Detection of variable-groove weld penetration based on cooperative awareness of melt pool vision and temperature field

• Published in: Optics and laser technology 2025-06, Vol.184, p.112432, Article 112432
• Main Authors: Yu, Rongwei, Zhang, Tianyang, Huang, Yong, Peng, Yong, Wang, Kehong
• Format: Article
• Language: English
• Subjects: Melt pool vision; Multi-information fusion; Penetration status; Temperature field; Variable-groove weld
• ISSN: 0030-3992
• DOI: 10.1016/j.optlastec.2025.112432

•The visual morphology and temperature distribution features of melt pool can be extracted from colour melt pool image.•A penetration state detection model for variable-groove weld seam is developed based on multi-information fusion.•The multi-information fusion-based detection model demonstrates high-precision detection for various penetration states. Welding of a variable-groove workpiece is challenging. A change in the groove angle of the workpiece can affect the weld penetration status, which, in turn, affects the forming quality of the workpiece. This study focuses on the development of an online detection technology for the penetration status of variable-groove welds and proposes a penetration status detection technology using cooperative awareness of the melt pool vision and temperature field. First, a vision sensor system for the melt pool was developed using a colour charge-coupled device (CCD) that collects high-quality pool images online. Second, the visual morphological features of the melt pool were extracted using image processing methods. In addition, the melt pool temperature field was measured using a colorimetric temperature measurement method, thereby extracting the temperature field distribution features of the melt pool. Finally, feature layer fusion was performed on the visual and temperature signals of the melt pool, and a detection model for the penetration status of variable-groove welds was developed using an artificial neural network. We verified the advantages of the multi-information fusion detection model, analysed the importance of multiple types of features, and validated the generalisation ability of the model. The test results showed that the measurement accuracy of the multi-information fusion-based variable-groove weld penetration status detection model established in this study was higher than 92%. In addition, the model has a powerful generalisation ability, which can provide effective technical support for the intelligent detection of the penetration status of butt welds.