Aero-engine sensor fault diagnosis based on stacked denoising autoencoders

In this paper, SDA (stacked denoising auto-encoders) model is introduced to solve the aero-engine sensor fault diagnosis problem. This model uses the principle of greedy layer-wise unsupervised training to initialize the network weight, then uses backpropagation algorithm to optimize the network parameters. This model applies SDA to extract features from aero-engine sensor fault signal and softmax classifier to diagnose fault based on these features. The result of sensor fault diagnosis experiment shows that this approach can make the accuracy of aero-engine sensor fault diagnosis achieve 99.286%.