Abstract
To prevent system failure or accidents, it is essential to detect their symptoms in their early stage. In this sense, the role of a safety monitoring system is very important. Further, in the safety monitoring system, two types of failure must be considered: failed-dangerous and failed-safe. The former does not issue alarms when it should, leading to a serious accident, while the latter issues alarms when it should not, causing unnecessary maintenance cost. This paper proposes a general framework for the design of a risk-based safety monitoring system of a dynamical system based on the current monitored data, which minimizes the expected loss caused by failure of the safety monitoring system. The overall system is represented by a dynamic Bayesian network, which can consider both transitions of component states such as one from its normal state to failed state, and cause-effect relations among components. Based on the monitored data, the dynamic Bayesian network updates state probabilities and shows their transition. Using this information, the risk caused by the diagnosis decision can be evaluated to determine the optimal logic structure. An illustrative example of a safety monitoring system of a simple dynamical system shows the details and merits of the proposed method.
