Robust Sensor Selection for Model-Based Aeroengine Diagnosis

The present study addressed the problem of the robust selection of sensors and their actual location on the aeroengine for diagnosis purposes. The objective was to infer the health status of the engine components from the reading of the pressure and temperature sensors. The method was model-based. Its development was illustrated via a direct application to the GE CF6-80 × 101 aeroengine. The novelty of the study lies in its methodological approach, namely, the use of a high-fidelity aeroengine model and a tensor-based algebraic formulation, rather than an AI-based method, for the inverse diagnosis problem. The baseline aeroengine model, formulated via the Proosis software, had nine inputs (eight health parameters plus one control parameter) and eleven outputs (thermodynamic cycle sensors’ readings). The model was used to generate an engine database in tensor form and a surrogate model which was the input for the diagnosis algorithm. The proposed methodology enabled the identification of sensors that are individually critical, as well as sensor subsets within which individual sensors can be removed without compromising the diagnostic accuracy. Another result was that, if the appropriate sensors were selected, the multi-point diagnosis could be used in situations with fewer sensors than health parameters. The global conclusion was that the proposed methodology could be used for generic aeroengines, even if, in the present study, it was illustrated by the application to a particular aeroengine.