摘要
熔化极气体保护焊(GMAW)以其低成本、高效率、易于实现自动化等优势在工程领域中具有非常广泛的应用,然而实际焊接过程中错边-间隙等组对因素的异变极易引起熔池失稳,导致GMAW打底焊熔透状态的实时监测与控制较为困难。为此,提出搭建一种基于结构光激光器+高动态范围(HDR)相机的坡口-熔池动态传感系统,用于焊前预扫获取错边-间隙坡口信息和实时焊接过程中GMAW熔池正面尺寸的动态测量;通过工艺试验采集了不同焊接工艺参数和组对参数(错边、间隙)下熔池正面尺寸-背面熔宽几何特征数据集。并在此基础上构建了基于工艺因素+熔池正面尺寸+组对因素的神经网络模型(ANN)来预测背面熔透,通过SHAP可解释算法分析影响熔透状态的关键因素。结果表明:间隙是影响背面熔透尺寸和稳定性的关键因素,这是由于间隙的异变会导致熔池底部受力与内部金属流动模式发生变化;间隙越大,向底部流动的液态金属越多,背面熔透稳定性越低。
Gas Metal Arc Welding(GMAW)has found extensive application in engineering fields due to its advantages of low cost,high efficiency,and ease of automation.However,variations in fit-up factors such as mismatch and gap during actual welding processes readily cause instability in the weld poo,rendering real-time monitoring and control of the penetration state in GMAW backing welding challenging.To address this,a dynamic sensing system for groove and weld pool,based on a structured light laser and a High Dynamic Range(HDR)camera,is proposed.This system is used for prewelding scanning to acquire mismatch and gap groove information,as well as for dynamic measurement of the front dimensions of the GMAW weld pool during real-time welding.A dataset of front weld poo images and geometric characteristics of the back bead width under different welding process parameters and fit-up parameters(mismatch and gap)is collected through orthogonal experiments.Furthermore,a Deep Neural Network(DNN)model based on process factors,weld poo images,and fit-up factors is constructed to predict back-side penetration.The SHapley Additive exPlanations(SHAP)interpretable algorithm is employed to analyze the key factors influencing the penetration state and optimize the DNN model.The results indicate that the performance of the neural network model considering fit-up factors is significantly superior to that of a model solely considering weld poo geometric characteristics.Gap is identified as a crucial factor influencing the size and stability of back-side penetration,as variations in gap can alter the force distribution at the bottom of the weld poo and the internal metal flow pattern.Larger gaps lead to increased liquid metal flowing to the bottom,reducing the stability of back-side penetration.The optimized DNN model meets the requirements of practical welding in terms of prediction accuracy and realtime performance.
作者
李春凯
谢若愚
石玗
王文楷
王程
LI Chunkai;XIE Ruoyu;SHI Yu;WANG Wenkai;WANG Cheng(Lanzhou University of Technology,State Key Lab.of Advanced Nonferrous Materials,Lanzhou 730050,China)
出处
《电焊机》
2025年第12期1-8,共8页
Electric Welding Machine
基金
兰州市青年科技人才创新项目(2024-QN-121)
国家自然科学基金(52365048)
甘肃省科技重大专项(24ZD13GA018)
宁夏自然科学基金(2023AAC03122)。
