Search

The underground coal mine environment is complex. The existing coal mine robot positioning methods have low positioning precision and low real-time performance caused by the non-line-of-sight (NLOS) error and other factors. In order to solve the above problems, a tightly combined positioning method of coal mine robot based on UWB (Ultra Wide Band) and IMU (Inertial Measurement Unit) is proposed. Firstly, the UWB module is used to measure distance between the coal mine robot and UWB base station. The least square support vector machine (LSSVM) model is trained by using real value and measured value of the distance between the coal mine robot and the UWB base station, and the modified LSSVM model is obtained. Secondly, the measured value measured by the UWB module during the positioning process of the coal mine robot is used as the input of the modified LSSVM model. The modified LSSVM model is used to correct the measured value of UWB, reduce the influence of NLOS error on positioning precision, and obtain more accurate distance information. Finally, the range information modified by modified LSSVM model is used as the measurement input of error-state Kalman filter (ESKF). The measurement equation is formed with the position information solved by inertial navigation. The ESKF is used to tightly combine the UWB ranging correction value with the range information calculated by the inertial navigation to complete the state update. The more precise position information of the coal mine robot is obtained, and the precise positioning of the coal mine robot is achieved. The simulation results under different layont schemes of UWB base stations show that using modified LSSVM model can make the UWB range information more accurate, and improve the positioning precision. In the static positioning experiment, when the four UWB base stations are symmetrically distributed at the same height, the root mean square error of the positioning is reduced from 0.146 4 m to 0.139 8 m. When the four UWB base stations are distributed symmetrically with unequal heights, the root mean square error decreases from 0.300 8 m to 0.200 6 m. When the four base stations are distributed irregularly, the root mean square error decreases from 0.317 5 m to 0.314 2 m. Therefore, in actual scenarios, the UWB base stations should be arranged symmetrically at the same height as possible. In the dynamic positioning experiment, the fusion positioning trajectory corrected by the modified LSSVM model is closer to the real trajectory of the coal mine robot than the fusion positioning trajectory before correction. The result verifies that the tightly combined positioning method can reduce the NLOS error and improve positioning precision.

There are many kinds of subsystems in the coal mine safety production monitoring system, and the types of equipments in the subsystems are multifarious, which leads to the low semantic completeness of data and the fragmentation of information interaction data caused by the heterogeneous information of equipment. At present, although the research of coal mine informatization construction has basically realized the network integration of each subsystem, the massive data obtained can not be effectively shared, and the integration analysis can not be carried out. In order to solve the above problems, this paper proposes an information model of coal mine safety production monitoring system based on OPC UA. According to the relevant information of coal mine safety production monitoring system and the general modeling rules of OPC UA information model, the mapping relationship between actual equipment of coal mine safety production monitoring system and information model is analyzed, and the overall structure of information model of coal mine safety production monitoring system is proposed. It is pointed out that when new functions need to be extended in coal mine safety production monitoring system, they can be extended in function set. When new equipment needs to be added to the subsystem, new components can be added to the equipment set to ensure the extensibility of the information model. Taking the information model of gas extraction monitoring system as an example, the information model of methane sensor is established by using UaModeler tool. After the information model is graphically designed, the XML description file is generated and imported into the address space of OPC UA server. Through the third-party client UaExpert connecting server to test the information model of the OPC UA, the results show that the information model can realize the mapping in the address space of the OPC UA according to the mapping rules, and access the server's address space through the OPC UA client. The attributes of any object in each coal mine safety production monitoring subsystems can be obtained, which verifies the feasibility of using the OPC UA information model to realize the information interconnection.

