图像自动增强与注意力机制深度学习的MIG焊缝跟踪系统

打开文本图片集

MIG weld seam tracking system based on image automatic enhancement and attention mechanism deep learning

ZHU Ming1，²，LEI Runji1，WENG Jun1 ，WANG Jincheng1， SHI Yu1， 2 （1.State KeyLaboratoryof Advanced Procesing and Recycling of Non-ferrous Metals，Lanzhou Universityof Technology， Lanzhou 73Oo5O，China;2. Key Laboratory of Non-ferous Metal Alloys and Processing of State Education Ministry， Lanzhou University of Technology，Lanzhou 730o5O，China）

Abstract：Aiming at the problem that conventional MIG welding is diffcult to adjust the welding position in real time according to the group deviation and thermal accumulation deformation，a weld seam tracking method based on passive vision is proposed. Through the image spatial domain filtering and automatic enhancement algorithm， the YOLO v7 deep learning model with attention mechanism is used to extract and analyze the groove alignment position andarc position in theregion of interest inreal time.The fuzzy control algorithm isused to control the MIG welding processin real time when the preset deviation occurs. The results show that，the image automatic enhancement algorithm is used to complete the preprocesing of the image，and the pixel gray value of the edge position is increased from 40 to about110，which significantly improves the accuracy of the edge position information extraction；Based on the YOLO v7 network structure，the attntion mechanism module is added to improve the eficiency of target detection，and the mAP index is as high as 99.27% .The preset deviation test shows thatthe pixel errorof the alignment deviation detection is within8 pixels，and thealignment deviation distance is controlled between ±0.5 mm.

Key words： weld tracking；passive vision； image enhancement； deep learning

0 前言

在工程实践中，由于焊件的坡口加工、组对与热积累变形等造成的偏差，会引起间隙、错边的不规则变化，并严重影响了焊接过程的稳定性与焊缝质量[1]。（剩余4193字）