1. China Nuclear Industry 23 Construction Co., Ltd., Nuclear Industry Research and Engineering Co., Ltd., Key Laboratory for Highly Efficient and Intelligent Welding, Beijing 101300, China
2. Tianjin Key Laboratory of Advanced Joining Technology, School of Materials Science and Engineering, Tianjin University, Tianjin 300350, China
3. Shaoxing Special Equipment Testing Institute, Shaoxing Key Laboratory of Special Equipment Intelligent Testing and Evaluation, Shaoxing 312071, China
Abstract：In view of the weld quality problems caused by unstable wire feeding in laser welding of ER316L stainless steels, this paper proposes an on-line monitoring method based on plasma emission spectrum diagnosis and builds a weld quality prediction model, which is of great significance to realize the adaptive control of welding process and intelligent laser welding. In order to further study the interaction mechanism between laser and welding material in laser welding, experiments of laser welding and laser wire filling welding were carried out. The laser-induced plasma’s spectral information was collected synchronously and compared with the arc spectrum of TIG welding process. The results showed that the spectrum during laser welding consisted of continuous spectrum and Fe Ⅰ 636.44 nm and Cr Ⅰ 427.48 nm line spectrum. During laser wire filling welding, the radiation intensity increased significantly, and many Cr Ⅰ lines were generated. The arc spectrum contained a large number of Ar Ⅰ and Ar Ⅱ lines and a small number of Fe Ⅰ lines. According to Boltzmann plotting and Stark broadening methods, the plasma electron temperature and electron density during laser wire filling welding were calculated. They were 5 024.9 K and 2.375×1016 cm-3, respectively, satisfying the local thermodynamic equilibrium state. The intrinsic relationship between laser welding quality and spectral features was explored on this basis. The results showed that the spectral line intensity and electron temperature strongly correlated with the weld quality. When the forming was good, the intensity of the CrⅠ spectral line was higher than that of the FeⅠ spectral line, and the electron temperature fluctuated steadily in a small range. The intensity of the CrⅠ line was lower than that of the FeⅠ line, and the electron temperature changed sharply when the bias defect occurred. Using the CrⅠ 529.83 nm spectral line intensity, FeⅠ 636.44 nm spectral line intensity and electron temperature as inputs, the weld quality classification model of a single hidden layer neural network was established to identify two states of well-formed and defects. The average accuracy of 10 tests was 88%. The t-distribution t-stochastic neighbor embedding was used to reduce the dimension of spectral data, and the three-dimensional embedding vectors were taken as the input features. The same neural network structure was used for weld quality pattern recognition, with an average accuracy of 97%. The results showed that the features obtained by dimensionality reduction of spectral data contained the information of line spectrum and continuous spectrum, characterizing the weld quality more accurately than the line spectrum selected by a human.
[1] Voisin T, Forien J B, Perron A, et al. Acta Materialia, 2021, 203: 116476.
[2] Wu D, Zhang P, Yu Z, et al. Journal of Manufacturing Processes, 2022, 75: 767.
[3] Gonzalez-Val C, Pallas A, Panadeiro V, et al. Journal of Intelligent Manufacturing, 2020, 31(3): 789.
[4] Cai W, Wang J Z, Jiang P, et al. Journal of Manufacturing Systems, 2020, 57: 1.
[5] Luo Y, Zhu L, Han J, et al. Mechanical Systems and Signal Processing, 2019, 124: 715.
[6] Huang Y, Hou S, Xu S, et al. Journal of Manufacturing Processes, 2019, 45: 642.
[7] Huang Y, Xu S, Yang L, et al. Journal of Materials Processing Technology, 2019, 271: 394.
[8] WANG Dong, ZHU De-cai, KANG Ze-tan(王 东, 朱德才, 康泽坛). Welding Technology(焊接技术), 2021, 50(3): 53.
[9] Turichin G, Zemlyakov E, Babkin K, et al. Physics Procedia, 2014, 56: 1232.
[10] Nilsen M, Sikstrom F, Christiansson A K, et al. Optics and Laser Technology, 2017, 96: 107.
[11] Chen Y, Chen B, Yao Y, et al. NDT & E International, 2019, 108: 102176.
[12] Lednev V N, Sdvizhenskii P A, Stavertiy A Y, et al. Spectrochimica Acta Part B: Atomic Spectroscopy, 2021, 175: 106032.
[13] Zhang Y, You D, Gao X, et al. Journal of Manufacturing Systems, 2019, 51: 87.
[14] Liu W, Liu S, Ma J, et al. Optics & Laser Technology, 2014, 57: 66.
[15] ZHAO Xiao-yan, YANG Li-jun, HUANG Yi-ming, et al(赵晓燕，杨立军，黄一鸣，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2021，41(8): 2397.
[16] Huang Y, Li S, Li J, et al. Measurement, 2019, 136: 405.
[17] Almodlej A, Alwadie N, Nessib N B, et al. Contrib. Astron. Obs. Skalnaté Pleso, 2020, 50: 147.
[18] Dimitrijevic M S, Ryabchikova T, Popovic L ( Cˇ ), et al. Astronomy & Astrophysics, 2005, 435(3): 1191.
[19] Wu D, Hu M, Huang Y, et al. Journal of Manufacturing Processes, 2021, 66: 153.
