Imbalanced Inter Shaft Bearing Fault Diagnosis Based on SAGAN-DANN

In the actual operation of aircraft engines, the early fault diagnosis of inter shaft bearings is extremely challenging due to their complex and variable working conditions and the difficulty in collecting sufficient vibration signals in advance. To address this issue, a cross-domain fault diagnosis model for inter shaft bearings based on small samples is proposed. This model first utilizes the Self-Attention Generative Adversarial Network (SAGAN) to augment limited vibration signals, capturing complex temporal patterns and structures through the self-attention mechanism. Secondly, it employs the Domain Adversarial Neural Network (DANN) to extract domain-invariant features through adversarial training. Finally, an intelligent diagnostic network is used to identify the health status of inter shaft bearings with unknown labels under variable working conditions. Experimental results demonstrate that the proposed method can accurately and effectively achieve cross-domain inter shaft bearing fault diagnosis with limited available samples, and performs excellently in cases of imbalanced source domain samples. This provides a new approach for early fault identification in aircraft engines.