Coal rock recognition technology can provide a basis for automatization improvement of shearer and is the key to achieving intelligent unmanned mining in coal mines. The existing coal rock recognition technologies include image recognition, process signal monitoring recognition, electromagnetic wave recognition, and ultrasonic detection recognition, multi-sensor fusion recognition. This article provides a detailed introduction to the principles and application status of the above-mentioned technologies. ① Image recognition technology is currently in the experimental stage, mainly involving large-scale coal rock image data annotation and recognition problems under complex geological conditions. ② Process signal monitoring and recognition technology can analyze relevant signals during coal mining and recognize potential coal rock interface information. But it needs to solve the problems of signal noise interference and complex coal rock interface recognition. ③ Electromagnetic wave recognition technology and ultrasonic detection recognition technology have been applied in actual coal rock interface detection. But there is still a need to improve recognition accuracy and reliability, especially for complex coal rock structures and interface situations. ④ Multi sensor fusion recognition technology needs to solve the problem of data fusion and matching, ensure accurate calibration and reliability between different sensors, and verify its feasibility and practicality in practical applications. In order to solve the above problems, the development directions of coal rock recognition technology are pointed out. ① Research on coal rock recognition should focus on improving the real-time performance and anti-interference capability of algorithms. It will ensure accurate recognition of coal rock under specific conditions and complex environmental interference, and meet the actual mining needs underground. ② Research on coal rock recognition should strengthen the research on mining sensors to improve their anti-interference performance. It is suggested to adopt advanced visual cameras and intelligent devices to combine with sensors to improve the precision and efficiency of coal rock recognition. ③ Research on coal rock recognition should focus on the cross fusion of multiple coal and rock recognition technologies. For coal and rock with different hardness, process signal monitoring recognition and multi-sensor fusion technology can be adopted. For cases with similar hardness, image recognition and electromagnetic wave recognition techniques can be combined to achieve accurate recognition of coal rock wall interfaces and coal seam thickness.

In response to the problems of small pressure monitoring range, low sensitivity, and high cost of existing fiber optic pressure sensors, an improved fiber optic pressure sensor is designed. A strain fiber optic grating is stuck on the cantilever beam and a temperature fiber optic grating is suspended (to make it stress free). The limit cover below the cantilever beam places the spring, corrugated pipe pressure cover, and corrugated pipe cover inside it. The inner upper plane of the limit cover contacts the upper plane of the spring, and the lower plane of the spring contacts the corrugated pipe pressure cover. When external pressure reaches the corrugated pipe through the pipeline at the bottom of the corrugated pipe, the high pressure causes it to undergo axial deformation. The deformation in turn compresses the spring. Finally, the spring undergoes deformation and transmits force to the cantilever beam, changing the stress situation of the strain grating. A spring with a larger stiffness coefficient is added to the single-layer corrugated pipe to limit its deformation when external pressure is generated, allowing the corrugated pipe and spring to jointly transmit pressure to the cantilever beam. The experimental test results show that the improved sensor has a pressure monitoring range of 0-5 MPa, which is 5 times higher than before. The sensitivity of the sensor is 0.379 98 nm/MPa, and the measurement error is within 0.02 MPa. The improved pressure sensor is validated in a water pipeline underground. The results show that compared with the measurement results of high-precision electronic pressure gauges, the pressure demodulation error of the sensor is within 0.02 MPa.

The existing deep learning based shovel detection methods fail to balance detection speed and precision well. In order to solve the above problem, an improved YOLOv7 model is proposed and applied to mining shovel detection. This model is based on the YOLOv7 model, using a lightweight GhostNet network for feature extraction in the backbone network. This model replaces some ordinary convolutions with lightweight GSConv in the neck network to reduce the number of model parameters and computation, and improve the detection speed of the model. Considering the impact of reduced model parameters on feature information extraction capability after lightweight improvement, the neck network is further improved without increasing computational complexity. The coordinate attention mechanism (CA) is embedded in the extended efficient layer aggregation network (ELAN). The bidirectional feature pyramid network (BiFPN) is used to improve path aggregation network (PANet) to enhance the network's capability to extract feature information. Furthermore, it effectively improves the precision of model detection. The experimental results show that compared with the YOLOv7 model, the improved YOLOv7 model reduces the number of parameters by 75.4%, reduces the number of floating-point operations per second by 82.9%, and improves the detection speed by 24.3%. Compared with other object detection models, the improved YOLOv7 model achieves a good balance between detection speed and precision, meeting the demand for real-time and accurate detection of electric shovels in open-pit coal mine scenarios. It provides favorable conditions for embedding into mobile devices.

