| Abstract Scope |
The current study focuses on the development of AI-driven control software that can monitor and control the defects forming during mass production of automobile components through the friction stir welding (FSW) process. FSW is preferred over other welding methods due to its ability to produce high-quality welding while avoiding common defects associated with conventional fusion welding. Although the FSW process produces a better-quality weld, attaining a defect-free weld strongly depends on several process parameters, such as welding speed, downward force, and tool rotational speed. Subsequently, incorrect parameters can be recognized by visually evaluating the corresponding surface defects, such as intensified flash formation, surface galling, and a few others. Human visual inspection for these defects is a costly process that requires the allocation of personnel. With the rapid development of vision sensing, artificial intelligence, and robotics technology, many excellent weld defect detection methods and models exist for identifying defects after welding. However, significant room remains for improvement in stability and accuracy. Real-time or in-situ tools for detecting weld defects are essential as they enable the identification of defects during the welding process, allowing for immediate corrective actions to minimize the consumption of resources in terms of material, energy, processing time, etc. This study employs a computer vision method to automatically detect irregularities on the weld surface and regulate the formation of defects by continuously adjusting the input parameters. The preliminary results of this study will serve as the first step for the future development of real-time monitoring and control, feedback loops, and optimization of parameters for FSW. |