Research On Downhole Drill Pipe Counting Method Based On Convolution Neural Network

Coal seam gas is a catastrophic gas that causes coal mine gas explosion and seriously threatens coal mine safety production.Coal seam gas pre-pumping is crucial to improve the efficiency of underground mining in coal mines and ensure the safe production of mines.Drilling depth is an important parameter of gas pre-pumping.Due to the protruding coal seam,there is a situation that the drilling depth of drill pipe is not up to the standard,which restricts the efficient production of coal mines.Drill hole depth is usually calculated by the number of drill pipes drilled,and the specific research on drill pipe counting method in this paper is as follows.By recording the complete workflow video of the drilling rig at the general mining face to obtain a large number of high-quality images of the working of the drilling rig,the working status data set of the drilling rig is established as the data basis for the research of the drill pipe counting method in this paper.Based on this,this paper analyzes the rig working dataset and studies the downhole drill pipe counting method for different environments under coal mines.A solution based on improved MobileNetV2(P-MobileNetV2)is proposed to solve the problem of low drill pipe counting accuracy in a stable working environment of drilling rigs.Firstly,the drilling rig working states are divided into four working states:unloading,stopping,hitting and loading;secondly,the accuracy of MobileNetV2 in recognizing the drilling rig working states is enhanced by adding three strategies:attention module,optimized objective function and migration learning;finally,PMobile Netv2 identifies the rig’s operating status,generates the corresponding confidence data and uses sliding window filtering to implement the drill pipe counting function.The experimental results show that P-MobileNetV2 outperforms seven other classical image recognition models in the task of detecting the work status of a drilling rig,and verifies the feasibility of this counting method for underground coal mine applications.A drilling rig target detection algorithm based on improved YOLOX is proposed for the problem that the image quality of drilling rigs in complex is poor and the rigs are difficult to detect.First,replace YOLOX’s CSPDark Net feature extraction network with the lightweight Ghost Net network to speed up model detection,then add the SCBAM attention module to enhance rig saliency in complex backgrounds to solve the problem of dim rig working images due to uneven lighting,and the RFB module is used to expand the perceptual field to solve the problem of difficult detection of the rig due to rig motion.Optimize the objective function to obtain a more suitable anchor frame for rig detection,and establish a rig target detection model(α-GCR-YOLOX).The experimental results prove that the accuracy and inference speed of α-GCR-YOLOX is better than other target detection algorithms for rig target detection in complex working scenarios of drilling rigs such as uneven illumination,existence of occlusion and motion blur.Aiming at the problem that the drilling machine is easy to lose the target in the drill pipe counting method,a drill pipe counting method combining improved YOLOX and Deep Sort is proposed.On the basis of Deep Sort target tracking algorithm,Gaussian smooth interpolation is used to compensate for the track gap caused by missed detection.The drill target detection model and Deep Sort are fused to achieve efficient and accurate drill target detection,and the drill target tracking model is established.Accurate counting of drill pipe is achieved by automatically,quickly and accurately tracking the drill rig to obtain coordinate motion information of the rig’s movement and processing the data to calculate the number of drill pipes.The experimental results prove that the tracking accuracy and precision of the rig target tracking model is better than other target tracking algorithms,solves the problem of easy target loss when tracking rig targets,and achieves accurate counting of drill pipes.Finally,the drill pipe counting method based on the improved YOLOX and Deep Sort algorithm is encapsulated into a drilling intelligent management system,and the system structure,software interface and use method of the drilling intelligent management system are introduced in detail.Figure [39] Table [17] Reference [87]...