Currently, research on neural network-based gas emission prediction models mainly focuses on the performance of gas emission problems, with less attention and improvement on the optimizer properties in model training. The training of gas emission prediction models based on neural networks often uses the Adam algorithm. But the non-convergence of the Adam algorithm can easily lead to the loss of the best hyperparameters of the prediction model, resulting in poor prediction performance. In order to solve the above problems, the Adam optimizer is improved by introducing a moment estimation parameter that updates iteratively in the Adam algorithm, achieving stronger convergence while ensuring convergence rate. Taking a certain mining face of Malan Mine in Xishan Coal and Power Group of Shanxi Coking Coal as an example, the training efficiency, model convergence, and prediction accuracy of the improved Adam optimizer in gas emission prediction are tested under the same recurrent neural network （RNN） prediction model. The test results show the following points. ① When the number of hidden layers is 2 and 3, the improved Adam algorithm reduces the running time by 18.83 and seconds 13.72 seconds respectively compared to the Adam algorithm. When the number of hidden layers is 2, the Adam algorithm reaches its maximum iteration number but still does not converge, while the improved Adam algorithm achieves convergence. ② Under different numbers of hidden layer nodes, the Adam algorithm does not converge within the maximum iteration step, while the improved Adam algorithm achieves convergence. The CPU running time is reduced by 16.17, 188.83 and 22.15 seconds respectively compared to the Adam algorithm. The improved Adam algorithm has higher accuracy in predicting trends. ③ When using the tanh function, the improved Adam algorithm reduces the running time by 22.15 seconds and 41.03 seconds respectively compared to the Adam algorithm. When using the ReLU function, the running time of the improved Adam algorithm and the Adam algorithm is not significantly different. ④ Using the improved Adam algorithm for traversal grid search, the optimal model hyperparameters are obtained as {3,20, tanh}, with mean square error, normalized mean square error, and running time of 0.078 5, 0.000 101, and 32.59 seconds, respectively. The optimal model given by the improved Adam's algorithm correctly judges the trends of several valleys and peaks that occur within the predicted range. The fitting degree on the training set is appropriate, and there is no obvious overfitting phenomenon.

The mining face is the main place for gas emission in mines. Accurately predicting the amount of gas emission from the mining face and proposing targeted prevention and control measures are of great significance for ensuring mine safety production. A prediction method for gas emission in mining face based on random forest regression algorithm has been proposed. Using the measured gas emission data from the working face as the original sample, the Bootstrap sampling method is used for random sampling. The out-of-bag (OOB) data assessment score oob_score is used as an evaluation indicator for the random forest regression model tuning parameter and importance of feature variables. The optimal parameters of the model and the percentage of importance of feature variables are calculated. The method ranks the importance proportion of each feature variable and conducts performance analysis of the random forest regression model according to the ranking. The results show that as the number of feature variables increases, the model performance does not show a regular change. When the number of feature variables is small, there may be overfitting. The test results show that the average absolute error and relative error between the predicted and measured values of the created random forest regression model decrease with the increase of the number of feature variables. The increase of the number of feature variables can improve the predictive performance of the model to a certain extent. Compared with the principal component regression analysis method, the random forest regression model reduces the average relative error by 14.29% for the same set of data, resulting in better prediction performance. The principle is simpler, parameter adjustment is easier, and the calculation speed is faster. The results can provide strong theoretical support for predicting gas emission in mining face.

The existing local path planning algorithms only achieve free movement of mobile robots in the scenario. But local path generation does not consider road constraints in the scenario, which is not applicable to some regularized structured roads. The OpenPlanner algorithm solves this problem well. But the local path planned by the traditional OpenPlanner algorithm does not meet the maximum turning curvature constraint of the mobile robot and cannot be effectively tracked by the mobile robot. In order to solve the above problem, the traditional OpenPlanner algorithm is improved from two aspects: state sampling and evaluation function. The improved OpenPlanner algorithm is applied to local path planning of mobile robots. In the state sampling stage, the optimal local path solution space is expanded by designing a double-layer local path cluster. The longitudinal sampling distance of the first layer local path cluster is linearly related to the driving speed in sections. The longitudinal sampling distance of the second layer local path cluster is 1.5 times that of the first layer local path cluster. In the path selection stage, the curvature cost of the path (obtained by summing the curvatures of each sampling point on the local path) is introduced into the evaluation function to ensure that the local path cluster satisfies the maximum turning curvature constraint of the mobile robot, thereby making the local path tracked by the mobile robot. The experimental results show that compared with the traditional OpenPlanner algorithm, the improved OpenPlanner algorithm filters the optimal local path with smoother turning. The average curvature is reduced by 31.3% and 6.2% in obstacle free and obstacle present scenarios, respectively. Moreover, the local path can be well tracked by mobile robots.

