High reliability (and availability) with low life-cycle costs are general goals for all maintenance programs. An effective Preventative Maintenance (PM) regime balances the cost of maintenance with the elimination of degradation failure mechanisms through preemptive intervention. Failures in operation require corrective maintenance (CM) and are often the most expensive to repair. PM is conducted to reduce the probability of specific failures and hence reduce the CM burden and cost. However, PM actions can themselves introduce additional damage and failure mechanisms. Determining the optimum PM frequency requires these competing factors to be quantified. Too little PM allows subsystems and components to wear out, decreasing overall system reliability. Conversely, too much PM will introduce an inordinate amount of damage (and failure mechanisms) that decreases system reliability, along with increasing the overall cost. In order to optimize maintenance, the failure modes of the individual components need to be analyzed and the maintenance program matched to how the equipment fails, how predictable the failure is and what the overall impact the failure has to the mission of the system it is a part of.

No abstract

Electrical utility reliability indices, such as SAIFI, SAIDI, CAIDI, CAIFI, and MAIFI are adequately defined in IEEE Std. 1366-2012 [1] and serve the purpose for which they were developed. However, the explosive growth of the electronics industry moving toward an "internet of things," along with the data center development necessary to support such explosive growth has had significant impact on an aging power distribution system. As the utility world evolves toward the notion of "smart grid," several new reliability metrics are needed to evaluate the electrical power to the customer's side of the meter.

PowerSecure (PS), a subsidiary of Southern Company, operates over 1600 microgrids including a large fleet of standby diesel generators. Machines range from 125 to 2800 kW. There are two missions: standby power after utility failure and load management when utility is available. This report updates generator reliability data published over 20 years ago. These results are first derived from uniform records from a distributed data acquisition system. The unique aspects of this effort include automated data collection, analysis, and storage producing a standardized record for all machines and events. This greatly reduces the effort required to prepare reports and analysis. The automated collection largely eliminates human errors in data entry, and prevents post hoc adjustment of mission success or failure. We are not aware of previous published generator data using automated data acquisition. This study counts all failures and excludes none, whereas previous studies exclude more than half. This analysis expresses reliability as experienced by the customers. The number of machines, years in service, and years of operation greatly exceed the sum of previous published studies, enabling calculation of useful confidence limits. The fleet is heterogenous with machines from over a dozen manufacturers and many power ratings. Machine reliability was consistent across brands, and there is no significant difference in reliability for different size machines. This common assumption was not supported by the prior studies. High-quality data enables reliability growth management practices. The fleet reliability for outage demands of all durations increased from 95% in 2011 to 98% in 2014-18. Analysis of failures arrival time shows that the failure rate is strongly dependent on mission duration. The failure rates after 14 h are approximately 1% of the value for the first 30 min of operation. This has important implications for designing systems to meet specified reliability targets, and in interpreting and comparing results from fleets with different mission durations.

No abstract