The effective attitude control of mine inspection unmanned helicopter is an important manifestation of the pros and cons of inspection capabilities. The existing unmanned helicopter attitude control changes along with the application scene, and the perturbation changes with it, resulting in the unmanned helicopter attitude fluctuation amplitude and error become larger. In order to solve the above problems, the cellular membrane computing is used to realize the attitude control of mine inspection unmanned helicopter. According an underground unmanned helicopter dynamics model, the unmanned helicopter attitude dynamics model is constructed. The cellular membrane system suitable for the underground unmanned helicopter attitude model is constructed, and the unmanned helicopter attitude membrane controller(MC) is designed. Through flight experiments over the ground and in simulated roadway, the effect of MC on the attitude control of unmanned helicopters is verified. And compared with traditional sliding mode controller (TSC) and linear feedback controller (LFC), the results are listed as follows. In the experimental environment over the ground, the attitude angle of the unmanned helicopter under MC is controlled at -0.8-0.8 rad, and the attitude fluctuation amplitude is less than that of the TSC and the LFC. In the simulated roadway environment, the attitude angle of the unmanned helicopter under MC is controlled at -1.8-2.0 rad, and the fluctuation amplitude becomes smaller. The attitude angle errors of unmanned helicopter under MC are smaller than those of TSC and LFC.

In order to solve the problem of inaccurate estimation and poor robustness of image Jacobian matrix based on traditional Kalman filtering (KF) in the uncalibrated visual servo control of mine intelligent inspection robot, Kalman filtering algorithm with long and short-term memory (LSTM) (KFLSTM algorithm) is proposed. The KFLSTM algorithm uses the LSTM to compensate for the estimation error generated by the KF algorithm, uses the filter gain error, state estimation vector error and observation error for online training of the LSTM, and uses the trained LSTM model for optimal estimation of the Jacobian matrix to improve the real-time and robustness of visual servo control by improving the accuracy and stability of the Jacobian matrix estimation. The uncalibrated visual servo model based on the KFLSTM algorithm is established, and the most optimal Jacobian matrix is used as the input of the controller, which makes the controller output more accurate joint angular velocity so as to control the real-time operation of the manipulator. The KFLSTM algorithm is applied to the six-degree-of-freedom visual servo simulation experiment of the mine intelligent inspection robot. The results show that the image error convergence speed obtained by the KFLSTM algorithm is 100%-142% higher than that of the traditional KF algorithm, the image characteristic error is smaller, the positioning precision is 0.5 pixels, and the robot end effector moves smoothly. Moreover, the method has strong anti-noise interference capability, which can improve the precision and efficiency of the mine intelligent inspection robot effectively and enhance its stability and safety.

The existing path planning algorithm of robot in coal mine uses fixed step size and serial mode to generate path, which has problems such as low success rate, poor real-time performance and low efficiency. Combining membrane computing (MC) with Informed RRT* algorithm, this study proposes a path planning algorithm of robot in coal mine, namely MC-IRRT* algorithm. The algorithm is divided into two stages, namely fast connectivity and path optimization. In the fast connectivity stage, the multi-step cellular membrane structure is constructed, and the step size is adjusted according to the size of the space area. The large step search is used in the area with larger feasible space to accelerate the search speed. The small step search is used in the narrow space to make the search space more refined and improve the success rate of the narrow space path search. In the path optimization stage, a multi-sampling cellular membrane structure is constructed, and multiple basic membranes are calculated in parallel, and the shortest feasible path is searched in parallel in multiple elliptical areas at the same time to save time and improve the efficiency of path optimization. The simple scene experimental results show that compared with Informed RRT* algorithm, the search efficiency of MC-IRRT* algorithm in the fast connectivity stage and the path optimization phase is increased by 76% and 40% respectively. The complex scene experimental results show that the path planning of RRT* algorithm and Informed RRT* algorithm fails, and both PQ-RRT* algorithm and MC-IRRT* algorithm can find feasible paths successfully. Compared with PQ-RRT* algorithm, the rate of MC-IRRT* algorithm is increased by 12.79%, and the planned path length is shortened by 8.18%. The MC-IRRT* algorithm can not only pass through narrow feasible areas quickly, but also can choose to use smaller step at the turning point of the path so as to make the path smoother.

