Research On Coal-gangue Recognition Method Of Top Coal Caving Face Based On Vibration Spectrum Image

Comprehensive mechanized topping coal mining method is an important mining method of extra thick coal seam,the use of manual coal discharge or fixed program coal discharge is very easy to cause top coal ’under-release’ or ’over-release’,seriously affecting the coal recovery rate and coal quality,the identification of coal gangue mixing degree in the coal discharge process is the premise of automatic coal release.Based on the gangue impact characteristics in the coal discharge process,this thesis studies the gangue mixing degree identification method,and collects the vibration signal of the gangue impact tail beam with different mixing degrees by building a simulated coal discharge test bench to realize the identification of gangue mixing degree based on vibration spectrum.The main work and results of this thesis are as follows:(1)Combined with the mining method and coal unloading process of the fully mechanized caving face,a simulated coal unloading experimental platform was built to overcome the difficulty of quantitatively describing the characteristic differences between vibration signals with different gangue contents in underground data.The mixing ratio parameter t was introduced to represent the coal-gangue mixing degree at different coal unloading stages.Parametric modeling of the coal unloading process was conducted and vibration data collection experiments of coal and gangue impact with 7 different mixing ratios were carried out,providing data support for the accurate identification of coal-gangue mixing ratio in subsequent studies.(2)Researched a signal decomposition and reconstruction-based denoising method for coal-gangue vibration signals,optimized the parameters of variational mode decomposition(VMD)using an improved ant-lion optimization algorithm,selected effective components through mutual information and multiscale permutation entropy screening,and reconstructed the signal.Based on this,compared and analyzed the advantages and disadvantages of existing vibration image coding methods,and selected the GASF image coding method that is more suitable for coal-gangue vibration signals to construct a vibration spectrum image dataset containing 7 different coal-gangue mixing ratios.(3)By comparing the performance of classical convolutional neural networks and lightweight convolutional neural networks in terms of accuracy and parameter quantity,Mobile Net V2 lightweight network was chosen.The model incorporated the coordinate attention mechanism to learn image position information and employed the Inception structure to enhance the capability of extracting edge features from vibration spectrum images.Thus,an improved Mobile Net V2 classification network model suitable for coal gangue vibration spectrum images was constructed.The model training was completed using transfer learning strategy,achieving a 5.04% increase in accuracy compared to the original Mobile Net V2 image classification network.(4)An underground industrial experiment was conducted at the Yuhua coal mine of Tongchuan Mining Company to collect coal-gangue impact vibration data during the roof-coal caving process.The signal denoising algorithm and coal-gangue vibration spectrum image classification model proposed in this thesis were validated.The results showed that the model trained using transfer learning achieved an accuracy of 97.91%,demonstrating that the proposed method can accurately identify the mixed gangue ratio at different stages of the roof-coal caving process.In this thesis,there are 82 figures,29 tables and 121 references.