Gas turbine fault detection and isolation using mlp artificial neural network

Resumo

This work deals with a nonlinear model, based on a particular form of artificial neural networks, ANN, for application to gas turbines fault diagnosis. The traditional multi-layer perceptron (MLP) is used, with error backpropagation and different activation functions. The application of the model is illustrated using test data from a gas turbine simulation computer program. A specially developed computer program is used to simulate the engine in operation, generating all needed engine data for both baseline and deteriorated engine. A test case using a turboshaft engine is used to demonstrate the capacity of this ANN to identify faults that may occur during engine operation. Copyright © 2007 by ASME.