| In view of the increasing emphasis on maintenance in the nuclear power industry, there is a need to develop analytical tools to evaluate maintenance policies and procedures. The field of operations research (OR) provides a rich source of analytical tools which potentially could be used to enhance existing maintenance practices. This dissertation develops a methodology and a computational scheme, based on dynamic programming (DP), to find the minimum cost schedule for a periodically maintained standby system. The DP formulation developed in the dissertation defines the component state as the number of unchecked maintenance periods and the system state as a vector of component states. The maintenance action consists of a check (inspection or test) followed by either full repair or replacement if a component malfunction is detected or restorative maintenance if no malfunction is detected. The maintenance action is assumed to be perfect and instantaneous. The maintenance action is assumed to return the component to a "as good as new" state, which is defined as the zero state. If no maintenance action is taken at a given period, the component state variable is incremented by one from its preexisting value. This formulation achieves a significant reduction of state space over which the optimization is to be performed. Further, the formulation allows a decomposition of the computational effort into two parts: a feasibility screening and an optimality search. A 5-component system is analyzed to test the viability and efficiency of the computational scheme. A full application of the scheme is demonstrated for a 10-component system, namely the high pressure injection system (HPIS) of a pressurized water reactor (PWR). The scheme is shown to be viable and efficient and particularly suited to exploit any economies of scale or scope that may be present with respect to maintenance activities. The scheme merits further development to extend its applications. The possible extensions of the scheme include incorporation of age-dependent failure behaviour, nonzero maintenance times and imperfect maintenance. |
