关键词: 非负矩阵分解, 图像识别, 焊接熔池, 焊接缺陷, 熔化极气体保护焊

Abstract: To explore the technology of visual sensing intelligent identification of welding defects,the continuous images of welding pool in Gas Metal Arc Welding (GMAW) were acquired by Charge Coupled Device (CCD) to research the behavior of the welding pool as well as the corresponding welding defects.The nonnegative matrix factorization algorithm was used for feature extraction of the welding pool images.The recognition value of tested image in feature matrix was calculated by least square method.The automatic identification method of welding defects was given.The results showed that the welding quality was related to the stability of the weld pool.Specifically,the unstable weld pool was characterized by profile fluctuation,welding slag dispersion and so on.The decrease of molten pool stability was accompanied by the decrease of welding quality,which leaded to appear the weld defects.The feature matrix of weld pool image based on nonnegative matrix factorization algorithm could describe the original images in general (such as weld pool profile) and in part-base (such as weld slag profile),which had physical interpretability and could be used to identify welding defects.

Key words: non-negative matrix factorization, image recognition, welding pool, welding defect, gas metal arc welding