Failures of turbomachinery are often caused by the dynamic behaviour of rotating blades in these machines. The financial and production implications of such failures may be very significant and appropriate blade condition monitoring methodologies are therefore of critical importance [2]. Blade tip timing (BTT) is a non-intrusive measurement technique for online measurement of turbomachine vibration. Essentially it senses when a blade passes a number of proximity probes distributed circumferentially and mounted radially through the turbomachine casing above the row of rotor blades being measured, to determine the time of arrival. This can be linked to the blade vibration by employing an accurate measure of the once per revolution reference signal. The technique is non-intrusive and online monitoring is possible.

BTT is therefore often regarded as a feasible technique to track the condition of turbomachine blades, thus preventing unexpected and catastrophic failures. The processing of BTT data to find the associated vibration characteristics is however non-trivial. In addition, these vibration characteristics are difficult to validate, therefore resulting in great uncertainty of the reliability of BTT techniques. To deal with the uncertainties of the method, various new concepts have been introduced [2,3]. These ideas deal primarily with diagnosis. Techniques for prognosis to assist with maintenance decision making is however becoming more important. Mishra et al. [4] explored a range of techniques of interest to accomplish this through the use of hybrid models that merge physics based and a data driven approaches into a unified approach.

This idea is pursued further in the context of turbomachinery blades, by proposing an approach comprising a stochastic Finite Element Model (FEM) based modal analysis and a Bayesian Linear Regression (BLR) based BTT technique. The use of this stochastic hybrid approach is demonstrated for the identification and classification of turbomachine blade damage. For the purposes of this demonstration, discrete damage is incrementally introduced to a simplified test blade of an experimental rotor setup. The damage identification and classification processes are further used to determine whether a damage threshold has been reached, therefore providing sufficient evidence to schedule a turbomachine outage. It is shown that the proposed stochastic hybrid approach may offer many short- and long-term benefits for practical implementation.

In the present work, some limitations of existing work [5] are critically discussed and further refinement of the methodology is explored.