Real-Time and Automatic Detection of Welding Joints Using Deep Learning

Welding technique plays a pivotal role in many industries, such as construction, automobile manufacturing, and nuclear power plants (NPPs). However, the shortage of skilled welding workers is still controversial due to the severe working environment and conditions. Therefore, to conserve human labor and improve manufacturing efficiency, an automated welding process is necessary. Also, welding efficiency and quality are vital indicators requiring attention for automatic welding. Notably, in NPPs, minor welding defects can occur serious safety issues. Therefore, our research's ultimate goal is to develop an automatic welding system to improve welding quality and manufacturing efficiency using visual sensors [e.g., a camera and light detection and ranging (LiDAR)], a robotic arm, and a welding machine. As the first step, this paper presents a method for automatically detecting different welding joints in real-time. Then, the different target joints are trained using a deep learning algorithm and detected by the camera. The results demonstrate the accuracy and effectiveness of the proposed method.