Adaptive choice of a safety management scheme upon an alarm under supervisory control of a large-complex system

Abstract This paper gives a probabilistic analysis of safety management schemes which should be used upon an alarm in a large-complex system under supervisory control configuration. Probabilistic models are developed for safety management schemes, where the importance of distinguishing two types of safety-control policies (the safety-preservation and the fault-warning safety-control policies) is stressed. Preference order relations among safety management schemes are analyzed by evaluating the conditional expectation of system damages, given that an alarm has been generated. It is proved rigorously that it is unwise to fix beforehand in an off-line manner a single safety management scheme so that the scheme is activated every time an alarm is given. An adequate safety management scheme must be chosen in an adaptive manner from a set of available safety management schemes, because an optimal safety management scheme varies depending not only on a given situation but also on the time-point when an alarm is given.