Maintenance Planning of Safety Protective Systems using Dynamic Bayesian Networks

Safety protective systems have two types of failure: Failed-Dangerous (FD) and Failed-Safe (FS). The former cannot responds to the system under an abnormal condition leading to a system accident, while the latter issues a spurious alarm leading to an unnecessary protective or maintenance action. Accordingly, consecutive data of no alarms can be interpreted in two different ways. One situation is that the system is abnormal with FD of the safety protective system, and the other is that the system is normal with normal safety protective system. Thus, 'non-alarm' condition does not always imply 'no system accident' condition. To prevent such a system accident due to the non-execution of system protective actions, this paper proposes a risk-based approach to determine whether a maintenance action should be taken or not. The expected loss caused by the non-execution of a maintenance action is estimated using a dynamic Bayesian network assuming the consecutive no-alarm data. An optimal maintenance interval can be obtained as the consecutive non-inspection interval if the expected loss caused by the execution is bigger than that caused by non-execution. An illustrative example of a simple voting system shows the property of the optimal maintenance interval.