Although the main steam turbine is not a safety-related system in NPPs, a worst-case scenario considers that catastrophic failure might result in flying objects (missiles) striking surrounding safety systems, thus increasing core damage frequency. However, these cases are remote. Son-catastrophic failures that result in turbine shutdown or power reduction are more frequent issues, sometimes activating safety systems, affecting the balance of plant and power generation capability. Economic issues include repair and replacement, as well as losses due to lack of power generation and plant restarts. The key elements within turbines are the blades. They are designed to create momentum from their interaction with steam flow, so they are highly stressed by a complex combination of forces, as well as by environmental conditions that induce several degradation mechanisms dependent on the turbine pressure stage. The inherent complexity and stochastic behavior of these mechanisms has resulted in a wide variety of approaches and methodologies that have been used to predict blade-failure probabilities; however, neither an established nor a preferred method has resulted. This remains true in spite of statistics that show approximately half of the hours of generation lost in PWR and BWR turbine issues are related to blade problems, and the significant amount of literature related to actual failure event post-mortem analysis. We report progress toward creating a quantitative methodology that allows the analyst to estimate the probability of blade failure-modes caused by typical degradation mechanisms in nuclear turbine units. The method also takes into account the effect of possible maintenance tasks as a way of optimizing the strategies based on the associated costs. The mechanisms and their failure modes, which affect nuclear turbine blade integrity, include pitting, droplet erosion, fatigue, corrosion fatigue, stress corrosion cracking, and fretting. It has been found that from a probabilistic perspective these mechanisms have a conditional behavior that can be described by a Bayesian Network. There are causal relationships between them (e.g., the phenomenology dictates that when pitting is found in a blade, the probability of corrosion fatigue increases) that can be estimated from turbine reliability databases. A prototype network has been constructed as a first qualitative approximation It is expected that introducing specific plant data from studies. inspections, and or nondestructive reports, failure modes can be computed as ending nodes of the network and vice versa; that is, if a failure mode occurs, then the most likely set of causes is revealed. The model described here will help optimize maintenance strategies to reduce costs.
展开▼
