This paper describes a quantitative methodology to estimate the probability of blade failure modes resulting from typical wear mechanisms in nuclear turbines, which can be used to optimize maintenance. The approach used to model time and spatial dependence of wear mechanisms that affect blades involves the coupling of a Static Bayesian Network to a Dynamic Bayesian Network. This prototype model has been designed to use conditional and time dependent Weibull-like failure rates that can be computed from reliability data bases (failure times and modes, associated causes, row and blade part that failed) to quantify Markov matrixes contained within dynamic nodes.
The model can be used to make inferences such as the most probable causes of failure in a row and blade part, and visualize the probability as a function of time. It can be also used to determine the riskier location given evidence such as failure mode or the wear mechanisms involved. Also, maintenance tasks acting over time dependent failure functions have been implemented to exemplify the effect of perfect and three kinds of imperfect actions and how they affect the mechanisms and failure mode evolution, given the conditional dependences among them.