Currently, research on the collaborative control of shearers and scraper conveyors has preliminarily established a collaborative control mechanism for mining and transportation systems. But none of them have taken into account the uncertainty and coupling features of factors that affect the stable operation of mining and transportation systems in unstructured fully mechanized mining face environments. And the coal flow state and scraper conveyor load current are affected by the underground electrical system and cannot truly reflect the changes in scraper conveyor load. In order to solve the above problems, a collaborative control method for mining and transportation in fully mechanized mining face based on scraper conveyor load current intensification and random self-attention capsule network (RSACNN) is proposed. Based on the electrical coupling features of the electric motor current of the scraper conveyor, a current intensification model is used to preprocess the original scraper conveyor current and obtain the current component that can reflect the real load of the coal flow system. There is a highly nonlinear and uncertain relationship between the operating state parameters of the mining and transportation system in the fully mechanized mining face and the traction speed of the shearer. It is difficult to establish an accurate mathematical model. In order to solve the above problem, based on capsule neural network (CNN), the features of fine-grained features such as sudden changes in the operating state of the mining and transportation system in the fully mechanized mining face can be preserved. A collaborative control model for mining and transportation in the fully mechanized mining face based on RSACNN is established. The verification results show that compared with the self-attention capsule neural network (SACNN) method and the CNN method, the proposed RSACNN method has higher precision in predicting the traction speed of the shearer. The fitting values between the predicted speed and the actual speed have increased by 0.032 05 and 0.075 04 respectively. The average absolute error decreases by 17.7% and 22.6% respectively. The average absolute percentage error decreases by 49.9% and 71.5% respectively. The root mean square error decreases by 13.3% and 34.6% respectively.

Foreign objects mixed in during coal mining may cause accidents such as blockage or even tearing of conveyor belt connections. Most existing machine learning algorithms for coal flow foreign objects use supervised learning to automatically recoginze item categories. However, in real industrial and mining scenarios, the scarcity of abnormal samples leads to problems of serious imbalanced sample distribution and significant features lost in the modeling dataset. In order to solve the above problems, a coal flow foreign object intelligent detection method based on dual-attention Skip-GANomaly (DA-GANomaly) is proposed. This method adopts a semi supervised learning approach, which only requires normal samples to complete the training of the foreign object detection model, effectively solving the problems of low recognition accuracy and poor robustness caused by imbalanced sample distribution. On the basis of Skip-GANomaly, a dual attention mechanism is introduced to enhance the information exchange between the encoder and decoder and suppress irrelevant features and noise. It highlights the interesting features that are conducive to distinguishing abnormal samples, and further improves the accuracy of model classification. The experimental results show that the classification accuracy of the DA-GANomaly model is 79.5%, the recall rate is 83.2%, and the area under the precision recall curve (AUPRC) is 85.1%. Compared with 5 classic anomaly detection models such as AnoGAN, the DA-GANomaly model has the best overall performance.

There are a large number of trackless rubber-tyred vehicles in underground coal mines. The transportation is easily affected by moving surfaces, emergencies, and other factors. Traditional manual scheduling methods are inefficient and prone to problems such as idle, empty, and wasted vehicles. However, existing auxiliary transportation vehicle scheduling methods mostly focus on fixed tasks using discrete event optimization schemes. It breaks down the global model into local models, and lacks analysis of the overall situation of underground coal mines. In order to solve the above problems, a global scheduling model for trackless rubber-tyred vehicle in underground coal mines based on Baidu industrial solver is proposed. The design scheme of the information collection module, data modeling module, and industrial solver module in this model are introduced, as well as the global scheduling process for trackless rubber-tyred vehicles. This model adopts a global scheduling algorithm for trackless rubber-tyred vehicles based on "batch solving and iterative optimization". The vehicle scheduling problem is optimized and solved by Baidu industrial solver based on action adjust heuristic algorithm. It solves the problems of long solving time and easy getting stuck in local optimal solutions in traditional scheduling models. The experimental results show that the global scheduling model for trackless rubber-tyred vehicles based on Baidu industrial solver significantly reduces the number of vehicles used and improves vehicle operation efficiency compared to manual scheduling methods. The solution time for scheduling optimization is lower than that of the local scheduling model based on Gurobi solver. It is more suitable for large-scale complex scheduling tasks in underground auxiliary transportation scenarios.

