Neural Network Based Diagnostic Indication of a Liquid Propellant Engine with Faulty Data Collection System

The aim of the paper is to describe a methodology of damage detection in the liquid propellant engine which is based on artificial neural networks in combination with stochastic analysis. It is assumed that the liquid propellant engine have faulty data collection system. Then a filtering algorithm for elimination perturbation data has been applied. The damage is defined as fuel and oxidizer channels clogging up. The key stone of the method is feed-forward multi layer network with back propagation algorithm. It is impossible to obtain appropriate training set for real engine, therefore stochastic analysis using mathematical model is carried out and dynamic simulation is made to get training set virtually. Engine channels clogging up leads to unwanted variation of pressure, flow rate of oxidizer and fuel and other main parameters of engine. Then variations considered as best input data for damage detection. The methodology was carried out using laboratory test.