Turbine engines are frequently used in critical systems including the power plants and propulsion systems of aircrafts and ships. Frequent inspection and periodic maintenance have been necessary to ensure their proper functionality. Condition based maintenance of jet engines can significantly reduce their operational and maintenance costs, and, in the mean time, enhance safety and reliability. This study investigates the feasibility of establishing the utility of a dynamic network, i.e., projection network, to recognize hot air pass faults from measurements of e.g., fan speed, core speed, compressor inlet and exit temperatures and pressures, turbine exit temperatures, etc. Projection network is a nonlinear dynamic network architecture that provides stable oscillatory or non-oscillatory attractors. In contrast to the static mapping provided by e.g., neural networks and fuzzy systems, the projection network offers more functionality through its rich dynamics. When properly setup, its nonlinear dynamics can filter out noise from measurements, and classify/recognize complex patterns. This study established the utility of projection network for detection and diagnosis of several aircraft engine faults. This paper will also describe methods for both structure training and parameter tuning. Using these methods, projection networks were setup to recognize baseline, fan damage, high pressure turbine fault, and customer bleed valve fault.

This content is only available via PDF.
You do not currently have access to this content.