Aircraft engine on-board diagnostics based on neural network

Abstract

It is offered to employ the neural network (NN) technology in the software of on-board aircraft engines automatic diagnostic system. But first it was necessary to solve the problem related to teaching NN the ability to adequately and certainly evaluate critical shift of an aeroengine technical condition in flight, including engine monitoring system, e.g. in case of one of data channel fault.
Current research recommends to move from trend analysis of gasdynamics parameters measured on the engine during flight to numerical analysis of engines air-gas path defects criteria using neural networks trained by using diagnostic matrix (DM) specifically worked out for this particular engine model with the purpose of deeper diagnostics of fitted on Fokker-50 PW-125B turboprop engine two-shaft gasgenerator. Based on the research results there is recommendation to introduce in the operations the set of neural networks, trained with abbreviated DM, with defined number of input data which will allow to perform more reliable engine diagnostics during flight in real time mode. Also it is noted that for two-shaft gasgenerator 4 parameters are the minimum parameters number below which the diagnostics become unreliable.