Prognostics under Different Available Information

In this work, we consider two practical situations with different information available, concerning the prediction of the Remaining Useful Life (RUL) of a creeping turbine blade for which a sequence of observations of the creep strain level is available. In the first case considered, we have available a stochastic model of the creep growth process and we know the value of the failure threshold, i.e., the maximum creep strain level beyond which the blade cracks. On this basis, a Monte Carlo-based filtering technique, called particle filtering, is set-up to predict the distribution of the system RUL and online-update it when new observations are collected. In the second case considered, the only available information is the sequence of observations of the creep strain of the blade of interest and the value of the failure threshold. On this basis, a data-driven method, based on an ensemble of bootstrap models, has been developed to estimate the turbine blade RUL and the uncertainty of the estimate caused by the uncertainty in the data, the variability of the blades behaviour and the imprecision of the empirical model. The two approaches are evaluated in terms of the assumptions they require and the accuracy of the RUL predictions they provide. The ability of providing measures of confidence in the outcomes is also considered.