and high dust environments. This network achieves an inference speed of 38.9 ms on the RTX 3060 mobile graphics card, occupying only 5 230 MB of memory and reducing deployment costs by more than 40%. It meets the real-time and low-power requirements of portable detection devices, and has a higher cost-effectiveness for adaptation. In the field test, the detection efficiency of UMTC-net is 8 times higher than that of manual inspection, and the missed detection rate is reduced from 18% to 3.2%. The research results provide an efficient and accurate new scheme for crack detection in underground mine tunnels, which is helpful to find potential safety hazards in time and ensure the safety of mine production and stable operation of equipment.

Fiber-reinforced wet-mix shotcrete has excellent properties such as deformation resistance and crack resistance. To evaluate the support performance of fiber-reinforced wet-mix shotcrete, a series of tests were conducted, including uniaxial compressive tests, notched beam flexural toughness tests, and disk flexural tests. The test results indicate that, although the addition of fibers slightly reduces the compressive strength of wet-mix shotcrete, the flexural strength and energy absorption capacity have been significantly enhanced. The roadway support test results shows that the compressive strength of the steel fiber wet-mix shotcrete can reach 25 MPa, the thickness of the spray layer is maintained at 100−150 mm, with a minimal rebound and notable support effectiveness.

Abandoned mine not only contains a large number of available resources, but also has many risk factors. In order to scientifically determine the reutilization mode and take into account the risk of many abandoned mines, a risk framework of abandoned mines including six dimensions of technical risk, safety risk, environmental risk, community risk, legal risk and financial risk was constructed, and the risk assessment of reutilization was carried out by LEC risk assessment method. Then, from the perspective of risk management, the method combining improved analytic hierarchy process (IAHP) and TOPSIS was proposed to determine the best reutilization mode of abandoned mines. Finally, taking the closed mine of Muchengjian Coal Mine in Jingxi Mining Area as an example, the proposed method was verified. The results show that the high risk factors of the closed mine are mainly concentrated on safety risk and social risk, and ecotourism is selected as the optimum reutilization mode. This risk management-based method can provide a reference for optimizing the reutilization mode of abandoned mines and effectively reduce the negative effects of mine closure.

In order to explore the potential value of a large amount of safety hazard data in the construction process of intelligent mines, taking a mine in Shandong as an example, comprehensive analysis of its historical safety hazard data from 2014 to 2023 was conducted, and a multidimensional analysis model for mine safety management was constructed. Firstly, a Multi-Layer Perceptron (MLP) was used to construct a personnel, equipment, and environmental classification model for identifying hazards and accidents. Using the Latent Dirichlet Allocation (LDA) topic model, equipment hazards were classified into eight topics of lighting, transportation, support, electrical, firefighting, blasting, ventilation, and miscellaneous. Then, based on the principle of Apriori algorithm, key information was extracted from unstructured hazard text, and the relationship between different hazard features and topics was explored and analyzed. Finally, deep analysis of the data mining results was conducted using a combination of multidimensional analysis and data visualization techniques. The results indicate that equipment related hazards are high-risk areas that require special attention in the safety management of the mine. The lack of support for the roof, potholes on sloping road surfaces, and installation of switch grounding are significant hazard topics and associated rules, and the areas such as S16181 and S18165 are gathering areas for this type of hazard. The multidimensional analysis model constructed by the research can provide a basis for the analysis of mining safety hazards.

In order to monitor the chlorophyll content of plants quickly and non-destructively, two different microbial reclamation treatments (inoculation group and control group) were set up, and six herbaceous plants (Astragalus adsurgens, Medicago sativa, Leymus chinensis, Agropyron cristatum, Elymus sibiricus, Bromus inermis) were selected according to four kinds of mixed sowing ratios (1∶1, 1∶2, 1∶3, 2∶1). The chlorophyll content and spectral reflectance of Astragalus adsurgens in the test area were measured respectively. Using the original spectrum, the logarithm of the reciprocal of the original spectrum, and the first-order differential, combined with three modeling methods of BP neural network regression, support vector machine (SVM) regression, and random forest (RF) regression, models were established for plant spectral characteristic curves under different treatments. The results show that the inoculation treatment increases the chlorophyll content, and the chlorophyll content is also different under different mixed sowing ratios. Compared with the original spectral curve, the modeling accuracy of the reciprocal logarithm and first order differential of original spectral is improved to varying degrees, and the modeling accuracy of FDR is the best. Under the condition of microbial reclamation, the RF regression model has the highest accuracy. Under the conditions of different planting ratios, the model established by BP neural network regression in the 1∶2 and 1∶3 regions of legumes has high accuracy, while the spectral samples in the 1∶1 and 2∶1 regions are more suitable for using RF regression method.

As one of the key processes in the stage open stope with subsequent filling mining method, the quality of the slot raise construction directly affects the effect of the subsequent blasting to form the cutting groove, and ultimately affects the quality of large-scale caving mining. In view of the problems of low intelligence, high safety risk and large occupational injury in the construction of traditional artificial sight distance operation raise boring machine, combined with the mining status and occurrence conditions of porphyry ore body in Shanxi Zijin, the existing construction mode of slot raise was optimized to realize the intelligent construction of slot raise. A dynamic optimization control technology for operation parameters was proposed, which used the mountain climbing method to achieve dynamic parameter optimization, and achieved autonomous and efficient continuous drilling through a remote intelligent control system based on 5G communication technology. The industrial application results show that the raise boring machine with remote intelligent control system can also maintain high-precision hole drilling and hole expansion operations in complex rock formations with fractured zones, with an average pilot hole footage of 1.75 m/hand an average reaming footage of 1.375 m/h, and a deviation rate of only 0.68%. The study can provide a reference for unmanned construction of wellbores in other scenarios.

