Prediction of weld back width based on top vision sensing during laser-MIG hybrid welding. (December 2022)

• Record Type: Journal Article
• Authors: Ye, Guangwen; Gao, Xiangdong; Liu, Qianwen; Wu, Jiakai; Zhang, Yanxi; Gao, Perry P.
• Is Part Of: Journal of manufacturing processes. Volume 84(2022)
• Page Start: 1376
• Page End: 1388
• Publication Date: 2022-12
• Subjects: Laser-arc hybrid welding -- Image processing -- Weld back width prediction -- Long and short term memory neural network -- Attention mechanism; Production management -- Data processing -- Periodicals; Manufacturing processes -- Periodicals; Procestechnologie; Productietechniek; Production -- Gestion -- Informatique -- Périodiques; Fabrication -- Périodiques; Manufacturing processes; Production management -- Data processing; Periodicals; 670.5
• DOI: 10.1016/j.jmapro.2022.11.021
• Languages: English
• ISSNs: 1526-6125

Abstract

Abstract: Online welding quality monitoring has gained increasing attention in modern automatic production. This work proposed a prediction method of weld back width based on top vision during laser-MIG (Metal Inert-Gas) hybrid welding. A high-speed photography system was used to monitor the laser-MIG hybrid welding, which could observe the visual information from top of the weldment. Visual information including the features of keyhole, arc and molten pool, were extracted as model inputs by image processing and signal processing, and the weld back width was extracted as model output by 3D (Three-Dimension) point cloud processing. An ATT-LSTM (Attention - Long Short Term Memory) prediction model was constructed to predict the weld back width. Experimental results show that ATT-LSTM can not only improve the accuracy and generalization ability of weld back width prediction but also deliver interpretable analysis, compared with other methods such as RNN (Recurrent Neural Network), LSTM, GRU (Gated Recurrent Unit), LSTM-ATT, which MSE (mean square error) was 0.0848 mm 2, R 2 (r-squared) was 0.5900, CORR (correlation coefficient) was 0.7694 and the comprehensive prediction error (2 σ ) was 0.5853 mm. The most attention was focused on the keyhole feature, and the feature information reflecting droplet transfer and laser-arc coupling stability was also discussed. Highlights: Weld back width was predicted by top visual information. Features of keyhole, arc and molten pool were extracted from visual sensors. Features of droplet transition were extracted by time-frequency analysis. The proposed prediction model could deliver interpretable analysis.