| Shield machine plays an irreplaceable role in tunnel construction,and the assembly efficiency of shield segments will affect the construction progress of shield machine.At present,the assembly effect of shield segments in China still needs to be controlled manually,which has high requirements for the proficiency,safety awareness and environmental insight of technical workers.Therefore,in order to improve the positioning accuracy,grasping efficiency and assembly effect of shield segments,based on the actual scene and combined with modern computer vision technology,the intelligent assembly of shield segments is studied as follows:Firstly,a vision measurement system is designed based on the vision sensor and shield erector.The system mainly considers three parts:image acquisition,data fusion and pose estimation,and then proposes to build a large-scale sensing system by calibrating multiple camera sensors.With the help of the motion control system of the shield assembly machine,the visual calibration task is completed,and a global unified perceptual coordinate system is finally established.Finally,the proposed multi-camera calibration method in large scenes is simulated and tested in practice.The results show that the repetition accuracy of the detected target plane is within 0.3°and the translation error is within 4mm even when the sensor span difference between two cameras is 2 m.Secondly,in order to solve the problem of image quality in the tunnel dark environment,a two-stage neural network framework is designed.The framework is mainly composed of two parts.The first part is composed of the brightness enhancement model and image denoising model which are trained from the public data collection without labels.The main function of this part is to establish the image enhancement data set in the tunnel dark environment.The second part is to design a network model LLE-Unet to realize the tasks of brightness enhancement and denoising for dim images at the same time according to the knowledge distillation method.The main purpose of the model is not only to learn the brightness enhancement and denoising ability of the one-stage model at the same time,but also to simplify the complexity of the model and improve the efficiency of image enhancement,which is very important for engineering application.Finally,the public data set and the image data in the actual tunnel environment are tested and analyzed.The results show that the image enhancement model LLE-Unet designed in this paper is suitable for most dark environments,and the enhancement effect is stable,which is also better than other methods.Finally,with the help of the actual shield assembly platform,this paper designs a set of intelligent assembly strategy of shield segments which aims to replace the current manual assembly task and improve segment assembly efficiency and construction safety.This strategy mainly extracts effective target pose information from multiple images collected by the visual measurement system,estimates the pose of shield segments with different sizes,and then transmits the positioning information to the shield assembly machine to guide the shield assembly machine to grasp and assemble the shield segments.Finally,by comparing the manual grasping and assembly process,the paper can ensure that the assembly strategy of the whole shield segment is feasible,and can ensure that the grasping error in practical application is controlled within 0.6° and 3mm,and the assembly error is controlled within 0.3° and 3mm. |
PDF Full Text Request