In order to solve the problems of poor predictability and low accuracy in the current fault diagnosis methods of mine ventilator, a predictive fault diagnosis method of mine ventilator based on digital twin and probabilistic neural network (PNN) is proposed.Unity3D, 3dsMax and SciFEA are used to build the digital twin model of ventilator to simulate the structural characteristics, physical properties and operation rules of the real ventilator, and the method uses PREspective to communicate with the PLC of the ventilator in real time to map the operation status of the ventilator to the digital twin model in real time. Based on the digital twin model of the ventilator, combined with expert knowledge, machine learning and historical data, the study constructs a predictive fault diagnosis model of the ventilator. The model continuously learns and updates the model parameters by analyzing the relationship between the real-time data and the operation status of the ventilator. The improved whale optimization algorithm (IWOA) is used to obtain the optimal value of the smoothing factor through the biological behaviors of surrounding prey, preying and searching prey, and assigns the optimal value to the PNN.The optimized PNN is applied to perform predictive fault diagnosis of the ventilator, and the result of the predictive fault model of the ventilator is compared with the actual situation to judge whether the results match the actual situation. If the diagnosis is wrong, the predictive fault diagnosis model needs to be corrected until the fault judgment is accurate.The experimental results show that compared with the PNN fault diagnosis accuracy optimized by the genetic algorithm (GA), particle swarm optimization algorithm (PSO) and whale optimization algorithm (WOA), the fault diagnosis accuracy of PNN optimized by IWOA reaches 97.5%, indicating that the predictive fault diagnosis method of mine ventilator based on digital twin and PNN can meet the requirements of real-time and accuracy of ventilator fault diagnosis.

In order to solve the problems of low detection precision, slow detection speed and lack of damage detection for small areas in existing conveyor belt damage detection methods, a conveyor belt damage detection method based on improved YOLOv4 is proposed. Based on YOLOv4, this method improves the PANet path fusion network part, increases the fusion with the shallow characteristic layer, increases the fusion of the original 3 scales of the characteristic layer to 4 scales, improves the characteristic extraction capability of the model for conveyor belt damage, and improves detection precision. The number of convolutions after fusion of each characteristic layer in the PANet part is reduced from 5 to 3 so as to reduce the amount of calculation and improve the detection speed. The conveyor belt damage images are labeled and input into the improved YOLOv4 model for training and testing. The experimental results show that the conveyor belt damage detection method based on improved YOLOv4 has a fast loss convergence speed and has a good model training effect. Based on improved YOLOv4 conveyor belt damage detection method, the average precision of the conveyor belt tear, surface wear and surface defect detection has reached 96.86%, and the detection speed has reached 20.66 frames/s. Compared with YOLOv4, YOLOv3 and Faster-RCNN, the average precision has increased by 1.4%, 6.35% and 2.16% respectively, and the detection speed has increased by 2.39, 2.34 and 15.25 frames/s respectively. Compared with YOLOv4, the conveyor belt damage detection method based on improved YOLOv4 has higher detection precision and better detection effect for small areas damages.

In order to meet the requirements of high accuracy and low energy consumption of time synchronization among massive nodes of coal mine underground network, a distributed accurate time synchronization algorithm is proposed based on the time-sensitive network (TSN) of coal mine with non-uniform clustering structure of multiple Sink nodes. The underground coal mine TSN is divided into 3 layers, including the TSN convergence layer containing all Sink nodes, the main network consisting of Sink nodes and all cluster head nodes within their respective communication radius, and the secondary network consisting of cluster head nodes and common nodes in the cluster. The TSN convergence layer uses the gPTP algorithm to achieve nanosecond time synchronization among Sink nodes. The main network uses an optimized algorithm based on Kalman filter to predict and compensate for frequency offset, phase offset and noise errors so as to improve the network time synchronization accuracy. The secondary network uses the broadcast-based single and two way hybrid synchronization algorithm to reduce the number of synchronization message packets. The simulation results show that the proposed algorithm can improve the accuracy and stability of network time synchronization effectively and reduce the energy consumption of network time synchronization.

In view of problems such as unreasonable path planning, slow planning speed and poor real -time performance when mine mobile robots use traditional dynamic window algorithm to plan path in complex environment, a dynamic window algorithm of mine mobile robot based on membrane computing and particle swarm optimization was proposed. The traditional dynamic window algorithm is optimized by using randomness of particle swarm optimization and distributed parallel computing ability of membrane computing. In the dynamic window algorithm, the velocity limit space of mine mobile robot is transformed into coordinate space, and the velocity coordinate of the mine mobile robot is regarded as particle position. The speed sampling mode is changed from uniform equal sampling to random sampling, the sample particles are evenly distributed to each basic membrane. The exchange between membranes and the renewal mechanism of particles in membrane are used to evaluate the renewal of particles. The optimal speed is output continuously. The path planning of the mine mobile robot is based on the optimal output speed in continuous time interval. The simulation results show that the algorithm optimizes the speed limit region of mine mobile robot by membrane computing and particle swarm optimization algorithm, and improves the randomness of speed sampling and the rationality of planning path. Compared with the traditional dynamic window algorithm, the proposed algorithm can not only reduce the number of planning steps and the evaluation times of each step, but also shorten the planning path length by 7% -10% and the planning time by 9% -32%, and can adapt to the special environment with U -shaped obstacles.

