部品の変色検査

抄録

At present, the pass/fail judgment of mechanical parts commonly used are manual detection. However, human inspection is expensive and not suitable all times. On the other hand, it is expected that computer vision systems will enable long-term detection and accurate judgment. The computer vision systems using mechanical learning have been investigated experimentally and numerically. Their investigation has primarily focused on image recognition. Since some mechanical parts have complicated shapes, they are sometimes conveyed in a state where they are partially overlapping with the preceding parts. In this paper, an actual alignment system for image recognition was proposed. In this paper, we propose a feeding device for image recognition. This device consists of a drum-type parts feeder and a linear feeder with a rail that has a hole. Any parts that are not in the correct position are dropped through the hole and returned to the original feeding area. This feeding device was tested to determine how the distance between parts that can be successfully recognized by the image recognition system changes with the number of drum rotations.