Takashi Ohnishi*, Yoshikazu Akaizawa*, Toshimasa Shimizu* and Tomimasa Yonezawa*

* Yokohama Research and Development Center, Mitsubishi Heavy Industries,LTD.

+ Correspondence should be addressed to Takashi Ohnishi:
(12, Nishiki-cho, Nakaku, Yokohama 231 Japan)

Abstract

In a bulky refuse disposal facility, dangerous articles such as fuel gas cylinders are removed by hand to prevent accidental explosions. However, it is very difficult to detect perfectly all dangerous articles which get mixed into various refuses. Especially, the detection of fuel gas cylinders is the most important, to assure the integrity of facilities and the security of operators. As refuses are complex and different from each other in shape, weight, components, etc., the concept of the conventional sensor system, which utilizes a deterministic relation between sensor outputs and properties of the object, can hardly apply to the detection of gas cylinders. In the present paper, we propose an automatic gas cylinder detection system which conteins a set of sensors and an inference block. The sensing system is composed of various kinds of sensors, and the refuse is characterized by fuzzy sets according to their output. The inference block infers the possibility of the refuse being a dangerous article, using fuzzy inference rules and neural network in parallel. The authors present test results of the detection system using real bulky wastes, and discuss the subject of practical application of the system.

Key words: bulky refuse, dangerous article, fuel gas cylinder, fuzzy inference, lassification of refuse