Turboshaft Engine Condition Monitoring by Bayesian Identification

The application of a Maximum Likelihood Estimate technique has been implemented in an Engine Condition Monitoring framework for a small turboshaft for power generation purposes. The turboshaft has been modeled in a fully non-linear way, by using actual turbomachine performance maps obtained from the manufacturer: the accuracy of the simulation proved to be very good with respect to real operating data. The model was used both to generate sample synthetic dataset (by adding Gaussian noise to the selected outputs-measurements) and as the core computational engine in the identification process. The results obtain show the very good robustness of the proposed identification process, and its capability of dealing with noisy or even malfunctioning transducers. This capability is provided by the possibility of determining a mathematically sound statistical framework, which is not only capable of identifying the most likely fault configuration but also to indicate the confidence level with which the identification is performed.