CLASSIFICATION OF AERO-ENGINE BLEED AIR STATES BASED ON WGAN-GP DATA AUGMENTATION

Aimed at the issue that the data driven method cannot be effectively applied due to the lack of real abnormal flight data over the aero-engine bleed air system, an engine bleed air states classification method based on data augmentation by Wasserstein generative adversarial networks with gradient penalty (WGAN-GP) is proposed. By enlarging and balancing the training dataset with WGAN-GP, the convolutional neural network (CNN) for states classification could learn richer data features, being able to accurately classify multiple states of engine bleed air. The experimental results indicate that the proposed model, with greater stability than the classic generative adversarial networks (GAN), can generate samples closer to the originals, and that the classification performance is also significantly improved.