Use of performance-monitoring to improve reliability of emergency diesel generators

Abstract

Emergency diesel generators are one of the most important contributors to the core damage failure rate of nuclear power plants. Current required testing and maintenance procedures are excessively strict and expensive without any real justification. Probabilistic risk assessment is used to propose a monitoring system and Technical Specification changes to reduce EDG unavailability without jeopardizing safety, and to ease the excessive deterministic requirements. The EDG fault tree is analyzed to identify the critical failure modes of the EDG, the failure of service water pumps, the failure of EDG building ventilation dampers, and the failure of the EDG "supercomponent," which includes the fuel oil, lubricating oil, cooling water, and starting air systems. We use data from the nuclear industry and the U.S. Navy to identify the most significant EDG supercomponent failure modes, including system fluid leakages, instrumentation & controls failures, electrical power output failures, and the fuel system governors. The monitoring system proposed includes instrumentation for twenty-one of the 121 basic events in the fault tree, for a total of 94.9% of EDG failure contributions. The failure modes identified with industry data are monitored, as are diesel engine mechanical failures currently assessed with teardown inspections. With a 50% reduction in these twenty-one basic event failure rates, the EDG system failure rate is reduced by 41.6%, from 0.097 per year to 0.059 per year. With this reduced failure rate, we propose to extend the EDG surveillance interval from one month to twelve months, to lengthen the running tests from one hour to twenty-four hours, and to eliminate the tear-down inspections conducted during refueling outages. To fully assess the benefits of these proposed changes, the monitoring system should be installed on an EDG on a trial basis. The work reported here demonstrates the feasible gains which can be realized, and proposes, a method for evaluating the efficacy of the system as realized through experimentation.