Alleviating Limited Dataset in Inter-Shaft Bearing Fault Diagnosis based on Multi-source Information with Information Discriminator

To alleviate limited dataset on the performance of intelligent fault diagnosis methods, this paper proposes a multi-source information with an information discriminator (MSI2D). Specifically, a multi-source information network with an information discriminator is designed to extract time-frequency representations from the transformed images, which can improve the input features while enhancing the model generalization performance. Moreover, we discovered that a single pooling method produces the limited feature outputs, leading to inaccurate fault diagnosis and identification. To address this, we present multiple pooling operations after feature fusion, which not only gathers more representative and finely grained features, but also improves diagnostic performance. Comprehensive experiments based on a real aeroengine validate the proposed MSI2D outperforms existing methods, striking a good trade-off between the limited data and diagnostic performance. It is shown that our MSI2D surpasses the comparative methods in accuracy.