In order to evaluate the implementation effect of safety management in non-coal underground mines, a safety performance evaluation index system of non-coal underground mines was established, including 6 first-level indicators and 30 second-level indicators. The safety performance evaluation method of non-coal underground mines based on ANP method and TOPSIS method was constructed. The weight of evaluation index was determined by ANP method, and the comprehensive evaluation was carried out by TOPSIS method. Finally, XD company was taken as an example for empirical research. The evaluation results show that the safety performance score of XD company in 2023 is the highest, and the safety performance is grade Ⅱ. The average annual growth rate of safety performance comprehensive score from 2020 to 2023 is 5.36%. XD company is relatively good in safety education and training, and relatively weak in risk classification control and hidden danger investigation and management. The evaluation results can provide guidance for the safety management of mining enterprises.

In order to study the effect of mineral composition content on Microcrack propagation and static mechanical parameters of granite, a numerical model considering the meso-components of rock was established by combining thin section identification analysis and particle flow discrete element software PFC2D. The mesoscopic parameters of the model were calibrated by static uniaxial compression tests, and the effects of mineral component content on microcrack propagation, compressive strength and elastic modulus of granite were discussed. The results show that with the increase of quartz content, the failure mode of rock changes from “y” -shaped shear failure to “X” -shaped shear failure. With the increase of potassium feldspar content, the failure mode of rock changes from oblique section shear failure to “y” -shaped shear failure. With the increase of plagioclase content, the failure mode of rock changes from “V” -shaped shear failure to “X” -shaped shear failure. Biotite has the greatest influence on the failure mode of rock, with the increase of biotite content, the failure mode of rock changes from “V” cross shear failure to monoclinic shear failure, and then to composite shear failure. The elastic modulus of granite is negatively correlated with potassium feldspar and plagioclase, and positively correlated with quartz and biotite. Biotite has the greatest influence on compressive strength of granite, followed by plagioclase, then quartz, and potassium feldspar has the least influence. The Stiffness Heterogeneity Factor (SHF) is further introduced to analyze the relationship between mineral composition content and rock mechanical parameters. It is found that the peak stress and elastic modulus of rock decrease linearly with the increase of SHF.

The engineering stability of jointed rock masses in cold regions is often affected by freeze-thaw cycles damage, which leads to geological hazards. The core control factor is the deterioration of shear strength of rock mass. In response to the engineering background of the coupled effects of freeze-thaw cycles and shear loads faced by jointed rock masses in cold regions, existing methods for calculating the shear strength of freeze-thaw joints have certain deviations. Therefore, a modified shear strength criterion was proposed that accounts for the strength degradation of jointed rock mass under freeze-thaw cycles. A systematic experimental study on freeze-thaw-shear coupling was conducted using similar material simulation methods. Based on direct shear test results of freeze-thaw joints, the degradation mechanism of rock bridges under freeze-thaw-shear coupling action was revealed as follows. The shear strength of non-persistent jointed rock mass consists of three components, including the shear strength of non-degraded rock bridges, the shear strength of degraded rock bridges, and the shear strength of the joint surfaces. A coupled damage model for rock bridges was established with shear modulus as the variable, achieving quantitative characterization of the degradation behavior of rock bridges. Building upon this foundation, the modified Jennings shear strength criterion demonstrates the following advantages: the mean relative error between theoretical predictions and test results is merely 3.28%, representing a significant improvement in prediction accuracy compared to the original Jennings criterion. This modified criterion effectively captures the strength characteristics of non-persistent jointed rock mass under freeze-thaw action, thereby providing a reliable theoretical basis for studying the freeze-thaw damage mechanisms and safety protection of jointed rock masses in cold regions.

