Accurate and rapid recognition of coal and gangue is an important prerequisite for safe and efficient mining in top coal caving face. In this paper, a novel coal-gangue recognition method is put forward based on improved ant-lion optimization algorithm (IALO), variational modal decomposition (VMD) and improved MobileNetV2 network. Firstly, two strategies of trap boundary adjustment and chaotic mapping are designed for ALO to sufficiently explore the solution space and prevent falling into local optimization. Subsequently, IALO is employed to search the optimal parameters of VMD and the superiority of IALO-VMD can be reasonably embodied through some simulation analysis. Then, the vibration signal-image mapping is performed to produce rich sample data for MobileNetV2. The coordinate attention mechanism and Inception structure are combined with MobileNetV2 to accelerate the training speed and improve the classification accuracy, and the improved MobileNetV2 based classifier is constructed to fulfill an automatic coal-gangue recognition. Finally, an experimental platform of coal-gangue impacting the tail beam of hydraulic support is built and many comparison experiments are carried out. The experiment results indicate that the proposed coal-gangue recognition method has the prediction accuracy of 99.66%, which is increased by 1.43% compared to the classical MobileNetV2. The underground on-site testing results show that the average prediction accuracy of the proposed coal-gangue recognition model can exceed 93%, and can effectively and accurately distinguish different top coal caving states.

Coal–rock recognition is a key technology to realize intelligent shearer and a prerequisite for achieving safe and efficient production in coal mining working face. Based on the analysis of the characteristics and defects of the current recognition methods, this article proposes a novel coal–rock recognition method based on laser scanning technology. First, a coal–rock point cloud data simplification method is designed based on feature points preserving. The purpose is to retain abundant information of coal–rock characteristics and simultaneously meet the requirements of simplicity and precision. Then, an improved ant colony optimization (IACO) algorithm is presented by the two strategies to enhance the optimization efficiency and search ability. The strategies are implemented by adaptively adjusting the pheromone volatilization coefficient and the update step of ant colony position. The IACO is combined with the 2-D OTSU (IACO-TOTSU) to determine the optimal intensity threshold of coal–rock point cloud data, which is utilized to consummate the growth rule of the region growing (RG) algorithm. Meanwhile, the initial seed point of RG is optimized through the curvature calculation of each point and an improved region growing algorithm (IACO-TOTSU-RG) is then performed to achieve the segmentation and recognition of coal–rock point cloud data. Experimental test results indicate that the proposed coal–rock recognition method outperforms others and the coal–rock recognition accuracy can reach above 90%. Finally, an industrial application is provided to prove the practicability and feasibility of the proposed method.

Current positioning methods for indoor or outdoor robots show poor applicability and limited performance for underground mobile equipment with complex scene interference, such as narrow roadways, heavy dust, and poor illumination. Therefore, a fusion positioning framework is proposed based on redundant inertial measurement units (IMUs) and ultra-wideband (UWB) by using a factor graph (FG). In this framework, six layout models of redundant IMUs are designed and compared to fundamentally inhibit the accumulation of positioning errors to a large extent. Then, the optimal layout parameters of base stations are explored through many simulation experiments in terms of the number, shape, and height to preliminarily improve the positioning accuracy of the UWB system. A hybrid calculation method based on bacterial foraging optimization algorithm and Taylor is provided to further enhance the positioning accuracy of UWB in a complex environment. Subsequently, the UWB credibility factor is developed to dynamically adjust the fusion weight of UWB information, to weaken the influence of the non-line of sight errors. Furthermore, FG architecture integrated with the UWB credibility factor is applied to fuse redundant IMUs and UWB for accurate positioning. Some simulations and experiments are carried out and the comparison results verify the superiority and practicability of the proposed positioning method for underground equipment.

To measure the support attitude of hydraulic support, a support attitude sensing system composed of an inertial measurement unit with microelectromechanical system (MEMS) was designed in this study. Yaw angle estimation with magnetometers is disturbed by the perturbed magnetic field generated by coal rock structure and high-power equipment of shearer in automatic coal mining working face. Roll and pitch angles are estimated using the MEMS gyroscope and accelerometer, and the accuracy is not reliable with time. In order to eliminate the measurement error of the sensors and obtain the high-accuracy attitude estimation of the system, an unscented Kalman filter based on quaternion according to the characteristics of complementation of the magnetometer, accelerometer and gyroscope is applied to optimize the solution of sensor data. Then the gradient descent algorithm is used to optimize the key parameter of unscented Kalman filter, namely process noise covariance, to improve the accuracy of attitude calculation. Finally, an experiment and industrial application show that the average measurement error of yaw angle is less than 2° and that of pitch angle and roll angle is less than 1°, which proves the efficiency and feasibility of the proposed system and method.