The communication system is the channel and bridge for information transmission in the hydraulic support electro-hydraulic control system of the fully mechanized mining face. Currently, CAN bus is commonly used as the communication bus. It is susceptible to interference from the complex electromagnetic environment underground, resulting in internal communication hardware failures of the support controller and causing the phenomenon of "disconnection" of the controller. In addition, the CAN bus communication system adopts a multi master communication mode. The disconnection of a single controller will cause the entire electro-hydraulic control system to malfunction, posing a safety hazard. A CAN communication protection circuit has been designed to ensure stable operation of the communication system under high load conditions and strong anti-interference capability in complex environments. A fault detection and diagnosis method for CAN bus communication is proposed based on the CAN bus communication protocol combined with the token ring network concept. By designing the frame structure and fault detection method of data reasonably, the defect of difficult positioning of nodes when lost in CAN bus communication mode is compensated. The impact of increasing data length on transmission load is minimized to ensure good communication performance. Two end controllers are combined with six hydraulic support controllers to form a ring network. The upper computer issues commands from time to time to simulate the actual load situation of the bus during underground operation. The experimental verification of the bus communication fault detection and diagnosis method for the hydraulic support electro-hydraulic control system is carried out. The results show that this method has a low impact on the system load rate and will not affect the normal operation of the system. When a faulty node occurs, the faulty controller can be detected within 300 ms and an alarm can be sent to the entire working face, with a fault elimination rate of 100%.

In traditional rockburst hazard assessment methods, there are problems of large computational complexity and low precision in dividing hazardous areas. In order to meet the development needs of intelligent and visual prevention and control of rockburst, an intelligent assessment method for rockburst hazard areas based on image recognition technology is proposed. Using a semi quantitative estimation method, the method quantitatively characterizes the main controlling factors of dynamic and static loads for 11 types of rockburst hazards. Based on OpenCV machine vision library and deep learning model, the method achieves image recognition for a single main control factor. By constructing a mapping matrix between the grayscale of the image and the stress concentration coefficient, linear and nonlinear superposition of a single influencing factor is achieved to obtain the stress concentration coefficient matrix of the assessment area. Using the min max standardization method to construct a 4-level discrimination standard of "no, weak, moderate, and strong" for the hazard area of rockburst, the method achieves graded and division assessment. A software for intelligent assessment of rockburst hazards is developed based on Python language, and the actual application effect of the software is tested. The results show that the software improves the traditional one-dimensional linear hazard area division method for roadways to a two-dimensional plane division method for the entire mining space. It significantly improvies the assessment efficiency and precision of hazard area division and reduces labor costs. The assessment results are highly consistent with the microseismic energy density cloud map and the on-site measured mining pressure pattern, which can provide effective guidance for the prevention and control of on-site rockburst.

At present, there is a lack of effective online detection methods for important parameters such as mineral composition that affect the flocculation and sedimentation effect of slime water. There are also lagging issues in the turbidity and interface of the overflow of the concentration tank, which limits the development of intelligent dosing for slime water in coal preparation plants. In order to solve the above problems, a fast detection method for slime water flocculation and sedimentation rate based on image grayscale recognition is proposed. Using a CCD camera to collect images of the sedimentation process of slime water online, and using the mean filtering method for noise reduction, the average grayscale and average grayscale change rate of the image are calculated. The sedimentation rate is obtained by using the relationship between the sedimentation rate and the average grayscale change rate. The method extracts feature values such as grayscale, energy, contrast, variance, and cross-correlation from images through flocculation sedimentation experiments for analysis and verification. The analysis results show the following points. ① Among the five image features, the change in grayscale mean conforms to the variation law of sedimentation rate during the sedimentation process of slime water batches. There are buffer zones, linear zones, and stable zones, and the variation features can be obtained within 30 seconds. ② There is a good linear correlation between the average grayscale change rate and sedimentation rate. When the concentration of slime water is 20 g/L, the linear correlation coefficient between the average grayscale change rate of the image and sedimentation rate under different flocculant addition amounts is 0.977 2. Under the conditions of slime water concentration of 5-25 g/L and flocculant addition amounts of 0.1-0.2 kg/t, the linear correlation coefficient between the two is 0.944 1. ③ The average grayscale change rate can adapt to the changes in the flocculation and sedimentation state of slime water within a large range. The average grayscale change rate can be used to quickly detect the flocculation and sedimentation rate of slime water and serve as the basis for intelligent adjustment of slime water dosing.