Based on the working condition, structure composition and working principle of the mine unanchoring robot, the whole machine model was established by using the three-dimensional software SolidWorks, and the variation law of the center of gravity position of the whole machine and the anchor-breaking manipulator in different postures and the dynamic characteristics of the crawler walking system were analyzed. The results show that under ideal conditions, the maximum lateral slope angle that can meet the normal operation of the unanchoring robot is 10°. Under the forward and turning conditions, the force between the crawler board and the driving sprocket during the movement of the mine unanchoring robot is very similar in the vertical, longitudinal and horizontal directions. The force in the Y direction is approximately sinusoidal, and the force in the Z direction is almost fluctuating near zero. The numerical comparison shows that the force on the crawler board when turning is 2-3 times of that when forward, so the crawler board is more likely to be damaged when turning, and its maximum bearing capacity should be fully considered in the design and strength check. The research results can provide theoretical basis and technical support for the research and development of mine unanchoring robot.

A collaborative optimization control method based on conflict right-of-way decision-making and driving risk field was proposed to address the problem of efficient and safe access of vehicles at unsignalized intersections in unmanned transportation systems of a mine. Firstly, a vehicle conflict state detection model was established by introducing safety margin information, and a three-level right-of-way system of first-come and first-leave, load priority, and priority right-of-way decision library was constructed. Secondly, based on the characteristics of mine working conditions, a driving risk field was established as the safety optimization basis for collaborative optimization control of unsignalized intersections. Finally, by integrating environmental factors, an optimization objective function was established with minimizing traffic risk as the primary objective and traffic efficiency as the secondary objective. A validation scenario for open-pit mine transportation was designed, which includes multiple conflict situations, to verify the collaborative control method. The simulation results show that the collaborative control method can achieve the coordinated and orderly passage of multiple vehicles through unsignalized intersections without stopping, while maintaining the minimum safe distance. The research results can provide a safe and efficient access plan for unsignalized intersections in unmanned transportation systems of smart mines.

In order to achieve the goals of balanced coal flow control, energy saving, and efficiency improvement for the belt conveyor in the mining area of the 4205 working face at Longquan Coal Mine, a mathematical model of the belt conveyor was established. The Multiverse Optimizer (MVO) algorithm was employed to optimize the fuzzy PID controller. Simulations were conducted based on MATLAB, and the results were compared and validated against actual field data. The results are as follows. Firstly, compared with the traditional PID control, the fuzzy PID control optimized by MVO algorithm reduces the adjustment time and overshoot by 46.4% and 34.49 percentage points respectively, and the fitness value converges from 1 022 of the initial iteration to 901.8, indicating that the response speed, stability and anti-interference ability of the system are significantly improved. Secondly, by comparing and analyzing the actual belt speed of the belt conveyor and the regulated belt speed of MVO-Fuzzy-PID within 300 s of five time periods on the belt conveyor, it is verified that the regulated belt speed can be closer to the set value more accurately. Thirdly, from the power point of view, the total power of the motor after MVO-Fuzzy-PID control is significantly reduced. Therefore, the MVO algorithm optimizes the belt speed to achieve energy-saving control and fast matching of coal flow.

Aiming at the problem of low fault diagnosis accuracy caused by unbalanced fault categories of mining excavator engine, an intelligent fault diagnosis method of mining excavator engine based on improved whale optimization algorithm (WOA) to optimize light gradient boosting machine (LightGBM) was proposed. Firstly, the recursive feature cross-validation elimination method (RFECV) was used to extract the features of the fault data of collected excavator engine, and the irrelevant features were deleted. Secondly, Focal-Loss was used to improve the loss function of LightGBM, and an improved WOA was proposed to optimize the hyperparameters of LightGBM, and a new diagnostic model was constructed. Finally, the engine fault data of a mine excavator was used for verification, and compared with common integrated models, tuning frameworks and diagnostic algorithms. The results show that the accuracy and F1 score of the proposed mining excavator engine fault diagnosis model IWOA-LightGBM are 98.08% and 98.53% respectively, and the diagnostic performance is good, which can provide reference for the intelligent diagnosis of mining machinery and equipment.

