| Coal roadway excavation is a special underground project to construct a stable passage in the coal stratum with large burial depth,poor lithology and complicated engineering environment.When tunneling,the randomly developed three-dimensional fissures and their combination of cut coal rock bodies constitute a complex engineering environment in the excavation area,often leading to deformation and collapse of the header,which is an important factor restricting the speed of excavation and affecting the long-term stability of the roadway surrounding rock.Therefore,the accurate detection and real-time stability analysis of the fissures in the surrounding rock of the coal tunnel boring face are crucial to the safety and intelligent tunneling of the roadway.However,the fissures are complex and diverse,and their distribution patterns dynamically evolve with the roadway excavation in real time.In addition,the imaging environment of fissure images is complex,and there are many interfering factors,and the existing technology relies heavily on the a priori knowledge of the human brain when processing such images,which has the limitations of slow processing speed,low generalization ability,and poor robustness,and is unable to effectively cope with the huge amount of fissure images in production,and it is difficult to satisfy the basic needs of batch,fast,and accurate acquisition of complex rock fissures in the intelligent roadway tunneling engineering scenario,and it lacks a method of accurate detection and stability analysis of multi-slice structural surface according to the roadway tunneling.There is a lack of a method to restore the real three-dimensional structural surface network in the direction of roadway excavation based on the multi-slice structural surface images of roadway excavation.The thesis focuses on the problems related to the intelligent tunneling condition perception link,including the intelligent recognition algorithm for the fissures of complex rock bodies in coal system,the 3D reconstruction method for the multi-slice structural surface image of roadway excavation,and the application of the above algorithms in the field of block identification and rapid classification of surrounding rocks,and the results are as follows:(1)It is proposed that the technical framework of intelligent tunneling in coal mine includes three major links,namely,"Condition Sensing,Decision Execution,and Production Guarantee".The traits,spatial distribution characteristics and combination status of Class IV structural face are clearly defined as the condition perception objects,the field description system of structural face is clarified,and the three-dimensional reconstruction characteristic parameters of structural face are clarified.(2)A set of intelligent recognition algorithms based on deep learning for the fissures of complex coal system rocks was developed.Firstly,according to the standard formulation procedure of "data acquisition→image preprocessing→data enhancement→data annotation" for deep learning semantic segmentation dataset,a typical mine condition is selected for data acquisition,and then a two-dimensional gamma function is used for adaptive correction of the uneven illumination image,and a manual annotation with uniform rules is adopted to annotate the crack image with pixel-level accuracy,and the crack image is annotated with pixel-level accuracy,and the crack image is annotated with pixel-level accuracy.A semantic segmentation dataset of coal mine roadway excavation headway images containing 1000 images was created by labeling with pixel-level accuracy using uniform rules manually.Secondly,an intelligent recognition algorithm for coal complex rock fissures based on the encoderdecoder structure(Deep Labv3+)is designed,and the residual block structure Mobile Net V2 is introduced into the encoder network structure to make the training process of the network more stable and efficient.Considering that the fissure prediction is a positive and negative sample imbalance problem,the binary cross-entropy loss function BCE and DICE coefficient are chosen as the hybrid loss function,which accelerates the convergence process of the network training and optimizes the segmentation accuracy to a certain extent.According to the ratio of "training set:validation set: test set=8:1:1",800 images were randomly selected as the training set,and 100 generations were trained on GPU platform,which took 5 hours in total to obtain the model weights network.The generalizability test under the roadway excavation scenario is designed,and the test shows that the algorithm can effectively identify the fissure structure surface in the fresh samples,with less leakage and misdetection,and higher continuity of the identification results.The robustness test under five common interference factors in coal mine tunneling,namely,low illumination,metal mesh occlusion,multi-scale edges,truncation indentation,and concentrated beams,is designed.The highest pixel accuracies predicted by the algorithm for the target categories under the five interference scenarios are 96.7%,76.86%,99.32%,87.44%,and 99.36%,respectively,which indicate that the algorithm can effectively cope with the situation and the recognition accuracy is high.For the conventional recognition conditions of coal tunneling,the highest pixel accuracy of the algorithm for the target category is 93.26%,indicating that the algorithm can realize the batch,fast and accurate acquisition of rock fissures.This algorithm has high application value in the field of intelligent identification of peripheral rock fissures in coal tunneling face.(3)Based on the two-dimensional fissure structural surface data of the roadway,a set of three-dimensional reconstruction algorithms for multi-slice structural surface of the roadway excavation is proposed,including four steps of similarity matching,threedimensional data set construction,three-dimensional reconstruction of the picture sequence,and computer visualization.Two judgment criteria,namely,similarity of morphological features and similarity of orientation features,are proposed to determine the attribution relationship of structural surfaces in adjacent roadway excavation sections,and the fractal dimension and apparent inclination are given as the metrics respectively.For a single fissure,a 3D reconstruction method is proposed in the steps of "edge refinement → corner detection → segment fitting → 3D data set establishment and visualization";for a cross fissure,a 3D reconstruction method is proposed in the steps of "fissure separation → single fissure reconstruction → cross fissure reassembly".The three-dimensional reconstruction method is proposed for the cross-fracture.The results of three-dimensional visualization of structural surface show that the three-dimensional reconstruction technology of discontinuous rock medium can intuitively and accurately display the discontinuous rock medium in the tunnel boring area,analyze the geological anomalies,and provide comprehensive geological visualization information and rich analysis means for the construction of tunnel boring.(4)The industrialized application process of intelligent tunneling technology in coal roadway was formulated,and the test sites were selected to test and apply the aforementioned results.One group of Class IV structural surfaces was detected in the condition-aware link,and its direction was approximately the same as the roadway boring direction.The three-dimensional reconstruction results of the structural surface show that the structural surface intersects with the next section,and the intersection points are located at the upper and right edges of the section.The judgment of diggability shows that when there is only one group of structural surface intersecting with one group of excavation surface in space,there is no movable block,and the roadway can be dug.The determination of empty-roof self-stabilization time shows that the empty-roof self-stabilization time of the 4.7 m span roadway is 7 days.The construction process was determined by checking the table of the relationship between the void roof self-stabilizing time and the appropriate tunneling process.The long-time pressure monitoring shows that the deformation process and section size of the roadway fully meet the production requirements.The thesis has 124 figures,34 tables and 176 references. |
PDF Full Text Request