基于深度学习的TIG焊背部熔池检测和熔宽提取  

Deep Learning Based Detection and Width Extraction of Back Molten Pool in TIG Welding

作　　者：卢振洋[1] 宫兆辉 闫志鸿[1] 翟思宽 LU Zhenyang;GONG Zhaohui;YAN Zhihong;ZHAI Sikuan(Engineering Research Center of Automobile Structural Parts Advanced Manufacturing Technology,Ministry of Education,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学汽车结构部件先进制造技术教育部工程研究中心,北京100124

出　　处：《北京工业大学学报》2020年第9期988-996,共9页Journal of Beijing University of Technology

基　　金：国家自然科学基金面上资助项目(51475009)。

摘　　要：为了保证焊接过程中熔池信息提取的实时性和准确性,解决在焊接监控领域传统图像处理算法抗干扰性弱,实时监测的可靠性差,以及自动化程度较低的问题,做到系统可以实时地对焊接过程的熔池进行宽度信息提取和分析,将图像处理算法与深度学习算法进行了结合.通过对TIG焊熔池的观察,针对熔透信息检测,将反面熔池分为3类,首先用图像处理的方法先筛选出烧穿熔池,然后用一个通过大数据样本训练的卷积神经网络对未熔透与熔透进行分类.区别于已有研究,该网络不仅获得了很好的熔透状态检测结果,同时找出了熔池最大宽度,并且保证了实时性,其结果达到了工程应用的要求.To ensure the real-time and accuracy of pool information extraction in welding process,and to solve the problems of weak anti-interference of traditional image processing algorithms in the field of welding monitoring,poor reliability of real-time monitoring and low degree of automation,and to enable the system to extract and analyze the pool width information in real-time,image processing algorithm and deep learning algorithm were combined in this paper.Through observation of TIG welding pool and detection of penetration information,the reverse pool was divided into three categories.First,the burning pool was screened out by image processing method,and then the incomplete and penetration were classified by a convolution neural network trained by large data samples,which was different from the existing research.Not only good penetration test results were obtained,but also the maximum width of the molten pool was found,and the real-time performance was guaranteed.The results meet the requirements of engineering application.

关 键 词：TIG焊 熔透检测 深度学习 图像处理 分类与回归 熔宽提取 

分 类 号：TG409[金属学及工艺—焊接]