Distributive watch using situation recognition

A novel method is presented for finding tendencies toward an accident in a large-scale plant. Modules are proposed as cooperative agents which perform a task to find tendencies. The task is distributed over many modules. Some of them find local tendencies in a manufacturing process and others find tendencies over several manufacturing processes. As the amount of information is tremendous in the case of finding tendencies over several manufacturing processes, skeletons are introduced to extract information relevant to the tendency to be found. The author develops a programming language, Soul and a virtual machine, SAMURAI, to implement modules. The distribution of the task over many modules reduces the complexity of the task. The extraction of information relevant to the tendency reduces the complexity of the information