Traditional convolutional neural network（CNN） is a single-task network. In order to realize simultaneous detection of coal quantity and deviation of belt conveyor, two CNNs are used to detect coal quantity and deviation respectively, resulting in large network volume, many parameters, large computation and long operation time, which seriously affect detection performance. In order to reduce complexity of network structure, a detection method of coal quantity and deviation of belt conveyor based on multi-task convolutional neural network (MT-CNN) was proposed, which could make two tasks of coal quantity detection and deviation detection to share the same network underlying structure and parameters. On the basis of VGGNet model, MT-CNN is constructed by increasing scale of convolution kernel and pooling kernel, reducing the number of channels in full connection layer, and changing structure of output layer. After preprocessing the acquired conveyor belt images, such as graying, median filtering and extracting region of interest, the training dataset and test dataset are acquired, and the MT-CNN is trained. The trained MT-CNN is used to identify and classify the conveyor belt images, so as to realize accurate and fast detection of coal quantity and deviation. The experimental results show that detection accuracy of the trained MT-CNN in the test dataset is 97.3%, and average processing time of each image is about 23.1 ms. The effectiveness of the method is verified by field operation.

The conveyor belt tear detection methods based on traditional two-dimensional Otsu segmentation algorithm adopts threshold selection method of exhaustive search. The image segmentation has a long time and poor real-time performance, which cannot meet requirements of tear detection accuracy and speed at the same time. For the above problems, a conveyor belt tear detection method based on lion group optimization two-dimensional Otsu was proposed. Firstly, the conveyor belt image is collected by the conveyor belt tear detection device, median filtering and histogram equalization were used to denoise and enhance the collected image to highlight the torn part. Then, the close to actual segmentation threshold of preprocessed image is obtained by lion group optimization two-dimensional Otsu algorithm, and the conveyor image is segmented by this threshold. Finally, the tear diagnosis is performed by calculating the number of black pixels in the segmented image. The simulation results show that the optimization ability of proposed method is more powerful than the traditional two-dimensional Otsu algorithm, and segmentation effect of conveyor belt is better, the cracks can be accurately segmented from the conveyor belt image. The correct rate of tear recognition of the method is 98.2%, improves the efficiency and shortens the running time, which can meet the accuracy and real-time requirements of the conveyor belt tear detection.

1.School of Electrical and Information Engineering, Anhui University of Science and Technology, Huainan 232001, China; 2.School of Electronic and Information Engineering, Suzhou University of Science and Technology, Suzhou 215009, China

In view of the problem that energy consumption of sensor nodes is fast and equipment life is cut caused by large amount of information transmission of wireless sensor network, a kind of compressed sensing image processing algorithm based on wavelet transform was put forward. The algorithm uses sym8 wavelet for image sparse and fragmental processing, and uses measurement matrix for adaptive sampling measurement, and finally reconstructs image through OMP algorithm and wavelet inverse transformation. The experimental results show that the proposed algorithm can obtain reconstructed image with high quality by lower sampling rate compared with traditional compressed sensing algorithm.

For location precision of LANDMARC affected by underground non-line-of-sight error, multipath effect and tag collision, LANDMARC location algorithm was improved from the view of measurement of received signal strength indication value, selection of received signal and tag anti-collision. The improved LANDMARC location algorithm judges the nearest reference tag by measuring the received signal strength indication value directly, abandons error signal through Gaussian filter, sets reasonable timeslot for signal sending according to tag number and resets the timeslot at tag collision time. The simulation results verify effectiveness of the improved LANDMARC location algorithm.