Aero-engine faults diagnosis based on K-means improved wasserstein GAN and relevant vector machine

The aero-engine faults diagnosis is essential to the safety of the long-endurance aircraft. The problem of fault diagnosis for aero-engines is essentially a sort of model classification problem. Due to the difficulty of the engine faults modeling, a data-driven approach is used in this paper, based on the Relevance Vector Machine for classification. However, the collection of the fault sample is so difficult that causes the imbalance learning problem. To solve this problem, a semi-supervised learning approach based on the Improved Wasserstein Generative Adversarial Networks and K-Means Cluster technique is proposed in this paper. The theoretical analysis and the experiment show that, compared with another sampling method synthetic minority oversampling technique (SMOTE), the proposed approach can better fit the fault sample distribution, generate much more appropriate new samples by learning from the small number of fault samples. It is more efficient to prevent over-fitting by training with the original samples that mixed with the Improved Wasserstein Generative Adversarial Networks generated samples.