In response to the problems of large mining range, ore dilution rate, low mechanization degree, and high safety risks in the shallow hole shrinkage mining method for sharply inclined thin ore veins in China, taking the occurrence conditions of a certain metal mine as the engineering application background, a mechanized mining process of shallow hole shrinkage mining method was proposed, and the mechanized mining idea of using micro rock drilling rigs to drill shallow holes upwards was determined. Firstly, based on the operating conditions of the experimental mining site, the design principles and main technical parameters of the micro rock drilling rig were determined. Secondly, the whole machine adopted the design scheme of modularization and 4-degree-of-freedom rock drilling positioning, and the passability, rock drilling positioning, machine stability, and in-situ steering drive system of the micro rock drilling rig were calculated and simulated in narrow and complex environments. The results show that the micro rock drilling rig can meet the practical engineering applications of narrow space transition driving, rock drilling positioning, and visual distance remote control operation in the mining of sharply inclined thin ore veins with shallow hole shrinkage method. The fastest single hole drilling speed can reach 2.2 m/min, the average of efficiency of artificialrock drilling is 1.35 m/min, and the drilling efficiency of equipment is 1.69 m/min, and the comprehensive rock drilling efficiency has been improved by more than 25% compared to manual rock drilling efficiency. At the same time, visual distance remote control can effectively reduce the number of operators, reduce mining amplitude, reduce safety risks, and improve comprehensive mining efficiency, which can provide equipment scheme reference for mechanized fransformation of steeply inclined thin veins with mining dip angle greater than 65° and pulse thickness less than 0.8 m.

Mine fire smoke target detection is one of the important preventive measures to prevent fire accidents in mines. In order to improve the accuracy and speed of fire smoke target detection in mines, a mine fire smoke target detection scheme based on improved YOLOv10s was proposed. Firstly, the collected mine fire smoke image dataset was divided into a training set, validation set and test set according to 7:2:1 after annotation on the online annotation platform. Before training, the images were subjected to wavelet denoising and Auto-Msrcr algorithm enhancement processing, then the original YOLOv10 loss function was reconstructed using Focaler-IoU. The attention module CBAM that strengthened the performance of convolutional neural networks was added, and the third C2f module in Neck was replaced by the cross-stage partial network module VoV-GSCSP, thereby improving the representation ability and decision-making performance of the model. Finally, the feature weights of mine fire smoke recognition were improved. The experimental results show that the improved YOLOv10s model has a 3.5% higher accuracy than the original model, the average precision mAP@50 and mAP@50−95 are increased by 0.7% and 2.2% respectively, and the detection frame rate is increased by 11.9 frames/s. Compared with other mainstream YOLO models, the improved YOLOv10s model has higher detection accuracy and speed, as well as better comprehensive detection performance, which is suitable for high-precision and high real-time mine fire smoke detection scenarios.

The implementation and application of AI management system module algorithms in coal mines is the core issues in the field of smart mine construction. Promoting integrated mine management and control can help improve the intelligent monitoring capabilities of mines. A three-level and five-layer mining AI management system architecture was constructed based on mining intelligent recognition technology, and mining AI management system algorithms including comprehensive mining module, excavation module, transportation module and general module were developed, forming a software+server intelligent visual application mode. The comprehensive mining module adopts CNN, YOLO Fastest V2 and YOLOv3 models to achieve coal mining equipment recognition, area target and support posture detection. The excavation module uses the YOLO model to identify the front beam support and human related detection. The transportation module uses the YOLACT model to identify the operation status of the belt, large coal blocks and foreign objects. The general module implements safety helmets, personnel behavior and environmental detection through OpenCV, YOLO and YOLOv4 models. Among them, the mean average precision of YOLACT model training prediction boxes is 86.79%, and the mean average precision of mask is 83.58%, both of which are at a high level. Real-time display of equipment, personnel, environment and other status results in the fully mechanized mining face. The excavation working face can realize the identification and alarm of personnel intrusion in hazardous areas and abnormal operation processes. The transportation system can realize monitoring of coal quantity of belt, large blocks, foreign objects, etc., real-time tracking and monitoring of the target and early warning of abnormal behavior. The general scenario timely alarms the unsafe behavior in the key underground areas to achieve comprehensive supervision of the entire mining area. The research results can promote the deep integration of AI technology and coal mine safety production, realize the intelligent application scenarios of reducing manpower and improving efficiency in coal mines, which is of great significance for the intelligent mining of coal mines in China.