The safety evaluation granularity of the current coal mine safety monitoring and management system is basically at the mine level or subsystem level. It cannot be finely managed for different areas of the mine. In order to solve the above problem, a unified area division method is proposed for safety risk assessment based on the features of operational scenarios in various areas of the mine. This method combines information such as risks, hidden dangers, disasters with area operating conditions, equipment maintenance management, and personnel positioning to unify and organize data from various safety systems. A comprehensive assessment of the safety indicator system is conducted from four dimensions: human, machine, environment, and management. The weights of each indicator in the area safety assessment are calculated through a combination of subjective and objective weighting methods. The subjective weighting is achieved through the analytic hierarchy process, while the objective weighting method is achieved through the entropy weighting method. The method constructs an area safety assessment model, which quantitatively evaluates and classifies the current safety situation of coal mines. The method uses normalized fusion weights to calculate the basic safety score, and considers high-risk combinations, historical trend changes, and inter regional coupling effects to modify the score. The comprehensive safety scores of all levels of coal mine areas are obtained. This model has been successfully applied to the intelligent comprehensive control platform of Shaanxi Xiaobaodang Mining Co., Ltd.. It provides effective reference for accurately evaluating underground safety risks in coal mines and improving the level of coal mine safety management.

The effective extraction layer of the overburden fracture zone in goaf is the basis for arranging high-level extraction boreholes to treat adjacent layers and gas in goaf. Based on the key layer theory, a mathematical model for the effective extraction layer in fracture zones is established, and the upper and lower boundaries of the effective extraction layer are determined. The lower boundary of the effective extraction layer is the first key layer above the collapse zone of the goaf, and the upper boundary is the first key layer below 10 times the mining height of the overburden layer in the goaf. The effective extraction layer includes the lower boundary rock layer and does not include the upper boundary rock layer. According to the mathematical model of the effective extraction layer of the fracture zone, it is calculated that the effective extraction layer of the fracture zone in the 8+9 coal seam of Duanwang Coal Mine is from the medium sandstone at 12.6 m above the coal seam roof to the No. 4 coal at 39.3 m. According to the drilling and observation results of the overburden fracture zone in the goaf, the fracture angle of the working face is about 62°. The height range of the fracture zone is 11.5-40.5 m above the coal seam roof. A high-level drilling and extraction test is conducted at Duanwang Coal Mine. It is found that the actual effective extraction layer of the fracture zone is from medium sandstone at 13.9 m above the coal seam roof to sandy mudstone at 37.4 m. The results of drilling observation analysis and high-level drilling extraction test have verified the accuracy of the mathematical model of effective extraction layer in the fracture zone. The research results can provide theoretical basis for the design of high-level extraction engineering in high gas and coal and gas outburst mines.

There are flammable and explosive gases such as gas underground in coal mines. The electromagnetic waves radiated by the 5G wireless communication system base station antenna are absorbed by the underground metal structure, generating discharge sparks at the metal structure breakpoint. When the energy of the electric spark reaches the minimum ignition energy of gas, an explosion may occur, which limits the application of 5G technology in coal mines. In order to evaluate the safety of the RF power of 5G wireless communication base stations, the relationship between RF power, maximum radiation field strength, and distance is obtained by analyzing the coupling of electromagnetic waves with metal structures. Using the minimum ignition energy as the safety criterion, it can be concluded that when the receiving power of the antenna load is less than 2.625 W, it can ensure that it will not cause gas explosions. The analysis shows that 700 MHz should be given priority as the 5G working frequency band in coal mines underground. By analyzing the directional coefficient, it is concluded that a symmetrical oscillator antenna metal structure with an arm length to wavelength ratio of 0.65 should be chosen for research. The safe electric field strength of the symmetrical oscillator antenna metal structure is 202.9 V/m, and the minimum safe distance is 0.2 m. The simulation results show that the electric field distribution is extremely uneven in areas less than 0.2 m away from the transmitting antenna. The electric field distribution is relatively even in areas more than 0.2 m away from the transmitting antenna. The minimum radio frequency power that causes a gas explosion in an area greater than 0.2 m from the transmitting antenna is 27.45 W.

