Abstract As a critical piece of industrial production equipment, the operational status of an air compressor directly affects the success of production. However, traditional fault diagnosis methods struggle to accurately obtain fault characteristics. The feature distribution differences between different working conditions are not sufficiently measured by domain adaptation, making it difficult to achieve high recognition accuracy. Additionally, background noise generated during the operation of air compressors introduces interference that impacts fault identification accuracy. To overcome these limitations, a feature fusion-based fault diagnosis method for air compressors is proposed. Firstly, Mel-frequency cepstral coefficients (MFCC) features and wavelet transform features of the air compressor are extracted separately. Then, at the decision layer, confidence scores and predicted bounding boxes are fused late in the process, and the best network model is selected based on evaluation metrics to complete the classification. Comparative experimental results show that this feature fusion method significantly improves fault identification accuracy.
