It is essential, in beer bottling plants, to reduce the ratio of defective products. One of the best ways to realize it is to detect the abnormally filled bottles as faster as they are filled, and find out which filling valve caused them. There have been a couple of technologies proposed to inspectt products at the last of the manufacturing process; however, there is none that can actualize the above requirement.
We propose, in this paper, a new method, which enables measuring the filling precision of each valve. It employs computer vision to gauge the filled level, and sensors to relate each bottle to the valve that has filled it. The problems of gauging are the frothed bubbles and the rough surface caused by jolts of the conveyer. In order to decrease the turbulent influences, we propose to apply statistical procedures to the measured data.
Through experiments at a test plant and an actual one, we confirmed that the distribution of measured data was approximately normal as we had hypothesized, and the mean value was quite stable. In addition, we verified that the measurement was accurate. As a result, we conclude that our proposed technique is quite effective.
抄録全体を表示