A novel surface temperature sensor and random forest-based welding quality prediction model

Temperature variation directly affects the melting and solidification process of welding and has a significant impact on weld quality and mechanical properties. Accurately acquiring real-time temperature variations during the welding process is crucial for the real-time detection of welding defects. In this study, a novel thin-film thermocouple (TFTC) sensor that offers fast response, easy installation and no damage to the temperature measurement surface was designed and developed to obtain real-time temperature variations during the metal inert gas (MIG) welding process of aluminium alloys. A random forest-based weld defect identification model was established with an accuracy of 97.14% for the four typical defects of incomplete penetration, nonfusion, undercutting and collapses, which occur in the three-layer, three-pass welding process. Subsequently, a random forest model based on the temperature signal was used to analyse the hardness, bending and tensile properties of the welded joints, demonstrating the feasibility of directly using the weld temperature signal to assess the mechanical properties of welded joints.