A deep learning vision and laser-assisted method for the automatic assembly and positioning of shield pipe pieces

Abstract

Abstract The positioning of the tunnel tube sheet is the key to realize the automatic assembly of the shield tube sheet. This paper proposes a deep learning vision and laser-assisted method for the automatic positioning of the shield tube sheet. The vision system and the laser ranging system are used to calculate the planar and depth positional information of the tube sheet to be assembled, and the positional information required by the assembler to assemble the tube sheet is calculated based on the positional information. The vision system is based on a specially designed two-stage convolutional neural network to achieve effective extraction of the contour features of the positioning marks on the surface of the tube sheet. The experiments show that the two-stage convolutional neural network algorithm proposed in this paper not only improves the extraction accuracy significantly compared to the existing feature extraction algorithms, but also increases the recognition rate by 28.4% in the case of contamination on the surface of the tube sheet. The experiments show that the proposed automatic tube sheet positioning method has a positioning accuracy of less than 3\xa0mm, which is higher than the manual positioning accuracy of 10\xa0mm, and can be effectively used for automatic tube sheet positioning.