Abstract
In a petroleum refining plant, the high-pressure gas leaks resulting from equipment failures may lead to disasters. To minimize these disasters, technology for the early detection of leak sound is indispensable. We assumed that background noise in a petroleum refining plant is a steady sound generated by various devices under normal conditions; there is no dominant sound, on the other hand high-pressure gas leak results in a dominant sound. We have employed chaos theory to identify these dominant sounds, and have already reported these results in papers listed in reference 1 and 2. However, the leak sound is not always more regular than the background noise depending on measuring places or leak conditions. In order to detect the high-pressure gas leak sound, it is necessary to estimate the steady state range quantitatively employing chaos theory from the background noise for every measuring point, and examine whether the observed sound is out of the range or not: the observed sound is more regular or more irregular than the background noise. Based on the concept stated above, employing chaos information criteria, we conducted the leak sound detection experiment by leaking steam artificially using a silencer nozzle near the benzene recovery unit in Idemitsu Kosan Chiba Refinery. This paper describes the leak sound detection algorism, the outline and results of experiment.
