Evaluation for Angular Distortion of Welding Plate

Welding is essential in our life. It is crucial to nurture welding skills in Japan nowadays. The experts have to evaluate the many beginners' welding. Since the experts' burden is critical, a computational assistant for evaluating beginners' welding is required. This paper describes a simple evaluation system of welding plates by beginners. The authors considered four types of beginners' typical defects: lack of welding metal, linear misalignment, welding metal unevenness, and angular distortion. To capture these defects simultaneously, the authors propose an original equipment to photograph the welding plates. The computer extracts only the part of the welding plate using color markers. CNN (Convolutional Neural Network) evaluates the defects. As a first step, the authors addressed evaluating only angular distortion. The angular distortion is one of the typical failures by beginners. In the experiment, the authors conducted the validation of CNN. In the conclusion part, we discuss the experimental result and future works.