Synergistic Feature Fusion With Deep Convolutional GAN for Fault Diagnosis in Imbalanced Rotating Machinery

In rotating machinery, accurate fault diagnosis is crucial for efficiency and preventing failures. Traditional models often struggle with imbalanced datasets. This study introduces strategies that use feature fusion deep convolutional generative adversarial network (DCGAN) architectures to improve fault diagnosis accuracy. Initially, we pretrain the DCGAN using a comprehensive dataset encompassing various general faults to robustly capture the underlying features. Then, we use rare fault samples to refine the DCGAN, enhancing its capability to extract features from these minority classes. Random noise is input into the feature fusion deep convolutional generative adversarial network (FFDCGAN) model to obtain pseudosamples of the rare faults. The generated faults are then combined with the original dataset and analyzed by a convolutional neural network to classify fault types. Based on experimental results from the ZHS-2 and HIT aero-engine fault datasets, comparative analysis with existing studies shows that the proposed FFDCGAN method generates samples with significantly greater diversity. In addition, the proposed imbalanced fault diagnosis approach achieves higher accuracy, thereby validating its efficacy in handling imbalanced datasets.