Extreme Learning Machine-Based Diagnostics for Component Degradation in a Microturbine

Abstract

Micro turbojets are used for propelling radio-controlled aircraft, aerial targets, and personal air vehicles. When compared to full-scale engines, they are characterized by relatively low efficiency and durability. In this context, the degraded performance of gas path components could lead to an unacceptable reduction in the overall engine performance. In this work, a data-driven model based on a conventional artificial neural network (ANN) and an extreme learning machine (ELM) was used for estimating the performance degradation of the micro turbojet. The training datasets containing the performance data of the engine with degraded components were generated using the validated GSP model and the Monte Carlo approach. In particular, compressor and turbine performance degradation were simulated for three different flight regimes. It was confirmed that component degradation had a similar impact in flight than at sea level. Finally, the datasets were used in the training and testing process of the ELM algorithm with four different input vectors. Two vectors had an extensive number of virtual sensors, and the other two were reduced to just fuel flow and exhaust gas temperature. Even with the small number of sensors, the high prediction accuracy of ELM was maintained for takeoff and cruise but was slightly worse for variable flight conditions.

Gas turbine aero engine fault detection using Geo-TLSVM and digital twin with multimodal data analysis

“…The GaT-AE is the combustion engine that converts chemical energy into mechanical energy to drive the engine and propeller [2,7]. But, the GaT had some faults like mechanical breakdown, gas path fault, etc [8]. The GaT-AE is regularly monitored and analyzed for FD in GaT-AE [9].…”