The surface point cloud data of coal mine roadways extracted using 3D laser scanning technology has a large amount of redundant data. The existing point cloud data simplification methods have the problem of insufficient detail preservation in the processing of large-scale point clouds. In order to solve the above problems, a surface point cloud data reduction method for coal mine roadways based on secondary feature extraction is proposed. Firstly, the method performs denoising preprocessing on the collected original roadway point cloud data. Secondly, the method establishes a K-d tree and uses principal component analysis to estimate the denoised point cloud data to fit the normal vector of the neighborhood plane. Thirdly, the point cloud is preliminarily divided into feature and non-feature regions using a smaller normal vector angle threshold, retaining the feature regions and randomly downsampling the non-feature regions. Fourthly, based on the larger normal vector angle threshold, the feature region point cloud is divided into feature points and non-feature points. And voxel random sampling is conducted on the non-feature points. Finally, the method merges the two point cloud simplification results with the feature points to obtain the final simplified data. The simulation results show that under million data level point clouds and high precision conditions, this method achieves better results in feature preservation and reconstruction precision compared to curvature simplification methods, random simplification methods, and grid reduction methods. The average standard deviation calculated after 3D reconstruction can be about 30% lower than other methods under the same reduction rate.

The drawing force of the steel wire rope in the joint area of the conveyor belt is an important indicator to measure the bearing capacity of the joint. At present, research on the joint of steel wire rope core conveyor belt mainly focuses on the structural parameters of the joint, vulcanization process, and adhesive performance of the rubber material. It has not pointed out the influence of lap length on the load-bearing capacity of the joint. To study the influence of lap length on the load-bearing capacity of steel wire rope core conveyor belt joints, the st1250 steel wire rope core conveyor belt is taken as the research object. A joint model is established by taking a single steel wire rope part at the conveyor belt joint. The bilinear cohesive zone model is used to simulate the bonding state between the steel wire rope and rubber. The model parameters are obtained through tangential tensile shear tests and normal tensile tests. By combining the bilinear cohesive zone model with the steel wire rope rubber contact interface, a simulation analysis is conducted on the damage evolution process of a single steel wire rope detached from rubber in a joint. It is found that the joint damage evolution process can be divided into four stages: linear loading, damage initiation, damage propagation, and complete failure. Moreover, the joint damage failure curve is consistent with the traction displacement curve of the bilinear cohesive zone model. It verifies that the bilinear cohesive zone model can effectively simulate the damage failure process of steel wire rope core conveyor belt joints. Simulation is conducted on joint models with different lap lengths. It is found that as the lap length increases from 350 mm to 750 mm, the overall stiffness of the joint shows a non-linear increase, and the maximum shear stress on the joint rubber shows a decreasing trend. Therefore, it is determined that the range of lap length should be controlled within 350 mm to 750 mm. The influence of lap length on joint bearing capacity under different wire rope diameters is simulated. The results show that the drawing force of wire rope increases nonlinearly with the increase of lap length. The larger the diameter of the steel wire rope, the greater the increase in the drawing force of the joint steel wire rope with the increase of the lap length. The functional relationship between joint lap length and single wire rope drawing force under different wire rope diameters is fitted, providing a theoretical basis for the rational selection of joint lap length under different bearing capacity requirements.