The existing object detection algorithms for shaking table concentrate bands have problems such as inability to balance detection accuracy and speed, high computational costs, difficulty in compressing model size, and slow inference speed. To address these problems, a lightweight fusion network for shaking tables (YC-Lightweight Net) object detection algorithm was proposed. The YC-Lightweight Net model firstly used a repetitive visual transformation network to extract features from the images of shaking table sub-banding. Then, by introducing group space convolution, multi-scale efficient cross stage fusion modules, and using skip connections, an efficient and lightweight neck network was designed. Finally, a weight based layer adaptive pruning algorithm was used to compress the model size. The experimental results show that the accuracy, recall, mean average precision, and FPS indicators of the YC-Lightweight Net model are 98.4%, 97.9%, 98.8% and 333 frame/s, respectively. The detection accuracy and speed are significantly better than those of the compared models. The number of parameters, floating-point operations, and model size after pruning are 13.9%, 15.4% and 17.5% of the original model, respectively. The pruning operation greatly reduces the computational complexity and model size of the model. The YC-Lightweight Net model has good detection accuracy and real-time performance, meeting the requirements of industrial equipment for lightweight models in shaking table mineral processing plants. The study can provide a technical support for accurate identification of separation points in mineral bands and intelligent upgrading of the shaking table mineral processing plant equipment.

In order to explore the influences of different cutting parameters on the cutting performance of the oscillating cutting disc, the Discrete Element Method was used to simulate the cutting process of the oscillating cutting disc, and the influences of eccentricity distance, oscillating frequency, feed rate and cutting depth on the cutting performance were studied. The results show that the average load of rock breaking with the oscillating cutting disc is obviously lower than that of the non-oscillating cutting disc. When the feed rate is 60 mm/s, 90 mm/s, 120 mm/s, 150 mm/s and 180 mm/s respectively, compared with the non-oscillating condition, the average load of the cutting disc under the oscillating condition is reduced by 37.37%, 44.19%, 57.47%, 60.32% and 61.25% respectively. Under the same condition, with the increase of eccentricity distance, the average load decreases firstly and then tends to be stable. With the increase of feed rate, the average load decreases gradually, and the maximum load increases gradually and then tends to be stable. When the feed rate is less than or equal to 90 mm/s, the average load increases with the increase of the oscillating frequency. When the feed rate is greater than 90 mm/s, the larger the oscillating frequency, the smaller the average load. When the cutting depth is 40 mm, the average load and the maximum load are the smallest. When the eccentricity distance, oscillating frequency, feed rate and cutting depth of the oscillating cutting disc are 3 mm, 60 Hz, 150 mm/s and 40 mm, the cutting performance is the best. The research results can provide a reference for the determination of the cutting parameters of the oscillating cutting disc.

To reveal the energy consumption patterns of irregular ore particles under impact crushing, impact crushing tests were conducted on six types of iron ore with different properties, and the fractal characteristics of irregular iron ore fragmentation were analyzed, as well as the size effect on average particle size of fragments, fractal dimension, and unit absorption energy. Then, the energy consumption model of irregular single particle crushing was established. The results show that the average particle size of fragments, fractal dimension, unit absorption energy all exhibit a power function relationship with initial isosphere diameter under the same impact conditions. As the isosphere diameter of the particles increases, the average particle size of the fragments gradually increases, and the fractal dimension and unit absorption energy gradually decrease. There is an increasing relationship with power function between unit absorption energy and average particle size of fragments. The relationship between the logarithm of unit absorption energy and fractal dimension shows a linear increase. Through the method of ore crushing tests of irregular single ore particle, relationship models of unit crushing energy consumption of irregular iron ore particles changing with the initial size, average particle size of fragments, and fractal dimension were established, with an average correlation coefficient of 0.789, which effectively describes the energy consumption pattern of ore crushing.

With the continuous progress of mineral resources exploration technology, the intelligent identification of rock minerals has become increasingly important in the field of mineral composition analysis. In order to analyze the influence of complex texture structure and variable mineral morphology of rock thin section images on intelligent identification technology, an intelligent identification model of rock minerals based on improved YOLOv8 algorithm (Mineral-YOLO model) was proposed. The Mineral-YOLO model innovatively integrates the LSK module to enhance the identification capacity of the model for different target and background information differences. The ODConv technology is introduced to reduce the influence of background interference, thereby improving the performance of the convolutional network. The loss function is optimized to improve the accuracy mAP of bounding box positioning. In the model training, the self-built data set was extended using the combination enhancement technology, so that the samples of the data set were more abundant. The validation set was used to verify the trained model. The results show that the mean average accuracy of the proposed mineral intelligent identification model is 83.3% and F₁ is 78% when identifying 6 kinds of minerals. Compared with the YOLOv8 model, it is increased by 3 percentage points and 1 percentage points respectively, which proves the high efficiency and accuracy of the Mineral-YOLO model in the intelligent identification of rock minerals.