The nuclear industry has recently been shifting to value-based maintenance in order to keep nuclear power competitive in the power generation market. A key challenge in value-based maintenance is the optimization of a maintenance schedule. With most components having ten to fifteen available maintenances, the complexity of the optimization grows quite quickly. This paper presents a methodology for combining maintenance effectiveness, cost estimates, failure impacts, and overall reliability data to estimate an expected life cycle cost (LCC) for a component. The maintenance types are categorized into three types: monitoring, wear-rate reducing (e.g. oil change), or life-restoring (e.g. refurbishment). Each maintenance type has a different effect on problem detection, degradation rate, and future life expectancy. A Markov model uses the maintenance effects to estimate the distribution of what state of degradation a component is in at a specific time and concurrently, the component failures, maintenance costs, and failure impacts are tallied up in order to provide an expected life cycle cost for a given maintenance schedule. Optimization of the maintenance schedule is performed using the genetic algorithm where multiple maintenance schedules are simultaneously calculated, compared, and evolved in order to find the lowest expected life cycle costs. The genetic algorithm was selected as a suitable optimization algorithm for its ability to find relatively close approximations to the global optimum with relative ease while concurrently being able to handle non-smooth objective functions.

This content is only available via PDF.
You do not currently have access to this content.