The flow and diffusion of high-temperature smoke in mine fires is an important cause of safety accidents. In response to the unclear relationship between fire hazard zones and time in typical mine roadways, a safety zone classification method for T-shaped roadway fire in deep coal mines is proposed. A three-dimensional numerical model is established using Pyrosim software to simulate the high-temperature smoke flow in the T-shaped roadway during the fire development stage under high temperature and humidity conditions. The variation law of temperature field and CO, CO₂ concentration field with time and spatial location in the T-shaped roadway during the fire development stage are revealed. Based on the simulated data of the height of the human mouth and nose (i.e. the height of 1.6 meters in the roadway), the relationship between the horizontal length of the roadway and temperature, CO concentration and CO₂ concentration is obtained. The airflow mixes high-temperature smoke through the roadway and spreads downwards along the top of the roadway. As the distance from the fire source increases, the temprature gradually decreases, and the longitudinal distribution of CO and CO₂ concentration contour lines becomes denser. On this basis, safety zones are classified based on the temperature of the smoke and the harm degree of CO and CO₂ concentration to human health. The smoke diffusion area is divided into four categories: safety zone (hazard level 1), mild hazard zone (hazard level 2), moderate hazard zone (hazard level 3), and severe hazard zone (hazard level 4). The analysis results show that in the temperature classification results, the measurement points in roadwayⅠare mainly concentrated in the severe hazard zone. In the toxic gas classification results, the safety zone range of CO₂ in roadway Ⅰ is larger than that of CO. The risk factors for CO are greater, mainly concentrated in mild and moderate hazard zones. In roadway Ⅱ, it is mainly concentrated in mild hazard zones. The range of hazard level 1 in roadway Ⅰ gradually decreases over time, while the range of hazard level 4 gradually increases over time, with the maximum change rate occurring at 40 seconds. The rates of change for hazard levels 2 and 3 are very small. The regional range changes of the two classification methods in roadway Ⅱ are similar, with the maximum change rate of hazard levels 2 and 3 occurring at 60 seconds.

The division of advanced support areas and support methods in roadways are key factors affecting the stability of surrounding rock in mining roadways. The existing research mostly divides the advanced support area under static load conditions. Further exploration is needed for the division of advanced support area under dynamic load impact and the relationship between roadway surrounding rock and hydraulic support. Taking the 5304 working face roadway of Zhaolou Coal Mine as the research object, the variation features of working resistance of hydraulic support under dynamic load impact and the relationship between surrounding rock and hydraulic support are analyzed. The concept of dynamic coefficient is proposed. Under the action of dynamic load disturbance, the peak point of advanced support pressure will transfer to the interior of the coal body, resulting in a new plastic zone. Therefore, the area affected by advanced support pressure is divided into fracture zone, plastic zone, elastic zone, original rock stress zone, and newly added plastic zone. According to the coal rock state and dynamic boundary points, the advanced support area is divided into reinforced support section, auxiliary support section, and original support section based on dynamic stress as the boundary. The reinforced support section is composed of fracture zone, plastic zone, and partially elastic zone, and requires high-strength advanced support equipment to strengthen roof support. The auxiliary support section is mainly composed of elastic zones and requires single hydraulic pillars or unit hydraulic supports for auxiliary support. The original support section is located in the original rock stress zone as a whole, and there is no need to strengthen the support. The numerical simulation is used to study the variation law of advanced support pressure under dynamic load, and establish a calculation model for advanced support pressure in roadways under dynamic load. The dynamic stress expression for each support section is derived. The on-site test results show that the support scheme designed based on the division of the advanced support area of the roadway has good support effect and can meet the quality requirements of the advanced support area.

The early fault characteristics of rolling bearings in coal mining equipment are weak, and they are easily affected by factors such as load and working conditions. The characteristics can be submerged by noise, making bearing fault diagnosis difficult. Most existing research uses a single algorithm to process bearing fault signals, and the accuracy of fault characteristic extraction and fault diagnosis needs to be further improved. A fault diagnosis method for coal mining equipment bearings is proposed, which combines local characteristic-scale decomposition (LCD) and singular value decomposition (SVD). The LCD method is used to decompose the vibration signal of coal mining equipment bearings into several intrinsic scale components (ISC), achieving preliminary signal denoising. The method calculates the Shannon entropy of each ISC, selects the ISC with the smallest Shannon entropy for SVD. The method constructs the singular value difference spectrum of the SVD signal. The method reconstructs the signal for the maximum abrupt component to achieve signal enhancement and denoising. The method performs Hilbert envelope demodulation on the reconstructed signal to obtain the characteristic frequency of bearing faults, and then determine the bearing faults. The on-site measured data is used to validate the bearing fault diagnosis method of coal mining equipment based on LCD-SVD. The results show that this method can accurately extract the characteristic frequency of bearing faults, thereby achieving early fault diagnosis of coal mining equipment bearings.