In this paper, we propose to monitor the Fuel Metering Valve (FMV) of a jet engine, subsystem controlling the fuel flow. The proposed control strategy aims to design a virtual sensor based on robust on-line estimation for LPV systems using Takagi-Sugeno formalism. The purpose of the original approach is to provide performances required in the specifications in spite of occurrence of an unexpected fault during the operation. The robustness is guaranteed by a Neural Kalman Filter accommodating, in real time, the model prediction. In the proposed methodology, an off-line closed-loop identification scheme is first used to elaborate multi local linear state space models, after that a multi-model observer based on Linear Matrix Inequalities (LMI) optimization is used to build the virtual sensor. The NEKF is added to circumvent online model accuracy problems.The efficiency and limit of the approach are shown and discussed through simulations on a complete numerical engine test bench.

In order to minimize the occurrence of unexpected costly flight failures modern aircraft engines industry focuses especially on increasing product’s availability. In this work, we propose to monitor the health of a Variable Stator Vane (VSV), subsystem controlling the amount of airflow through the High Pressure Compressor (HPC), allowing optimum compressor performance. This control of airflow prevents the engine from stalling. The proposed methodology is based on an original approach for real time on-board Passive Fault Tolerant Control (PFTC). The objective of the proposed PFTC is to provide acceptable performance and preserve stability when faults occur. The method relies on the design of a specific Robust Virtual Sensor in a Linear Parameter Variable (LPV) polytopic framework. The robustness to model uncertainties is ensured by a Neural Extended Kalman Filter (NEKF) accommodating, in real time, the model prediction. In the proposed methodology, an off-line closed-loop identification scheme is first used to elaborate a multi local linear state space models, after that a multi-model observer based on Linear Matrix Inequalities (LMI) optimization is used to build the virtual sensor. The NEKF is added to circumvent online model accuracy problems. The efficiency and limit of the approach are shown and discussed through simulations on a complete numerical engine test bench.