Abstract: To overcome the weakness of dependence on the accurate model, a method using self-organizing neuralnetworks for aeroengine fault diagnosis is developed.Because self-organizing neural networks can be trained to adjusttheir own structures and eventually identify the faults only with the sensed data, the precise engine model is not necessary. With the help of the data preprocession,which extracts the features of faults, the neural networks finally getthe correct results of fault diagnosis of aeroengine gas path components. Then the autoassociative neural network isused to test the noise rekection abilities of the self-organizing networks. Since they can work without the enginemodel and are robust for noise rejection, the self-organizing neural networks are able to meet the application requirements.

Key words: self organizing neural networks, aeroengine, fault diagnosis, autoassociative neural networks, aeroengine model